A trail of breadcrumbs

A trail of breadcrumbs

[Chris Wilson]

A couple of patches left at the start to clear the regression (LIFO), and
then an almost completely reviewed set of breadcrumbs...
-Chris

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

[PATCH 01/25] drm/i915: Preserve current RPS frequency across init

[Chris Wilson]

Select idle frequency during initialisation, then reset the last known
frequency when re-enabling. This allows us to preserve the user selected
frequency across resets.

v2: Stop CHV from overriding the user's choice in cherryview_enable_rps()

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
---
 drivers/gpu/drm/i915/intel_pm.c | 43 ++++++++++++++++-------------------------
 1 file changed, 17 insertions(+), 26 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index c94521cc476c..78b35b86e80d 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -5098,6 +5098,7 @@ static void gen6_init_rps_frequencies(struct drm_i915_private *dev_priv)
 	}
 
 	dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
+	dev_priv->rps.cur_freq = dev_priv->rps.idle_freq;
 
 	/* Preserve min/max settings in case of re-init */
 	if (dev_priv->rps.max_freq_softlimit == 0)
@@ -5114,6 +5115,15 @@ static void gen6_init_rps_frequencies(struct drm_i915_private *dev_priv)
 	}
 }
 
+static void reset_rps(struct drm_i915_private *dev_priv,
+		      void (*set)(struct drm_i915_private *, u8))
+{
+	u8 freq = dev_priv->rps.cur_freq; /* force a reset */
+	dev_priv->rps.power = -1;
+	dev_priv->rps.cur_freq = -1;
+	set(dev_priv, freq);
+}
+
 /* See the Gen9_GT_PM_Programming_Guide doc for the below */
 static void gen9_enable_rps(struct drm_i915_private *dev_priv)
 {
@@ -5150,8 +5160,7 @@ static void gen9_enable_rps(struct drm_i915_private *dev_priv)
 	/* Leaning on the below call to gen6_set_rps to program/setup the
 	 * Up/Down EI & threshold registers, as well as the RP_CONTROL,
 	 * RP_INTERRUPT_LIMITS & RPNSWREQ registers */
-	dev_priv->rps.power = HIGH_POWER; /* force a reset */
-	gen6_set_rps(dev_priv, dev_priv->rps.idle_freq);
+	reset_rps(dev_priv, gen6_set_rps);
 
 	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
 }
@@ -5297,8 +5306,7 @@ static void gen8_enable_rps(struct drm_i915_private *dev_priv)
 
 	/* 6: Ring frequency + overclocking (our driver does this later */
 
-	dev_priv->rps.power = HIGH_POWER; /* force a reset */
-	gen6_set_rps(dev_priv, dev_priv->rps.idle_freq);
+	reset_rps(dev_priv, gen6_set_rps);
 
 	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
 }
@@ -5391,8 +5399,7 @@ static void gen6_enable_rps(struct drm_i915_private *dev_priv)
 		dev_priv->rps.max_freq = pcu_mbox & 0xff;
 	}
 
-	dev_priv->rps.power = HIGH_POWER; /* force a reset */
-	gen6_set_rps(dev_priv, dev_priv->rps.idle_freq);
+	reset_rps(dev_priv, gen6_set_rps);
 
 	rc6vids = 0;
 	ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
@@ -5756,6 +5763,7 @@ static void valleyview_init_gt_powersave(struct drm_i915_private *dev_priv)
 			 dev_priv->rps.min_freq);
 
 	dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
+	dev_priv->rps.cur_freq = dev_priv->rps.idle_freq;
 
 	/* Preserve min/max settings in case of re-init */
 	if (dev_priv->rps.max_freq_softlimit == 0)
@@ -5820,6 +5828,7 @@ static void cherryview_init_gt_powersave(struct drm_i915_private *dev_priv)
 		  "Odd GPU freq values\n");
 
 	dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
+	dev_priv->rps.cur_freq = dev_priv->rps.idle_freq;
 
 	/* Preserve min/max settings in case of re-init */
 	if (dev_priv->rps.max_freq_softlimit == 0)
@@ -5919,16 +5928,7 @@ static void cherryview_enable_rps(struct drm_i915_private *dev_priv)
 	DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE));
 	DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
 
-	dev_priv->rps.cur_freq = (val >> 8) & 0xff;
-	DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
-			 intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
-			 dev_priv->rps.cur_freq);
-
-	DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
-			 intel_gpu_freq(dev_priv, dev_priv->rps.idle_freq),
-			 dev_priv->rps.idle_freq);
-
-	valleyview_set_rps(dev_priv, dev_priv->rps.idle_freq);
+	reset_rps(dev_priv, valleyview_set_rps);
 
 	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
 }
@@ -6008,16 +6008,7 @@ static void valleyview_enable_rps(struct drm_i915_private *dev_priv)
 	DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE));
 	DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
 
-	dev_priv->rps.cur_freq = (val >> 8) & 0xff;
-	DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
-			 intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
-			 dev_priv->rps.cur_freq);
-
-	DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
-			 intel_gpu_freq(dev_priv, dev_priv->rps.idle_freq),
-			 dev_priv->rps.idle_freq);
-
-	valleyview_set_rps(dev_priv, dev_priv->rps.idle_freq);
+	reset_rps(dev_priv, valleyview_set_rps);
 
 	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
 }
-- 
2.8.1

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

[PATCH 02/25] drm/i915: Remove superfluous powersave work flushing

[Chris Wilson]

Instead of flushing the outstanding enabling, remember the requested
frequency to apply when the powersave work runs.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
---
 drivers/gpu/drm/i915/i915_debugfs.c | 30 ++------------------------
 drivers/gpu/drm/i915/i915_sysfs.c   | 42 +++++++------------------------------
 2 files changed, 10 insertions(+), 62 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index f664884ad27f..2858415a7d0e 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -1193,8 +1193,6 @@ static int i915_frequency_info(struct seq_file *m, void *unused)
 
 	intel_runtime_pm_get(dev_priv);
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
 	if (IS_GEN5(dev)) {
 		u16 rgvswctl = I915_READ16(MEMSWCTL);
 		u16 rgvstat = I915_READ16(MEMSTAT_ILK);
@@ -1882,8 +1880,6 @@ static int i915_ring_freq_table(struct seq_file *m, void *unused)
 
 	intel_runtime_pm_get(dev_priv);
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
 	ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
 	if (ret)
 		goto out;
@@ -4976,20 +4972,11 @@ i915_max_freq_get(void *data, u64 *val)
 {
 	struct drm_device *dev = data;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-	int ret;
 
 	if (INTEL_INFO(dev)->gen < 6)
 		return -ENODEV;
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
-	ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
-	if (ret)
-		return ret;
-
 	*val = intel_gpu_freq(dev_priv, dev_priv->rps.max_freq_softlimit);
-	mutex_unlock(&dev_priv->rps.hw_lock);
-
 	return 0;
 }
 
@@ -5004,8 +4991,6 @@ i915_max_freq_set(void *data, u64 val)
 	if (INTEL_INFO(dev)->gen < 6)
 		return -ENODEV;
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
 	DRM_DEBUG_DRIVER("Manually setting max freq to %llu\n", val);
 
 	ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
@@ -5043,20 +5028,11 @@ i915_min_freq_get(void *data, u64 *val)
 {
 	struct drm_device *dev = data;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-	int ret;
 
-	if (INTEL_INFO(dev)->gen < 6)
+	if (INTEL_GEN(dev_priv) < 6)
 		return -ENODEV;
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
-	ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
-	if (ret)
-		return ret;
-
 	*val = intel_gpu_freq(dev_priv, dev_priv->rps.min_freq_softlimit);
-	mutex_unlock(&dev_priv->rps.hw_lock);
-
 	return 0;
 }
 
@@ -5068,11 +5044,9 @@ i915_min_freq_set(void *data, u64 val)
 	u32 hw_max, hw_min;
 	int ret;
 
-	if (INTEL_INFO(dev)->gen < 6)
+	if (INTEL_GEN(dev_priv) < 6)
 		return -ENODEV;
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
 	DRM_DEBUG_DRIVER("Manually setting min freq to %llu\n", val);
 
 	ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
diff --git a/drivers/gpu/drm/i915/i915_sysfs.c b/drivers/gpu/drm/i915/i915_sysfs.c
index a6e90fe05a1e..915e97cdc4d5 100644
--- a/drivers/gpu/drm/i915/i915_sysfs.c
+++ b/drivers/gpu/drm/i915/i915_sysfs.c
@@ -271,8 +271,6 @@ static ssize_t gt_act_freq_mhz_show(struct device *kdev,
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	int ret;
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
 	intel_runtime_pm_get(dev_priv);
 
 	mutex_lock(&dev_priv->rps.hw_lock);
@@ -303,19 +301,9 @@ static ssize_t gt_cur_freq_mhz_show(struct device *kdev,
 	struct drm_minor *minor = dev_to_drm_minor(kdev);
 	struct drm_device *dev = minor->dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-	int ret;
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
-	intel_runtime_pm_get(dev_priv);
-
-	mutex_lock(&dev_priv->rps.hw_lock);
-	ret = intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq);
-	mutex_unlock(&dev_priv->rps.hw_lock);
-
-	intel_runtime_pm_put(dev_priv);
-
-	return snprintf(buf, PAGE_SIZE, "%d\n", ret);
+	return snprintf(buf, PAGE_SIZE, "%d\n",
+			intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq));
 }
 
 static ssize_t vlv_rpe_freq_mhz_show(struct device *kdev,
@@ -335,15 +323,10 @@ static ssize_t gt_max_freq_mhz_show(struct device *kdev, struct device_attribute
 	struct drm_minor *minor = dev_to_drm_minor(kdev);
 	struct drm_device *dev = minor->dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-	int ret;
-
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
 
-	mutex_lock(&dev_priv->rps.hw_lock);
-	ret = intel_gpu_freq(dev_priv, dev_priv->rps.max_freq_softlimit);
-	mutex_unlock(&dev_priv->rps.hw_lock);
-
-	return snprintf(buf, PAGE_SIZE, "%d\n", ret);
+	return snprintf(buf, PAGE_SIZE, "%d\n",
+			intel_gpu_freq(dev_priv,
+				       dev_priv->rps.max_freq_softlimit));
 }
 
 static ssize_t gt_max_freq_mhz_store(struct device *kdev,
@@ -360,8 +343,6 @@ static ssize_t gt_max_freq_mhz_store(struct device *kdev,
 	if (ret)
 		return ret;
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
 	intel_runtime_pm_get(dev_priv);
 
 	mutex_lock(&dev_priv->rps.hw_lock);
@@ -403,15 +384,10 @@ static ssize_t gt_min_freq_mhz_show(struct device *kdev, struct device_attribute
 	struct drm_minor *minor = dev_to_drm_minor(kdev);
 	struct drm_device *dev = minor->dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-	int ret;
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
-	mutex_lock(&dev_priv->rps.hw_lock);
-	ret = intel_gpu_freq(dev_priv, dev_priv->rps.min_freq_softlimit);
-	mutex_unlock(&dev_priv->rps.hw_lock);
-
-	return snprintf(buf, PAGE_SIZE, "%d\n", ret);
+	return snprintf(buf, PAGE_SIZE, "%d\n",
+			intel_gpu_freq(dev_priv,
+				       dev_priv->rps.min_freq_softlimit));
 }
 
 static ssize_t gt_min_freq_mhz_store(struct device *kdev,
@@ -428,8 +404,6 @@ static ssize_t gt_min_freq_mhz_store(struct device *kdev,
 	if (ret)
 		return ret;
 
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
 	intel_runtime_pm_get(dev_priv);
 
 	mutex_lock(&dev_priv->rps.hw_lock);
-- 
2.8.1

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

[PATCH 03/25] drm/i915: Defer enabling rc6 til after we submit the first batch/context

[Chris Wilson]

Some hardware requires a valid render context before it can initiate
rc6 power gating of the GPU; the default state of the GPU is not
sufficient and may lead to undefined behaviour. The first execution of
any batch will load the "golden render state", at which point it is safe
to enable rc6. As we do not forcibly load the kernel context at resume,
we have to hook into the batch submission to be sure that the render
state is setup before enabling rc6.

However, since we don't enable powersaving until that first batch, we
queued a delayed task in order to guarantee that the batch is indeed
submitted.

v2: Rearrange intel_disable_gt_powersave() to match.
v3: Apply user specified cur_freq (or idle_freq if not set).
v4: Give in, and supply a delayed work to autoenable rc6

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
---
 drivers/gpu/drm/i915/i915_drv.c      |   7 +-
 drivers/gpu/drm/i915/intel_display.c |   3 +-
 drivers/gpu/drm/i915/intel_drv.h     |   2 +
 drivers/gpu/drm/i915/intel_pm.c      | 151 +++++++++++++++++++++--------------
 drivers/gpu/drm/i915/intel_uncore.c  |   2 +-
 5 files changed, 98 insertions(+), 67 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
index b98afbd33235..e5f2551a2fd7 100644
--- a/drivers/gpu/drm/i915/i915_drv.c
+++ b/drivers/gpu/drm/i915/i915_drv.c
@@ -2021,6 +2021,7 @@ static int i915_drm_resume(struct drm_device *dev)
 
 	intel_opregion_notify_adapter(dev_priv, PCI_D0);
 
+	intel_autoenable_gt_powersave(dev_priv);
 	drm_kms_helper_poll_enable(dev);
 
 	enable_rpm_wakeref_asserts(dev_priv);
@@ -2217,8 +2218,7 @@ int i915_reset(struct drm_i915_private *dev_priv)
 	 * previous concerns that it doesn't respond well to some forms
 	 * of re-init after reset.
 	 */
-	if (INTEL_INFO(dev)->gen > 5)
-		intel_enable_gt_powersave(dev_priv);
+	intel_autoenable_gt_powersave(dev_priv);
 
 	return 0;
 
@@ -2841,7 +2841,6 @@ static int intel_runtime_resume(struct device *device)
 	 * we can do is to hope that things will still work (and disable RPM).
 	 */
 	i915_gem_init_swizzling(dev);
-	gen6_update_ring_freq(dev_priv);
 
 	intel_runtime_pm_enable_interrupts(dev_priv);
 
@@ -2853,7 +2852,7 @@ static int intel_runtime_resume(struct device *device)
 	if (!IS_VALLEYVIEW(dev_priv) && !IS_CHERRYVIEW(dev_priv))
 		intel_hpd_init(dev_priv);
 
-	intel_enable_gt_powersave(dev_priv);
+	intel_autoenable_gt_powersave(dev_priv);
 
 	enable_rpm_wakeref_asserts(dev_priv);
 
diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index c3b5dc8f5e80..699da067bb85 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -10976,6 +10976,7 @@ void intel_mark_busy(struct drm_i915_private *dev_priv)
 		return;
 
 	intel_runtime_pm_get(dev_priv);
+	intel_enable_gt_powersave(dev_priv);
 	i915_update_gfx_val(dev_priv);
 	if (INTEL_GEN(dev_priv) >= 6)
 		gen6_rps_busy(dev_priv);
@@ -15545,7 +15546,6 @@ void intel_modeset_init_hw(struct drm_device *dev)
 	dev_priv->atomic_cdclk_freq = dev_priv->cdclk_freq;
 
 	intel_init_clock_gating(dev);
-	intel_enable_gt_powersave(dev_priv);
 }
 
 /*
@@ -16340,6 +16340,7 @@ void intel_modeset_cleanup(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
 
+	intel_suspend_gt_powersave(dev_priv);
 	intel_disable_gt_powersave(dev_priv);
 
 	/*
diff --git a/drivers/gpu/drm/i915/intel_drv.h b/drivers/gpu/drm/i915/intel_drv.h
index 3156d8df7921..1f06e7ff5931 100644
--- a/drivers/gpu/drm/i915/intel_drv.h
+++ b/drivers/gpu/drm/i915/intel_drv.h
@@ -1662,7 +1662,9 @@ void intel_gpu_ips_init(struct drm_i915_private *dev_priv);
 void intel_gpu_ips_teardown(void);
 void intel_init_gt_powersave(struct drm_i915_private *dev_priv);
 void intel_cleanup_gt_powersave(struct drm_i915_private *dev_priv);
+void intel_sanitize_gt_powersave(struct drm_i915_private *dev_priv);
 void intel_enable_gt_powersave(struct drm_i915_private *dev_priv);
+void intel_autoenable_gt_powersave(struct drm_i915_private *dev_priv);
 void intel_disable_gt_powersave(struct drm_i915_private *dev_priv);
 void intel_suspend_gt_powersave(struct drm_i915_private *dev_priv);
 void intel_reset_gt_powersave(struct drm_i915_private *dev_priv);
diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index 78b35b86e80d..4f612b96224e 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -6522,13 +6522,6 @@ void intel_cleanup_gt_powersave(struct drm_i915_private *dev_priv)
 		intel_runtime_pm_put(dev_priv);
 }
 
-static void gen6_suspend_rps(struct drm_i915_private *dev_priv)
-{
-	flush_delayed_work(&dev_priv->rps.delayed_resume_work);
-
-	gen6_disable_rps_interrupts(dev_priv);
-}
-
 /**
  * intel_suspend_gt_powersave - suspend PM work and helper threads
  * @dev_priv: i915 device
@@ -6542,50 +6535,65 @@ void intel_suspend_gt_powersave(struct drm_i915_private *dev_priv)
 	if (INTEL_GEN(dev_priv) < 6)
 		return;
 
-	gen6_suspend_rps(dev_priv);
+	cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work);
+
+	gen6_disable_rps_interrupts(dev_priv);
 
 	/* Force GPU to min freq during suspend */
 	gen6_rps_idle(dev_priv);
 }
 
+void intel_sanitize_gt_powersave(struct drm_i915_private *dev_priv)
+{
+	dev_priv->rps.enabled = true; /* force disabling */
+	intel_disable_gt_powersave(dev_priv);
+
+	gen6_reset_rps_interrupts(dev_priv);
+}
+
 void intel_disable_gt_powersave(struct drm_i915_private *dev_priv)
 {
-	if (IS_IRONLAKE_M(dev_priv)) {
-		ironlake_disable_drps(dev_priv);
-	} else if (INTEL_INFO(dev_priv)->gen >= 6) {
-		intel_suspend_gt_powersave(dev_priv);
+	if (!READ_ONCE(dev_priv->rps.enabled))
+		return;
 
-		mutex_lock(&dev_priv->rps.hw_lock);
-		if (INTEL_INFO(dev_priv)->gen >= 9) {
-			gen9_disable_rc6(dev_priv);
-			gen9_disable_rps(dev_priv);
-		} else if (IS_CHERRYVIEW(dev_priv))
-			cherryview_disable_rps(dev_priv);
-		else if (IS_VALLEYVIEW(dev_priv))
-			valleyview_disable_rps(dev_priv);
-		else
-			gen6_disable_rps(dev_priv);
+	mutex_lock(&dev_priv->rps.hw_lock);
 
-		dev_priv->rps.enabled = false;
-		mutex_unlock(&dev_priv->rps.hw_lock);
+	if (INTEL_GEN(dev_priv) >= 9) {
+		gen9_disable_rc6(dev_priv);
+		gen9_disable_rps(dev_priv);
+	} else if (IS_CHERRYVIEW(dev_priv)) {
+		cherryview_disable_rps(dev_priv);
+	} else if (IS_VALLEYVIEW(dev_priv)) {
+		valleyview_disable_rps(dev_priv);
+	} else if (INTEL_GEN(dev_priv) >= 6) {
+		gen6_disable_rps(dev_priv);
+	}  else if (IS_IRONLAKE_M(dev_priv)) {
+		ironlake_disable_drps(dev_priv);
 	}
+
+	dev_priv->rps.enabled = false;
+	mutex_unlock(&dev_priv->rps.hw_lock);
 }
 
-static void intel_gen6_powersave_work(struct work_struct *work)
+void intel_enable_gt_powersave(struct drm_i915_private *dev_priv)
 {
-	struct drm_i915_private *dev_priv =
-		container_of(work, struct drm_i915_private,
-			     rps.delayed_resume_work.work);
+	/* We shouldn't be disabling as we submit, so this should be less
+	 * racy than it appears!
+	 */
+	if (READ_ONCE(dev_priv->rps.enabled))
+		return;
 
-	mutex_lock(&dev_priv->rps.hw_lock);
+	/* Powersaving is controlled by the host when inside a VM */
+	if (intel_vgpu_active(dev_priv))
+		return;
 
-	gen6_reset_rps_interrupts(dev_priv);
+	mutex_lock(&dev_priv->rps.hw_lock);
 
 	if (IS_CHERRYVIEW(dev_priv)) {
 		cherryview_enable_rps(dev_priv);
 	} else if (IS_VALLEYVIEW(dev_priv)) {
 		valleyview_enable_rps(dev_priv);
-	} else if (INTEL_INFO(dev_priv)->gen >= 9) {
+	} else if (INTEL_GEN(dev_priv) >= 9) {
 		gen9_enable_rc6(dev_priv);
 		gen9_enable_rps(dev_priv);
 		if (IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
@@ -6593,9 +6601,12 @@ static void intel_gen6_powersave_work(struct work_struct *work)
 	} else if (IS_BROADWELL(dev_priv)) {
 		gen8_enable_rps(dev_priv);
 		__gen6_update_ring_freq(dev_priv);
-	} else {
+	} else if (INTEL_GEN(dev_priv) >= 6) {
 		gen6_enable_rps(dev_priv);
 		__gen6_update_ring_freq(dev_priv);
+	} else if (IS_IRONLAKE_M(dev_priv)) {
+		ironlake_enable_drps(dev_priv);
+		intel_init_emon(dev_priv);
 	}
 
 	WARN_ON(dev_priv->rps.max_freq < dev_priv->rps.min_freq);
@@ -6604,43 +6615,61 @@ static void intel_gen6_powersave_work(struct work_struct *work)
 	WARN_ON(dev_priv->rps.efficient_freq < dev_priv->rps.min_freq);
 	WARN_ON(dev_priv->rps.efficient_freq > dev_priv->rps.max_freq);
 
+	if (INTEL_GEN(dev_priv) >= 6)
+		gen6_enable_rps_interrupts(dev_priv);
+
 	dev_priv->rps.enabled = true;
+	mutex_unlock(&dev_priv->rps.hw_lock);
+}
 
-	gen6_enable_rps_interrupts(dev_priv);
+static void __intel_autoenable_gt_powersave(struct work_struct *work)
+{
+	struct drm_i915_private *dev_priv =
+		container_of(work, typeof(*dev_priv), rps.delayed_resume_work.work);
+	struct intel_engine_cs *rcs;
+	struct drm_i915_gem_request *req;
+	int ret;
 
-	mutex_unlock(&dev_priv->rps.hw_lock);
+	if (READ_ONCE(dev_priv->rps.enabled))
+		return;
+
+	rcs = &dev_priv->engine[RCS];
+	if (rcs->last_context)
+		return;
+
+	if (!rcs->init_context)
+		return;
+
+	intel_runtime_pm_get(dev_priv);
+	mutex_lock(&dev_priv->dev->struct_mutex);
+
+	req = i915_gem_request_alloc(rcs, dev_priv->kernel_context);
+	if (IS_ERR(req))
+		goto unlock;
+
+	ret = 0;
+	if (!i915.enable_execlists)
+		ret = i915_switch_context(req);
+	if (ret == 0)
+		rcs->init_context(req);
+	i915_add_request_no_flush(req);
 
+unlock:
+	mutex_unlock(&dev_priv->dev->struct_mutex);
 	intel_runtime_pm_put(dev_priv);
 }
 
-void intel_enable_gt_powersave(struct drm_i915_private *dev_priv)
+void intel_autoenable_gt_powersave(struct drm_i915_private *dev_priv)
 {
-	/* Powersaving is controlled by the host when inside a VM */
-	if (intel_vgpu_active(dev_priv))
+	if (READ_ONCE(dev_priv->rps.enabled))
 		return;
 
-	if (IS_IRONLAKE_M(dev_priv)) {
-		ironlake_enable_drps(dev_priv);
-		mutex_lock(&dev_priv->dev->struct_mutex);
-		intel_init_emon(dev_priv);
-		mutex_unlock(&dev_priv->dev->struct_mutex);
-	} else if (INTEL_INFO(dev_priv)->gen >= 6) {
-		/*
-		 * PCU communication is slow and this doesn't need to be
-		 * done at any specific time, so do this out of our fast path
-		 * to make resume and init faster.
-		 *
-		 * We depend on the HW RC6 power context save/restore
-		 * mechanism when entering D3 through runtime PM suspend. So
-		 * disable RPM until RPS/RC6 is properly setup. We can only
-		 * get here via the driver load/system resume/runtime resume
-		 * paths, so the _noresume version is enough (and in case of
-		 * runtime resume it's necessary).
-		 */
-		if (schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
-					   round_jiffies_up_relative(HZ)))
-			intel_runtime_pm_get_noresume(dev_priv);
-	}
+	if (INTEL_GEN(dev_priv) < 6)
+		return;
+
+	queue_delayed_work(dev_priv->wq,
+			   &dev_priv->rps.delayed_resume_work,
+			   round_jiffies_up_relative(2*HZ));
 }
 
 void intel_reset_gt_powersave(struct drm_i915_private *dev_priv)
@@ -6648,7 +6677,7 @@ void intel_reset_gt_powersave(struct drm_i915_private *dev_priv)
 	if (INTEL_INFO(dev_priv)->gen < 6)
 		return;
 
-	gen6_suspend_rps(dev_priv);
+	gen6_disable_rps_interrupts(dev_priv);
 	dev_priv->rps.enabled = false;
 }
 
@@ -7746,7 +7775,7 @@ void intel_pm_setup(struct drm_device *dev)
 	spin_lock_init(&dev_priv->rps.client_lock);
 
 	INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
-			  intel_gen6_powersave_work);
+			  __intel_autoenable_gt_powersave);
 	INIT_LIST_HEAD(&dev_priv->rps.clients);
 	INIT_LIST_HEAD(&dev_priv->rps.semaphores.link);
 	INIT_LIST_HEAD(&dev_priv->rps.mmioflips.link);
diff --git a/drivers/gpu/drm/i915/intel_uncore.c b/drivers/gpu/drm/i915/intel_uncore.c
index c1ca458d688e..1c7ae9dd8cae 100644
--- a/drivers/gpu/drm/i915/intel_uncore.c
+++ b/drivers/gpu/drm/i915/intel_uncore.c
@@ -435,7 +435,7 @@ void intel_uncore_sanitize(struct drm_i915_private *dev_priv)
 	i915.enable_rc6 = sanitize_rc6_option(dev_priv, i915.enable_rc6);
 
 	/* BIOS often leaves RC6 enabled, but disable it for hw init */
-	intel_disable_gt_powersave(dev_priv);
+	intel_sanitize_gt_powersave(dev_priv);
 }
 
 static void __intel_uncore_forcewake_get(struct drm_i915_private *dev_priv,
-- 
2.8.1

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

[PATCH 04/25] drm: Restore double clflush on the last partial cacheline

[Chris Wilson]

This effectively reverts

commit afcd950cafea6e27b739fe7772cbbeed37d05b8b
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Jun 10 15:58:01 2015 +0100

    drm: Avoid the double clflush on the last cache line in drm_clflush_virt_range()

as we have observed issues with serialisation of the clflush operations
on Baytrail+ Atoms with partial updates. Applying the double flush on the
last cacheline forces that clflush to be ordered with respect to the
previous clflush, and the mfence then protects against prefetches crossing
the clflush boundary.

