Received-SPF: pass (google.com: domain of linux-kernel+bounces-137807-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
From: "Huang, Ying" <ying.huang@intel.com>
To: Zhaoyu Liu <liuzhaoyu.zackary@bytedance.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,  ryncsn@gmail.com,
  nphamcs@gmail.com,  songmuchun@bytedance.com,  david@redhat.com,
  chrisl@kernel.org,  guo.ziliang@zte.com.cn,  yosryahmed@google.com,
  linux-kernel@vger.kernel.org,  linux-mm@kvack.org
Subject: Re: [PATCH v2] mm: swap: prejudgement swap_has_cache to avoid page
 allocation
In-Reply-To: <20240409144132.GA542987@bytedance> (Zhaoyu Liu's message of
	"Tue, 9 Apr 2024 22:41:32 +0800")
References: <20240408121439.GA252652@bytedance>
	<20240408132704.f966adc8d3928df4d3b8c0a9@linux-foundation.org>
	<20240409144132.GA542987@bytedance>
Date: Wed, 10 Apr 2024 09:43:19 +0800
Message-ID: <87bk6i6lhk.fsf@yhuang6-desk2.ccr.corp.intel.com>
User-Agent: Gnus/5.13 (Gnus v5.13)
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=ascii

Zhaoyu Liu <liuzhaoyu.zackary@bytedance.com> writes:

> On Mon, Apr 08, 2024 at 01:27:04PM -0700, Andrew Morton wrote:
>> On Mon, 8 Apr 2024 20:14:39 +0800 Zhaoyu Liu <liuzhaoyu.zackary@bytedance.com> wrote:
>> 
>> > Based on qemu arm64 - latest kernel + 100M memory + 1024M swapfile.
>> > Create 1G anon mmap and set it to shared, and has two processes
>> > randomly access the shared memory. When they are racing on swap cache,
>> > on average, each "alloc_pages_mpol + swapcache_prepare + folio_put"
>> > took about 1475 us.
>> 
>> And what effect does this patch have upon the measured time?  ANd upon
>> overall runtime?
>
> Hi Andrew,
>
> When share memory between two or more processes has swapped and pagefault now,
> it would readahead swap and call __read_swap_cache_async().
> If one of the processes calls swapcache_prepare() and finds that the cache 
> has been EXIST(another process added), it will folio_put on the basis of the 
> alloc_pages_mpol() that has been called, and then try filemap_get_folio() again.
>
> I think the page alloc in this process is wasteful.
> when the memory pressure is large, alloc_pages_mpol() will be time-consuming, 
> so the purpose of my patch is to judge whether the page has cache before page alloc, 
> then skip page alloc and retry filemap_get_folio() to save the time of the function.

Please prove your theory with data, better with benchmark score.

--
Best Regards,
Huang, Ying

>> 
>> > So skip page allocation if SWAP_HAS_CACHE was set, just
>> > schedule_timeout_uninterruptible and continue to acquire page
>> > via filemap_get_folio() from swap cache, to speedup
>> > __read_swap_cache_async.