[Intel-wired-lan] [PATCH 2/2] Trigger proper interrupts in igc_xsk_wakeup

[Jeff Evanson]

Hi Vinicius. Thank you for the reply.

The scenario only happens when the transmit queue interrupt is mapped to a different interrupt from the receive queue. In the case where XDP_WAKEUP_TX is set in the flags argument, the previous code would only trigger the interrupt for the receive queue, causing the transmit queue's napi_struct to never be scheduled.

In the scenario where XDP_WAKEUP_TX and XDP_WAKEUP_RX are both set in flags, the receive interrupt is always triggered and the transmit interrupt is only triggered when the transmit q_vector differs from the receive q_vector.

Regards,
Jeff Evanson

-----Original Message-----
From: Vinicius Costa Gomes <vinicius.gomes at intel.com> 
Sent: Monday, April 18, 2022 11:45 AM
To: Jeff Evanson <jeff.evanson at gmail.com>; Jesse Brandeburg <jesse.brandeburg at intel.com>; Tony Nguyen <anthony.l.nguyen at intel.com>; David S. Miller <davem at davemloft.net>; Jakub Kicinski <kuba at kernel.org>; intel-wired-lan at lists.osuosl.org; netdev at vger.kernel.org; linux-kernel at vger.kernel.org
Cc: Jeff Evanson <Jeff.Evanson at qsc.com>; jeff.evanson at gmail.com
Subject: Re: [PATCH 2/2] Trigger proper interrupts in igc_xsk_wakeup

-External-

Jeff Evanson <jeff.evanson at gmail.com> writes:

> in igc_xsk_wakeup, trigger the proper interrupt based on whether flags 
> contains XDP_WAKEUP_RX and/or XDP_WAKEUP_TX
>
> Signed-off-by: Jeff Evanson <jeff.evanson at qsc.com>
> ---
>  drivers/net/ethernet/intel/igc/igc_main.c | 36 
> +++++++++++++++++------
>  1 file changed, 27 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/net/ethernet/intel/igc/igc_main.c 
> b/drivers/net/ethernet/intel/igc/igc_main.c
> index a36a18c84aeb..d706de95dc06 100644
> --- a/drivers/net/ethernet/intel/igc/igc_main.c
> +++ b/drivers/net/ethernet/intel/igc/igc_main.c
> @@ -6073,7 +6073,7 @@ static void igc_trigger_rxtxq_interrupt(struct 
> igc_adapter *adapter,  int igc_xsk_wakeup(struct net_device *dev, u32 
> queue_id, u32 flags)  {
>  	struct igc_adapter *adapter = netdev_priv(dev);
> -	struct igc_q_vector *q_vector;
> +	struct igc_q_vector *txq_vector = 0, *rxq_vector = 0;
>  	struct igc_ring *ring;
>  
>  	if (test_bit(__IGC_DOWN, &adapter->state)) @@ -6082,17 +6082,35 @@ 
> int igc_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
>  	if (!igc_xdp_is_enabled(adapter))
>  		return -ENXIO;
>  
> -	if (queue_id >= adapter->num_rx_queues)
> -		return -EINVAL;
> +	if (flags & XDP_WAKEUP_RX) {
> +		if (queue_id >= adapter->num_rx_queues)
> +			return -EINVAL;
>  
> -	ring = adapter->rx_ring[queue_id];
> +		ring = adapter->rx_ring[queue_id];
> +		if (!ring->xsk_pool)
> +			return -ENXIO;
>  
> -	if (!ring->xsk_pool)
> -		return -ENXIO;
> +		rxq_vector = ring->q_vector;
> +	}
> +
> +	if (flags & XDP_WAKEUP_TX) {
> +		if (queue_id >= adapter->num_tx_queues)
> +			return -EINVAL;
> +
> +		ring = adapter->tx_ring[queue_id];
> +		if (!ring->xsk_pool)
> +			return -ENXIO;
> +
> +		txq_vector = ring->q_vector;
> +	}
> +
> +	if (rxq_vector &&
> +	    !napi_if_scheduled_mark_missed(&rxq_vector->napi))
> +		igc_trigger_rxtxq_interrupt(adapter, rxq_vector);
>  
> -	q_vector = adapter->q_vector[queue_id];
> -	if (!napi_if_scheduled_mark_missed(&q_vector->napi))
> -		igc_trigger_rxtxq_interrupt(adapter, q_vector);
> +	if (txq_vector && txq_vector != rxq_vector &&
> +	    !napi_if_scheduled_mark_missed(&txq_vector->napi))
> +		igc_trigger_rxtxq_interrupt(adapter, txq_vector);

Two things:
 - My imagination was not able to produce a scenario where this commit  would solve any problems. Can you explain better the case where the  current code would cause the wrong interrupt to be generated (or miss  generating an interrupt)? (this should be in the commit message)
 - I think that with this patch applied, there would cases (both TX and  RX flags set) that we cause two writes into the EICS register. That  could cause unnecessary wakeups.

>  
>  	return 0;
>  }
> --
> 2.17.1
>

Cheers,
-- 
Vinicius