linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [bug] wrong result of android callchain
@ 2016-01-05  9:14 He Kuang
  2016-01-05 10:20 ` Peter Zijlstra
  0 siblings, 1 reply; 6+ messages in thread
From: He Kuang @ 2016-01-05  9:14 UTC (permalink / raw)
  To: peterz, mingo, Arnaldo Carvalho de Melo, Jiri Olsa,
	rabin.vincent, namhyung, Wang Nan
  Cc: linux-kernel

I found a wrong result of aarch64 callchain when using perf script on
a android phone.

Here's the callchain record fragment from the output of perf script:

   init   369 [002]   339.970607: raw_syscalls:sys_enter: NR 22 (b, 7fd9e360a0, 10, ffffffff, 0, 8)
                      ...
                    230ac [unknown] (/system/lib64/libsurfaceflinger.so)
                     11a0 main (/system/bin/surfaceflinger)
                    1c3fc __libc_init (/system/lib64/libc.so)
                      fd0 _start (/system/bin/surfaceflinger)
                     29ec __dl__start (/system/bin/linker64)

The fault occured in the '[unknown]' line, from objdump result of
/system/bin/surfaceflinger, we can see the branch instruction before
0x11a0:

  # objdump /system/bin/surfaceflinger
     1198:       f9400fe0        ldr     x0, [sp,#24]
     119c:       97ffff05        bl      db0 <_ZN7android14SurfaceFlinger3runEv@plt>
     11a0:       f9400be8        ldr     x8, [sp,#16]
     11a4:       b40000c8        cbz     x8, 11bc <main+0x150>

The function '_ZN7android14SurfaceFlinger3runEv' is located at 0x3a094
~ 0x3a0ac in libsurfaceflinger.so, but perf misparsed that value to
0x230ac:

  # objdump libsurfaceflinger.so
   000000000003a094 <_ZN7android14SurfaceFlinger3runEv>:
     3a094:       a9be4ff4        stp     x20, x19, [sp,#-32]!
     3a098:       a9017bfd        stp     x29, x30, [sp,#16]
     3a09c:       910043fd        add     x29, sp, #0x10
     3a0a0:       910c0013        add     x19, x0, #0x300
     3a0a4:       aa1303e0        mov     x0, x19
     3a0a8:       97fff12f        bl      36564 <_ZN7android12MessageQueue11waitMessageEv>
     3a0ac:       17fffffe        b       3a0a4 <_ZN7android14SurfaceFlinger3runEv+0x10>

There's a difference of 0x17000 between those two offsets, it seems
that this value is the VirtAddr of this dynamic library.

  # readelf -a libsurfaceflinger.so
   Program Headers:
     Type           Offset             VirtAddr           PhysAddr
                    FileSiz            MemSiz              Flags  Align
     LOAD           0x0000000000000000 0x0000000000017000 0x0000000000017000
                    0x0000000000057258 0x0000000000057258  R E    1000


                                                    


^ permalink raw reply	[flat|nested] 6+ messages in thread
* Re: [bug] wrong result of android callchain
  2016-01-05  9:14 [bug] wrong result of android callchain He Kuang
@ 2016-01-05 10:20 ` Peter Zijlstra
  2016-01-05 11:13   ` Will Deacon
  0 siblings, 1 reply; 6+ messages in thread
From: Peter Zijlstra @ 2016-01-05 10:20 UTC (permalink / raw)
  To: He Kuang
  Cc: mingo, Arnaldo Carvalho de Melo, Jiri Olsa, rabin.vincent,
	namhyung, Wang Nan, linux-kernel, mark.rutland, will.deacon

On Tue, Jan 05, 2016 at 05:14:37PM +0800, He Kuang wrote:
> I found a wrong result of aarch64 callchain when using perf script on
> a android phone.

Might help to include the AARGH64 people then.. seeing I have no clue
about all that. Cc's added, email preserved etc..

