2012-02-04 00:45:34

by Dan McGee

[permalink] [raw]
Subject: [perf] perf top segfaulting

On i686, version 3.2-2, but looks like annotate.c hasn't changed much
since. It sometimes happens within 5 seconds of starting perf,
sometimes much later, but almost always if I leave it running I well
come back to it having segfaulted. When ran with gdb here it took
about 3 minutes; I had a 5 second segfault and a 5 minute segfault
before and after this run as well. I'm not sure what triggers it other
than it isn't user input, as I can start `perf top`, not touch it, and
it will eventually segfault.

-Dan

Program received signal SIGSEGV, Segmentation fault.
symbol__inc_addr_samples (sym=0x91ca9e8, map=0x9153010, evidx=0,
addr=385376) at util/annotate.c:73
73 util/annotate.c: No such file or directory.
(gdb) bt
#0 symbol__inc_addr_samples (sym=0x91ca9e8, map=0x9153010, evidx=0,
addr=385376) at util/annotate.c:73
#1 0x08066593 in record_precise_ip (ip=<optimized out>, counter=0,
he=<optimized out>) at builtin-top.c:223
#2 perf_event__process_sample (session=0x827bf50, sample=0xbffff9f4,
evsel=<optimized out>, event=<optimized out>) at builtin-top.c:801
#3 perf_session__mmap_read_idx (self=0x827bf50, idx=0) at builtin-top.c:825
#4 0x08068489 in perf_session__mmap_read (self=0x827bf50) at builtin-top.c:839
#5 __cmd_top () at builtin-top.c:1003
#6 cmd_top (argc=<optimized out>, argv=0xbffffd38, prefix=0x0) at
builtin-top.c:1274
#7 0x08055e90 in run_builtin (p=0x80f2624, argc=1, argv=0xbffffd38)
at perf.c:286
#8 0x08055651 in handle_internal_command (argv=0xbffffd38, argc=1) at
perf.c:358
#9 run_argv (argv=0xbffffbd8, argcp=0xbffffbdc) at perf.c:402
#10 main (argc=1, argv=0xbffffd38) at perf.c:512
(gdb) bt full
#0 symbol__inc_addr_samples (sym=0x91ca9e8, map=0x9153010, evidx=0,
addr=385376) at util/annotate.c:73
offset = 4294557696
h = 0xb6a8101c
__func__ = "symbol__inc_addr_samples"
#1 0x08066593 in record_precise_ip (ip=<optimized out>, counter=0,
he=<optimized out>) at builtin-top.c:223
sym = 0x91ca9e8
#2 perf_event__process_sample (session=0x827bf50, sample=0xbffff9f4,
evsel=<optimized out>, event=<optimized out>) at builtin-top.c:801
he = 0xfff9c000
parent = 0x0
machine = 0x0
err = <optimized out>
origin = <optimized out>
ip = 3076591968
al = {thread = 0x91fa360, map = 0x91fa650, sym = 0x91ca9e8,
addr = 794976, level = 46 '.', filtered = false, cpumode = 2 '\002',
cpu = -1}
#3 perf_session__mmap_read_idx (self=0x827bf50, idx=0) at builtin-top.c:825
sample = {ip = 3076591968, pid = 16947, tid = 16947, time =
18446744073709551615, addr = 0, id = 18446744073709551615, stream_id =
18446744073709551615, period = 1057945,
cpu = 4294967295, raw_size = 4294967295, raw_data = 0x0,
callchain = 0x8}
evsel = <optimized out>
event = <optimized out>
ret = -1230499812
__PRETTY_FUNCTION__ = "perf_session__mmap_read_idx"
#4 0x08068489 in perf_session__mmap_read (self=0x827bf50) at builtin-top.c:839
i = <optimized out>
#5 __cmd_top () at builtin-top.c:1003
hits = <optimized out>
thread = 3075533632
#6 cmd_top (argc=<optimized out>, argv=0xbffffd38, prefix=0x0) at
builtin-top.c:1274
pos = <optimized out>
status = -12
#7 0x08055e90 in run_builtin (p=0x80f2624, argc=1, argv=0xbffffd38)
at perf.c:286
status = <optimized out>
st = {st_dev = 13258589575124611012, __pad1 = 49030, __st_ino
= 3087003588, st_mode = 3086926771, st_nlink = 3087006384, st_uid =
3079435296, st_gid = 1, st_rdev = 1,
__pad2 = 256, st_size = -4611691086352719864, st_blksize =
-1073742952, st_blocks = 577833303592140800, st_atim = {tv_sec =
135210216, tv_nsec = 1}, st_mtim = {
tv_sec = 135105751, tv_nsec = -1}, st_ctim = {tv_sec =
-1213274024, tv_nsec = 0}, st_ino = 135811744}
#8 0x08055651 in handle_internal_command (argv=0xbffffd38, argc=1) at
perf.c:358
p = 0x80f2624
cmd = 0xbffffe6b "top"
i = <optimized out>
commands = {{cmd = 0x80d20f7 "buildid-cache", fn = 0x805e8a0
<cmd_buildid_cache>, option = 0}, {cmd = 0x80ce581 "buildid-list", fn
= 0x805e320 <cmd_buildid_list>,
option = 0}, {cmd = 0x80ce58e "diff", fn = 0x8058630
<cmd_diff>, option = 0}, {cmd = 0x80ce593 "evlist", fn = 0x8058a60
<cmd_evlist>, option = 0}, {
cmd = 0x80ce3ea "help", fn = 0x8059420 <cmd_help>, option
= 0}, {cmd = 0x80d13b1 "list", fn = 0x805e920 <cmd_list>, option = 0},
{cmd = 0x80d6b7c "record",
fn = 0x8060360 <cmd_record>, option = 0}, {cmd = 0x80d5a46
"report", fn = 0x8060d90 <cmd_report>, option = 0}, {cmd = 0x80ce59a
"bench", fn = 0x8056850 <cmd_bench>,
option = 0}, {cmd = 0x80ce5a0 "stat", fn = 0x8062950
<cmd_stat>, option = 0}, {cmd = 0x80ce5a5 "timechart", fn = 0x8065550
<cmd_timechart>, option = 0}, {
cmd = 0x80ce5af "top", fn = 0x8067de0 <cmd_top>, option =
0}, {cmd = 0x80ceb3b "annotate", fn = 0x8056350 <cmd_annotate>, option
= 0}, {cmd = 0x80ce3f1 "version",
fn = 0x807c780 <cmd_version>, option = 0}, {cmd =
0x80d5b37 "script", fn = 0x806a5f0 <cmd_script>, option = 0}, {cmd =
0x80ce5b3 "sched", fn = 0x805c5d0 <cmd_sched>,
option = 0}, {cmd = 0x80ce5b9 "probe", fn = 0x806b740
<cmd_probe>, option = 0}, {cmd = 0x80ce5bf "kmem", fn = 0x806cf80
<cmd_kmem>, option = 0}, {
cmd = 0x80d9295 "lock", fn = 0x806e180 <cmd_lock>, option
= 0}, {cmd = 0x80ce5c4 "kvm", fn = 0x806eaf0 <cmd_kvm>, option = 0},
{cmd = 0x80ce5c8 "test",
fn = 0x8071880 <cmd_test>, option = 0}, {cmd = 0x80ce5cd
"inject", fn = 0x8071ed0 <cmd_inject>, option = 0}}
#9 run_argv (argv=0xbffffbd8, argcp=0xbffffbdc) at perf.c:402
done_alias = 0
#10 main (argc=1, argv=0xbffffd38) at perf.c:512
done_help = 0
was_alias = 0
cmd = 0xbffffe6b "top"


