Re: Question about x86/mm/gup.c's use of disabled interrupts

[Avi Kivity <avi@redhat.com>]

Jeremy Fitzhardinge wrote:
>> I thought you were concerned about cpu 0 doing a gup_fast(), cpu 1 
>> doing P->N, and cpu 2 doing N->P.  In this case cpu 2 is waiting on 
>> the pte lock.
>
> The issue is that if cpu 0 is doing a gup_fast() and other cpus are 
> doing P->P updates, then gup_fast() can potentially get a mix of old 
> and new pte values - where P->P is any aggregate set of unsynchronized 
> P->N and N->P operations on any number of other cpus.  Ah, but if 
> every P->N is followed by a tlb flush, then disabling interrupts will 
> hold off any following N->P, allowing gup_fast to get a consistent pte 
> snapshot.
>

Right.

> Hm, awkward if flush_tlb_others doesn't IPI...
>

How can it avoid flushing the tlb on cpu [01]?  It's it's gup_fast()ing 
a pte, it may as well load it into the tlb.

>
> Simplest fix is to make gup_get_pte() a pvop, but that does seem like 
> putting a red flag in front of an inner-loop hotspot, or something...
>
> The per-cpu tlb-flush exclusion flag might really be the way to go.

I don't see how it will work, without changing Xen to look at the flag?

local_irq_disable() is used here to lock out a remote cpu, I don't see 
why deferring the flush helps.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.