2019-10-28 15:33:25

by Jiangfeng Xiao

Subject: [PATCH v2] net: hisilicon: Fix ping latency when deal with high throughput

This is due to error in over budget processing.
When dealing with high throughput, the used buffers
that exceeds the budget is not cleaned up. In addition,
it takes a lot of cycles to clean up the used buffer,
and then the buffer where the valid data is located can take effect.

Signed-off-by: Jiangfeng Xiao <[email protected]>
---
ChangeLog v1->v2:
- Make rx_cnt_remaining part of struct hip04_priv.
---
drivers/net/ethernet/hisilicon/hip04_eth.c | 15 +++++++++------
1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/hisilicon/hip04_eth.c b/drivers/net/ethernet/hisilicon/hip04_eth.c
index c841674..f51bc02 100644
--- a/drivers/net/ethernet/hisilicon/hip04_eth.c
+++ b/drivers/net/ethernet/hisilicon/hip04_eth.c
@@ -237,6 +237,7 @@ struct hip04_priv {
dma_addr_t rx_phys[RX_DESC_NUM];
unsigned int rx_head;
unsigned int rx_buf_size;
+ unsigned int rx_cnt_remaining;

struct device_node *phy_node;
struct phy_device *phy;
@@ -575,7 +576,6 @@ static int hip04_rx_poll(struct napi_struct *napi, int budget)
struct hip04_priv *priv = container_of(napi, struct hip04_priv, napi);
struct net_device *ndev = priv->ndev;
struct net_device_stats *stats = &ndev->stats;
- unsigned int cnt = hip04_recv_cnt(priv);
struct rx_desc *desc;
struct sk_buff *skb;
unsigned char *buf;
@@ -588,8 +588,8 @@ static int hip04_rx_poll(struct napi_struct *napi, int budget)

/* clean up tx descriptors */
tx_remaining = hip04_tx_reclaim(ndev, false);
-
- while (cnt && !last) {
+ priv->rx_cnt_remaining += hip04_recv_cnt(priv);
+ while (priv->rx_cnt_remaining && !last) {
buf = priv->rx_buf[priv->rx_head];
skb = build_skb(buf, priv->rx_buf_size);
if (unlikely(!skb)) {
@@ -635,11 +635,13 @@ static int hip04_rx_poll(struct napi_struct *napi, int budget)
hip04_set_recv_desc(priv, phys);

priv->rx_head = RX_NEXT(priv->rx_head);
- if (rx >= budget)
+ if (rx >= budget) {
+ --priv->rx_cnt_remaining;
goto done;
+ }

- if (--cnt == 0)
- cnt = hip04_recv_cnt(priv);
+ if (--priv->rx_cnt_remaining == 0)
+ priv->rx_cnt_remaining += hip04_recv_cnt(priv);
}

if (!(priv->reg_inten & RCV_INT)) {
@@ -724,6 +726,7 @@ static int hip04_mac_open(struct net_device *ndev)
int i;

priv->rx_head = 0;
+ priv->rx_cnt_remaining = 0;
priv->tx_head = 0;
priv->tx_tail = 0;
hip04_reset_ppe(priv);
--
1.8.5.6

2019-10-29 23:36:13

by David Miller

Subject: Re: [PATCH v2] net: hisilicon: Fix ping latency when deal with high throughput

From: Jiangfeng Xiao <[email protected]>
Date: Mon, 28 Oct 2019 13:09:46 +0800

> This is due to error in over budget processing.
> When dealing with high throughput, the used buffers
> that exceeds the budget is not cleaned up. In addition,
> it takes a lot of cycles to clean up the used buffer,
> and then the buffer where the valid data is located can take effect.
>
> Signed-off-by: Jiangfeng Xiao <[email protected]>

Applied and queued up for -stable, thanks.