[U-Boot] [PATCH] net: configure DWMAC DMA by default AXI burst length
From: Sonic Zhang sonic.zhang@analog.com
Board can define its own AXI burst length to improve DWMAC DMA performance.
Signed-off-by: Sonic Zhang sonic.zhang@analog.com ---
drivers/net/designware.c | 2 ++ drivers/net/designware.h | 5 +++++ 2 files changed, 7 insertions(+)
diff --git a/drivers/net/designware.c b/drivers/net/designware.c index 9ded895..d8d6792 100644 --- a/drivers/net/designware.c +++ b/drivers/net/designware.c @@ -256,6 +256,8 @@ static int dw_eth_init(struct eth_device *dev, bd_t *bis)
writel(readl(&dma_p->opmode) | RXSTART | TXSTART, &dma_p->opmode);
+ writel((CONFIG_DW_AXI_BURST_LEN & 0x1FF >> 1), &dma_p->axibus); + /* Start up the PHY */ if (phy_startup(priv->phydev)) { printf("Could not initialize PHY %s\n", diff --git a/drivers/net/designware.h b/drivers/net/designware.h index ce51102..d8bd84d 100644 --- a/drivers/net/designware.h +++ b/drivers/net/designware.h @@ -102,6 +102,11 @@ struct eth_dma_regs { #define TXSECONDFRAME (1 << 2) #define RXSTART (1 << 1)
+/* Default AXI BUS Burst length */ +#ifndef CONFIG_DW_AXI_BURST_LEN +#define CONFIG_DW_AXI_BURST_LEN 0 +#endif + /* Descriptior related definitions */ #define MAC_MAX_FRAME_SZ (1600)
On Mon, Jan 26, 2015 at 8:54 PM, sonic.adi@gmail.com wrote:
From: Sonic Zhang sonic.zhang@analog.com
Board can define its own AXI burst length to improve DWMAC DMA
performance.
Signed-off-by: Sonic Zhang sonic.zhang@analog.com
drivers/net/designware.c | 2 ++ drivers/net/designware.h | 5 +++++ 2 files changed, 7 insertions(+)
diff --git a/drivers/net/designware.c b/drivers/net/designware.c index 9ded895..d8d6792 100644 --- a/drivers/net/designware.c +++ b/drivers/net/designware.c @@ -256,6 +256,8 @@ static int dw_eth_init(struct eth_device *dev, bd_t
*bis)
writel(readl(&dma_p->opmode) | RXSTART | TXSTART, &dma_p->opmode);
This should probably include: +#if CONFIG_DW_AXI_BURST_LEN so that this register is only accessed if configured.
writel((CONFIG_DW_AXI_BURST_LEN & 0x1FF >> 1), &dma_p->axibus);/* Start up the PHY */ if (phy_startup(priv->phydev)) { printf("Could not initialize PHY %s\n",diff --git a/drivers/net/designware.h b/drivers/net/designware.h index ce51102..d8bd84d 100644 --- a/drivers/net/designware.h +++ b/drivers/net/designware.h @@ -102,6 +102,11 @@ struct eth_dma_regs { #define TXSECONDFRAME (1 << 2) #define RXSTART (1 << 1)
+/* Default AXI BUS Burst length */ +#ifndef CONFIG_DW_AXI_BURST_LEN +#define CONFIG_DW_AXI_BURST_LEN 0 +#endif
/* Descriptior related definitions */ #define MAC_MAX_FRAME_SZ (1600)
-- 1.7.9.5
Hi Joe,
On Wed, Jan 28, 2015 at 4:43 AM, Joe Hershberger joe.hershberger@gmail.com wrote:
On Mon, Jan 26, 2015 at 8:54 PM, sonic.adi@gmail.com wrote:
From: Sonic Zhang sonic.zhang@analog.com
Board can define its own AXI burst length to improve DWMAC DMA performance.
Signed-off-by: Sonic Zhang sonic.zhang@analog.com
drivers/net/designware.c | 2 ++ drivers/net/designware.h | 5 +++++ 2 files changed, 7 insertions(+)
diff --git a/drivers/net/designware.c b/drivers/net/designware.c index 9ded895..d8d6792 100644 --- a/drivers/net/designware.c +++ b/drivers/net/designware.c @@ -256,6 +256,8 @@ static int dw_eth_init(struct eth_device *dev, bd_t *bis)
writel(readl(&dma_p->opmode) | RXSTART | TXSTART, &dma_p->opmode);This should probably include: +#if CONFIG_DW_AXI_BURST_LEN so that this register is only accessed if configured.
