From: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
To: Chris Wilson <chris@chris-wilson.co.uk>, intel-gfx@lists.freedesktop.org
Subject: Re: [PATCH 11/11] drm/i915: Allow user control over preempt timeout on the important context
Date: Mon, 26 Mar 2018 18:09:29 +0100 [thread overview]
Message-ID: <8fc9fc36-1fa9-947b-85f4-d52668563a0b@linux.intel.com> (raw)
In-Reply-To: <20180326115044.2505-12-chris@chris-wilson.co.uk>
On 26/03/2018 12:50, Chris Wilson wrote:
> EGL_NV_realtime_priority?
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
> drivers/gpu/drm/i915/i915_gem_context.c | 22 ++++++++++++++++++++++
> drivers/gpu/drm/i915/i915_gem_context.h | 13 +++++++++++++
> drivers/gpu/drm/i915/i915_request.c | 8 ++++++--
> drivers/gpu/drm/i915/intel_lrc.c | 2 +-
> include/uapi/drm/i915_drm.h | 12 ++++++++++++
> 5 files changed, 54 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
> index 5cfac0255758..dfb0a2b698c3 100644
> --- a/drivers/gpu/drm/i915/i915_gem_context.c
> +++ b/drivers/gpu/drm/i915/i915_gem_context.c
> @@ -749,6 +749,15 @@ int i915_gem_context_getparam_ioctl(struct drm_device *dev, void *data,
> case I915_CONTEXT_PARAM_PRIORITY:
> args->value = ctx->priority;
> break;
> + case I915_CONTEXT_PARAM_PREEMPT_TIMEOUT:
> + if (!(to_i915(dev)->caps.scheduler & I915_SCHEDULER_CAP_PREEMPTION))
> + ret = -ENODEV;
> + else if (args->size)
> + ret = -EINVAL;
> + else
> + args->value = ctx->preempt_timeout;
> + break;
> +
> default:
> ret = -EINVAL;
> break;
> @@ -824,6 +833,19 @@ int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data,
> }
> break;
>
> + case I915_CONTEXT_PARAM_PREEMPT_TIMEOUT:
> + if (args->size)
> + ret = -EINVAL;
> + else if (args->value >= NSEC_PER_SEC)
> + ret = -EINVAL;
> + else if (!(to_i915(dev)->caps.scheduler & I915_SCHEDULER_CAP_PREEMPTION))
> + ret = -ENODEV;
> + else if (args->value && !capable(CAP_SYS_ADMIN))
> + ret = -EPERM;
> + else
> + ctx->preempt_timeout = args->value;
> + break;
> +
> default:
> ret = -EINVAL;
> break;
> diff --git a/drivers/gpu/drm/i915/i915_gem_context.h b/drivers/gpu/drm/i915/i915_gem_context.h
> index 7854262ddfd9..74d4cadd729e 100644
> --- a/drivers/gpu/drm/i915/i915_gem_context.h
> +++ b/drivers/gpu/drm/i915/i915_gem_context.h
> @@ -150,6 +150,19 @@ struct i915_gem_context {
> */
> int priority;
>
> + /**
> + * @preempt_timeout: QoS guarantee for the high priority context
> + *
> + * Some clients need a guarantee that they will start executing
> + * within a certain window, even at the expense of others. This entails
> + * that if a preemption request is not honoured by the active context
> + * within the timeout, we will reset the GPU to evict the hog and
> + * run the high priority context instead.
> + *
> + * Timeout is stored in nanoseconds; exclusive max of 1s.
Why did you think we would want to limit it to 1s?
> + */
> + u32 preempt_timeout;
> +
> /** ggtt_offset_bias: placement restriction for context objects */
> u32 ggtt_offset_bias;
>
> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> index 9d8dcebd9649..2cd4ea75d127 100644
> --- a/drivers/gpu/drm/i915/i915_request.c
> +++ b/drivers/gpu/drm/i915/i915_request.c
> @@ -1101,8 +1101,12 @@ void __i915_request_add(struct i915_request *request, bool flush_caches)
> * run at the earliest possible convenience.
> */
> rcu_read_lock();
> - if (engine->schedule)
> - engine->schedule(request, request->ctx->priority, 0);
> + if (engine->schedule) {
> + unsigned int timeout = request->ctx->preempt_timeout;
> + int priority = request->ctx->priority;
> +
> + engine->schedule(request, priority, timeout);
> + }
> rcu_read_unlock();
>
> local_bh_disable();
> diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
> index e266657851e1..e782a621b40b 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/intel_lrc.c
> @@ -1148,7 +1148,7 @@ static void execlists_submit_request(struct i915_request *request)
> spin_lock_irqsave(&engine->timeline->lock, flags);
>
> queue_request(engine, &request->priotree, rq_prio(request));
> - submit_queue(engine, rq_prio(request), 0);
> + submit_queue(engine, rq_prio(request), request->ctx->preempt_timeout);
>
> GEM_BUG_ON(!engine->execlists.first);
> GEM_BUG_ON(list_empty(&request->priotree.link));
> diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
> index 7f5634ce8e88..6f10bbe90304 100644
> --- a/include/uapi/drm/i915_drm.h
> +++ b/include/uapi/drm/i915_drm.h
> @@ -1456,6 +1456,18 @@ struct drm_i915_gem_context_param {
> #define I915_CONTEXT_MAX_USER_PRIORITY 1023 /* inclusive */
> #define I915_CONTEXT_DEFAULT_PRIORITY 0
> #define I915_CONTEXT_MIN_USER_PRIORITY -1023 /* inclusive */
> +
> +/*
> + * I915_CONTEXT_PARAM_PREEMPT_TIMEOUT:
> + *
> + * Preemption timeout give in nanoseconds.