The same issue can be demonstrated in userspace with igt/gem_exec_flush.

Fixes: afcd950cafea6 (drm: Avoid the double clflush on the last cache...)
Testcase: igt/gem_concurrent_blit
Testcase: igt/gem_partial_pread_pwrite
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92845
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: dri-devel@lists.freedesktop.org
Cc: Akash Goel <akash.goel@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jason Ekstrand <jason.ekstrand@intel.com>
Cc: stable@vger.kernel.org
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
---
 drivers/gpu/drm/drm_cache.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/gpu/drm/drm_cache.c b/drivers/gpu/drm/drm_cache.c
index 059f7c39c582..a7916e5f8864 100644
--- a/drivers/gpu/drm/drm_cache.c
+++ b/drivers/gpu/drm/drm_cache.c
@@ -136,6 +136,7 @@ drm_clflush_virt_range(void *addr, unsigned long length)
 		mb();
 		for (; addr < end; addr += size)
 			clflushopt(addr);
+		clflushopt(end - 1); /* force serialisation */
 		mb();
 		return;
 	}
-- 
2.8.1

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

[PATCH 05/25] drm/i915/shrinker: Flush active on objects before counting

[Chris Wilson]

As we inspect obj->active to decide how many objects we can shrink (we
only shrink idle objects), it helps to flush the active lists first
in order to have a more accurate count of available objects.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_gem_shrinker.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_gem_shrinker.c b/drivers/gpu/drm/i915/i915_gem_shrinker.c
index 886a8797566d..1bf14544d8ad 100644
--- a/drivers/gpu/drm/i915/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/i915_gem_shrinker.c
@@ -265,6 +265,8 @@ i915_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc)
 	if (!i915_gem_shrinker_lock(dev, &unlock))
 		return 0;
 
+	i915_gem_retire_requests(dev_priv);
+
 	count = 0;
 	list_for_each_entry(obj, &dev_priv->mm.unbound_list, global_list)
 		if (can_release_pages(obj))
-- 
2.8.1

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

[PATCH 06/25] drm/i915: Delay queuing hangcheck to wait-request

[Chris Wilson]

We can forgo queuing the hangcheck from the start of every request to
until we wait upon a request. This reduces the overhead of every
request, but may increase the latency of detecting a hang. Howeever, if
nothing every waits upon a hang, did it ever hang? It also improves the
robustness of the wait-request by ensuring that the hangchecker is
indeed running before we sleep indefinitely (and thereby ensuring that
we never actually sleep forever waiting for a dead GPU).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c |  9 +++++----
 drivers/gpu/drm/i915/i915_irq.c | 10 ++++------
 2 files changed, 9 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 51191b879747..f8b31114999a 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1531,6 +1531,9 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 			break;
 		}
 
+		/* Ensure that even if the GPU hangs, we get woken up. */
+		i915_queue_hangcheck(dev_priv);
+
 		timer.function = NULL;
 		if (timeout || missed_irq(dev_priv, engine)) {
 			unsigned long expire;
@@ -2918,8 +2921,6 @@ void __i915_add_request(struct drm_i915_gem_request *request,
 	/* Not allowed to fail! */
 	WARN(ret, "emit|add_request failed: %d!\n", ret);
 
-	i915_queue_hangcheck(engine->i915);
-
 	queue_delayed_work(dev_priv->wq,
 			   &dev_priv->mm.retire_work,
 			   round_jiffies_up_relative(HZ));
@@ -3264,8 +3265,8 @@ i915_gem_retire_requests(struct drm_i915_private *dev_priv)
 
 	if (idle)
 		mod_delayed_work(dev_priv->wq,
-				   &dev_priv->mm.idle_work,
-				   msecs_to_jiffies(100));
+				 &dev_priv->mm.idle_work,
+				 msecs_to_jiffies(100));
 
 	return idle;
 }
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 4378a659d962..5614582ca240 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -3135,10 +3135,10 @@ static void i915_hangcheck_elapsed(struct work_struct *work)
 	intel_uncore_arm_unclaimed_mmio_detection(dev_priv);
 
 	for_each_engine_id(engine, dev_priv, id) {
+		bool busy = waitqueue_active(&engine->irq_queue);
 		u64 acthd;
 		u32 seqno;
 		unsigned user_interrupts;
-		bool busy = true;
 
 		semaphore_clear_deadlocks(dev_priv);
 
@@ -3161,12 +3161,11 @@ static void i915_hangcheck_elapsed(struct work_struct *work)
 		if (engine->hangcheck.seqno == seqno) {
 			if (ring_idle(engine, seqno)) {
 				engine->hangcheck.action = HANGCHECK_IDLE;
-				if (waitqueue_active(&engine->irq_queue)) {
+				if (busy) {
 					/* Safeguard against driver failure */
 					user_interrupts = kick_waiters(engine);
 					engine->hangcheck.score += BUSY;
-				} else
-					busy = false;
+				}
 			} else {
 				/* We always increment the hangcheck score
 				 * if the ring is busy and still processing
@@ -3240,9 +3239,8 @@ static void i915_hangcheck_elapsed(struct work_struct *work)
 		goto out;
 	}
 
+	/* Reset timer in case GPU hangs without another request being added */
 	if (busy_count)
-		/* Reset timer case chip hangs without another request
-		 * being added */
 		i915_queue_hangcheck(dev_priv);
 
 out:
-- 
2.8.1

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

[PATCH 07/25] drm/i915: Remove the dedicated hangcheck workqueue

[Chris Wilson]

The queue only ever contains at most one item and has no special flags.
It is just a very simple wrapper around the system-wq - a complication
with no benefits.

v2: Use the system_long_wq as we may wish to capture the error state
after detecting the hang - which may take a bit of time.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.c | 8 --------
 drivers/gpu/drm/i915/i915_drv.h | 1 -
 drivers/gpu/drm/i915/i915_irq.c | 7 ++++---
 3 files changed, 4 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
index e5f2551a2fd7..ed3eef1eb398 100644
--- a/drivers/gpu/drm/i915/i915_drv.c
+++ b/drivers/gpu/drm/i915/i915_drv.c
@@ -1169,15 +1169,8 @@ static int i915_workqueues_init(struct drm_i915_private *dev_priv)
 	if (dev_priv->hotplug.dp_wq == NULL)
 		goto out_free_wq;
 
-	dev_priv->gpu_error.hangcheck_wq =
-		alloc_ordered_workqueue("i915-hangcheck", 0);
-	if (dev_priv->gpu_error.hangcheck_wq == NULL)
-		goto out_free_dp_wq;
-
 	return 0;
 
-out_free_dp_wq:
-	destroy_workqueue(dev_priv->hotplug.dp_wq);
 out_free_wq:
 	destroy_workqueue(dev_priv->wq);
 out_err:
@@ -1188,7 +1181,6 @@ out_err:
 
 static void i915_workqueues_cleanup(struct drm_i915_private *dev_priv)
 {
-	destroy_workqueue(dev_priv->gpu_error.hangcheck_wq);
 	destroy_workqueue(dev_priv->hotplug.dp_wq);
 	destroy_workqueue(dev_priv->wq);
 }
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 9ef491975812..13bec2c777c3 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1374,7 +1374,6 @@ struct i915_gpu_error {
 	/* Hang gpu twice in this window and your context gets banned */
 #define DRM_I915_CTX_BAN_PERIOD DIV_ROUND_UP(8*DRM_I915_HANGCHECK_PERIOD, 1000)
 
-	struct workqueue_struct *hangcheck_wq;
 	struct delayed_work hangcheck_work;
 
 	/* For reset and error_state handling. */
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 5614582ca240..3ad4ef9250d8 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -3249,7 +3249,7 @@ out:
 
 void i915_queue_hangcheck(struct drm_i915_private *dev_priv)
 {
-	struct i915_gpu_error *e = &dev_priv->gpu_error;
+	unsigned long delay;
 
 	if (!i915.enable_hangcheck)
 		return;
@@ -3259,8 +3259,9 @@ void i915_queue_hangcheck(struct drm_i915_private *dev_priv)
 	 * we will ignore a hung ring if a second ring is kept busy.
 	 */
 
-	queue_delayed_work(e->hangcheck_wq, &e->hangcheck_work,
-			   round_jiffies_up_relative(DRM_I915_HANGCHECK_JIFFIES));
+	delay = round_jiffies_up_relative(DRM_I915_HANGCHECK_JIFFIES);
+	queue_delayed_work(system_long_wq,
+			   &dev_priv->gpu_error.hangcheck_work, delay);
 }
 
 static void ibx_irq_reset(struct drm_device *dev)
-- 
2.8.1

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

[PATCH 08/25] drm/i915: Make queueing the hangcheck work inline

[Chris Wilson]

Since the function is a small wrapper around schedule_delayed_work(),
move it inline to remove the function call overhead for the principle
caller.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h | 18 +++++++++++++++++-
 drivers/gpu/drm/i915/i915_irq.c | 17 -----------------
 2 files changed, 17 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 13bec2c777c3..903f213580aa 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2933,7 +2933,23 @@ void intel_hpd_cancel_work(struct drm_i915_private *dev_priv);
 bool intel_hpd_pin_to_port(enum hpd_pin pin, enum port *port);
 
 /* i915_irq.c */
-void i915_queue_hangcheck(struct drm_i915_private *dev_priv);
+static inline void i915_queue_hangcheck(struct drm_i915_private *dev_priv)
+{
+	unsigned long delay;
+
+	if (unlikely(!i915.enable_hangcheck))
+		return;
+
+	/* Don't continually defer the hangcheck so that it is always run at
+	 * least once after work has been scheduled on any ring. Otherwise,
+	 * we will ignore a hung ring if a second ring is kept busy.
+	 */
+
+	delay = round_jiffies_up_relative(DRM_I915_HANGCHECK_JIFFIES);
+	queue_delayed_work(system_long_wq,
+			   &dev_priv->gpu_error.hangcheck_work, delay);
+}
+
 __printf(3, 4)
 void i915_handle_error(struct drm_i915_private *dev_priv,
 		       u32 engine_mask,
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 3ad4ef9250d8..83f40baeb1f3 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -3247,23 +3247,6 @@ out:
 	ENABLE_RPM_WAKEREF_ASSERTS(dev_priv);
 }
 
-void i915_queue_hangcheck(struct drm_i915_private *dev_priv)
-{
-	unsigned long delay;
-
-	if (!i915.enable_hangcheck)
-		return;
-
-	/* Don't continually defer the hangcheck so that it is always run at
-	 * least once after work has been scheduled on any ring. Otherwise,
-	 * we will ignore a hung ring if a second ring is kept busy.
-	 */
-
-	delay = round_jiffies_up_relative(DRM_I915_HANGCHECK_JIFFIES);
-	queue_delayed_work(system_long_wq,
-			   &dev_priv->gpu_error.hangcheck_work, delay);
-}
-
 static void ibx_irq_reset(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
-- 
2.8.1

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

[PATCH 09/25] drm/i915: Separate GPU hang waitqueue from advance

[Chris Wilson]

Currently __i915_wait_request uses a per-engine wait_queue_t for the dual
purpose of waking after the GPU advances or for waking after an error.
In the future, we may add even more wake sources and require greater
separation, but for now we can conceptually simplify wakeups by separating
the two sources. In particular, this allows us to use different wait-queues
(e.g. one on the engine advancement, a global one for errors and one on
each requests) without any hassle.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h |  6 ++++++
 drivers/gpu/drm/i915/i915_gem.c |  5 +++++
 drivers/gpu/drm/i915/i915_irq.c | 19 ++++---------------
 3 files changed, 15 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 903f213580aa..97bda64184de 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1410,6 +1410,12 @@ struct i915_gpu_error {
 #define I915_WEDGED			(1 << 31)
 
 	/**
+	 * Waitqueue to signal when a hang is detected. Used to for waiters
+	 * to release the struct_mutex for the reset to procede.
+	 */
+	wait_queue_head_t wait_queue;
+
+	/**
 	 * Waitqueue to signal when the reset has completed. Used by clients
 	 * that wait for dev_priv->mm.wedged to settle.
 	 */
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index f8b31114999a..b586c5e79eb0 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1455,6 +1455,7 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 	const bool irq_test_in_progress =
 		ACCESS_ONCE(dev_priv->gpu_error.test_irq_rings) & intel_engine_flag(engine);
 	int state = interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
+	DEFINE_WAIT(reset);
 	DEFINE_WAIT(wait);
 	unsigned long timeout_expire;
 	s64 before = 0; /* Only to silence a compiler warning. */
@@ -1499,6 +1500,7 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 		goto out;
 	}
 
+	add_wait_queue(&dev_priv->gpu_error.wait_queue, &reset);
 	for (;;) {
 		struct timer_list timer;
 
@@ -1550,6 +1552,8 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 			destroy_timer_on_stack(&timer);
 		}
 	}
+	remove_wait_queue(&dev_priv->gpu_error.wait_queue, &reset);
+
 	if (!irq_test_in_progress)
 		engine->irq_put(engine);
 
@@ -5281,6 +5285,7 @@ i915_gem_load_init(struct drm_device *dev)
 			  i915_gem_retire_work_handler);
 	INIT_DELAYED_WORK(&dev_priv->mm.idle_work,
 			  i915_gem_idle_work_handler);
+	init_waitqueue_head(&dev_priv->gpu_error.wait_queue);
 	init_waitqueue_head(&dev_priv->gpu_error.reset_queue);
 
 	dev_priv->relative_constants_mode = I915_EXEC_CONSTANTS_REL_GENERAL;
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 83f40baeb1f3..6c17596d75dd 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2488,11 +2488,8 @@ static irqreturn_t gen8_irq_handler(int irq, void *arg)
 	return ret;
 }
 
-static void i915_error_wake_up(struct drm_i915_private *dev_priv,
-			       bool reset_completed)
+static void i915_error_wake_up(struct drm_i915_private *dev_priv)
 {
-	struct intel_engine_cs *engine;
-
 	/*
 	 * Notify all waiters for GPU completion events that reset state has
 	 * been changed, and that they need to restart their wait after
@@ -2501,18 +2498,10 @@ static void i915_error_wake_up(struct drm_i915_private *dev_priv,
 	 */
 
 	/* Wake up __wait_seqno, potentially holding dev->struct_mutex. */
-	for_each_engine(engine, dev_priv)
-		wake_up_all(&engine->irq_queue);
+	wake_up_all(&dev_priv->gpu_error.wait_queue);
 
 	/* Wake up intel_crtc_wait_for_pending_flips, holding crtc->mutex. */
 	wake_up_all(&dev_priv->pending_flip_queue);
-
-	/*
-	 * Signal tasks blocked in i915_gem_wait_for_error that the pending
-	 * reset state is cleared.
-	 */
-	if (reset_completed)
-		wake_up_all(&dev_priv->gpu_error.reset_queue);
 }
 
 /**
@@ -2577,7 +2566,7 @@ static void i915_reset_and_wakeup(struct drm_i915_private *dev_priv)
 		 * Note: The wake_up also serves as a memory barrier so that
 		 * waiters see the update value of the reset counter atomic_t.
 		 */
-		i915_error_wake_up(dev_priv, true);
+		wake_up_all(&dev_priv->gpu_error.reset_queue);
 	}
 }
 
@@ -2714,7 +2703,7 @@ void i915_handle_error(struct drm_i915_private *dev_priv,
 		 * ensure that the waiters see the updated value of the reset
 		 * counter atomic_t.
 		 */
-		i915_error_wake_up(dev_priv, false);
+		i915_error_wake_up(dev_priv);
 	}
 
 	i915_reset_and_wakeup(dev_priv);
-- 
2.8.1

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

[PATCH 10/25] drm/i915: Slaughter the thundering i915_wait_request herd

[Chris Wilson]

One particularly stressful scenario consists of many independent tasks
all competing for GPU time and waiting upon the results (e.g. realtime
transcoding of many, many streams). One bottleneck in particular is that
each client waits on its own results, but every client is woken up after
every batchbuffer - hence the thunder of hooves as then every client must
do its heavyweight dance to read a coherent seqno to see if it is the
lucky one.

Ideally, we only want one client to wake up after the interrupt and
check its request for completion. Since the requests must retire in
order, we can select the first client on the oldest request to be woken.
Once that client has completed his wait, we can then wake up the
next client and so on. However, all clients then incur latency as every
process in the chain may be delayed for scheduling - this may also then
cause some priority inversion. To reduce the latency, when a client
is added or removed from the list, we scan the tree for completed
seqno and wake up all the completed waiters in parallel.

Using igt/benchmarks/gem_latency, we can demonstrate this effect. The
benchmark measures the number of GPU cycles between completion of a
batch and the client waking up from a call to wait-ioctl. With many
concurrent waiters, with each on a different request, we observe that
the wakeup latency before the patch scales nearly linearly with the
number of waiters (before external factors kick in making the scaling much
worse). After applying the patch, we can see that only the single waiter
for the request is being woken up, providing a constant wakeup latency
for every operation. However, the situation is not quite as rosy for
many waiters on the same request, though to the best of my knowledge this
is much less likely in practice. Here, we can observe that the
concurrent waiters incur extra latency from being woken up by the
solitary bottom-half, rather than directly by the interrupt. This
appears to be scheduler induced (having discounted adverse effects from
having a rbtree walk/erase in the wakeup path), each additional
wake_up_process() costs approximately 1us on big core. Another effect of
performing the secondary wakeups from the first bottom-half is the
incurred delay this imposes on high priority threads - rather than
immediately returning to userspace and leaving the interrupt handler to
wake the others.

To offset the delay incurred with additional waiters on a request, we
could use a hybrid scheme that did a quick read in the interrupt handler
and dequeued all the completed waiters (incurring the overhead in the
interrupt handler, not the best plan either as we then incur GPU
submission latency) but we would still have to wake up the bottom-half
every time to do the heavyweight slow read. Or we could only kick the
waiters on the seqno with the same priority as the current task (i.e. in
the realtime waiter scenario, only it is woken up immediately by the
interrupt and simply queues the next waiter before returning to userspace,
minimising its delay at the expense of the chain, and also reducing
contention on its scheduler runqueue). This is effective at avoid long
pauses in the interrupt handler and at avoiding the extra latency in
realtime/high-priority waiters.

v2: Convert from a kworker per engine into a dedicated kthread for the
bottom-half.
v3: Rename request members and tweak comments.
v4: Use a per-engine spinlock in the breadcrumbs bottom-half.
v5: Fix race in locklessly checking waiter status and kicking the task on
adding a new waiter.
v6: Fix deciding when to force the timer to hide missing interrupts.
v7: Move the bottom-half from the kthread to the first client process.
v8: Reword a few comments
v9: Break the busy loop when the interrupt is unmasked or has fired.
v10: Comments, unnecessary churn, better debugging from Tvrtko
v11: Wake all completed waiters on removing the current bottom-half to
reduce the latency of waking up a herd of clients all waiting on the
same request.
v12: Rearrange missed-interrupt fault injection so that it works with
igt/drv_missed_irq_hang
v13: Rename intel_breadcrumb and friends to intel_wait in preparation
for signal handling.
v14: RCU commentary, assert_spin_locked
v15: Hide BUG_ON behind the compiler; report on gem_latency findings.
v16: Sort seqno-groups by priority so that first-waiter has the highest
task priority (and so avoid priority inversion).
v17: Add waiters to post-mortem GPU hang state.
v18: Return early for a completed wait after acquiring the spinlock.
Avoids adding ourselves to the tree if the is already complete, and
skips the awkward question of why we don't do completion wakeups for
waits earlier than or equal to ourselves.
v19: Prepare for init_breadcrumbs to fail. Later patches may want to
allocate during init, so be prepared to propagate back the error code.

Testcase: igt/gem_concurrent_blit
Testcase: igt/benchmarks/gem_latency
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: "Rogozhkin, Dmitry V" <dmitry.v.rogozhkin@intel.com>
Cc: "Gong, Zhipeng" <zhipeng.gong@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Dave Gordon <david.s.gordon@intel.com>
Cc: "Goel, Akash" <akash.goel@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com> #v18
---
 drivers/gpu/drm/i915/Makefile            |   1 +
 drivers/gpu/drm/i915/i915_debugfs.c      |  15 +-
 drivers/gpu/drm/i915/i915_drv.h          |  39 +++-
 drivers/gpu/drm/i915/i915_gem.c          | 140 +++++-------
 drivers/gpu/drm/i915/i915_gpu_error.c    |  59 ++++-
 drivers/gpu/drm/i915/i915_irq.c          |  20 +-
 drivers/gpu/drm/i915/intel_breadcrumbs.c | 375 +++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/intel_lrc.c         |   7 +-
 drivers/gpu/drm/i915/intel_ringbuffer.c  |  12 +-
 drivers/gpu/drm/i915/intel_ringbuffer.h  |  72 +++++-
 10 files changed, 632 insertions(+), 108 deletions(-)
 create mode 100644 drivers/gpu/drm/i915/intel_breadcrumbs.c

diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile
index a3d2b789c9ec..618293c8c9d9 100644
--- a/drivers/gpu/drm/i915/Makefile
+++ b/drivers/gpu/drm/i915/Makefile
@@ -38,6 +38,7 @@ i915-y += i915_cmd_parser.o \
 	  i915_gem_userptr.o \
 	  i915_gpu_error.o \
 	  i915_trace_points.o \
+	  intel_breadcrumbs.o \
 	  intel_lrc.o \
 	  intel_mocs.o \
 	  intel_ringbuffer.o \
diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 2858415a7d0e..c81c27dca74d 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -788,10 +788,21 @@ static int i915_gem_request_info(struct seq_file *m, void *data)
 static void i915_ring_seqno_info(struct seq_file *m,
 				 struct intel_engine_cs *engine)
 {
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	struct rb_node *rb;
+
 	seq_printf(m, "Current sequence (%s): %x\n",
 		   engine->name, engine->get_seqno(engine));
 	seq_printf(m, "Current user interrupts (%s): %x\n",
 		   engine->name, READ_ONCE(engine->user_interrupts));
+
+	spin_lock(&b->lock);
+	for (rb = rb_first(&b->waiters); rb != NULL; rb = rb_next(rb)) {
+		struct intel_wait *w = container_of(rb, typeof(*w), node);
+		seq_printf(m, "Waiting (%s): %s [%d] on %x\n",
+			   engine->name, w->tsk->comm, w->tsk->pid, w->seqno);
+	}
+	spin_unlock(&b->lock);
 }
 
 static int i915_gem_seqno_info(struct seq_file *m, void *data)
@@ -1426,6 +1437,8 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
 			   engine->hangcheck.seqno,
 			   seqno[id],
 			   engine->last_submitted_seqno);
+		seq_printf(m, "\twaiters? %d\n",
+			   intel_engine_has_waiter(engine));
 		seq_printf(m, "\tuser interrupts = %x [current %x]\n",
 			   engine->hangcheck.user_interrupts,
 			   READ_ONCE(engine->user_interrupts));
@@ -2411,7 +2424,7 @@ static int count_irq_waiters(struct drm_i915_private *i915)
 	int count = 0;
 
 	for_each_engine(engine, i915)
-		count += engine->irq_refcount;
+		count += intel_engine_has_waiter(engine);
 
 	return count;
 }
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 97bda64184de..ed92c434f4e6 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -506,6 +506,7 @@ struct drm_i915_error_state {
 		bool valid;
 		/* Software tracked state */
 		bool waiting;
+		int num_waiters;
 		int hangcheck_score;
 		enum intel_ring_hangcheck_action hangcheck_action;
 		int num_requests;
@@ -551,6 +552,12 @@ struct drm_i915_error_state {
 			u32 tail;
 		} *requests;
 
+		struct drm_i915_error_waiter {
+			char comm[TASK_COMM_LEN];
+			pid_t pid;
+			u32 seqno;
+		} *waiters;
+
 		struct {
 			u32 gfx_mode;
 			union {
@@ -1429,7 +1436,7 @@ struct i915_gpu_error {
 #define I915_STOP_RING_ALLOW_WARN      (1 << 30)
 
