Re: Re: [PATCH 3/4] mm/memofy-failure.c: optimize hwpoison_filter

[Naoya Horiguchi <naoya.horiguchi@linux.dev>]

On Sat, May 07, 2022 at 08:28:05AM +0800, zhenwei pi wrote:
> 
> On 5/7/22 00:28, David Hildenbrand wrote:
> > On 06.05.22 15:38, zhenwei pi wrote:
> > > 
> > > 
> > > On 5/6/22 16:59, Naoya Horiguchi wrote:
> > > > On Fri, Apr 29, 2022 at 10:22:05PM +0800, zhenwei pi wrote:
> > > > > In the memory failure procedure, hwpoison_filter has higher priority,
> > > > > if memory_filter() filters the error event, there is no need to do
> > > > > the further work.
> > > > 
> > > > Could you clarify what problem you are trying to solve (what does
> > > > "optimize" mean in this context or what is the benefit)?
> > > > 
> > > 
> > > OK. The background of this work:
> > > As well known, the memory failure mechanism handles memory corrupted
> > > event, and try to send SIGBUS to the user process which uses this
> > > corrupted page.
> > > 
> > > For the virtualization case, QEMU catches SIGBUS and tries to inject MCE
> > > into the guest, and the guest handles memory failure again. Thus the
> > > guest gets the minimal effect from hardware memory corruption.
> > > 
> > > The further step I'm working on:
> > > 1, try to modify code to decrease poisoned pages in a single place
> > > (mm/memofy-failure.c: simplify num_poisoned_pages_dec in this series).

This is fine to me.

> > > 
> > > 2, try to use page_handle_poison() to handle SetPageHWPoison() and
> > > num_poisoned_pages_inc() together. It would be best to call
> > > num_poisoned_pages_inc() in a single place too. I'm not sure if this is
> > > possible or not, please correct me if I misunderstand.

SetPageHWPoison() can be cancelled in memory_failure(), so simply bundling
it with num_poisoned_pages_inc() might not be optimal.  I think that
action_result() is supposed to be called when memory error handling is
effective (not filtered, not cancelled). So moving num_poisoned_pages_inc()
(and notification code in your plan) into this function might be good.

> > > 
> > > 3, introduce memory failure notifier list in memory-failure.c: notify
> > > the corrupted PFN to someone who registers this list.
> > > If I can complete [1] and [2] part, [3] will be quite easy(just call
> > > notifier list after increasing poisoned page).
> > > 
> > > 4, introduce memory recover VQ for memory balloon device, and registers
> > > memory failure notifier list. During the guest kernel handles memory
> > > failure, balloon device gets notified by memory failure notifier list,
> > > and tells the host to recover the corrupted PFN(GPA) by the new VQ.
> > 
> > Most probably you might want to do that asynchronously, and once the
> > callback succeeds, un-poison the page.
> > 
> Yes!
> 
> > > 
> > > 5, host side remaps the corrupted page(HVA), and tells the guest side to
> > > unpoison the PFN(GPA). Then the guest fixes the corrupted page(GPA)
> > > dynamically.
> > 
> > I think QEMU already does that during reboots. Now it would be triggered
> > by the guest for individual pages.
> > 
> Yes, currently QEMU supports to un-poison corrupted pages during
> reset/reboot. We can reuse some code to do the work in this case, this
> allows a VM to fix corrupted pages as soon as possible(also no need to
> reset/reboot).

So this finally allows to replace broken page mapped to guest with
a healthy page without rebooting the guest. That sounds helpful.

Thanks,
Naoya Horiguchi

> 
> > > 
> > > Because [4] and [5] are related to balloon device, also CC Michael,
> > > David and Jason.
> > 
> > Doesn't sound too crazy for me, although it's a shame that we always
> > have to use virtio-balloon for such fairly balloon-unrelated things.
> > 
> Thanks!
> 
> -- 
> zhenwei pi