Re: [PATCH] mm/page_alloc: Clean up check_for_memory

From: Oscar Salvador <osalvador@techadventures.net>
To: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: "akpm@linux-foundation.org" <akpm@linux-foundation.org>,
	"mhocko@suse.com" <mhocko@suse.com>,
	"vbabka@suse.cz" <vbabka@suse.cz>,
	"sfr@canb.auug.org.au" <sfr@canb.auug.org.au>,
	"iamjoonsoo.kim@lge.com" <iamjoonsoo.kim@lge.com>,
	"linux-mm@kvack.org" <linux-mm@kvack.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Oscar Salvador <osalvador@suse.de>
Subject: Re: [PATCH] mm/page_alloc: Clean up check_for_memory
Date: Fri, 31 Aug 2018 22:45:43 +0200	[thread overview]
Message-ID: <20180831204543.GA3885@techadventures.net> (raw)
In-Reply-To: <b2fea9ef-84e9-84dc-c847-5b944a8d832f@microsoft.com>

On Fri, Aug 31, 2018 at 02:04:59PM +0000, Pasha Tatashin wrote:
> Are you saying the code that is in mainline is broken? Because we set
> node_set_state(nid, N_NORMAL_MEMORY); even on node with N_HIGH_MEMORY:
> 
> 6826			if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
> 6827			    zone_type <= ZONE_NORMAL)
> 6828				node_set_state(nid, N_NORMAL_MEMORY);

Yes, and that is fine. Although the curent code is subtle for the reasons
I expplained in the changelog.
What I am saying is that the code you suggested would not work
because your code either sets N_NORMAL_MEMORY or N_HIGH_MEMORY and then
breaks the loop.

That is wrong because when we are on a CONFIG_HIGHMEM system,
it can happen that we have a node with both types, so we have to set
both types of memory.

N_HIGH_MEMORY, and N_NORMAL_MEMORY if the zone is <= ZONE_NORMAL.

Thanks
-- 
Oscar Salvador
SUSE L3