[PATCH RESEND 0/2] kbuild: dead code elimination: ftrace fixes

[PATCH RESEND 0/2] kbuild: dead code elimination: ftrace fixes

[Marcin Nowakowski]

Enabling dead code & data elimination currently breaks ftrace operation,
as the __mcount_loc section is removed (as it is not referenced directly
anywhere in the code).
Moreover, there are a lot of entries missing in the __mcount_loc section
as the recordmcount tool doesn't currently properly handle the section
names as created by the use of -ffunction-sections.

The following 2 patches fix that behaviour.
They are based on next-20161208.

Marcin Nowakowski (2):
  kbuild: keep __mcount_loc table through dead code elimination
  recordmcount: fix mcount recording with -ffunction-sections

 include/asm-generic/vmlinux.lds.h |  2 +-
 scripts/Makefile                  |  5 +++++
 scripts/recordmcount.c            | 26 +++++++++++++++++++++++++-
 3 files changed, 31 insertions(+), 2 deletions(-)

-- 
2.7.4

[PATCH 1/2] kbuild: keep __mcount_loc table through dead code elimination

[Marcin Nowakowski]

When CONFIG_LD_DEAD_CODE_DATA_ELIMINATION is enabled we must ensure
__mcount_loc is preserved, as otherwise there are no valid entries for
ftrace to work with.

Fixes: 4b89b7f7aad5 ('kbuild: keep data tables through dead code elimination')

Signed-off-by: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
Acked-by: Nicholas Piggin <npiggin@gmail.com>
---
 include/asm-generic/vmlinux.lds.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index 0968d13..f492aa4 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -89,7 +89,7 @@
 #ifdef CONFIG_FTRACE_MCOUNT_RECORD
 #define MCOUNT_REC()	. = ALIGN(8);				\
 			VMLINUX_SYMBOL(__start_mcount_loc) = .; \
-			*(__mcount_loc)				\
+			KEEP(*(__mcount_loc))			\
 			VMLINUX_SYMBOL(__stop_mcount_loc) = .;
 #else
 #define MCOUNT_REC()
-- 
2.7.4

[PATCH 2/2] recordmcount: fix mcount recording with -ffunction-sections

[Marcin Nowakowski]

When CONFIG_LD_DEAD_CODE_DATA_ELIMINATION is enabled, each function is
added to its own new section called .text.functionname.
As described in vmlinux.lds.h, "those enabling
LD_DEAD_CODE_DATA_ELIMINATION must ensure they don't have conflicting
section names, and must pull in .text.[0-9a-zA-Z_]*", add a similar
pattern matching to recordmcount script for section detection.

To simplify the basic C implementation, replace the regex with a string
match on the prefix and a single character match following the '.text.'
prefix.

Signed-off-by: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
Acked-by: Nicholas Piggin <npiggin@gmail.com>
---
 scripts/Makefile       |  5 +++++
 scripts/recordmcount.c | 26 +++++++++++++++++++++++++-
 2 files changed, 30 insertions(+), 1 deletion(-)

diff --git a/scripts/Makefile b/scripts/Makefile
index 1d80897..9b88250 100644
--- a/scripts/Makefile
+++ b/scripts/Makefile
@@ -23,6 +23,11 @@ hostprogs-$(CONFIG_SYSTEM_EXTRA_CERTIFICATE) += insert-sys-cert
 
 HOSTCFLAGS_sortextable.o = -I$(srctree)/tools/include
 HOSTCFLAGS_asn1_compiler.o = -I$(srctree)/include
+
+ifdef CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
+HOSTCFLAGS_recordmcount.o = -DMCOUNT_INCLUDE_FUNCTION_SECTIONS
+endif
+
 HOSTLOADLIBES_sign-file = -lcrypto
 HOSTLOADLIBES_extract-cert = -lcrypto
 
diff --git a/scripts/recordmcount.c b/scripts/recordmcount.c
index aeb3422..bb517ef 100644
--- a/scripts/recordmcount.c
+++ b/scripts/recordmcount.c
@@ -32,6 +32,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
+#include <ctype.h>
 
 /*
  * glibc synced up and added the metag number but didn't add the relocations.
@@ -408,9 +409,31 @@ static uint32_t (*w)(uint32_t);
 static uint32_t (*w2)(uint16_t);
 
 /* Names of the sections that could contain calls to mcount. */
+#ifdef MCOUNT_INCLUDE_FUNCTION_SECTIONS
+static int
+is_mcounted_function_section_name(char const *const txtname)
+{
+	const size_t len = strlen(".text.");
+
+	return	strncmp(".text.", txtname, len) == 0 &&
+		strlen(txtname) > len &&
+		(isalnum(txtname[len]) || txtname[len] == '_');
+}
+#else
+static int
+is_mcounted_function_section_name(char const *const txtname)
+{
+	return 0;
+}
+#endif
+
+
 static int
 is_mcounted_section_name(char const *const txtname)
 {
+	int ffunc_section =
+		is_mcounted_function_section_name(txtname);
+
 	return strcmp(".text",           txtname) == 0 ||
 		strcmp(".ref.text",      txtname) == 0 ||
 		strcmp(".sched.text",    txtname) == 0 ||
@@ -419,7 +442,8 @@ 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;
+		strcmp(".text.unlikely", txtname) == 0 ||
+		ffunc_section;
 }
 
 /* 32 bit and 64 bit are very similar */
-- 
2.7.4

Re: [PATCH RESEND 0/2] kbuild: dead code elimination: ftrace fixes

[Nicholas Piggin]

On Thu, 22 Dec 2016 09:51:45 +0100
Marcin Nowakowski <marcin.nowakowski@imgtec.com> wrote:

> Enabling dead code & data elimination currently breaks ftrace operation,
> as the __mcount_loc section is removed (as it is not referenced directly
> anywhere in the code).
> Moreover, there are a lot of entries missing in the __mcount_loc section
> as the recordmcount tool doesn't currently properly handle the section
> names as created by the use of -ffunction-sections.

Thanks for keeping on top of these. I didn't have any objections, so
if you would send them to Michal Marek for merge and cc linux-kbuild
that would be appreciated.

Thanks,
Nick

Re: [PATCH RESEND 0/2] kbuild: dead code elimination: ftrace fixes

[Marcin Nowakowski]

Hi Nick,

On 22.12.2016 10:08, Nicholas Piggin wrote:
> On Thu, 22 Dec 2016 09:51:45 +0100
> Marcin Nowakowski <marcin.nowakowski@imgtec.com> wrote:
>
>> Enabling dead code & data elimination currently breaks ftrace operation,
>> as the __mcount_loc section is removed (as it is not referenced directly
>> anywhere in the code).
>> Moreover, there are a lot of entries missing in the __mcount_loc section
>> as the recordmcount tool doesn't currently properly handle the section
>> names as created by the use of -ffunction-sections.
>
> Thanks for keeping on top of these. I didn't have any objections, so
> if you would send them to Michal Marek for merge and cc linux-kbuild
> that would be appreciated.

I had previously sent them to Michal and cc:kbuild, but Michal has not 
picked them up yet hence I've sent them to lkml now.
It's not clear to me if Michal is the person responsible for 
recordmcount changes, but I initially posted this to kbuild as this is 
where all of your dead code & data elimination patches were going through.

Michal - do you plan to pick up these changes or would they need to be 
acked/merged by somebody else?

thanks,
Marcin