 	/* For missed irq/seqno simulation. */
-	unsigned int test_irq_rings;
+	unsigned long test_irq_rings;
 };
 
 enum modeset_restore {
@@ -3054,7 +3061,6 @@ ibx_disable_display_interrupt(struct drm_i915_private *dev_priv, uint32_t bits)
 	ibx_display_interrupt_update(dev_priv, bits, 0);
 }
 
-
 /* i915_gem.c */
 int i915_gem_create_ioctl(struct drm_device *dev, void *data,
 			  struct drm_file *file_priv);
@@ -3964,4 +3970,33 @@ static inline void i915_trace_irq_get(struct intel_engine_cs *engine,
 		i915_gem_request_assign(&engine->trace_irq_req, req);
 }
 
+static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
+{
+	/* Ensure our read of the seqno is coherent so that we
+	 * do not "miss an interrupt" (i.e. if this is the last
+	 * request and the seqno write from the GPU is not visible
+	 * by the time the interrupt fires, we will see that the
+	 * request is incomplete and go back to sleep awaiting
+	 * another interrupt that will never come.)
+	 *
+	 * Strictly, we only need to do this once after an interrupt,
+	 * but it is easier and safer to do it every time the waiter
+	 * is woken.
+	 */
+	if (i915_gem_request_completed(req, false))
+		return true;
+
+	/* We need to check whether any gpu reset happened in between
+	 * the request being submitted and now. If a reset has occurred,
+	 * the request is effectively complete (we either are in the
+	 * process of or have discarded the rendering and completely
+	 * reset the GPU. The results of the request are lost and we
+	 * are free to continue on with the original operation.
+	 */
+	if (req->reset_counter != i915_reset_counter(&req->i915->gpu_error))
+		return true;
+
+	return false;
+}
+
 #endif
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index b586c5e79eb0..fe5ba4aa336c 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1343,17 +1343,6 @@ i915_gem_check_wedge(unsigned reset_counter, bool interruptible)
 	return 0;
 }
 
-static void fake_irq(unsigned long data)
-{
-	wake_up_process((struct task_struct *)data);
-}
-
-static bool missed_irq(struct drm_i915_private *dev_priv,
-		       struct intel_engine_cs *engine)
-{
-	return test_bit(engine->id, &dev_priv->gpu_error.missed_irq_rings);
-}
-
 static unsigned long local_clock_us(unsigned *cpu)
 {
 	unsigned long t;
@@ -1386,7 +1375,7 @@ static bool busywait_stop(unsigned long timeout, unsigned cpu)
 	return this_cpu != cpu;
 }
 
-static int __i915_spin_request(struct drm_i915_gem_request *req, int state)
+static bool __i915_spin_request(struct drm_i915_gem_request *req, int state)
 {
 	unsigned long timeout;
 	unsigned cpu;
@@ -1401,17 +1390,14 @@ static int __i915_spin_request(struct drm_i915_gem_request *req, int state)
 	 * takes to sleep on a request, on the order of a microsecond.
 	 */
 
-	if (req->engine->irq_refcount)
-		return -EBUSY;
-
 	/* Only spin if we know the GPU is processing this request */
 	if (!i915_gem_request_started(req, true))
-		return -EAGAIN;
+		return false;
 
 	timeout = local_clock_us(&cpu) + 5;
-	while (!need_resched()) {
+	do {
 		if (i915_gem_request_completed(req, true))
-			return 0;
+			return true;
 
 		if (signal_pending_state(state, current))
 			break;
@@ -1420,12 +1406,9 @@ static int __i915_spin_request(struct drm_i915_gem_request *req, int state)
 			break;
 
 		cpu_relax_lowlatency();
-	}
-
-	if (i915_gem_request_completed(req, false))
-		return 0;
+	} while (!need_resched());
 
-	return -EAGAIN;
+	return false;
 }
 
 /**
@@ -1450,18 +1433,14 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 			s64 *timeout,
 			struct intel_rps_client *rps)
 {
-	struct intel_engine_cs *engine = i915_gem_request_get_engine(req);
-	struct drm_i915_private *dev_priv = req->i915;
-	const bool irq_test_in_progress =
-		ACCESS_ONCE(dev_priv->gpu_error.test_irq_rings) & intel_engine_flag(engine);
 	int state = interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
 	DEFINE_WAIT(reset);
-	DEFINE_WAIT(wait);
-	unsigned long timeout_expire;
+	struct intel_wait wait;
+	unsigned long timeout_remain;
 	s64 before = 0; /* Only to silence a compiler warning. */
-	int ret;
+	int ret = 0;
 
-	WARN(!intel_irqs_enabled(dev_priv), "IRQs disabled");
+	might_sleep();
 
 	if (list_empty(&req->list))
 		return 0;
@@ -1469,7 +1448,7 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 	if (i915_gem_request_completed(req, true))
 		return 0;
 
-	timeout_expire = 0;
+	timeout_remain = MAX_SCHEDULE_TIMEOUT;
 	if (timeout) {
 		if (WARN_ON(*timeout < 0))
 			return -EINVAL;
@@ -1477,7 +1456,7 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 		if (*timeout == 0)
 			return -ETIME;
 
-		timeout_expire = jiffies + nsecs_to_jiffies_timeout(*timeout);
+		timeout_remain = nsecs_to_jiffies_timeout(*timeout);
 
 		/*
 		 * Record current time in case interrupted by signal, or wedged.
@@ -1485,81 +1464,60 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 		before = ktime_get_raw_ns();
 	}
 
-	if (INTEL_INFO(dev_priv)->gen >= 6)
-		gen6_rps_boost(dev_priv, rps, req->emitted_jiffies);
-
 	trace_i915_gem_request_wait_begin(req);
 
-	/* Optimistic spin for the next jiffie before touching IRQs */
-	ret = __i915_spin_request(req, state);
-	if (ret == 0)
-		goto out;
+	if (INTEL_INFO(req->i915)->gen >= 6)
+		gen6_rps_boost(req->i915, rps, req->emitted_jiffies);
 
-	if (!irq_test_in_progress && WARN_ON(!engine->irq_get(engine))) {
-		ret = -ENODEV;
-		goto out;
-	}
+	/* Optimistic spin for the next ~jiffie before touching IRQs */
+	if (__i915_spin_request(req, state))
+		goto complete;
 
-	add_wait_queue(&dev_priv->gpu_error.wait_queue, &reset);
-	for (;;) {
-		struct timer_list timer;
-
-		prepare_to_wait(&engine->irq_queue, &wait, state);
+	set_current_state(state);
+	add_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
 
-		/* We need to check whether any gpu reset happened in between
-		 * the request being submitted and now. If a reset has occurred,
-		 * the request is effectively complete (we either are in the
-		 * process of or have discarded the rendering and completely
-		 * reset the GPU. The results of the request are lost and we
-		 * are free to continue on with the original operation.
+	intel_wait_init(&wait, req->seqno);
+	if (intel_engine_add_wait(req->engine, &wait))
+		/* In order to check that we haven't missed the interrupt
+		 * as we enabled it, we need to kick ourselves to do a
+		 * coherent check on the seqno before we sleep.
 		 */
-		if (req->reset_counter != i915_reset_counter(&dev_priv->gpu_error)) {
-			ret = 0;
-			break;
-		}
-
-		if (i915_gem_request_completed(req, false)) {
-			ret = 0;
-			break;
-		}
+		goto wakeup;
 
+	for (;;) {
 		if (signal_pending_state(state, current)) {
 			ret = -ERESTARTSYS;
 			break;
 		}
 
-		if (timeout && time_after_eq(jiffies, timeout_expire)) {
+		/* Ensure that even if the GPU hangs, we get woken up. */
+		i915_queue_hangcheck(req->i915);
+
+		timeout_remain = io_schedule_timeout(timeout_remain);
+		if (timeout_remain == 0) {
 			ret = -ETIME;
 			break;
 		}
 
-		/* Ensure that even if the GPU hangs, we get woken up. */
-		i915_queue_hangcheck(dev_priv);
-
-		timer.function = NULL;
-		if (timeout || missed_irq(dev_priv, engine)) {
-			unsigned long expire;
-
-			setup_timer_on_stack(&timer, fake_irq, (unsigned long)current);
-			expire = missed_irq(dev_priv, engine) ? jiffies + 1 : timeout_expire;
-			mod_timer(&timer, expire);
-		}
+		if (intel_wait_complete(&wait))
+			break;
 
-		io_schedule();
+wakeup:
+		set_current_state(state);
 
-		if (timer.function) {
-			del_singleshot_timer_sync(&timer);
-			destroy_timer_on_stack(&timer);
-		}
+		/* Carefully check if the request is complete, giving time
+		 * for the seqno to be visible following the interrupt.
+		 * We also have to check in case we are kicked by the GPU
+		 * reset in order to drop the struct_mutex.
+		 */
+		if (__i915_request_irq_complete(req))
+			break;
 	}
-	remove_wait_queue(&dev_priv->gpu_error.wait_queue, &reset);
+	remove_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
 
-	if (!irq_test_in_progress)
-		engine->irq_put(engine);
-
-	finish_wait(&engine->irq_queue, &wait);
-
-out:
+	intel_engine_remove_wait(req->engine, &wait);
+	__set_current_state(TASK_RUNNING);
+complete:
 	trace_i915_gem_request_wait_end(req);
 
 	if (timeout) {
@@ -2789,6 +2747,12 @@ i915_gem_init_seqno(struct drm_i915_private *dev_priv, u32 seqno)
 	}
 	i915_gem_retire_requests(dev_priv);
 
+	/* If the seqno wraps around, we need to clear the breadcrumb rbtree */
+	if (!i915_seqno_passed(seqno, dev_priv->next_seqno)) {
+		while (intel_kick_waiters(dev_priv))
+			yield();
+	}
+
 	/* Finally reset hw state */
 	for_each_engine(engine, dev_priv)
 		intel_ring_init_seqno(engine, seqno);
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index 34ff2459ceea..606f74e15208 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -463,6 +463,18 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
 			}
 		}
 
+		if (error->ring[i].num_waiters) {
+			err_printf(m, "%s --- %d waiters\n",
+				   dev_priv->engine[i].name,
+				   error->ring[i].num_waiters);
+			for (j = 0; j < error->ring[i].num_waiters; j++) {
+				err_printf(m, " seqno 0x%08x for %s [%d]\n",
+					   error->ring[i].waiters[j].seqno,
+					   error->ring[i].waiters[j].comm,
+					   error->ring[i].waiters[j].pid);
+			}
+		}
+
 		if ((obj = error->ring[i].ringbuffer)) {
 			err_printf(m, "%s --- ringbuffer = 0x%08x\n",
 				   dev_priv->engine[i].name,
@@ -605,8 +617,9 @@ static void i915_error_state_free(struct kref *error_ref)
 		i915_error_object_free(error->ring[i].ringbuffer);
 		i915_error_object_free(error->ring[i].hws_page);
 		i915_error_object_free(error->ring[i].ctx);
-		kfree(error->ring[i].requests);
 		i915_error_object_free(error->ring[i].wa_ctx);
+		kfree(error->ring[i].requests);
+		kfree(error->ring[i].waiters);
 	}
 
 	i915_error_object_free(error->semaphore_obj);
@@ -892,6 +905,47 @@ static void gen6_record_semaphore_state(struct drm_i915_private *dev_priv,
 	}
 }
 
+static void engine_record_waiters(struct intel_engine_cs *engine,
+				  struct drm_i915_error_ring *ering)
+{
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	struct drm_i915_error_waiter *waiter;
+	struct rb_node *rb;
+	int count;
+
+	ering->num_waiters = 0;
+	ering->waiters = NULL;
+
+	spin_lock(&b->lock);
+	count = 0;
+	for (rb = rb_first(&b->waiters); rb != NULL; rb = rb_next(rb))
+		count++;
+	spin_unlock(&b->lock);
+
+	waiter = NULL;
+	if (count)
+		waiter = kmalloc(count*sizeof(struct drm_i915_error_waiter),
+				 GFP_ATOMIC);
+	if (!waiter)
+		return;
+
+	ering->waiters = waiter;
+
+	spin_lock(&b->lock);
+	for (rb = rb_first(&b->waiters); rb != NULL; rb = rb_next(rb)) {
+		struct intel_wait *w = container_of(rb, typeof(*w), node);
+
+		strcpy(waiter->comm, w->tsk->comm);
+		waiter->pid = w->tsk->pid;
+		waiter->seqno = w->seqno;
+		waiter++;
+
+		if (++ering->num_waiters == count)
+			break;
+	}
+	spin_unlock(&b->lock);
+}
+
 static void i915_record_ring_state(struct drm_i915_private *dev_priv,
 				   struct drm_i915_error_state *error,
 				   struct intel_engine_cs *engine,
@@ -926,7 +980,7 @@ static void i915_record_ring_state(struct drm_i915_private *dev_priv,
 		ering->instdone = I915_READ(GEN2_INSTDONE);
 	}
 
-	ering->waiting = waitqueue_active(&engine->irq_queue);
+	ering->waiting = intel_engine_has_waiter(engine);
 	ering->instpm = I915_READ(RING_INSTPM(engine->mmio_base));
 	ering->acthd = intel_ring_get_active_head(engine);
 	ering->seqno = engine->get_seqno(engine);
@@ -1032,6 +1086,7 @@ static void i915_gem_record_rings(struct drm_i915_private *dev_priv,
 		error->ring[i].valid = true;
 
 		i915_record_ring_state(dev_priv, error, engine, &error->ring[i]);
+		engine_record_waiters(engine, &error->ring[i]);
 
 		request = i915_gem_find_active_request(engine);
 		if (request) {
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 6c17596d75dd..a11ab00cdee0 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -976,13 +976,10 @@ static void ironlake_rps_change_irq_handler(struct drm_i915_private *dev_priv)
 
 static void notify_ring(struct intel_engine_cs *engine)
 {
-	if (!intel_engine_initialized(engine))
-		return;
-
-	trace_i915_gem_request_notify(engine);
-	engine->user_interrupts++;
-
-	wake_up_all(&engine->irq_queue);
+	if (intel_engine_wakeup(engine)) {
+		trace_i915_gem_request_notify(engine);
+		engine->user_interrupts++;
+	}
 }
 
 static void vlv_c0_read(struct drm_i915_private *dev_priv,
@@ -1063,7 +1060,7 @@ static bool any_waiters(struct drm_i915_private *dev_priv)
 	struct intel_engine_cs *engine;
 
 	for_each_engine(engine, dev_priv)
-		if (engine->irq_refcount)
+		if (intel_engine_has_waiter(engine))
 			return true;
 
 	return false;
@@ -3074,13 +3071,14 @@ static unsigned kick_waiters(struct intel_engine_cs *engine)
 
 	if (engine->hangcheck.user_interrupts == user_interrupts &&
 	    !test_and_set_bit(engine->id, &i915->gpu_error.missed_irq_rings)) {
-		if (!(i915->gpu_error.test_irq_rings & intel_engine_flag(engine)))
+		if (!test_bit(engine->id, &i915->gpu_error.test_irq_rings))
 			DRM_ERROR("Hangcheck timer elapsed... %s idle\n",
 				  engine->name);
 		else
 			DRM_INFO("Fake missed irq on %s\n",
 				 engine->name);
-		wake_up_all(&engine->irq_queue);
+
+		intel_engine_enable_fake_irq(engine);
 	}
 
 	return user_interrupts;
@@ -3124,7 +3122,7 @@ static void i915_hangcheck_elapsed(struct work_struct *work)
 	intel_uncore_arm_unclaimed_mmio_detection(dev_priv);
 
 	for_each_engine_id(engine, dev_priv, id) {
-		bool busy = waitqueue_active(&engine->irq_queue);
+		bool busy = intel_engine_has_waiter(engine);
 		u64 acthd;
 		u32 seqno;
 		unsigned user_interrupts;
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
new file mode 100644
index 000000000000..11ec077660a1
--- /dev/null
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -0,0 +1,375 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ */
+
+#include "i915_drv.h"
+
+static void intel_breadcrumbs_fake_irq(unsigned long data)
+{
+	struct intel_engine_cs *engine = (struct intel_engine_cs *)data;
+
+	/*
+	 * The timer persists in case we cannot enable interrupts,
+	 * or if we have previously seen seqno/interrupt incoherency
+	 * ("missed interrupt" syndrome). Here the worker will wake up
+	 * every jiffie in order to kick the oldest waiter to do the
+	 * coherent seqno check.
+	 */
+	rcu_read_lock();
+	if (intel_engine_wakeup(engine))
+		mod_timer(&engine->breadcrumbs.fake_irq, jiffies + 1);
+	rcu_read_unlock();
+}
+
+static void irq_enable(struct intel_engine_cs *engine)
+{
+	WARN_ON(!engine->irq_get(engine));
+}
+
+static void irq_disable(struct intel_engine_cs *engine)
+{
+	engine->irq_put(engine);
+}
+
+static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
+{
+	struct intel_engine_cs *engine =
+		container_of(b, struct intel_engine_cs, breadcrumbs);
+	struct drm_i915_private *i915 = engine->i915;
+	bool irq_posted = false;
+
+	assert_spin_locked(&b->lock);
+	if (b->rpm_wakelock)
+		return false;
+
+	/* Since we are waiting on a request, the GPU should be busy
+	 * and should have its own rpm reference. For completeness,
+	 * record an rpm reference for ourselves to cover the
+	 * interrupt we unmask.
+	 */
+	intel_runtime_pm_get_noresume(i915);
+	b->rpm_wakelock = true;
+
+	/* No interrupts? Kick the waiter every jiffie! */
+	if (intel_irqs_enabled(i915)) {
+		if (!test_bit(engine->id, &i915->gpu_error.test_irq_rings)) {
+			irq_enable(engine);
+			irq_posted = true;
+		}
+		b->irq_enabled = true;
+	}
+
+	if (!b->irq_enabled ||
+	    test_bit(engine->id, &i915->gpu_error.missed_irq_rings))
+		mod_timer(&b->fake_irq, jiffies + 1);
+
+	return irq_posted;
+}
+
+static void __intel_breadcrumbs_disable_irq(struct intel_breadcrumbs *b)
+{
+	struct intel_engine_cs *engine =
+		container_of(b, struct intel_engine_cs, breadcrumbs);
+
+	assert_spin_locked(&b->lock);
+	if (!b->rpm_wakelock)
+		return;
+
+	if (b->irq_enabled) {
+		irq_disable(engine);
+		b->irq_enabled = false;
+	}
+
+	intel_runtime_pm_put(engine->i915);
+	b->rpm_wakelock = false;
+}
+
+static inline struct intel_wait *to_wait(struct rb_node *node)
+{
+	return container_of(node, struct intel_wait, node);
+}
+
+static inline void __intel_breadcrumbs_finish(struct intel_breadcrumbs *b,
+					      struct intel_wait *wait)
+{
+	assert_spin_locked(&b->lock);
+
+	/* This request is completed, so remove it from the tree, mark it as
+	 * complete, and *then* wake up the associated task.
+	 */
+	rb_erase(&wait->node, &b->waiters);
+	RB_CLEAR_NODE(&wait->node);
+
+	wake_up_process(wait->tsk); /* implicit smp_wmb() */
+}
+
+static bool __intel_engine_add_wait(struct intel_engine_cs *engine,
+				    struct intel_wait *wait)
+{
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	struct rb_node **p, *parent, *completed;
+	bool first;
+	u32 seqno;
+
+	/* Insert the request into the retirement ordered list
+	 * of waiters by walking the rbtree. If we are the oldest
+	 * seqno in the tree (the first to be retired), then
+	 * set ourselves as the bottom-half.
+	 *
+	 * As we descend the tree, prune completed branches since we hold the
+	 * spinlock we know that the first_waiter must be delayed and can
+	 * reduce some of the sequential wake up latency if we take action
+	 * ourselves and wake up the completed tasks in parallel. Also, by
+	 * removing stale elements in the tree, we may be able to reduce the
+	 * ping-pong between the old bottom-half and ourselves as first-waiter.
+	 */
+	first = true;
+	parent = NULL;
+	completed = NULL;
+	seqno = engine->get_seqno(engine);
+
+	 /* If the request completed before we managed to grab the spinlock,
+	  * return now before adding ourselves to the rbtree. We let the
+	  * current bottom-half handle any pending wakeups and instead
+	  * try and get out of the way quickly.
+	  */
+	if (i915_seqno_passed(seqno, wait->seqno)) {
+		RB_CLEAR_NODE(&wait->node);
+		return first;
+	}
+
+	p = &b->waiters.rb_node;
+	while (*p) {
+		parent = *p;
+		if (wait->seqno == to_wait(parent)->seqno) {
+			/* We have multiple waiters on the same seqno, select
+			 * the highest priority task (that with the smallest
+			 * task->prio) to serve as the bottom-half for this
+			 * group.
+			 */
+			if (wait->tsk->prio > to_wait(parent)->tsk->prio) {
+				p = &parent->rb_right;
+				first = false;
+			} else
+				p = &parent->rb_left;
+		} else if (i915_seqno_passed(wait->seqno,
+					     to_wait(parent)->seqno)) {
+			p = &parent->rb_right;
+			if (i915_seqno_passed(seqno, to_wait(parent)->seqno))
+				completed = parent;
+			else
+				first = false;
+		} else
+			p = &parent->rb_left;
+	}
+	rb_link_node(&wait->node, parent, p);
+	rb_insert_color(&wait->node, &b->waiters);
+	GEM_BUG_ON(!first && !b->tasklet);
+
+	if (completed) {
+		struct rb_node *next = rb_next(completed);
+
+		GEM_BUG_ON(!next && !first);
+		if (next && next != &wait->node) {
+			GEM_BUG_ON(first);
+			b->first_wait = to_wait(next);
+			smp_store_mb(b->tasklet, b->first_wait->tsk);
+			/* As there is a delay between reading the current
+			 * seqno, processing the completed tasks and selecting
+			 * the next waiter, we may have missed the interrupt
+			 * and so need for the next bottom-half to wakeup.
+			 *
+			 * Also as we enable the IRQ, we may miss the
+			 * interrupt for that seqno, so we have to wake up
+			 * the next bottom-half in order to do a coherent check
+			 * in case the seqno passed.
+			 */
+			__intel_breadcrumbs_enable_irq(b);
+			wake_up_process(to_wait(next)->tsk);
+		}
+
+		do {
+			struct intel_wait *crumb = to_wait(completed);
+			completed = rb_prev(completed);
+			__intel_breadcrumbs_finish(b, crumb);
+		} while (completed);
+	}
+
+	if (first) {
+		GEM_BUG_ON(rb_first(&b->waiters) != &wait->node);
+		b->first_wait = wait;
+		smp_store_mb(b->tasklet, wait->tsk);
+		first =__intel_breadcrumbs_enable_irq(b);
+	}
+	GEM_BUG_ON(!b->tasklet);
+	GEM_BUG_ON(!b->first_wait);
+	GEM_BUG_ON(rb_first(&b->waiters) != &b->first_wait->node);
+
+	return first;
+}
+
+bool intel_engine_add_wait(struct intel_engine_cs *engine,
+			   struct intel_wait *wait)
+{
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	bool first;
+
+	spin_lock(&b->lock);
+	first = __intel_engine_add_wait(engine, wait);
+	spin_unlock(&b->lock);
+
+	return first;
+}
+
+void intel_engine_enable_fake_irq(struct intel_engine_cs *engine)
+{
+	mod_timer(&engine->breadcrumbs.fake_irq, jiffies + 1);
+}
+
+static inline bool chain_wakeup(struct rb_node *rb, int priority)
+{
+	return rb && to_wait(rb)->tsk->prio <= priority;
+}
+
+void intel_engine_remove_wait(struct intel_engine_cs *engine,
+			      struct intel_wait *wait)
+{
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+
+	/* Quick check to see if this waiter was already decoupled from
+	 * the tree by the bottom-half to avoid contention on the spinlock
+	 * by the herd.
+	 */
+	if (RB_EMPTY_NODE(&wait->node))
+		return;
+
+	spin_lock(&b->lock);
+
+	if (RB_EMPTY_NODE(&wait->node))
+		goto out_unlock;
+
+	if (b->first_wait == wait) {
+		struct rb_node *next;
+		const int priority = wait->tsk->prio;
+
+		GEM_BUG_ON(b->tasklet != wait->tsk);
+
+		/* We are the current bottom-half. Find the next candidate,
+		 * the first waiter in the queue on the remaining oldest
+		 * request. As multiple seqnos may complete in the time it
+		 * takes us to wake up and find the next waiter, we have to
+		 * wake up that waiter for it to perform its own coherent
+		 * completion check.
+		 */
+		next = rb_next(&wait->node);
+		if (chain_wakeup(next, priority)) {
+			/* If the next waiter is already complete,
+			 * wake it up and continue onto the next waiter. So
+			 * if have a small herd, they will wake up in parallel
+			 * rather than sequentially, which should reduce
+			 * the overall latency in waking all the completed
+			 * clients.
+			 *
+			 * However, waking up a chain adds extra latency to
+			 * the first_waiter. This is undesirable if that
+			 * waiter is a high priority task.
+			 */
+			u32 seqno = engine->get_seqno(engine);
+			while (i915_seqno_passed(seqno, to_wait(next)->seqno)) {
+				struct rb_node *n = rb_next(next);
+				__intel_breadcrumbs_finish(b, to_wait(next));
+				next = n;
+				if (!chain_wakeup(next, priority))
+					break;
+			}
+		}
+
+		if (next) {
+			/* In our haste, we may have completed the first waiter
+			 * before we enabled the interrupt. Do so now as we
+			 * have a second waiter for a future seqno. Afterwards,
+			 * we have to wake up that waiter in case we missed
+			 * the interrupt, or if we have to handle an
+			 * exception rather than a seqno completion.
+			 */
+			b->first_wait = to_wait(next);
+			smp_store_mb(b->tasklet, b->first_wait->tsk);
+			if (b->first_wait->seqno != wait->seqno)
+				__intel_breadcrumbs_enable_irq(b);
+			wake_up_process(b->tasklet);
+		} else {
+			b->first_wait = NULL;
+			WRITE_ONCE(b->tasklet, NULL);
+			__intel_breadcrumbs_disable_irq(b);
+		}
+	} else {
+		GEM_BUG_ON(rb_first(&b->waiters) == &wait->node);
+	}
+
+	GEM_BUG_ON(RB_EMPTY_NODE(&wait->node));
+	rb_erase(&wait->node, &b->waiters);
+
+out_unlock:
+	GEM_BUG_ON(b->first_wait == wait);
+	GEM_BUG_ON(rb_first(&b->waiters) != (b->first_wait ? &b->first_wait->node : NULL));
+	GEM_BUG_ON(!b->tasklet ^ RB_EMPTY_ROOT(&b->waiters));
+	spin_unlock(&b->lock);
+}
+
+int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
+{
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+
+	spin_lock_init(&b->lock);
+	setup_timer(&b->fake_irq,
+		    intel_breadcrumbs_fake_irq,
+		    (unsigned long)engine);
+
+	return 0;
+}
+
+void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
+{
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+
+	del_timer_sync(&b->fake_irq);
+}
+
+unsigned intel_kick_waiters(struct drm_i915_private *i915)
+{
+	struct intel_engine_cs *engine;
+	unsigned mask = 0;
+
+	/* To avoid the task_struct disappearing beneath us as we wake up
+	 * the process, we must first inspect the task_struct->state under the
+	 * RCU lock, i.e. as we call wake_up_process() we must be holding the
+	 * rcu_read_lock().
+	 */
+	rcu_read_lock();
+	for_each_engine(engine, i915)
+		if (unlikely(intel_engine_wakeup(engine)))
+			mask |= intel_engine_flag(engine);
+	rcu_read_unlock();
+
+	return mask;
+}
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 62b0dc6c2642..faa62d626211 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1958,6 +1958,8 @@ void intel_logical_ring_cleanup(struct intel_engine_cs *engine)
 	i915_cmd_parser_fini_ring(engine);
 	i915_gem_batch_pool_fini(&engine->batch_pool);
 
