| Message ID | 20240214151204.2976-1-paul.barker.ct@bp.renesas.com (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | ed4adc07207d9165a4b3b36199231a22e9f51a55 |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net,v3] net: ravb: Count packets instead of descriptors in GbEth RX path | expand |
On 2/14/24 6:12 PM, Paul Barker wrote: > The units of "work done" in the RX path should be packets instead of > descriptors, as large packets can be spread over multiple descriptors. > > Fixes: 1c59eb678cbd ("ravb: Fillup ravb_rx_gbeth() stub") > Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com> Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru> [...] MBR, Sergey
On Wed, 14 Feb 2024 15:12:04 +0000 Paul Barker wrote: > The units of "work done" in the RX path should be packets instead of > descriptors, as large packets can be spread over multiple descriptors. FWIW one of y'all may also want to look into processing Tx completions before Rx. Tx completions can free memory which Rx can then consume. More efficient. Not to mention netconsole being able to reap Tx with budget of 0.
On 15/02/2024 16:03, Jakub Kicinski wrote: > On Wed, 14 Feb 2024 15:12:04 +0000 Paul Barker wrote: >> The units of "work done" in the RX path should be packets instead of >> descriptors, as large packets can be spread over multiple descriptors. > > FWIW one of y'all may also want to look into processing Tx completions > before Rx. Tx completions can free memory which Rx can then consume. > More efficient. Not to mention netconsole being able to reap Tx with > budget of 0. Thanks for the recommendation - I'll make this change as part of v3 of the series "Improve GbEth performance on Renesas RZ/G2L and related SoCs" [1] once we've completed gPTP testing. [1]: https://lore.kernel.org/netdev/20240206091909.3191-1-paul.barker.ct@bp.renesas.com/
Hello: This patch was applied to netdev/net.git (main) by Jakub Kicinski <kuba@kernel.org>: On Wed, 14 Feb 2024 15:12:04 +0000 you wrote: > The units of "work done" in the RX path should be packets instead of > descriptors, as large packets can be spread over multiple descriptors. > > Fixes: 1c59eb678cbd ("ravb: Fillup ravb_rx_gbeth() stub") > Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com> > --- > This patch has been broken out from my previous series "Improve GbEth > performance on Renesas RZ/G2L and related SoCs" and submitted as a > bugfix as requested by Sergey. I've labeled it as 'v3' so the ordering > is clear. Remaining patches from the series will follow once we've done > gPTP testing. > > [...] Here is the summary with links: - [net,v3] net: ravb: Count packets instead of descriptors in GbEth RX path https://git.kernel.org/netdev/net/c/ed4adc07207d You are awesome, thank you!
diff --git a/drivers/net/ethernet/renesas/ravb_main.c b/drivers/net/ethernet/renesas/ravb_main.c index 0e3731f50fc2..f7566cfa45ca 100644 --- a/drivers/net/ethernet/renesas/ravb_main.c +++ b/drivers/net/ethernet/renesas/ravb_main.c @@ -772,29 +772,25 @@ static bool ravb_rx_gbeth(struct net_device *ndev, int *quota, int q) struct ravb_rx_desc *desc; struct sk_buff *skb; dma_addr_t dma_addr; + int rx_packets = 0; u8 desc_status; - int boguscnt; u16 pkt_len; u8 die_dt; int entry; int limit; + int i; entry = priv->cur_rx[q] % priv->num_rx_ring[q]; - boguscnt = priv->dirty_rx[q] + priv->num_rx_ring[q] - priv->cur_rx[q]; + limit = priv->dirty_rx[q] + priv->num_rx_ring[q] - priv->cur_rx[q]; stats = &priv->stats[q]; - boguscnt = min(boguscnt, *quota); - limit = boguscnt; desc = &priv->gbeth_rx_ring[entry]; - while (desc->die_dt != DT_FEMPTY) { + for (i = 0; i < limit && rx_packets < *quota && desc->die_dt != DT_FEMPTY; i++) { /* Descriptor type must be checked before all other reads */ dma_rmb(); desc_status = desc->msc; pkt_len = le16_to_cpu(desc->ds_cc) & RX_DS; - if (--boguscnt < 0) - break; - /* We use 0-byte descriptors to mark the DMA mapping errors */ if (!pkt_len) continue; @@ -820,7 +816,7 @@ static bool ravb_rx_gbeth(struct net_device *ndev, int *quota, int q) skb_put(skb, pkt_len); skb->protocol = eth_type_trans(skb, ndev); napi_gro_receive(&priv->napi[q], skb); - stats->rx_packets++; + rx_packets++; stats->rx_bytes += pkt_len; break; case DT_FSTART: @@ -848,7 +844,7 @@ static bool ravb_rx_gbeth(struct net_device *ndev, int *quota, int q) eth_type_trans(priv->rx_1st_skb, ndev); napi_gro_receive(&priv->napi[q], priv->rx_1st_skb); - stats->rx_packets++; + rx_packets++; stats->rx_bytes += pkt_len; break; } @@ -887,9 +883,9 @@ static bool ravb_rx_gbeth(struct net_device *ndev, int *quota, int q) desc->die_dt = DT_FEMPTY; } - *quota -= limit - (++boguscnt); - - return boguscnt <= 0; + stats->rx_packets += rx_packets; + *quota -= rx_packets; + return *quota == 0; } /* Packet receive function for Ethernet AVB */
The units of "work done" in the RX path should be packets instead of descriptors, as large packets can be spread over multiple descriptors. Fixes: 1c59eb678cbd ("ravb: Fillup ravb_rx_gbeth() stub") Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com> --- This patch has been broken out from my previous series "Improve GbEth performance on Renesas RZ/G2L and related SoCs" and submitted as a bugfix as requested by Sergey. I've labeled it as 'v3' so the ordering is clear. Remaining patches from the series will follow once we've done gPTP testing. drivers/net/ethernet/renesas/ravb_main.c | 22 +++++++++------------- 1 file changed, 9 insertions(+), 13 deletions(-)