2012-02-04 02:24:57

by David Daney

[permalink] [raw]
Subject: Re: [perf] perf top segfaulting

On 02/03/2012 04:45 PM, Dan McGee wrote:
> On i686, version 3.2-2, but looks like annotate.c hasn't changed much
> since. It sometimes happens within 5 seconds of starting perf,
> sometimes much later, but almost always if I leave it running I well
> come back to it having segfaulted. When ran with gdb here it took
> about 3 minutes; I had a 5 second segfault and a 5 minute segfault
> before and after this run as well. I'm not sure what triggers it other
> than it isn't user input, as I can start `perf top`, not touch it, and
> it will eventually segfault.
>


I have seen the same thing (basically the same stack trace), so I think
what I see is probably closely related. My failures however are on
mips64 based systems.

My debugging suggests that this happens when the ABIs used by the
running processes are heterogeneous (A mixture of 32-bit and 64-bit
processes). What I see is that all processes use a library with a
common name, but differing in paths (/lib32/libc-2.11.3.so and
/lib64/libc-2.11.3.so for example). It looks like perf is confusing the
offsets it caches from one library to look up information in the other
and since the symbols are in different locations, the resulting
erroneous address calculations result in accesses to unmapped portions
of perf's address space and you get SIGSEGV.

I haven't dug into the code enough to suggest a fix, but I think that at
a high hand-waving level, this is what is happening. I have never
observed the failure when using only a single ABI on the system

David Daney


> -Dan
>
> Program received signal SIGSEGV, Segmentation fault.
> symbol__inc_addr_samples (sym=0x91ca9e8, map=0x9153010, evidx=0,
> addr=385376) at util/annotate.c:73
> 73 util/annotate.c: No such file or directory.
> (gdb) bt
> #0 symbol__inc_addr_samples (sym=0x91ca9e8, map=0x9153010, evidx=0,
> addr=385376) at util/annotate.c:73
> #1 0x08066593 in record_precise_ip (ip=<optimized out>, counter=0,
> he=<optimized out>) at builtin-top.c:223
> #2 perf_event__process_sample (session=0x827bf50, sample=0xbffff9f4,
> evsel=<optimized out>, event=<optimized out>) at builtin-top.c:801
> #3 perf_session__mmap_read_idx (self=0x827bf50, idx=0) at builtin-top.c:825
> #4 0x08068489 in perf_session__mmap_read (self=0x827bf50) at builtin-top.c:839
> #5 __cmd_top () at builtin-top.c:1003
> #6 cmd_top (argc=<optimized out>, argv=0xbffffd38, prefix=0x0) at
> builtin-top.c:1274
> #7 0x08055e90 in run_builtin (p=0x80f2624, argc=1, argv=0xbffffd38)
> at perf.c:286
> #8 0x08055651 in handle_internal_command (argv=0xbffffd38, argc=1) at
> perf.c:358
> #9 run_argv (argv=0xbffffbd8, argcp=0xbffffbdc) at perf.c:402
> #10 main (argc=1, argv=0xbffffd38) at perf.c:512
> (gdb) bt full
> #0 symbol__inc_addr_samples (sym=0x91ca9e8, map=0x9153010, evidx=0,
> addr=385376) at util/annotate.c:73
> offset = 4294557696
> h = 0xb6a8101c
> __func__ = "symbol__inc_addr_samples"
> #1 0x08066593 in record_precise_ip (ip=<optimized out>, counter=0,
> he=<optimized out>) at builtin-top.c:223
> sym = 0x91ca9e8
> #2 perf_event__process_sample (session=0x827bf50, sample=0xbffff9f4,
> evsel=<optimized out>, event=<optimized out>) at builtin-top.c:801
> he = 0xfff9c000
> parent = 0x0
> machine = 0x0
> err =<optimized out>
> origin =<optimized out>
> ip = 3076591968