> 
> Here's the callchain record fragment from the output of perf script:
> 
>   init   369 [002]   339.970607: raw_syscalls:sys_enter: NR 22 (b, 7fd9e360a0, 10, ffffffff, 0, 8)
>                      ...
>                    230ac [unknown] (/system/lib64/libsurfaceflinger.so)
>                     11a0 main (/system/bin/surfaceflinger)
>                    1c3fc __libc_init (/system/lib64/libc.so)
>                      fd0 _start (/system/bin/surfaceflinger)
>                     29ec __dl__start (/system/bin/linker64)
> 
> The fault occured in the '[unknown]' line, from objdump result of
> /system/bin/surfaceflinger, we can see the branch instruction before
> 0x11a0:
> 
>  # objdump /system/bin/surfaceflinger
>     1198:       f9400fe0        ldr     x0, [sp,#24]
>     119c:       97ffff05        bl      db0 <_ZN7android14SurfaceFlinger3runEv@plt>
>     11a0:       f9400be8        ldr     x8, [sp,#16]
>     11a4:       b40000c8        cbz     x8, 11bc <main+0x150>
> 
> The function '_ZN7android14SurfaceFlinger3runEv' is located at 0x3a094
> ~ 0x3a0ac in libsurfaceflinger.so, but perf misparsed that value to
> 0x230ac:
> 
>  # objdump libsurfaceflinger.so
>   000000000003a094 <_ZN7android14SurfaceFlinger3runEv>:
>     3a094:       a9be4ff4        stp     x20, x19, [sp,#-32]!
>     3a098:       a9017bfd        stp     x29, x30, [sp,#16]
>     3a09c:       910043fd        add     x29, sp, #0x10
>     3a0a0:       910c0013        add     x19, x0, #0x300
>     3a0a4:       aa1303e0        mov     x0, x19
>     3a0a8:       97fff12f        bl      36564 <_ZN7android12MessageQueue11waitMessageEv>
>     3a0ac:       17fffffe        b       3a0a4 <_ZN7android14SurfaceFlinger3runEv+0x10>
> 
> There's a difference of 0x17000 between those two offsets, it seems
> that this value is the VirtAddr of this dynamic library.
> 
>  # readelf -a libsurfaceflinger.so
>   Program Headers:
>     Type           Offset             VirtAddr           PhysAddr
>                    FileSiz            MemSiz              Flags  Align
>     LOAD           0x0000000000000000 0x0000000000017000 0x0000000000017000
>                    0x0000000000057258 0x0000000000057258  R E    1000
> 
> 
> 

^ permalink raw reply	[flat|nested] 6+ messages in thread
* Re: [bug] wrong result of android callchain
  2016-01-05 10:20 ` Peter Zijlstra
@ 2016-01-05 11:13   ` Will Deacon
  2016-01-06  8:26     ` [PATCH] perf symbols: Fix wrong symbol offset according to p_addr He Kuang
  0 siblings, 1 reply; 6+ messages in thread
From: Will Deacon @ 2016-01-05 11:13 UTC (permalink / raw)
  To: Peter Zijlstra
  Cc: He Kuang, mingo, Arnaldo Carvalho de Melo, Jiri Olsa,
	rabin.vincent, namhyung, Wang Nan, linux-kernel, mark.rutland

On Tue, Jan 05, 2016 at 11:20:55AM +0100, Peter Zijlstra wrote:
> On Tue, Jan 05, 2016 at 05:14:37PM +0800, He Kuang wrote:
> > I found a wrong result of aarch64 callchain when using perf script on
> > a android phone.
> 
> Might help to include the AARGH64 people then.. seeing I have no clue
> about all that. Cc's added, email preserved etc..

It looks like the issue is in the perf tool, where it fails to resolve
the address to a symbol correctly.

