linux-kernel - Re: [PATCH 2/2] drm/i915: Limit the busy wait on requests to 2us not 10ms!

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Mon, 16 Nov 2015 10:24:45 +0000
From:	Tvrtko Ursulin <tvrtko.ursulin@...ux.intel.com>
To:	Chris Wilson <chris@...is-wilson.co.uk>,
	Jens Axboe <axboe@...nel.dk>, intel-gfx@...ts.freedesktop.org,
	linux-kernel@...r.kernel.org
Cc:	dri-devel@...ts.freedesktop.org,
	Daniel Vetter <daniel.vetter@...ll.ch>,
	Eero Tamminen <eero.t.tamminen@...el.com>,
	"Rantala, Valtteri" <valtteri.rantala@...el.com>,
	stable@...nel.vger.org
Subject: Re: [PATCH 2/2] drm/i915: Limit the busy wait on requests to 2us not
 10ms!


Hi,

On 15/11/15 13:32, Chris Wilson wrote:
> When waiting for high frequency requests, the finite amount of time
> required to set up the irq and wait upon it limits the response rate. By
> busywaiting on the request completion for a short while we can service
> the high frequency waits as quick as possible. However, if it is a slow
> request, we want to sleep as quickly as possible. The tradeoff between
> waiting and sleeping is roughly the time it takes to sleep on a request,
> on the order of a microsecond. Based on measurements from big core, I
> have set the limit for busywaiting as 2 microseconds.

Sounds like solid reasoning. Would it also be worth finding the trade 
off limit for small core?

> The code currently uses the jiffie clock, but that is far too coarse (on
> the order of 10 milliseconds) and results in poor interactivity as the
> CPU ends up being hogged by slow requests. To get microsecond resolution
> we need to use a high resolution timer. The cheapest of which is polling
> local_clock(), but that is only valid on the same CPU. If we switch CPUs
> because the task was preempted, we can also use that as an indicator that
>   the system is too busy to waste cycles on spinning and we should sleep
> instead.

Hm, need_resched would not cover the CPU switch anyway? Or maybe 
need_resched means something other than I thought which is "there are 
other runnable tasks"?

This would also have impact on the patch subject line.I thought we would 
burn a jiffie of CPU cycles only if there are no other runnable tasks - 
so how come an impact on interactivity?

Also again I think the commit message needs some data on how this was 
found and what is the impact.

Btw as it happens, just last week as I was playing with perf, I did 
notice busy spinning is the top cycle waster in some benchmarks. I was 
in the process of trying to quantize the difference with it on or off 
but did not complete it.

> __i915_spin_request was introduced in
> commit 2def4ad99befa25775dd2f714fdd4d92faec6e34 [v4.2]
> Author: Chris Wilson <chris@...is-wilson.co.uk>
> Date:   Tue Apr 7 16:20:41 2015 +0100
>
>       drm/i915: Optimistically spin for the request completion
>
> Reported-by: Jens Axboe <axboe@...nel.dk>
> Link: https://lkml.org/lkml/2015/11/12/621
> Cc: Jens Axboe <axboe@...nel.dk>
> Cc; "Rogozhkin, Dmitry V" <dmitry.v.rogozhkin@...el.com>
> Cc: Daniel Vetter <daniel.vetter@...ll.ch>
> Cc: Tvrtko Ursulin <tvrtko.ursulin@...ux.intel.com>
> Cc: Eero Tamminen <eero.t.tamminen@...el.com>
> Cc: "Rantala, Valtteri" <valtteri.rantala@...el.com>
> Cc: stable@...nel.vger.org
> ---
>   drivers/gpu/drm/i915/i915_gem.c | 28 +++++++++++++++++++++++++---
>   1 file changed, 25 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 740530c571d1..2a88158bd1f7 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -1146,14 +1146,36 @@ static bool missed_irq(struct drm_i915_private *dev_priv,
>   	return test_bit(ring->id, &dev_priv->gpu_error.missed_irq_rings);
>   }
>
> +static u64 local_clock_us(unsigned *cpu)
> +{
> +	u64 t;
> +
> +	*cpu = get_cpu();
> +	t = local_clock() >> 10;

Needs comment I think to explicitly mention the approximation, or maybe 
drop the _us suffix?

> +	put_cpu();
> +
> +	return t;
> +}
> +
> +static bool busywait_stop(u64 timeout, unsigned cpu)
> +{
> +	unsigned this_cpu;
> +
> +	if (time_after64(local_clock_us(&this_cpu), timeout))
> +		return true;
> +
> +	return this_cpu != cpu;
> +}
> +
>   static int __i915_spin_request(struct drm_i915_gem_request *req, int state)
>   {
> -	unsigned long timeout;
> +	u64 timeout;
> +	unsigned cpu;
>
>   	if (i915_gem_request_get_ring(req)->irq_refcount)
>   		return -EBUSY;
>
> -	timeout = jiffies + 1;
> +	timeout = local_clock_us(&cpu) + 2;
>   	while (!need_resched()) {
>   		if (i915_gem_request_completed(req, true))
>   			return 0;
> @@ -1161,7 +1183,7 @@ static int __i915_spin_request(struct drm_i915_gem_request *req, int state)
>   		if (signal_pending_state(state, current))
>   			break;
>
> -		if (time_after_eq(jiffies, timeout))
> +		if (busywait_stop(timeout, cpu))
>   			break;
>
>   		cpu_relax_lowlatency();
>

Otherwise looks good. Not sure what would you convert to 32-bit from 
your follow up reply since you need us resolution?

Regards,

Tvrtko
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/