[PATCH V2 2/3] perf symbols: improve abi compliance in arm mapping symbols handling

[Victor Kamensky]

On 22 January 2015 at 07:29, Russell King - ARM Linux
<linux at arm.linux.org.uk> wrote:
> On Wed, Jan 21, 2015 at 08:52:57PM -0800, Victor Kamensky wrote:
>> Both Arm and Aarch64 ELF ABI allow mapping symbols be in from
>> either "$d" or "$d.<any>". But current code that handles mapping
>> symbols only deals with the first, dollar character and a single
>> letter, case.
>>
>> The patch adds handling of the second case with period
>> followed by any characters.
>>
>> Suggested-by: Russell King <linux at arm.linux.org.uk>
>> Suggested-by: Will Deacon <will.deacon at arm.com>
>> Signed-off-by: Victor Kamensky <victor.kamensky at linaro.org>
>> Acked-by: Will Deacon <will.deacon at arm.com>
>> Cc: Peter Zijlstra <a.p.zijlstra at chello.nl>
>> Cc: Paul Mackerras <paulus at samba.org>
>> Cc: Ingo Molnar <mingo at redhat.com>
>> Cc: Arnaldo Carvalho de Melo <acme at kernel.org>
>> Cc: Adrian Hunter <adrian.hunter at intel.com>
>> Cc: Jiri Olsa <jolsa at redhat.com>
>> Cc: Namhyung Kim <namhyung at kernel.org>
>> Cc: Avi Kivity <avi at cloudius-systems.com>
>> Cc: Masami Hiramatsu <masami.hiramatsu.pt at hitachi.com>
>> Cc: Anton Blanchard <anton at samba.org>
>> Cc: David Ahern <dsahern at gmail.com>
>> Cc: Dave Martin <Dave.Martin at arm.com>
>> ---
>>  tools/perf/util/symbol-elf.c | 9 ++++-----
>>  1 file changed, 4 insertions(+), 5 deletions(-)
>>
>> diff --git a/tools/perf/util/symbol-elf.c b/tools/perf/util/symbol-elf.c
>> index 1e188dd..7fd35d6 100644
>> --- a/tools/perf/util/symbol-elf.c
>> +++ b/tools/perf/util/symbol-elf.c
>> @@ -857,17 +857,16 @@ int dso__load_sym(struct dso *dso, struct map *map,
>>                * don't identify functions, so will confuse the profile
>>                * output: */
>>               if (ehdr.e_machine == EM_ARM) {
>> -                     if (!strcmp(elf_name, "$a") ||
>> -                         !strcmp(elf_name, "$d") ||
>> -                         !strcmp(elf_name, "$t"))
>> +                     if (elf_name[0] == '$' && strchr("adt", elf_name[1])
>> +                         && (elf_name[2] == '\0' || elf_name[2] == '.'))
>>                               continue;
>>               }
>>               /* Reject Aarch64 ELF "mapping symbols": these aren't unique and
>>                * don't identify functions, so will confuse the profile
>>                * output: */
>>               if (ehdr.e_machine == EM_AARCH64) {
>> -                     if (!strcmp(elf_name, "$x") ||
>> -                         !strcmp(elf_name, "$d"))
>> +                     if (elf_name[0] == '$' && strchr("dx", elf_name[1])
>> +                         && (elf_name[2] == '\0' || elf_name[2] == '.'))
>>                               continue;
>>               }
>
> Why not fix this first?  Then, all you'd need to do would be to change
> the test for EM_ARM to EM_ARM or EM_AARCH64, and add the 'x' into the
> strchr() string.

In previously discussed [1] I proposed two variants and ask
for people's preferences. Will replied that he did not care and
let me choose that I like.

It seems to me that you more like variant 2 (provided here
for reference again). It would replace patch 1 and patch 2 of
version 2. Please confirm. I don't practically care much and
can use either. I personally preferred variant 1 because in
variant 2 $x will be filtered on ARM V7, but V7 ABI does not
say anything about $x, only Aarch64 does. I.e special symbols
do to match between V7 and V8 exactly. But as it is
already in other places like kallsyms, I agree practically it
may not matter. If you give your node to variant 2 (as below)
I can report this miniseries: Please let me know

Variant 2:

>From c8d08ebddc61203daf21b17c891c26c1d08e14f1 Mon Sep 17 00:00:00 2001
From: Victor Kamensky <victor.kamensky at linaro.org>
Date: Mon, 12 Jan 2015 14:13:36 -0800
Subject: [PATCH 1/2] perf symbols: Ignore mapping symbols on aarch64

Aarch64 ELF files use mapping symbols with special names $x, $d
to identify regions of Aarch64 code (see Aarch64 ELF ABI - "ARM
IHI 0056B", section "4.5.4 Mapping symbols").

The patch filters out these symbols at load time, similar to
"696b97a perf symbols: Ignore mapping symbols on ARM" changes
done for ARM before V8.

Also added handling of mapping symbols that has format
"$d.<any>" and similar for both cases.

Note we are not making difference between EM_ARM and
EM_AARCH64 mapping symbols instead code handles superset
of both.

Signed-off-by: Victor Kamensky <victor.kamensky at linaro.org>
Cc: Peter Zijlstra <a.p.zijlstra at chello.nl>
Cc: Paul Mackerras <paulus at samba.org>
Cc: Ingo Molnar <mingo at redhat.com>
Cc: Arnaldo Carvalho de Melo <acme at kernel.org>
Cc: Adrian Hunter <adrian.hunter at intel.com>
Cc: Jiri Olsa <jolsa at redhat.com>
Cc: Namhyung Kim <namhyung at kernel.org>
Cc: Avi Kivity <avi at cloudius-systems.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt at hitachi.com>
Cc: Anton Blanchard <anton at samba.org>
Cc: David Ahern <dsahern at gmail.com>
Cc: Will Deacon <will.deacon at arm.com>
Cc: Russell King <linux at arm.linux.org.uk>
Cc: Dave Martin <Dave.Martin at arm.com>
---
 tools/perf/util/symbol-elf.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/tools/perf/util/symbol-elf.c b/tools/perf/util/symbol-elf.c
index 06fcd1b..b2eb0f9 100644
--- a/tools/perf/util/symbol-elf.c
+++ b/tools/perf/util/symbol-elf.c
@@ -856,10 +856,9 @@ int dso__load_sym(struct dso *dso, struct map *map,
         /* Reject ARM ELF "mapping symbols": these aren't unique and
          * don't identify functions, so will confuse the profile
          * output: */
-        if (ehdr.e_machine == EM_ARM) {
-            if (!strcmp(elf_name, "$a") ||
-                !strcmp(elf_name, "$d") ||
-                !strcmp(elf_name, "$t"))
+        if (ehdr.e_machine == EM_ARM || ehdr.e_machine == EM_AARCH64) {
+                        if (elf_name[0] == '$' && strchr("adtx", elf_name[1])
+                        && (elf_name[2] == '\0' || elf_name[2] == '.'))
                 continue;
         }

-- 
1.9.3

Thanks,
Victor

[1] https://lkml.org/lkml/2015/1/14/1019

> --
> FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
> according to speedtest.net.