intel-gfx.lists.freedesktop.org archive mirror
 help / color / mirror / Atom feed
From: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
To: Chris Wilson <chris@chris-wilson.co.uk>, intel-gfx@lists.freedesktop.org
Subject: Re: [PATCH v3 12/14] drm/i915/scheduler: Support user-defined priorities
Date: Mon, 14 Nov 2016 11:32:41 +0000	[thread overview]
Message-ID: <b4fc350d-7607-1a27-f190-35bdd59c6fb7@linux.intel.com> (raw)
In-Reply-To: <20161114085703.16540-12-chris@chris-wilson.co.uk>


On 14/11/2016 08:57, Chris Wilson wrote:
> Use a priority stored in the context as the initial value when
> submitting a request. This allows us to change the default priority on a
> per-context basis, allowing different contexts to be favoured with GPU
> time at the expense of lower importance work. The user can adjust the
> context's priority via I915_CONTEXT_PARAM_PRIORITY, with more positive
> values being higher priority (they will be serviced earlier, after their
> dependencies have been resolved). Any prerequisite work for an execbuf
> will have its priority raised to match the new request as required.
>
> Normal users can specify any value in the range of -1023 to 0 [default],
> i.e. they can reduce the priority of their workloads (and temporarily
> boost it back to normal if so desired).
>
> Privileged users can specify any value in the range of -1023 to 1023,
> [default is 0], i.e. they can raise their priority above all overs and
> so potentially starve the system.
>
> Note that the existing schedulers are not fair, nor load balancing, the
> execution is strictly by priority on a first-come, first-served basis,
> and the driver may choose to boost some requests above the range
> available to users.
>
> This priority was originally based around nice(2), but evolved to allow
> clients to adjust their priority within a small range, and allow for a
> privileged high priority range.
>
> For example, this can be used to implement EGL_IMG_context_priority
> https://www.khronos.org/registry/egl/extensions/IMG/EGL_IMG_context_priority.txt
>
> 	EGL_CONTEXT_PRIORITY_LEVEL_IMG determines the priority level of
>         the context to be created. This attribute is a hint, as an
>         implementation may not support multiple contexts at some
>         priority levels and system policy may limit access to high
>         priority contexts to appropriate system privilege level. The
>         default value for EGL_CONTEXT_PRIORITY_LEVEL_IMG is
>         EGL_CONTEXT_PRIORITY_MEDIUM_IMG."
>
> so we can map
>
> 	PRIORITY_HIGH -> 1023 [privileged, will failback to 0]
> 	PRIORITY_MED -> 0 [default]
> 	PRIORITY_LOW -> -1023
>
> They also map onto the priorities used by VkQueue (and a VkQueue is
> essentially a timeline, our i915_gem_context under full-ppgtt).
>
> Testcase: igt/gem_exec_schedule
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/i915/i915_gem_context.c | 20 ++++++++++++++++++++
>  include/uapi/drm/i915_drm.h             |  3 +++
>  2 files changed, 23 insertions(+)
>
> diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
> index 1f94b8d6d83d..1f74ab266f6b 100644
> --- a/drivers/gpu/drm/i915/i915_gem_context.c
> +++ b/drivers/gpu/drm/i915/i915_gem_context.c
> @@ -1101,6 +1101,9 @@ int i915_gem_context_getparam_ioctl(struct drm_device *dev, void *data,
>  	case I915_CONTEXT_PARAM_NO_ERROR_CAPTURE:
>  		args->value = !!(ctx->flags & CONTEXT_NO_ERROR_CAPTURE);
>  		break;
> +	case I915_CONTEXT_PARAM_PRIORITY:
> +		args->value = ctx->priority;
> +		break;
>  	default:
>  		ret = -EINVAL;
>  		break;
> @@ -1156,6 +1159,23 @@ int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data,
>  				ctx->flags &= ~CONTEXT_NO_ERROR_CAPTURE;
>  		}
>  		break;
> +
> +	case I915_CONTEXT_PARAM_PRIORITY:
> +		{
> +			int priority = args->value;
> +
> +			if (args->size)
> +				ret = -EINVAL;
> +			else if (priority >= I915_PRIORITY_MAX ||
> +				 priority <= -I915_PRIORITY_MAX)
> +				ret = -EINVAL;
> +			else if (priority > 0 && !capable(CAP_SYS_ADMIN))
> +				ret = -EPERM;
> +			else
> +				ctx->priority = priority;
> +		}
> +		break;
> +
>  	default:
>  		ret = -EINVAL;
>  		break;
> diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
> index 1c12a350eca3..47901a8ad682 100644
> --- a/include/uapi/drm/i915_drm.h
> +++ b/include/uapi/drm/i915_drm.h
> @@ -391,6 +391,8 @@ typedef struct drm_i915_irq_wait {
>
>  /* Query whether DRM_I915_GEM_EXECBUFFER2 supports user defined execution
>   * priorities and the driver will attempt to execute batches in priority order.
> + * The initial priority for each batch is supplied by the context and is
> + * controlled via I915_CONTEXT_PARAM_PRIORITY.
>   */
>  #define I915_PARAM_HAS_SCHEDULER	 41
>
> @@ -1224,6 +1226,7 @@ struct drm_i915_gem_context_param {
>  #define I915_CONTEXT_PARAM_NO_ZEROMAP	0x2
>  #define I915_CONTEXT_PARAM_GTT_SIZE	0x3
>  #define I915_CONTEXT_PARAM_NO_ERROR_CAPTURE	0x4
> +#define I915_CONTEXT_PARAM_PRIORITY	0x5
>  	__u64 value;
>  };
>
>

Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Regards,

Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

  reply	other threads:[~2016-11-14 11:32 UTC|newest]

Thread overview: 82+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-11-07 13:59 Trivial scheduler, take 2 Chris Wilson
2016-11-07 13:59 ` [PATCH v2 01/11] drm/i915: Create distinct lockclasses for execution vs user timelines Chris Wilson
2016-11-08  7:43   ` Joonas Lahtinen
2016-11-08  8:50     ` Chris Wilson
2016-11-07 13:59 ` [PATCH v2 02/11] drm/i915: Split request submit/execute phase into two Chris Wilson
2016-11-08  9:06   ` Joonas Lahtinen
2016-11-07 13:59 ` [PATCH v2 03/11] drm/i915: Defer transfer onto execution timeline to actual hw submission Chris Wilson
2016-11-10 10:43   ` Tvrtko Ursulin
2016-11-10 11:11     ` Chris Wilson
2016-11-10 11:51       ` Tvrtko Ursulin
2016-11-10 14:43         ` Chris Wilson
2016-11-10 11:23     ` [PATCH v3] " Chris Wilson
2016-11-07 13:59 ` [PATCH v2 04/11] drm/i915: Remove engine->execlist_lock Chris Wilson
2016-11-07 13:59 ` [PATCH v2 05/11] drm/i915/scheduler: Signal the arrival of a new request Chris Wilson
2016-11-07 13:59 ` [PATCH v2 06/11] drm/i915/scheduler: Record all dependencies upon request construction Chris Wilson
2016-11-08 12:20   ` Chris Wilson
2016-11-10 10:44     ` Tvrtko Ursulin
2016-11-10 10:55       ` Chris Wilson
2016-11-10 11:54         ` Tvrtko Ursulin
2016-11-10 12:10           ` Chris Wilson
2016-11-10 14:45   ` Tvrtko Ursulin
2016-11-10 15:01     ` Chris Wilson
2016-11-10 15:36       ` Tvrtko Ursulin
2016-11-10 15:55         ` Chris Wilson
2016-11-07 13:59 ` [PATCH v2 07/11] drm/i915/scheduler: Boost priorities for flips Chris Wilson
2016-11-10 10:52   ` Tvrtko Ursulin
2016-11-07 13:59 ` [PATCH v2 08/11] HACK drm/i915/scheduler: emulate a scheduler for guc Chris Wilson
2016-11-07 13:59 ` [PATCH v2 09/11] drm/i915/scheduler: Support user-defined priorities Chris Wilson
2016-11-10 13:02   ` Tvrtko Ursulin
2016-11-10 13:10     ` Chris Wilson
2016-11-07 13:59 ` [PATCH v2 10/11] drm/i915: Enable userspace to opt-out of implicit fencing Chris Wilson
2016-11-07 13:59 ` [PATCH v2 11/11] drm/i915: Support explicit fencing for execbuf Chris Wilson
2016-11-07 15:18 ` ✓ Fi.CI.BAT: success for series starting with [v2,01/11] drm/i915: Create distinct lockclasses for execution vs user timelines Patchwork
2016-11-10 11:45 ` ✓ Fi.CI.BAT: success for series starting with [v2,01/11] drm/i915: Create distinct lockclasses for execution vs user timelines (rev2) Patchwork
2016-11-10 12:04   ` Saarinen, Jani
2016-11-14  8:56 ` [PATCH v3 01/14] drm/i915: Give each sw_fence its own lockclass Chris Wilson
2016-11-14  8:56   ` [PATCH v3 02/14] drm/i915: Create distinct lockclasses for execution vs user timelines Chris Wilson
2016-11-14  8:56   ` [PATCH v3 03/14] drm/i915: Split request submit/execute phase into two Chris Wilson
