lkml.org 
[lkml]   [2016]   [Feb]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 1/2] mm: fix bogus VM_BUG_ON_PAGE() in isolate_lru_page()
On Mon 01-02-16 16:38:53, Kirill A. Shutemov wrote:
> On Mon, Feb 01, 2016 at 03:24:46PM +0100, Michal Hocko wrote:
> > On Mon 01-02-16 16:26:08, Kirill A. Shutemov wrote:
> > > We don't care if there's a tail pages which is not on LRU. We are not
> > > going to isolate them anyway.
> >
> > yes we are not going to isolate them but calling this function on a
> > tail page is wrong in principle, no? PageLRU check is racy outside of
> > lru_lock so what if we are racing here. I know, highly unlikely but not
> > impossible. So I am not really sure this is an improvement. When would
> > we hit this VM_BUG_ON and it wouldn't be a bug or at least suspicious
> > usage?
>
> Yes, there is no point in calling isolate_lru_page() for tail pages, but
> we do this anyway -- see the second patch.

yes, I have seen it and that is a bug as well AFAIU. So the VM_BUG_ON
triggered for the real bug.

> And we need to validate all drivers, that they don't forget to set VM_IO
> or make vma_migratable() return false in other way.

Yes, some drivers will do it incorrectly but this is VM_BUG_ON so it is
usually disabled no?

> Alternative approach would be to downgrate the VM_BUG_ON_PAGE() to
> WARN_ONCE_ON(). This way we would have chance to catch bad callers.

a ratelimitted WARN_ON would work as well.
--
Michal Hocko
SUSE Labs

\
 
 \ /
  Last update: 2016-02-02 14:41    [W:0.035 / U:2.396 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site