> > 
> > Here's the callchain record fragment from the output of perf script:
> > 
> >   init   369 [002]   339.970607: raw_syscalls:sys_enter: NR 22 (b, 7fd9e360a0, 10, ffffffff, 0, 8)
> >                      ...
> >                    230ac [unknown] (/system/lib64/libsurfaceflinger.so)
> >                     11a0 main (/system/bin/surfaceflinger)
> >                    1c3fc __libc_init (/system/lib64/libc.so)
> >                      fd0 _start (/system/bin/surfaceflinger)
> >                     29ec __dl__start (/system/bin/linker64)
> > 
> > The fault occured in the '[unknown]' line, from objdump result of
> > /system/bin/surfaceflinger, we can see the branch instruction before
> > 0x11a0:
> > 
> >  # objdump /system/bin/surfaceflinger
> >     1198:       f9400fe0        ldr     x0, [sp,#24]
> >     119c:       97ffff05        bl      db0 <_ZN7android14SurfaceFlinger3runEv@plt>
> >     11a0:       f9400be8        ldr     x8, [sp,#16]
> >     11a4:       b40000c8        cbz     x8, 11bc <main+0x150>
> > 
> > The function '_ZN7android14SurfaceFlinger3runEv' is located at 0x3a094
> > ~ 0x3a0ac in libsurfaceflinger.so, but perf misparsed that value to
> > 0x230ac:
> > 
> >  # objdump libsurfaceflinger.so
> >   000000000003a094 <_ZN7android14SurfaceFlinger3runEv>:
> >     3a094:       a9be4ff4        stp     x20, x19, [sp,#-32]!
> >     3a098:       a9017bfd        stp     x29, x30, [sp,#16]
> >     3a09c:       910043fd        add     x29, sp, #0x10
> >     3a0a0:       910c0013        add     x19, x0, #0x300
> >     3a0a4:       aa1303e0        mov     x0, x19
> >     3a0a8:       97fff12f        bl      36564 <_ZN7android12MessageQueue11waitMessageEv>
> >     3a0ac:       17fffffe        b       3a0a4 <_ZN7android14SurfaceFlinger3runEv+0x10>
> > 
> > There's a difference of 0x17000 between those two offsets, it seems
> > that this value is the VirtAddr of this dynamic library.
> > 
> >  # readelf -a libsurfaceflinger.so
> >   Program Headers:
> >     Type           Offset             VirtAddr           PhysAddr
> >                    FileSiz            MemSiz              Flags  Align
> >     LOAD           0x0000000000000000 0x0000000000017000 0x0000000000017000
> >                    0x0000000000057258 0x0000000000057258  R E    1000

This reminds me of:

  http://lkml.kernel.org/r/1435162623-20075-1-git-send-email-will.deacon@arm.com

but that was all vdso-specific. Can you look at what gets passed to objdump?

Will

^ permalink raw reply	[flat|nested] 6+ messages in thread
* [PATCH] perf symbols: Fix wrong symbol offset according to p_addr
  2016-01-05 11:13   ` Will Deacon
@ 2016-01-06  8:26     ` He Kuang
  2016-01-11 11:25       ` Ingo Molnar
  0 siblings, 1 reply; 6+ messages in thread
From: He Kuang @ 2016-01-06  8:26 UTC (permalink / raw)
  To: peterz, mingo, acme, jolsa, rabin.vincent, namhyung, wangnan0,
	mark.rutland, hekuang
  Cc: linux-kernel

Perf missed the 'VirtAddr' value in elf program_headers when adjusting
symbol address for dynamic libraries. This patch considers the p_addr
value and gets the right symbol offset.

Before this patch, some symbols can not be right parsed on android phone:

  init   369 [002]   339.970607: raw_syscalls:sys_enter: NR 22 (b, 7fd9e360a0, 10, ffffffff, 0, 8)
                     ...
                   230ac [unknown] (/system/lib64/libsurfaceflinger.so)
                    11a0 main (/system/bin/surfaceflinger)
                   1c3fc __libc_init (/system/lib64/libc.so)
                     fd0 _start (/system/bin/surfaceflinger)
                    29ec __dl__start (/system/bin/linker64)

After this patch:

  init   369 [002]   339.970607: raw_syscalls:sys_enter: NR 22 (b, 7fd9e360a0, 10, ffffffff, 0, 8)
                     ...
                   3a0ac _ZN7android14SurfaceFlinger3runEv (/system/lib64/libsurfaceflinger.so)
                    11a0 main (/system/bin/surfaceflinger)
                   1c3fc __libc_init (/system/lib64/libc.so)
                     fd0 _start (/system/bin/surfaceflinger)
                    29ec __dl__start (/system/bin/linker64)

