diff mbox series

net: designware: improve ethernet DMA reset

Message ID 20210721061910.776977-1-art@khadas.com
State Changes Requested
Delegated to: Ramon Fried
Headers show
Series net: designware: improve ethernet DMA reset | expand

Commit Message

Art Nikpal July 21, 2021, 6:19 a.m. UTC
Problem: uboot ethernet always have "DMA reset timeout" after
linux usage, detected on VIM2 VIM3 VIM3L amlogic devices,
mainline uboot and mainline linux > 5.1? .

How-to reproduce:

1) Poweron device => ethernet uboot work fine.
2) Linux start and ethernet interface up and same works.
3) Reboot to uboot again and we cant use ethernet anymore.
uboot always have "DMA reset timeout" inside designware_eth_init()
only poweroff/poweron cycle can solve this problem.

Solution: add phy_reset() into waiting DMA reset cycle,
this insertion doesn't break normal behaviour (maybe).

Note: same need explore real problem why its happens! because this 
patch may be just hack.

Signed-off-by: Artem Lapkin <art@khadas.com>
---
 drivers/net/designware.c | 4 ++++
 1 file changed, 4 insertions(+)

Comments

Neil Armstrong July 26, 2021, 7:38 a.m. UTC | #1
Hi,

On 21/07/2021 08:19, Artem Lapkin wrote:
> Problem: uboot ethernet always have "DMA reset timeout" after
> linux usage, detected on VIM2 VIM3 VIM3L amlogic devices,
> mainline uboot and mainline linux > 5.1? .
> 
> How-to reproduce:
> 
> 1) Poweron device => ethernet uboot work fine.
> 2) Linux start and ethernet interface up and same works.
> 3) Reboot to uboot again and we cant use ethernet anymore.
> uboot always have "DMA reset timeout" inside designware_eth_init()
> only poweroff/poweron cycle can solve this problem.
> 
> Solution: add phy_reset() into waiting DMA reset cycle,
> this insertion doesn't break normal behaviour (maybe).
> 
> Note: same need explore real problem why its happens! because this 
> patch may be just hack.

Interesting, maybe add this phy_reset() call only if the GPIO reset is not available ?

Neil

> 
> Signed-off-by: Artem Lapkin <art@khadas.com>
> ---
>  drivers/net/designware.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/net/designware.c b/drivers/net/designware.c
> index 5d92257e..6485c46d 100644
> --- a/drivers/net/designware.c
> +++ b/drivers/net/designware.c
> @@ -371,6 +371,7 @@ int designware_eth_init(struct dw_eth_dev *priv, u8 *enetaddr)
>  	struct eth_dma_regs *dma_p = priv->dma_regs_p;
>  	unsigned int start;
>  	int ret;
> +	int phy_reset_force = 0;
>  
>  	writel(readl(&dma_p->busmode) | DMAMAC_SRST, &dma_p->busmode);
>  
> @@ -389,6 +390,9 @@ int designware_eth_init(struct dw_eth_dev *priv, u8 *enetaddr)
>  			printf("DMA reset timeout\n");
>  			return -ETIMEDOUT;
>  		}
> +		// forced phy reset after 0.2s can help with DMAMAC_SRST
> +		if (phy_reset_force++ == 2)
> +			phy_reset(priv->phydev);
>  
>  		mdelay(100);
>  	};
>
diff mbox series

Patch

diff --git a/drivers/net/designware.c b/drivers/net/designware.c
index 5d92257e..6485c46d 100644
--- a/drivers/net/designware.c
+++ b/drivers/net/designware.c
@@ -371,6 +371,7 @@  int designware_eth_init(struct dw_eth_dev *priv, u8 *enetaddr)
 	struct eth_dma_regs *dma_p = priv->dma_regs_p;
 	unsigned int start;
 	int ret;
+	int phy_reset_force = 0;
 
 	writel(readl(&dma_p->busmode) | DMAMAC_SRST, &dma_p->busmode);
 
@@ -389,6 +390,9 @@  int designware_eth_init(struct dw_eth_dev *priv, u8 *enetaddr)
 			printf("DMA reset timeout\n");
 			return -ETIMEDOUT;
 		}
+		// forced phy reset after 0.2s can help with DMAMAC_SRST
+		if (phy_reset_force++ == 2)
+			phy_reset(priv->phydev);
 
 		mdelay(100);
 	};