android_kernel_motorola_sm6225/drivers/s390/block/dasd.c

2514 lines
66 KiB
C
Raw Normal View History

/*
* File...........: linux/drivers/s390/block/dasd.c
* Author(s)......: Holger Smolinski <Holger.Smolinski@de.ibm.com>
* Horst Hummel <Horst.Hummel@de.ibm.com>
* Carsten Otte <Cotte@de.ibm.com>
* Martin Schwidefsky <schwidefsky@de.ibm.com>
* Bugreports.to..: <Linux390@de.ibm.com>
* (C) IBM Corporation, IBM Deutschland Entwicklung GmbH, 1999-2001
*
*/
#include <linux/kmod.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/ctype.h>
#include <linux/major.h>
#include <linux/slab.h>
#include <linux/buffer_head.h>
#include <linux/hdreg.h>
#include <asm/ccwdev.h>
#include <asm/ebcdic.h>
#include <asm/idals.h>
#include <asm/todclk.h>
/* This is ugly... */
#define PRINTK_HEADER "dasd:"
#include "dasd_int.h"
/*
* SECTION: Constant definitions to be used within this file
*/
#define DASD_CHANQ_MAX_SIZE 4
/*
* SECTION: exported variables of dasd.c
*/
debug_info_t *dasd_debug_area;
struct dasd_discipline *dasd_diag_discipline_pointer;
void dasd_int_handler(struct ccw_device *, unsigned long, struct irb *);
MODULE_AUTHOR("Holger Smolinski <Holger.Smolinski@de.ibm.com>");
MODULE_DESCRIPTION("Linux on S/390 DASD device driver,"
" Copyright 2000 IBM Corporation");
MODULE_SUPPORTED_DEVICE("dasd");
MODULE_LICENSE("GPL");
/*
* SECTION: prototypes for static functions of dasd.c
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_alloc_queue(struct dasd_block *);
static void dasd_setup_queue(struct dasd_block *);
static void dasd_free_queue(struct dasd_block *);
static void dasd_flush_request_queue(struct dasd_block *);
static int dasd_flush_block_queue(struct dasd_block *);
static void dasd_device_tasklet(struct dasd_device *);
static void dasd_block_tasklet(struct dasd_block *);
static void do_kick_device(struct work_struct *);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_return_cqr_cb(struct dasd_ccw_req *, void *);
/*
* SECTION: Operations on the device structure.
*/
static wait_queue_head_t dasd_init_waitq;
static wait_queue_head_t dasd_flush_wq;
/*
* Allocate memory for a new device structure.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_device *dasd_alloc_device(void)
{
struct dasd_device *device;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = kzalloc(sizeof(struct dasd_device), GFP_ATOMIC);
if (!device)
return ERR_PTR(-ENOMEM);
/* Get two pages for normal block device operations. */
device->ccw_mem = (void *) __get_free_pages(GFP_ATOMIC | GFP_DMA, 1);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (!device->ccw_mem) {
kfree(device);
return ERR_PTR(-ENOMEM);
}
/* Get one page for error recovery. */
device->erp_mem = (void *) get_zeroed_page(GFP_ATOMIC | GFP_DMA);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (!device->erp_mem) {
free_pages((unsigned long) device->ccw_mem, 1);
kfree(device);
return ERR_PTR(-ENOMEM);
}
dasd_init_chunklist(&device->ccw_chunks, device->ccw_mem, PAGE_SIZE*2);
dasd_init_chunklist(&device->erp_chunks, device->erp_mem, PAGE_SIZE);
spin_lock_init(&device->mem_lock);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
atomic_set(&device->tasklet_scheduled, 0);
tasklet_init(&device->tasklet,
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
(void (*)(unsigned long)) dasd_device_tasklet,
(unsigned long) device);
INIT_LIST_HEAD(&device->ccw_queue);
init_timer(&device->timer);
INIT_WORK(&device->kick_work, do_kick_device);
device->state = DASD_STATE_NEW;
device->target = DASD_STATE_NEW;
return device;
}
/*
* Free memory of a device structure.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_free_device(struct dasd_device *device)
{
kfree(device->private);
free_page((unsigned long) device->erp_mem);
free_pages((unsigned long) device->ccw_mem, 1);
kfree(device);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/*
* Allocate memory for a new device structure.
*/
struct dasd_block *dasd_alloc_block(void)
{
struct dasd_block *block;
block = kzalloc(sizeof(*block), GFP_ATOMIC);
if (!block)
return ERR_PTR(-ENOMEM);
/* open_count = 0 means device online but not in use */
atomic_set(&block->open_count, -1);
spin_lock_init(&block->request_queue_lock);
atomic_set(&block->tasklet_scheduled, 0);
tasklet_init(&block->tasklet,
(void (*)(unsigned long)) dasd_block_tasklet,
(unsigned long) block);
INIT_LIST_HEAD(&block->ccw_queue);
spin_lock_init(&block->queue_lock);
init_timer(&block->timer);
return block;
}
/*
* Free memory of a device structure.
*/
void dasd_free_block(struct dasd_block *block)
{
kfree(block);
}
/*
* Make a new device known to the system.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_state_new_to_known(struct dasd_device *device)
{
int rc;
/*
* As long as the device is not in state DASD_STATE_NEW we want to
* keep the reference count > 0.
*/
dasd_get_device(device);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (device->block) {
rc = dasd_alloc_queue(device->block);
if (rc) {
dasd_put_device(device);
return rc;
}
}
device->state = DASD_STATE_KNOWN;
return 0;
}
/*
* Let the system forget about a device.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_state_known_to_new(struct dasd_device *device)
{
/* Disable extended error reporting for this device. */
dasd_eer_disable(device);
/* Forget the discipline information. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (device->discipline) {
if (device->discipline->uncheck_device)
device->discipline->uncheck_device(device);
module_put(device->discipline->owner);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
}
device->discipline = NULL;
if (device->base_discipline)
module_put(device->base_discipline->owner);
device->base_discipline = NULL;
device->state = DASD_STATE_NEW;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (device->block)
dasd_free_queue(device->block);
/* Give up reference we took in dasd_state_new_to_known. */
dasd_put_device(device);
return 0;
}
/*
* Request the irq line for the device.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_state_known_to_basic(struct dasd_device *device)
{
int rc;
/* Allocate and register gendisk structure. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (device->block) {
rc = dasd_gendisk_alloc(device->block);
if (rc)
return rc;
}
/* register 'device' debug area, used for all DBF_DEV_XXX calls */
device->debug_area = debug_register(device->cdev->dev.bus_id, 1, 1,
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
8 * sizeof(long));
debug_register_view(device->debug_area, &debug_sprintf_view);
debug_set_level(device->debug_area, DBF_WARNING);
DBF_DEV_EVENT(DBF_EMERG, device, "%s", "debug area created");
device->state = DASD_STATE_BASIC;
return 0;
}
/*
* Release the irq line for the device. Terminate any running i/o.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_state_basic_to_known(struct dasd_device *device)
{
int rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (device->block) {
dasd_gendisk_free(device->block);
dasd_block_clear_timer(device->block);
}
rc = dasd_flush_device_queue(device);
if (rc)
return rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_device_clear_timer(device);
DBF_DEV_EVENT(DBF_EMERG, device, "%p debug area deleted", device);
if (device->debug_area != NULL) {
debug_unregister(device->debug_area);
device->debug_area = NULL;
}
device->state = DASD_STATE_KNOWN;
return 0;
}
/*
* Do the initial analysis. The do_analysis function may return
* -EAGAIN in which case the device keeps the state DASD_STATE_BASIC
* until the discipline decides to continue the startup sequence
* by calling the function dasd_change_state. The eckd disciplines
* uses this to start a ccw that detects the format. The completion
* interrupt for this detection ccw uses the kernel event daemon to
* trigger the call to dasd_change_state. All this is done in the
* discipline code, see dasd_eckd.c.
* After the analysis ccw is done (do_analysis returned 0) the block
* device is setup.
* In case the analysis returns an error, the device setup is stopped
* (a fake disk was already added to allow formatting).
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_state_basic_to_ready(struct dasd_device *device)
{
int rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_block *block;
rc = 0;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
block = device->block;
/* make disk known with correct capacity */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (block) {
if (block->base->discipline->do_analysis != NULL)
rc = block->base->discipline->do_analysis(block);
if (rc) {
if (rc != -EAGAIN)
device->state = DASD_STATE_UNFMT;
return rc;
}
dasd_setup_queue(block);
set_capacity(block->gdp,
block->blocks << block->s2b_shift);
device->state = DASD_STATE_READY;
rc = dasd_scan_partitions(block);
if (rc)
device->state = DASD_STATE_BASIC;
} else {
device->state = DASD_STATE_READY;
}
return rc;
}
/*
* Remove device from block device layer. Destroy dirty buffers.
* Forget format information. Check if the target level is basic
* and if it is create fake disk for formatting.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_state_ready_to_basic(struct dasd_device *device)
{
int rc;
device->state = DASD_STATE_BASIC;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (device->block) {
struct dasd_block *block = device->block;
rc = dasd_flush_block_queue(block);
if (rc) {
device->state = DASD_STATE_READY;
return rc;
}
dasd_destroy_partitions(block);
dasd_flush_request_queue(block);
block->blocks = 0;
block->bp_block = 0;
block->s2b_shift = 0;
}
return 0;
}
/*
* Back to basic.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_state_unfmt_to_basic(struct dasd_device *device)
{
device->state = DASD_STATE_BASIC;
return 0;
}
/*
* Make the device online and schedule the bottom half to start
* the requeueing of requests from the linux request queue to the
* ccw queue.