+	intel_engine_fini_breadcrumbs(engine);
+
 	if (engine->status_page.obj) {
 		i915_gem_object_unpin_map(engine->status_page.obj);
 		engine->status_page.obj = NULL;
@@ -1995,7 +1997,6 @@ logical_ring_default_irqs(struct intel_engine_cs *engine, unsigned shift)
 {
 	engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT << shift;
 	engine->irq_keep_mask = GT_CONTEXT_SWITCH_INTERRUPT << shift;
-	init_waitqueue_head(&engine->irq_queue);
 }
 
 static int
@@ -2022,6 +2023,10 @@ logical_ring_init(struct intel_engine_cs *engine)
 	struct i915_gem_context *dctx = engine->i915->kernel_context;
 	int ret;
 
+	ret = intel_engine_init_breadcrumbs(engine);
+	if (ret)
+		goto error;
+
 	ret = i915_cmd_parser_init_ring(engine);
 	if (ret)
 		goto error;
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 04a2d141e690..d9dc6bc98194 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -2390,7 +2390,9 @@ static int intel_init_ring_buffer(struct drm_device *dev,
 	memset(engine->semaphore.sync_seqno, 0,
 	       sizeof(engine->semaphore.sync_seqno));
 
-	init_waitqueue_head(&engine->irq_queue);
+	ret = intel_engine_init_breadcrumbs(engine);
+	if (ret)
+		goto error;
 
 	/* We may need to do things with the shrinker which
 	 * require us to immediately switch back to the default
@@ -2470,6 +2472,7 @@ void intel_cleanup_engine(struct intel_engine_cs *engine)
 
 	i915_cmd_parser_fini_ring(engine);
 	i915_gem_batch_pool_fini(&engine->batch_pool);
+	intel_engine_fini_breadcrumbs(engine);
 
 	intel_ring_context_unpin(dev_priv->kernel_context, engine);
 
@@ -2672,6 +2675,13 @@ void intel_ring_init_seqno(struct intel_engine_cs *engine, u32 seqno)
 	engine->last_submitted_seqno = seqno;
 
 	engine->hangcheck.seqno = seqno;
+
+	/* After manually advancing the seqno, fake the interrupt in case
+	 * there are any waiters for that seqno.
+	 */
+	rcu_read_lock();
+	intel_engine_wakeup(engine);
+	rcu_read_unlock();
 }
 
 static void gen6_bsd_ring_write_tail(struct intel_engine_cs *engine,
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index b33c876fed20..601d6549d0db 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -160,6 +160,32 @@ struct intel_engine_cs {
 	struct intel_ringbuffer *buffer;
 	struct list_head buffers;
 
+	/* Rather than have every client wait upon all user interrupts,
+	 * with the herd waking after every interrupt and each doing the
+	 * heavyweight seqno dance, we delegate the task (of being the
+	 * bottom-half of the user interrupt) to the first client. After
+	 * every interrupt, we wake up one client, who does the heavyweight
+	 * coherent seqno read and either goes back to sleep (if incomplete),
+	 * or wakes up all the completed clients in parallel, before then
+	 * transferring the bottom-half status to the next client in the queue.
+	 *
+	 * Compared to walking the entire list of waiters in a single dedicated
+	 * bottom-half, we reduce the latency of the first waiter by avoiding
+	 * a context switch, but incur additional coherent seqno reads when
+	 * following the chain of request breadcrumbs. Since it is most likely
+	 * that we have a single client waiting on each seqno, then reducing
+	 * the overhead of waking that client is much preferred.
+	 */
+	struct intel_breadcrumbs {
+		spinlock_t lock; /* protects the lists of requests */
+		struct rb_root waiters; /* sorted by retirement, priority */
+		struct intel_wait *first_wait; /* oldest waiter by retirement */
+		struct task_struct *tasklet; /* bh for user interrupts */
+		struct timer_list fake_irq; /* used after a missed interrupt */
+		bool irq_enabled;
+		bool rpm_wakelock;
+	} breadcrumbs;
+
 	/*
 	 * A pool of objects to use as shadow copies of client batch buffers
 	 * when the command parser is enabled. Prevents the client from
@@ -308,8 +334,6 @@ struct intel_engine_cs {
 
 	bool gpu_caches_dirty;
 
-	wait_queue_head_t irq_queue;
-
 	struct i915_gem_context *last_context;
 
 	struct intel_ring_hangcheck hangcheck;
@@ -495,4 +519,48 @@ static inline u32 intel_hws_seqno_address(struct intel_engine_cs *engine)
 	return engine->status_page.gfx_addr + I915_GEM_HWS_INDEX_ADDR;
 }
 
+/* intel_breadcrumbs.c -- user interrupt bottom-half for waiters */
+struct intel_wait {
+	struct rb_node node;
+	struct task_struct *tsk;
+	u32 seqno;
+};
+int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
+static inline void intel_wait_init(struct intel_wait *wait, u32 seqno)
+{
+	wait->tsk = current;
+	wait->seqno = seqno;
+}
+static inline bool intel_wait_complete(const struct intel_wait *wait)
+{
+	return RB_EMPTY_NODE(&wait->node);
+}
+bool intel_engine_add_wait(struct intel_engine_cs *engine,
+			   struct intel_wait *wait);
+void intel_engine_remove_wait(struct intel_engine_cs *engine,
+			      struct intel_wait *wait);
+static inline bool intel_engine_has_waiter(struct intel_engine_cs *engine)
+{
+	return READ_ONCE(engine->breadcrumbs.tasklet);
+}
+static inline bool intel_engine_wakeup(struct intel_engine_cs *engine)
+{
+	bool wakeup = false;
+	struct task_struct *tsk = READ_ONCE(engine->breadcrumbs.tasklet);
+	/* Note that for this not to dangerously chase a dangling pointer,
+	 * the caller is responsible for ensure that the task remain valid for
+	 * wake_up_process() i.e. that the RCU grace period cannot expire.
+	 *
+	 * Also note that tsk is likely to be in !TASK_RUNNING state so an
+	 * early test for tsk->state != TASK_RUNNING before wake_up_process()
+	 * is unlikely to be beneficial.
+	 */
+	if (tsk)
+		wakeup = wake_up_process(tsk);
+	return wakeup;
+}
+void intel_engine_enable_fake_irq(struct intel_engine_cs *engine);
+void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
+unsigned intel_kick_waiters(struct drm_i915_private *i915);
+
 #endif /* _INTEL_RINGBUFFER_H_ */
-- 
2.8.1

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

[PATCH 11/25] drm/i915: Spin after waking up for an interrupt

[Chris Wilson]

When waiting for an interrupt (waiting for the GPU to complete some
work), we know we are the single waiter for the GPU. We also know when
the GPU has nearly completed our request (or at least started processing
it), so after being woken and we detect that the GPU is actively working
on our request, allow the bottom-half (the first waiter who wakes up to
handle checking the seqno after the interrupt) to spin for a very short
while to reduce client latencies.

The impact is minimal, there was an improvement to the realtime-vs-many
clients case, but exporting the function proves useful later.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c  |  2 +-
 drivers/gpu/drm/i915/i915_drv.h      | 26 +++++++++++++++--------
 drivers/gpu/drm/i915/i915_gem.c      | 40 +++++++++++++++++++++---------------
 drivers/gpu/drm/i915/intel_display.c |  2 +-
 drivers/gpu/drm/i915/intel_pm.c      |  4 ++--
 5 files changed, 45 insertions(+), 29 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index c81c27dca74d..dba298c7a8b8 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -663,7 +663,7 @@ static int i915_gem_pageflip_info(struct seq_file *m, void *data)
 					   i915_gem_request_get_seqno(work->flip_queued_req),
 					   dev_priv->next_seqno,
 					   engine->get_seqno(engine),
-					   i915_gem_request_completed(work->flip_queued_req, true));
+					   i915_gem_request_completed(work->flip_queued_req));
 			} else
 				seq_printf(m, "Flip not associated with any ring\n");
 			seq_printf(m, "Flip queued on frame %d, (was ready on frame %d), now %d\n",
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index ed92c434f4e6..4cc372df3534 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3277,24 +3277,27 @@ i915_seqno_passed(uint32_t seq1, uint32_t seq2)
 	return (int32_t)(seq1 - seq2) >= 0;
 }
 
-static inline bool i915_gem_request_started(struct drm_i915_gem_request *req,
-					   bool lazy_coherency)
+static inline bool i915_gem_request_started(const struct drm_i915_gem_request *req)
 {
-	if (!lazy_coherency && req->engine->irq_seqno_barrier)
-		req->engine->irq_seqno_barrier(req->engine);
 	return i915_seqno_passed(req->engine->get_seqno(req->engine),
 				 req->previous_seqno);
 }
 
-static inline bool i915_gem_request_completed(struct drm_i915_gem_request *req,
-					      bool lazy_coherency)
+static inline bool i915_gem_request_completed(const struct drm_i915_gem_request *req)
 {
-	if (!lazy_coherency && req->engine->irq_seqno_barrier)
-		req->engine->irq_seqno_barrier(req->engine);
 	return i915_seqno_passed(req->engine->get_seqno(req->engine),
 				 req->seqno);
 }
 
+bool __i915_spin_request(const struct drm_i915_gem_request *request,
+			 int state, unsigned long timeout_us);
+static inline bool i915_spin_request(const struct drm_i915_gem_request *request,
+				     int state, unsigned long timeout_us)
+{
+	return (i915_gem_request_started(request) &&
+		__i915_spin_request(request, state, timeout_us));
+}
+
 int __must_check i915_gem_get_seqno(struct drm_i915_private *dev_priv, u32 *seqno);
 int __must_check i915_gem_set_seqno(struct drm_device *dev, u32 seqno);
 
@@ -3972,6 +3975,8 @@ static inline void i915_trace_irq_get(struct intel_engine_cs *engine,
 
 static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
 {
+	struct intel_engine_cs *engine = req->engine;
+
 	/* Ensure our read of the seqno is coherent so that we
 	 * do not "miss an interrupt" (i.e. if this is the last
 	 * request and the seqno write from the GPU is not visible
@@ -3983,7 +3988,10 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
 	 * but it is easier and safer to do it every time the waiter
 	 * is woken.
 	 */
-	if (i915_gem_request_completed(req, false))
+	if (engine->irq_seqno_barrier)
+		engine->irq_seqno_barrier(engine);
+
+	if (i915_gem_request_completed(req))
 		return true;
 
 	/* We need to check whether any gpu reset happened in between
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index fe5ba4aa336c..30bd0d0d917b 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1375,9 +1375,9 @@ static bool busywait_stop(unsigned long timeout, unsigned cpu)
 	return this_cpu != cpu;
 }
 
-static bool __i915_spin_request(struct drm_i915_gem_request *req, int state)
+bool __i915_spin_request(const struct drm_i915_gem_request *req,
+			 int state, unsigned long timeout_us)
 {
-	unsigned long timeout;
 	unsigned cpu;
 
 	/* When waiting for high frequency requests, e.g. during synchronous
@@ -1390,19 +1390,15 @@ static bool __i915_spin_request(struct drm_i915_gem_request *req, int state)
 	 * takes to sleep on a request, on the order of a microsecond.
 	 */
 
-	/* Only spin if we know the GPU is processing this request */
-	if (!i915_gem_request_started(req, true))
-		return false;
-
-	timeout = local_clock_us(&cpu) + 5;
+	timeout_us += local_clock_us(&cpu);
 	do {
-		if (i915_gem_request_completed(req, true))
+		if (i915_gem_request_completed(req))
 			return true;
 
 		if (signal_pending_state(state, current))
 			break;
 
-		if (busywait_stop(timeout, cpu))
+		if (busywait_stop(timeout_us, cpu))
 			break;
 
 		cpu_relax_lowlatency();
@@ -1445,7 +1441,7 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 	if (list_empty(&req->list))
 		return 0;
 
-	if (i915_gem_request_completed(req, true))
+	if (i915_gem_request_completed(req))
 		return 0;
 
 	timeout_remain = MAX_SCHEDULE_TIMEOUT;
@@ -1470,7 +1466,7 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 		gen6_rps_boost(req->i915, rps, req->emitted_jiffies);
 
 	/* Optimistic spin for the next ~jiffie before touching IRQs */
-	if (__i915_spin_request(req, state))
+	if (i915_spin_request(req, state, 5))
 		goto complete;
 
 	set_current_state(state);
@@ -1512,6 +1508,10 @@ wakeup:
 		 */
 		if (__i915_request_irq_complete(req))
 			break;
+
+		/* Only spin if we know the GPU is processing this request */
+		if (i915_spin_request(req, state, 2))
+			break;
 	}
 	remove_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
 
@@ -3050,8 +3050,16 @@ i915_gem_find_active_request(struct intel_engine_cs *engine)
 {
 	struct drm_i915_gem_request *request;
 
+	/* We are called by the error capture and reset at a random
+	 * point in time. In particular, note that neither is crucially
+	 * ordered with an interrupt. After a hang, the GPU is dead and we
+	 * assume that no more writes can happen (we waited long enough for
+	 * all writes that were in transaction to be flushed) - adding an
+	 * extra delay for a recent interrupt is pointless. Hence, we do
+	 * not need an engine->irq_seqno_barrier() before the seqno reads.
+	 */
 	list_for_each_entry(request, &engine->request_list, list) {
-		if (i915_gem_request_completed(request, false))
+		if (i915_gem_request_completed(request))
 			continue;
 
 		return request;
@@ -3183,7 +3191,7 @@ i915_gem_retire_requests_ring(struct intel_engine_cs *engine)
 					   struct drm_i915_gem_request,
 					   list);
 
-		if (!i915_gem_request_completed(request, true))
+		if (!i915_gem_request_completed(request))
 			break;
 
 		i915_gem_request_retire(request);
@@ -3207,7 +3215,7 @@ i915_gem_retire_requests_ring(struct intel_engine_cs *engine)
 	}
 
 	if (unlikely(engine->trace_irq_req &&
-		     i915_gem_request_completed(engine->trace_irq_req, true))) {
+		     i915_gem_request_completed(engine->trace_irq_req))) {
 		engine->irq_put(engine);
 		i915_gem_request_assign(&engine->trace_irq_req, NULL);
 	}
@@ -3305,7 +3313,7 @@ i915_gem_object_flush_active(struct drm_i915_gem_object *obj)
 		if (req == NULL)
 			continue;
 
-		if (i915_gem_request_completed(req, true))
+		if (i915_gem_request_completed(req))
 			i915_gem_object_retire__read(obj, i);
 	}
 
@@ -3413,7 +3421,7 @@ __i915_gem_object_sync(struct drm_i915_gem_object *obj,
 	if (to == from)
 		return 0;
 
-	if (i915_gem_request_completed(from_req, true))
+	if (i915_gem_request_completed(from_req))
 		return 0;
 
 	if (!i915_semaphore_is_enabled(to_i915(obj->base.dev))) {
diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index 699da067bb85..adeafbc96760 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -11594,7 +11594,7 @@ static bool __pageflip_stall_check_cs(struct drm_i915_private *dev_priv,
 	vblank = intel_crtc_get_vblank_counter(intel_crtc);
 	if (work->flip_ready_vblank == 0) {
 		if (work->flip_queued_req &&
-		    !i915_gem_request_completed(work->flip_queued_req, true))
+		    !i915_gem_request_completed(work->flip_queued_req))
 			return false;
 
 		work->flip_ready_vblank = vblank;
diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index 4f612b96224e..2f54fc5d1131 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -7739,7 +7739,7 @@ static void __intel_rps_boost_work(struct work_struct *work)
 	struct request_boost *boost = container_of(work, struct request_boost, work);
 	struct drm_i915_gem_request *req = boost->req;
 
-	if (!i915_gem_request_completed(req, true))
+	if (!i915_gem_request_completed(req))
 		gen6_rps_boost(req->i915, NULL, req->emitted_jiffies);
 
 	i915_gem_request_unreference(req);
@@ -7753,7 +7753,7 @@ void intel_queue_rps_boost_for_request(struct drm_i915_gem_request *req)
 	if (req == NULL || INTEL_GEN(req->i915) < 6)
 		return;
 
-	if (i915_gem_request_completed(req, true))
+	if (i915_gem_request_completed(req))
 		return;
 
 	boost = kmalloc(sizeof(*boost), GFP_ATOMIC);
-- 
2.8.1

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

Re: [PATCH 11/25] drm/i915: Spin after waking up for an interrupt

[Tvrtko Ursulin]

On 25/06/16 11:12, Chris Wilson wrote:
> When waiting for an interrupt (waiting for the GPU to complete some
> work), we know we are the single waiter for the GPU. We also know when

You mean we know we were the first waiter, if we have been woken up? Or 
perhaps the single waiter for this request? (more or less, maybe close 
enough) Or also s/", we know we are/, and we know we are" ?

Regards,

Tvrtko

> the GPU has nearly completed our request (or at least started processing
> it), so after being woken and we detect that the GPU is actively working
> on our request, allow the bottom-half (the first waiter who wakes up to
> handle checking the seqno after the interrupt) to spin for a very short
> while to reduce client latencies.
>
> The impact is minimal, there was an improvement to the realtime-vs-many
> clients case, but exporting the function proves useful later.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_debugfs.c  |  2 +-
>   drivers/gpu/drm/i915/i915_drv.h      | 26 +++++++++++++++--------
>   drivers/gpu/drm/i915/i915_gem.c      | 40 +++++++++++++++++++++---------------
>   drivers/gpu/drm/i915/intel_display.c |  2 +-
>   drivers/gpu/drm/i915/intel_pm.c      |  4 ++--
>   5 files changed, 45 insertions(+), 29 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
> index c81c27dca74d..dba298c7a8b8 100644
> --- a/drivers/gpu/drm/i915/i915_debugfs.c
> +++ b/drivers/gpu/drm/i915/i915_debugfs.c
> @@ -663,7 +663,7 @@ static int i915_gem_pageflip_info(struct seq_file *m, void *data)
>   					   i915_gem_request_get_seqno(work->flip_queued_req),
>   					   dev_priv->next_seqno,
>   					   engine->get_seqno(engine),
> -					   i915_gem_request_completed(work->flip_queued_req, true));
> +					   i915_gem_request_completed(work->flip_queued_req));
>   			} else
>   				seq_printf(m, "Flip not associated with any ring\n");
>   			seq_printf(m, "Flip queued on frame %d, (was ready on frame %d), now %d\n",
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index ed92c434f4e6..4cc372df3534 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -3277,24 +3277,27 @@ i915_seqno_passed(uint32_t seq1, uint32_t seq2)
>   	return (int32_t)(seq1 - seq2) >= 0;
>   }
>
> -static inline bool i915_gem_request_started(struct drm_i915_gem_request *req,
> -					   bool lazy_coherency)
> +static inline bool i915_gem_request_started(const struct drm_i915_gem_request *req)
>   {
> -	if (!lazy_coherency && req->engine->irq_seqno_barrier)
> -		req->engine->irq_seqno_barrier(req->engine);
>   	return i915_seqno_passed(req->engine->get_seqno(req->engine),
>   				 req->previous_seqno);
>   }
>
> -static inline bool i915_gem_request_completed(struct drm_i915_gem_request *req,
> -					      bool lazy_coherency)
> +static inline bool i915_gem_request_completed(const struct drm_i915_gem_request *req)
>   {
> -	if (!lazy_coherency && req->engine->irq_seqno_barrier)
> -		req->engine->irq_seqno_barrier(req->engine);
>   	return i915_seqno_passed(req->engine->get_seqno(req->engine),
>   				 req->seqno);
>   }
>
> +bool __i915_spin_request(const struct drm_i915_gem_request *request,
> +			 int state, unsigned long timeout_us);
> +static inline bool i915_spin_request(const struct drm_i915_gem_request *request,
> +				     int state, unsigned long timeout_us)
> +{
> +	return (i915_gem_request_started(request) &&
> +		__i915_spin_request(request, state, timeout_us));
> +}
> +
>   int __must_check i915_gem_get_seqno(struct drm_i915_private *dev_priv, u32 *seqno);
>   int __must_check i915_gem_set_seqno(struct drm_device *dev, u32 seqno);
>
> @@ -3972,6 +3975,8 @@ static inline void i915_trace_irq_get(struct intel_engine_cs *engine,
>
>   static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
>   {
> +	struct intel_engine_cs *engine = req->engine;
> +
>   	/* Ensure our read of the seqno is coherent so that we
>   	 * do not "miss an interrupt" (i.e. if this is the last
>   	 * request and the seqno write from the GPU is not visible
> @@ -3983,7 +3988,10 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
>   	 * but it is easier and safer to do it every time the waiter
>   	 * is woken.
>   	 */
> -	if (i915_gem_request_completed(req, false))
> +	if (engine->irq_seqno_barrier)
> +		engine->irq_seqno_barrier(engine);
> +
> +	if (i915_gem_request_completed(req))
>   		return true;
>
>   	/* We need to check whether any gpu reset happened in between
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index fe5ba4aa336c..30bd0d0d917b 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -1375,9 +1375,9 @@ static bool busywait_stop(unsigned long timeout, unsigned cpu)
>   	return this_cpu != cpu;
>   }
>
> -static bool __i915_spin_request(struct drm_i915_gem_request *req, int state)
> +bool __i915_spin_request(const struct drm_i915_gem_request *req,
> +			 int state, unsigned long timeout_us)
>   {
> -	unsigned long timeout;
>   	unsigned cpu;
>
>   	/* When waiting for high frequency requests, e.g. during synchronous
> @@ -1390,19 +1390,15 @@ static bool __i915_spin_request(struct drm_i915_gem_request *req, int state)
>   	 * takes to sleep on a request, on the order of a microsecond.
>   	 */
>
> -	/* Only spin if we know the GPU is processing this request */
> -	if (!i915_gem_request_started(req, true))
> -		return false;
> -
> -	timeout = local_clock_us(&cpu) + 5;
> +	timeout_us += local_clock_us(&cpu);
>   	do {
> -		if (i915_gem_request_completed(req, true))
> +		if (i915_gem_request_completed(req))
>   			return true;
>
>   		if (signal_pending_state(state, current))
>   			break;
>
> -		if (busywait_stop(timeout, cpu))
> +		if (busywait_stop(timeout_us, cpu))
>   			break;
>
>   		cpu_relax_lowlatency();
> @@ -1445,7 +1441,7 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
>   	if (list_empty(&req->list))
>   		return 0;
>
> -	if (i915_gem_request_completed(req, true))
> +	if (i915_gem_request_completed(req))
>   		return 0;
>
>   	timeout_remain = MAX_SCHEDULE_TIMEOUT;
> @@ -1470,7 +1466,7 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
>   		gen6_rps_boost(req->i915, rps, req->emitted_jiffies);
>
>   	/* Optimistic spin for the next ~jiffie before touching IRQs */
> -	if (__i915_spin_request(req, state))
> +	if (i915_spin_request(req, state, 5))
>   		goto complete;
>
>   	set_current_state(state);
> @@ -1512,6 +1508,10 @@ wakeup:
>   		 */
>   		if (__i915_request_irq_complete(req))
>   			break;
> +
> +		/* Only spin if we know the GPU is processing this request */
> +		if (i915_spin_request(req, state, 2))
> +			break;
>   	}
>   	remove_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
>
> @@ -3050,8 +3050,16 @@ i915_gem_find_active_request(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_gem_request *request;
>
> +	/* We are called by the error capture and reset at a random
> +	 * point in time. In particular, note that neither is crucially
> +	 * ordered with an interrupt. After a hang, the GPU is dead and we
> +	 * assume that no more writes can happen (we waited long enough for
> +	 * all writes that were in transaction to be flushed) - adding an
> +	 * extra delay for a recent interrupt is pointless. Hence, we do
> +	 * not need an engine->irq_seqno_barrier() before the seqno reads.
> +	 */
>   	list_for_each_entry(request, &engine->request_list, list) {
> -		if (i915_gem_request_completed(request, false))
> +		if (i915_gem_request_completed(request))
>   			continue;
>
>   		return request;
> @@ -3183,7 +3191,7 @@ i915_gem_retire_requests_ring(struct intel_engine_cs *engine)
>   					   struct drm_i915_gem_request,
>   					   list);
>
> -		if (!i915_gem_request_completed(request, true))
> +		if (!i915_gem_request_completed(request))
>   			break;
>
>   		i915_gem_request_retire(request);
> @@ -3207,7 +3215,7 @@ i915_gem_retire_requests_ring(struct intel_engine_cs *engine)
>   	}
>
>   	if (unlikely(engine->trace_irq_req &&
> -		     i915_gem_request_completed(engine->trace_irq_req, true))) {
> +		     i915_gem_request_completed(engine->trace_irq_req))) {
>   		engine->irq_put(engine);
>   		i915_gem_request_assign(&engine->trace_irq_req, NULL);
>   	}
> @@ -3305,7 +3313,7 @@ i915_gem_object_flush_active(struct drm_i915_gem_object *obj)
>   		if (req == NULL)
>   			continue;
>
> -		if (i915_gem_request_completed(req, true))
> +		if (i915_gem_request_completed(req))
>   			i915_gem_object_retire__read(obj, i);
>   	}
>
> @@ -3413,7 +3421,7 @@ __i915_gem_object_sync(struct drm_i915_gem_object *obj,
>   	if (to == from)
>   		return 0;
>
> -	if (i915_gem_request_completed(from_req, true))
> +	if (i915_gem_request_completed(from_req))
>   		return 0;
>
>   	if (!i915_semaphore_is_enabled(to_i915(obj->base.dev))) {
> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> index 699da067bb85..adeafbc96760 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -11594,7 +11594,7 @@ static bool __pageflip_stall_check_cs(struct drm_i915_private *dev_priv,
>   	vblank = intel_crtc_get_vblank_counter(intel_crtc);
>   	if (work->flip_ready_vblank == 0) {
>   		if (work->flip_queued_req &&
> -		    !i915_gem_request_completed(work->flip_queued_req, true))
> +		    !i915_gem_request_completed(work->flip_queued_req))
>   			return false;
>
>   		work->flip_ready_vblank = vblank;
> diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
> index 4f612b96224e..2f54fc5d1131 100644
> --- a/drivers/gpu/drm/i915/intel_pm.c
> +++ b/drivers/gpu/drm/i915/intel_pm.c
> @@ -7739,7 +7739,7 @@ static void __intel_rps_boost_work(struct work_struct *work)
>   	struct request_boost *boost = container_of(work, struct request_boost, work);
>   	struct drm_i915_gem_request *req = boost->req;
>
> -	if (!i915_gem_request_completed(req, true))
> +	if (!i915_gem_request_completed(req))
>   		gen6_rps_boost(req->i915, NULL, req->emitted_jiffies);
>
>   	i915_gem_request_unreference(req);
> @@ -7753,7 +7753,7 @@ void intel_queue_rps_boost_for_request(struct drm_i915_gem_request *req)
>   	if (req == NULL || INTEL_GEN(req->i915) < 6)
>   		return;
>
> -	if (i915_gem_request_completed(req, true))
> +	if (i915_gem_request_completed(req))
>   		return;
>
>   	boost = kmalloc(sizeof(*boost), GFP_ATOMIC);
>
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 11/25] drm/i915: Spin after waking up for an interrupt

