From: Naresh Kumar Kachhi <naresh.kumar.kac...@intel.com>

GPU idleness is tracked by checking the request queue. Whenever
request queue is empty we assume that GPU is idle. When a new
set of commands sheduled on ring we call i915_add_request to
make sure these commands are tracked properly. However there are
few places which are not being treacked currently.
This patch introduces a new function add_request_wo_flush to track
such requests. add_request_wo_flush is same as add_request, only
difference is that it will not cause a flush. This is to avoid
any extra overhead while adding new request.

To make sure Gfx is in D0 while there are still commands pending
on ring following is done.
- All the ioctls are already covered with get/put this makes sure
  at the time of scheduling commands on GPU, Gfx is in D0
- Once command scheduling is done, we call add_request to track
  ring activity.
- We call get_noresume if this is first request (ioctl is already
  covered with get_sync).
- put is called only when request_list becomes empty. i.e GPU is
  idle and there are no pending commands on the rings

Note: Make sure we don't do multiple add_request in same
ioctl/callback only one in the end is enough

Signed-off-by: Naresh Kumar Kachhi <naresh.kumar.kac...@intel.com>
---
 drivers/gpu/drm/i915/i915_dma.c            |  5 ++++
 drivers/gpu/drm/i915/i915_drv.h            | 10 +++++--
 drivers/gpu/drm/i915/i915_gem.c            | 27 ++++++++++++-----
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |  4 +--
 drivers/gpu/drm/i915/intel_display.c       |  1 +
 drivers/gpu/drm/i915/intel_drv.h           |  3 ++
 drivers/gpu/drm/i915/intel_pm.c            | 47 ++++++++++++++++++++++++++++++
 7 files changed, 84 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
index ee9502b..b5af745 100644
--- a/drivers/gpu/drm/i915/i915_dma.c
+++ b/drivers/gpu/drm/i915/i915_dma.c
@@ -454,6 +454,7 @@ static int i915_dispatch_cmdbuffer(struct drm_device * dev,
                                   struct drm_clip_rect *cliprects,
                                   void *cmdbuf)
 {
+       struct drm_i915_private *dev_priv = dev->dev_private;
        int nbox = cmd->num_cliprects;
        int i = 0, count, ret;
 
@@ -480,6 +481,7 @@ static int i915_dispatch_cmdbuffer(struct drm_device * dev,
        }
 
        i915_emit_breadcrumb(dev);
+       i915_add_request_wo_flush(LP_RING(dev_priv));
        return 0;
 }
 
@@ -542,6 +544,7 @@ static int i915_dispatch_batchbuffer(struct drm_device * 
dev,
        }
 
        i915_emit_breadcrumb(dev);
+       i915_add_request_wo_flush(LP_RING(dev_priv));
        return 0;
 }
 
@@ -595,6 +598,7 @@ static int i915_dispatch_flip(struct drm_device * dev)
                ADVANCE_LP_RING();
        }
 
+       i915_add_request_wo_flush(LP_RING(dev_priv));
        master_priv->sarea_priv->pf_current_page = dev_priv->dri1.current_page;
        return 0;
 }
@@ -768,6 +772,7 @@ static int i915_emit_irq(struct drm_device * dev)
                OUT_RING(dev_priv->dri1.counter);
                OUT_RING(MI_USER_INTERRUPT);
                ADVANCE_LP_RING();
+               i915_add_request_wo_flush(LP_RING(dev_priv));
        }
 
        return dev_priv->dri1.counter;
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 56c720b..d1399f9 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1324,6 +1324,7 @@ struct i915_package_c8 {
 
 struct i915_runtime_pm {
        bool suspended;
+       bool gpu_idle;
 };
 
 enum intel_pipe_crc_source {
@@ -2063,7 +2064,7 @@ static inline void i915_gem_object_unpin_pages(struct 
drm_i915_gem_object *obj)
 