Signed-off-by: He Kuang <hekuang@huawei.com>
---
 tools/perf/util/dso.c        |  6 ++++++
 tools/perf/util/dso.h        |  1 +
 tools/perf/util/event.c      |  6 +++++-
 tools/perf/util/map.h        |  7 +++++--
 tools/perf/util/symbol-elf.c | 20 ++++++++++++++++++++
 5 files changed, 37 insertions(+), 3 deletions(-)

diff --git a/tools/perf/util/dso.c b/tools/perf/util/dso.c
index e8e9a9d..93bf13a 100644
--- a/tools/perf/util/dso.c
+++ b/tools/perf/util/dso.c
@@ -1063,6 +1063,7 @@ struct dso *dso__new(const char *name)
 		dso->needs_swap = DSO_SWAP__UNSET;
 		RB_CLEAR_NODE(&dso->rb_node);
 		dso->root = NULL;
+		dso->vaddr = 0;
 		INIT_LIST_HEAD(&dso->node);
 		INIT_LIST_HEAD(&dso->data.open_entry);
 		pthread_mutex_init(&dso->lock, NULL);
@@ -1367,3 +1368,8 @@ int dso__strerror_load(struct dso *dso, char *buf, size_t buflen)
 	scnprintf(buf, buflen, "%s", dso_load__error_str[idx]);
 	return 0;
 }
+
+u64 dso__get_vaddr(struct dso *dso)
+{
+	return dso->vaddr;
+}
diff --git a/tools/perf/util/dso.h b/tools/perf/util/dso.h
index 45ec4d0..b6c894f 100644
--- a/tools/perf/util/dso.h
+++ b/tools/perf/util/dso.h
@@ -166,6 +166,7 @@ struct dso {
 	const char	 *long_name;
 	u16		 long_name_len;
 	u16		 short_name_len;
+	u64		vaddr;
 	void		*dwfl;			/* DWARF debug info */
 	struct auxtrace_cache *auxtrace_cache;
 
diff --git a/tools/perf/util/event.c b/tools/perf/util/event.c
index cd61bb1..ef964c0 100644
--- a/tools/perf/util/event.c
+++ b/tools/perf/util/event.c
@@ -1273,8 +1273,12 @@ try_again:
 		 * Kernel maps might be changed when loading symbols so loading
 		 * must be done prior to using kernel maps.
 		 */
+		map__load(al->map, machine->symbol_filter);
+
+		/* exclude kernel kallsyms */
 		if (load_map)
-			map__load(al->map, machine->symbol_filter);
+			al->map->dso->vaddr = 0;
+
 		al->addr = al->map->map_ip(al->map, al->addr);
 	}
 }
diff --git a/tools/perf/util/map.h b/tools/perf/util/map.h
index 7309d64..68ab703 100644
--- a/tools/perf/util/map.h
+++ b/tools/perf/util/map.h
@@ -86,14 +86,17 @@ void map_groups__put(struct map_groups *mg);
 struct kmap *map__kmap(struct map *map);
 struct map_groups *map__kmaps(struct map *map);
 
+u64 dso__get_vaddr(struct dso *dso);
 static inline u64 map__map_ip(struct map *map, u64 ip)
 {
-	return ip - map->start + map->pgoff;
+	return ip - map->start + map->pgoff +
+		dso__get_vaddr(map->dso);
 }
 
 static inline u64 map__unmap_ip(struct map *map, u64 ip)
 {
-	return ip + map->start - map->pgoff;
+	return ip + map->start - map->pgoff -
+		dso__get_vaddr(map->dso);
 }
 
 static inline u64 identity__map_ip(struct map *map __maybe_unused, u64 ip)
diff --git a/tools/perf/util/symbol-elf.c b/tools/perf/util/symbol-elf.c
index 562b8eb..de0129f 100644
--- a/tools/perf/util/symbol-elf.c
+++ b/tools/perf/util/symbol-elf.c
@@ -15,6 +15,8 @@
 #define EM_AARCH64	183  /* ARM 64 bit */
 #endif
 
+static int elf_read_maps_ex(Elf * elf, bool exe, mapfn_t mapfn, void *data,
+			    struct map *map);
 
 #ifdef HAVE_CPLUS_DEMANGLE_SUPPORT
 extern char *cplus_demangle(const char *, int);
@@ -831,6 +833,10 @@ int dso__load_sym(struct dso *dso, struct map *map,
 	sec = syms_ss->symtab;
 	shdr = syms_ss->symshdr;
 
