svn commit: r341586 - head/sys/dev/mlx5/mlx5_en
John Baldwin
jhb at FreeBSD.org
Mon Apr 29 21:14:23 UTC 2019
On 4/25/19 12:10 AM, Slava Shwartsman wrote:
>
>
> On 17-Apr-19 00:28, John Baldwin wrote:
>> On 4/16/19 8:32 AM, Hans Petter Selasky wrote:
>>> On 4/16/19 4:39 PM, Andrey V. Elsukov wrote:
>>>> On 05.12.2018 17:25, Slava Shwartsman wrote:
>>>>> Author: slavash
>>>>> Date: Wed Dec 5 14:25:03 2018
>>>>> New Revision: 341586
>>>>> URL: https://svnweb.freebsd.org/changeset/base/341586
>>>>>
>>>>> Log:
>>>>> mlx5en: Implement backpressure indication.
>>>>>
>>>>> The backpressure indication is implemented using an unlimited rate type of
>>>>> mbuf send tag. When the upper layers typically the socket layer has obtained such
>>>>> a tag, it can then query the destination driver queue for the current
>>>>> amount of space available in the send queue.
>>>>>
>>>>> A single mbuf send tag may be referenced multiple times and a refcount has been added
>>>>> to the mlx5e_priv structure to track its usage. Because the send tag resides
>>>>> in the mlx5e_channel structure, there is no need to wait for refcounts to reach
>>>>> zero until the mlx4en(4) driver is detached. The channels structure is persistant
>>>>> during the lifetime of the mlx5en(4) driver it belongs to and can so be accessed
>>>>> without any need of synchronization.
>>>>>
>>>>> The mlx5e_snd_tag structure was extended to contain a type field, because there are now
>>>>> two different tag types which end up in the driver which need to be distinguished.
>>>>>
>>>>> Submitted by: hselasky@
>>>>> Approved by: hselasky (mentor)
>>>>> MFC after: 1 week
>>>>> Sponsored by: Mellanox Technologies
>>>>> @@ -587,27 +609,33 @@ mlx5e_xmit(struct ifnet *ifp, struct mbuf *mb)
>>>>> struct mlx5e_sq *sq;
>>>>> int ret;
>>>>>
>>>>> - sq = mlx5e_select_queue(ifp, mb);
>>>>> - if (unlikely(sq == NULL)) {
>>>>> -#ifdef RATELIMIT
>>>>> - /* Check for route change */
>>>>> - if (mb->m_pkthdr.snd_tag != NULL &&
>>>>> - mb->m_pkthdr.snd_tag->ifp != ifp) {
>>>>> + if (mb->m_pkthdr.snd_tag != NULL) {
>>>>> + sq = mlx5e_select_queue_by_send_tag(ifp, mb);
>>>>> + if (unlikely(sq == NULL)) {
>>>>> + /* Check for route change */
>>>>> + if (mb->m_pkthdr.snd_tag->ifp != ifp) {
>>>>> + /* Free mbuf */
>>>>> + m_freem(mb);
>>>>> +
>>>>> + /*
>>>>> + * Tell upper layers about route
>>>>> + * change and to re-transmit this
>>>>> + * packet:
>>>>> + */
>>>>> + return (EAGAIN);
>>>>> + }
>>>>
>>>> Hi,
>>>>
>>>> I just discovered something strange and found that this commit is the
>>>> cause.
>>>> The test system has mlx5en 100G interface. It has two vlans: vlan500 and
>>>> vlan100.
>>>> Via vlan500 it receives some packets flows. Then it routes these packets
>>>> into vlan100.
>>>> But packets are dropped in mlx5e_xmit() with EAGAIN error code.
>>>>
>>>> # dtrace -n 'fbt::ip6_output:return {printf("%d", arg1);}'
>>>> dtrace: description 'fbt::ip6_output:return ' matched 1 probe
>>>> CPU ID FUNCTION:NAME
>>>> 23 54338 ip6_output:return 35
>>>> 16 54338 ip6_output:return 35
>>>> 21 54338 ip6_output:return 35
>>>> 22 54338 ip6_output:return 35
>>>> 24 54338 ip6_output:return 35
>>>> 23 54338 ip6_output:return 35
>>>> 14 54338 ip6_output:return 35
>>>> ^C
>>>>
>>>> # dtrace -n 'fbt::mlx5e_xmit:return {printf("%d", arg1);}'
>>>> dtrace: description 'fbt::mlx5e_xmit:return ' matched 1 probe
>>>> CPU ID FUNCTION:NAME
>>>> 16 69030 mlx5e_xmit:return 35
>>>> 23 69030 mlx5e_xmit:return 35
>>>> 26 69030 mlx5e_xmit:return 35
>>>> 25 69030 mlx5e_xmit:return 35
>>>> 24 69030 mlx5e_xmit:return 35
>>>> 21 69030 mlx5e_xmit:return 35
>>>> 26 69030 mlx5e_xmit:return 35
>>>> ^C
>>>>
>>>> The kernel config is GENERIC.
>>>> 13.0-CURRENT #9 r345758+82f3d57(svn_head)-dirty
>>>>
>>>
>>> Hi,
>>>
>>> This might be a case where rcvif in the mbuf's pktheader is not cleared
>>> before the packet is fed back on the wire.
>>>
>>> John Baldwin is working on the send tags implementation, to eliminate
>>> the EAGAIN handling in the network drivers.
>>
>> I will try to push this branch sooner then since it affects more than just
>> TLS. Part of the change includes a new flag we can use to assert that we
> Thanks John!
>> aren't just getting a stale rcvif (though there are also now assertions in
>> ip_output that should catch this case I think).
>>
>
>
>
> Hi Andrey,
>
> Yes, we were able to reproduce this issue in house. If you don't mind, I
> prefer to wait for John's update - where he eliminates the EAGAIN
> handling in the network drivers.
I have rebased the branch for this, but for now it will just panic sooner
I believe by tripping an assertion. Can you grab the diff (or just the branch)
from the 'send_tags' branch at github/bsdjhb/freebsd and reproduce under a
kernel with INVARIANTS? I think we will have to explicitly clear the 'rcvif'
pointer somewhere, but I want to see what the stack trace looks like so I can
think about the "right" place to clear it.
--
John Baldwin
More information about the svn-src-all
mailing list