Re: [PATCH net-next v1 11/11] eth: fbnic: Support AF_XDP page pool zero copy

["Björn Töpel" <bjorn@kernel.org>]

netdev-bot+sashiko@kernel.org writes:

[...]

> Won't an idle xsk queue then report a fast-growing alloc_fail count that
> has nothing to do with memory pressure? With need_wakeup, each wakeup
> that finds the FILL ring empty would also be counted.

mlx5 counts it the same way, ice doesn't. It's already in
fill_ring_empty_descs, so v2 doesn't count it on rings the provider
feeds.

[...]

> The xsk_rxq_pool() comment says "The pool stays valid while the queue
> that the caller allocates runs", but that does not seem to hold on this
> failure path. Part of the cause is the core error handling. Still, the
> fbnic TX code added here is what keeps using the raw pointer.

Same root cause as the failed removal in 05, and the same fix. The
provider then keeps the pool, its rings and the UMEM until the queue DMA
has stopped, so the comment holds. The socket is off the pool's TX list
by then, so TX finds nothing, and completions go to a live ring.


Björn