On Sat 10-06-17 20:57:46, Tetsuo Handa wrote: > Michal Hocko wrote: > > And just to clarify a bit. The OOM killer should be invoked whenever > > appropriate from the allocation context. If we decide to fail the > > allocation in the PF path then we can safely roll back and retry the > > whole PF. This has an advantage that any locks held while doing the > > allocation will be released and that alone can help to make a further > > progress. Moreover we can relax retry-for-ever _inside_ the allocator > > semantic for the PF path and fail allocations when we cannot make > > further progress even after we hit the OOM condition or we do stall for > > too long. > > What!? Are you saying that leave the allocator loop rather than invoke > the OOM killer if it is from page fault event without __GFP_FS set? > With below patch applied (i.e. ignore __GFP_FS for emulation purpose), > I can trivially observe systemwide lockup where the OOM killer is > never called. Because you have ruled the OOM out of the game completely from the PF path AFICS. So that is clearly _not_ what I meant (read the second sentence). What I meant was that page fault allocations _could_ fail _after_ we have used _all_ the reclaim opportunities. Without this patch this would be impossible. Note that I am not proposing that change now because that would require a deeper audit but it sounds like a viable way to go long term. > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index b896897..c79dfd5 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -3255,6 +3255,9 @@ void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...) > > *did_some_progress = 0; > > + if (current->in_pagefault) > + return NULL; > + > /* > * Acquire the oom lock. If that fails, somebody else is > * making progress for us. -- Michal Hocko SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