> + *
> + * Only allowed for privileged clients (CAP_SYS_ADMIN), this property allows
> + * the preempting context to kick out a GPU hog using a GPU reset if they do
> + * not honour the preemption request in time.
> + */
> +#define I915_CONTEXT_PARAM_PREEMPT_TIMEOUT 0x7
> +
Since I expressed a concern on the "force preemption" naming in the
other thread, I feel obliged to say that I am OK with PREEMPT_TIMEOUT. :)
Regards,
Tvrtko
> __u64 value;
> };
>
>
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
next prev parent reply other threads:[~2018-03-26 17:09 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-03-26 11:50 Forced preemption gedankenexperiment Chris Wilson
2018-03-26 11:50 ` [PATCH 01/11] drm/i915/execlists: Avoid kicking the submission too early for rescheduling Chris Wilson
2018-03-27 11:34 ` Mika Kuoppala
2018-03-27 11:47 ` Chris Wilson
2018-03-27 12:18 ` Mika Kuoppala
2018-03-27 13:34 ` Chris Wilson
2018-03-27 11:42 ` Mika Kuoppala
2018-03-26 11:50 ` [PATCH 02/11] drm/i915/execlists: Clear user-active flag on preemption completion Chris Wilson
2018-03-27 10:00 ` Chris Wilson
2018-03-27 10:01 ` Chris Wilson
2018-03-26 11:50 ` [PATCH 03/11] drm/i915: Include submission tasklet state in engine dump Chris Wilson
2018-03-27 8:37 ` Mika Kuoppala
2018-03-26 11:50 ` [PATCH 04/11] drm/i915/execlists: Refactor out complete_preempt_context() Chris Wilson
2018-03-26 11:50 ` [PATCH 05/11] drm/i915: Move engine reset prepare/finish to backends Chris Wilson
2018-03-26 11:50 ` [PATCH 06/11] drm/i915: Split execlists/guc reset prepartions Chris Wilson
2018-03-26 11:50 ` [PATCH 07/11] drm/i915/execlists: Flush pending preemption events during reset Chris Wilson
2018-03-27 11:44 ` [PATCH v2] " Chris Wilson
2018-03-27 15:33 ` Jeff McGee
2018-03-26 11:50 ` [PATCH 08/11] drm/i915/execlists: Force preemption via reset on timeout Chris Wilson
2018-03-27 8:51 ` Tvrtko Ursulin
2018-03-27 9:10 ` Chris Wilson
2018-03-27 9:23 ` Chris Wilson
2018-03-28 8:59 ` Chris Wilson
2018-03-27 15:40 ` Jeff McGee
2018-03-27 15:50 ` Jeff McGee
2018-03-26 11:50 ` [PATCH 09/11] drm/i915/preemption: Select timeout when scheduling Chris Wilson
2018-03-26 11:50 ` [PATCH 10/11] drm/i915: Use a preemption timeout to enforce interactivity Chris Wilson
2018-03-27 8:35 ` Tvrtko Ursulin
2018-03-27 8:39 ` Chris Wilson
2018-03-27 8:57 ` Tvrtko Ursulin
2018-03-26 11:50 ` [PATCH 11/11] drm/i915: Allow user control over preempt timeout on the important context Chris Wilson
2018-03-26 17:09 ` Tvrtko Ursulin [this message]
2018-03-26 19:52 ` Chris Wilson
2018-03-26 20:49 ` Chris Wilson
2018-03-26 12:08 ` ✗ Fi.CI.CHECKPATCH: warning for series starting with [01/11] drm/i915/execlists: Avoid kicking the submission too early for rescheduling Patchwork
2018-03-26 12:23 ` ✓ Fi.CI.BAT: success " Patchwork
2018-03-26 14:56 ` ✗ Fi.CI.IGT: failure " Patchwork
2018-03-27 12:28 ` ✗ Fi.CI.CHECKPATCH: warning for series starting with [01/11] drm/i915/execlists: Avoid kicking the submission too early for rescheduling (rev2) Patchwork
2018-03-27 12:44 ` ✓ Fi.CI.BAT: success " Patchwork
2018-03-27 15:30 ` ✗ Fi.CI.IGT: failure " Patchwork
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=8fc9fc36-1fa9-947b-85f4-d52668563a0b@linux.intel.com \
--to=tvrtko.ursulin@linux.intel.com \
--cc=chris@chris-wilson.co.uk \
--cc=intel-gfx@lists.freedesktop.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox