Re: [PATCH 0/5] arm: shmobile: Lager Add USB and SATA support

[Khiem Nguyen <khiem.nguyen.xt@renesas.com>]

Hi Hiep-san,

Thanks for your test result.

On 9/11/2014 6:35 PM, カオ　ミン　ヒェップ wrote:
> Hi Khiem-san
> 
> Sorry for late replying!
> 
> On 09/10/2014 06:32 PM, Khiem Nguyen wrote:
>> Hi Hiep-san, Sakato-san,
>>
>> Sorry for interruption.
>>
>> Could you also confirm performance of SATA driver
>> in case that CPUFreq uses performance governor ?
>>
>> If SATA performance is degraded in that case,
>> I guess we still need countermeasure.
>> (because CPUFreq will definitely be enabled in multi-platform defconfig.)
> Here is the results of SATA driver writing speed in case CPUFreq uses governor:
> Data size: 350MB,
> root@linaro-nano:~# uname -a
> Linux linaro-nano 3.17.0-rc2 #3 SMP Thu Sep 11 17:15:17 JST 2014 armv7l armv7l armv7l GNU/Lix
> root@linaro-nano:~# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
> ondemand
> root@linaro-nano:~# cat /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
> ondemand
> root@linaro-nano:~# cat /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor
> ondemand
> root@linaro-nano:~# cat /sys/devices/system/cpu/cpu3/cpufreq/scaling_governor
> ondemand
> 
> # mount -t tmpfs -o size@0m tmpfs /tmp
> root@linaro-nano:~# dd if=/dev/urandom of=/tmp/file-350mb bs=1M count50
> 
> 350+0 records in
> 350+0 records out
> 367001600 bytes (367 MB) copied, 88.5655 s, 4.1 MB/s
> root@linaro-nano:~# mount /dev/sda1 /mnt/sata/
> kjournald starting.  Commit interval 5 seconds
> EXT3-fs (sda1): using internal journal
> EXT3-fs (sda1): mounted filesystem with ordered data mode
>  oot@linaro-nano:~# time cp /tmp/file-350mb /mnt/sata/; time umount /mnt/sata/
> 
> real    0m6.665s
> user    0m0.020s
> sys     0m5.690s
> 
> real    0m1.243s
> user    0m0.010s
> sys     0m0.660s
> 
> Total: 7.9s
> Speed: 46.45MB/s
> =========================
> 
> root@linaro-nano:~# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
> performance
> root@linaro-nano:~# cat /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
> performance
> root@linaro-nano:~# cat /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor
> performance
> root@linaro-nano:~# cat /sys/devices/system/cpu/cpu3/cpufreq/scaling_governor
> performance
> # time cp /tmp/file-350mb /mnt/sata/; time umount /mnt/sata/
> 
> real    0m5.331s
> user    0m0.010s
> sys     0m5.250s
> 
> real    0m1.084s
> user    0m0.030s
> sys     0m0.500s
> 
> Total: 6.3s
> Speed: 58.3MB/s

It seems SATA performance (write speed) is degraded after CPUFreq is enabled
(even performance governor is selected).

Base on your test result so far:
T1 Upstream-v3.16-rc5: 66 MB/s
T2 v3.17-rc2 (CPUFreq disabled): ~ 66 MB/s
T3 v3.17-rc2 (CPUFreq enable / on-demand governor): 42.61MB/s
T4 v3.17-rc2 (CPUFreq disabled / performance governor ): 58.3MB/s

The diff between T2 and T4 is ~8 MB/s (or 8% of 66 MB/s).
I guess we should try to find the cause of this regression.
(I wonder similar phenomenon occurs in backport version ...)

(I added Inami-san, CPUFreq developer for his info.)
> 
> 
> Thank you!
> Cao Minh Hiep.

-- 
Best regards,
KHIEM Nguyen