> al = {thread = 0x91fa360, map = 0x91fa650, sym = 0x91ca9e8,
> addr = 794976, level = 46 '.', filtered = false, cpumode = 2 '\002',
> cpu = -1}
> #3 perf_session__mmap_read_idx (self=0x827bf50, idx=0) at builtin-top.c:825
> sample = {ip = 3076591968, pid = 16947, tid = 16947, time =
> 18446744073709551615, addr = 0, id = 18446744073709551615, stream_id =
> 18446744073709551615, period = 1057945,
> cpu = 4294967295, raw_size = 4294967295, raw_data = 0x0,
> callchain = 0x8}
> evsel =<optimized out>
> event =<optimized out>
> ret = -1230499812
> __PRETTY_FUNCTION__ = "perf_session__mmap_read_idx"
> #4 0x08068489 in perf_session__mmap_read (self=0x827bf50) at builtin-top.c:839
> i =<optimized out>
> #5 __cmd_top () at builtin-top.c:1003
> hits =<optimized out>
> thread = 3075533632
> #6 cmd_top (argc=<optimized out>, argv=0xbffffd38, prefix=0x0) at
> builtin-top.c:1274
> pos =<optimized out>
> status = -12
> #7 0x08055e90 in run_builtin (p=0x80f2624, argc=1, argv=0xbffffd38)
> at perf.c:286
> status =<optimized out>
> st = {st_dev = 13258589575124611012, __pad1 = 49030, __st_ino
> = 3087003588, st_mode = 3086926771, st_nlink = 3087006384, st_uid =
> 3079435296, st_gid = 1, st_rdev = 1,
> __pad2 = 256, st_size = -4611691086352719864, st_blksize =
> -1073742952, st_blocks = 577833303592140800, st_atim = {tv_sec =
> 135210216, tv_nsec = 1}, st_mtim = {
> tv_sec = 135105751, tv_nsec = -1}, st_ctim = {tv_sec =
> -1213274024, tv_nsec = 0}, st_ino = 135811744}
> #8 0x08055651 in handle_internal_command (argv=0xbffffd38, argc=1) at
> perf.c:358
> p = 0x80f2624
> cmd = 0xbffffe6b "top"
> i =<optimized out>
> commands = {{cmd = 0x80d20f7 "buildid-cache", fn = 0x805e8a0
> <cmd_buildid_cache>, option = 0}, {cmd = 0x80ce581 "buildid-list", fn
> = 0x805e320<cmd_buildid_list>,
> option = 0}, {cmd = 0x80ce58e "diff", fn = 0x8058630
> <cmd_diff>, option = 0}, {cmd = 0x80ce593 "evlist", fn = 0x8058a60
> <cmd_evlist>, option = 0}, {
> cmd = 0x80ce3ea "help", fn = 0x8059420<cmd_help>, option
> = 0}, {cmd = 0x80d13b1 "list", fn = 0x805e920<cmd_list>, option = 0},
> {cmd = 0x80d6b7c "record",
> fn = 0x8060360<cmd_record>, option = 0}, {cmd = 0x80d5a46
> "report", fn = 0x8060d90<cmd_report>, option = 0}, {cmd = 0x80ce59a
> "bench", fn = 0x8056850<cmd_bench>,
> option = 0}, {cmd = 0x80ce5a0 "stat", fn = 0x8062950
> <cmd_stat>, option = 0}, {cmd = 0x80ce5a5 "timechart", fn = 0x8065550
> <cmd_timechart>, option = 0}, {
> cmd = 0x80ce5af "top", fn = 0x8067de0<cmd_top>, option =
> 0}, {cmd = 0x80ceb3b "annotate", fn = 0x8056350<cmd_annotate>, option
> = 0}, {cmd = 0x80ce3f1 "version",
> fn = 0x807c780<cmd_version>, option = 0}, {cmd =
> 0x80d5b37 "script", fn = 0x806a5f0<cmd_script>, option = 0}, {cmd =
> 0x80ce5b3 "sched", fn = 0x805c5d0<cmd_sched>,
> option = 0}, {cmd = 0x80ce5b9 "probe", fn = 0x806b740
> <cmd_probe>, option = 0}, {cmd = 0x80ce5bf "kmem", fn = 0x806cf80
> <cmd_kmem>, option = 0}, {
> cmd = 0x80d9295 "lock", fn = 0x806e180<cmd_lock>, option
> = 0}, {cmd = 0x80ce5c4 "kvm", fn = 0x806eaf0<cmd_kvm>, option = 0},
> {cmd = 0x80ce5c8 "test",
> fn = 0x8071880<cmd_test>, option = 0}, {cmd = 0x80ce5cd
> "inject", fn = 0x8071ed0<cmd_inject>, option = 0}}
> #9 run_argv (argv=0xbffffbd8, argcp=0xbffffbdc) at perf.c:402
> done_alias = 0
> #10 main (argc=1, argv=0xbffffd38) at perf.c:512
> done_help = 0
> was_alias = 0
> cmd = 0xbffffe6b "top"
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>