+	err = elf_read_maps_ex(elf, ehdr.e_type == ET_EXEC ||
+			       ehdr.e_type == ET_REL,
+			       NULL, NULL, map);
+
 	if (runtime_ss->opdsec)
 		opddata = elf_rawdata(runtime_ss->opdsec, NULL);
 
@@ -1116,6 +1122,13 @@ out_elf_end:
 
 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
 {
+
+	return elf_read_maps_ex(elf, exe, mapfn, data, NULL);
+}
+
+static int elf_read_maps_ex(Elf *elf, bool exe, mapfn_t mapfn, void *data,
+			    struct map *map)
+{
 	GElf_Phdr phdr;
 	size_t i, phdrnum;
 	int err;
@@ -1135,10 +1148,17 @@ static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
 		} else {
 			if (!(phdr.p_flags & PF_R))
 				continue;
+			if (map && (phdr.p_flags & PF_X))
+				map->dso->vaddr = phdr.p_vaddr;
 		}
+
 		sz = min(phdr.p_memsz, phdr.p_filesz);
 		if (!sz)
 			continue;
+
+		if (!mapfn)
+			continue;
+
 		err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
 		if (err)
 			return err;
-- 
1.8.5.2


^ permalink raw reply related	[flat|nested] 6+ messages in thread
* Re: [PATCH] perf symbols: Fix wrong symbol offset according to p_addr
  2016-01-06  8:26     ` [PATCH] perf symbols: Fix wrong symbol offset according to p_addr He Kuang
@ 2016-01-11 11:25       ` Ingo Molnar
  2016-01-12  2:58         ` [PATCH v2] " He Kuang
  0 siblings, 1 reply; 6+ messages in thread
From: Ingo Molnar @ 2016-01-11 11:25 UTC (permalink / raw)
  To: He Kuang
  Cc: peterz, mingo, acme, jolsa, rabin.vincent, namhyung, wangnan0,
	mark.rutland, linux-kernel


* He Kuang <hekuang@huawei.com> wrote:

> Perf missed the 'VirtAddr' value in elf program_headers when adjusting
> symbol address for dynamic libraries. This patch considers the p_addr
> value and gets the right symbol offset.

s/gets/calculates

> 
> Before this patch, some symbols can not be right parsed on android phone:

s/Before this patch, some symbols are not parsed correctly on android phones:

> 
>   init   369 [002]   339.970607: raw_syscalls:sys_enter: NR 22 (b, 7fd9e360a0, 10, ffffffff, 0, 8)
>                      ...
>                    230ac [unknown] (/system/lib64/libsurfaceflinger.so)
>                     11a0 main (/system/bin/surfaceflinger)
>                    1c3fc __libc_init (/system/lib64/libc.so)
>                      fd0 _start (/system/bin/surfaceflinger)
>                     29ec __dl__start (/system/bin/linker64)
> 
> After this patch:
> 
>   init   369 [002]   339.970607: raw_syscalls:sys_enter: NR 22 (b, 7fd9e360a0, 10, ffffffff, 0, 8)
>                      ...
>                    3a0ac _ZN7android14SurfaceFlinger3runEv (/system/lib64/libsurfaceflinger.so)
>                     11a0 main (/system/bin/surfaceflinger)
>                    1c3fc __libc_init (/system/lib64/libc.so)
>                      fd0 _start (/system/bin/surfaceflinger)
>                     29ec __dl__start (/system/bin/linker64)

> --- a/tools/perf/util/dso.h
> +++ b/tools/perf/util/dso.h
> @@ -166,6 +166,7 @@ struct dso {
>  	const char	 *long_name;
>  	u16		 long_name_len;
>  	u16		 short_name_len;
> +	u64		vaddr;
>  	void		*dwfl;			/* DWARF debug info */

btw., looks like there's some extra spaces in front of those other field 
definitions.

Also, might make sense to describe what 'vaddr' stands for precisely, the naming 
is pretty ambiguous.