*/
static int
dasd_state_ready_to_online(struct dasd_device * device)
{
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int rc;
if (device->discipline->ready_to_online) {
rc = device->discipline->ready_to_online(device);
if (rc)
return rc;
}
device->state = DASD_STATE_ONLINE;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (device->block)
dasd_schedule_block_bh(device->block);
return 0;
}
/*
* Stop the requeueing of requests again.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_state_online_to_ready(struct dasd_device *device)
{
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int rc;
if (device->discipline->online_to_ready) {
rc = device->discipline->online_to_ready(device);
if (rc)
return rc;
}
device->state = DASD_STATE_READY;
return 0;
}
/*
* Device startup state changes.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_increase_state(struct dasd_device *device)
{
int rc;
rc = 0;
if (device->state == DASD_STATE_NEW &&
device->target >= DASD_STATE_KNOWN)
rc = dasd_state_new_to_known(device);
if (!rc &&
device->state == DASD_STATE_KNOWN &&
device->target >= DASD_STATE_BASIC)
rc = dasd_state_known_to_basic(device);
if (!rc &&
device->state == DASD_STATE_BASIC &&
device->target >= DASD_STATE_READY)
rc = dasd_state_basic_to_ready(device);
if (!rc &&
device->state == DASD_STATE_UNFMT &&
device->target > DASD_STATE_UNFMT)
rc = -EPERM;
if (!rc &&
device->state == DASD_STATE_READY &&
device->target >= DASD_STATE_ONLINE)
rc = dasd_state_ready_to_online(device);
return rc;
}
/*
* Device shutdown state changes.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_decrease_state(struct dasd_device *device)
{
int rc;
rc = 0;
if (device->state == DASD_STATE_ONLINE &&
device->target <= DASD_STATE_READY)
rc = dasd_state_online_to_ready(device);
if (!rc &&
device->state == DASD_STATE_READY &&
device->target <= DASD_STATE_BASIC)
rc = dasd_state_ready_to_basic(device);
if (!rc &&
device->state == DASD_STATE_UNFMT &&
device->target <= DASD_STATE_BASIC)
rc = dasd_state_unfmt_to_basic(device);
if (!rc &&
device->state == DASD_STATE_BASIC &&
device->target <= DASD_STATE_KNOWN)
rc = dasd_state_basic_to_known(device);
if (!rc &&
device->state == DASD_STATE_KNOWN &&
device->target <= DASD_STATE_NEW)
rc = dasd_state_known_to_new(device);
return rc;
}
/*
* This is the main startup/shutdown routine.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_change_state(struct dasd_device *device)
{
int rc;
if (device->state == device->target)
/* Already where we want to go today... */
return;
if (device->state < device->target)
rc = dasd_increase_state(device);
else
rc = dasd_decrease_state(device);
if (rc && rc != -EAGAIN)
device->target = device->state;
if (device->state == device->target)
wake_up(&dasd_init_waitq);
/* let user-space know that the device status changed */
kobject_uevent(&device->cdev->dev.kobj, KOBJ_CHANGE);
}
/*
* Kick starter for devices that did not complete the startup/shutdown
* procedure or were sleeping because of a pending state.
* dasd_kick_device will schedule a call do do_kick_device to the kernel
* event daemon.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void do_kick_device(struct work_struct *work)
{
struct dasd_device *device = container_of(work, struct dasd_device, kick_work);
dasd_change_state(device);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_schedule_device_bh(device);
dasd_put_device(device);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_kick_device(struct dasd_device *device)
{
dasd_get_device(device);
/* queue call to dasd_kick_device to the kernel event daemon. */
schedule_work(&device->kick_work);
}
/*
* Set the target state for a device and starts the state change.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_set_target_state(struct dasd_device *device, int target)
{
/* If we are in probeonly mode stop at DASD_STATE_READY. */
if (dasd_probeonly && target > DASD_STATE_READY)
target = DASD_STATE_READY;
if (device->target != target) {
if (device->state == target)
wake_up(&dasd_init_waitq);
device->target = target;
}
if (device->state != device->target)
dasd_change_state(device);
}
/*
* Enable devices with device numbers in [from..to].
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static inline int _wait_for_device(struct dasd_device *device)
{
return (device->state == device->target);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_enable_device(struct dasd_device *device)
{
dasd_set_target_state(device, DASD_STATE_ONLINE);
if (device->state <= DASD_STATE_KNOWN)
/* No discipline for device found. */
dasd_set_target_state(device, DASD_STATE_NEW);
/* Now wait for the devices to come up. */
wait_event(dasd_init_waitq, _wait_for_device(device));
}
/*
* SECTION: device operation (interrupt handler, start i/o, term i/o ...)
*/
#ifdef CONFIG_DASD_PROFILE
struct dasd_profile_info_t dasd_global_profile;
unsigned int dasd_profile_level = DASD_PROFILE_OFF;
/*
* Increments counter in global and local profiling structures.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
#define dasd_profile_counter(value, counter, block) \
{ \
int index; \
for (index = 0; index < 31 && value >> (2+index); index++); \
dasd_global_profile.counter[index]++; \
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
block->profile.counter[index]++; \
}
/*
* Add profiling information for cqr before execution.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_profile_start(struct dasd_block *block,
struct dasd_ccw_req *cqr,
struct request *req)
{
struct list_head *l;
unsigned int counter;
if (dasd_profile_level != DASD_PROFILE_ON)
return;
/* count the length of the chanq for statistics */
counter = 0;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
list_for_each(l, &block->ccw_queue)
if (++counter >= 31)
break;
dasd_global_profile.dasd_io_nr_req[counter]++;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
block->profile.dasd_io_nr_req[counter]++;
}
/*
* Add profiling information for cqr after execution.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_profile_end(struct dasd_block *block,
struct dasd_ccw_req *cqr,
struct request *req)
{
long strtime, irqtime, endtime, tottime; /* in microseconds */
long tottimeps, sectors;
if (dasd_profile_level != DASD_PROFILE_ON)
return;
sectors = req->nr_sectors;
if (!cqr->buildclk || !cqr->startclk ||
!cqr->stopclk || !cqr->endclk ||
!sectors)
return;
strtime = ((cqr->startclk - cqr->buildclk) >> 12);
irqtime = ((cqr->stopclk - cqr->startclk) >> 12);
endtime = ((cqr->endclk - cqr->stopclk) >> 12);
tottime = ((cqr->endclk - cqr->buildclk) >> 12);
tottimeps = tottime / sectors;
if (!dasd_global_profile.dasd_io_reqs)
memset(&dasd_global_profile, 0,
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
sizeof(struct dasd_profile_info_t));
dasd_global_profile.dasd_io_reqs++;
dasd_global_profile.dasd_io_sects += sectors;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (!block->profile.dasd_io_reqs)
memset(&block->profile, 0,
sizeof(struct dasd_profile_info_t));
block->profile.dasd_io_reqs++;
block->profile.dasd_io_sects += sectors;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_profile_counter(sectors, dasd_io_secs, block);
dasd_profile_counter(tottime, dasd_io_times, block);
dasd_profile_counter(tottimeps, dasd_io_timps, block);
dasd_profile_counter(strtime, dasd_io_time1, block);
dasd_profile_counter(irqtime, dasd_io_time2, block);
dasd_profile_counter(irqtime / sectors, dasd_io_time2ps, block);
dasd_profile_counter(endtime, dasd_io_time3, block);
}
#else
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
#define dasd_profile_start(block, cqr, req) do {} while (0)
#define dasd_profile_end(block, cqr, req) do {} while (0)
#endif /* CONFIG_DASD_PROFILE */
/*
* Allocate memory for a channel program with 'cplength' channel
* command words and 'datasize' additional space. There are two
* variantes: 1) dasd_kmalloc_request uses kmalloc to get the needed
* memory and 2) dasd_smalloc_request uses the static ccw memory
* that gets allocated for each device.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_ccw_req *dasd_kmalloc_request(char *magic, int cplength,
int datasize,
struct dasd_device *device)
{
struct dasd_ccw_req *cqr;
/* Sanity checks */
BUG_ON( magic == NULL || datasize > PAGE_SIZE ||
(cplength*sizeof(struct ccw1)) > PAGE_SIZE);
cqr = kzalloc(sizeof(struct dasd_ccw_req), GFP_ATOMIC);
if (cqr == NULL)
return ERR_PTR(-ENOMEM);
cqr->cpaddr = NULL;
if (cplength > 0) {
cqr->cpaddr = kcalloc(cplength, sizeof(struct ccw1),
GFP_ATOMIC | GFP_DMA);
if (cqr->cpaddr == NULL) {
kfree(cqr);
return ERR_PTR(-ENOMEM);
}
}
cqr->data = NULL;
if (datasize > 0) {
cqr->data = kzalloc(datasize, GFP_ATOMIC | GFP_DMA);
if (cqr->data == NULL) {
kfree(cqr->cpaddr);
kfree(cqr);
return ERR_PTR(-ENOMEM);
}
}
strncpy((char *) &cqr->magic, magic, 4);
ASCEBC((char *) &cqr->magic, 4);
set_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
dasd_get_device(device);
return cqr;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_ccw_req *dasd_smalloc_request(char *magic, int cplength,
int datasize,
struct dasd_device *device)
{
unsigned long flags;
struct dasd_ccw_req *cqr;
char *data;
int size;
/* Sanity checks */
BUG_ON( magic == NULL || datasize > PAGE_SIZE ||
(cplength*sizeof(struct ccw1)) > PAGE_SIZE);
size = (sizeof(struct dasd_ccw_req) + 7L) & -8L;
if (cplength > 0)
size += cplength * sizeof(struct ccw1);
if (datasize > 0)
size += datasize;
spin_lock_irqsave(&device->mem_lock, flags);
cqr = (struct dasd_ccw_req *)
dasd_alloc_chunk(&device->ccw_chunks, size);
spin_unlock_irqrestore(&device->mem_lock, flags);
if (cqr == NULL)
return ERR_PTR(-ENOMEM);
memset(cqr, 0, sizeof(struct dasd_ccw_req));
data = (char *) cqr + ((sizeof(struct dasd_ccw_req) + 7L) & -8L);
cqr->cpaddr = NULL;
if (cplength > 0) {
cqr->cpaddr = (struct ccw1 *) data;
data += cplength*sizeof(struct ccw1);
memset(cqr->cpaddr, 0, cplength*sizeof(struct ccw1));
}
cqr->data = NULL;
if (datasize > 0) {
cqr->data = data;
memset(cqr->data, 0, datasize);
}
strncpy((char *) &cqr->magic, magic, 4);
ASCEBC((char *) &cqr->magic, 4);
set_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
dasd_get_device(device);
return cqr;
}
/*
* Free memory of a channel program. This function needs to free all the
* idal lists that might have been created by dasd_set_cda and the
* struct dasd_ccw_req itself.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_kfree_request(struct dasd_ccw_req *cqr, struct dasd_device *device)
{
#ifdef CONFIG_64BIT
struct ccw1 *ccw;
/* Clear any idals used for the request. */
ccw = cqr->cpaddr;
do {
clear_normalized_cda(ccw);
} while (ccw++->flags & (CCW_FLAG_CC | CCW_FLAG_DC));
#endif
kfree(cqr->cpaddr);
kfree(cqr->data);
kfree(cqr);
dasd_put_device(device);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_sfree_request(struct dasd_ccw_req *cqr, struct dasd_device *device)
{
unsigned long flags;
spin_lock_irqsave(&device->mem_lock, flags);
dasd_free_chunk(&device->ccw_chunks, cqr);
spin_unlock_irqrestore(&device->mem_lock, flags);
dasd_put_device(device);
}
/*
* Check discipline magic in cqr.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static inline int dasd_check_cqr(struct dasd_ccw_req *cqr)
{
struct dasd_device *device;
if (cqr == NULL)
return -EINVAL;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = cqr->startdev;
if (strncmp((char *) &cqr->magic, device->discipline->ebcname, 4)) {
DEV_MESSAGE(KERN_WARNING, device,
" dasd_ccw_req 0x%08x magic doesn't match"
" discipline 0x%08x",
cqr->magic,
*(unsigned int *) device->discipline->name);
return -EINVAL;
}
return 0;
}
/*
* Terminate the current i/o and set the request to clear_pending.