 int __must_check i915_mutex_lock_interruptible(struct drm_device *dev);
 int i915_gem_object_sync(struct drm_i915_gem_object *obj,
-                        struct intel_ring_buffer *to);
+                        struct intel_ring_buffer *to, bool add_request);
 void i915_vma_move_to_active(struct i915_vma *vma,
                             struct intel_ring_buffer *ring);
 int i915_gem_dumb_create(struct drm_file *file_priv,
@@ -2139,9 +2140,12 @@ int __must_check i915_gem_suspend(struct drm_device 
*dev);
 int __i915_add_request(struct intel_ring_buffer *ring,
                       struct drm_file *file,
                       struct drm_i915_gem_object *batch_obj,
-                      u32 *seqno);
+                      u32 *seqno,
+                      bool flush_caches);
 #define i915_add_request(ring, seqno) \
-       __i915_add_request(ring, NULL, NULL, seqno)
+       __i915_add_request(ring, NULL, NULL, seqno, true)
+#define i915_add_request_wo_flush(ring) \
+       __i915_add_request(ring, NULL, NULL, NULL, false)
 int __must_check i915_wait_seqno(struct intel_ring_buffer *ring,
                                 uint32_t seqno);
 int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 024e454..3e8202e 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2136,7 +2136,8 @@ i915_gem_get_seqno(struct drm_device *dev, u32 *seqno)
 int __i915_add_request(struct intel_ring_buffer *ring,
                       struct drm_file *file,
                       struct drm_i915_gem_object *obj,
-                      u32 *out_seqno)
+                      u32 *out_seqno,
+                      bool flush_caches)
 {
        drm_i915_private_t *dev_priv = ring->dev->dev_private;
        struct drm_i915_gem_request *request;
@@ -2152,9 +2153,11 @@ int __i915_add_request(struct intel_ring_buffer *ring,
         * is that the flush _must_ happen before the next request, no matter
         * what.
         */
-       ret = intel_ring_flush_all_caches(ring);
-       if (ret)
-               return ret;
+       if (flush_caches) {
+               ret = intel_ring_flush_all_caches(ring);
+               if (ret)
+                       return ret;
+       }
 
        request = ring->preallocated_lazy_request;
        if (WARN_ON(request == NULL))
@@ -2219,6 +2222,7 @@ int __i915_add_request(struct intel_ring_buffer *ring,
                                           &dev_priv->mm.retire_work,
                                           round_jiffies_up_relative(HZ));
                        intel_mark_busy(dev_priv->dev);
+                       intel_runtime_pm_gpu_busy(dev_priv);
                }
        }
 
@@ -2544,10 +2548,12 @@ i915_gem_retire_requests(struct drm_device *dev)
                idle &= list_empty(&ring->request_list);
        }
 
-       if (idle)
+       if (idle) {
                mod_delayed_work(dev_priv->wq,
                                   &dev_priv->mm.idle_work,
                                   msecs_to_jiffies(100));
+               intel_runtime_pm_gpu_idle(dev_priv);
+       }
 
        return idle;
 }
@@ -2691,6 +2697,8 @@ out:
  *
  * @obj: object which may be in use on another ring.
  * @to: ring we wish to use the object on. May be NULL.
+ * @add_request: do we need to add a request to track operations
+ *    submitted on ring with sync_to function
  *
  * This code is meant to abstract object synchronization with the GPU.
  * Calling with NULL implies synchronizing the object with the CPU
@@ -2700,7 +2708,7 @@ out:
  */
 int
 i915_gem_object_sync(struct drm_i915_gem_object *obj,
-                    struct intel_ring_buffer *to)
+                    struct intel_ring_buffer *to, bool add_request)
 {
        struct intel_ring_buffer *from = obj->ring;
        u32 seqno;
@@ -2724,12 +2732,15 @@ i915_gem_object_sync(struct drm_i915_gem_object *obj,
 
        trace_i915_gem_ring_sync_to(from, to, seqno);
        ret = to->sync_to(to, from, seqno);
-       if (!ret)
+       if (!ret) {
                /* We use last_read_seqno because sync_to()
                 * might have just caused seqno wrap under
                 * the radar.
                 */
                from->sync_seqno[idx] = obj->last_read_seqno;
+               if (add_request)
+                       i915_add_request_wo_flush(to);
+       }
 
        return ret;
 }
@@ -3707,7 +3718,7 @@ i915_gem_object_pin_to_display_plane(struct 
drm_i915_gem_object *obj,
        int ret;
 
        if (pipelined != obj->ring) {
-               ret = i915_gem_object_sync(obj, pipelined);
+               ret = i915_gem_object_sync(obj, pipelined, true);
                if (ret)
                        return ret;
        }
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c 
b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 0c6bcff..bda7a06 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -832,7 +832,7 @@ i915_gem_execbuffer_move_to_gpu(struct intel_ring_buffer 
*ring,
 
        list_for_each_entry(vma, vmas, exec_list) {
                struct drm_i915_gem_object *obj = vma->obj;
-               ret = i915_gem_object_sync(obj, ring);
+               ret = i915_gem_object_sync(obj, ring, false);
                if (ret)
                        return ret;
 
@@ -969,7 +969,7 @@ i915_gem_execbuffer_retire_commands(struct drm_device *dev,
        ring->gpu_caches_dirty = true;
 
