FROMLIST: kbuild: fix dynamic ftrace with clang LTO

With CONFIG_LTO_CLANG enabled, LLVM IR won't be compiled into object
files until modpost_link. This change postpones calls to recordmcount
until after this step.

In order to exclude ftrace_process_locs from inspection, we add a new
code section .text..ftrace, which we tell recordmcount to ignore, and
a __norecordmcount attribute for moving functions to this section.

Bug: 62093296
Bug: 67506682
Bug: 133186739
Change-Id: Iba2c053968206acf533fadab1eb34a743b5088ee
(am from https://patchwork.kernel.org/patch/10060327/)
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>

arch/Kconfig

@@ -497,7 +497,7 @@ config LTO_NONE
 config LTO_CLANG
 	bool "Use clang Link Time Optimization (LTO) (EXPERIMENTAL)"
 	depends on ARCH_SUPPORTS_LTO_CLANG
-	depends on !FTRACE_MCOUNT_RECORD
+	depends on !FTRACE_MCOUNT_RECORD || HAVE_C_RECORDMCOUNT
 	depends on CC_IS_CLANG && LD_IS_LLD
 	select LTO
 	help

include/asm-generic/vmlinux.lds.h

@@ -492,6 +492,7 @@
 		ALIGN_FUNCTION();					\
 		*(.text.hot TEXT_MAIN .text.fixup .text.unlikely)	\
 		*(.text..refcount)					\
+		*(.text..ftrace)					\
 		*(.ref.text)						\
 	MEM_KEEP(init.text*)						\
 	MEM_KEEP(exit.text*)						\

include/linux/compiler-clang.h

@@ -43,3 +43,10 @@
 #define __must_be_array(a) BUILD_BUG_ON_ZERO(__same_type((a), &(a)[0]))
 #define __assume_aligned(a, ...)	\
 	__attribute__((__assume_aligned__(a, ## __VA_ARGS__)))
+
+#ifdef CONFIG_LTO_CLANG
+#ifdef CONFIG_FTRACE_MCOUNT_RECORD
+#define __norecordmcount \
+	__attribute__((__section__(".text..ftrace")))
+#endif
+#endif

include/linux/compiler_types.h

@@ -248,6 +248,10 @@ struct ftrace_likely_data {
 # define __gnu_inline
 #endif
 
+#ifndef __norecordmcount
+#define __norecordmcount
+#endif
+
 /*
  * Force always-inline if the user requests it so via the .config.
  * GCC does not warn about unused static inline functions for

kernel/trace/ftrace.c

@@ -5500,9 +5500,9 @@ static int ftrace_cmp_ips(const void *a, const void *b)
 	return 0;
 }
 
-static int ftrace_process_locs(struct module *mod,
-			       unsigned long *start,
-			       unsigned long *end)
+static int __norecordmcount ftrace_process_locs(struct module *mod,
+						unsigned long *start,
+						unsigned long *end)
 {
 	struct ftrace_page *start_pg;
 	struct ftrace_page *pg;

