[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <aLgJ9iqQhq-LT9S0@arm.com>
Date: Wed, 3 Sep 2025 10:27:18 +0100
From: Catalin Marinas <catalin.marinas@....com>
To: Ankur Arora <ankur.a.arora@...cle.com>
Cc: linux-kernel@...r.kernel.org, linux-arch@...r.kernel.org,
linux-arm-kernel@...ts.infradead.org, bpf@...r.kernel.org,
arnd@...db.de, will@...nel.org, peterz@...radead.org,
akpm@...ux-foundation.org, mark.rutland@....com,
harisokn@...zon.com, cl@...two.org, ast@...nel.org,
memxor@...il.com, zhenglifeng1@...wei.com,
xueshuai@...ux.alibaba.com, joao.m.martins@...cle.com,
boris.ostrovsky@...cle.com, konrad.wilk@...cle.com
Subject: Re: [PATCH v4 0/5] barrier: Add smp_cond_load_*_timewait()
On Tue, Sep 02, 2025 at 03:46:52PM -0700, Ankur Arora wrote:
> Catalin Marinas <catalin.marinas@....com> writes:
> > Can you have a go at poll_idle() to see how it would look like using
> > this API? It doesn't necessarily mean we have to merge them all at once
> > but it gives us a better idea of the suitability of the interface.
>
> So, I've been testing with some version of the following:
>
> diff --git a/drivers/cpuidle/poll_state.c b/drivers/cpuidle/poll_state.c
> index 9b6d90a72601..361879396d0c 100644
> --- a/drivers/cpuidle/poll_state.c
> +++ b/drivers/cpuidle/poll_state.c
> @@ -8,35 +8,25 @@
> #include <linux/sched/clock.h>
> #include <linux/sched/idle.h>
>
> -#define POLL_IDLE_RELAX_COUNT 200
> -
> static int __cpuidle poll_idle(struct cpuidle_device *dev,
> struct cpuidle_driver *drv, int index)
> {
> - u64 time_start;
> -
> - time_start = local_clock_noinstr();
> + unsigned long flags;
>
> dev->poll_time_limit = false;
>
> raw_local_irq_enable();
> if (!current_set_polling_and_test()) {
> - unsigned int loop_count = 0;
> - u64 limit;
> + u64 limit, time_end;
>
> limit = cpuidle_poll_time(drv, dev);
> + time_end = local_clock_noinstr() + limit;
>
> - while (!need_resched()) {
> - cpu_relax();
> - if (loop_count++ < POLL_IDLE_RELAX_COUNT)
> - continue;
> + flags = smp_cond_load_relaxed_timewait(¤t_thread_info()->flags,
> + VAL & _TIF_NEED_RESCHED,
> + (local_clock_noinstr() >= time_end));
It makes sense to have the non-strict comparison, though it changes the
original behaviour slightly. Just mention it in the commit log.
>
> - loop_count = 0;
> - if (local_clock_noinstr() - time_start > limit) {
> - dev->poll_time_limit = true;
> - break;
> - }
> - }
> + dev->poll_time_limit = (local_clock_noinstr() >= time_end);
Could we do this instead and avoid another clock read:
dev->poll_time_limit = !(flags & _TIF_NEED_RESCHED);
In the original code, it made sense since it had to check the clock
anyway and break the loop.
When you repost, please include the rqspinlock and poll_idle changes as
well to show how the interface is used.
--
Catalin
Powered by blists - more mailing lists