Re: [PATCH 6/8] perf symbols: Improve DSO long names lookup speed with rbtree

[Arnaldo Carvalho de Melo <acme@kernel.org>]

Em Tue, Oct 14, 2014 at 11:09:58AM +0200, Jiri Olsa escreveu:
> On Wed, Oct 01, 2014 at 04:50:41PM -0300, Arnaldo Carvalho de Melo wrote:
> > From: Waiman Long <Waiman.Long@hp.com>
> > 
> > With workload that spawns and destroys many threads and processes, it
> > was found that perf-mem could took a long time to post-process the perf
> > data after the target workload had completed its operation.
> > 
> > The performance bottleneck was found to be the lookup and insertion of
> > the new DSO structures (thousands of them in this case).
> 
> this change segfaults (below) some tests, but only if I compiled
> without DEBUG when I revert this commit, I can no longer reproduce..

Reproduced, looking at it... 
 
> jirka
> 
> (gdb) set follow-fork-mode child
> (gdb) r test 31
> Starting program: /home/jolsa/kernel.org/linux-perf/tools/perf/perf test 31
> warning: section  not found in /usr/lib/debug/lib/modules/3.16.3-200.fc20.x86_64/vdso/vdso64.so.debug
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib64/libthread_db.so.1".
> 31: Test output sorting of hist entries                    :[New process 15477]
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib64/libthread_db.so.1".
> 
> Program received signal SIGSEGV, Segmentation fault.
> [Switching to Thread 0x7ffff7b9d7c0 (LWP 15477)]
> __strcmp_ssse3 () at ../sysdeps/x86_64/strcmp.S:210
> 210             movlpd  (%rsi), %xmm2
> (gdb) bt
> #0  __strcmp_ssse3 () at ../sysdeps/x86_64/strcmp.S:210
> #1  0x0000000000477967 in dso__findlink_by_longname (name=<optimized out>, dso=0x0, root=0x7fffffffdbf0)
>     at util/dso.c:674
> #2  dso__find_by_longname (name=0x7fffffffcae8 "perf", root=0x7fffffffdbf0) at util/dso.c:712
> #3  dsos__find (cmp_short=false, name=0x7fffffffcae8 "perf", dsos=0x7fffffffdbe0) at util/dso.c:935
> #4  __dsos__findnew (dsos=dsos@entry=0x7fffffffdbe0, name=name@entry=0x7fffffffcae8 "perf") at util/dso.c:940
> #5  0x00000000004915d9 in map__new (machine=machine@entry=0x7fffffffdb90, start=4194304, len=1048576, pgoff=0, 
>     pid=<optimized out>, d_maj=d_maj@entry=0, d_min=d_min@entry=0, ino=ino@entry=0, ino_gen=ino_gen@entry=0, 
>     prot=prot@entry=0, flags=flags@entry=0, filename=filename@entry=0x7fffffffcae8 "perf", type=MAP__FUNCTION, 
>     thread=thread@entry=0x90d1f0) at util/map.c:180
> #6  0x00000000004900f4 in machine__process_mmap_event (machine=machine@entry=0x7fffffffdb90, 
>     event=event@entry=0x7fffffffcac0, sample=sample@entry=0x0) at util/machine.c:1182
> #7  0x00000000004d12bb in setup_fake_machine (machines=machines@entry=0x7fffffffdb90)
>     at tests/hists_common.c:116
> #8  0x00000000004d4478 in test__hists_output () at tests/hists_output.c:600
> #9  0x0000000000448fe4 in run_test (test=0x8166a0 <tests+480>) at tests/builtin-test.c:210
> #10 __cmd_test (skiplist=0x0, argv=0x7fffffffe2d0, argc=1) at tests/builtin-test.c:255
> #11 cmd_test (argc=1, argv=0x7fffffffe2d0, prefix=<optimized out>) at tests/builtin-test.c:320
> #12 0x000000000041c8f5 in run_builtin (p=p@entry=0x814fc0 <commands+480>, argc=argc@entry=2, 
>     argv=argv@entry=0x7fffffffe2d0) at perf.c:331
> #13 0x000000000041c110 in handle_internal_command (argv=0x7fffffffe2d0, argc=2) at perf.c:390
> #14 run_argv (argv=0x7fffffffe050, argcp=0x7fffffffe05c) at perf.c:434
> #15 main (argc=2, argv=0x7fffffffe2d0) at perf.c:549
> (gdb)