netdev - Re: [WIP] net+mlx4: auto doorbell

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-next>] [day] [month] [year] [list]

Message-ID: <CAADnVQLSxkeOkZMAq=PRdPevSLTC6un-pj_QKD1wNeFZSYY-Sw@mail.gmail.com>
Date:   Tue, 29 Nov 2016 23:28:36 -0800
From:   Alexei Starovoitov <alexei.starovoitov@...il.com>
To:     Eric Dumazet <eric.dumazet@...il.com>
Cc:     Jesper Dangaard Brouer <brouer@...hat.com>,
        Rick Jones <rick.jones2@....com>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        Saeed Mahameed <saeedm@...lanox.com>,
        Tariq Toukan <tariqt@...lanox.com>
Subject: Re: [WIP] net+mlx4: auto doorbell

On Mon, Nov 28, 2016 at 10:58 PM, Eric Dumazet <eric.dumazet@...il.com> wrote:
>  {
> @@ -496,8 +531,13 @@ static bool mlx4_en_process_tx_cq(struct net_device *dev,
>         wmb();
>
>         /* we want to dirty this cache line once */
> -       ACCESS_ONCE(ring->last_nr_txbb) = last_nr_txbb;
> -       ACCESS_ONCE(ring->cons) = ring_cons + txbbs_skipped;
> +       WRITE_ONCE(ring->last_nr_txbb, last_nr_txbb);
> +       ring_cons += txbbs_skipped;
> +       WRITE_ONCE(ring->cons, ring_cons);
> +       WRITE_ONCE(ring->ncons, ring_cons + last_nr_txbb);
> +
> +       if (dev->doorbell_opt)
> +               mlx4_en_xmit_doorbell(dev, ring);
...
> +       /* Doorbell avoidance : We can omit doorbell if we know a TX completion
> +        * will happen shortly.
> +        */
> +       if (send_doorbell &&
> +           dev->doorbell_opt &&
> +           (s32)(READ_ONCE(ring->prod_bell) - READ_ONCE(ring->ncons)) > 0)
> +               send_doorbell = false;

this is awesome idea!
I'm missing though why you need all these read_once/write_once.
It's a tx ring that is already protected by tx queue lock
or completion is happening without grabbing the lock?
Then how do we make sure that we don't race here
and indeed above prod_bell - ncons > 0 condition is safe?
Good description of algorithm would certainly help ;)