Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Fri, 5 Feb 2021 14:28:08 -0800 (PST)
From:   David Rientjes <rientjes@google.com>
To:     Charan Teja Kalla <charante@codeaurora.org>
cc:     akpm@linux-foundation.org, vbabka@suse.cz, mhocko@suse.com,
        vinmenon@codeaurora.org, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] mm: page_alloc: update the COMPACT[STALL|FAIL] events
 properly
In-Reply-To: <160ba3b5-2cd4-5ff0-1348-fb477cefd33d@codeaurora.org>
Message-ID: <1213f4c6-7557-268d-253e-23f8fea55b19@google.com>
References: <1612187338-19100-1-git-send-email-charante@codeaurora.org> <d9d1dd7b-1fe7-67b6-6ba4-fb1a6faa8fae@google.com> <160ba3b5-2cd4-5ff0-1348-fb477cefd33d@codeaurora.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Precedence: bulk

On Tue, 2 Feb 2021, Charan Teja Kalla wrote:

> >> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> >> index 519a60d..531f244 100644
> >> --- a/mm/page_alloc.c
> >> +++ b/mm/page_alloc.c
> >> @@ -4152,6 +4152,8 @@ __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
> >>  	memalloc_noreclaim_restore(noreclaim_flag);
> >>  	psi_memstall_leave(&pflags);
> >>  
> >> +	if (*compact_result == COMPACT_SKIPPED)
> >> +		return NULL;
> >>  	/*
> >>  	 * At least in one zone compaction wasn't deferred or skipped, so let's
> >>  	 * count a compaction stall
> > 
> > This makes sense, I wonder if it would also be useful to check that 
> > page == NULL, either in try_to_compact_pages() or here for 
> > COMPACT_SKIPPED?
> 
> In the code, when COMPACT_SKIPPED is being returned, the page will
> always be NULL. So, I'm not sure how much useful it is for the page ==
> NULL check here. Or I failed to understand your point here?
> 

Your code is short-circuiting the rest of  __alloc_pages_direct_compact() 
where the return value is dictated by whether page is NULL or non-NULL.  
We can't leak a captured page if we are testing for it being NULL or 
non-NULL, which is what the rest of __alloc_pages_direct_compact() does 
*before* your change.  So the idea was to add a check the page is actually 
NULL here since you are now relying on the return value of 
compact_zone_order() to be COMPACT_SKIPPED to infer page == NULL.

I agree that's currently true in the code, I was trying to catch any 
errors where current->capture_control.page was non-NULL but 
try_to_compact_pages() returns COMPACT_SKIPPED.  There's some complexity 
here.

So my idea was the expand this out to:

	if (*compact_result == COMPACT_SKIPPED) {
		VM_BUG_ON(page);
		return NULL;
	}