* Timer keeps device runnig.
* ccw_device_clear can fail if the i/o subsystem
* is in a bad mood.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_term_IO(struct dasd_ccw_req *cqr)
{
struct dasd_device *device;
int retries, rc;
/* Check the cqr */
rc = dasd_check_cqr(cqr);
if (rc)
return rc;
retries = 0;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = (struct dasd_device *) cqr->startdev;
while ((retries < 5) && (cqr->status == DASD_CQR_IN_IO)) {
rc = ccw_device_clear(device->cdev, (long) cqr);
switch (rc) {
case 0: /* termination successful */
cqr->retries--;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr->status = DASD_CQR_CLEAR_PENDING;
cqr->stopclk = get_clock();
cqr->starttime = 0;
DBF_DEV_EVENT(DBF_DEBUG, device,
"terminate cqr %p successful",
cqr);
break;
case -ENODEV:
DBF_DEV_EVENT(DBF_ERR, device, "%s",
"device gone, retry");
break;
case -EIO:
DBF_DEV_EVENT(DBF_ERR, device, "%s",
"I/O error, retry");
break;
case -EINVAL:
case -EBUSY:
DBF_DEV_EVENT(DBF_ERR, device, "%s",
"device busy, retry later");
break;
default:
DEV_MESSAGE(KERN_ERR, device,
"line %d unknown RC=%d, please "
"report to linux390@de.ibm.com",
__LINE__, rc);
BUG();
break;
}
retries++;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_schedule_device_bh(device);
return rc;
}
/*
* Start the i/o. This start_IO can fail if the channel is really busy.
* In that case set up a timer to start the request later.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_start_IO(struct dasd_ccw_req *cqr)
{
struct dasd_device *device;
int rc;
/* Check the cqr */
rc = dasd_check_cqr(cqr);
if (rc)
return rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = (struct dasd_device *) cqr->startdev;
if (cqr->retries < 0) {
DEV_MESSAGE(KERN_DEBUG, device,
"start_IO: request %p (%02x/%i) - no retry left.",
cqr, cqr->status, cqr->retries);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr->status = DASD_CQR_ERROR;
return -EIO;
}
cqr->startclk = get_clock();
cqr->starttime = jiffies;
cqr->retries--;
rc = ccw_device_start(device->cdev, cqr->cpaddr, (long) cqr,
cqr->lpm, 0);
switch (rc) {
case 0:
cqr->status = DASD_CQR_IN_IO;
DBF_DEV_EVENT(DBF_DEBUG, device,
"start_IO: request %p started successful",
cqr);
break;
case -EBUSY:
DBF_DEV_EVENT(DBF_ERR, device, "%s",
"start_IO: device busy, retry later");
break;
case -ETIMEDOUT:
DBF_DEV_EVENT(DBF_ERR, device, "%s",
"start_IO: request timeout, retry later");
break;
case -EACCES:
/* -EACCES indicates that the request used only a
* subset of the available pathes and all these
* pathes are gone.
* Do a retry with all available pathes.
*/
cqr->lpm = LPM_ANYPATH;
DBF_DEV_EVENT(DBF_ERR, device, "%s",
"start_IO: selected pathes gone,"
" retry on all pathes");
break;
case -ENODEV:
case -EIO:
DBF_DEV_EVENT(DBF_ERR, device, "%s",
"start_IO: device gone, retry");
break;
default:
DEV_MESSAGE(KERN_ERR, device,
"line %d unknown RC=%d, please report"
" to linux390@de.ibm.com", __LINE__, rc);
BUG();
break;
}
return rc;
}
/*
* Timeout function for dasd devices. This is used for different purposes
* 1) missing interrupt handler for normal operation
* 2) delayed start of request where start_IO failed with -EBUSY
* 3) timeout for missing state change interrupts
* The head of the ccw queue will have status DASD_CQR_IN_IO for 1),
* DASD_CQR_QUEUED for 2) and 3).
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_device_timeout(unsigned long ptr)
{
unsigned long flags;
struct dasd_device *device;
device = (struct dasd_device *) ptr;
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
/* re-activate request queue */
device->stopped &= ~DASD_STOPPED_PENDING;
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_schedule_device_bh(device);
}
/*
* Setup timeout for a device in jiffies.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_device_set_timer(struct dasd_device *device, int expires)
{
if (expires == 0) {
if (timer_pending(&device->timer))
del_timer(&device->timer);
return;
}
if (timer_pending(&device->timer)) {
if (mod_timer(&device->timer, jiffies + expires))
return;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device->timer.function = dasd_device_timeout;
device->timer.data = (unsigned long) device;
device->timer.expires = jiffies + expires;
add_timer(&device->timer);
}
/*
* Clear timeout for a device.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_device_clear_timer(struct dasd_device *device)
{
if (timer_pending(&device->timer))
del_timer(&device->timer);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_handle_killed_request(struct ccw_device *cdev,
unsigned long intparm)
{
struct dasd_ccw_req *cqr;
struct dasd_device *device;
cqr = (struct dasd_ccw_req *) intparm;
if (cqr->status != DASD_CQR_IN_IO) {
MESSAGE(KERN_DEBUG,
"invalid status in handle_killed_request: "
"bus_id %s, status %02x",
cdev->dev.bus_id, cqr->status);
return;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = (struct dasd_device *) cqr->startdev;
if (device == NULL ||
device != dasd_device_from_cdev_locked(cdev) ||
strncmp(device->discipline->ebcname, (char *) &cqr->magic, 4)) {
MESSAGE(KERN_DEBUG, "invalid device in request: bus_id %s",
cdev->dev.bus_id);
return;
}
/* Schedule request to be retried. */
cqr->status = DASD_CQR_QUEUED;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_device_clear_timer(device);
dasd_schedule_device_bh(device);
dasd_put_device(device);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_generic_handle_state_change(struct dasd_device *device)
{
/* First of all start sense subsystem status request. */
dasd_eer_snss(device);
device->stopped &= ~DASD_STOPPED_PENDING;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_schedule_device_bh(device);
if (device->block)
dasd_schedule_block_bh(device->block);
}
/*
* Interrupt handler for "normal" ssch-io based dasd devices.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_int_handler(struct ccw_device *cdev, unsigned long intparm,
struct irb *irb)
{
struct dasd_ccw_req *cqr, *next;
struct dasd_device *device;
unsigned long long now;
int expires;
if (IS_ERR(irb)) {
switch (PTR_ERR(irb)) {
case -EIO:
dasd_handle_killed_request(cdev, intparm);
break;
case -ETIMEDOUT:
printk(KERN_WARNING"%s(%s): request timed out\n",
__FUNCTION__, cdev->dev.bus_id);
//FIXME - dasd uses own timeout interface...
