Re: [RFC 1/3] mm, oom: refactor oom detection

[Kamezawa Hiroyuki]

On 2015/10/30 17:23, Michal Hocko wrote:
> On Fri 30-10-15 14:23:59, KAMEZAWA Hiroyuki wrote:
> > On 2015/10/30 0:17, mhocko@xxxxxxxxxx wrote:
> [...]
> > > @@ -3135,13 +3145,56 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
> > >    	if (gfp_mask & __GFP_NORETRY)
> > >    		goto noretry;
> > >
> > > -	/* Keep reclaiming pages as long as there is reasonable progress */
> > > +	/*
> > > +	 * Do not retry high order allocations unless they are __GFP_REPEAT
> > > +	 * and even then do not retry endlessly.
> > > +	 */
> > >    	pages_reclaimed += did_some_progress;
> > > -	if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
> > > -	    ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
> > > -		/* Wait for some write requests to complete then retry */
> > > -		wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
> > > -		goto retry;
> > > +	if (order > PAGE_ALLOC_COSTLY_ORDER) {
> > > +		if (!(gfp_mask & __GFP_REPEAT) || pages_reclaimed >= (1<<order))
> > > +			goto noretry;
> > > +
> > > +		if (did_some_progress)
> > > +			goto retry;
> >
> > why directly retry here ?
>
> Because I wanted to preserve the previous logic for GFP_REPEAT as much
> as possible here and do an incremental change in the later patch.

I see.

> [...]
>
> > > @@ -3150,8 +3203,10 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
> > >    		goto got_pg;
> > >
> > >    	/* Retry as long as the OOM killer is making progress */
> > > -	if (did_some_progress)
> > > +	if (did_some_progress) {
> > > +		stall_backoff = 0;
> > >    		goto retry;
> > > +	}
> >
> > Umm ? I'm sorry that I didn't notice page allocation may fail even
> > if order < PAGE_ALLOC_COSTLY_ORDER.  I thought old logic ignores
> > did_some_progress. It seems a big change.
>
> __alloc_pages_may_oom will set did_some_progress
>
> > So, now, 0-order page allocation may fail in a OOM situation ?
>
> No they don't normally and this patch doesn't change the logic here.

I understand your patch doesn't change the behavior.
Looking into __alloc_pages_may_oom(), *did_some_progress is finally set by

     if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
                *did_some_progress = 1;

...depends on out_of_memory() return value.
Now, allocation may fail if oom-killer is disabled.... Isn't it complicated ?

Shouldn't we have

 if (order < PAGE_ALLOC_COSTLY_ORDER)
    goto retry;

here ?

Thanks,
-Kame






--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/