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:   Wed, 12 Aug 2020 17:32:34 +0800
From:   Gao Xiang <hsiangkao@redhat.com>
To:     Daeho Jeong <daeho43@gmail.com>
Cc:     linux-kernel@vger.kernel.org,
        linux-f2fs-devel@lists.sourceforge.net, kernel-team@android.com,
        Daeho Jeong <daehojeong@google.com>
Subject: Re: [f2fs-dev] [PATCH v3] f2fs: change virtual mapping way for
 compression pages
Message-ID: <20200812093234.GA759@xiangao.remote.csb>
References: <20200812051711.2147716-1-daeho43@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200812051711.2147716-1-daeho43@gmail.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Aug 12, 2020 at 02:17:11PM +0900, Daeho Jeong wrote:
> From: Daeho Jeong <daehojeong@google.com>
> 
> By profiling f2fs compression works, I've found vmap() callings have
> unexpected hikes in the execution time in our test environment and
> those are bottlenecks of f2fs decompression path. Changing these with
> vm_map_ram(), we can enhance f2fs decompression speed pretty much.
> 
> [Verification]
> Android Pixel 3(ARM64, 6GB RAM, 128GB UFS)
> Turned on only 0-3 little cores(at 1.785GHz)
> 
> dd if=/dev/zero of=dummy bs=1m count=1000
> echo 3 > /proc/sys/vm/drop_caches
> dd if=dummy of=/dev/zero bs=512k
> 
> - w/o compression -
> 1048576000 bytes (0.9 G) copied, 2.082554 s, 480 M/s
> 1048576000 bytes (0.9 G) copied, 2.081634 s, 480 M/s
> 1048576000 bytes (0.9 G) copied, 2.090861 s, 478 M/s
> 
> - before patch -
> 1048576000 bytes (0.9 G) copied, 7.407527 s, 135 M/s
> 1048576000 bytes (0.9 G) copied, 7.283734 s, 137 M/s
> 1048576000 bytes (0.9 G) copied, 7.291508 s, 137 M/s
> 
> - after patch -
> 1048576000 bytes (0.9 G) copied, 1.998959 s, 500 M/s
> 1048576000 bytes (0.9 G) copied, 1.987554 s, 503 M/s
> 1048576000 bytes (0.9 G) copied, 1.986380 s, 503 M/s
>

The reason why I raised up this was that I once observed vmap() vs
vm_map_ram() on arm64 kirin platform as well. it indeed had some
impact (~10%) but not as huge as this. Anyway, such description
with test environment looks ok.

Thanks,
Gao Xiang