2016-11-14  8:56   ` [PATCH v3 04/14] drm/i915: Defer transfer onto execution timeline to actual hw submission Chris Wilson
2016-11-14 10:59     ` Tvrtko Ursulin
2016-11-14  8:56   ` [PATCH v3 05/14] drm/i915: Remove engine->execlist_lock Chris Wilson
2016-11-14  8:56   ` [PATCH v3 06/14] drm/i915/scheduler: Signal the arrival of a new request Chris Wilson
2016-11-14  8:56   ` [PATCH v3 07/14] drm/i915/scheduler: Record all dependencies upon request construction Chris Wilson
2016-11-14 11:09     ` Tvrtko Ursulin
2016-11-14  8:56   ` [PATCH v3 08/14] drm/i915/scheduler: Execute requests in order of priorities Chris Wilson
2016-11-14 11:15     ` Tvrtko Ursulin
2016-11-14 11:41       ` Chris Wilson
2016-11-14 11:48         ` Tvrtko Ursulin
2016-11-14 14:25           ` Chris Wilson
2016-11-14  8:56   ` [PATCH v3 09/14] drm/i915: Store the execution priority on the context Chris Wilson
2016-11-14 11:16     ` Tvrtko Ursulin
2016-11-14  8:56   ` [PATCH v3 10/14] drm/i915/scheduler: Boost priorities for flips Chris Wilson
2016-11-14  8:57   ` [PATCH v3 11/14] HACK drm/i915/scheduler: emulate a scheduler for guc Chris Wilson
2016-11-14 11:31     ` Tvrtko Ursulin
2016-11-14 14:40       ` Chris Wilson
2016-12-01 10:45     ` Tvrtko Ursulin
2016-12-01 11:18       ` Chris Wilson
2016-12-01 12:45         ` Tvrtko Ursulin
2016-12-01 13:01           ` Chris Wilson
2016-11-14  8:57   ` [PATCH v3 12/14] drm/i915/scheduler: Support user-defined priorities Chris Wilson
2016-11-14 11:32     ` Tvrtko Ursulin [this message]
2016-11-14  8:57   ` [PATCH v3 13/14] drm/i915: Enable userspace to opt-out of implicit fencing Chris Wilson
2017-01-25 20:38     ` Chad Versace
2017-01-26 10:32       ` Chris Wilson
2017-01-26 10:58         ` [PATCH] i965: Share the workaround bo between all contexts Chris Wilson
2017-01-26 17:39           ` [Mesa-dev] " Chad Versace
2017-01-26 18:05             ` Chris Wilson
2017-01-26 23:40               ` Chad Versace
2017-01-26 18:46             ` Chris Wilson
2017-01-27  0:01             ` Chad Versace
2017-01-27 18:20               ` [Intel-gfx] " Emil Velikov
2017-01-27 18:30                 ` [Mesa-dev] " Chris Wilson
2017-01-27 18:37                   ` [Intel-gfx] " Emil Velikov
2017-01-27  0:07         ` [PATCH v3 13/14] drm/i915: Enable userspace to opt-out of implicit fencing Chad Versace
2016-11-14  8:57   ` [PATCH v3 14/14] drm/i915: Support explicit fencing for execbuf Chris Wilson
2016-11-14 22:29     ` Rafael Antognolli
2017-01-25 20:27     ` Chad Versace
2016-11-14  9:01   ` [PATCH v3 01/14] drm/i915: Give each sw_fence its own lockclass Tvrtko Ursulin
2016-11-14  9:05     ` Chris Wilson
2016-11-14 10:57   ` Tvrtko Ursulin
2016-11-14 14:48   ` Joonas Lahtinen
2016-11-14 15:13     ` Chris Wilson

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=b4fc350d-7607-1a27-f190-35bdd59c6fb7@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;
as well as URLs for NNTP newsgroup(s).