scripts/Makefile.build

@@ -223,6 +223,12 @@ ifdef BUILD_C_RECORDMCOUNT
 ifeq ("$(origin RECORDMCOUNT_WARN)", "command line")
   RECORDMCOUNT_FLAGS = -w
 endif
+
+ifdef CONFIG_LTO_CLANG
+# With LTO, we postpone running recordmcount until after the LTO link step, so
+# let's export the parameters for the link script.
+export RECORDMCOUNT_FLAGS
+else
 # Due to recursion, we must skip empty.o.
 # The empty.o file is created in the make process in order to determine
 # the target endianness and word size. It is made before all other C
@@ -231,22 +237,28 @@ sub_cmd_record_mcount =					\
 	if [ $(@) != "scripts/mod/empty.o" ]; then	\
 		$(objtree)/scripts/recordmcount $(RECORDMCOUNT_FLAGS) "$(@)";	\
 	fi;
+endif
+
 recordmcount_source := $(srctree)/scripts/recordmcount.c \
 		    $(srctree)/scripts/recordmcount.h
-else
+else # !BUILD_C_RECORDMCOUNT
 sub_cmd_record_mcount = set -e ; perl $(srctree)/scripts/recordmcount.pl "$(ARCH)" \
 	"$(if $(CONFIG_CPU_BIG_ENDIAN),big,little)" \
 	"$(if $(CONFIG_64BIT),64,32)" \
 	"$(OBJDUMP)" "$(OBJCOPY)" "$(CC) $(KBUILD_CPPFLAGS) $(KBUILD_CFLAGS)" \
 	"$(LD) $(KBUILD_LDFLAGS)" "$(NM)" "$(RM)" "$(MV)" \
 	"$(if $(part-of-module),1,0)" "$(@)";
+
 recordmcount_source := $(srctree)/scripts/recordmcount.pl
 endif # BUILD_C_RECORDMCOUNT
+
+ifndef CONFIG_LTO_CLANG
 cmd_record_mcount =						\
 	if [ "$(findstring $(CC_FLAGS_FTRACE),$(_c_flags))" =	\
 	     "$(CC_FLAGS_FTRACE)" ]; then			\
 		$(sub_cmd_record_mcount)			\
 	fi;
+endif # CONFIG_LTO_CLANG
 endif # CC_USING_RECORD_MCOUNT
 endif # CONFIG_FTRACE_MCOUNT_RECORD
 

scripts/Makefile.modpost

@@ -143,6 +143,10 @@ ifdef CONFIG_LTO_CLANG
 			echo -T $(@:.ko=.o.symversions))  		\
 		 -o $@ --whole-archive					\
 		 $(filter-out FORCE,$(^:$(modpost-ext).o=.o))
+
+  ifdef CONFIG_FTRACE_MCOUNT_RECORD
+      cmd_ld_ko_o += ; $(objtree)/scripts/recordmcount $(RECORDMCOUNT_FLAGS) $@
+  endif
 else
       cmd_ld_ko_o =                                                     \
 	$(LD) -r $(KBUILD_LDFLAGS)                                      \

scripts/link-vmlinux.sh

@@ -116,6 +116,19 @@ modpost_link()
 	${LD} ${KBUILD_LDFLAGS} -r -o ${1} $(modversions) ${objects}
 }
 
+# If CONFIG_LTO_CLANG is selected, we postpone running recordmcount until
+# we have compiled LLVM IR to an object file.
+recordmcount()
+{
+	if [ -z "${CONFIG_LTO_CLANG}" ]; then
+		return
+	fi
+
+	if [ -n "${CONFIG_FTRACE_MCOUNT_RECORD}" ]; then
+		scripts/recordmcount ${RECORDMCOUNT_FLAGS} $*
+	fi
+}
+
 # Link of vmlinux
 # ${1} - optional extra .o files
 # ${2} - output file
@@ -270,6 +283,11 @@ modpost_link vmlinux.o
 # modpost vmlinux.o to check for section mismatches
 ${MAKE} -f "${srctree}/scripts/Makefile.modpost" vmlinux.o
 
+if [ -n "${CONFIG_LTO_CLANG}" ]; then
+	# Call recordmcount if needed
+	recordmcount vmlinux.o
+fi
+
 kallsymso=""
 kallsyms_vmlinux=""
 if [ -n "${CONFIG_KALLSYMS}" ]; then

scripts/recordmcount.c

@@ -406,7 +406,8 @@ is_mcounted_section_name(char const *const txtname)
 		strcmp(".softirqentry.text", txtname) == 0 ||
 		strcmp(".kprobes.text", txtname) == 0 ||
 		strcmp(".cpuidle.text", txtname) == 0 ||
-		strcmp(".text.unlikely", txtname) == 0;
+		(strncmp(".text.",       txtname, 6) == 0 &&
+		 strcmp(".text..ftrace", txtname) != 0);
 }
 
 /* 32 bit and 64 bit are very similar */