netdev - Re: [PATCH net-next v4 01/10] tsnep: Use spin_lock

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:   Tue, 10 Jan 2023 20:44:33 +0100
From:   Gerhard Engleder <gerhard@...leder-embedded.com>
To:     Alexander H Duyck <alexander.duyck@...il.com>,
        netdev@...r.kernel.org
Cc:     davem@...emloft.net, kuba@...nel.org, edumazet@...gle.com,
        pabeni@...hat.com
Subject: Re: [PATCH net-next v4 01/10] tsnep: Use spin_lock_bh for TX

On 10.01.23 16:49, Alexander H Duyck wrote:
> On Mon, 2023-01-09 at 20:15 +0100, Gerhard Engleder wrote:
>> TX processing is done only within BH context. Therefore, _irqsafe
>> variant is not necessary.
>>
>> Signed-off-by: Gerhard Engleder <gerhard@...leder-embedded.com>
> 
> Rather than reducing the context of this why not drop it completely? It
> doesn't look like you are running with the NETIF_F_LLTX flag set so
> from what I can tell it looks like you are taking the Tx lock in the
> xmit path. So Tx queues are protected with the Tx queue lock at the
> netdev level via the HARD_TX_LOCK macro.
> 
> Since it is already being used in the Tx path to protect multiple
> access you could probably just look at getting rid of it entirely.
> 
> The only caveat you would need to watch out for is a race between the
> cleaning and transmitting which can be addressed via a few barriers
> like what was done in the intel drivers via something like the
> __ixgbe_maybe_stop_tx function and the logic to wake the queue in the
> clean function.

I know these barriers in the intel drivers. But I chose to use a lock
like the microchip driver to be on the safe side rather than having a
fully optimized driver.

> Alternatively if you really feel you need this in the non-xmit path
> functions you could just drop the lock and instead use __netif_tx_lock
> for those spots that are accessing the queue outside the normal
> transmit path.

Ok, I will work on that. One of your suggestions will be done.

Gerhard