break;
default:
printk(KERN_WARNING"%s(%s): unknown error %ld\n",
__FUNCTION__, cdev->dev.bus_id, PTR_ERR(irb));
}
return;
}
now = get_clock();
DBF_EVENT(DBF_ERR, "Interrupt: bus_id %s CS/DS %04x ip %08x",
cdev->dev.bus_id, ((irb->scsw.cstat<<8)|irb->scsw.dstat),
(unsigned int) intparm);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/* check for unsolicited interrupts */
cqr = (struct dasd_ccw_req *) intparm;
if (!cqr || ((irb->scsw.cc == 1) &&
(irb->scsw.fctl & SCSW_FCTL_START_FUNC) &&
(irb->scsw.stctl & SCSW_STCTL_STATUS_PEND)) ) {
if (cqr && cqr->status == DASD_CQR_IN_IO)
cqr->status = DASD_CQR_QUEUED;
device = dasd_device_from_cdev_locked(cdev);
if (!IS_ERR(device)) {
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_device_clear_timer(device);
device->discipline->handle_unsolicited_interrupt(device,
irb);
dasd_put_device(device);
}
return;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = (struct dasd_device *) cqr->startdev;
if (!device ||
strncmp(device->discipline->ebcname, (char *) &cqr->magic, 4)) {
MESSAGE(KERN_DEBUG, "invalid device in request: bus_id %s",
cdev->dev.bus_id);
return;
}
/* Check for clear pending */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (cqr->status == DASD_CQR_CLEAR_PENDING &&
irb->scsw.fctl & SCSW_FCTL_CLEAR_FUNC) {
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr->status = DASD_CQR_CLEARED;
dasd_device_clear_timer(device);
wake_up(&dasd_flush_wq);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_schedule_device_bh(device);
return;
}
/* check status - the request might have been killed by dyn detach */
if (cqr->status != DASD_CQR_IN_IO) {
MESSAGE(KERN_DEBUG,
"invalid status: bus_id %s, status %02x",
cdev->dev.bus_id, cqr->status);
return;
}
DBF_DEV_EVENT(DBF_DEBUG, device, "Int: CS/DS 0x%04x for cqr %p",
((irb->scsw.cstat << 8) | irb->scsw.dstat), cqr);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
next = NULL;
expires = 0;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (irb->scsw.dstat == (DEV_STAT_CHN_END | DEV_STAT_DEV_END) &&
irb->scsw.cstat == 0 && !irb->esw.esw0.erw.cons) {
/* request was completed successfully */
cqr->status = DASD_CQR_SUCCESS;
cqr->stopclk = now;
/* Start first request on queue if possible -> fast_io. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (cqr->devlist.next != &device->ccw_queue) {
next = list_entry(cqr->devlist.next,
struct dasd_ccw_req, devlist);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
} else { /* error */
memcpy(&cqr->irb, irb, sizeof(struct irb));
if (device->features & DASD_FEATURE_ERPLOG) {
dasd_log_sense(cqr, irb);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/* If we have no sense data, or we just don't want complex ERP
* for this request, but if we have retries left, then just
* reset this request and retry it in the fastpath
*/
if (!(cqr->irb.esw.esw0.erw.cons &&
test_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags)) &&
cqr->retries > 0) {
DEV_MESSAGE(KERN_DEBUG, device,
"default ERP in fastpath (%i retries left)",
cqr->retries);
cqr->lpm = LPM_ANYPATH;
cqr->status = DASD_CQR_QUEUED;
next = cqr;
} else
cqr->status = DASD_CQR_ERROR;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
}
if (next && (next->status == DASD_CQR_QUEUED) &&
(!device->stopped)) {
if (device->discipline->start_IO(next) == 0)
expires = next->expires;
else
DEV_MESSAGE(KERN_DEBUG, device, "%s",
"Interrupt fastpath "
"failed!");
}
if (expires != 0)
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_device_set_timer(device, expires);
else
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_device_clear_timer(device);
dasd_schedule_device_bh(device);
}
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* If we have an error on a dasd_block layer request then we cancel
* and return all further requests from the same dasd_block as well.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void __dasd_device_recovery(struct dasd_device *device,
struct dasd_ccw_req *ref_cqr)
{
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct list_head *l, *n;
struct dasd_ccw_req *cqr;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/*
* only requeue request that came from the dasd_block layer
*/
if (!ref_cqr->block)
return;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
list_for_each_safe(l, n, &device->ccw_queue) {
cqr = list_entry(l, struct dasd_ccw_req, devlist);
if (cqr->status == DASD_CQR_QUEUED &&
ref_cqr->block == cqr->block) {
cqr->status = DASD_CQR_CLEARED;
}
}
};
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* Remove those ccw requests from the queue that need to be returned
* to the upper layer.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void __dasd_device_process_ccw_queue(struct dasd_device *device,
struct list_head *final_queue)
{
struct list_head *l, *n;
struct dasd_ccw_req *cqr;
/* Process request with final status. */
list_for_each_safe(l, n, &device->ccw_queue) {
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr = list_entry(l, struct dasd_ccw_req, devlist);
/* Stop list processing at the first non-final request. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (cqr->status == DASD_CQR_QUEUED ||
cqr->status == DASD_CQR_IN_IO ||
cqr->status == DASD_CQR_CLEAR_PENDING)
break;
if (cqr->status == DASD_CQR_ERROR) {
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
__dasd_device_recovery(device, cqr);
}
/* Rechain finished requests to final queue */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
list_move_tail(&cqr->devlist, final_queue);
}
}
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* the cqrs from the final queue are returned to the upper layer
* by setting a dasd_block state and calling the callback function
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void __dasd_device_process_final_queue(struct dasd_device *device,
struct list_head *final_queue)
{
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct list_head *l, *n;
struct dasd_ccw_req *cqr;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
list_for_each_safe(l, n, final_queue) {
cqr = list_entry(l, struct dasd_ccw_req, devlist);
list_del_init(&cqr->devlist);
if (cqr->block)
spin_lock_bh(&cqr->block->queue_lock);
switch (cqr->status) {
case DASD_CQR_SUCCESS:
cqr->status = DASD_CQR_DONE;
break;
case DASD_CQR_ERROR:
cqr->status = DASD_CQR_NEED_ERP;
break;
case DASD_CQR_CLEARED:
cqr->status = DASD_CQR_TERMINATED;
break;
default:
DEV_MESSAGE(KERN_ERR, device,
"wrong cqr status in __dasd_process_final_queue "
"for cqr %p, status %x",
cqr, cqr->status);
BUG();
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (cqr->block)
spin_unlock_bh(&cqr->block->queue_lock);
if (cqr->callback != NULL)
(cqr->callback)(cqr, cqr->callback_data);
}
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/*
* Take a look at the first request on the ccw queue and check
* if it reached its expire time. If so, terminate the IO.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void __dasd_device_check_expire(struct dasd_device *device)
{
struct dasd_ccw_req *cqr;
if (list_empty(&device->ccw_queue))
return;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr = list_entry(device->ccw_queue.next, struct dasd_ccw_req, devlist);
if ((cqr->status == DASD_CQR_IN_IO && cqr->expires != 0) &&
(time_after_eq(jiffies, cqr->expires + cqr->starttime))) {
if (device->discipline->term_IO(cqr) != 0) {
/* Hmpf, try again in 5 sec */
DEV_MESSAGE(KERN_ERR, device,
"internal error - timeout (%is) expired "
"for cqr %p, termination failed, "
"retrying in 5s",
(cqr->expires/HZ), cqr);
cqr->expires += 5*HZ;
dasd_device_set_timer(device, 5*HZ);
} else {
DEV_MESSAGE(KERN_ERR, device,
"internal error - timeout (%is) expired "
"for cqr %p (%i retries left)",
(cqr->expires/HZ), cqr, cqr->retries);
}
}
}
/*
* Take a look at the first request on the ccw queue and check
* if it needs to be started.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void __dasd_device_start_head(struct dasd_device *device)
{
struct dasd_ccw_req *cqr;
int rc;
if (list_empty(&device->ccw_queue))
return;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr = list_entry(device->ccw_queue.next, struct dasd_ccw_req, devlist);
if (cqr->status != DASD_CQR_QUEUED)
return;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/* when device is stopped, return request to previous layer */
if (device->stopped) {
cqr->status = DASD_CQR_CLEARED;
dasd_schedule_device_bh(device);
return;
}
rc = device->discipline->start_IO(cqr);
if (rc == 0)
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_device_set_timer(device, cqr->expires);
else if (rc == -EACCES) {
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_schedule_device_bh(device);
} else
/* Hmpf, try again in 1/2 sec */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_device_set_timer(device, 50);
}
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* Go through all request on the dasd_device request queue,
* terminate them on the cdev if necessary, and return them to the
* submitting layer via callback.
* Note:
* Make sure that all 'submitting layers' still exist when
* this function is called!. In other words, when 'device' is a base
* device then all block layer requests must have been removed before
* via dasd_flush_block_queue.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_flush_device_queue(struct dasd_device *device)
{
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_ccw_req *cqr, *n;
int rc;
struct list_head flush_queue;
INIT_LIST_HEAD(&flush_queue);
spin_lock_irq(get_ccwdev_lock(device->cdev));
rc = 0;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
list_for_each_entry_safe(cqr, n, &device->ccw_queue, devlist) {
/* Check status and move request to flush_queue */
switch (cqr->status) {
case DASD_CQR_IN_IO:
rc = device->discipline->term_IO(cqr);
if (rc) {
/* unable to terminate requeust */
DEV_MESSAGE(KERN_ERR, device,
"dasd flush ccw_queue is unable "
" to terminate request %p",
cqr);
/* stop flush processing */
goto finished;
}
break;
case DASD_CQR_QUEUED:
cqr->stopclk = get_clock();
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr->status = DASD_CQR_CLEARED;
break;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
default: /* no need to modify the others */
break;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
list_move_tail(&cqr->devlist, &flush_queue);
}
finished:
spin_unlock_irq(get_ccwdev_lock(device->cdev));
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/*
* After this point all requests must be in state CLEAR_PENDING,
* CLEARED, SUCCESS or ERROR. Now wait for CLEAR_PENDING to become
* one of the others.