[Chris Wilson]

On Mon, Jun 27, 2016 at 11:32:10AM +0100, Tvrtko Ursulin wrote:
> 
> On 25/06/16 11:12, Chris Wilson wrote:
> >When waiting for an interrupt (waiting for the GPU to complete some
> >work), we know we are the single waiter for the GPU. We also know when
> 
> You mean we know we were the first waiter, if we have been woken up?
> Or perhaps the single waiter for this request? (more or less, maybe
> close enough) Or also s/", we know we are/, and we know we are" ?

Single waiter for the engine. s/and//; it follows from and is not
conditional. To be here we know we are the only piece of code performing
this check. There may be threads entering i915_wait_request that also
get to spin before adding themselves to the waitqueue, but of all waiters,
we are the only one allowed to spin on this engine.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 11/25] drm/i915: Spin after waking up for an interrupt

[Chris Wilson]

On Tue, Jun 28, 2016 at 09:55:17AM +0100, Chris Wilson wrote:
> On Mon, Jun 27, 2016 at 11:32:10AM +0100, Tvrtko Ursulin wrote:
> > 
> > On 25/06/16 11:12, Chris Wilson wrote:
> > >When waiting for an interrupt (waiting for the GPU to complete some
> > >work), we know we are the single waiter for the GPU. We also know when
> > 
> > You mean we know we were the first waiter, if we have been woken up?
> > Or perhaps the single waiter for this request? (more or less, maybe
> > close enough) Or also s/", we know we are/, and we know we are" ?
> 
> Single waiter for the engine. s/and//; it follows from and is not
> conditional. To be here we know we are the only piece of code performing
> this check. There may be threads entering i915_wait_request that also
> get to spin before adding themselves to the waitqueue, but of all waiters,
> we are the only one allowed to spin on this engine.

Changed to:

When waiting for an interrupt (waiting for the engine to complete some
work), we know we are the only waiter to be woken on this engine. We also 
know when the GPU has nearly completed our request (or at least started
processing it), so after being woken and we detect that the GPU is
active and working on our request, allow us the bottom-half (the first
waiter who wakes up to handle checking the seqno after the interrupt) to
spin for a very short while to reduce client latencies.

The impact is minimal, there was an improvement to the realtime-vs-many
clients case, but exporting the function proves useful later. However,
it is tempting to adjust irq_seqno_barrier to include the spin. The
problem is first ensuring that the "start-of-request" seqno is coherent
as we use that as our basis for judging when it is ok to spin. If we
could, spinning there could dramatically shorten some sleeps, and allow
us to make the barriers more conservative to handle missed seqno writes
on more platforms (all gen7+ are known to have the occasional issue, at 
least).

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 11/25] drm/i915: Spin after waking up for an interrupt

[Tvrtko Ursulin]

On 28/06/16 10:17, Chris Wilson wrote:
> On Tue, Jun 28, 2016 at 09:55:17AM +0100, Chris Wilson wrote:
>> On Mon, Jun 27, 2016 at 11:32:10AM +0100, Tvrtko Ursulin wrote:
>>>
>>> On 25/06/16 11:12, Chris Wilson wrote:
>>>> When waiting for an interrupt (waiting for the GPU to complete some
>>>> work), we know we are the single waiter for the GPU. We also know when
>>>
>>> You mean we know we were the first waiter, if we have been woken up?
>>> Or perhaps the single waiter for this request? (more or less, maybe
>>> close enough) Or also s/", we know we are/, and we know we are" ?
>>
>> Single waiter for the engine. s/and//; it follows from and is not
>> conditional. To be here we know we are the only piece of code performing
>> this check. There may be threads entering i915_wait_request that also
>> get to spin before adding themselves to the waitqueue, but of all waiters,
>> we are the only one allowed to spin on this engine.
>
> Changed to:
>
> When waiting for an interrupt (waiting for the engine to complete some
> work), we know we are the only waiter to be woken on this engine. We also
> know when the GPU has nearly completed our request (or at least started
> processing it), so after being woken and we detect that the GPU is
> active and working on our request, allow us the bottom-half (the first
> waiter who wakes up to handle checking the seqno after the interrupt) to
> spin for a very short while to reduce client latencies.
>
> The impact is minimal, there was an improvement to the realtime-vs-many
> clients case, but exporting the function proves useful later. However,
> it is tempting to adjust irq_seqno_barrier to include the spin. The
> problem is first ensuring that the "start-of-request" seqno is coherent
> as we use that as our basis for judging when it is ok to spin. If we
> could, spinning there could dramatically shorten some sleeps, and allow
> us to make the barriers more conservative to handle missed seqno writes
> on more platforms (all gen7+ are known to have the occasional issue, at
> least).

Super clear now;

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

[PATCH 12/25] drm/i915: Use HWS for seqno tracking everywhere

[Chris Wilson]

By using the same address for storing the HWS on every platform, we can
remove the platform specific vfuncs and reduce the get-seqno routine to
a single read of a cached memory location.

v2: Fix semaphore_passed() to look at the signaling engine (not the
waiter's)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c      |  6 +--
 drivers/gpu/drm/i915/i915_drv.h          |  4 +-
 drivers/gpu/drm/i915/i915_gpu_error.c    |  2 +-
 drivers/gpu/drm/i915/i915_irq.c          |  4 +-
 drivers/gpu/drm/i915/i915_trace.h        |  2 +-
 drivers/gpu/drm/i915/intel_breadcrumbs.c |  4 +-
 drivers/gpu/drm/i915/intel_lrc.c         | 26 +---------
 drivers/gpu/drm/i915/intel_ringbuffer.c  | 83 ++++++++------------------------
 drivers/gpu/drm/i915/intel_ringbuffer.h  |  7 +--
 9 files changed, 36 insertions(+), 102 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index dba298c7a8b8..30dcae2bf303 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -662,7 +662,7 @@ static int i915_gem_pageflip_info(struct seq_file *m, void *data)
 					   engine->name,
 					   i915_gem_request_get_seqno(work->flip_queued_req),
 					   dev_priv->next_seqno,
-					   engine->get_seqno(engine),
+					   intel_engine_get_seqno(engine),
 					   i915_gem_request_completed(work->flip_queued_req));
 			} else
 				seq_printf(m, "Flip not associated with any ring\n");
@@ -792,7 +792,7 @@ static void i915_ring_seqno_info(struct seq_file *m,
 	struct rb_node *rb;
 
 	seq_printf(m, "Current sequence (%s): %x\n",
-		   engine->name, engine->get_seqno(engine));
+		   engine->name, intel_engine_get_seqno(engine));
 	seq_printf(m, "Current user interrupts (%s): %x\n",
 		   engine->name, READ_ONCE(engine->user_interrupts));
 
@@ -1417,7 +1417,7 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
 
 	for_each_engine_id(engine, dev_priv, id) {
 		acthd[id] = intel_ring_get_active_head(engine);
-		seqno[id] = engine->get_seqno(engine);
+		seqno[id] = intel_engine_get_seqno(engine);
 	}
 
 	i915_get_extra_instdone(dev_priv, instdone);
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 4cc372df3534..f4b883a58b68 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3279,13 +3279,13 @@ i915_seqno_passed(uint32_t seq1, uint32_t seq2)
 
 static inline bool i915_gem_request_started(const struct drm_i915_gem_request *req)
 {
-	return i915_seqno_passed(req->engine->get_seqno(req->engine),
+	return i915_seqno_passed(intel_engine_get_seqno(req->engine),
 				 req->previous_seqno);
 }
 
 static inline bool i915_gem_request_completed(const struct drm_i915_gem_request *req)
 {
-	return i915_seqno_passed(req->engine->get_seqno(req->engine),
+	return i915_seqno_passed(intel_engine_get_seqno(req->engine),
 				 req->seqno);
 }
 
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index 606f74e15208..bf0851d13916 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -983,7 +983,7 @@ static void i915_record_ring_state(struct drm_i915_private *dev_priv,
 	ering->waiting = intel_engine_has_waiter(engine);
 	ering->instpm = I915_READ(RING_INSTPM(engine->mmio_base));
 	ering->acthd = intel_ring_get_active_head(engine);
-	ering->seqno = engine->get_seqno(engine);
+	ering->seqno = intel_engine_get_seqno(engine);
 	ering->last_seqno = engine->last_submitted_seqno;
 	ering->start = I915_READ_START(engine);
 	ering->head = I915_READ_HEAD(engine);
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index a11ab00cdee0..7c379afcff2f 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2952,7 +2952,7 @@ static int semaphore_passed(struct intel_engine_cs *engine)
 	if (signaller->hangcheck.deadlock >= I915_NUM_ENGINES)
 		return -1;
 
-	if (i915_seqno_passed(signaller->get_seqno(signaller), seqno))
+	if (i915_seqno_passed(intel_engine_get_seqno(signaller), seqno))
 		return 1;
 
 	/* cursory check for an unkickable deadlock */
@@ -3140,7 +3140,7 @@ static void i915_hangcheck_elapsed(struct work_struct *work)
 			engine->irq_seqno_barrier(engine);
 
 		acthd = intel_ring_get_active_head(engine);
-		seqno = engine->get_seqno(engine);
+		seqno = intel_engine_get_seqno(engine);
 
 		/* Reset stuck interrupts between batch advances */
 		user_interrupts = 0;
diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h
index 6768db032f84..3d13fde95fdf 100644
--- a/drivers/gpu/drm/i915/i915_trace.h
+++ b/drivers/gpu/drm/i915/i915_trace.h
@@ -558,7 +558,7 @@ TRACE_EVENT(i915_gem_request_notify,
 	    TP_fast_assign(
 			   __entry->dev = engine->i915->dev->primary->index;
 			   __entry->ring = engine->id;
-			   __entry->seqno = engine->get_seqno(engine);
+			   __entry->seqno = intel_engine_get_seqno(engine);
 			   ),
 
 	    TP_printk("dev=%u, ring=%u, seqno=%u",
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index 11ec077660a1..cb5e85192669 100644
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -146,7 +146,7 @@ static bool __intel_engine_add_wait(struct intel_engine_cs *engine,
 	first = true;
 	parent = NULL;
 	completed = NULL;
-	seqno = engine->get_seqno(engine);
+	seqno = intel_engine_get_seqno(engine);
 
 	 /* If the request completed before we managed to grab the spinlock,
 	  * return now before adding ourselves to the rbtree. We let the
@@ -294,7 +294,7 @@ void intel_engine_remove_wait(struct intel_engine_cs *engine,
 			 * the first_waiter. This is undesirable if that
 			 * waiter is a high priority task.
 			 */
-			u32 seqno = engine->get_seqno(engine);
+			u32 seqno = intel_engine_get_seqno(engine);
 			while (i915_seqno_passed(seqno, to_wait(next)->seqno)) {
 				struct rb_node *n = rb_next(next);
 				__intel_breadcrumbs_finish(b, to_wait(next));
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index faa62d626211..97dc2ea59632 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1780,16 +1780,6 @@ static int gen8_emit_flush_render(struct drm_i915_gem_request *request,
 	return 0;
 }
 
-static u32 gen8_get_seqno(struct intel_engine_cs *engine)
-{
-	return intel_read_status_page(engine, I915_GEM_HWS_INDEX);
-}
-
-static void gen8_set_seqno(struct intel_engine_cs *engine, u32 seqno)
-{
-	intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
-}
-
 static void bxt_a_seqno_barrier(struct intel_engine_cs *engine)
 {
 	/*
@@ -1805,14 +1795,6 @@ static void bxt_a_seqno_barrier(struct intel_engine_cs *engine)
 	intel_flush_status_page(engine, I915_GEM_HWS_INDEX);
 }
 
-static void bxt_a_set_seqno(struct intel_engine_cs *engine, u32 seqno)
-{
-	intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
-
-	/* See bxt_a_get_seqno() explaining the reason for the clflush. */
-	intel_flush_status_page(engine, I915_GEM_HWS_INDEX);
-}
-
 /*
  * Reserve space for 2 NOOPs at the end of each request to be
  * used as a workaround for not being allowed to do lite
@@ -1838,7 +1820,7 @@ static int gen8_emit_request(struct drm_i915_gem_request *request)
 				intel_hws_seqno_address(request->engine) |
 				MI_FLUSH_DW_USE_GTT);
 	intel_logical_ring_emit(ringbuf, 0);
-	intel_logical_ring_emit(ringbuf, i915_gem_request_get_seqno(request));
+	intel_logical_ring_emit(ringbuf, request->seqno);
 	intel_logical_ring_emit(ringbuf, MI_USER_INTERRUPT);
 	intel_logical_ring_emit(ringbuf, MI_NOOP);
 	return intel_logical_ring_advance_and_submit(request);
@@ -1984,12 +1966,8 @@ logical_ring_default_vfuncs(struct intel_engine_cs *engine)
 	engine->irq_get = gen8_logical_ring_get_irq;
 	engine->irq_put = gen8_logical_ring_put_irq;
 	engine->emit_bb_start = gen8_emit_bb_start;
-	engine->get_seqno = gen8_get_seqno;
-	engine->set_seqno = gen8_set_seqno;
-	if (IS_BXT_REVID(engine->i915, 0, BXT_REVID_A1)) {
+	if (IS_BXT_REVID(engine->i915, 0, BXT_REVID_A1))
 		engine->irq_seqno_barrier = bxt_a_seqno_barrier;
-		engine->set_seqno = bxt_a_set_seqno;
-	}
 }
 
 static inline void
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index d9dc6bc98194..96a80e96dbcf 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -1362,19 +1362,17 @@ static int gen8_rcs_signal(struct drm_i915_gem_request *signaller_req,
 		return ret;
 
 	for_each_engine_id(waiter, dev_priv, id) {
-		u32 seqno;
 		u64 gtt_offset = signaller->semaphore.signal_ggtt[id];
 		if (gtt_offset == MI_SEMAPHORE_SYNC_INVALID)
 			continue;
 
-		seqno = i915_gem_request_get_seqno(signaller_req);
 		intel_ring_emit(signaller, GFX_OP_PIPE_CONTROL(6));
 		intel_ring_emit(signaller, PIPE_CONTROL_GLOBAL_GTT_IVB |
 					   PIPE_CONTROL_QW_WRITE |
 					   PIPE_CONTROL_CS_STALL);
 		intel_ring_emit(signaller, lower_32_bits(gtt_offset));
 		intel_ring_emit(signaller, upper_32_bits(gtt_offset));
-		intel_ring_emit(signaller, seqno);
+		intel_ring_emit(signaller, signaller_req->seqno);
 		intel_ring_emit(signaller, 0);
 		intel_ring_emit(signaller, MI_SEMAPHORE_SIGNAL |
 					   MI_SEMAPHORE_TARGET(waiter->hw_id));
@@ -1403,18 +1401,16 @@ static int gen8_xcs_signal(struct drm_i915_gem_request *signaller_req,
 		return ret;
 
 	for_each_engine_id(waiter, dev_priv, id) {
-		u32 seqno;
 		u64 gtt_offset = signaller->semaphore.signal_ggtt[id];
 		if (gtt_offset == MI_SEMAPHORE_SYNC_INVALID)
 			continue;
 
-		seqno = i915_gem_request_get_seqno(signaller_req);
 		intel_ring_emit(signaller, (MI_FLUSH_DW + 1) |
 					   MI_FLUSH_DW_OP_STOREDW);
 		intel_ring_emit(signaller, lower_32_bits(gtt_offset) |
 					   MI_FLUSH_DW_USE_GTT);
 		intel_ring_emit(signaller, upper_32_bits(gtt_offset));
-		intel_ring_emit(signaller, seqno);
+		intel_ring_emit(signaller, signaller_req->seqno);
 		intel_ring_emit(signaller, MI_SEMAPHORE_SIGNAL |
 					   MI_SEMAPHORE_TARGET(waiter->hw_id));
 		intel_ring_emit(signaller, 0);
@@ -1445,11 +1441,9 @@ static int gen6_signal(struct drm_i915_gem_request *signaller_req,
 		i915_reg_t mbox_reg = signaller->semaphore.mbox.signal[id];
 
 		if (i915_mmio_reg_valid(mbox_reg)) {
-			u32 seqno = i915_gem_request_get_seqno(signaller_req);
-
 			intel_ring_emit(signaller, MI_LOAD_REGISTER_IMM(1));
 			intel_ring_emit_reg(signaller, mbox_reg);
-			intel_ring_emit(signaller, seqno);
+			intel_ring_emit(signaller, signaller_req->seqno);
 		}
 	}
 
@@ -1485,7 +1479,7 @@ gen6_add_request(struct drm_i915_gem_request *req)
 	intel_ring_emit(engine, MI_STORE_DWORD_INDEX);
 	intel_ring_emit(engine,
 			I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
-	intel_ring_emit(engine, i915_gem_request_get_seqno(req));
+	intel_ring_emit(engine, req->seqno);
 	intel_ring_emit(engine, MI_USER_INTERRUPT);
 	__intel_ring_advance(engine);
 
@@ -1624,7 +1618,9 @@ static int
 pc_render_add_request(struct drm_i915_gem_request *req)
 {
 	struct intel_engine_cs *engine = req->engine;
-	u32 scratch_addr = engine->scratch.gtt_offset + 2 * CACHELINE_BYTES;
+	u32 addr = engine->status_page.gfx_addr +
+		(I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
+	u32 scratch_addr = addr;
 	int ret;
 
 	/* For Ironlake, MI_USER_INTERRUPT was deprecated and apparently
@@ -1640,12 +1636,12 @@ pc_render_add_request(struct drm_i915_gem_request *req)
 		return ret;
 
 	intel_ring_emit(engine,
-			GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE |
+			GFX_OP_PIPE_CONTROL(4) |
+			PIPE_CONTROL_QW_WRITE |
 			PIPE_CONTROL_WRITE_FLUSH |
 			PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE);
-	intel_ring_emit(engine,
-			engine->scratch.gtt_offset | PIPE_CONTROL_GLOBAL_GTT);
-	intel_ring_emit(engine, i915_gem_request_get_seqno(req));
+	intel_ring_emit(engine, addr | PIPE_CONTROL_GLOBAL_GTT);
+	intel_ring_emit(engine, req->seqno);
 	intel_ring_emit(engine, 0);
 	PIPE_CONTROL_FLUSH(engine, scratch_addr);
 	scratch_addr += 2 * CACHELINE_BYTES; /* write to separate cachelines */
@@ -1660,13 +1656,12 @@ pc_render_add_request(struct drm_i915_gem_request *req)
 	PIPE_CONTROL_FLUSH(engine, scratch_addr);
 
 	intel_ring_emit(engine,
-			GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE |
+		       	GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE |
 			PIPE_CONTROL_WRITE_FLUSH |
 			PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE |
 			PIPE_CONTROL_NOTIFY);
-	intel_ring_emit(engine,
-			engine->scratch.gtt_offset | PIPE_CONTROL_GLOBAL_GTT);
-	intel_ring_emit(engine, i915_gem_request_get_seqno(req));
+	intel_ring_emit(engine, addr | PIPE_CONTROL_GLOBAL_GTT);
+	intel_ring_emit(engine, req->seqno);
 	intel_ring_emit(engine, 0);
 	__intel_ring_advance(engine);
 
@@ -1698,30 +1693,6 @@ gen6_seqno_barrier(struct intel_engine_cs *engine)
 	spin_unlock_irq(&dev_priv->uncore.lock);
 }
 
-static u32
-ring_get_seqno(struct intel_engine_cs *engine)
-{
-	return intel_read_status_page(engine, I915_GEM_HWS_INDEX);
-}
-
-static void
-ring_set_seqno(struct intel_engine_cs *engine, u32 seqno)
-{
-	intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
-}
-
-static u32
-pc_render_get_seqno(struct intel_engine_cs *engine)
-{
-	return engine->scratch.cpu_page[0];
-}
-
-static void
-pc_render_set_seqno(struct intel_engine_cs *engine, u32 seqno)
-{
-	engine->scratch.cpu_page[0] = seqno;
-}
-
 static bool
 gen5_ring_get_irq(struct intel_engine_cs *engine)
 {
@@ -1851,8 +1822,8 @@ i9xx_add_request(struct drm_i915_gem_request *req)
 
 	intel_ring_emit(engine, MI_STORE_DWORD_INDEX);
 	intel_ring_emit(engine,
-			I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
-	intel_ring_emit(engine, i915_gem_request_get_seqno(req));
+		       	I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
+	intel_ring_emit(engine, req->seqno);
 	intel_ring_emit(engine, MI_USER_INTERRUPT);
 	__intel_ring_advance(engine);
 
@@ -2671,7 +2642,9 @@ void intel_ring_init_seqno(struct intel_engine_cs *engine, u32 seqno)
 	memset(engine->semaphore.sync_seqno, 0,
 	       sizeof(engine->semaphore.sync_seqno));
 