CONFIG_DW_AXI_BURST_LEN is always defined in designware.h in the same patch.
+/* Default AXI BUS Burst length */ +#ifndef CONFIG_DW_AXI_BURST_LEN +#define CONFIG_DW_AXI_BURST_LEN 0 +#endif +
Regards,
Sonic
On Tue, Jan 27, 2015 at 9:24 PM, Sonic Zhang sonic.adi@gmail.com wrote:
Hi Joe,
On Wed, Jan 28, 2015 at 4:43 AM, Joe Hershberger joe.hershberger@gmail.com wrote:
On Mon, Jan 26, 2015 at 8:54 PM, sonic.adi@gmail.com wrote:
From: Sonic Zhang sonic.zhang@analog.com
Board can define its own AXI burst length to improve DWMAC DMA performance.
Signed-off-by: Sonic Zhang sonic.zhang@analog.com
drivers/net/designware.c | 2 ++ drivers/net/designware.h | 5 +++++ 2 files changed, 7 insertions(+)
diff --git a/drivers/net/designware.c b/drivers/net/designware.c index 9ded895..d8d6792 100644 --- a/drivers/net/designware.c +++ b/drivers/net/designware.c @@ -256,6 +256,8 @@ static int dw_eth_init(struct eth_device *dev, bd_t *bis)
writel(readl(&dma_p->opmode) | RXSTART | TXSTART,
&dma_p->opmode);
This should probably include: +#if CONFIG_DW_AXI_BURST_LEN so that this register is only accessed if configured.
CONFIG_DW_AXI_BURST_LEN is always defined in designware.h in the same
patch.
+/* Default AXI BUS Burst length */ +#ifndef CONFIG_DW_AXI_BURST_LEN +#define CONFIG_DW_AXI_BURST_LEN 0 +#endif
Yes, I saw that... sorry I wasn't clear. I meant don't we want to avoid the write (preserving existing behavior) instead of ensuring that it is set, and still adding a register write that is (presumably) a NOP?
-Joe
Hi Joe,
On Wed, Jan 28, 2015 at 5:42 PM, Joe Hershberger joe.hershberger@gmail.com wrote:
On Tue, Jan 27, 2015 at 9:24 PM, Sonic Zhang sonic.adi@gmail.com wrote:
Hi Joe,
On Wed, Jan 28, 2015 at 4:43 AM, Joe Hershberger joe.hershberger@gmail.com wrote:
On Mon, Jan 26, 2015 at 8:54 PM, sonic.adi@gmail.com wrote:
From: Sonic Zhang sonic.zhang@analog.com
Board can define its own AXI burst length to improve DWMAC DMA performance.
Signed-off-by: Sonic Zhang sonic.zhang@analog.com
drivers/net/designware.c | 2 ++ drivers/net/designware.h | 5 +++++ 2 files changed, 7 insertions(+)
diff --git a/drivers/net/designware.c b/drivers/net/designware.c index 9ded895..d8d6792 100644 --- a/drivers/net/designware.c +++ b/drivers/net/designware.c @@ -256,6 +256,8 @@ static int dw_eth_init(struct eth_device *dev, bd_t *bis)
writel(readl(&dma_p->opmode) | RXSTART | TXSTART,&dma_p->opmode);
This should probably include: +#if CONFIG_DW_AXI_BURST_LEN so that this register is only accessed if configured.
CONFIG_DW_AXI_BURST_LEN is always defined in designware.h in the same patch.
+/* Default AXI BUS Burst length */ +#ifndef CONFIG_DW_AXI_BURST_LEN +#define CONFIG_DW_AXI_BURST_LEN 0 +#endif
Yes, I saw that... sorry I wasn't clear. I meant don't we want to avoid the write (preserving existing behavior) instead of ensuring that it is set, and still adding a register write that is (presumably) a NOP?
OK, I will update the patch.
Sonic
participants (3)
-
Joe Hershberger -
Sonic Zhang -
sonic.adi＠gmail.com