Re: [PATCH v6] Documentation: userspace-api: Document perf ring buffer mechanism

[Namhyung Kim]

Hi Leo,

On Sat, Aug 26, 2023 at 12:30 AM Leo Yan <leo.yan@xxxxxxxxxx> wrote:
>
> Hi Namhyung,
>
> On Tue, Aug 22, 2023 at 09:26:05PM -0700, Namhyung Kim wrote:
>
> [...]
>
> > > > > +Why ring buffers are mapped with above two different modes?  Here the
> > > > > +write direction matters.  The forward writing starts to save data from
> > > > > +the beginning of the ring buffer and wrap around when overflow, which is
> > > > > +used with the read-write mode in the normal ring buffer.  When the
> > > > > +consumer doesn't keep up with the producer, it would lose some data, the
> > > > > +kernel keeps how many records it lost and generates the
> > > > > +``PERF_RECORD_LOST`` records in the next time when it finds a space in the
> > > > > +ring buffer.
> > > >
> > > > Thanks for the update.  It's unclear to me if all 4 combination of
> > > > (rw, ro) x (fwd, bwd) are possible (yes!).  The rw mode and back-
> > > > ward is also possible but just not used for perf tool.
> > >
> > > I can add a matrix for the combinations:
> > >
> > >   The combination is supported in perf tool:
> > >
> > >   ---+------------+-----------
> > >      |  Forward   | Backward
> > >   ---+------------+-----------
> > >   rw |  Yes       |   No
> > >   ---+------------+-----------
> > >   ro |  X         |   Yes
> > >   ---+------------+-----------
> > >
> > >   Yes: is supported
> > >   No: is not supported
> > >   X: is not feasible
> >
> > I think they are all supported.  You can use rw mode with backward
> > direction but it's just not intuitive.  Also ro mode with forward direction
> > is working but there's a chance to miss the start position of the
> > previous event.
>
> I am a bit confused for the all four modes are supported.
>
> From the code [1], we can see there have only two combinations:
>
> - overwrite + read-only mode (PROT_READ);
> - no-overwrite + read-write mode (PROT_READ | PROT_WRITE);
>
> And I think perf tool must use the backward writing for "overwrite",
> and use the forward writing for "non-overwrite" mode.  I think the
> code [2] can help us to conclude this.

I mean the kernel supports all the combinations but perf tool uses
the selected two.  IIUC you are describing the kernel behavior, right?

Thanks,
Namhyung

>
> In the end, the perf tool doesn't support the combinations:
>
> - overwrite + read-write mode
> - no-overwrite + read-only mode
>
> I have sent a v6 patch, please let me know if you have any concerns or
> if I misunderstand anything.  Thanks!
>
> Leo
>
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/lib/perf/evlist.c#n474
> [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/lib/perf/mmap.c#n141