Re: Possible duplicate page fault accounting on some archs after commit 4064b9827063

[Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>]

On Wed, Jun 10, 2020 at 8:48 AM Gerald Schaefer
<gerald.schaefer@xxxxxxxxxx> wrote:
>
> This was found by coincidence in s390 code, and a quick check showed that
> there are quite a lot of other architectures that seem to be affected in a
> similar way. I'm preparing a fix for s390, by moving the accounting behind
> the retry loop, similar to x86. It is not completely straight-forward, so
> I leave the fix for other archs to the respective maintainers.

Hmm. I wonder if we could move the handling into  handle_mm_fault() itself.

It's _fairly_ trivial to do on the arch side, just as long as you
remember to make the VM_FAULT_MAJOR bit sticky like x86 does with that

        major |= fault & VM_FAULT_MAJOR;

right after handle_mm_fault(). But it certainly doesn't seem like it
would be hard to move into common code in handle_mm_fault() either, by
just not doing the accounting if it's about to return VM_FAULT_RETRY
or VM_FAULT_ERROR.

That said, we want that perf_sw_event() accounting too, so we'd have
to pass in a 'struct regs *' as well. And it's not clear which way
accounting should go for other callers of handle_mm_fault() (ie gup
etc).

So I guess just having architectures fix it up individually and make
sure they don't do it for retry conditions is the right thing to do..

             Linus