<div dir="ltr"><div dir="ltr">Hi Tvrtko, </div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Tue, Nov 6, 2018 at 3:14 PM Tvrtko Ursulin <<a href="mailto:tvrtko.ursulin@linux.intel.com">tvrtko.ursulin@linux.intel.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><br>
On 06/11/2018 04:13, Ankit Navik wrote:<br>
> From: Praveen Diwakar <<a href="mailto:praveen.diwakar@intel.com" target="_blank">praveen.diwakar@intel.com</a>><br>
> <br>
> This patch gives us the active pending request count which is yet<br>
> to be submitted to the GPU<br>
> <br>
> Signed-off-by: Praveen Diwakar <<a href="mailto:praveen.diwakar@intel.com" target="_blank">praveen.diwakar@intel.com</a>><br>
> Signed-off-by: Yogesh Marathe <<a href="mailto:yogesh.marathe@intel.com" target="_blank">yogesh.marathe@intel.com</a>><br>
> Signed-off-by: Aravindan Muthukumar <<a href="mailto:aravindan.muthukumar@intel.com" target="_blank">aravindan.muthukumar@intel.com</a>><br>
> Signed-off-by: Kedar J Karanje <<a href="mailto:kedar.j.karanje@intel.com" target="_blank">kedar.j.karanje@intel.com</a>><br>
> Signed-off-by: Ankit Navik <<a href="mailto:ankit.p.navik@intel.com" target="_blank">ankit.p.navik@intel.com</a>><br>
> Suggested-by: Tvrtko Ursulin <<a href="mailto:tvrtko.ursulin@linux.intel.com" target="_blank">tvrtko.ursulin@linux.intel.com</a>><br>
> ---<br>
> drivers/gpu/drm/i915/i915_drv.c | 1 +<br>
> drivers/gpu/drm/i915/i915_drv.h | 5 +++++<br>
> drivers/gpu/drm/i915/i915_gem_context.c | 1 +<br>
> drivers/gpu/drm/i915/i915_gem_context.h | 6 ++++++<br>
> drivers/gpu/drm/i915/i915_gem_execbuffer.c | 5 +++++<br>
> drivers/gpu/drm/i915/intel_lrc.c | 6 ++++++<br>
> 6 files changed, 24 insertions(+)<br>
> <br>
> diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c<br>
> index f8cfd16..d37c46e 100644<br>
> --- a/drivers/gpu/drm/i915/i915_drv.c<br>
> +++ b/drivers/gpu/drm/i915/i915_drv.c<br>
> @@ -903,6 +903,7 @@ static int i915_driver_init_early(struct drm_i915_private *dev_priv,<br>
> mutex_init(&dev_priv->av_mutex);<br>
> mutex_init(&dev_priv->wm.wm_mutex);<br>
> mutex_init(&dev_priv->pps_mutex);<br>
> + mutex_init(&dev_priv->pred_mutex);<br>
> <br>
> i915_memcpy_init_early(dev_priv);<br>
> <br>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h<br>
> index 4aca534..137ec33 100644<br>
> --- a/drivers/gpu/drm/i915/i915_drv.h<br>
> +++ b/drivers/gpu/drm/i915/i915_drv.h<br>
> @@ -1609,6 +1609,11 @@ struct drm_i915_private {<br>
> * controller on different i2c buses. */<br>
> struct mutex gmbus_mutex;<br>
> <br>
> + /** pred_mutex protects against councurrent usage of pending<br>
> + * request counter for multiple contexts<br>
> + */<br>
> + struct mutex pred_mutex;<br>
> +<br>
> /**<br>
> * Base address of the gmbus and gpio block.<br>
> */<br>
> diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c<br>
> index b10770c..0bcbe32 100644<br>
> --- a/drivers/gpu/drm/i915/i915_gem_context.c<br>
> +++ b/drivers/gpu/drm/i915/i915_gem_context.c<br>
> @@ -387,6 +387,7 @@ i915_gem_create_context(struct drm_i915_private *dev_priv,<br>
> }<br>
> <br>
> trace_i915_context_create(ctx);<br>
> + atomic_set(&ctx->req_cnt, 0);<br>
> <br>
> return ctx;<br>
> }<br>
