Re: [PATCH 1/1] mm: do not increment pgfault stats when page fault handler retries

[Suren Baghdasaryan]

On Fri, Apr 14, 2023 at 4:49 PM Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote:
>
> On Fri, Apr 14, 2023 at 3:35 PM Peter Xu <peterx@xxxxxxxxxx> wrote:
> >
> > Hi, Suren,
> >
> > On Fri, Apr 14, 2023 at 03:14:23PM -0700, Suren Baghdasaryan wrote:
> > > > It also already ignores invalid faults:
> > > >
> > > >         if (ret & (VM_FAULT_ERROR | VM_FAULT_RETRY))
> > > >                 return;
> > >
> > > Can there be a case of (!VM_FAULT_ERROR && VM_FAULT_RETRY) - basically
> > > we need to retry but no errors happened? If so then this condition
> > > would double-count pagefaults in such cases.
> >
> > If ret==VM_FAULT_RETRY it should return here already, so I assume
> > mm_account_fault() itself is fine regarding fault retries?
> >
> > Note that I think "ret & (VM_FAULT_ERROR | VM_FAULT_RETRY)" above means
> > "either ERROR or RETRY we'll skip the accounting".
> >
> > IMHO we should have 3 cases here:
> >
> >   - ERROR && !RETRY
> >     error triggered of any kind
> >
> >   - RETRY && !ERROR
> >     we need to try one more time
> >
> >   - !RETRY && !ERROR
> >     we finished the fault
>
> After looking some more into mm_account_fault(), I think it would be
> fine to count the faults which produced errors. IIUC these counters
> represent the total number of faults, not the number of valid and
> successful faults. If so then I think simply using VM_FAULT_RETRY
> should be ok without considering all possible combinations. WDYT?

I posted v2 at https://lore.kernel.org/all/20230415000818.1955007-1-surenb@xxxxxxxxxx/
Hopefully it's closer to what we want it to be.

>
> >
> > I don't think ERROR & RETRY can even be set at the same time so I assume
> > there's no option 4) - a RETRY should imply no ERROR already, even though
> > it's still incomplete so need another attempt.
> >
> > Thanks,
> >
> > --
> > Peter Xu
> >