-	engine->set_seqno(engine, seqno);
+	intel_write_status_page(engine, I915_GEM_HWS_INDEX, seqno);
+	if (engine->irq_seqno_barrier)
+		engine->irq_seqno_barrier(engine);
 	engine->last_submitted_seqno = seqno;
 
 	engine->hangcheck.seqno = seqno;
@@ -2920,8 +2893,6 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 		engine->irq_get = gen8_ring_get_irq;
 		engine->irq_put = gen8_ring_put_irq;
 		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT;
-		engine->get_seqno = ring_get_seqno;
-		engine->set_seqno = ring_set_seqno;
 		if (i915_semaphore_is_enabled(dev_priv)) {
 			WARN_ON(!dev_priv->semaphore_obj);
 			engine->semaphore.sync_to = gen8_ring_sync;
@@ -2938,8 +2909,6 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 		engine->irq_put = gen6_ring_put_irq;
 		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT;
 		engine->irq_seqno_barrier = gen6_seqno_barrier;
-		engine->get_seqno = ring_get_seqno;
-		engine->set_seqno = ring_set_seqno;
 		if (i915_semaphore_is_enabled(dev_priv)) {
 			engine->semaphore.sync_to = gen6_ring_sync;
 			engine->semaphore.signal = gen6_signal;
@@ -2964,8 +2933,6 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 	} else if (IS_GEN5(dev_priv)) {
 		engine->add_request = pc_render_add_request;
 		engine->flush = gen4_render_ring_flush;
-		engine->get_seqno = pc_render_get_seqno;
-		engine->set_seqno = pc_render_set_seqno;
 		engine->irq_get = gen5_ring_get_irq;
 		engine->irq_put = gen5_ring_put_irq;
 		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT |
@@ -2976,8 +2943,6 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 			engine->flush = gen2_render_ring_flush;
 		else
 			engine->flush = gen4_render_ring_flush;
-		engine->get_seqno = ring_get_seqno;
-		engine->set_seqno = ring_set_seqno;
 		if (IS_GEN2(dev_priv)) {
 			engine->irq_get = i8xx_ring_get_irq;
 			engine->irq_put = i8xx_ring_put_irq;
@@ -3055,8 +3020,6 @@ int intel_init_bsd_ring_buffer(struct drm_device *dev)
 		engine->flush = gen6_bsd_ring_flush;
 		engine->add_request = gen6_add_request;
 		engine->irq_seqno_barrier = gen6_seqno_barrier;
-		engine->get_seqno = ring_get_seqno;
-		engine->set_seqno = ring_set_seqno;
 		if (INTEL_GEN(dev_priv) >= 8) {
 			engine->irq_enable_mask =
 				GT_RENDER_USER_INTERRUPT << GEN8_VCS1_IRQ_SHIFT;
@@ -3094,8 +3057,6 @@ int intel_init_bsd_ring_buffer(struct drm_device *dev)
 		engine->mmio_base = BSD_RING_BASE;
 		engine->flush = bsd_ring_flush;
 		engine->add_request = i9xx_add_request;
-		engine->get_seqno = ring_get_seqno;
-		engine->set_seqno = ring_set_seqno;
 		if (IS_GEN5(dev_priv)) {
 			engine->irq_enable_mask = ILK_BSD_USER_INTERRUPT;
 			engine->irq_get = gen5_ring_get_irq;
@@ -3130,8 +3091,6 @@ int intel_init_bsd2_ring_buffer(struct drm_device *dev)
 	engine->flush = gen6_bsd_ring_flush;
 	engine->add_request = gen6_add_request;
 	engine->irq_seqno_barrier = gen6_seqno_barrier;
-	engine->get_seqno = ring_get_seqno;
-	engine->set_seqno = ring_set_seqno;
 	engine->irq_enable_mask =
 			GT_RENDER_USER_INTERRUPT << GEN8_VCS2_IRQ_SHIFT;
 	engine->irq_get = gen8_ring_get_irq;
@@ -3163,8 +3122,6 @@ int intel_init_blt_ring_buffer(struct drm_device *dev)
 	engine->flush = gen6_ring_flush;
 	engine->add_request = gen6_add_request;
 	engine->irq_seqno_barrier = gen6_seqno_barrier;
-	engine->get_seqno = ring_get_seqno;
-	engine->set_seqno = ring_set_seqno;
 	if (INTEL_GEN(dev_priv) >= 8) {
 		engine->irq_enable_mask =
 			GT_RENDER_USER_INTERRUPT << GEN8_BCS_IRQ_SHIFT;
@@ -3223,8 +3180,6 @@ int intel_init_vebox_ring_buffer(struct drm_device *dev)
 	engine->flush = gen6_ring_flush;
 	engine->add_request = gen6_add_request;
 	engine->irq_seqno_barrier = gen6_seqno_barrier;
-	engine->get_seqno = ring_get_seqno;
-	engine->set_seqno = ring_set_seqno;
 
 	if (INTEL_GEN(dev_priv) >= 8) {
 		engine->irq_enable_mask =
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 601d6549d0db..76e1efd87533 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -219,9 +219,6 @@ struct intel_engine_cs {
 	 * monotonic, even if not coherent.
 	 */
 	void		(*irq_seqno_barrier)(struct intel_engine_cs *ring);
-	u32		(*get_seqno)(struct intel_engine_cs *ring);
-	void		(*set_seqno)(struct intel_engine_cs *ring,
-				     u32 seqno);
 	int		(*dispatch_execbuffer)(struct drm_i915_gem_request *req,
 					       u64 offset, u32 length,
 					       unsigned dispatch_flags);
@@ -497,6 +494,10 @@ int intel_init_blt_ring_buffer(struct drm_device *dev);
 int intel_init_vebox_ring_buffer(struct drm_device *dev);
 
 u64 intel_ring_get_active_head(struct intel_engine_cs *engine);
+static inline u32 intel_engine_get_seqno(struct intel_engine_cs *engine)
+{
+	return intel_read_status_page(engine, I915_GEM_HWS_INDEX);
+}
 
 int init_workarounds_ring(struct intel_engine_cs *engine);
 
-- 
2.8.1

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

[PATCH 13/25] drm/i915: Stop mapping the scratch page into CPU space

[Chris Wilson]

After the elimination of using the scratch page for Ironlake's
breadcrumb, we no longer need to kmap the object. We therefore can move
it into the high unmappable space and do not need to force the object to
be coherent (i.e. snooped on !llc platforms).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/intel_ringbuffer.c | 40 +++++++++------------------------
 drivers/gpu/drm/i915/intel_ringbuffer.h |  1 -
 2 files changed, 11 insertions(+), 30 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 96a80e96dbcf..8474b7fea710 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -643,58 +643,40 @@ out:
 	return ret;
 }
 
-void
-intel_fini_pipe_control(struct intel_engine_cs *engine)
+void intel_fini_pipe_control(struct intel_engine_cs *engine)
 {
 	if (engine->scratch.obj == NULL)
 		return;
 
-	if (INTEL_GEN(engine->i915) >= 5) {
-		kunmap(sg_page(engine->scratch.obj->pages->sgl));
-		i915_gem_object_ggtt_unpin(engine->scratch.obj);
-	}
-
+	i915_gem_object_ggtt_unpin(engine->scratch.obj);
 	drm_gem_object_unreference(&engine->scratch.obj->base);
 	engine->scratch.obj = NULL;
 }
 
-int
-intel_init_pipe_control(struct intel_engine_cs *engine)
+int intel_init_pipe_control(struct intel_engine_cs *engine)
 {
+	struct drm_i915_gem_object *obj;
 	int ret;
 
 	WARN_ON(engine->scratch.obj);
 
-	engine->scratch.obj = i915_gem_object_create(engine->i915->dev, 4096);
-	if (IS_ERR(engine->scratch.obj)) {
-		DRM_ERROR("Failed to allocate seqno page\n");
-		ret = PTR_ERR(engine->scratch.obj);
-		engine->scratch.obj = NULL;
+	obj = i915_gem_object_create(engine->i915->dev, 4096);
+	if (IS_ERR(obj)) {
+		DRM_ERROR("Failed to allocate scratch page\n");
+		ret = PTR_ERR(obj);
 		goto err;
 	}
 
-	ret = i915_gem_object_set_cache_level(engine->scratch.obj,
-					      I915_CACHE_LLC);
+	ret = i915_gem_obj_ggtt_pin(obj, 4096, PIN_HIGH);
 	if (ret)
 		goto err_unref;
 
-	ret = i915_gem_obj_ggtt_pin(engine->scratch.obj, 4096, 0);
-	if (ret)
-		goto err_unref;
-
-	engine->scratch.gtt_offset = i915_gem_obj_ggtt_offset(engine->scratch.obj);
-	engine->scratch.cpu_page = kmap(sg_page(engine->scratch.obj->pages->sgl));
-	if (engine->scratch.cpu_page == NULL) {
-		ret = -ENOMEM;
-		goto err_unpin;
-	}
-
+	engine->scratch.obj = obj;
+	engine->scratch.gtt_offset = i915_gem_obj_ggtt_offset(obj);
 	DRM_DEBUG_DRIVER("%s pipe control offset: 0x%08x\n",
 			 engine->name, engine->scratch.gtt_offset);
 	return 0;
 
-err_unpin:
-	i915_gem_object_ggtt_unpin(engine->scratch.obj);
 err_unref:
 	drm_gem_object_unreference(&engine->scratch.obj->base);
 err:
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 76e1efd87533..2ce679f8eaae 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -338,7 +338,6 @@ struct intel_engine_cs {
 	struct {
 		struct drm_i915_gem_object *obj;
 		u32 gtt_offset;
-		volatile u32 *cpu_page;
 	} scratch;
 
 	bool needs_cmd_parser;
-- 
2.8.1

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

[PATCH 14/25] drm/i915: Allocate scratch page from stolen

[Chris Wilson]

With the last direct CPU access to the scratch page removed, we can now
allocate it from our small amount of reserved system pages (stolen
memory).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/intel_ringbuffer.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 8474b7fea710..4de89ae90d9e 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -660,7 +660,9 @@ int intel_init_pipe_control(struct intel_engine_cs *engine)
 
 	WARN_ON(engine->scratch.obj);
 
-	obj = i915_gem_object_create(engine->i915->dev, 4096);
+	obj = i915_gem_object_create_stolen(engine->i915->dev, 4096);
+	if (obj == NULL)
+		obj = i915_gem_object_create(engine->i915->dev, 4096);
 	if (IS_ERR(obj)) {
 		DRM_ERROR("Failed to allocate scratch page\n");
 		ret = PTR_ERR(obj);
-- 
2.8.1

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

[PATCH 15/25] drm/i915: Refactor scratch object allocation for gen2 w/a buffer

[Chris Wilson]

The gen2 w/a buffer is stuffed into the same slot as the gen5+ scratch
buffer. If we pass in the size we want to allocate for the scratch
buffer, both callers can use the same routine.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/intel_lrc.c        |  2 +-
 drivers/gpu/drm/i915/intel_ringbuffer.c | 31 ++++++++-----------------------
 drivers/gpu/drm/i915/intel_ringbuffer.h |  2 +-
 3 files changed, 10 insertions(+), 25 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 97dc2ea59632..526c71975f11 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -2053,7 +2053,7 @@ static int logical_render_ring_init(struct intel_engine_cs *engine)
 	engine->emit_flush = gen8_emit_flush_render;
 	engine->emit_request = gen8_emit_request_render;
 
-	ret = intel_init_pipe_control(engine);
+	ret = intel_init_pipe_control(engine, 4096);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 4de89ae90d9e..695241dc3b38 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -653,16 +653,16 @@ void intel_fini_pipe_control(struct intel_engine_cs *engine)
 	engine->scratch.obj = NULL;
 }
 
-int intel_init_pipe_control(struct intel_engine_cs *engine)
+int intel_init_pipe_control(struct intel_engine_cs *engine, int size)
 {
 	struct drm_i915_gem_object *obj;
 	int ret;
 
 	WARN_ON(engine->scratch.obj);
 
-	obj = i915_gem_object_create_stolen(engine->i915->dev, 4096);
+	obj = i915_gem_object_create_stolen(engine->i915->dev, size);
 	if (obj == NULL)
-		obj = i915_gem_object_create(engine->i915->dev, 4096);
+		obj = i915_gem_object_create(engine->i915->dev, size);
 	if (IS_ERR(obj)) {
 		DRM_ERROR("Failed to allocate scratch page\n");
 		ret = PTR_ERR(obj);
@@ -2953,31 +2953,16 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 	engine->init_hw = init_render_ring;
 	engine->cleanup = render_ring_cleanup;
 
-	/* Workaround batchbuffer to combat CS tlb bug. */
-	if (HAS_BROKEN_CS_TLB(dev_priv)) {
-		obj = i915_gem_object_create(dev, I830_WA_SIZE);
-		if (IS_ERR(obj)) {
-			DRM_ERROR("Failed to allocate batch bo\n");
-			return PTR_ERR(obj);
-		}
-
-		ret = i915_gem_obj_ggtt_pin(obj, 0, 0);
-		if (ret != 0) {
-			drm_gem_object_unreference(&obj->base);
-			DRM_ERROR("Failed to ping batch bo\n");
-			return ret;
-		}
-
-		engine->scratch.obj = obj;
-		engine->scratch.gtt_offset = i915_gem_obj_ggtt_offset(obj);
-	}
-
 	ret = intel_init_ring_buffer(dev, engine);
 	if (ret)
 		return ret;
 
 	if (INTEL_GEN(dev_priv) >= 5) {
-		ret = intel_init_pipe_control(engine);
+		ret = intel_init_pipe_control(engine, 4096);
+		if (ret)
+			return ret;
+	} else if (HAS_BROKEN_CS_TLB(dev_priv)) {
+		ret = intel_init_pipe_control(engine, I830_WA_SIZE);
 		if (ret)
 			return ret;
 	}
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 2ce679f8eaae..7e53e4d66b6c 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -483,8 +483,8 @@ void intel_ring_init_seqno(struct intel_engine_cs *engine, u32 seqno);
 int intel_ring_flush_all_caches(struct drm_i915_gem_request *req);
 int intel_ring_invalidate_all_caches(struct drm_i915_gem_request *req);
 
+int intel_init_pipe_control(struct intel_engine_cs *engine, int size);
 void intel_fini_pipe_control(struct intel_engine_cs *engine);
-int intel_init_pipe_control(struct intel_engine_cs *engine);
 
 int intel_init_render_ring_buffer(struct drm_device *dev);
 int intel_init_bsd_ring_buffer(struct drm_device *dev);
-- 
2.8.1

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

[PATCH 16/25] drm/i915: Add a delay between interrupt and inspecting the final seqno (ilk)

[Chris Wilson]

On Ironlake, there is no command nor register to ensure that the write
from a MI_STORE command is completed (and coherent on the CPU) before the
command parser continues. This means that the ordering between the seqno
write and the subsequent user interrupt is undefined (like gen6+). So to
ensure that the seqno write is completed after the final user interrupt
we need to delay the read sufficiently to allow the write to complete.
This delay is undefined by the bspec, and empirically requires 75us even
though a register read combined with a clflush is less than 500ns. Hence,
the delay is due to an on-chip buffer rather than the latency of the write
to memory.

Note that the render ring controls this by filling the PIPE_CONTROL fifo
with stalling commands that force the earliest pipe-control with the
seqno to be completed before the command parser continues. Given that we
need a barrier operation for BSD, we may as well forgo the extra
per-batch latency by using a common per-interrupt barrier.

Studying the impact of adding the usleep shows that in both sequences of
and individual synchronous no-op batches is negligible for the media
engine (where the write now is unordered with the interrupt). Converting
the render engine over from the current glutton of pie-controls over to
the per-interrupt delays speeds up both the sequential and individual
synchronous no-ops by 20% and 60%, respectively. This speed up holds
even when looking at the throughput of small copies (4KiB->4MiB), both
serial and synchronous, by about 20%. This is because despite adding a
significant delay to the interrupt, in all likelihood we will see the
seqno write without having to apply the barrier (only in the rare corner
cases where the write is delayed on the last required is the delay
necessary).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94307
Testcase: igt/gem_sync #ilk
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_irq.c         | 10 ++---
 drivers/gpu/drm/i915/intel_ringbuffer.c | 80 ++++++++-------------------------
 2 files changed, 21 insertions(+), 69 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 7c379afcff2f..be7f0b9b27e0 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -1264,8 +1264,7 @@ static void ivybridge_parity_error_irq_handler(struct drm_i915_private *dev_priv
 static void ilk_gt_irq_handler(struct drm_i915_private *dev_priv,
 			       u32 gt_iir)
 {
-	if (gt_iir &
-	    (GT_RENDER_USER_INTERRUPT | GT_RENDER_PIPECTL_NOTIFY_INTERRUPT))
+	if (gt_iir & GT_RENDER_USER_INTERRUPT)
 		notify_ring(&dev_priv->engine[RCS]);
 	if (gt_iir & ILK_BSD_USER_INTERRUPT)
 		notify_ring(&dev_priv->engine[VCS]);
@@ -1274,9 +1273,7 @@ static void ilk_gt_irq_handler(struct drm_i915_private *dev_priv,
 static void snb_gt_irq_handler(struct drm_i915_private *dev_priv,
 			       u32 gt_iir)
 {
-
-	if (gt_iir &
-	    (GT_RENDER_USER_INTERRUPT | GT_RENDER_PIPECTL_NOTIFY_INTERRUPT))
+	if (gt_iir & GT_RENDER_USER_INTERRUPT)
 		notify_ring(&dev_priv->engine[RCS]);
 	if (gt_iir & GT_BSD_USER_INTERRUPT)
 		notify_ring(&dev_priv->engine[VCS]);
@@ -3601,8 +3598,7 @@ static void gen5_gt_irq_postinstall(struct drm_device *dev)
 
 	gt_irqs |= GT_RENDER_USER_INTERRUPT;
 	if (IS_GEN5(dev)) {
-		gt_irqs |= GT_RENDER_PIPECTL_NOTIFY_INTERRUPT |
-			   ILK_BSD_USER_INTERRUPT;
+		gt_irqs |= ILK_BSD_USER_INTERRUPT;
 	} else {
 		gt_irqs |= GT_BLT_USER_INTERRUPT | GT_BSD_USER_INTERRUPT;
 	}
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 695241dc3b38..ad9b54071159 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -1589,67 +1589,22 @@ gen6_ring_sync(struct drm_i915_gem_request *waiter_req,
 	return 0;
 }
 
-#define PIPE_CONTROL_FLUSH(ring__, addr__)					\
-do {									\
-	intel_ring_emit(ring__, GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE |		\
-		 PIPE_CONTROL_DEPTH_STALL);				\
-	intel_ring_emit(ring__, (addr__) | PIPE_CONTROL_GLOBAL_GTT);			\
-	intel_ring_emit(ring__, 0);							\
-	intel_ring_emit(ring__, 0);							\
-} while (0)
-
-static int
-pc_render_add_request(struct drm_i915_gem_request *req)
+static void
+gen5_seqno_barrier(struct intel_engine_cs *ring)
 {
-	struct intel_engine_cs *engine = req->engine;
-	u32 addr = engine->status_page.gfx_addr +
-		(I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
-	u32 scratch_addr = addr;
-	int ret;
-
-	/* For Ironlake, MI_USER_INTERRUPT was deprecated and apparently
-	 * incoherent with writes to memory, i.e. completely fubar,
-	 * so we need to use PIPE_NOTIFY instead.
+	/* MI_STORE are internally buffered by the GPU and not flushed
+	 * either by MI_FLUSH or SyncFlush or any other combination of
+	 * MI commands.
 	 *
-	 * However, we also need to workaround the qword write
-	 * incoherence by flushing the 6 PIPE_NOTIFY buffers out to
-	 * memory before requesting an interrupt.
+	 * "Only the submission of the store operation is guaranteed.
+	 * The write result will be complete (coherent) some time later
+	 * (this is practically a finite period but there is no guaranteed
+	 * latency)."
+	 *
+	 * Empirically, we observe that we need a delay of at least 75us to
+	 * be sure that the seqno write is visible by the CPU.
 	 */
-	ret = intel_ring_begin(req, 32);
-	if (ret)
-		return ret;
-
-	intel_ring_emit(engine,
-			GFX_OP_PIPE_CONTROL(4) |
-			PIPE_CONTROL_QW_WRITE |
-			PIPE_CONTROL_WRITE_FLUSH |
-			PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE);
-	intel_ring_emit(engine, addr | PIPE_CONTROL_GLOBAL_GTT);
-	intel_ring_emit(engine, req->seqno);
-	intel_ring_emit(engine, 0);
-	PIPE_CONTROL_FLUSH(engine, scratch_addr);
-	scratch_addr += 2 * CACHELINE_BYTES; /* write to separate cachelines */
-	PIPE_CONTROL_FLUSH(engine, scratch_addr);
-	scratch_addr += 2 * CACHELINE_BYTES;
-	PIPE_CONTROL_FLUSH(engine, scratch_addr);
-	scratch_addr += 2 * CACHELINE_BYTES;
-	PIPE_CONTROL_FLUSH(engine, scratch_addr);
-	scratch_addr += 2 * CACHELINE_BYTES;
-	PIPE_CONTROL_FLUSH(engine, scratch_addr);
-	scratch_addr += 2 * CACHELINE_BYTES;
-	PIPE_CONTROL_FLUSH(engine, scratch_addr);
-
-	intel_ring_emit(engine,
-		       	GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE |
-			PIPE_CONTROL_WRITE_FLUSH |
-			PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE |
-			PIPE_CONTROL_NOTIFY);
-	intel_ring_emit(engine, addr | PIPE_CONTROL_GLOBAL_GTT);
-	intel_ring_emit(engine, req->seqno);
-	intel_ring_emit(engine, 0);
-	__intel_ring_advance(engine);
-
-	return 0;
+	usleep_range(75, 250);
 }
 
 static void
@@ -2915,12 +2870,12 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 			engine->semaphore.mbox.signal[VCS2] = GEN6_NOSYNC;
 		}
 	} else if (IS_GEN5(dev_priv)) {
-		engine->add_request = pc_render_add_request;
+		engine->add_request = i9xx_add_request;
 		engine->flush = gen4_render_ring_flush;
 		engine->irq_get = gen5_ring_get_irq;
 		engine->irq_put = gen5_ring_put_irq;
-		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT |
-					GT_RENDER_PIPECTL_NOTIFY_INTERRUPT;
+		engine->irq_seqno_barrier = gen5_seqno_barrier;
+		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT;
 	} else {
 		engine->add_request = i9xx_add_request;
 		if (INTEL_GEN(dev_priv) < 4)
@@ -2957,7 +2912,7 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 	if (ret)
 		return ret;
 
-	if (INTEL_GEN(dev_priv) >= 5) {
+	if (INTEL_GEN(dev_priv) >= 6) {
 		ret = intel_init_pipe_control(engine, 4096);
 		if (ret)
 			return ret;
@@ -3030,6 +2985,7 @@ int intel_init_bsd_ring_buffer(struct drm_device *dev)
 			engine->irq_enable_mask = ILK_BSD_USER_INTERRUPT;
 			engine->irq_get = gen5_ring_get_irq;
 			engine->irq_put = gen5_ring_put_irq;
+			engine->irq_seqno_barrier = gen5_seqno_barrier;
 		} else {
 			engine->irq_enable_mask = I915_BSD_USER_INTERRUPT;
 			engine->irq_get = i9xx_ring_get_irq;
-- 
2.8.1

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

[PATCH 17/25] drm/i915: Check the CPU cached value in HWS of seqno after waking the waiter

[Chris Wilson]

If we have multiple waiters, we may find that many complete on the same
wake up. If we first inspect the seqno from the CPU cache, we may reduce
the number of heavyweight coherent seqno reads we require.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_drv.h | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index f4b883a58b68..9a3890f95fb1 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3977,6 +3977,12 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
 {
 	struct intel_engine_cs *engine = req->engine;
 
+	/* Before we do the heavier coherent read of the seqno,
+	 * check the value (hopefully) in the CPU cacheline.
+	 */
+	if (i915_gem_request_completed(req))
+		return true;
+
 	/* Ensure our read of the seqno is coherent so that we
 	 * do not "miss an interrupt" (i.e. if this is the last
 	 * request and the seqno write from the GPU is not visible
@@ -3988,11 +3994,11 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
 	 * but it is easier and safer to do it every time the waiter
 	 * is woken.
 	 */
-	if (engine->irq_seqno_barrier)
+	if (engine->irq_seqno_barrier) {
 		engine->irq_seqno_barrier(engine);
-
-	if (i915_gem_request_completed(req))
-		return true;
+		if (i915_gem_request_completed(req))
+			return true;
+	}
 
 	/* We need to check whether any gpu reset happened in between
 	 * the request being submitted and now. If a reset has occurred,
-- 
2.8.1

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

[PATCH 18/25] drm/i915: Only apply one barrier after a breadcrumb interrupt is posted

[Chris Wilson]

If we flag the seqno as potentially stale upon receiving an interrupt,
we can use that information to reduce the frequency that we apply the
heavyweight coherent seqno read (i.e. if we wake up a chain of waiters).

v2: Use cmpxchg to replace READ_ONCE/WRITE_ONCE for more explicit
control of the ordering wrt to interrupt generation and interrupt
checking in the bottom-half.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h          | 15 ++++++++++++++-
 drivers/gpu/drm/i915/i915_irq.c          |  1 +
 drivers/gpu/drm/i915/intel_breadcrumbs.c | 16 ++++++++++------
 drivers/gpu/drm/i915/intel_ringbuffer.h  |  1 +
 4 files changed, 26 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 9a3890f95fb1..d09b96d193a5 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3994,7 +3994,20 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
 	 * but it is easier and safer to do it every time the waiter
 	 * is woken.
 	 */
-	if (engine->irq_seqno_barrier) {
+	if (engine->irq_seqno_barrier &&
+	    cmpxchg_relaxed(&engine->irq_posted, 1, 0)) {
+		/* The ordering of irq_posted versus applying the barrier
+		 * is crucial. The clearing of the current irq_posted must
+		 * be visible before we perform the barrier operation,
+		 * such that if a subsequent interrupt arrives, irq_posted
+		 * is reasserted and our task rewoken (which causes us to
+		 * do another __i915_request_irq_complete() immediately
+		 * and reapply the barrier). Conversely, if the clear
+		 * occurs after the barrier, then an interrupt that arrived
+		 * whilst we waited on the barrier would not trigger a
+		 * barrier on the next pass, and the read may not see the
+		 * seqno update.
+		 */
 		engine->irq_seqno_barrier(engine);
 		if (i915_gem_request_completed(req))
 			return true;
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index be7f0b9b27e0..7724bae27bcf 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -976,6 +976,7 @@ static void ironlake_rps_change_irq_handler(struct drm_i915_private *dev_priv)
 
 static void notify_ring(struct intel_engine_cs *engine)
 {
+	smp_store_mb(engine->irq_posted, true);
 	if (intel_engine_wakeup(engine)) {
 		trace_i915_gem_request_notify(engine);
 		engine->user_interrupts++;
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index cb5e85192669..84c2a449dd0e 100644
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -43,12 +43,18 @@ static void intel_breadcrumbs_fake_irq(unsigned long data)
 
 static void irq_enable(struct intel_engine_cs *engine)
 {
+	/* Enabling the IRQ may miss the generation of the interrupt, but
+	 * we still need to force the barrier before reading the seqno,
+	 * just in case.
+	 */
+	engine->irq_posted = true;
 	WARN_ON(!engine->irq_get(engine));
 }
 
 static void irq_disable(struct intel_engine_cs *engine)
 {
 	engine->irq_put(engine);
+	engine->irq_posted = false;
 }
 
 static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
@@ -56,7 +62,6 @@ static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
 	struct intel_engine_cs *engine =
 		container_of(b, struct intel_engine_cs, breadcrumbs);
 	struct drm_i915_private *i915 = engine->i915;
-	bool irq_posted = false;
 
 	assert_spin_locked(&b->lock);
 	if (b->rpm_wakelock)
@@ -72,10 +77,8 @@ static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
 
