Re: [PATCH] kvm: Faults which trigger IO release the mmap_sem

[Andres Lagar-Cavilla]

On Wed, Sep 17, 2014 at 12:43 AM, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
> Il 16/09/2014 20:42, Andres Lagar-Cavilla ha scritto:
>> On Tue, Sep 16, 2014 at 11:29 AM, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
>>>>     I think a first patch should introduce kvm_get_user_page_retry ("Retry a
>>>>     fault after a gup with FOLL_NOWAIT.") and the second would add
>>>>     FOLL_TRIED ("This properly relinquishes mmap semaphore if the
>>>>     filemap/swap has to wait on page lock (and retries the gup to completion
>>>>     after that").
>>>>
>>>> That's not what FOLL_TRIED does. The relinquishing of mmap semaphore is
>>>> done by this patch minus the FOLL_TRIED bits. FOLL_TRIED will let the
>>>> fault handler (e.g. filemap) know that we've been there and waited on
>>>> the IO already, so in the common case we won't need to redo the IO.
>>>
>>> Yes, that's not what FOLL_TRIED does.  But it's the difference between
>>> get_user_pages and kvm_get_user_page_retry, right?
>>
>> Unfortunately get_user_pages does not expose the param (int
>> *nonblocking) that __gup will use to set FAULT_FLAG_ALLOW_RETRY. So
>> that's one difference. The second difference is that kvm_gup_retry
>> will call two times if necessary (the second without _RETRY but with
>> _TRIED).
>
> Yeah, that's how it is in your patch.  I can see that.
>
> What I'm saying is that your patch is two changes in one:
>
> 1) do not use gup_fast in hva_to_pfn_slow, instead use gup as in
> async_pf_execute.  This change can already introduce a function called
> kvm_get_user_page_retry, and can already use it in async_pf_execute and
> hva_to_pfn_slow
>
> 2) introduce the two-phase RETRY + TRIED mechanism in
> kvm_get_user_page_retry, so that the mmap semaphore is relinquished
> properly if the filemap or swap has to wait on the page lock.
>
> I would prefer to split it in two patches.  Is it clearer now?

Understood. So in patch 1, would kvm_gup_retry be ... just a wrapper
around gup? That looks thin to me, and the naming of the function will
not be accurate. Plus, considering Radim's suggestion that the naming
is not optimal.

I can have patch 1 just s/gup_fast/gup (one liner), and then patch 2
do the rest of the work.

Andres
>
> Paolo



-- 
Andres Lagar-Cavilla | Google Kernel Team | andreslc@xxxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/