RE: [PATCH net] net: xilinx: axienet: Increment Rx skb ring head pointer after BD is successfully allocated in dmaengine flow

[Public]

-----Original Message-----
From: Jakub Kicinski <kuba@kernel.org>
Sent: Monday, August 11, 2025 10:13 PM
To: Pandey, Radhey Shyam <radhey.shyam.pandey@amd.com>
Cc: Gupta, Suraj <redacted>; andrew+netdev@lunn.ch;
davem@davemloft.net; edumazet@google.com; pabeni@redhat.com; Simek,
Michal [off-list ref]; sean.anderson@linux.dev;
horms@kernel.org; netdev@vger.kernel.org; linux-arm-
kernel@lists.infradead.org; linux-kernel@vger.kernel.org; Katakam, Harini
[off-list ref]
Subject: Re: [PATCH net] net: xilinx: axienet: Increment Rx skb ring head
pointer after BD is successfully allocated in dmaengine flow

Caution: This message originated from an External Source. Use proper
caution when opening attachments, clicking links, or responding.


On Mon, 11 Aug 2025 15:55:02 +0000 Pandey, Radhey Shyam wrote:

quoted

quoted

That wasn't my reading, maybe I misinterpreted the code.

From what I could tell the driver tries to give one new buffer for
each buffer completed. So it never tries to "catch up" on previously
missed allocations. IOW say we have a queue with 16 indexes, after
16 failures (which may be spread out over
time) the ring will be empty.

Yes, IIRC there is 1:1 mapping for RX DMA callback and
axienet_rx_submit_desc(). In case there are failure in
axienet_rx_submit_desc() it is not able to reattempt in current
implementation. Theoretically there could be other error in
rx_submit_desc() (like dma_mapping/netdev
allocation)

One thought is to have some flag/index to tell that it should be
reattempted in subsequent axienet_rx_submit_desc() ?

Yes, some kind of counter of buffer that need to be allocated.
The other problem to solve is when the buffers are completely depleted there
will be no callback so no opportunity to refill.
For drivers which refill from NAPI this is usually solved by periodically
scheduling NAPI.

I think even after this same problem will exist in dmaengine flow because main problem here is axidma descriptor is freed after invoking callback. Callback (axienet_rx_cb()) will try to request new descriptor in rx_submit_desc(), but it won't be free till that time.
One possible solution I can think of is together with NAPI, having some DMAengine interface invoked by AXIDMA (similar to callback, say post_cb()) after freeing descriptor, which requests new descriptor from AXIDMA?