[PATCH net-next v4 0/2] net: renesas: rswitch: Improve perfromance of TX/RX

[PATCH net-next v4 0/2] net: renesas: rswitch: Improve perfromance of TX/RX

[Yoshihiro Shimoda]

This patch series is based on net-next.git / main branch [1]. This patch
series can improve perfromance of TX in a specific condition. The previous code
used "global rate limiter" feature so that this is possible to cause
performance down if we use multiple ports at the same time. To resolve this
issue, use "hardware pause" features of GWCA and COMA. Note that this is not
related to the ethernet PAUSE frames.

< UDP TX by iperf3 >
 before: about 450Mbps on both tsn0 and tsn1
 after:  about 950Mbps on both tsn0 and tsn1

Also, this patch series can improve performance of RX by using
napi_gro_receive().

< TCP RX by iperf >
 before: about 670Mbps on tsn0
 after:  about 840Mbps on tsn0

[1]
The commit e06bd5e3adae ("Merge branch 'followup-fixes-for-the-dwmac-and-altera-lynx-conversion'")

Changes from v3:
https://lore.kernel.org/all/20230607015641.1724057-1-yoshihiro.shimoda.uh@renesas.com/
 - Rebased on the latest net-next.git / main branch.
 - Added Reviewed-by in the patch 2/2. (Maciej, thanks!)
 - Fix typos in the commit description in the patch 2/2.

Changes from v2:
https://lore.kernel.org/all/20230606085558.1708766-1-yoshihiro.shimoda.uh@renesas.com/
 - Rebased on the latest net-next.git / main branch.
 - Added Reviewed-by in the patch 1/2. (Maciej, thanks!)
 - Revise the commit description in the patch 2/2.
 - Add definition to remove magic hardcoded numbers in the patch 2/2.

Changes from v1:
https://lore.kernel.org/all/20230529080840.1156458-1-yoshihiro.shimoda.uh@renesas.com/
 - Rebased on the latest net-next.git / main branch.
 - Use "hardware pause" feature instead of "per-queue limiter" feature.
 - Drop refactaring for "per-queue limiter".
 - Drop dt-bindings update because "hardware pause" doesn't need additional
   clock information.
 - Use napi_gro_receive() to improve RX performance.

Yoshihiro Shimoda (2):
  net: renesas: rswitch: Use napi_gro_receive() in RX
  net: renesas: rswitch: Use hardware pause features

 drivers/net/ethernet/renesas/rswitch.c | 38 ++++++++++----------------
 drivers/net/ethernet/renesas/rswitch.h |  7 +++++
 2 files changed, 22 insertions(+), 23 deletions(-)

-- 
2.25.1

[PATCH net-next v4 1/2] net: renesas: rswitch: Use napi_gro_receive() in RX

[Yoshihiro Shimoda]

This hardware can receive multiple frames so that using
napi_gro_receive() instead of netif_receive_skb() gets good
performance of RX.

Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/renesas/rswitch.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/renesas/rswitch.c b/drivers/net/ethernet/renesas/rswitch.c
index aace87139cea..7bb0a6d594a0 100644
--- a/drivers/net/ethernet/renesas/rswitch.c
+++ b/drivers/net/ethernet/renesas/rswitch.c
@@ -729,7 +729,7 @@ static bool rswitch_rx(struct net_device *ndev, int *quota)
 		}
 		skb_put(skb, pkt_len);
 		skb->protocol = eth_type_trans(skb, ndev);
-		netif_receive_skb(skb);
+		napi_gro_receive(&rdev->napi, skb);
 		rdev->ndev->stats.rx_packets++;
 		rdev->ndev->stats.rx_bytes += pkt_len;
 
-- 
2.25.1

[PATCH net-next v4 2/2] net: renesas: rswitch: Use hardware pause features

[Yoshihiro Shimoda]

Since this driver used the "global rate limiter" feature of GWCA,
the TX performance of each port was reduced when multiple ports
transmitted frames simultaneously. To improve performance, remove
the use of the "global rate limiter" feature and use "hardware pause"
features of the following:
 - "per priority pause" of GWCA
 - "global pause" of COMA

Note that these features are not related to the ethernet PAUSE frame.

Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
---
 drivers/net/ethernet/renesas/rswitch.c | 36 ++++++++++----------------
 drivers/net/ethernet/renesas/rswitch.h |  7 +++++
 2 files changed, 21 insertions(+), 22 deletions(-)

diff --git a/drivers/net/ethernet/renesas/rswitch.c b/drivers/net/ethernet/renesas/rswitch.c
index 7bb0a6d594a0..84f62c77eb8f 100644
--- a/drivers/net/ethernet/renesas/rswitch.c
+++ b/drivers/net/ethernet/renesas/rswitch.c
@@ -90,6 +90,11 @@ static int rswitch_bpool_config(struct rswitch_private *priv)
 	return rswitch_reg_wait(priv->addr, CABPIRM, CABPIRM_BPR, CABPIRM_BPR);
 }
 
+static void rswitch_coma_init(struct rswitch_private *priv)
+{
+	iowrite32(CABPPFLC_INIT_VALUE, priv->addr + CABPPFLC0);
+}
+
 /* R-Switch-2 block (TOP) */
 static void rswitch_top_init(struct rswitch_private *priv)
 {
@@ -156,24 +161,6 @@ static int rswitch_gwca_axi_ram_reset(struct rswitch_private *priv)
 	return rswitch_reg_wait(priv->addr, GWARIRM, GWARIRM_ARR, GWARIRM_ARR);
 }
 
-static void rswitch_gwca_set_rate_limit(struct rswitch_private *priv, int rate)
-{
-	u32 gwgrlulc, gwgrlc;
-
-	switch (rate) {
-	case 1000:
-		gwgrlulc = 0x0000005f;
-		gwgrlc = 0x00010260;
-		break;
-	default:
-		dev_err(&priv->pdev->dev, "%s: This rate is not supported (%d)\n", __func__, rate);
-		return;
-	}
-
-	iowrite32(gwgrlulc, priv->addr + GWGRLULC);
-	iowrite32(gwgrlc, priv->addr + GWGRLC);
-}
-
 static bool rswitch_is_any_data_irq(struct rswitch_private *priv, u32 *dis, bool tx)
 {
 	u32 *mask = tx ? priv->gwca.tx_irq_bits : priv->gwca.rx_irq_bits;
@@ -402,7 +389,7 @@ static int rswitch_gwca_queue_format(struct net_device *ndev,
 	linkfix->die_dt = DT_LINKFIX;
 	rswitch_desc_set_dptr(linkfix, gq->ring_dma);
 
-	iowrite32(GWDCC_BALR | (gq->dir_tx ? GWDCC_DQT : 0) | GWDCC_EDE,
+	iowrite32(GWDCC_BALR | (gq->dir_tx ? GWDCC_DCP(GWCA_IPV_NUM) | GWDCC_DQT : 0) | GWDCC_EDE,
 		  priv->addr + GWDCC_OFFS(gq->index));
 
 	return 0;
@@ -500,7 +487,8 @@ static int rswitch_gwca_queue_ext_ts_format(struct net_device *ndev,
 	linkfix->die_dt = DT_LINKFIX;
 	rswitch_desc_set_dptr(linkfix, gq->ring_dma);
 
-	iowrite32(GWDCC_BALR | (gq->dir_tx ? GWDCC_DQT : 0) | GWDCC_ETS | GWDCC_EDE,
+	iowrite32(GWDCC_BALR | (gq->dir_tx ? GWDCC_DCP(GWCA_IPV_NUM) | GWDCC_DQT : 0) |
+		  GWDCC_ETS | GWDCC_EDE,
 		  priv->addr + GWDCC_OFFS(gq->index));
 
 	return 0;
@@ -649,7 +637,8 @@ static int rswitch_gwca_hw_init(struct rswitch_private *priv)
 	iowrite32(lower_32_bits(priv->gwca.ts_queue.ring_dma), priv->addr + GWTDCAC10);
 	iowrite32(upper_32_bits(priv->gwca.ts_queue.ring_dma), priv->addr + GWTDCAC00);
 	iowrite32(GWCA_TS_IRQ_BIT, priv->addr + GWTSDCC0);
-	rswitch_gwca_set_rate_limit(priv, priv->gwca.speed);
+
+	iowrite32(GWTPC_PPPL(GWCA_IPV_NUM), priv->addr + GWTPC0);
 
 	for (i = 0; i < RSWITCH_NUM_PORTS; i++) {
 		err = rswitch_rxdmac_init(priv, i);
@@ -1502,7 +1491,8 @@ static netdev_tx_t rswitch_start_xmit(struct sk_buff *skb, struct net_device *nd
 	rswitch_desc_set_dptr(&desc->desc, dma_addr);
 	desc->desc.info_ds = cpu_to_le16(skb->len);
 
