Re: [PATCH v5 02/25] mm: userfault: return VM_FAULT_RETRY on signals

[Peter Xu]

On Wed, Jun 26, 2019 at 09:59:58AM +0800, Linus Torvalds wrote:
> On Tue, Jun 25, 2019 at 1:31 PM Peter Xu <peterx@xxxxxxxxxx> wrote:
> >
> > Yes that sounds reasonable to me, and that matches perfectly with
> > TASK_INTERRUPTIBLE and TASK_KILLABLE.  The only thing that I am a bit
> > uncertain is whether we should define FAULT_FLAG_INTERRUPTIBLE as a
> > new bit or make it simply a combination of:
> >
> >   FAULT_FLAG_KILLABLE | FAULT_FLAG_USER
> 
> It needs to be a new bit, I think.
> 
> Some things could potentially care about the difference between "can I
> abort this thing because the task will *die* and never see the end
> result" and "can I abort this thing because it will be retried".
> 
> For a regular page fault, maybe FAULT_FLAG_INTERRUPTBLE will always be
> set for the same things that set FAULT_FLAG_KILLABLE when it happens
> from user mode, but at least conceptually I think they are different,
> and it could make a difference for things like get_user_pages() or
> similar.
> 
> Also, I actually don't think we should ever expose FAULT_FLAG_USER to
> any fault handlers anyway. It has a very specific meaning for memory
> cgroup handling, and no other fault handler should likely ever care
> about "was this a user fault". So I'd actually prefer for people to
> ignore and forget that hacky flag entirely, rather than give it subtle
> semantic meaning together with KILLABLE.

OK.

> 
> [ Side note: this is the point where I may soon lose internet access,
> so I'll probably not be able to participate in the discussion any more
> for a while ]

Appreciate for these suggestions.  I'll prepare something with that
new bit and see whether that could be accepted.  I'll also try to
split those out of the bigger series.

Thanks,

-- 
Peter Xu