linux-kernel - Re: [PATCH] i2c: designware: do not disable adapter after transfer

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 8 Apr 2016 15:17:49 +0300
From:	Jarkko Nikula <jarkko.nikula@...ux.intel.com>
To:	Lucas De Marchi <lucas.de.marchi@...il.com>,
	linux-i2c@...r.kernel.org
Cc:	Lucas De Marchi <lucas.demarchi@...el.com>,
	linux-kernel@...r.kernel.org, Wolfram Sang <wsa@...-dreams.de>,
	Mika Westerberg <mika.westerberg@...ux.intel.com>,
	christian.ruppert@...tech.com
Subject: Re: [PATCH] i2c: designware: do not disable adapter after transfer

Hi

On 04/01/2016 05:47 AM, Lucas De Marchi wrote:
> From: Lucas De Marchi <lucas.demarchi@...el.com>
>
> Disabling the adapter after each transfer is pretty bad for sensors and
> other devices doing small transfers at a high rate. It slows down the
> transfer rate a lot since each of them have to wait the adapter to be
> enabled again.
>
> It was done in order to avoid the adapter to generate interrupts when
> it's not being used. Instead of doing that here we just disable the
> interrupt generation in the controller. With a small program test to
> read/write registers in a sensor the speed doubled. Example below with
> write sequences of 16 bytes:
>
> Before:
> 	i2c-transfer-time -w -a 0x40 -x 6 -n 20000 -- 0 0 0xd0 0x07 0 0 0xd0 0x07 0 0 0xd0 0x07 0 0 0xd0 0x07
> 	num_transfers=20000
> 	transfer_time_avg=1032.728500us
>
> After:
> 	i2c-transfer-time -w -a 0x40 -x 6 -n 20000 -- 0 0 0xd0 0x07 0 0 0xd0 0x07 0 0 0xd0 0x07 0 0 0xd0 0x07
> 	num_transfers=20000
> 	transfer_time_avg=470.256050us
>
I gave a test to this patch and saw similar improvements when dumping 
large set of registers from I2C connected audio codec.

echo Y > /sys/kernel/debug/regmap/i2c-10EC5640\:00/cache_bypass
time cat /sys/kernel/debug/regmap/i2c-10EC5640\:00/registers >/dev/null

I checked the runtime PM status of adapter was suspended before running 
above cat command in order to get comparable results.

Before patch time was ~0.55 - ~0.76 s and with patch applied time was 
~0.16 - ~0.25 s.

Hopefully we'll find how to prevent regression on Christian's machines.

-- 
Jarkko