> diff --git a/drivers/gpu/drm/i915/i915_gem_context.h b/drivers/gpu/drm/i915/i915_gem_context.h<br>
> index b116e49..04e3ff7 100644<br>
> --- a/drivers/gpu/drm/i915/i915_gem_context.h<br>
> +++ b/drivers/gpu/drm/i915/i915_gem_context.h<br>
> @@ -194,6 +194,12 @@ struct i915_gem_context {<br>
> * context close.<br>
> */<br>
> struct list_head handles_list;<br>
> +<br>
> + /** req_cnt: tracks the pending commands, based on which we decide to<br>
> + * go for low/medium/high load configuration of the GPU, this is<br>
> + * controlled via a mutex<br>
> + */<br>
> + atomic_t req_cnt;<br>
> };<br>
> <br>
> static inline bool i915_gem_context_is_closed(const struct i915_gem_context *ctx)<br>
> diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c<br>
> index 3f0c612..8afa2a5 100644<br>
> --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c<br>
> +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c<br>
> @@ -2178,6 +2178,7 @@ i915_gem_do_execbuffer(struct drm_device *dev,<br>
> struct drm_syncobj **fences)<br>
> {<br>
> struct i915_execbuffer eb;<br>
> + struct drm_i915_private *dev_priv = to_i915(dev);<br>
> struct dma_fence *in_fence = NULL;<br>
> struct sync_file *out_fence = NULL;<br>
> int out_fence_fd = -1;<br>
> @@ -2390,6 +2391,10 @@ i915_gem_do_execbuffer(struct drm_device *dev,<br>
> */<br>
> eb.request->batch = eb.batch;<br>
> <br>
> + mutex_lock(&dev_priv->pred_mutex);<br>
> + atomic_inc(&eb.ctx->req_cnt);<br>
<br>
Point of going to atomic_t was to remove need for the mutex.<br>
<br>
> + mutex_unlock(&dev_priv->pred_mutex);<br>
> +<br>
> trace_i915_request_queue(eb.request, eb.batch_flags);<br>
> err = eb_submit(&eb);<br>
> err_request:<br>
> diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c<br>
> index 1744792..bcbb66b 100644<br>
> --- a/drivers/gpu/drm/i915/intel_lrc.c<br>
> +++ b/drivers/gpu/drm/i915/intel_lrc.c<br>
> @@ -728,6 +728,12 @@ static void execlists_dequeue(struct intel_engine_cs *engine)<br>
> trace_i915_request_in(rq, port_index(port, execlists));<br>
> last = rq;<br>
> submit = true;<br>
> +<br>
> + mutex_lock(&rq->i915->pred_mutex);<br>
> + if (atomic_read(&rq->gem_context->req_cnt) > 0)<br>
> + atomic_dec(&rq->gem_context->req_cnt);<br>
<br>
Hitting underflow is a hint accounting does not work as expected. I <br>
really think you need to fix it by gathering some ideas from the patches <br>
I've pointed at in the previous round.<br></blockquote><div><br></div><div>I have submitted the patch v4. <br>I have tried with point which you have suggested, but didnt see much</div><div>power benefit with that. </div><div><br></div><div>Regards, Ankit</div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<br>
And there is also GuC to think about.<br>
<br>
Regards,<br>
<br>
Tvrtko<br>
<br>
> +<br>
> + mutex_unlock(&rq->i915->pred_mutex);<br>
> }<br>
> <br>
> rb_erase_cached(&p->node, &execlists->queue);<br>
> <br>
_______________________________________________<br>
Intel-gfx mailing list<br>
<a href="mailto:Intel-gfx@lists.freedesktop.org" target="_blank">Intel-gfx@lists.freedesktop.org</a><br>
<a href="https://lists.freedesktop.org/mailman/listinfo/intel-gfx" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/intel-gfx</a><br>
</blockquote></div></div>