Re: [PATCH] mm,page_alloc,cma: conditionally prefer cma pageblocks for movable allocations

From: Vlastimil Babka <vbabka@suse.cz>
To: Joonsoo Kim <js1304@gmail.com>
Cc: Rik van Riel <riel@surriel.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Linux Memory Management List <linux-mm@kvack.org>,
	LKML <linux-kernel@vger.kernel.org>,
	kernel-team@fb.com, Roman Gushchin <guro@fb.com>,
	Qian Cai <cai@lca.pw>, Mel Gorman <mgorman@techsingularity.net>,
	Anshuman Khandual <anshuman.khandual@arm.com>,
	Joonsoo Kim <iamjoonsoo.kim@lge.com>
Subject: Re: [PATCH] mm,page_alloc,cma: conditionally prefer cma pageblocks for movable allocations
Date: Wed, 11 Mar 2020 18:41:38 +0100	[thread overview]
Message-ID: <89865b7b-f0ed-999d-0f23-4dbfb45115db@suse.cz> (raw)
In-Reply-To: <CAAmzW4P6+3O_RLvgy_QOKD4iXw+Hk3HE7Toc4Ky7kvQbCozCeA@mail.gmail.com>

On 3/11/20 11:13 AM, Joonsoo Kim wrote:
>>
>> Few years ago Joonsoo wanted to fix these kinds of weird MIGRATE_CMA corner
>> cases by using ZONE_MOVABLE instead [1]. Unfortunately it was reverted due to
>> unresolved bugs. Perhaps the idea could be resurrected now?
>>
>> [1]
>> https://lore.kernel.org/linux-mm/1512114786-5085-1-git-send-email-iamjoonsoo.kim@lge.com/
> 
> Thanks for ccing, Vlastimil.

No problem. Also, welcome back :)

> Recently, I'm working for resurrecting this idea.
> I will send the preparation patches in this or next week.

Good to hear!

> Unresolved bugs of my patchset comes from the fact that ZONE_MOVABLE
> which is used for
> serving CMA memory in my patchset could have both lowmem(direct mapped) and
> highmem(no direct mapped) pages on CONFIG_HIGHMEM enabled system.
> 
> For someone to use this memory, PageHighMem() should be called to
> check if there is direct
> mapping or not. Current PageHighMem() implementation is:
> 
> #define PageHighMem(__p) is_highmem_idx(page_zonenum(__p))
> 
> Since ZONE_MOVABLE has both typed pages, ZONE_MOVABLE should be considered
> as highmem zone. In this case, PageHighMem() always returns TRUE for
> all pages on
> ZONE_MOVABLE and lowmem pages on ZONE_MOVABLE could make some troubles.

Doh, HighMem brings only troubles these days [1].

[1] https://lwn.net/Articles/813201/

> My plan to fix this problem is to change the PageHighMem() implementation.
> 
> #define PageHighMem(__p) (page_to_pfn(__p) >= max_low_pfn)
> 
> In fact, PageHighMem() is used to check whether direct mapping exists or not.
> With this new implementation, regardless of the zone type of the page, we can
> correctly check if the page is direct mapped or not. Changing the
> name, PageHighMem(),
> to !PageDirectMapped() is also planned but it will be done after
> everything have settle down.
> 
> Unfortunately, before changing the implementation, I should check the
> all call-sites
> of PageHighMem() since there is some callers who use PageHighMem() to check
> the zone type.
> 
> What my preparation patch will does is to correct this PageHighMem() usage.
> After fixing it, I will try to merge the patchset [1].
> 
> Thanks.
> 