ethernet and usb host storage very slow on android-omap-panda-3.0

[m-ric]

Hi,

I run AOSP-4.0.4 on a pandaboard rev A3. I got linux kernel from
remotes/android-omap/android-omap-panda-3.0 branch, 52f4764 commit
"Revert gpu: pvr: Revert to 1.8@550175".

With this software:
1. Ethernet bandwidth is very low, barely usable for file transfers.
TX gets stuck, I measured 20KBps. RX is quite slow too.
2. USB storage on USB host port suffers same limitations. I measured
~100KBps.

Problem is known though, fedora- and ubuntu guys saw it:
http://article.gmane.org/gmane.linux.redhat.fedora.arm/1266
https://bugs.launchpad.net/linux-linaro/+bug/709245

My kernel embeds Ming Lei's barrier patch in ehci-q.c:
usb: ehci: make HC see up-to-date qh/qtd descriptor ASAP
and Will Deacon's ARM errata CONFIG_PL310_ERRATA_769419:
ARM: 7161/1: errata: no automatic store buffer drain
I checked that asm on this latter commit is fine (dsb instr.).

Still above bugs are still present. For fedora and ubuntu guys though,
above fixes have solved the problem.
I am reluctant to change my (entire) kernel because everything else
that I need works fine. I am in a maturation phase, not in a
development phase anymore.

Does anyone experience that on pandaboard and aosp?
Did you find any workaround?

I have been on this bug for 4 weeks now. I am kind of desperate right
now.

[Sagar Shinde]

Hi,
We are facing the same issue.
Did you get any solution for this?

Thanks in advance.
BR
Sagar
sangi.hs_gmail.com

[m-ric]

Good to know someone shares the problem. I posted on pretty much all the mailng list I know, without any "me too" effect.
I have not fixed this issue yet. I know it is related to USB host driver (ehci-omap.c) because USB mass storage - panda being the host, usb stick being the device - also has transfer issue (Write). An interrupt is not fired properly in TX,  RX is fine. I will resume my investigation not before the end of october.

Post in here if you make any progress in the meantime.

Thanks,
Emeric

[Satish Patel]

Check the dirty ration while USB transfer. It might getting change..

--
unsubscribe: android-kerne...@googlegroups.com
website: http://groups.google.com/group/android-kernel

--
Regards,
satish patel

[m-ric]

"might getting change"? What is that supposed to mean?

Emeric

[Chyifeng Ding]

Hi all,

I wonder if you guys have solution for it already? If so, please post it.

I ran AOSP Android 4.3_r2.1 on pandaboard ES rev B2 and it has similar problem. I got 10KB/s for TX. RX is ok, which is around 10MB/s.

The kernel is

Linux version 3.0.31-gcb5fc50 (j...@jqueru.mtv.corp.google.com) (gcc version 4.6.x-google 20120106 (prerelease) (GCC) ) #1 SMP PREEMPT Mon Dec 3 18:52:22 PST 2012

Our project requires large amount of data transfer.

Please help!

Thanks,

 

Chyifeng

[Enzeneer]

Any chance it could be related to this: http://thread.gmane.org/gmane.linux.ports.arm.kernel/271108 ?

[Chyifeng Ding]

Hi Enzeneer,

No, it is different issue. The link you mentioned is related to SATA and SMP. I tried single CPU and nothing changes. It has not SATA either. So far I can only say that TX DMA is not triggered in this case. Moving USB mouse helps increasing TX data rate.

Thanks for trying.

Chyifeng

On Thursday, July 12, 2012 12:26:32 PM UTC-7, m-ric wrote:

[m-ric]

Hi Chyifeng,

I made no real progress on this topic.

I worked around it by using an USB/ethernet adapter which worked fine.
I wish you more luck with your investigation.

Emeric

[Chyifeng Ding]

Hi Emeric,

Thanks for replying. It is sad that this issue was not resolved. At least I know now there is a workaround by connecting external USB-ethernet.

Chyifeng

On Thursday, July 12, 2012 12:26:32 PM UTC-7, m-ric wrote: