Re: [PATCH v2] mm: Stop kswapd early when nothing's waiting for it to free pages

[Sultan Alsawaf]

On Fri, Feb 21, 2020 at 10:22:02AM -0800, Ira Weiny wrote:
> On Thu, Feb 20, 2020 at 08:30:52PM -0800, Sultan Alsawaf wrote:
> > From: Sultan Alsawaf <sultan@xxxxxxxxxxxxxxx>
>  
> [snip]
> 
> > @@ -4640,9 +4647,12 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
> >  		goto retry;
> >  	}
> >  fail:
> > -	warn_alloc(gfp_mask, ac->nodemask,
> > -			"page allocation failure: order:%u", order);
> >  got_pg:
> 
> I have no insight into if this is masking a deeper problem or if this fixes
> something but doesn't the above result in 'fail' and 'got_pg' being the same
> label?
> 
> Ira
> 
> > +	if (woke_kswapd)
> > +		atomic_dec(&pgdat->kswapd_waiters);
> > +	if (!page)
> > +		warn_alloc(gfp_mask, ac->nodemask,
> > +				"page allocation failure: order:%u", order);
> >  	return page;
> >  }
>   
> [snip]

Yes,. This was to reduce the patch delta for the initial submission so it's
clearer what's going on; it can be altered of course to coalesce the labels into
a single one. I typically produce my patches to upstream components to be as
uninvasive as possible to aid in backporting and forward porting in case it's
rejected and I want to keep it for myself.

Sultan