Thanks,

	Ingo

^ permalink raw reply	[flat|nested] 6+ messages in thread
* [PATCH v2] perf symbols: Fix wrong symbol offset according to p_addr
  2016-01-11 11:25       ` Ingo Molnar
@ 2016-01-12  2:58         ` He Kuang
  0 siblings, 0 replies; 6+ messages in thread
From: He Kuang @ 2016-01-12  2:58 UTC (permalink / raw)
  To: peterz, mingo, acme, jolsa, rabin.vincent, namhyung, wangnan0,
	mark.rutland, hekuang
  Cc: linux-kernel

Perf missed the 'VirtAddr' value in elf program_headers when adjusting
symbol address for dynamic libraries. This patch considers the p_addr
value and calculates the right symbol offset.

Before this patch, some symbols are not parsed correctly on android phone:

  init   369 [002]   339.970607: raw_syscalls:sys_enter: NR 22 (b, 7fd9e360a0, 10, ffffffff, 0, 8)
                     ...
                   230ac [unknown] (/system/lib64/libsurfaceflinger.so)
                    11a0 main (/system/bin/surfaceflinger)
                   1c3fc __libc_init (/system/lib64/libc.so)
                     fd0 _start (/system/bin/surfaceflinger)
                    29ec __dl__start (/system/bin/linker64)

After this patch:

  init   369 [002]   339.970607: raw_syscalls:sys_enter: NR 22 (b, 7fd9e360a0, 10, ffffffff, 0, 8)
                     ...
                   3a0ac _ZN7android14SurfaceFlinger3runEv (/system/lib64/libsurfaceflinger.so)
                    11a0 main (/system/bin/surfaceflinger)
                   1c3fc __libc_init (/system/lib64/libc.so)
                     fd0 _start (/system/bin/surfaceflinger)
                    29ec __dl__start (/system/bin/linker64)

Signed-off-by: He Kuang <hekuang@huawei.com>
---

Patch v1 to v2:
  - Modify commit message by following Ingo's suggestion.
  - Change ambiguous virable name 'vaddr' to 'load_virtaddr'.
  
---
 tools/perf/util/dso.c        |  6 ++++++
 tools/perf/util/dso.h        |  3 ++-
 tools/perf/util/event.c      |  6 +++++-
 tools/perf/util/map.h        |  7 +++++--
 tools/perf/util/symbol-elf.c | 20 ++++++++++++++++++++
 5 files changed, 38 insertions(+), 4 deletions(-)

diff --git a/tools/perf/util/dso.c b/tools/perf/util/dso.c
index e8e9a9d..afc51ce 100644
--- a/tools/perf/util/dso.c
+++ b/tools/perf/util/dso.c
@@ -1063,6 +1063,7 @@ struct dso *dso__new(const char *name)
 		dso->needs_swap = DSO_SWAP__UNSET;
 		RB_CLEAR_NODE(&dso->rb_node);
 		dso->root = NULL;
+		dso->load_virtaddr = 0;
 		INIT_LIST_HEAD(&dso->node);
 		INIT_LIST_HEAD(&dso->data.open_entry);
 		pthread_mutex_init(&dso->lock, NULL);
@@ -1367,3 +1368,8 @@ int dso__strerror_load(struct dso *dso, char *buf, size_t buflen)
 	scnprintf(buf, buflen, "%s", dso_load__error_str[idx]);
 	return 0;
 }
+
+u64 dso__get_load_virtaddr(struct dso *dso)
+{
+	return dso->load_virtaddr;
+}
diff --git a/tools/perf/util/dso.h b/tools/perf/util/dso.h
index 45ec4d0..2a03b87 100644
--- a/tools/perf/util/dso.h
+++ b/tools/perf/util/dso.h
@@ -166,7 +166,8 @@ struct dso {
 	const char	 *long_name;
 	u16		 long_name_len;
 	u16		 short_name_len;
-	void		*dwfl;			/* DWARF debug info */
+	u64		 load_virtaddr;		/* Load virtual address */
+	void		 *dwfl;			/* DWARF debug info */
 	struct auxtrace_cache *auxtrace_cache;
 
