Re: [PATCH 1/4] mm: Trial do_wp_page() simplification

From: John Hubbard
Date: Thu Sep 17 2020 - 16:19:07 EST

On 9/17/20 1:06 PM, Jason Gunthorpe wrote:
On Thu, Sep 17, 2020 at 12:42:11PM -0700, Linus Torvalds wrote:
Is there possibly somethign else we can filter on than just
GUP_PIN_COUNTING_BIAS? Because it could be as simple as just marking
the vma itself and saying "this vma has had a page pinning event done
on it".

We'd have to give up pin_user_pages_fast() to do that as we can't fast
walk and get vmas?

oops, yes. I'd forgotten about that point. Basically all of the O_DIRECT
callers need _fast. This is a big problem.

John Hubbard

Hmm, there are many users. I remember that the hfi1 folks really
wanted the fast version for some reason..

Because if we only start copying the page *iff* the vma is marked by
that "this vma had page pinning" _and_ the page count is bigger than
GUP_PIN_COUNTING_BIAS, than I think we can rest pretty easily knowing
that we aren't going to hit some regular old-fashioned UNIX server
cases with a lot of forks..


Given that this is a user visible regression, it is nearly rc6, what
do you prefer for next steps?

Sorting out this for fork, especially if it has the vma change is
probably more than a weeks time.

Revert this patch and try again next cycle?