Re: [PATCH v9 00/14] Add support for remote unwind

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

Em Tue, Jun 07, 2016 at 12:12:15PM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Sat, Jun 04, 2016 at 01:39:04PM +0800, Hekuang escreveu:
> > 在 2016/6/4 5:09, Arnaldo Carvalho de Melo 写道:
> > > Em Fri, Jun 03, 2016 at 06:06:02PM -0300, Arnaldo Carvalho de Melo escreveu:
> > > > Anyway, please try your patchkit with 'perf top' and 'perf trace', as
> > > > both don't use perf.data files, i.e. they work 'live', so probably
> > > > things that you touch in normalize_arch() are not initialized and need
> > > > to be setup.
> > > Ah, no need to resend the whole patchkit, just find out what is the bug
> > > and send me a patch and I'll insert it at the right point to avoid
> > > introducing a bisect breaking point.
> 
> > > Your patchkit is in my perf/unwind branch at my tree, I already added
> > > Jiri's Acked-by in all the patches.
> 
> > > git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git
> 
> > I send the updated one after PATCH 10/14,  env->arch is null in live mode,
> > comments are added and "perf top/trace" works now.
> 
> Ok, now it fails 'perf test unwind' and I bisected it to:

Further info:

[root@jouet ~]# perf test -v unwind
48: Test dwarf unwind                                        :
--- start ---
test child forked, pid 25078
got wrong number of stack entries 0 != 8
test child finished with -1
---- end ----
Test dwarf unwind: FAILED!
[root@jouet ~]#
 
> [acme@jouet linux]$ git bisect good
> e2b6ce1ca32cedeb2cef08f49aa812f0eadcc3c5 is the first bad commit
> commit e2b6ce1ca32cedeb2cef08f49aa812f0eadcc3c5
> Author: He Kuang <hekuang@huawei.com>
> Date:   Fri Jun 3 03:33:19 2016 +0000
> 
>     perf unwind: Check the target platform before assigning unwind methods
>     
>     Currently, 'perf script' uses host unwind methods to parse perf.data
>     callchain info without taking the target architecture into account, i.e.
>     assuming the perf.data file was generated on the same machine where the
>     analysis is being performed. So we get wrong result without any warnings
>     when unwinding callchains of x86(32-bit) on x86(64-bit) machine.
>     
>     This patch adds an extra step that checks the target platform before
>     assigning unwind methods. In later patches in this series, we can use
>     this info to assign the right unwind methods for supported platforms.
>