-	desc->info1 = cpu_to_le64(INFO1_DV(BIT(rdev->etha->index)) | INFO1_FMT);
+	desc->info1 = cpu_to_le64(INFO1_DV(BIT(rdev->etha->index)) |
+				  INFO1_IPV(GWCA_IPV_NUM) | INFO1_FMT);
 	if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) {
 		struct rswitch_gwca_ts_info *ts_info;
 
@@ -1772,6 +1762,8 @@ static int rswitch_init(struct rswitch_private *priv)
 	if (err < 0)
 		return err;
 
+	rswitch_coma_init(priv);
+
 	err = rswitch_gwca_linkfix_alloc(priv);
 	if (err < 0)
 		return -ENOMEM;
diff --git a/drivers/net/ethernet/renesas/rswitch.h b/drivers/net/ethernet/renesas/rswitch.h
index b3e0411b408e..bb9ed971a97c 100644
--- a/drivers/net/ethernet/renesas/rswitch.h
+++ b/drivers/net/ethernet/renesas/rswitch.h
@@ -48,6 +48,7 @@
 #define GWCA_NUM_IRQS		8
 #define GWCA_INDEX		0
 #define AGENT_INDEX_GWCA	3
+#define GWCA_IPV_NUM		0
 #define GWRO			RSWITCH_GWCA0_OFFSET
 
 #define GWCA_TS_IRQ_RESOURCE_NAME	"gwca0_rxts0"
@@ -768,11 +769,14 @@ enum rswitch_gwca_mode {
 #define GWARIRM_ARR		BIT(1)
 
 #define GWDCC_BALR		BIT(24)
+#define GWDCC_DCP_MASK		GENMASK(18, 16)
+#define GWDCC_DCP(prio)		FIELD_PREP(GWDCC_DCP_MASK, (prio))
 #define GWDCC_DQT		BIT(11)
 #define GWDCC_ETS		BIT(9)
 #define GWDCC_EDE		BIT(8)
 
 #define GWTRC(queue)		(GWTRC0 + (queue) / 32 * 4)
+#define GWTPC_PPPL(ipv)		BIT(ipv)
 #define GWDCC_OFFS(queue)	(GWDCC0 + (queue) * 4)
 
 #define GWDIS(i)		(GWDIS0 + (i) * 0x10)
@@ -789,6 +793,8 @@ enum rswitch_gwca_mode {
 #define CABPIRM_BPIOG		BIT(0)
 #define CABPIRM_BPR		BIT(1)
 
+#define CABPPFLC_INIT_VALUE	0x00800080
+
 /* MFWD */
 #define FWPC0_LTHTA		BIT(0)
 #define FWPC0_IP4UE		BIT(3)
@@ -863,6 +869,7 @@ enum DIE_DT {
 
 /* For transmission */
 #define INFO1_TSUN(val)		((u64)(val) << 8ULL)
+#define INFO1_IPV(prio)		((u64)(prio) << 28ULL)
 #define INFO1_CSD0(index)	((u64)(index) << 32ULL)
 #define INFO1_CSD1(index)	((u64)(index) << 40ULL)
 #define INFO1_DV(port_vector)	((u64)(port_vector) << 48ULL)
-- 
2.25.1

Re: [PATCH net-next v4 0/2] net: renesas: rswitch: Improve perfromance of TX/RX

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (main)
by David S. Miller <davem@davemloft.net>:

On Thu,  8 Jun 2023 11:20:05 +0900 you wrote:
> This patch series is based on net-next.git / main branch [1]. This patch
> series can improve perfromance of TX in a specific condition. The previous code
> used "global rate limiter" feature so that this is possible to cause
> performance down if we use multiple ports at the same time. To resolve this
> issue, use "hardware pause" features of GWCA and COMA. Note that this is not
> related to the ethernet PAUSE frames.
> 
> [...]

Here is the summary with links:
  - [net-next,v4,1/2] net: renesas: rswitch: Use napi_gro_receive() in RX
    https://git.kernel.org/netdev/net-next/c/dc510c6d2ecf
  - [net-next,v4,2/2] net: renesas: rswitch: Use hardware pause features
    https://git.kernel.org/netdev/net-next/c/c87bd91e34e1

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html