Re: Mellanox ConnectX-5 crashes and mbuf leak

[Martin Weiser <martin.weiser@allegro-packets.com>]

Hi Yongseok,

I can confirm that this patch fixes the crashes and freezing in my tests
so far.

We still see an issue that once the mbufs run low and reference counts
are used as well as freeing of mbufs in processing lcores happens we
suddenly lose a large amount of mbufs that will never return to the
pool. But I can also reproduce this with ixgbe so this is not specific
to the mlx5 driver but rather an issue of the current dpdk-net-next
state. I will write up a separate mail with details how to reproduce this.

Thank you for your support!

Best regards,
Martin


On 08.10.17 00:19, Yongseok Koh wrote:
>> On Oct 6, 2017, at 3:30 PM, Yongseok Koh <yskoh@mellanox.com> wrote:
>>
>> Hi, Martin
>>
>> Even though I had done quite serious tests before sending out the patch,
>> I figured out deadlock could happen if the Rx queue size is smaller. It is 128
>> by default in testpmd while I usually use 256.
>>
>> I've fixed the bug and submitted a new patch [1], which actually reverts the
>> previous patch. So, you can apply the attached with disregarding the old one.
>>
>> And I have also done extensive tests for this new patch but please let me know
>> your test results.
>>
>> [1]
>> "net/mlx5: fix deadlock due to buffered slots in Rx SW ring"
>> at http://dpdk.org/dev/patchwork/patch/29847
> Hi Martin
>
> I've submitted v2 of the patch [1]. I just replaced vector insns with regular
> statements.  This is just for ease of maintenance because I'm about to add
> vectorized PMD for ARM NEON.  In terms of functionality and performance it is
> identical.
>
> Please proceed your testing with this and let me know the result.
>
> [1]
> [dpdk-dev,v2] net/mlx5: fix deadlock due to buffered slots in Rx SW ring
> , which is at http://dpdk.org/dev/patchwork/patch/29879/
>
> Thanks,
> Yongseok
>