On 09/01/2015 02:21 AM, yzhu1 wrote:
On 09/01/2015 04:23 PM, Daniel Borkmann wrote:
On 09/01/2015 09:10 AM, yzhu1 wrote:
On 09/01/2015 03:00 PM, David Miller wrote:
From: yzhu1 <yanjun....@windriver.com>
Date: Tue, 1 Sep 2015 14:46:38 +0800

After I applied this patch, the skb->xmit_more is not always zero.
There have been thousands upon thousands of commits since that
change.

You should be testing the tree as it currently stands, to see
if xmit_more behaves correctly or not.

If xmit_more were incorrectly set to 1 in the current tree, it
would stall the TX queue of the networking device and we would
be seeing lots of reports of this.

Thanks for your reply.
Yes. After running for several days, the following messages will appear.

Your below trace says 3.14.29ltsi-WR7.0.0.0 ...

As Dave said, please retest with something up to date, like 4.2 kernel,
or latest -net git tree.

Besides, the *upstream* xmit_more changes first went into 3.18 ...
nearest git describe is at:

  $ git describe 0b725a2ca61bedc33a2a63d0451d528b268cf975
  v3.17-rc1-251-g0b725a2

So, that only tells me, that you are reporting a possible bug based on
some non-upstream kernel ... ? Thus, it's not even possible to verify
if the actual backport was correct ?

Sorry. There is something wrong with backporting this patch.

Thanks for your help.

Zhu Yanjun

You may very well be missing the code that forced the tail to write if the Tx descriptor ring was full. Double check your igb driver code and compare it to the upstream kernel. You should double check and verify that you backported commit 6f19e12f62306 "igb: flush when in xmit_more mode and under descriptor pressure".

You need to have both checked before you skip writing the next descriptor to the ring. The original code only checked one and this can result in Tx hangs if the ring fills without ever notifying the hardware.

- Alex
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to