Re: [PATCH RFC 03/24] mm: allow VM_FAULT_RETRY for multiple times

[Jerome Glisse <jglisse@redhat.com>]

On Tue, Jan 22, 2019 at 04:22:38PM +0800, Peter Xu wrote:
> On Mon, Jan 21, 2019 at 10:55:36AM -0500, Jerome Glisse wrote:
> > On Mon, Jan 21, 2019 at 03:57:01PM +0800, Peter Xu wrote:
> > > The idea comes from a discussion between Linus and Andrea [1].
> > > 
> > > Before this patch we only allow a page fault to retry once.  We achieved
> > > this by clearing the FAULT_FLAG_ALLOW_RETRY flag when doing
> > > handle_mm_fault() the second time.  This was majorly used to avoid
> > > unexpected starvation of the system by looping over forever to handle
> > > the page fault on a single page.  However that should hardly happen, and
> > > after all for each code path to return a VM_FAULT_RETRY we'll first wait
> > > for a condition (during which time we should possibly yield the cpu) to
> > > happen before VM_FAULT_RETRY is really returned.
> > > 
> > > This patch removes the restriction by keeping the FAULT_FLAG_ALLOW_RETRY
> > > flag when we receive VM_FAULT_RETRY.  It means that the page fault
> > > handler now can retry the page fault for multiple times if necessary
> > > without the need to generate another page fault event. Meanwhile we
> > > still keep the FAULT_FLAG_TRIED flag so page fault handler can still
> > > identify whether a page fault is the first attempt or not.
> > 
> > So there is nothing protecting starvation after this patch ? AFAICT.
> > Do we sufficient proof that we never have a scenario where one process
> > might starve fault another ?
> > 
> > For instance some page locking could starve one process.
> 
> Hi, Jerome,
> 
> Do you mean lock_page()?
> 
> AFAIU lock_page() will only yield the process itself until the lock is
> released, so IMHO it's not really starving the process but a natural
> behavior.  After all the process may not continue without handling the
> page fault correctly.
> 
> Or when you say "starvation" do you mean that we might return
> VM_FAULT_RETRY from handle_mm_fault() continuously so we'll looping
> over and over inside the page fault handler?

That one ie every time we retry someone else is holding the lock and
thus lock_page_or_retry() will continuously retry. Some process just
get unlucky ;)

With existing code because we remove the retry flag then on the second
try we end up waiting for the page lock while holding the mmap_sem so
we know that we are in line for the page lock and we will get it once
it is our turn.

Cheers,
Jérôme