i.MX6 quad SATA performance getting worse.

[ju-hakpark]

Hi NXP support,

I was requested from one of my customer to check SATA HDD access speed between kernel version 3.14.28 and 3.14.52.

So I built 3.14.28, 3.14.52 and 4.1.15 and tested all then found HDD access speed is significantly getting worse than 3.14.28 kernel.

I can get the result over 100MB/s access speed both read and write on 3.14.28.

But, I got disappointable access speed with 3.14.52 and 4.1.15 that is less than 35MB/s for writing, 60MB/s for reading.

It is less than 1 of 3 speed for writing and half speed for reading when it compares to 3.14.28.

They expect to get HDD access speed as same as 3.14.28 that is "Read over 100MB/s and Write over 80MB/s" with SATA-2 HDD.

I have 2 questions regarding this issue.

1. Could you please let me know what is the root cause of this issue and how to fix this issue?

2. Can you guarantee whether it is no problem if I use 'ata related' codes in driver/ata directory from 3.14.28 with 3.14.52 kernel?

Please check.

Thanks in advance.

[igorpadykov]

Hi Ju-hak Park

please try to measure performance with Demo images from i.MX6

official product page using nxp i.MX6Q Sabre SD/AI reference board

http://www.nxp.com/products/microcontrollers-and-processors/arm-processors/i.mx-applications-process...

Please note that performance issues on custom board greatly depends on use case

and may be supported using NXP Professional Services:

http://www.nxp.com/support/nxp-professional-services:PROFESSIONAL-SERVICE
Pro-Support contact www.nxp.com/prosupport

================================================================
Best regards
igor
-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------

[ju-hakpark]

Hi Igor,

I am using SABRE-AI with i.MX6Quad and build images without any modification.

Please check my questions.

[igorpadykov]

Hi Ju-hak Park

reason is errata ERR009598 workaround (MLK-11444 ata) has been implemented from the Linux BSP 3.14.28_1.0.0_ga.
Note ERR009598 is occurred at a very low rate and difficult to reproduce. If user can't
encounter this issue, they can ignore this errata and workaround to maintain a better performance of SATA disk.

Best regards
igor

[joerg_albert]

Hi Igor,

I currently scan through all SATA Errata of the iMX.6 as we have some problems with this interface on our custom board. We run some customized kernel based on Linux-stable v4.14.113. Was the workaround for ERR009598 (aka MLK-11444 ata) ever distributed upstream? I cannot find any of the changes for MLK-11444 (https://git.congatec.com/arm/imx6_kernel_3.14/commit/3d5ca5ece0dbc2ffd351470944744bd947298c3d) in our kernel sources.

Best Regards,

Joerg