linaro-release team mailing list archive

[Ming Lei <709245@xxxxxxxxxxxxxxxxxx>]

Hi,

I think that the problem is caused by ehci driver instead of ARM SMP
scheduler, and I
have verified that patch below can fix the problem. And the patch has
been posted
on usb/arm/omap mail list for discussion.

But I am wondering that why no such problem after passing  'nosmp' to
kernel.

>From d85a08714ed23ec8688013b464dc90c6386db0d8 Mon Sep 17 00:00:00 2001
From: Ming Lei <ming.lei@xxxxxxxxxxxxx>
Date: Sat, 27 Aug 2011 22:29:15 +0800
Subject: [PATCH] usb: ehci: fix update qtd->token in qh_append_tds

This patch fixs one performance bug on ARM Cortex A9 dual core platform,
which has been reported on quite a few ARM machines(OMAP4, Tegra 2,
snowball...),
see details from link of https://bugs.launchpad.net/bugs/709245.

In fact, one mb() on ARM is enough to flush L2 cache, but
'dummy->hw_token = token;' after mb() is added just for obeying
correct mb() usage.

The patch has been tested ok on OMAP4 panda A1 board, the performance
of 'dd' over usb mass storage can be increased from 4~5MB/sec to
14~16MB/sec after applying this patch.

Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxxxxx>
---
 drivers/usb/host/ehci-q.c |   14 ++++++++++++++
 1 files changed, 14 insertions(+), 0 deletions(-)

diff --git a/drivers/usb/host/ehci-q.c b/drivers/usb/host/ehci-q.c
index 0917e3a..65b5021 100644
--- a/drivers/usb/host/ehci-q.c
+++ b/drivers/usb/host/ehci-q.c
@@ -1082,6 +1082,20 @@ static struct ehci_qh *qh_append_tds (
 			wmb ();
 			dummy->hw_token = token;

+			/* The mb() below is added to make sure that
+			 * 'token' can be writen into qtd, so that ehci
+			 * HC can see the up-to-date qtd descriptor. On
+			 * some archs(at least on ARM Cortex A9 dual core),
+			 * writing into coherenet memory doesn't mean the
+			 * value written can reach physical memory
+			 * immediately, and the value may be buffered
+			 * inside L2 cache. 'dummy->hw_token = token;'
+			 * after mb() is added for obeying correct mb()
+			 * usage.
+			 * */
+			mb();
+			token = dummy->hw_token;
+
 			urb->hcpriv = qh_get (qh);
 		}
 	}
-- 
1.7.4.1

-- 
You received this bug notification because you are a member of Linaro
Release Team, which is subscribed to the bug report.
https://bugs.launchpad.net/bugs/709245

Title:
  ARM SMP scheduler performance bug

Status in Linaro Ubuntu Evaluation Builds:
  Confirmed
Status in Linaro Linux:
  Confirmed
Status in OEM Priority Project:
  New
Status in “linux-ti-omap4” package in Ubuntu:
  Confirmed
Status in “linux-ti-omap4” source package in Maverick:
  Confirmed
Status in “linux-ti-omap4” source package in Natty:
  Confirmed
Status in “linux-ti-omap4” source package in Oneiric:
  Confirmed

Bug description:
  Original Bug name: "panda: USB disk IO slow"

  This bug effects ARM Cortex A9 cores, snowball, nvidia, OMAP 4, and
  other Cortex A9 processors.  Problem is in Fedora ARM builds also so
  not limited to Ubuntu.

  My Panda's USB seems to be significantly slower than a Beagle C4.

  hdparm shows buffered reads as ~12MB/s on the Panda, and about ~20-25MB/s on a Beagle C4 from the same
  external Lacie USB disk.

  Kernel is 2.6.37-1002-linaro-omap

  Disk shows as:

  [    5.170440] scsi 0:0:0:0: Direct-Access     LaCie    d2 quadra             PQ: 0 ANSI: 4
  [    5.172546] sd 0:0:0:0: Attached scsi generic sg0 type 0
  [    5.175415] sd 0:0:0:0: [sda] 976773168 512-byte logical blocks: (500 GB/465 GiB)

  The board is otherwise idle during the test.

  Doing perf_2.6.37-12 record -a dd if=/dev/sda of=/dev/null bs=4096
  count=100000

  shows :
      81.41%         swapper  [kernel.kallsyms]     [k] default_idle
       6.33%              dd  [kernel.kallsyms]     [k] __copy_to_user
       0.94%         swapper  [kernel.kallsyms]     [k] cpu_idle
       0.51%              dd  [kernel.kallsyms]     [k] __make_request
       0.51%  perf_2.6.37-12  [kernel.kallsyms]     [k] __copy_from_user

  which suggests it's not CPU constrained.

  Dave

To manage notifications about this bug go to:
https://bugs.launchpad.net/linaro-ubuntu/+bug/709245/+subscriptions