*/
list_for_each_entry_safe(cqr, n, &flush_queue, devlist)
wait_event(dasd_flush_wq,
(cqr->status != DASD_CQR_CLEAR_PENDING));
/*
* Now set each request back to TERMINATED, DONE or NEED_ERP
* and call the callback function of flushed requests
*/
__dasd_device_process_final_queue(device, &flush_queue);
return rc;
}
/*
* Acquire the device lock and process queues for the device.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_device_tasklet(struct dasd_device *device)
{
struct list_head final_queue;
atomic_set (&device->tasklet_scheduled, 0);
INIT_LIST_HEAD(&final_queue);
spin_lock_irq(get_ccwdev_lock(device->cdev));
/* Check expire time of first request on the ccw queue. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
__dasd_device_check_expire(device);
/* find final requests on ccw queue */
__dasd_device_process_ccw_queue(device, &final_queue);
spin_unlock_irq(get_ccwdev_lock(device->cdev));
/* Now call the callback function of requests with final status */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
__dasd_device_process_final_queue(device, &final_queue);
spin_lock_irq(get_ccwdev_lock(device->cdev));
/* Now check if the head of the ccw queue needs to be started. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
__dasd_device_start_head(device);
spin_unlock_irq(get_ccwdev_lock(device->cdev));
dasd_put_device(device);
}
/*
* Schedules a call to dasd_tasklet over the device tasklet.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_schedule_device_bh(struct dasd_device *device)
{
/* Protect against rescheduling. */
if (atomic_cmpxchg (&device->tasklet_scheduled, 0, 1) != 0)
return;
dasd_get_device(device);
tasklet_hi_schedule(&device->tasklet);
}
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* Queue a request to the head of the device ccw_queue.
* Start the I/O if possible.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_add_request_head(struct dasd_ccw_req *cqr)
{
struct dasd_device *device;
unsigned long flags;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = cqr->startdev;
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr->status = DASD_CQR_QUEUED;
list_add(&cqr->devlist, &device->ccw_queue);
/* let the bh start the request to keep them in order */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_schedule_device_bh(device);
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
}
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* Queue a request to the tail of the device ccw_queue.
* Start the I/O if possible.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_add_request_tail(struct dasd_ccw_req *cqr)
{
struct dasd_device *device;
unsigned long flags;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = cqr->startdev;
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr->status = DASD_CQR_QUEUED;
list_add_tail(&cqr->devlist, &device->ccw_queue);
/* let the bh start the request to keep them in order */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_schedule_device_bh(device);
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
}
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* Wakeup helper for the 'sleep_on' functions.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_wakeup_cb(struct dasd_ccw_req *cqr, void *data)
{
wake_up((wait_queue_head_t *) data);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static inline int _wait_for_wakeup(struct dasd_ccw_req *cqr)
{
struct dasd_device *device;
int rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = cqr->startdev;
spin_lock_irq(get_ccwdev_lock(device->cdev));
rc = ((cqr->status == DASD_CQR_DONE ||
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr->status == DASD_CQR_NEED_ERP ||
cqr->status == DASD_CQR_TERMINATED) &&
list_empty(&cqr->devlist));
spin_unlock_irq(get_ccwdev_lock(device->cdev));
return rc;
}
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* Queue a request to the tail of the device ccw_queue and wait for
* it's completion.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_sleep_on(struct dasd_ccw_req *cqr)
{
wait_queue_head_t wait_q;
struct dasd_device *device;
int rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = cqr->startdev;
init_waitqueue_head (&wait_q);
cqr->callback = dasd_wakeup_cb;
cqr->callback_data = (void *) &wait_q;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_add_request_tail(cqr);
wait_event(wait_q, _wait_for_wakeup(cqr));
/* Request status is either done or failed. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
rc = (cqr->status == DASD_CQR_DONE) ? 0 : -EIO;
return rc;
}
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* Queue a request to the tail of the device ccw_queue and wait
* interruptible for it's completion.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_sleep_on_interruptible(struct dasd_ccw_req *cqr)
{
wait_queue_head_t wait_q;
struct dasd_device *device;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = cqr->startdev;
init_waitqueue_head (&wait_q);
cqr->callback = dasd_wakeup_cb;
cqr->callback_data = (void *) &wait_q;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_add_request_tail(cqr);
rc = wait_event_interruptible(wait_q, _wait_for_wakeup(cqr));
if (rc == -ERESTARTSYS) {
dasd_cancel_req(cqr);
/* wait (non-interruptible) for final status */
wait_event(wait_q, _wait_for_wakeup(cqr));
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
rc = (cqr->status == DASD_CQR_DONE) ? 0 : -EIO;
return rc;
}
/*
* Whoa nelly now it gets really hairy. For some functions (e.g. steal lock
* for eckd devices) the currently running request has to be terminated
* and be put back to status queued, before the special request is added
* to the head of the queue. Then the special request is waited on normally.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static inline int _dasd_term_running_cqr(struct dasd_device *device)
{
struct dasd_ccw_req *cqr;
if (list_empty(&device->ccw_queue))
return 0;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr = list_entry(device->ccw_queue.next, struct dasd_ccw_req, devlist);
return device->discipline->term_IO(cqr);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_sleep_on_immediatly(struct dasd_ccw_req *cqr)
{
wait_queue_head_t wait_q;
struct dasd_device *device;
int rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device = cqr->startdev;
spin_lock_irq(get_ccwdev_lock(device->cdev));
rc = _dasd_term_running_cqr(device);
if (rc) {
spin_unlock_irq(get_ccwdev_lock(device->cdev));
return rc;
}
init_waitqueue_head (&wait_q);
cqr->callback = dasd_wakeup_cb;
cqr->callback_data = (void *) &wait_q;
cqr->status = DASD_CQR_QUEUED;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
list_add(&cqr->devlist, &device->ccw_queue);
/* let the bh start the request to keep them in order */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_schedule_device_bh(device);
spin_unlock_irq(get_ccwdev_lock(device->cdev));
wait_event(wait_q, _wait_for_wakeup(cqr));
/* Request status is either done or failed. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
rc = (cqr->status == DASD_CQR_DONE) ? 0 : -EIO;
return rc;
}
/*
* Cancels a request that was started with dasd_sleep_on_req.
* This is useful to timeout requests. The request will be
* terminated if it is currently in i/o.
* Returns 1 if the request has been terminated.
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* 0 if there was no need to terminate the request (not started yet)
* negative error code if termination failed
* Cancellation of a request is an asynchronous operation! The calling
* function has to wait until the request is properly returned via callback.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_cancel_req(struct dasd_ccw_req *cqr)
{
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_device *device = cqr->startdev;
unsigned long flags;
int rc;
rc = 0;
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
switch (cqr->status) {
case DASD_CQR_QUEUED:
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/* request was not started - just set to cleared */
cqr->status = DASD_CQR_CLEARED;
break;
case DASD_CQR_IN_IO:
/* request in IO - terminate IO and release again */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
rc = device->discipline->term_IO(cqr);
if (rc) {
DEV_MESSAGE(KERN_ERR, device,
"dasd_cancel_req is unable "
" to terminate request %p, rc = %d",
cqr, rc);
} else {
cqr->stopclk = get_clock();
rc = 1;
}
break;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
default: /* already finished or clear pending - do nothing */
break;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
}
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
dasd_schedule_device_bh(device);
return rc;
}
/*
* SECTION: Operations of the dasd_block layer.
*/
/*
* Timeout function for dasd_block. This is used when the block layer
* is waiting for something that may not come reliably, (e.g. a state
* change interrupt)
*/
static void dasd_block_timeout(unsigned long ptr)
{
unsigned long flags;
struct dasd_block *block;
block = (struct dasd_block *) ptr;
spin_lock_irqsave(get_ccwdev_lock(block->base->cdev), flags);
/* re-activate request queue */
block->base->stopped &= ~DASD_STOPPED_PENDING;
spin_unlock_irqrestore(get_ccwdev_lock(block->base->cdev), flags);
dasd_schedule_block_bh(block);
}
/*
* Setup timeout for a dasd_block in jiffies.
*/
void dasd_block_set_timer(struct dasd_block *block, int expires)
{
if (expires == 0) {
if (timer_pending(&block->timer))
del_timer(&block->timer);
return;
}
if (timer_pending(&block->timer)) {
if (mod_timer(&block->timer, jiffies + expires))
return;
}
block->timer.function = dasd_block_timeout;
block->timer.data = (unsigned long) block;
block->timer.expires = jiffies + expires;
add_timer(&block->timer);
}
/*
* Clear timeout for a dasd_block.
*/
void dasd_block_clear_timer(struct dasd_block *block)
{
if (timer_pending(&block->timer))
del_timer(&block->timer);
}
/*
* posts the buffer_cache about a finalized request
*/
static inline void dasd_end_request(struct request *req, int error)
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
{
if (__blk_end_request(req, error, blk_rq_bytes(req)))
BUG();
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
}
/*
* Process finished error recovery ccw.