 	/* No interrupts? Kick the waiter every jiffie! */
 	if (intel_irqs_enabled(i915)) {
-		if (!test_bit(engine->id, &i915->gpu_error.test_irq_rings)) {
+		if (!test_bit(engine->id, &i915->gpu_error.test_irq_rings))
 			irq_enable(engine);
-			irq_posted = true;
-		}
 		b->irq_enabled = true;
 	}
 
@@ -83,7 +86,7 @@ static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
 	    test_bit(engine->id, &i915->gpu_error.missed_irq_rings))
 		mod_timer(&b->fake_irq, jiffies + 1);
 
-	return irq_posted;
+	return engine->irq_posted;
 }
 
 static void __intel_breadcrumbs_disable_irq(struct intel_breadcrumbs *b)
@@ -205,7 +208,8 @@ static bool __intel_engine_add_wait(struct intel_engine_cs *engine,
 			 * in case the seqno passed.
 			 */
 			__intel_breadcrumbs_enable_irq(b);
-			wake_up_process(to_wait(next)->tsk);
+			if (READ_ONCE(engine->irq_posted))
+				wake_up_process(to_wait(next)->tsk);
 		}
 
 		do {
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 7e53e4d66b6c..4d6c2b773a64 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -197,6 +197,7 @@ struct intel_engine_cs {
 	struct i915_ctx_workarounds wa_ctx;
 
 	unsigned irq_refcount; /* protected by dev_priv->irq_lock */
+	bool		irq_posted;
 	u32		irq_enable_mask;	/* bitmask to enable ring interrupt */
 	struct drm_i915_gem_request *trace_irq_req;
 	bool __must_check (*irq_get)(struct intel_engine_cs *ring);
-- 
2.8.1

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

Re: [PATCH 18/25] drm/i915: Only apply one barrier after a breadcrumb interrupt is posted

[Tvrtko Ursulin]

On 25/06/16 11:13, Chris Wilson wrote:
> If we flag the seqno as potentially stale upon receiving an interrupt,
> we can use that information to reduce the frequency that we apply the
> heavyweight coherent seqno read (i.e. if we wake up a chain of waiters).
>
> v2: Use cmpxchg to replace READ_ONCE/WRITE_ONCE for more explicit
> control of the ordering wrt to interrupt generation and interrupt
> checking in the bottom-half.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_drv.h          | 15 ++++++++++++++-
>   drivers/gpu/drm/i915/i915_irq.c          |  1 +
>   drivers/gpu/drm/i915/intel_breadcrumbs.c | 16 ++++++++++------
>   drivers/gpu/drm/i915/intel_ringbuffer.h  |  1 +
>   4 files changed, 26 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 9a3890f95fb1..d09b96d193a5 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -3994,7 +3994,20 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
>   	 * but it is easier and safer to do it every time the waiter
>   	 * is woken.
>   	 */
> -	if (engine->irq_seqno_barrier) {
> +	if (engine->irq_seqno_barrier &&
> +	    cmpxchg_relaxed(&engine->irq_posted, 1, 0)) {
> +		/* The ordering of irq_posted versus applying the barrier
> +		 * is crucial. The clearing of the current irq_posted must
> +		 * be visible before we perform the barrier operation,
> +		 * such that if a subsequent interrupt arrives, irq_posted
> +		 * is reasserted and our task rewoken (which causes us to
> +		 * do another __i915_request_irq_complete() immediately
> +		 * and reapply the barrier). Conversely, if the clear
> +		 * occurs after the barrier, then an interrupt that arrived
> +		 * whilst we waited on the barrier would not trigger a
> +		 * barrier on the next pass, and the read may not see the
> +		 * seqno update.
> +		 */
>   		engine->irq_seqno_barrier(engine);
>   		if (i915_gem_request_completed(req))
>   			return true;
> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> index be7f0b9b27e0..7724bae27bcf 100644
> --- a/drivers/gpu/drm/i915/i915_irq.c
> +++ b/drivers/gpu/drm/i915/i915_irq.c
> @@ -976,6 +976,7 @@ static void ironlake_rps_change_irq_handler(struct drm_i915_private *dev_priv)
>
>   static void notify_ring(struct intel_engine_cs *engine)
>   {
> +	smp_store_mb(engine->irq_posted, true);
>   	if (intel_engine_wakeup(engine)) {
>   		trace_i915_gem_request_notify(engine);
>   		engine->user_interrupts++;
> diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> index cb5e85192669..84c2a449dd0e 100644
> --- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
> +++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> @@ -43,12 +43,18 @@ static void intel_breadcrumbs_fake_irq(unsigned long data)
>
>   static void irq_enable(struct intel_engine_cs *engine)
>   {
> +	/* Enabling the IRQ may miss the generation of the interrupt, but
> +	 * we still need to force the barrier before reading the seqno,
> +	 * just in case.
> +	 */
> +	engine->irq_posted = true;
>   	WARN_ON(!engine->irq_get(engine));
>   }
>
>   static void irq_disable(struct intel_engine_cs *engine)
>   {
>   	engine->irq_put(engine);
> +	engine->irq_posted = false;
>   }
>
>   static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
> @@ -56,7 +62,6 @@ static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
>   	struct intel_engine_cs *engine =
>   		container_of(b, struct intel_engine_cs, breadcrumbs);
>   	struct drm_i915_private *i915 = engine->i915;
> -	bool irq_posted = false;
>
>   	assert_spin_locked(&b->lock);
>   	if (b->rpm_wakelock)
> @@ -72,10 +77,8 @@ static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
>
>   	/* No interrupts? Kick the waiter every jiffie! */
>   	if (intel_irqs_enabled(i915)) {
> -		if (!test_bit(engine->id, &i915->gpu_error.test_irq_rings)) {
> +		if (!test_bit(engine->id, &i915->gpu_error.test_irq_rings))
>   			irq_enable(engine);
> -			irq_posted = true;
> -		}
>   		b->irq_enabled = true;
>   	}
>
> @@ -83,7 +86,7 @@ static bool __intel_breadcrumbs_enable_irq(struct intel_breadcrumbs *b)
>   	    test_bit(engine->id, &i915->gpu_error.missed_irq_rings))
>   		mod_timer(&b->fake_irq, jiffies + 1);
>
> -	return irq_posted;
> +	return engine->irq_posted;
>   }
>
>   static void __intel_breadcrumbs_disable_irq(struct intel_breadcrumbs *b)
> @@ -205,7 +208,8 @@ static bool __intel_engine_add_wait(struct intel_engine_cs *engine,
>   			 * in case the seqno passed.
>   			 */
>   			__intel_breadcrumbs_enable_irq(b);
> -			wake_up_process(to_wait(next)->tsk);
> +			if (READ_ONCE(engine->irq_posted))
> +				wake_up_process(to_wait(next)->tsk);
>   		}
>
>   		do {
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index 7e53e4d66b6c..4d6c2b773a64 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -197,6 +197,7 @@ struct intel_engine_cs {
>   	struct i915_ctx_workarounds wa_ctx;
>
>   	unsigned irq_refcount; /* protected by dev_priv->irq_lock */
> +	bool		irq_posted;
>   	u32		irq_enable_mask;	/* bitmask to enable ring interrupt */
>   	struct drm_i915_gem_request *trace_irq_req;
>   	bool __must_check (*irq_get)(struct intel_engine_cs *ring);
>

AFAIR this looked OK to me last time round apart from the a little bit 
of unclarity in __i915_request_irq_complete which you resolved with cmpxchg.

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

[PATCH 19/25] drm/i915: Stop setting wraparound seqno on initialisation

[Chris Wilson]

We have testcases to ensure that seqno wraparound works fine, so we can
forgo forcing everyone to encounter seqno wraparound during early
uptime. seqno wraparound incurs a full GPU stall so not forcing it
will eliminate one jitter from the early system. Using the testcases, we
have very deterministic testing which given how difficult it would be to
debug an issue (GPU hang) stemming from a wraparound using pure
postmortem analysis I see no value in forcing a wrap during boot.

Advancing the global next_seqno after a GPU reset is equally pointless.

References? https://bugs.freedesktop.org/show_bug.cgi?id=95023
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
---
 drivers/gpu/drm/i915/i915_gem.c | 14 --------------
 1 file changed, 14 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 30bd0d0d917b..486636021009 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -5114,12 +5114,6 @@ i915_gem_init_hw(struct drm_device *dev)
 	if (ret)
 		goto out;
 
-	/*
-	 * Increment the next seqno by 0x100 so we have a visible break
-	 * on re-initialisation
-	 */
-	ret = i915_gem_set_seqno(dev, dev_priv->next_seqno+0x100);
-
 out:
 	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
 	return ret;
@@ -5262,14 +5256,6 @@ i915_gem_load_init(struct drm_device *dev)
 
 	dev_priv->relative_constants_mode = I915_EXEC_CONSTANTS_REL_GENERAL;
 
-	/*
-	 * Set initial sequence number for requests.
-	 * Using this number allows the wraparound to happen early,
-	 * catching any obvious problems.
-	 */
-	dev_priv->next_seqno = ((u32)~0 - 0x1100);
-	dev_priv->last_seqno = ((u32)~0 - 0x1101);
-
 	INIT_LIST_HEAD(&dev_priv->mm.fence_list);
 
 	init_waitqueue_head(&dev_priv->pending_flip_queue);
-- 
2.8.1

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

[PATCH 20/25] drm/i915: Only query timestamp when measuring elapsed time

[Chris Wilson]

Avoid the two calls to ktime_get_raw_ns() (at best it reads the TSC) as
we only need to compute the elapsed time for a timed wait.

v2: Eliminate the unused local variable reducing the function size by 64
bytes (using the storage space on the callers stack rather than adding
to our stack frame). Writing the code this emits smaller and faster code
for the normal case.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c | 14 +++++---------
 1 file changed, 5 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 486636021009..fd59b30a024d 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1433,7 +1433,6 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 	DEFINE_WAIT(reset);
 	struct intel_wait wait;
 	unsigned long timeout_remain;
-	s64 before = 0; /* Only to silence a compiler warning. */
 	int ret = 0;
 
 	might_sleep();
@@ -1452,12 +1451,9 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
 		if (*timeout == 0)
 			return -ETIME;
 
+		/* Record current time in case interrupted, or wedged */
 		timeout_remain = nsecs_to_jiffies_timeout(*timeout);
-
-		/*
-		 * Record current time in case interrupted by signal, or wedged.
-		 */
-		before = ktime_get_raw_ns();
+		*timeout += ktime_get_raw_ns();
 	}
 
 	trace_i915_gem_request_wait_begin(req);
@@ -1521,9 +1517,9 @@ complete:
 	trace_i915_gem_request_wait_end(req);
 
 	if (timeout) {
-		s64 tres = *timeout - (ktime_get_raw_ns() - before);
-
-		*timeout = tres < 0 ? 0 : tres;
+		*timeout -= ktime_get_raw_ns();
+		if (*timeout < 0)
+			*timeout = 0;
 
 		/*
 		 * Apparently ktime isn't accurate enough and occasionally has a
-- 
2.8.1

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

Re: [PATCH 20/25] drm/i915: Only query timestamp when measuring elapsed time

[Tvrtko Ursulin]

On 25/06/16 11:13, Chris Wilson wrote:
> Avoid the two calls to ktime_get_raw_ns() (at best it reads the TSC) as
> we only need to compute the elapsed time for a timed wait.
>
> v2: Eliminate the unused local variable reducing the function size by 64
> bytes (using the storage space on the callers stack rather than adding
> to our stack frame). Writing the code this emits smaller and faster code
> for the normal case.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_gem.c | 14 +++++---------
>   1 file changed, 5 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 486636021009..fd59b30a024d 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -1433,7 +1433,6 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
>   	DEFINE_WAIT(reset);
>   	struct intel_wait wait;
>   	unsigned long timeout_remain;
> -	s64 before = 0; /* Only to silence a compiler warning. */
>   	int ret = 0;
>
>   	might_sleep();
> @@ -1452,12 +1451,9 @@ int __i915_wait_request(struct drm_i915_gem_request *req,
>   		if (*timeout == 0)
>   			return -ETIME;
>
> +		/* Record current time in case interrupted, or wedged */
>   		timeout_remain = nsecs_to_jiffies_timeout(*timeout);
> -
> -		/*
> -		 * Record current time in case interrupted by signal, or wedged.
> -		 */
> -		before = ktime_get_raw_ns();
> +		*timeout += ktime_get_raw_ns();
>   	}
>
>   	trace_i915_gem_request_wait_begin(req);
> @@ -1521,9 +1517,9 @@ complete:
>   	trace_i915_gem_request_wait_end(req);
>
>   	if (timeout) {
> -		s64 tres = *timeout - (ktime_get_raw_ns() - before);
> -
> -		*timeout = tres < 0 ? 0 : tres;
> +		*timeout -= ktime_get_raw_ns();
> +		if (*timeout < 0)
> +			*timeout = 0;
>
>   		/*
>   		 * Apparently ktime isn't accurate enough and occasionally has a
>

Still a no from me.

Regards,

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

[PATCH 21/25] drm/i915: Convert trace-irq to the breadcrumb waiter

[Chris Wilson]

If we convert the tracing over from direct use of ring->irq_get() and
over to the breadcrumb infrastructure, we only have a single user of the
ring->irq_get and so we will be able to simplify the driver routines
(eliminating the redundant validation and irq refcounting).

Process context is preferred over softirq (or even hardirq) for a couple
of reasons:

 - we already utilize process context to have fast wakeup of a single
   client (i.e. the client waiting for the GPU inspects the seqno for
   itself following an interrupt to avoid the overhead of a context
   switch before it returns to userspace)

 - engine->irq_seqno() is not suitable for use from an softirq/hardirq
   context as we may require long waits (100-250us) to ensure the seqno
   write is posted before we read it from the CPU

A signaling framework is a requirement for enabling dma-fences.

v2: Move to a signaling framework based upon the waiter.
v3: Track the first-signal to avoid having to walk the rbtree everytime.
v4: Mark the signaler thread as RT priority to reduce latency in the
indirect wakeups.
v5: Make failure to allocate the thread fatal.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h          |   8 --
 drivers/gpu/drm/i915/i915_gem.c          |   9 +-
 drivers/gpu/drm/i915/i915_trace.h        |   2 +-
 drivers/gpu/drm/i915/intel_breadcrumbs.c | 192 ++++++++++++++++++++++++++++++-
 drivers/gpu/drm/i915/intel_ringbuffer.h  |   8 +-
 5 files changed, 201 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index d09b96d193a5..b7089cedb80c 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3965,14 +3965,6 @@ wait_remaining_ms_from_jiffies(unsigned long timestamp_jiffies, int to_wait_ms)
 			    schedule_timeout_uninterruptible(remaining_jiffies);
 	}
 }
-
-static inline void i915_trace_irq_get(struct intel_engine_cs *engine,
-				      struct drm_i915_gem_request *req)
-{
-	if (engine->trace_irq_req == NULL && engine->irq_get(engine))
-		i915_gem_request_assign(&engine->trace_irq_req, req);
-}
-
 static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
 {
 	struct intel_engine_cs *engine = req->engine;
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index fd59b30a024d..e6daa1ee4d2d 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2745,7 +2745,8 @@ i915_gem_init_seqno(struct drm_i915_private *dev_priv, u32 seqno)
 
 	/* If the seqno wraps around, we need to clear the breadcrumb rbtree */
 	if (!i915_seqno_passed(seqno, dev_priv->next_seqno)) {
-		while (intel_kick_waiters(dev_priv))
+		while (intel_kick_waiters(dev_priv) ||
+		       intel_kick_signalers(dev_priv))
 			yield();
 	}
 
@@ -3210,12 +3211,6 @@ i915_gem_retire_requests_ring(struct intel_engine_cs *engine)
 		i915_gem_object_retire__read(obj, engine->id);
 	}
 
-	if (unlikely(engine->trace_irq_req &&
-		     i915_gem_request_completed(engine->trace_irq_req))) {
-		engine->irq_put(engine);
-		i915_gem_request_assign(&engine->trace_irq_req, NULL);
-	}
-
 	WARN_ON(i915_verify_lists(engine->dev));
 }
 
diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h
index 3d13fde95fdf..f59cf07184ae 100644
--- a/drivers/gpu/drm/i915/i915_trace.h
+++ b/drivers/gpu/drm/i915/i915_trace.h
@@ -490,7 +490,7 @@ TRACE_EVENT(i915_gem_ring_dispatch,
 			   __entry->ring = req->engine->id;
 			   __entry->seqno = req->seqno;
 			   __entry->flags = flags;
-			   i915_trace_irq_get(req->engine, req);
+			   intel_engine_enable_signaling(req);
 			   ),
 
 	    TP_printk("dev=%u, ring=%u, seqno=%u, flags=%x",
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index 84c2a449dd0e..bc0e9e8b7871 100644
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -22,6 +22,8 @@
  *
  */
 
+#include <linux/kthread.h>
+
 #include "i915_drv.h"
 
 static void intel_breadcrumbs_fake_irq(unsigned long data)
@@ -255,6 +257,15 @@ static inline bool chain_wakeup(struct rb_node *rb, int priority)
 	return rb && to_wait(rb)->tsk->prio <= priority;
 }
 
+static inline int wakeup_priority(struct intel_breadcrumbs *b,
+				  struct task_struct *tsk)
+{
+	if (tsk == b->signaler)
+		return INT_MIN;
+	else
+		return tsk->prio;
+}
+
 void intel_engine_remove_wait(struct intel_engine_cs *engine,
 			      struct intel_wait *wait)
 {
@@ -273,8 +284,8 @@ void intel_engine_remove_wait(struct intel_engine_cs *engine,
 		goto out_unlock;
 
 	if (b->first_wait == wait) {
+		const int priority = wakeup_priority(b, wait->tsk);
 		struct rb_node *next;
-		const int priority = wait->tsk->prio;
 
 		GEM_BUG_ON(b->tasklet != wait->tsk);
 
@@ -340,15 +351,176 @@ out_unlock:
 	spin_unlock(&b->lock);
 }
 
+struct signal {
+	struct rb_node node;
+	struct intel_wait wait;
+	struct drm_i915_gem_request *request;
+};
+
+static bool signal_complete(struct signal *signal)
+{
+	if (signal == NULL)
+		return false;
+
+	/* If another process served as the bottom-half it may have already
+	 * signalled that this wait is already completed.
+	 */
+	if (intel_wait_complete(&signal->wait))
+		return true;
+
+	/* Carefully check if the request is complete, giving time for the
+	 * seqno to be visible or if the GPU hung.
+	 */
+	if (__i915_request_irq_complete(signal->request))
+		return true;
+
+	return false;
+}
+
+static struct signal *to_signal(struct rb_node *rb)
+{
+	return container_of(rb, struct signal, node);
+}
+
+static void signaler_set_rtpriority(void)
+{
+	 struct sched_param param = { .sched_priority = 1 };
+	 sched_setscheduler_nocheck(current, SCHED_FIFO, &param);
+}
+
+static int intel_breadcrumbs_signaler(void *arg)
+{
+	struct intel_engine_cs *engine = arg;
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	struct signal *signal;
+
+	/* Install ourselves with high priority to reduce signalling latency */
+	signaler_set_rtpriority();
+
+	do {
+		set_current_state(TASK_INTERRUPTIBLE);
+
+		/* We are either woken up by the interrupt bottom-half,
+		 * or by a client adding a new signaller. In both cases,
+		 * the GPU seqno may have advanced beyond our oldest signal.
+		 * If it has, propagate the signal, remove the waiter and
+		 * check again with the next oldest signal. Otherwise we
+		 * need to wait for a new interrupt from the GPU or for
+		 * a new client.
+		 */
+		signal = READ_ONCE(b->first_signal);
+		if (signal_complete(signal)) {
+			/* Wake up all other completed waiters and select the
+			 * next bottom-half for the next user interrupt.
+			 */
+			intel_engine_remove_wait(engine, &signal->wait);
+
+			i915_gem_request_unreference(signal->request);
+
+			/* Find the next oldest signal. Note that as we have
+			 * not been holding the lock, another client may
+			 * have installed an even older signal than the one
+			 * we just completed - so double check we are still
+			 * the oldest before picking the next one.
+			 */
+			spin_lock(&b->lock);
+			if (signal == b->first_signal)
+				b->first_signal = rb_next(&signal->node);
+			rb_erase(&signal->node, &b->signals);
+			spin_unlock(&b->lock);
+
+			kfree(signal);
+		} else {
+			if (kthread_should_stop())
+				break;
+
+			schedule();
+		}
+	} while (1);
+	__set_current_state(TASK_RUNNING);
+
+	return 0;
+}
+
+int intel_engine_enable_signaling(struct drm_i915_gem_request *request)
+{
+	struct intel_engine_cs *engine = request->engine;
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	struct rb_node *parent, **p;
+	struct signal *signal;
+	bool first, wakeup;
+
+	signal = kmalloc(sizeof(*signal), GFP_ATOMIC);
+	if (unlikely(!signal))
+		return -ENOMEM;
+
+	signal->wait.tsk = b->signaler;
+	signal->wait.seqno = request->seqno;
+
+	signal->request = i915_gem_request_reference(request);
+
+	/* First add ourselves into the list of waiters, but register our
+	 * bottom-half as the signaller thread. As per usual, only the oldest
+	 * waiter (not just signaller) is tasked as the bottom-half waking
+	 * up all completed waiters after the user interrupt.
+	 *
+	 * If we are the oldest waiter, enable the irq (after which we
+	 * must double check that the seqno did not complete).
+	 */
+	wakeup = intel_engine_add_wait(engine, &signal->wait);
+
+	/* Now insert ourselves into the retirement ordered list of signals
+	 * on this engine. We track the oldest seqno as that will be the
+	 * first signal to complete.
+	 */
+	spin_lock(&b->lock);
+	parent = NULL;
+	first = true;
+	p = &b->signals.rb_node;
+	while (*p) {
+		parent = *p;
+		if (i915_seqno_passed(signal->wait.seqno,
+				      to_signal(parent)->wait.seqno)) {
+			p = &parent->rb_right;
+			first = false;
+		} else
+			p = &parent->rb_left;
+	}
+	rb_link_node(&signal->node, parent, p);
+	rb_insert_color(&signal->node, &b->signals);
+	if (first)
+		smp_store_mb(b->first_signal, signal);
+	spin_unlock(&b->lock);
+
+	if (wakeup)
+		wake_up_process(b->signaler);
+
+	return 0;
+}
+
 int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
 {
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	struct task_struct *tsk;
 
 	spin_lock_init(&b->lock);
 	setup_timer(&b->fake_irq,
 		    intel_breadcrumbs_fake_irq,
 		    (unsigned long)engine);
 
+	/* Spawn a thread to provide a common bottom-half for all signals.
+	 * As this is an asynchronous interface we cannot steal the current
+	 * task for handling the bottom-half to the user interrupt, therefore
+	 * we create a thread to do the coherent seqno dance after the
+	 * interrupt and then signal the waitqueue (via the dma-buf/fence).
+	 */
+	tsk = kthread_run(intel_breadcrumbs_signaler, engine,
+			  "irq/i915:%d", engine->id);
+	if (IS_ERR(tsk))
+		return PTR_ERR(tsk);
+
+	b->signaler = tsk;
+
 	return 0;
 }
 
@@ -356,6 +528,9 @@ void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
 {
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
 
+	if (!IS_ERR_OR_NULL(b->signaler))
+		kthread_stop(b->signaler);
+
 	del_timer_sync(&b->fake_irq);
 }
 
@@ -377,3 +552,18 @@ unsigned intel_kick_waiters(struct drm_i915_private *i915)
 
 	return mask;
 }
+
+unsigned intel_kick_signalers(struct drm_i915_private *i915)
+{
+	struct intel_engine_cs *engine;
+	unsigned mask = 0;
+
+	for_each_engine(engine, i915) {
+		if (unlikely(READ_ONCE(engine->breadcrumbs.first_signal))) {
+			wake_up_process(engine->breadcrumbs.signaler);
+			mask |= intel_engine_flag(engine);
+		}
+	}
+
+	return mask;
+}
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 4d6c2b773a64..c05b45727f7d 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -141,6 +141,8 @@ struct  i915_ctx_workarounds {
 	struct drm_i915_gem_object *obj;
 };
 
+struct drm_i915_gem_request;
+
 struct intel_engine_cs {
 	struct drm_i915_private *i915;
 	const char	*name;
@@ -179,8 +181,11 @@ struct intel_engine_cs {
 	struct intel_breadcrumbs {
 		spinlock_t lock; /* protects the lists of requests */
 		struct rb_root waiters; /* sorted by retirement, priority */
+		struct rb_root signals; /* sorted by retirement */
 		struct intel_wait *first_wait; /* oldest waiter by retirement */
 		struct task_struct *tasklet; /* bh for user interrupts */
+		struct task_struct *signaler; /* used for fence signalling */
+		void *first_signal;
 		struct timer_list fake_irq; /* used after a missed interrupt */
 		bool irq_enabled;
 		bool rpm_wakelock;
@@ -199,7 +204,6 @@ struct intel_engine_cs {
 	unsigned irq_refcount; /* protected by dev_priv->irq_lock */
 	bool		irq_posted;
 	u32		irq_enable_mask;	/* bitmask to enable ring interrupt */
-	struct drm_i915_gem_request *trace_irq_req;
 	bool __must_check (*irq_get)(struct intel_engine_cs *ring);
 	void		(*irq_put)(struct intel_engine_cs *ring);
 