        /* Add a breadcrumb for the completion of the batch buffer */
-       (void)__i915_add_request(ring, file, obj, NULL);
+       (void)__i915_add_request(ring, file, obj, NULL, true);
 }
 
 static int
diff --git a/drivers/gpu/drm/i915/intel_display.c 
b/drivers/gpu/drm/i915/intel_display.c
index ec96002..25eae03 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -8624,6 +8624,7 @@ static int intel_gen7_queue_flip(struct drm_device *dev,
 
        intel_mark_page_flip_active(intel_crtc);
        __intel_ring_advance(ring);
+       i915_add_request_wo_flush(ring);
        return 0;
 
 err_unpin:
diff --git a/drivers/gpu/drm/i915/intel_drv.h b/drivers/gpu/drm/i915/intel_drv.h
index 7b3c209..9061aa7 100644
--- a/drivers/gpu/drm/i915/intel_drv.h
+++ b/drivers/gpu/drm/i915/intel_drv.h
@@ -881,9 +881,12 @@ void gen6_rps_boost(struct drm_i915_private *dev_priv);
 void intel_aux_display_runtime_get(struct drm_i915_private *dev_priv);
 void intel_aux_display_runtime_put(struct drm_i915_private *dev_priv);
 void intel_runtime_pm_get(struct drm_i915_private *dev_priv);
+void intel_runtime_pm_get_noresume(struct drm_i915_private *dev_priv);
 void intel_runtime_pm_put(struct drm_i915_private *dev_priv);
 void intel_init_runtime_pm(struct drm_i915_private *dev_priv);
 void intel_fini_runtime_pm(struct drm_i915_private *dev_priv);
+void intel_runtime_pm_gpu_busy(struct drm_i915_private *dev_priv);
+void intel_runtime_pm_gpu_idle(struct drm_i915_private *dev_priv);
 void ilk_wm_get_hw_state(struct drm_device *dev);
 
 
diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index b9b4fe4..991ff62 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -5470,6 +5470,37 @@ void intel_aux_display_runtime_put(struct 
drm_i915_private *dev_priv)
        hsw_enable_package_c8(dev_priv);
 }
 
+void intel_runtime_pm_gpu_idle(struct drm_i915_private *dev_priv)
+{
+       if (!HAS_RUNTIME_PM(dev_priv->dev))
+               return;
+
+       /* don't need a seperate mutex here as callers are
+        * already under struct_mutex
+        */
+       WARN_ON(!mutex_is_locked(&dev_priv->dev->struct_mutex));
+       if (!dev_priv->pm.gpu_idle) {
+               dev_priv->pm.gpu_idle = true;
+               /* match with get in gpu_busy */
+               intel_runtime_pm_put(dev_priv);
+       }
+}
+
+void intel_runtime_pm_gpu_busy(struct drm_i915_private *dev_priv)
+{
+       if (!HAS_RUNTIME_PM(dev_priv->dev))
+               return;
+
+       WARN_ON(!mutex_is_locked(&dev_priv->dev->struct_mutex));
+       if (dev_priv->pm.gpu_idle) {
+               dev_priv->pm.gpu_idle = false;
+               /* make sure that we keep the GPU on until request list
+                * is empty
+                */
+               intel_runtime_pm_get_noresume(dev_priv);
+       }
+}
+
 void intel_runtime_pm_get(struct drm_i915_private *dev_priv)
 {
        struct drm_device *dev = dev_priv->dev;
@@ -5482,6 +5513,21 @@ void intel_runtime_pm_get(struct drm_i915_private 
*dev_priv)
        WARN(dev_priv->pm.suspended, "Device still suspended.\n");
 }
 
+void intel_runtime_pm_get_noresume(struct drm_i915_private *dev_priv)
+{
+       struct drm_device *dev = dev_priv->dev;
+       struct device *device = &dev->pdev->dev;
+
+       if (!HAS_RUNTIME_PM(dev))
+               return;
+
+       /* driver calls no resume when it is sure that device is
+        * already active and just want to increment the ref count
+        */
+       WARN(dev_priv->pm.suspended, "Device suspended. call get_sync?\n");
+       pm_runtime_get_noresume(device);
+}
+
 void intel_runtime_pm_put(struct drm_i915_private *dev_priv)
 {
        struct drm_device *dev = dev_priv->dev;
@@ -5500,6 +5546,7 @@ void intel_init_runtime_pm(struct drm_i915_private 
*dev_priv)
        struct device *device = &dev->pdev->dev;
 
        dev_priv->pm.suspended = false;
+       dev_priv->pm.gpu_idle = true;
 
        if (!HAS_RUNTIME_PM(dev))
                return;
-- 
1.8.1.2

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

Reply via email to