 	/* dso data file */
diff --git a/tools/perf/util/event.c b/tools/perf/util/event.c
index cd61bb1..7f1dbdf 100644
--- a/tools/perf/util/event.c
+++ b/tools/perf/util/event.c
@@ -1273,8 +1273,12 @@ try_again:
 		 * Kernel maps might be changed when loading symbols so loading
 		 * must be done prior to using kernel maps.
 		 */
+		map__load(al->map, machine->symbol_filter);
+
+		/* exclude kernel kallsyms */
 		if (load_map)
-			map__load(al->map, machine->symbol_filter);
+			al->map->dso->load_virtaddr = 0;
+
 		al->addr = al->map->map_ip(al->map, al->addr);
 	}
 }
diff --git a/tools/perf/util/map.h b/tools/perf/util/map.h
index 7309d64..db284a4 100644
--- a/tools/perf/util/map.h
+++ b/tools/perf/util/map.h
@@ -86,14 +86,17 @@ void map_groups__put(struct map_groups *mg);
 struct kmap *map__kmap(struct map *map);
 struct map_groups *map__kmaps(struct map *map);
 
+u64 dso__get_load_virtaddr(struct dso *dso);
 static inline u64 map__map_ip(struct map *map, u64 ip)
 {
-	return ip - map->start + map->pgoff;
+	return ip - map->start + map->pgoff +
+		dso__get_load_virtaddr(map->dso);
 }
 
 static inline u64 map__unmap_ip(struct map *map, u64 ip)
 {
-	return ip + map->start - map->pgoff;
+	return ip + map->start - map->pgoff -
+		dso__get_load_virtaddr(map->dso);
 }
 
 static inline u64 identity__map_ip(struct map *map __maybe_unused, u64 ip)
diff --git a/tools/perf/util/symbol-elf.c b/tools/perf/util/symbol-elf.c
index 562b8eb..9a02b3e 100644
--- a/tools/perf/util/symbol-elf.c
+++ b/tools/perf/util/symbol-elf.c
@@ -15,6 +15,8 @@
 #define EM_AARCH64	183  /* ARM 64 bit */
 #endif
 
+static int elf_read_maps_ex(Elf * elf, bool exe, mapfn_t mapfn, void *data,
+			    struct map *map);
 
 #ifdef HAVE_CPLUS_DEMANGLE_SUPPORT
 extern char *cplus_demangle(const char *, int);
@@ -831,6 +833,10 @@ int dso__load_sym(struct dso *dso, struct map *map,
 	sec = syms_ss->symtab;
 	shdr = syms_ss->symshdr;
 
+	err = elf_read_maps_ex(elf, ehdr.e_type == ET_EXEC ||
+			       ehdr.e_type == ET_REL,
+			       NULL, NULL, map);
+
 	if (runtime_ss->opdsec)
 		opddata = elf_rawdata(runtime_ss->opdsec, NULL);
 
@@ -1116,6 +1122,13 @@ out_elf_end:
 
 static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
 {
+
+	return elf_read_maps_ex(elf, exe, mapfn, data, NULL);
+}
+
+static int elf_read_maps_ex(Elf *elf, bool exe, mapfn_t mapfn, void *data,
+			    struct map *map)
+{
 	GElf_Phdr phdr;
 	size_t i, phdrnum;
 	int err;
@@ -1135,10 +1148,17 @@ static int elf_read_maps(Elf *elf, bool exe, mapfn_t mapfn, void *data)
 		} else {
 			if (!(phdr.p_flags & PF_R))
 				continue;
+			if (map && (phdr.p_flags & PF_X))
+				map->dso->load_virtaddr = phdr.p_vaddr;
 		}
+
 		sz = min(phdr.p_memsz, phdr.p_filesz);
 		if (!sz)
 			continue;
+
+		if (!mapfn)
+			continue;
+
 		err = mapfn(phdr.p_vaddr, sz, phdr.p_offset, data);
 		if (err)
 			return err;
-- 
1.8.5.2

^ permalink raw reply related	[flat|nested] 6+ messages in thread
end of thread, other threads:[~2016-01-12  3:05 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-01-05  9:14 [bug] wrong result of android callchain He Kuang
2016-01-05 10:20 ` Peter Zijlstra
2016-01-05 11:13   ` Will Deacon
2016-01-06  8:26     ` [PATCH] perf symbols: Fix wrong symbol offset according to p_addr He Kuang
2016-01-11 11:25       ` Ingo Molnar
2016-01-12  2:58         ` [PATCH v2] " He Kuang

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).