ANDROID: 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: 145210207
Change-Id: Ib77f7c431fce54243c46d584b55761ed2342965c
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>

arch/Kconfig

@@ -590,7 +590,7 @@ config LTO_CLANG
 	bool "Use Clang's Link Time Optimization (LTO) (EXPERIMENTAL)"
 	depends on ARCH_SUPPORTS_LTO_CLANG
 	depends on !KASAN
-	depends on !FTRACE_MCOUNT_RECORD
+	depends on !FTRACE_MCOUNT_RECORD || HAVE_C_RECORDMCOUNT
 	depends on CC_IS_CLANG && CLANG_VERSION >= 100000 && LD_IS_LLD
 	select LTO
 	help

include/asm-generic/vmlinux.lds.h

@@ -521,6 +521,7 @@
 		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

@@ -48,3 +48,10 @@
 #else
 # define __noscs
 #endif
+
+#ifdef CONFIG_LTO_CLANG
+#ifdef CONFIG_FTRACE_MCOUNT_RECORD
+#define __norecordmcount \
+	__attribute__((__section__(".text..ftrace")))
+#endif
+#endif

include/linux/compiler_types.h

@@ -206,6 +206,10 @@ struct ftrace_likely_data {
 # define __noscs
 #endif
 
+#ifndef __norecordmcount
+# define __norecordmcount
+#endif
+
 #ifndef asm_volatile_goto
 #define asm_volatile_goto(x...) asm goto(x)
 #endif

kernel/trace/ftrace.c

@@ -5564,9 +5564,9 @@ 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

@@ -177,6 +177,12 @@ 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
@@ -185,6 +191,8 @@ 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
@@ -194,10 +202,13 @@ sub_cmd_record_mcount = perl $(srctree)/scripts/recordmcount.pl "$(ARCH)" \
 	"$(OBJDUMP)" "$(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 $(strip $(CC_FLAGS_FTRACE)),$(_c_flags)),	\
 	$(sub_cmd_record_mcount))
+endif # CONFIG_LTO_CLANG
 endif # CC_USING_RECORD_MCOUNT
 endif # CONFIG_FTRACE_MCOUNT_RECORD
 

scripts/Makefile.modfinal

@@ -41,6 +41,10 @@ ifdef CONFIG_LTO_CLANG
 			echo -T $(@:.ko=.o.symversions))  		\
 		 -o $@ --whole-archive $(filter %.o, $^);		\
 	$(if $(ARCH_POSTLINK), $(MAKE) -f $(ARCH_POSTLINK) $@, true)
+
+  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

@@ -87,6 +87,19 @@ 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} - output file
 # ${2}, ${3}, ... - optional extra .o files
@@ -287,6 +300,11 @@ modpost_link vmlinux.o
 # modpost vmlinux.o to check for section mismatches
 ${MAKE} -f "${srctree}/scripts/Makefile.modpost" MODPOST_VMLINUX=1
 
+if [ -n "${CONFIG_LTO_CLANG}" ]; then
+	# Call recordmcount if needed
+	recordmcount vmlinux.o
+fi
+
 info MODINFO modules.builtin.modinfo
 ${OBJCOPY} -j .modinfo -O binary vmlinux.o modules.builtin.modinfo
 

scripts/recordmcount.c

@@ -408,7 +408,9 @@ static int is_mcounted_section_name(char const *const txtname)
 		strcmp(".irqentry.text", txtname) == 0 ||
 		strcmp(".softirqentry.text", txtname) == 0 ||
 		strcmp(".kprobes.text", txtname) == 0 ||
-		strcmp(".cpuidle.text", txtname) == 0;
+		strcmp(".cpuidle.text", txtname) == 0 ||
+		(strncmp(".text.",       txtname, 6) == 0 &&
+		 strcmp(".text..ftrace", txtname) != 0);
 }
 
 static char const *already_has_rel_mcount = "success"; /* our work here is done! */