netdev - Re: [PATCH net-next] virtio-net: invoke zerocopy callback on xmit path if no tx napi

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 1 Sep 2017 12:17:08 -0400
From:   Willem de Bruijn <willemdebruijn.kernel@...il.com>
To:     Jason Wang <jasowang@...hat.com>
Cc:     "Michael S. Tsirkin" <mst@...hat.com>,
        Koichiro Den <den@...ipeden.com>,
        virtualization@...ts.linux-foundation.org,
        Network Development <netdev@...r.kernel.org>
Subject: Re: [PATCH net-next] virtio-net: invoke zerocopy callback on xmit
 path if no tx napi

>>> This is not a 50/50 split, which impliesTw that some packets from the
>>> large
>>> packet flow are still converted to copying. Without the change the rate
>>> without queue was 80k zerocopy vs 80k copy, so this choice of
>>> (vq->num >> 2) appears too conservative.
>>>
>>> However, testing with (vq->num >> 1) was not as effective at mitigating
>>> stalls. I did not save that data, unfortunately. Can run more tests on
>>> fine
>>> tuning this variable, if the idea sounds good.
>>
>>
>> Looks like there're still two cases were left:
>
> To be clear, this patch is not intended to fix all issues. It is a small
> improvement to avoid HoL blocking due to queued zerocopy skbs.
>
> The trade-off is that reverting to copying in these cases increases
> cycle cost. I think that that is a trade-off worth making compared to
> the alternative drop in throughput. It probably would be good to be
> able to measure this without kernel instrumentation: export
> counters similar to net->tx_zcopy_err and net->tx_packets (though
> without reset to zero, as in vhost_net_tx_packet).
>
>> 1) sndbuf is not INT_MAX
>
> You mean the case where the device stalls, later zerocopy notifications
> are queued, but these are never cleaned in free_old_xmit_skbs,
> because it requires a start_xmit and by now the (only) socket is out of
> descriptors?

Typo, sorry. I meant out of sndbuf.

> A watchdog would help somewhat. With tx-napi, this case cannot occur,
> either, as free_old_xmit_skbs no longer depends on a call to start_xmit.
>
>> 2) tx napi is used for virtio-net
>
> I am not aware of any issue specific to the use of tx-napi?
>
>> 1) could be a corner case, and for 2) what your suggest here may not solve
>> the issue since it still do in order completion.
>
> Somewhat tangential, but it might also help to break the in-order
> completion processing in vhost_zerocopy_signal_used. Complete
> all descriptors between done_idx and upend_idx. done_idx should
> then only be forward to the oldest still not-completed descriptor.
>
> In the test I ran, where the oldest descriptors are held in a queue and
> all newer ones are tail-dropped, this would avoid blocking a full ring
> of completions, when only a small number (or 1) is actually delayed.
>
> Dynamic switching between copy and zerocopy using zcopy_used
> already returns completions out-of-order, so this is not a huge leap.