2012-02-04 02:46:33

by Dan McGee

[permalink] [raw]
Subject: Re: [perf] perf top segfaulting

On Fri, Feb 3, 2012 at 8:24 PM, David Daney <[email protected]> wrote:
> On 02/03/2012 04:45 PM, Dan McGee wrote:
>>
>> On i686, version 3.2-2, but looks like annotate.c hasn't changed much
>> since. It sometimes happens within 5 seconds of starting perf,
>> sometimes much later, but almost always if I leave it running I well
>> come back to it having segfaulted. When ran with gdb here it took
>> about 3 minutes; I had a 5 second segfault and a 5 minute segfault
>> before and after this run as well. I'm not sure what triggers it other
>> than it isn't user input, as I can start `perf top`, not touch it, and
>> it will eventually segfault.
>>
>
>
> I have seen the same thing (basically the same stack trace), so I think what
> I see is probably closely related.  My failures however are on mips64 based
> systems.
>
> My debugging suggests that this happens when the ABIs used by the running
> processes are heterogeneous (A mixture of 32-bit and 64-bit processes).
>  What I see is that all processes use a library with a common name, but
> differing in paths (/lib32/libc-2.11.3.so and /lib64/libc-2.11.3.so for
> example).  It looks like perf is confusing the offsets it caches from one
> library to look up information in the other and since the symbols are in
> different locations, the resulting erroneous address calculations result in
> accesses to unmapped portions of perf's address space and you get SIGSEGV.
>
> I haven't dug into the code enough to suggest a fix, but I think that at a
> high hand-waving level, this is what is happening.  I have never observed
> the failure when using only a single ABI on the system