*/
static inline void __dasd_block_process_erp(struct dasd_block *block,
struct dasd_ccw_req *cqr)
{
dasd_erp_fn_t erp_fn;
struct dasd_device *device = block->base;
if (cqr->status == DASD_CQR_DONE)
DBF_DEV_EVENT(DBF_NOTICE, device, "%s", "ERP successful");
else
DEV_MESSAGE(KERN_ERR, device, "%s", "ERP unsuccessful");
erp_fn = device->discipline->erp_postaction(cqr);
erp_fn(cqr);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/*
* Fetch requests from the block device queue.
*/
static void __dasd_process_request_queue(struct dasd_block *block)
{
struct request_queue *queue;
struct request *req;
struct dasd_ccw_req *cqr;
struct dasd_device *basedev;
unsigned long flags;
queue = block->request_queue;
basedev = block->base;
/* No queue ? Then there is nothing to do. */
if (queue == NULL)
return;
/*
* We requeue request from the block device queue to the ccw
* queue only in two states. In state DASD_STATE_READY the
* partition detection is done and we need to requeue requests
* for that. State DASD_STATE_ONLINE is normal block device
* operation.
*/
if (basedev->state < DASD_STATE_READY)
return;
/* Now we try to fetch requests from the request queue */
while (!blk_queue_plugged(queue) &&
elv_next_request(queue)) {
req = elv_next_request(queue);
if (basedev->features & DASD_FEATURE_READONLY &&
rq_data_dir(req) == WRITE) {
DBF_DEV_EVENT(DBF_ERR, basedev,
"Rejecting write request %p",
req);
blkdev_dequeue_request(req);
dasd_end_request(req, -EIO);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
continue;
}
cqr = basedev->discipline->build_cp(basedev, block, req);
if (IS_ERR(cqr)) {
if (PTR_ERR(cqr) == -EBUSY)
break; /* normal end condition */
if (PTR_ERR(cqr) == -ENOMEM)
break; /* terminate request queue loop */
if (PTR_ERR(cqr) == -EAGAIN) {
/*
* The current request cannot be build right
* now, we have to try later. If this request
* is the head-of-queue we stop the device
* for 1/2 second.
*/
if (!list_empty(&block->ccw_queue))
break;
spin_lock_irqsave(get_ccwdev_lock(basedev->cdev), flags);
basedev->stopped |= DASD_STOPPED_PENDING;
spin_unlock_irqrestore(get_ccwdev_lock(basedev->cdev), flags);
dasd_block_set_timer(block, HZ/2);
break;
}
DBF_DEV_EVENT(DBF_ERR, basedev,
"CCW creation failed (rc=%ld) "
"on request %p",
PTR_ERR(cqr), req);
blkdev_dequeue_request(req);
dasd_end_request(req, -EIO);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
continue;
}
/*
* Note: callback is set to dasd_return_cqr_cb in
* __dasd_block_start_head to cover erp requests as well
*/
cqr->callback_data = (void *) req;
cqr->status = DASD_CQR_FILLED;
blkdev_dequeue_request(req);
list_add_tail(&cqr->blocklist, &block->ccw_queue);
dasd_profile_start(block, cqr, req);
}
}
static void __dasd_cleanup_cqr(struct dasd_ccw_req *cqr)
{
struct request *req;
int status;
int error = 0;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
req = (struct request *) cqr->callback_data;
dasd_profile_end(cqr->block, cqr, req);
status = cqr->memdev->discipline->free_cp(cqr, req);
if (status <= 0)
error = status ? status : -EIO;
dasd_end_request(req, error);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
}
/*
* Process ccw request queue.
*/
static void __dasd_process_block_ccw_queue(struct dasd_block *block,
struct list_head *final_queue)
{
struct list_head *l, *n;
struct dasd_ccw_req *cqr;
dasd_erp_fn_t erp_fn;
unsigned long flags;
struct dasd_device *base = block->base;
restart:
/* Process request with final status. */
list_for_each_safe(l, n, &block->ccw_queue) {
cqr = list_entry(l, struct dasd_ccw_req, blocklist);
if (cqr->status != DASD_CQR_DONE &&
cqr->status != DASD_CQR_FAILED &&
cqr->status != DASD_CQR_NEED_ERP &&
cqr->status != DASD_CQR_TERMINATED)
continue;
if (cqr->status == DASD_CQR_TERMINATED) {
base->discipline->handle_terminated_request(cqr);
goto restart;
}
/* Process requests that may be recovered */
if (cqr->status == DASD_CQR_NEED_ERP) {
if (cqr->irb.esw.esw0.erw.cons &&
test_bit(DASD_CQR_FLAGS_USE_ERP,
&cqr->flags)) {
erp_fn = base->discipline->erp_action(cqr);
erp_fn(cqr);
}
goto restart;
}
/* First of all call extended error reporting. */
if (dasd_eer_enabled(base) &&
cqr->status == DASD_CQR_FAILED) {
dasd_eer_write(base, cqr, DASD_EER_FATALERROR);
/* restart request */
cqr->status = DASD_CQR_FILLED;
cqr->retries = 255;
spin_lock_irqsave(get_ccwdev_lock(base->cdev), flags);
base->stopped |= DASD_STOPPED_QUIESCE;
spin_unlock_irqrestore(get_ccwdev_lock(base->cdev),
flags);
goto restart;
}
/* Process finished ERP request. */
if (cqr->refers) {
__dasd_block_process_erp(block, cqr);
goto restart;
}
/* Rechain finished requests to final queue */
cqr->endclk = get_clock();
list_move_tail(&cqr->blocklist, final_queue);
}
}
static void dasd_return_cqr_cb(struct dasd_ccw_req *cqr, void *data)
{
dasd_schedule_block_bh(cqr->block);
}
static void __dasd_block_start_head(struct dasd_block *block)
{
struct dasd_ccw_req *cqr;
if (list_empty(&block->ccw_queue))
return;
/* We allways begin with the first requests on the queue, as some
* of previously started requests have to be enqueued on a
* dasd_device again for error recovery.
*/
list_for_each_entry(cqr, &block->ccw_queue, blocklist) {
if (cqr->status != DASD_CQR_FILLED)
continue;
/* Non-temporary stop condition will trigger fail fast */
if (block->base->stopped & ~DASD_STOPPED_PENDING &&
test_bit(DASD_CQR_FLAGS_FAILFAST, &cqr->flags) &&
(!dasd_eer_enabled(block->base))) {
cqr->status = DASD_CQR_FAILED;
dasd_schedule_block_bh(block);
continue;
}
/* Don't try to start requests if device is stopped */
if (block->base->stopped)
return;
/* just a fail safe check, should not happen */
if (!cqr->startdev)
cqr->startdev = block->base;
/* make sure that the requests we submit find their way back */
cqr->callback = dasd_return_cqr_cb;
dasd_add_request_tail(cqr);
}
}
/*
* Central dasd_block layer routine. Takes requests from the generic
* block layer request queue, creates ccw requests, enqueues them on
* a dasd_device and processes ccw requests that have been returned.
*/
static void dasd_block_tasklet(struct dasd_block *block)
{
struct list_head final_queue;
struct list_head *l, *n;
struct dasd_ccw_req *cqr;
atomic_set(&block->tasklet_scheduled, 0);
INIT_LIST_HEAD(&final_queue);
spin_lock(&block->queue_lock);
/* Finish off requests on ccw queue */
__dasd_process_block_ccw_queue(block, &final_queue);
spin_unlock(&block->queue_lock);
/* Now call the callback function of requests with final status */
spin_lock_irq(&block->request_queue_lock);
list_for_each_safe(l, n, &final_queue) {
cqr = list_entry(l, struct dasd_ccw_req, blocklist);
list_del_init(&cqr->blocklist);
__dasd_cleanup_cqr(cqr);
}
spin_lock(&block->queue_lock);
/* Get new request from the block device request queue */
__dasd_process_request_queue(block);
/* Now check if the head of the ccw queue needs to be started. */
__dasd_block_start_head(block);
spin_unlock(&block->queue_lock);
spin_unlock_irq(&block->request_queue_lock);
dasd_put_device(block->base);
}
static void _dasd_wake_block_flush_cb(struct dasd_ccw_req *cqr, void *data)
{
wake_up(&dasd_flush_wq);
}
/*
* Go through all request on the dasd_block request queue, cancel them
* on the respective dasd_device, and return them to the generic
* block layer.
*/
static int dasd_flush_block_queue(struct dasd_block *block)
{
struct dasd_ccw_req *cqr, *n;
int rc, i;
struct list_head flush_queue;
INIT_LIST_HEAD(&flush_queue);
spin_lock_bh(&block->queue_lock);
rc = 0;
restart:
list_for_each_entry_safe(cqr, n, &block->ccw_queue, blocklist) {
/* if this request currently owned by a dasd_device cancel it */
if (cqr->status >= DASD_CQR_QUEUED)
rc = dasd_cancel_req(cqr);
if (rc < 0)
break;
/* Rechain request (including erp chain) so it won't be
* touched by the dasd_block_tasklet anymore.
* Replace the callback so we notice when the request
* is returned from the dasd_device layer.
*/
cqr->callback = _dasd_wake_block_flush_cb;
for (i = 0; cqr != NULL; cqr = cqr->refers, i++)
list_move_tail(&cqr->blocklist, &flush_queue);
if (i > 1)
/* moved more than one request - need to restart */
goto restart;
}
spin_unlock_bh(&block->queue_lock);
/* Now call the callback function of flushed requests */
restart_cb:
list_for_each_entry_safe(cqr, n, &flush_queue, blocklist) {
wait_event(dasd_flush_wq, (cqr->status < DASD_CQR_QUEUED));
/* Process finished ERP request. */
if (cqr->refers) {
__dasd_block_process_erp(block, cqr);
/* restart list_for_xx loop since dasd_process_erp
* might remove multiple elements */
goto restart_cb;
}
/* call the callback function */
cqr->endclk = get_clock();
list_del_init(&cqr->blocklist);
__dasd_cleanup_cqr(cqr);
}
return rc;
}
/*
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
* Schedules a call to dasd_tasklet over the device tasklet.
