[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <a9ad456d-eeff-4fac-a18d-0219fcc9f5ed@linux.dev>
Date: Thu, 5 Sep 2024 10:34:15 -0400
From: Sean Anderson <sean.anderson@...ux.dev>
To: Simon Horman <horms@...nel.org>
Cc: Radhey Shyam Pandey <radhey.shyam.pandey@....com>,
"David S. Miller" <davem@...emloft.net>, Eric Dumazet <edumazet@...gle.com>,
Jakub Kicinski <kuba@...nel.org>, Paolo Abeni <pabeni@...hat.com>,
netdev@...r.kernel.org, Michal Simek <michal.simek@....com>,
linux-arm-kernel@...ts.infradead.org,
Ariane Keller <ariane.keller@....ee.ethz.ch>, linux-kernel@...r.kernel.org,
Daniel Borkmann <daniel@...earbox.net>, Andy Chiu <andy.chiu@...ive.com>
Subject: Re: [PATCH net] net: xilinx: axienet: Fix IRQ coalescing packet count
overflow
On 9/4/24 12:00, Simon Horman wrote:
> On Tue, Sep 03, 2024 at 02:00:59PM -0400, Sean Anderson wrote:
>> If coalesce_count is greater than 255 it will not fit in the register and
>> will overflow. Clamp it to 255 for more-predictable results.
>
> Hi Sean,
>
> Can this occur in practice?
Yes. Simply do `ethtool -C ethX rx-frames 300` or something similar and
you will end up with a limit of 44 instead. I ran into this with DIM and
was wondering why the highest-throughput setting (256) was behaving so
poorly...
>>
>> Signed-off-by: Sean Anderson <sean.anderson@...ux.dev>
>> Fixes: 8a3b7a252dca ("drivers/net/ethernet/xilinx: added Xilinx AXI Ethernet driver")
>
> nit: I think it is usual for the order of these tags to be reversed.
OK
>> ---
>>
>> drivers/net/ethernet/xilinx/xilinx_axienet_main.c | 6 ++++--
>> 1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
>> index 9aeb7b9f3ae4..5f27fc1c4375 100644
>> --- a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
>> +++ b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
>> @@ -252,7 +252,8 @@ 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 = (min(lp->coalesce_count_rx, 255) <<
>> + XAXIDMA_COALESCE_SHIFT) |
>> XAXIDMA_IRQ_IOC_MASK | XAXIDMA_IRQ_ERROR_MASK;
>
> nit: it would be nice to avoid using a naked 255 here.
> Perhaps: #define XAXIDMA_COALESCE_MAX 0xff
OK, but this is the same as the limit used in axienet_usec_to_timer.
--Sean
>> /* 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 +265,8 @@ 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 = (min(lp->coalesce_count_tx, 255) <<
>> + XAXIDMA_COALESCE_SHIFT) |
>> 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
>>
>>
Powered by blists - more mailing lists