Re: [PATCH 03/10] perf record: Turns auxtrace_snapshot_enable into 3 states

[Wangnan (F)]

On 2016/4/14 16:30, Adrian Hunter wrote:
> On 14/04/16 10:50, Wangnan (F) wrote:
> > On 2016/4/14 15:15, Adrian Hunter wrote:
> > > On 13/04/16 18:55, Arnaldo Carvalho de Melo wrote:
> > > > Em Wed, Apr 13, 2016 at 08:21:06AM +0000, Wang Nan escreveu:
> > > > > auxtrace_snapshot_enable has only two states (0/1). Turns it into a
> > > > > triple states enum so SIGUSR2 handler can safely do other works without
> > > > > triggering auxtrace snapshot.
> > > > Adrian, can you take a look at this? Is it ok with you?
> > > Please forgive me if these are stupid questions:
> > >
> > > First I am wondering why we wouldn't want to snapshot auxtrace data at the
> > > same time as the perf buffer?
> > This patch doesn't prevent taking snapshot when receiving SIGUSR2.
> So it was a stupid question ;-)

Still thank you for pointing this. I suddenly realized the 
'switch_output_started'
in patch 5/10 is also need to be turned to a 3 state enum. If not, a 
SIGUSR2 incorrectly
triggers output switching even '--switch-output' is not provided when 
'--snapshot' exist.

> > If both --snapshot and --switch-outupt is provided, when SIGUSR2 received,
> > perf takes auxtrace snapshot and other perf buffer together.
> How do you keep from losing tracking information like MMAP events?  Are they
> is a different buffer?

Please see patch 8/10 and 9/10. MMAP events are resynthesized each time
when output file switched, so at the *head* of each 'perf.data' you can find
many MMAP/COMM/FORK... events.

After overwritable ring buffer is supported, there is a more aggresive
patch [1] resynthesize tracking events and put them at the *end* of
perf.data.

[1] 
https://git.kernel.org/cgit/linux/kernel/git/pi3orama/linux.git/commit/?h=perf/overwrite&id=747e10300397b9c28b01bca5bfad943c8cf2dcce