RE: [PATCH RFC V4 5/6] perf top: switch to backward overwrite mode

[Liang, Kan]

> On Fri, Sep 29, 2017 at 07:47:56AM -0700, kan.liang@xxxxxxxxx wrote:
> > From: Kan Liang <kan.liang@xxxxxxxxx>
> >
> > perf_top__mmap_read has severe performance issue in Knights
> > Landing/Mill, when monitoring in heavy load system. It costs several
> > minutes to finish, which is unacceptable.
> >
> > perf top was overwrite mode. But it is changed to non overwrite mode
> > since commit 93fc64f14472 ("perf top: Switch to non overwrite mode").
> > For non overwrite mode, it tries to read everything in the ring buffer
> > and does not check the messup. Once there are lots of samples
> > delivered shortly, the processing time could be very long.
> > Knights Landing/Mill as a manycore processor contains a large number
> > of small cores. Because of the huge core number, it will generated
> > lots of samples in a heavy load system. Also, since the huge sample#,
> > the mmap writer probably bite the tail and mess up the samples.
> >
> > Also, to avoid the problems which is described in commit 9ecda41acb97
> > ("perf/core: Add ::write_backward attribute to perf event"), switch to
> > backward overwrite mode.
> > Pausing the ring-buffer during perf_top__mmap_read to ensure the
> > ring-buffer is stable.
> > There would be some records lost in backward overwrite mode. Removing
> > the lost events checking.
> 
> I'm getting perf top hogging the cpu completely with this change
> 

I think I find the root cause of the cpu hogging.
perf_mmap__read_catchup discards the md->prev from previous mmap_read.
Current mmap_read doesn't know which data has already been processed by
previous mmap_read. So it has to go through all the valid data in the ring buffer,
even most of the data has been processed by previous mmap_read.

Also, it looks perf record has the similar issue.
The previous location will be discarded as well in backward overwrite mode.
That will be an issue when --overwrite and --switch-output are enabled.
The new output will always include the old data in the previous output, which
should be wrong.

I think I will rewrite the perf_mmap__read_backward and perf_mmap__read_catchup
to fix this issue in a separate thread. Those functions should be common backward
mmap_read functions for all tools and tests.

BTW, are you OK with patch 1-4?
Those patches multithreading the machine__synthesize_threads, which is
irrelevant with the overwrite mode.
I think they can be merged separately.

Thanks,
Kan