Re: [PATCH net v2] net: xilinx: axienet: Fix IRQ coalescing packet count overflow

[Sean Anderson]

On 9/11/24 03:01, Pandey, Radhey Shyam wrote:
>> -----Original Message-----
>> From: Sean Anderson <sean.anderson@xxxxxxxxx>
>> Sent: Tuesday, September 10, 2024 4:39 AM
>> To: Pandey, Radhey Shyam <radhey.shyam.pandey@xxxxxxx>; David S .
>> Miller <davem@xxxxxxxxxxxxx>; Eric Dumazet <edumazet@xxxxxxxxxx>;
>> Jakub Kicinski <kuba@xxxxxxxxxx>; Paolo Abeni <pabeni@xxxxxxxxxx>;
>> netdev@xxxxxxxxxxxxxxx
>> Cc: Andy Chiu <andy.chiu@xxxxxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx; Simon
>> Horman <horms@xxxxxxxxxx>; Ariane Keller <ariane.keller@xxxxxxxxxxxxxx>;
>> Daniel Borkmann <daniel@xxxxxxxxxxxxx>; linux-arm-
>> kernel@xxxxxxxxxxxxxxxxxxx; Simek, Michal <michal.simek@xxxxxxx>; Sean
>> Anderson <sean.anderson@xxxxxxxxx>
>> Subject: [PATCH net v2] net: xilinx: axienet: Fix IRQ coalescing packet count
>> overflow
>> 
>> If coalece_count is greater than 255 it will not fit in the register and
>> will overflow. This can be reproduced by running
>> 
>>     # ethtool -C ethX rx-frames 256
>> 
>> which will result in a timeout of 0us instead. Fix this by clamping the
>> counts to the maximum value.
> After this fix - what is o/p we get on rx-frames read? I think silent clamping is not a great 
> idea and user won't know about it.  One alternative is to add check in set_coalesc 
> count for valid range? (Similar to axienet_ethtools_set_ringparam so that user is notified 
> for incorrect range)

The value reported will be unclamped. In [1] I improve the driver to
return the actual (clamped) value.

Remember that without this commit, we have silent wraparound instead. I
think clamping is much friendlier, since you at least get something
close to the rx-frames value, instead of zero!

This commit is just a fix for the overflow issue. To ensure it is
appropriate for backporting I have omitted any other
changes/improvements.

--Sean

[1] https://lore.kernel.org/netdev/20240909235208.1331065-6-sean.anderson@xxxxxxxxx/

>> 
>> Signed-off-by: Sean Anderson <sean.anderson@xxxxxxxxx>
>> Fixes: 8a3b7a252dca ("drivers/net/ethernet/xilinx: added Xilinx AXI Ethernet
>> driver")
>> ---
>> 
>> Changes in v2:
>> - Use FIELD_MAX to extract the max value from the mask
>> - Expand the commit message with an example on how to reproduce this
>>   issue
>> 
>>  drivers/net/ethernet/xilinx/xilinx_axienet.h      | 5 ++---
>>  drivers/net/ethernet/xilinx/xilinx_axienet_main.c | 8 ++++++--
>>  2 files changed, 8 insertions(+), 5 deletions(-)
>> 
>> diff --git a/drivers/net/ethernet/xilinx/xilinx_axienet.h
>> b/drivers/net/ethernet/xilinx/xilinx_axienet.h
>> index 1223fcc1a8da..54db69893565 100644
>> --- a/drivers/net/ethernet/xilinx/xilinx_axienet.h
>> +++ b/drivers/net/ethernet/xilinx/xilinx_axienet.h
>> @@ -109,11 +109,10 @@
>>  #define XAXIDMA_BD_CTRL_TXEOF_MASK	0x04000000 /* Last tx packet
>> */
>>  #define XAXIDMA_BD_CTRL_ALL_MASK	0x0C000000 /* All control bits
>> */
>> 
>> -#define XAXIDMA_DELAY_MASK		0xFF000000 /* Delay timeout
>> counter */
>> -#define XAXIDMA_COALESCE_MASK		0x00FF0000 /* Coalesce
>> counter */
>> +#define XAXIDMA_DELAY_MASK		((u32)0xFF000000) /* Delay
>> timeout counter */
> 
> Adding typecast here looks odd. Any reason for it? 
> If needed we do it in specific case where it is required.
> 
>> +#define XAXIDMA_COALESCE_MASK		((u32)0x00FF0000) /*
>> Coalesce counter */
>> 
>>  #define XAXIDMA_DELAY_SHIFT		24
>> -#define XAXIDMA_COALESCE_SHIFT		16
>> 
>>  #define XAXIDMA_IRQ_IOC_MASK		0x00001000 /* Completion
>> intr */
>>  #define XAXIDMA_IRQ_DELAY_MASK		0x00002000 /* Delay
>> interrupt */
>> diff --git a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
>> b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
>> index 9eb300fc3590..89b63695293d 100644
>> --- a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
>> +++ b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
>> @@ -252,7 +252,9 @@ static u32 axienet_usec_to_timer(struct axienet_local
>> *lp, u32 coalesce_usec)
>>  static void axienet_dma_start(struct axienet_local *lp)
>>  {
>>  	/* Start updating the Rx channel control register */
>> -	lp->rx_dma_cr = (lp->coalesce_count_rx <<
>> XAXIDMA_COALESCE_SHIFT) |
>> +	lp->rx_dma_cr = FIELD_PREP(XAXIDMA_COALESCE_MASK,
>> +				   min(lp->coalesce_count_rx,
>> +
>> FIELD_MAX(XAXIDMA_COALESCE_MASK))) |
>>  			XAXIDMA_IRQ_IOC_MASK |
>> XAXIDMA_IRQ_ERROR_MASK;
>>  	/* Only set interrupt delay timer if not generating an interrupt on
>>  	 * the first RX packet. Otherwise leave at 0 to disable delay interrupt.
>> @@ -264,7 +266,9 @@ static void axienet_dma_start(struct axienet_local
>> *lp)
>>  	axienet_dma_out32(lp, XAXIDMA_RX_CR_OFFSET, lp->rx_dma_cr);
>> 
>>  	/* Start updating the Tx channel control register */
>> -	lp->tx_dma_cr = (lp->coalesce_count_tx <<
>> XAXIDMA_COALESCE_SHIFT) |
>> +	lp->tx_dma_cr = FIELD_PREP(XAXIDMA_COALESCE_MASK,
>> +				   min(lp->coalesce_count_tx,
>> +
>> FIELD_MAX(XAXIDMA_COALESCE_MASK))) |
>>  			XAXIDMA_IRQ_IOC_MASK |
>> XAXIDMA_IRQ_ERROR_MASK;
>>  	/* Only set interrupt delay timer if not generating an interrupt on
>>  	 * the first TX packet. Otherwise leave at 0 to disable delay interrupt.
>> --
>> 2.35.1.1320.gc452695387.dirty
>