*/
void dasd_schedule_block_bh(struct dasd_block *block)
{
/* Protect against rescheduling. */
if (atomic_cmpxchg(&block->tasklet_scheduled, 0, 1) != 0)
return;
/* life cycle of block is bound to it's base device */
dasd_get_device(block->base);
tasklet_hi_schedule(&block->tasklet);
}
/*
* SECTION: external block device operations
* (request queue handling, open, release, etc.)
*/
/*
* Dasd request queue function. Called from ll_rw_blk.c
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void do_dasd_request(struct request_queue *queue)
{
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_block *block;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
block = queue->queuedata;
spin_lock(&block->queue_lock);
/* Get new request from the block device request queue */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
__dasd_process_request_queue(block);
/* Now check if the head of the ccw queue needs to be started. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
__dasd_block_start_head(block);
spin_unlock(&block->queue_lock);
}
/*
* Allocate and initialize request queue and default I/O scheduler.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_alloc_queue(struct dasd_block *block)
{
int rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
block->request_queue = blk_init_queue(do_dasd_request,
&block->request_queue_lock);
if (block->request_queue == NULL)
return -ENOMEM;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
block->request_queue->queuedata = block;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
elevator_exit(block->request_queue->elevator);
rc = elevator_init(block->request_queue, "deadline");
if (rc) {
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
blk_cleanup_queue(block->request_queue);
return rc;
}
return 0;
}
/*
* Allocate and initialize request queue.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_setup_queue(struct dasd_block *block)
{
int max;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
blk_queue_hardsect_size(block->request_queue, block->bp_block);
max = block->base->discipline->max_blocks << block->s2b_shift;
blk_queue_max_sectors(block->request_queue, max);
blk_queue_max_phys_segments(block->request_queue, -1L);
blk_queue_max_hw_segments(block->request_queue, -1L);
blk_queue_max_segment_size(block->request_queue, -1L);
blk_queue_segment_boundary(block->request_queue, -1L);
blk_queue_ordered(block->request_queue, QUEUE_ORDERED_DRAIN, NULL);
}
/*
* Deactivate and free request queue.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_free_queue(struct dasd_block *block)
{
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (block->request_queue) {
blk_cleanup_queue(block->request_queue);
block->request_queue = NULL;
}
}
/*
* Flush request on the request queue.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static void dasd_flush_request_queue(struct dasd_block *block)
{
struct request *req;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (!block->request_queue)
return;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
spin_lock_irq(&block->request_queue_lock);
while ((req = elv_next_request(block->request_queue))) {
blkdev_dequeue_request(req);
dasd_end_request(req, -EIO);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
spin_unlock_irq(&block->request_queue_lock);
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_open(struct inode *inp, struct file *filp)
{
struct gendisk *disk = inp->i_bdev->bd_disk;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_block *block = disk->private_data;
struct dasd_device *base = block->base;
int rc;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
atomic_inc(&block->open_count);
if (test_bit(DASD_FLAG_OFFLINE, &base->flags)) {
rc = -ENODEV;
goto unlock;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (!try_module_get(base->discipline->owner)) {
rc = -EINVAL;
goto unlock;
}
if (dasd_probeonly) {
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
DEV_MESSAGE(KERN_INFO, base, "%s",
"No access to device due to probeonly mode");
rc = -EPERM;
goto out;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (base->state <= DASD_STATE_BASIC) {
DBF_DEV_EVENT(DBF_ERR, base, " %s",
" Cannot open unrecognized device");
rc = -ENODEV;
goto out;
}
return 0;
out:
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
module_put(base->discipline->owner);
unlock:
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
atomic_dec(&block->open_count);
return rc;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_release(struct inode *inp, struct file *filp)
{
struct gendisk *disk = inp->i_bdev->bd_disk;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_block *block = disk->private_data;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
atomic_dec(&block->open_count);
module_put(block->base->discipline->owner);
return 0;
}
/*
* Return disk geometry.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int dasd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
{
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_block *block;
struct dasd_device *base;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
block = bdev->bd_disk->private_data;
base = block->base;
if (!block)
return -ENODEV;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (!base->discipline ||
!base->discipline->fill_geometry)
return -EINVAL;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
base->discipline->fill_geometry(block, geo);
geo->start = get_start_sect(bdev) >> block->s2b_shift;
return 0;
}
struct block_device_operations
dasd_device_operations = {
.owner = THIS_MODULE,
.open = dasd_open,
.release = dasd_release,
.ioctl = dasd_ioctl,
.compat_ioctl = dasd_compat_ioctl,
.getgeo = dasd_getgeo,
};
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/*******************************************************************************
* end of block device operations
*/
static void
dasd_exit(void)
{
#ifdef CONFIG_PROC_FS
dasd_proc_exit();
#endif
dasd_eer_exit();
if (dasd_page_cache != NULL) {
kmem_cache_destroy(dasd_page_cache);
dasd_page_cache = NULL;
}
dasd_gendisk_exit();
dasd_devmap_exit();
if (dasd_debug_area != NULL) {
debug_unregister(dasd_debug_area);
dasd_debug_area = NULL;
}
}
/*
* SECTION: common functions for ccw_driver use
*/
/*
* Initial attempt at a probe function. this can be simplified once
* the other detection code is gone.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_generic_probe(struct ccw_device *cdev,
struct dasd_discipline *discipline)
{
int ret;
ret = ccw_device_set_options(cdev, CCWDEV_DO_PATHGROUP);
if (ret) {
printk(KERN_WARNING
"dasd_generic_probe: could not set ccw-device options "
"for %s\n", cdev->dev.bus_id);
return ret;
}
ret = dasd_add_sysfs_files(cdev);
if (ret) {
printk(KERN_WARNING
"dasd_generic_probe: could not add sysfs entries "
"for %s\n", cdev->dev.bus_id);
return ret;
}
cdev->handler = &dasd_int_handler;
/*
* Automatically online either all dasd devices (dasd_autodetect)
* or all devices specified with dasd= parameters during
* initial probe.
*/
if ((dasd_get_feature(cdev, DASD_FEATURE_INITIAL_ONLINE) > 0 ) ||
(dasd_autodetect && dasd_busid_known(cdev->dev.bus_id) != 0))
ret = ccw_device_set_online(cdev);
if (ret)
printk(KERN_WARNING
"dasd_generic_probe: could not initially "
"online ccw-device %s; return code: %d\n",
cdev->dev.bus_id, ret);
return 0;
}
/*
* This will one day be called from a global not_oper handler.
* It is also used by driver_unregister during module unload.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
void dasd_generic_remove(struct ccw_device *cdev)
{
struct dasd_device *device;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_block *block;
cdev->handler = NULL;
dasd_remove_sysfs_files(cdev);
device = dasd_device_from_cdev(cdev);
if (IS_ERR(device))
return;
if (test_and_set_bit(DASD_FLAG_OFFLINE, &device->flags)) {
/* Already doing offline processing */
dasd_put_device(device);
return;
}
/*
* This device is removed unconditionally. Set offline
* flag to prevent dasd_open from opening it while it is
* no quite down yet.
*/
dasd_set_target_state(device, DASD_STATE_NEW);
/* dasd_delete_device destroys the device reference. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
block = device->block;
device->block = NULL;
dasd_delete_device(device);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/*
* life cycle of block is bound to device, so delete it after
* device was safely removed
*/
if (block)
dasd_free_block(block);
}
/*
* Activate a device. This is called from dasd_{eckd,fba}_probe() when either
* the device is detected for the first time and is supposed to be used
* or the user has started activation through sysfs.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_generic_set_online(struct ccw_device *cdev,
struct dasd_discipline *base_discipline)
{
struct dasd_discipline *discipline;
struct dasd_device *device;
int rc;
/* first online clears initial online feature flag */
dasd_set_feature(cdev, DASD_FEATURE_INITIAL_ONLINE, 0);
device = dasd_create_device(cdev);
if (IS_ERR(device))
return PTR_ERR(device);
discipline = base_discipline;
if (device->features & DASD_FEATURE_USEDIAG) {
if (!dasd_diag_discipline_pointer) {
printk (KERN_WARNING
"dasd_generic couldn't online device %s "
"- discipline DIAG not available\n",
cdev->dev.bus_id);
dasd_delete_device(device);
return -ENODEV;
}
discipline = dasd_diag_discipline_pointer;
}
if (!try_module_get(base_discipline->owner)) {
dasd_delete_device(device);
return -EINVAL;
}
if (!try_module_get(discipline->owner)) {
module_put(base_discipline->owner);
dasd_delete_device(device);
return -EINVAL;
}
device->base_discipline = base_discipline;
device->discipline = discipline;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/* check_device will allocate block device if necessary */
rc = discipline->check_device(device);
if (rc) {
printk (KERN_WARNING
"dasd_generic couldn't online device %s "
"with discipline %s rc=%i\n",
cdev->dev.bus_id, discipline->name, rc);
module_put(discipline->owner);
module_put(base_discipline->owner);
dasd_delete_device(device);
return rc;
}
dasd_set_target_state(device, DASD_STATE_ONLINE);
if (device->state <= DASD_STATE_KNOWN) {
printk (KERN_WARNING
"dasd_generic discipline not found for %s\n",
cdev->dev.bus_id);
rc = -ENODEV;
dasd_set_target_state(device, DASD_STATE_NEW);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (device->block)
dasd_free_block(device->block);
dasd_delete_device(device);
} else
pr_debug("dasd_generic device %s found\n",
cdev->dev.bus_id);
/* FIXME: we have to wait for the root device but we don't want
* to wait for each single device but for all at once. */
wait_event(dasd_init_waitq, _wait_for_device(device));
dasd_put_device(device);
return rc;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_generic_set_offline(struct ccw_device *cdev)
{
struct dasd_device *device;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
struct dasd_block *block;
int max_count, open_count;
device = dasd_device_from_cdev(cdev);
if (IS_ERR(device))
return PTR_ERR(device);
if (test_and_set_bit(DASD_FLAG_OFFLINE, &device->flags)) {
/* Already doing offline processing */
dasd_put_device(device);
return 0;
}
/*
* We must make sure that this device is currently not in use.