Note that in this case, it is a pure 32-bit x86 system, and no library
changes were going on in the background. So I wouldn't be surprised if
the causes are similar (or the same), but I don't think I can chalk it
up to being a single ABI vs multiple ABI problem; i686 only has one
ABI.

-Dan

2012-02-04 08:59:50

by Sorin Dumitru

[permalink] [raw]
Subject: Re: [perf] perf top segfaulting

On Sat, Feb 4, 2012 at 4:46 AM, Dan McGee <[email protected]> wrote:
> On Fri, Feb 3, 2012 at 8:24 PM, David Daney <[email protected]> wrote:
>> On 02/03/2012 04:45 PM, Dan McGee wrote:
>>>
>>> On i686, version 3.2-2, but looks like annotate.c hasn't changed much
>>> since. It sometimes happens within 5 seconds of starting perf,
>>> sometimes much later, but almost always if I leave it running I well
>>> come back to it having segfaulted. When ran with gdb here it took
>>> about 3 minutes; I had a 5 second segfault and a 5 minute segfault
>>> before and after this run as well. I'm not sure what triggers it other
>>> than it isn't user input, as I can start `perf top`, not touch it, and
>>> it will eventually segfault.
>>>
>>
>>
>> I have seen the same thing (basically the same stack trace), so I think what
>> I see is probably closely related. ?My failures however are on mips64 based
>> systems.
>>
>> My debugging suggests that this happens when the ABIs used by the running
>> processes are heterogeneous (A mixture of 32-bit and 64-bit processes).
>> ?What I see is that all processes use a library with a common name, but
>> differing in paths (/lib32/libc-2.11.3.so and /lib64/libc-2.11.3.so for
>> example). ?It looks like perf is confusing the offsets it caches from one
>> library to look up information in the other and since the symbols are in
>> different locations, the resulting erroneous address calculations result in
>> accesses to unmapped portions of perf's address space and you get SIGSEGV.
>>
>> I haven't dug into the code enough to suggest a fix, but I think that at a
>> high hand-waving level, this is what is happening. ?I have never observed
>> the failure when using only a single ABI on the system
>
> Note that in this case, it is a pure 32-bit x86 system, and no library
> changes were going on in the background. So I wouldn't be surprised if
> the causes are similar (or the same), but I don't think I can chalk it
> up to being a single ABI vs multiple ABI problem; i686 only has one
> ABI.
>
> -Dan
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at ?http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at ?http://www.tux.org/lkml/

I've seen this same problem on a pure 32-bit x86 system. So it definitely isn't
an ABI problem.
>From what i can tell the problem is in perf_event__process_sample. When calling
perf_event__process_sample, we set al->sym based on al->address. The symbol
in the hist_entry is set to the one from al but in the call to
perf_top__record_precise_ip
we pass in the address from the event struct which is sometimes
different than the one
in the al structure. When this situation occurs, when calculating the offset in
symbol__inc_addr_samples, because addr is not in the symbol [start,end] range,
we get a very big value which causes the segfault when we use it to
index something.
I've sent a patch that works for me, but i don't know if it's the
right solution at [1].

[1] https://lkml.org/lkml/2012/1/29/59

2012-02-23 20:35:43

by Jérôme Carretero

[permalink] [raw]
Subject: Re: [perf] perf top segfaulting

On Sat, 4 Feb 2012 10:59:28 +0200
Sorin Dumitru <[email protected]> wrote:

> I've sent a patch that works for me, but i don't know if it's the
> right solution at [1].
>
> [1] https://lkml.org/lkml/2012/1/29/59

Hi,

Just to say that I had the issue on ARM and this patch works for me.

Regards,

--
cJ