From: Harini Katakam <[email protected]>
Reset TX when halt times out i.e. disable TX, clean up TX BDs,
interrupts (already done) and enable TX.
This addresses the issue observed when iperf is run at 10Mps Half
duplex where, after multiple collisions and retries, TX halts.
Signed-off-by: Harini Katakam <[email protected]>
Signed-off-by: Michal Simek <[email protected]>
Signed-off-by: Radhey Shyam Pandey <[email protected]>
---
drivers/net/ethernet/cadence/macb_main.c | 10 ++++++++--
1 file changed, 8 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
index 96fd2aa9ee90..473c2d0174ad 100644
--- a/drivers/net/ethernet/cadence/macb_main.c
+++ b/drivers/net/ethernet/cadence/macb_main.c
@@ -1021,6 +1021,7 @@ static void macb_tx_error_task(struct work_struct *work)
struct sk_buff *skb;
unsigned int tail;
unsigned long flags;
+ bool halt_timeout = false;
netdev_vdbg(bp->dev, "macb_tx_error_task: q = %u, t = %u, h = %u\n",
(unsigned int)(queue - bp->queues),
@@ -1042,9 +1043,11 @@ static void macb_tx_error_task(struct work_struct *work)
* (in case we have just queued new packets)
* macb/gem must be halted to write TBQP register
*/
- if (macb_halt_tx(bp))
- /* Just complain for now, reinitializing TX path can be good */
+ if (macb_halt_tx(bp)) {
netdev_err(bp->dev, "BUG: halt tx timed out\n");
+ macb_writel(bp, NCR, macb_readl(bp, NCR) & (~MACB_BIT(TE)));
+ halt_timeout = true;
+ }
/* Treat frames in TX queue including the ones that caused the error.
* Free transmit buffers in upper layer.
@@ -1115,6 +1118,9 @@ static void macb_tx_error_task(struct work_struct *work)
macb_writel(bp, TSR, macb_readl(bp, TSR));
queue_writel(queue, IER, MACB_TX_INT_FLAGS);
+ if (halt_timeout)
+ macb_writel(bp, NCR, macb_readl(bp, NCR) | MACB_BIT(TE));
+
/* Now we are ready to start transmission again */
netif_tx_start_all_queues(bp->dev);
macb_writel(bp, NCR, macb_readl(bp, NCR) | MACB_BIT(TSTART));
--
2.17.1
On Thu, Mar 16, 2023 at 02:05:54PM +0530, Harini Katakam wrote:
> From: Harini Katakam <[email protected]>
>
> Reset TX when halt times out i.e. disable TX, clean up TX BDs,
> interrupts (already done) and enable TX.
> This addresses the issue observed when iperf is run at 10Mps Half
> duplex where, after multiple collisions and retries, TX halts.
>
> Signed-off-by: Harini Katakam <[email protected]>
> Signed-off-by: Michal Simek <[email protected]>
> Signed-off-by: Radhey Shyam Pandey <[email protected]>
> ---
> drivers/net/ethernet/cadence/macb_main.c | 10 ++++++++--
> 1 file changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
> index 96fd2aa9ee90..473c2d0174ad 100644
> --- a/drivers/net/ethernet/cadence/macb_main.c
> +++ b/drivers/net/ethernet/cadence/macb_main.c
> @@ -1021,6 +1021,7 @@ static void macb_tx_error_task(struct work_struct *work)
> struct sk_buff *skb;
> unsigned int tail;
> unsigned long flags;
> + bool halt_timeout = false;
RCT
Otherwise looks fine
Reviewed-by: Michal Swiatkowski <[email protected]>
[...]
> --
> 2.17.1
>
On 16/03/2023 at 11:54, Michal Swiatkowski wrote:
> On Thu, Mar 16, 2023 at 02:05:54PM +0530, Harini Katakam wrote:
>> From: Harini Katakam <[email protected]>
>>
>> Reset TX when halt times out i.e. disable TX, clean up TX BDs,
>> interrupts (already done) and enable TX.
>> This addresses the issue observed when iperf is run at 10Mps Half
>> duplex where, after multiple collisions and retries, TX halts.
>>
>> Signed-off-by: Harini Katakam <[email protected]>
>> Signed-off-by: Michal Simek <[email protected]>
>> Signed-off-by: Radhey Shyam Pandey <[email protected]>
>> ---
>> drivers/net/ethernet/cadence/macb_main.c | 10 ++++++++--
>> 1 file changed, 8 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
>> index 96fd2aa9ee90..473c2d0174ad 100644
>> --- a/drivers/net/ethernet/cadence/macb_main.c
>> +++ b/drivers/net/ethernet/cadence/macb_main.c
>> @@ -1021,6 +1021,7 @@ static void macb_tx_error_task(struct work_struct *work)
>> struct sk_buff *skb;
>> unsigned int tail;
>> unsigned long flags;
>> + bool halt_timeout = false;
> RCT
Yes, might not pass the netdev checks.
> Otherwise looks fine
> Reviewed-by: Michal Swiatkowski <[email protected]>
Likewise, this fixed:
Acked-by: Nicolas Ferre <[email protected]>
Best regards,
Nicolas
>
> [...]
>
>> --
>> 2.17.1
>>
--
Nicolas Ferre