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  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:   Sun, 18 Oct 2020 15:28:42 +0200
From:   Maciej Fijalkowski <maciej.fijalkowski@...el.com>
To:     sven.auhagen@...eatech.de
Cc:     anthony.l.nguyen@...el.com, davem@...emloft.net,
        intel-wired-lan@...ts.osuosl.org, netdev@...r.kernel.org,
        nhorman@...hat.com, sassmann@...hat.com,
        sandeep.penigalapati@...el.com, brouer@...hat.com
Subject: Re: [PATCH v2 6/6] igb: avoid transmit queue timeout in xdp path

On Sat, Oct 17, 2020 at 09:12:38AM +0200, sven.auhagen@...eatech.de wrote:
> From: Sven Auhagen <sven.auhagen@...eatech.de>
> 
> Since we share the transmit queue with the slow path,
> it is possible that we run into a transmit queue timeout.
> This will reset the queue.
> This happens under high load when the fast path is using the
> transmit queue pretty much exclusively.

I'm kinda not leaning towards slow/fast path distinction here, IMHO it
would be better to state that transmit queues are shared between network
stack and XDP, but that's just a rant.

> 
> By setting the transmit queues trans_start variable to jiffies
> in the two xdp xmit functions we avoid the timeout.

Probably a few more words of explanation would help here, specifically I
would say that netdev_start_xmit() sets trans_start to jiffies which is
later utilized by dev_watchdog(), so to avoid timeout, let stack know that
XDP xmit happened by bumping the trans_start within XDP Tx routines.

> 
> Signed-off-by: Sven Auhagen <sven.auhagen@...eatech.de>
> ---
>  drivers/net/ethernet/intel/igb/igb_main.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/drivers/net/ethernet/intel/igb/igb_main.c b/drivers/net/ethernet/intel/igb/igb_main.c
> index 55e708f75187..4a082c06f48d 100644
> --- a/drivers/net/ethernet/intel/igb/igb_main.c
> +++ b/drivers/net/ethernet/intel/igb/igb_main.c
> @@ -2916,6 +2916,8 @@ static int igb_xdp_xmit_back(struct igb_adapter *adapter, struct xdp_buff *xdp)
>  
>  	nq = txring_txq(tx_ring);
>  	__netif_tx_lock(nq, cpu);
> +	/* Avoid transmit queue timeout since we share it with the slow path */
> +	nq->trans_start = jiffies;
>  	ret = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
>  	__netif_tx_unlock(nq);
>  
> @@ -2948,6 +2950,9 @@ static int igb_xdp_xmit(struct net_device *dev, int n,
>  	nq = txring_txq(tx_ring);
>  	__netif_tx_lock(nq, cpu);
>  
> +	/* Avoid transmit queue timeout since we share it with the slow path */
> +	nq->trans_start = jiffies;
> +
>  	for (i = 0; i < n; i++) {
>  		struct xdp_frame *xdpf = frames[i];
>  		int err;
> -- 
> 2.20.1
> 

Powered by blists - more mailing lists