Re: [PATCH v2] io_uring/napi: don't end the single NAPI busy poll early on need_resched()

[Cinq Prairie]

On Mon, 09 Nov 2026 14:49:42 -0700, Jens Axboe wrote:
> Applied, thanks!
>
> [1/1] io_uring/napi: don't end the single NAPI busy poll early on need_resched()
>       commit: 39c7210e970200f8b17de3767c73efe64173a95f

Thanks.

On the cond_resched() follow-up: I no longer think it is worth sending.
Since 7dadeaa6e851 ("sched: Further restrict the preemption modes") in
v7.0, PREEMPT_NONE/PREEMPT_VOLUNTARY can't be selected on archs with
lazy preemption (x86, arm64, ...), so CONFIG_PREEMPTION is always set
there and cond_resched() compiles to a no-op. On those kernels the
need_resched() break in __napi_busy_loop() is already followed by a
preempt_enable() that reschedules, so a yield between laps wouldn't
add anything but a dead cond_resched() call while they are actively
being removed tree-wide.

Unless you'd like it for the few architectures still able to build
PREEMPT_VOLUNTARY, I'll leave it there.