[PATCH net-next] ice, xsk: Move Rx alloction out of while-loop

[PATCH net-next] ice, xsk: Move Rx alloction out of while-loop

[Björn Töpel]

From: Björn Töpel <bjorn.topel@intel.com>

Instead of trying to allocate for each packet, move it outside the
while loop and try to allocate once every NAPI loop.

This change boosts the xdpsock rxdrop scenario with 15% more
packets-per-second.

Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
---
 drivers/net/ethernet/intel/ice/ice_xsk.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/intel/ice/ice_xsk.c b/drivers/net/ethernet/intel/ice/ice_xsk.c
index 797886524054..39757b4cf8f4 100644
--- a/drivers/net/ethernet/intel/ice/ice_xsk.c
+++ b/drivers/net/ethernet/intel/ice/ice_xsk.c
@@ -570,12 +570,6 @@ int ice_clean_rx_irq_zc(struct ice_ring *rx_ring, int budget)
 		u16 vlan_tag = 0;
 		u8 rx_ptype;
 
-		if (cleaned_count >= ICE_RX_BUF_WRITE) {
-			failure |= ice_alloc_rx_bufs_zc(rx_ring,
-							cleaned_count);
-			cleaned_count = 0;
-		}
-
 		rx_desc = ICE_RX_DESC(rx_ring, rx_ring->next_to_clean);
 
 		stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_DD_S);
@@ -642,6 +636,9 @@ int ice_clean_rx_irq_zc(struct ice_ring *rx_ring, int budget)
 		ice_receive_skb(rx_ring, skb, vlan_tag);
 	}
 
+	if (cleaned_count >= ICE_RX_BUF_WRITE)
+		failure = !ice_alloc_rx_bufs_zc(rx_ring, cleaned_count);
+
 	ice_finalize_xdp_rx(rx_ring, xdp_xmit);
 	ice_update_rx_ring_stats(rx_ring, total_rx_packets, total_rx_bytes);
 

base-commit: a7105e3472bf6bb3099d1293ea7d70e7783aa582
-- 
2.27.0

Re: [PATCH net-next] ice, xsk: Move Rx alloction out of while-loop

[Maciej Fijalkowski]

On Thu, Dec 10, 2020 at 01:19:15PM +0100, Björn Töpel wrote:
> From: Björn Töpel <bjorn.topel@intel.com>
> 
> Instead of trying to allocate for each packet, move it outside the
> while loop and try to allocate once every NAPI loop.

To rectify above, it wasn't for each packet but per ICE_RX_BUF_WRITE
cleaned frames (16).

You also have a typo in subject (alloction).

Is spinning a v2 worth it?

> 
> This change boosts the xdpsock rxdrop scenario with 15% more
> packets-per-second.
> 
> Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
> Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
> ---
>  drivers/net/ethernet/intel/ice/ice_xsk.c | 9 +++------
>  1 file changed, 3 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ice/ice_xsk.c b/drivers/net/ethernet/intel/ice/ice_xsk.c
> index 797886524054..39757b4cf8f4 100644
> --- a/drivers/net/ethernet/intel/ice/ice_xsk.c
> +++ b/drivers/net/ethernet/intel/ice/ice_xsk.c
> @@ -570,12 +570,6 @@ int ice_clean_rx_irq_zc(struct ice_ring *rx_ring, int budget)
>  		u16 vlan_tag = 0;
>  		u8 rx_ptype;
>  
> -		if (cleaned_count >= ICE_RX_BUF_WRITE) {
> -			failure |= ice_alloc_rx_bufs_zc(rx_ring,
> -							cleaned_count);
> -			cleaned_count = 0;
> -		}
> -
>  		rx_desc = ICE_RX_DESC(rx_ring, rx_ring->next_to_clean);
>  
>  		stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_DD_S);
> @@ -642,6 +636,9 @@ int ice_clean_rx_irq_zc(struct ice_ring *rx_ring, int budget)
>  		ice_receive_skb(rx_ring, skb, vlan_tag);
>  	}
>  
> +	if (cleaned_count >= ICE_RX_BUF_WRITE)
> +		failure = !ice_alloc_rx_bufs_zc(rx_ring, cleaned_count);
> +
>  	ice_finalize_xdp_rx(rx_ring, xdp_xmit);
>  	ice_update_rx_ring_stats(rx_ring, total_rx_packets, total_rx_bytes);
>  
> 
> base-commit: a7105e3472bf6bb3099d1293ea7d70e7783aa582
> -- 
> 2.27.0
> 

Re: [PATCH net-next] ice, xsk: Move Rx alloction out of while-loop

[Björn Töpel]

On Thu, 10 Dec 2020 at 17:20, Maciej Fijalkowski
<maciej.fijalkowski@intel.com> wrote:
>
> On Thu, Dec 10, 2020 at 01:19:15PM +0100, Björn Töpel wrote:
> > From: Björn Töpel <bjorn.topel@intel.com>
> >
> > Instead of trying to allocate for each packet, move it outside the
> > while loop and try to allocate once every NAPI loop.
>
> To rectify above, it wasn't for each packet but per ICE_RX_BUF_WRITE
> cleaned frames (16).
>
> You also have a typo in subject (alloction).
>
> Is spinning a v2 worth it?
>

Sure! I'll rephrase, and resend!

Thanks,
Björn