Re: [PATCH RFC 1/4] mm/gup: Add FOLL_INTERRUPTIBLE

[Peter Xu <peterx@redhat.com>]

On Tue, Jun 21, 2022 at 10:23:32AM +0200, David Hildenbrand wrote:
> On 17.06.22 03:41, Peter Xu wrote:
> > We have had FAULT_FLAG_INTERRUPTIBLE but it was never applied to GUPs.  One
> > issue with it is that not all GUP paths are able to handle signal delivers
> > besides SIGKILL.
> > 
> > That's not ideal for the GUP users who are actually able to handle these
> > cases, like KVM.
> > 
> > KVM uses GUP extensively on faulting guest pages, during which we've got
> > existing infrastructures to retry a page fault at a later time.  Allowing
> > the GUP to be interrupted by generic signals can make KVM related threads
> > to be more responsive.  For examples:
> > 
> >   (1) SIGUSR1: which QEMU/KVM uses to deliver an inter-process IPI,
> >       e.g. when the admin issues a vm_stop QMP command, SIGUSR1 can be
> >       generated to kick the vcpus out of kernel context immediately,
> > 
> >   (2) SIGINT: which can be used with interactive hypervisor users to stop a
> >       virtual machine with Ctrl-C without any delays/hangs,
> > 
> >   (3) SIGTRAP: which grants GDB capability even during page faults that are
> >       stuck for a long time.
> > 
> > Normally hypervisor will be able to receive these signals properly, but not
> > if we're stuck in a GUP for a long time for whatever reason.  It happens
> > easily with a stucked postcopy migration when e.g. a network temp failure
> > happens, then some vcpu threads can hang death waiting for the pages.  With
> > the new FOLL_INTERRUPTIBLE, we can allow GUP users like KVM to selectively
> > enable the ability to trap these signals.
> 
> This makes sense to me. I assume relevant callers will detect "GUP
> failed" but also "well, there is a signal to handle" and cleanly back
> off, correct?

Correct, via an -EINTR.

One thing to mention is that the gup user behavior will be the same as
before if the caller didn't explicilty pass in FOLL_INTERRUPTIBLE with the
gup call.  So after the whole series applied only kvm (and only some path
of kvm, not all GUP; I only touched up the x86 slow page fault path) to
handle this, but that'll be far enough to cover 99.99% use cases that I
wanted to take care of.

E.g., some kvm request to gup on some guest apic page may not still be able
to respond to a SIGUSR1 but that's very very rare, and we can always add
more users of FOLL_INTERRUPTIBLE when the code is ready to benefit from the
fast respondings.

Thanks,

-- 
Peter Xu