* The open_count is increased for every opener, that includes
* the blkdev_get in dasd_scan_partitions. We are only interested
* in the other openers.
*/
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
if (device->block) {
struct dasd_block *block = device->block;
max_count = block->bdev ? 0 : -1;
open_count = (int) atomic_read(&block->open_count);
if (open_count > max_count) {
if (open_count > 0)
printk(KERN_WARNING "Can't offline dasd "
"device with open count = %i.\n",
open_count);
else
printk(KERN_WARNING "%s",
"Can't offline dasd device due "
"to internal use\n");
clear_bit(DASD_FLAG_OFFLINE, &device->flags);
dasd_put_device(device);
return -EBUSY;
}
}
dasd_set_target_state(device, DASD_STATE_NEW);
/* dasd_delete_device destroys the device reference. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
block = device->block;
device->block = NULL;
dasd_delete_device(device);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
/*
* life cycle of block is bound to device, so delete it after
* device was safely removed
*/
if (block)
dasd_free_block(block);
return 0;
}
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
int dasd_generic_notify(struct ccw_device *cdev, int event)
{
struct dasd_device *device;
struct dasd_ccw_req *cqr;
unsigned long flags;
int ret;
device = dasd_device_from_cdev(cdev);
if (IS_ERR(device))
return 0;
spin_lock_irqsave(get_ccwdev_lock(cdev), flags);
ret = 0;
switch (event) {
case CIO_GONE:
case CIO_NO_PATH:
/* First of all call extended error reporting. */
dasd_eer_write(device, NULL, DASD_EER_NOPATH);
if (device->state < DASD_STATE_BASIC)
break;
/* Device is active. We want to keep it. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
list_for_each_entry(cqr, &device->ccw_queue, devlist)
if (cqr->status == DASD_CQR_IN_IO) {
cqr->status = DASD_CQR_QUEUED;
cqr->retries++;
}
device->stopped |= DASD_STOPPED_DC_WAIT;
dasd_device_clear_timer(device);
dasd_schedule_device_bh(device);
ret = 1;
break;
case CIO_OPER:
/* FIXME: add a sanity check. */
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
device->stopped &= ~DASD_STOPPED_DC_WAIT;
dasd_schedule_device_bh(device);
if (device->block)
dasd_schedule_block_bh(device->block);
ret = 1;
break;
}
spin_unlock_irqrestore(get_ccwdev_lock(cdev), flags);
dasd_put_device(device);
return ret;
}
static struct dasd_ccw_req *dasd_generic_build_rdc(struct dasd_device *device,
void *rdc_buffer,
int rdc_buffer_size,
char *magic)
{
struct dasd_ccw_req *cqr;
struct ccw1 *ccw;
cqr = dasd_smalloc_request(magic, 1 /* RDC */, rdc_buffer_size, device);
if (IS_ERR(cqr)) {
DEV_MESSAGE(KERN_WARNING, device, "%s",
"Could not allocate RDC request");
return cqr;
}
ccw = cqr->cpaddr;
ccw->cmd_code = CCW_CMD_RDC;
ccw->cda = (__u32)(addr_t)rdc_buffer;
ccw->count = rdc_buffer_size;
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
cqr->startdev = device;
cqr->memdev = device;
cqr->expires = 10*HZ;
clear_bit(DASD_CQR_FLAGS_USE_ERP, &cqr->flags);
cqr->retries = 2;
cqr->buildclk = get_clock();
cqr->status = DASD_CQR_FILLED;
return cqr;
}
int dasd_generic_read_dev_chars(struct dasd_device *device, char *magic,
void **rdc_buffer, int rdc_buffer_size)
{
int ret;
struct dasd_ccw_req *cqr;
cqr = dasd_generic_build_rdc(device, *rdc_buffer, rdc_buffer_size,
magic);
if (IS_ERR(cqr))
return PTR_ERR(cqr);
ret = dasd_sleep_on(cqr);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
dasd_sfree_request(cqr, cqr->memdev);
return ret;
}
EXPORT_SYMBOL_GPL(dasd_generic_read_dev_chars);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
static int __init dasd_init(void)
{
int rc;
init_waitqueue_head(&dasd_init_waitq);
init_waitqueue_head(&dasd_flush_wq);
/* register 'common' DASD debug area, used for all DBF_XXX calls */
dasd_debug_area = debug_register("dasd", 1, 1, 8 * sizeof(long));
if (dasd_debug_area == NULL) {
rc = -ENOMEM;
goto failed;
}
debug_register_view(dasd_debug_area, &debug_sprintf_view);
debug_set_level(dasd_debug_area, DBF_WARNING);
DBF_EVENT(DBF_EMERG, "%s", "debug area created");
dasd_diag_discipline_pointer = NULL;
rc = dasd_devmap_init();
if (rc)
goto failed;
rc = dasd_gendisk_init();
if (rc)
goto failed;
rc = dasd_parse();
if (rc)
goto failed;
rc = dasd_eer_init();
if (rc)
goto failed;
#ifdef CONFIG_PROC_FS
rc = dasd_proc_init();
if (rc)
goto failed;
#endif
return 0;
failed:
MESSAGE(KERN_INFO, "%s", "initialization not performed due to errors");
dasd_exit();
return rc;
}
module_init(dasd_init);
module_exit(dasd_exit);
EXPORT_SYMBOL(dasd_debug_area);
EXPORT_SYMBOL(dasd_diag_discipline_pointer);
EXPORT_SYMBOL(dasd_add_request_head);
EXPORT_SYMBOL(dasd_add_request_tail);
EXPORT_SYMBOL(dasd_cancel_req);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
EXPORT_SYMBOL(dasd_device_clear_timer);
EXPORT_SYMBOL(dasd_block_clear_timer);
EXPORT_SYMBOL(dasd_enable_device);
EXPORT_SYMBOL(dasd_int_handler);
EXPORT_SYMBOL(dasd_kfree_request);
EXPORT_SYMBOL(dasd_kick_device);
EXPORT_SYMBOL(dasd_kmalloc_request);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
EXPORT_SYMBOL(dasd_schedule_device_bh);
EXPORT_SYMBOL(dasd_schedule_block_bh);
EXPORT_SYMBOL(dasd_set_target_state);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
EXPORT_SYMBOL(dasd_device_set_timer);
EXPORT_SYMBOL(dasd_block_set_timer);
EXPORT_SYMBOL(dasd_sfree_request);
EXPORT_SYMBOL(dasd_sleep_on);
EXPORT_SYMBOL(dasd_sleep_on_immediatly);
EXPORT_SYMBOL(dasd_sleep_on_interruptible);
EXPORT_SYMBOL(dasd_smalloc_request);
EXPORT_SYMBOL(dasd_start_IO);
EXPORT_SYMBOL(dasd_term_IO);
EXPORT_SYMBOL_GPL(dasd_generic_probe);
EXPORT_SYMBOL_GPL(dasd_generic_remove);
EXPORT_SYMBOL_GPL(dasd_generic_notify);
EXPORT_SYMBOL_GPL(dasd_generic_set_online);
EXPORT_SYMBOL_GPL(dasd_generic_set_offline);
[S390] dasd: add hyper PAV support to DASD device driver, part 1 Parallel access volumes (PAV) is a storage server feature, that allows to start multiple channel programs on the same DASD in parallel. It defines alias devices which can be used as alternative paths to the same disk. With the old base PAV support we only needed rudimentary functionality in the DASD device driver. As the mapping between base and alias devices was static, we just had to export an identifier (uid) and could leave the combining of devices to external layers like a device mapper multipath. Now hyper PAV removes the requirement to dedicate alias devices to specific base devices. Instead each alias devices can be combined with multiple base device on a per request basis. This requires full support by the DASD device driver as now each channel program itself has to identify the target base device. The changes to the dasd device driver and the ECKD discipline are: - Separate subchannel device representation (dasd_device) from block device representation (dasd_block). Only base devices are block devices. - Gather information about base and alias devices and possible combinations. - For each request decide which dasd_device should be used (base or alias) and build specific channel program. - Support summary unit checks, which allow the storage server to upgrade / downgrade between base and hyper PAV at runtime (support is mandatory). Signed-off-by: Stefan Weinhuber <wein@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2008-01-26 14:11:23 +01:00
EXPORT_SYMBOL_GPL(dasd_generic_handle_state_change);
EXPORT_SYMBOL_GPL(dasd_flush_device_queue);
EXPORT_SYMBOL_GPL(dasd_alloc_block);
EXPORT_SYMBOL_GPL(dasd_free_block);