@@ -540,6 +544,7 @@ bool intel_engine_add_wait(struct intel_engine_cs *engine,
 			   struct intel_wait *wait);
 void intel_engine_remove_wait(struct intel_engine_cs *engine,
 			      struct intel_wait *wait);
+int intel_engine_enable_signaling(struct drm_i915_gem_request *request);
 static inline bool intel_engine_has_waiter(struct intel_engine_cs *engine)
 {
 	return READ_ONCE(engine->breadcrumbs.tasklet);
@@ -563,5 +568,6 @@ static inline bool intel_engine_wakeup(struct intel_engine_cs *engine)
 void intel_engine_enable_fake_irq(struct intel_engine_cs *engine);
 void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
 unsigned intel_kick_waiters(struct drm_i915_private *i915);
+unsigned intel_kick_signalers(struct drm_i915_private *i915);
 
 #endif /* _INTEL_RINGBUFFER_H_ */
-- 
2.8.1

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

Re: [PATCH 21/25] drm/i915: Convert trace-irq to the breadcrumb waiter

[Tvrtko Ursulin]

On 25/06/16 11:13, Chris Wilson wrote:
> If we convert the tracing over from direct use of ring->irq_get() and
> over to the breadcrumb infrastructure, we only have a single user of the
> ring->irq_get and so we will be able to simplify the driver routines
> (eliminating the redundant validation and irq refcounting).
>
> Process context is preferred over softirq (or even hardirq) for a couple
> of reasons:
>
>   - we already utilize process context to have fast wakeup of a single
>     client (i.e. the client waiting for the GPU inspects the seqno for
>     itself following an interrupt to avoid the overhead of a context
>     switch before it returns to userspace)
>
>   - engine->irq_seqno() is not suitable for use from an softirq/hardirq
>     context as we may require long waits (100-250us) to ensure the seqno
>     write is posted before we read it from the CPU
>
> A signaling framework is a requirement for enabling dma-fences.
>
> v2: Move to a signaling framework based upon the waiter.
> v3: Track the first-signal to avoid having to walk the rbtree everytime.
> v4: Mark the signaler thread as RT priority to reduce latency in the
> indirect wakeups.
> v5: Make failure to allocate the thread fatal.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_drv.h          |   8 --
>   drivers/gpu/drm/i915/i915_gem.c          |   9 +-
>   drivers/gpu/drm/i915/i915_trace.h        |   2 +-
>   drivers/gpu/drm/i915/intel_breadcrumbs.c | 192 ++++++++++++++++++++++++++++++-
>   drivers/gpu/drm/i915/intel_ringbuffer.h  |   8 +-
>   5 files changed, 201 insertions(+), 18 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index d09b96d193a5..b7089cedb80c 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -3965,14 +3965,6 @@ wait_remaining_ms_from_jiffies(unsigned long timestamp_jiffies, int to_wait_ms)
>   			    schedule_timeout_uninterruptible(remaining_jiffies);
>   	}
>   }
> -
> -static inline void i915_trace_irq_get(struct intel_engine_cs *engine,
> -				      struct drm_i915_gem_request *req)
> -{
> -	if (engine->trace_irq_req == NULL && engine->irq_get(engine))
> -		i915_gem_request_assign(&engine->trace_irq_req, req);
> -}
> -
>   static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
>   {
>   	struct intel_engine_cs *engine = req->engine;
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index fd59b30a024d..e6daa1ee4d2d 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -2745,7 +2745,8 @@ i915_gem_init_seqno(struct drm_i915_private *dev_priv, u32 seqno)
>
>   	/* If the seqno wraps around, we need to clear the breadcrumb rbtree */
>   	if (!i915_seqno_passed(seqno, dev_priv->next_seqno)) {
> -		while (intel_kick_waiters(dev_priv))
> +		while (intel_kick_waiters(dev_priv) ||
> +		       intel_kick_signalers(dev_priv))
>   			yield();
>   	}
>
> @@ -3210,12 +3211,6 @@ i915_gem_retire_requests_ring(struct intel_engine_cs *engine)
>   		i915_gem_object_retire__read(obj, engine->id);
>   	}
>
> -	if (unlikely(engine->trace_irq_req &&
> -		     i915_gem_request_completed(engine->trace_irq_req))) {
> -		engine->irq_put(engine);
> -		i915_gem_request_assign(&engine->trace_irq_req, NULL);
> -	}
> -
>   	WARN_ON(i915_verify_lists(engine->dev));
>   }
>
> diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h
> index 3d13fde95fdf..f59cf07184ae 100644
> --- a/drivers/gpu/drm/i915/i915_trace.h
> +++ b/drivers/gpu/drm/i915/i915_trace.h
> @@ -490,7 +490,7 @@ TRACE_EVENT(i915_gem_ring_dispatch,
>   			   __entry->ring = req->engine->id;
>   			   __entry->seqno = req->seqno;
>   			   __entry->flags = flags;
> -			   i915_trace_irq_get(req->engine, req);
> +			   intel_engine_enable_signaling(req);
>   			   ),
>
>   	    TP_printk("dev=%u, ring=%u, seqno=%u, flags=%x",
> diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> index 84c2a449dd0e..bc0e9e8b7871 100644
> --- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
> +++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> @@ -22,6 +22,8 @@
>    *
>    */
>
> +#include <linux/kthread.h>
> +
>   #include "i915_drv.h"
>
>   static void intel_breadcrumbs_fake_irq(unsigned long data)
> @@ -255,6 +257,15 @@ static inline bool chain_wakeup(struct rb_node *rb, int priority)
>   	return rb && to_wait(rb)->tsk->prio <= priority;
>   }
>
> +static inline int wakeup_priority(struct intel_breadcrumbs *b,
> +				  struct task_struct *tsk)
> +{
> +	if (tsk == b->signaler)
> +		return INT_MIN;
> +	else
> +		return tsk->prio;
> +}
> +
>   void intel_engine_remove_wait(struct intel_engine_cs *engine,
>   			      struct intel_wait *wait)
>   {
> @@ -273,8 +284,8 @@ void intel_engine_remove_wait(struct intel_engine_cs *engine,
>   		goto out_unlock;
>
>   	if (b->first_wait == wait) {
> +		const int priority = wakeup_priority(b, wait->tsk);
>   		struct rb_node *next;
> -		const int priority = wait->tsk->prio;
>
>   		GEM_BUG_ON(b->tasklet != wait->tsk);
>
> @@ -340,15 +351,176 @@ out_unlock:
>   	spin_unlock(&b->lock);
>   }
>
> +struct signal {
> +	struct rb_node node;
> +	struct intel_wait wait;
> +	struct drm_i915_gem_request *request;
> +};
> +
> +static bool signal_complete(struct signal *signal)
> +{
> +	if (signal == NULL)
> +		return false;
> +
> +	/* If another process served as the bottom-half it may have already
> +	 * signalled that this wait is already completed.
> +	 */
> +	if (intel_wait_complete(&signal->wait))
> +		return true;
> +
> +	/* Carefully check if the request is complete, giving time for the
> +	 * seqno to be visible or if the GPU hung.
> +	 */
> +	if (__i915_request_irq_complete(signal->request))
> +		return true;
> +
> +	return false;
> +}
> +
> +static struct signal *to_signal(struct rb_node *rb)
> +{
> +	return container_of(rb, struct signal, node);
> +}
> +
> +static void signaler_set_rtpriority(void)
> +{
> +	 struct sched_param param = { .sched_priority = 1 };
> +	 sched_setscheduler_nocheck(current, SCHED_FIFO, &param);
> +}
> +
> +static int intel_breadcrumbs_signaler(void *arg)
> +{
> +	struct intel_engine_cs *engine = arg;
> +	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> +	struct signal *signal;
> +
> +	/* Install ourselves with high priority to reduce signalling latency */
> +	signaler_set_rtpriority();
> +
> +	do {
> +		set_current_state(TASK_INTERRUPTIBLE);
> +
> +		/* We are either woken up by the interrupt bottom-half,
> +		 * or by a client adding a new signaller. In both cases,
> +		 * the GPU seqno may have advanced beyond our oldest signal.
> +		 * If it has, propagate the signal, remove the waiter and
> +		 * check again with the next oldest signal. Otherwise we
> +		 * need to wait for a new interrupt from the GPU or for
> +		 * a new client.
> +		 */
> +		signal = READ_ONCE(b->first_signal);
> +		if (signal_complete(signal)) {
> +			/* Wake up all other completed waiters and select the
> +			 * next bottom-half for the next user interrupt.
> +			 */
> +			intel_engine_remove_wait(engine, &signal->wait);
> +
> +			i915_gem_request_unreference(signal->request);
> +
> +			/* Find the next oldest signal. Note that as we have
> +			 * not been holding the lock, another client may
> +			 * have installed an even older signal than the one
> +			 * we just completed - so double check we are still
> +			 * the oldest before picking the next one.
> +			 */
> +			spin_lock(&b->lock);
> +			if (signal == b->first_signal)
> +				b->first_signal = rb_next(&signal->node);
> +			rb_erase(&signal->node, &b->signals);
> +			spin_unlock(&b->lock);
> +
> +			kfree(signal);
> +		} else {
> +			if (kthread_should_stop())
> +				break;
> +
> +			schedule();
> +		}
> +	} while (1);
> +	__set_current_state(TASK_RUNNING);
> +
> +	return 0;
> +}
> +
> +int intel_engine_enable_signaling(struct drm_i915_gem_request *request)
> +{
> +	struct intel_engine_cs *engine = request->engine;
> +	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> +	struct rb_node *parent, **p;
> +	struct signal *signal;
> +	bool first, wakeup;
> +
> +	signal = kmalloc(sizeof(*signal), GFP_ATOMIC);
> +	if (unlikely(!signal))
> +		return -ENOMEM;
> +
> +	signal->wait.tsk = b->signaler;
> +	signal->wait.seqno = request->seqno;
> +
> +	signal->request = i915_gem_request_reference(request);
> +
> +	/* First add ourselves into the list of waiters, but register our
> +	 * bottom-half as the signaller thread. As per usual, only the oldest
> +	 * waiter (not just signaller) is tasked as the bottom-half waking
> +	 * up all completed waiters after the user interrupt.
> +	 *
> +	 * If we are the oldest waiter, enable the irq (after which we
> +	 * must double check that the seqno did not complete).
> +	 */
> +	wakeup = intel_engine_add_wait(engine, &signal->wait);
> +
> +	/* Now insert ourselves into the retirement ordered list of signals
> +	 * on this engine. We track the oldest seqno as that will be the
> +	 * first signal to complete.
> +	 */
> +	spin_lock(&b->lock);
> +	parent = NULL;
> +	first = true;
> +	p = &b->signals.rb_node;
> +	while (*p) {
> +		parent = *p;
> +		if (i915_seqno_passed(signal->wait.seqno,
> +				      to_signal(parent)->wait.seqno)) {
> +			p = &parent->rb_right;
> +			first = false;
> +		} else
> +			p = &parent->rb_left;
> +	}
> +	rb_link_node(&signal->node, parent, p);
> +	rb_insert_color(&signal->node, &b->signals);
> +	if (first)
> +		smp_store_mb(b->first_signal, signal);
> +	spin_unlock(&b->lock);
> +
> +	if (wakeup)
> +		wake_up_process(b->signaler);
> +
> +	return 0;
> +}
> +
>   int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
>   {
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> +	struct task_struct *tsk;
>
>   	spin_lock_init(&b->lock);
>   	setup_timer(&b->fake_irq,
>   		    intel_breadcrumbs_fake_irq,
>   		    (unsigned long)engine);
>
> +	/* Spawn a thread to provide a common bottom-half for all signals.
> +	 * As this is an asynchronous interface we cannot steal the current
> +	 * task for handling the bottom-half to the user interrupt, therefore
> +	 * we create a thread to do the coherent seqno dance after the
> +	 * interrupt and then signal the waitqueue (via the dma-buf/fence).
> +	 */
> +	tsk = kthread_run(intel_breadcrumbs_signaler, engine,
> +			  "irq/i915:%d", engine->id);
> +	if (IS_ERR(tsk))
> +		return PTR_ERR(tsk);
> +
> +	b->signaler = tsk;
> +
>   	return 0;
>   }
>
> @@ -356,6 +528,9 @@ void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine)
>   {
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
>
> +	if (!IS_ERR_OR_NULL(b->signaler))
> +		kthread_stop(b->signaler);
> +
>   	del_timer_sync(&b->fake_irq);
>   }
>
> @@ -377,3 +552,18 @@ unsigned intel_kick_waiters(struct drm_i915_private *i915)
>
>   	return mask;
>   }
> +
> +unsigned intel_kick_signalers(struct drm_i915_private *i915)
> +{
> +	struct intel_engine_cs *engine;
> +	unsigned mask = 0;
> +
> +	for_each_engine(engine, i915) {
> +		if (unlikely(READ_ONCE(engine->breadcrumbs.first_signal))) {
> +			wake_up_process(engine->breadcrumbs.signaler);
> +			mask |= intel_engine_flag(engine);
> +		}
> +	}
> +
> +	return mask;
> +}
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index 4d6c2b773a64..c05b45727f7d 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -141,6 +141,8 @@ struct  i915_ctx_workarounds {
>   	struct drm_i915_gem_object *obj;
>   };
>
> +struct drm_i915_gem_request;
> +
>   struct intel_engine_cs {
>   	struct drm_i915_private *i915;
>   	const char	*name;
> @@ -179,8 +181,11 @@ struct intel_engine_cs {
>   	struct intel_breadcrumbs {
>   		spinlock_t lock; /* protects the lists of requests */
>   		struct rb_root waiters; /* sorted by retirement, priority */
> +		struct rb_root signals; /* sorted by retirement */
>   		struct intel_wait *first_wait; /* oldest waiter by retirement */
>   		struct task_struct *tasklet; /* bh for user interrupts */
> +		struct task_struct *signaler; /* used for fence signalling */
> +		void *first_signal;
>   		struct timer_list fake_irq; /* used after a missed interrupt */
>   		bool irq_enabled;
>   		bool rpm_wakelock;
> @@ -199,7 +204,6 @@ struct intel_engine_cs {
>   	unsigned irq_refcount; /* protected by dev_priv->irq_lock */
>   	bool		irq_posted;
>   	u32		irq_enable_mask;	/* bitmask to enable ring interrupt */
> -	struct drm_i915_gem_request *trace_irq_req;
>   	bool __must_check (*irq_get)(struct intel_engine_cs *ring);
>   	void		(*irq_put)(struct intel_engine_cs *ring);
>
> @@ -540,6 +544,7 @@ bool intel_engine_add_wait(struct intel_engine_cs *engine,
>   			   struct intel_wait *wait);
>   void intel_engine_remove_wait(struct intel_engine_cs *engine,
>   			      struct intel_wait *wait);
> +int intel_engine_enable_signaling(struct drm_i915_gem_request *request);
>   static inline bool intel_engine_has_waiter(struct intel_engine_cs *engine)
>   {
>   	return READ_ONCE(engine->breadcrumbs.tasklet);
> @@ -563,5 +568,6 @@ static inline bool intel_engine_wakeup(struct intel_engine_cs *engine)
>   void intel_engine_enable_fake_irq(struct intel_engine_cs *engine);
>   void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
>   unsigned intel_kick_waiters(struct drm_i915_private *i915);
> +unsigned intel_kick_signalers(struct drm_i915_private *i915);
>
>   #endif /* _INTEL_RINGBUFFER_H_ */
>

Looks good to me.

Only thing I would possibly change is the thread name to not mislead 
anyone into thinking this is the main and only i915 irq activity. Looks 
to me it should just be called something like i915/signaller:%u or similar.

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

Re: [PATCH 21/25] drm/i915: Convert trace-irq to the breadcrumb waiter

[Chris Wilson]

On Mon, Jun 27, 2016 at 12:38:37PM +0100, Tvrtko Ursulin wrote:
> Looks good to me.
> 
> Only thing I would possibly change is the thread name to not mislead
> anyone into thinking this is the main and only i915 irq activity.
> Looks to me it should just be called something like
> i915/signaller:%u or similar.

I wanted it fairly short for top (about 10 characters on a normal
terminal iirc). There is nothing preventing us from having multiple irq
threads (and preceding with irq seems to fit in with others).

If we need more, I would suggest irq/i915:%u-signal
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 22/25] drm/i915: Embed signaling node into the GEM request

[Chris Wilson]

Under the assumption that enabling signaling will be a frequent
operation, lets preallocate our attachments for signaling inside the
(rather large) request struct (and so benefiting from the slab cache).

v2: Convert from void * to more meaningful names and types.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h          |  1 +
 drivers/gpu/drm/i915/intel_breadcrumbs.c | 76 ++++++++++++++++----------------
 drivers/gpu/drm/i915/intel_ringbuffer.h  | 10 ++++-
 3 files changed, 46 insertions(+), 41 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index b7089cedb80c..440f6267ec27 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2381,6 +2381,7 @@ struct drm_i915_gem_request {
 	struct drm_i915_private *i915;
 	struct intel_engine_cs *engine;
 	unsigned reset_counter;
+	struct intel_signal_node signaling;
 
 	 /** GEM sequence number associated with the previous request,
 	  * when the HWS breadcrumb is equal to this the GPU is processing
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index bc0e9e8b7871..661d6b415ac3 100644
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -351,35 +351,29 @@ out_unlock:
 	spin_unlock(&b->lock);
 }
 
-struct signal {
-	struct rb_node node;
-	struct intel_wait wait;
-	struct drm_i915_gem_request *request;
-};
-
-static bool signal_complete(struct signal *signal)
+static bool signal_complete(struct drm_i915_gem_request *request)
 {
-	if (signal == NULL)
+	if (request == NULL)
 		return false;
 
 	/* If another process served as the bottom-half it may have already
 	 * signalled that this wait is already completed.
 	 */
-	if (intel_wait_complete(&signal->wait))
+	if (intel_wait_complete(&request->signaling.wait))
 		return true;
 
 	/* Carefully check if the request is complete, giving time for the
 	 * seqno to be visible or if the GPU hung.
 	 */
-	if (__i915_request_irq_complete(signal->request))
+	if (__i915_request_irq_complete(request))
 		return true;
 
 	return false;
 }
 
-static struct signal *to_signal(struct rb_node *rb)
+static struct drm_i915_gem_request *to_signaler(struct rb_node *rb)
 {
-	return container_of(rb, struct signal, node);
+	return container_of(rb, struct drm_i915_gem_request, signaling.node);
 }
 
 static void signaler_set_rtpriority(void)
@@ -392,7 +386,7 @@ static int intel_breadcrumbs_signaler(void *arg)
 {
 	struct intel_engine_cs *engine = arg;
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
-	struct signal *signal;
+	struct drm_i915_gem_request *request;
 
 	/* Install ourselves with high priority to reduce signalling latency */
 	signaler_set_rtpriority();
@@ -408,14 +402,13 @@ static int intel_breadcrumbs_signaler(void *arg)
 		 * need to wait for a new interrupt from the GPU or for
 		 * a new client.
 		 */
-		signal = READ_ONCE(b->first_signal);
-		if (signal_complete(signal)) {
+		request = READ_ONCE(b->first_signal);
+		if (signal_complete(request)) {
 			/* Wake up all other completed waiters and select the
 			 * next bottom-half for the next user interrupt.
 			 */
-			intel_engine_remove_wait(engine, &signal->wait);
-
-			i915_gem_request_unreference(signal->request);
+			intel_engine_remove_wait(engine,
+						 &request->signaling.wait);
 
 			/* Find the next oldest signal. Note that as we have
 			 * not been holding the lock, another client may
@@ -424,12 +417,15 @@ static int intel_breadcrumbs_signaler(void *arg)
 			 * the oldest before picking the next one.
 			 */
 			spin_lock(&b->lock);
-			if (signal == b->first_signal)
-				b->first_signal = rb_next(&signal->node);
-			rb_erase(&signal->node, &b->signals);
+			if (request == b->first_signal) {
+				struct rb_node *rb =
+					rb_next(&request->signaling.node);
+				b->first_signal = rb ? to_signaler(rb) : NULL;
+			}
+			rb_erase(&request->signaling.node, &b->signals);
 			spin_unlock(&b->lock);
 
-			kfree(signal);
+			i915_gem_request_unreference(request);
 		} else {
 			if (kthread_should_stop())
 				break;
@@ -442,22 +438,25 @@ static int intel_breadcrumbs_signaler(void *arg)
 	return 0;
 }
 
-int intel_engine_enable_signaling(struct drm_i915_gem_request *request)
+void intel_engine_enable_signaling(struct drm_i915_gem_request *request)
 {
 	struct intel_engine_cs *engine = request->engine;
 	struct intel_breadcrumbs *b = &engine->breadcrumbs;
 	struct rb_node *parent, **p;
-	struct signal *signal;
 	bool first, wakeup;
 
-	signal = kmalloc(sizeof(*signal), GFP_ATOMIC);
-	if (unlikely(!signal))
-		return -ENOMEM;
+	if (unlikely(READ_ONCE(request->signaling.wait.tsk)))
+		return;
 
-	signal->wait.tsk = b->signaler;
-	signal->wait.seqno = request->seqno;
+	spin_lock(&b->lock);
+	if (unlikely(request->signaling.wait.tsk)) {
+		wakeup = false;
+		goto unlock;
+	}
 
-	signal->request = i915_gem_request_reference(request);
+	request->signaling.wait.tsk = b->signaler;
+	request->signaling.wait.seqno = request->seqno;
+	i915_gem_request_reference(request);
 
 	/* First add ourselves into the list of waiters, but register our
 	 * bottom-half as the signaller thread. As per usual, only the oldest
@@ -467,35 +466,34 @@ int intel_engine_enable_signaling(struct drm_i915_gem_request *request)
 	 * If we are the oldest waiter, enable the irq (after which we
 	 * must double check that the seqno did not complete).
 	 */
-	wakeup = intel_engine_add_wait(engine, &signal->wait);
+	wakeup = __intel_engine_add_wait(engine, &request->signaling.wait);
 
 	/* Now insert ourselves into the retirement ordered list of signals
 	 * on this engine. We track the oldest seqno as that will be the
 	 * first signal to complete.
 	 */
-	spin_lock(&b->lock);
 	parent = NULL;
 	first = true;
 	p = &b->signals.rb_node;
 	while (*p) {
 		parent = *p;
-		if (i915_seqno_passed(signal->wait.seqno,
-				      to_signal(parent)->wait.seqno)) {
+		if (i915_seqno_passed(request->seqno,
+				      to_signaler(parent)->seqno)) {
 			p = &parent->rb_right;
 			first = false;
 		} else
 			p = &parent->rb_left;
 	}
-	rb_link_node(&signal->node, parent, p);
-	rb_insert_color(&signal->node, &b->signals);
+	rb_link_node(&request->signaling.node, parent, p);
+	rb_insert_color(&request->signaling.node, &b->signals);
 	if (first)
-		smp_store_mb(b->first_signal, signal);
+		smp_store_mb(b->first_signal, request);
+
+unlock:
 	spin_unlock(&b->lock);
 
 	if (wakeup)
 		wake_up_process(b->signaler);
-
-	return 0;
 }
 
 int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index c05b45727f7d..b55b9961d9f6 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -185,7 +185,7 @@ struct intel_engine_cs {
 		struct intel_wait *first_wait; /* oldest waiter by retirement */
 		struct task_struct *tasklet; /* bh for user interrupts */
 		struct task_struct *signaler; /* used for fence signalling */
-		void *first_signal;
+		struct drm_i915_gem_request *first_signal;
 		struct timer_list fake_irq; /* used after a missed interrupt */
 		bool irq_enabled;
 		bool rpm_wakelock;
@@ -530,6 +530,12 @@ struct intel_wait {
 	struct task_struct *tsk;
 	u32 seqno;
 };
+
+struct intel_signal_node {
+	struct rb_node node;
+	struct intel_wait wait;
+};
+
 int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
 static inline void intel_wait_init(struct intel_wait *wait, u32 seqno)
 {
@@ -544,7 +550,7 @@ bool intel_engine_add_wait(struct intel_engine_cs *engine,
 			   struct intel_wait *wait);
 void intel_engine_remove_wait(struct intel_engine_cs *engine,
 			      struct intel_wait *wait);
-int intel_engine_enable_signaling(struct drm_i915_gem_request *request);
+void intel_engine_enable_signaling(struct drm_i915_gem_request *request);
 static inline bool intel_engine_has_waiter(struct intel_engine_cs *engine)
 {
 	return READ_ONCE(engine->breadcrumbs.tasklet);
-- 
2.8.1

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

Re: [PATCH 22/25] drm/i915: Embed signaling node into the GEM request

[Tvrtko Ursulin]

On 25/06/16 11:13, Chris Wilson wrote:
> Under the assumption that enabling signaling will be a frequent
> operation, lets preallocate our attachments for signaling inside the
> (rather large) request struct (and so benefiting from the slab cache).
>
> v2: Convert from void * to more meaningful names and types.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_drv.h          |  1 +
>   drivers/gpu/drm/i915/intel_breadcrumbs.c | 76 ++++++++++++++++----------------
>   drivers/gpu/drm/i915/intel_ringbuffer.h  | 10 ++++-
>   3 files changed, 46 insertions(+), 41 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index b7089cedb80c..440f6267ec27 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -2381,6 +2381,7 @@ struct drm_i915_gem_request {
>   	struct drm_i915_private *i915;
>   	struct intel_engine_cs *engine;
>   	unsigned reset_counter;
> +	struct intel_signal_node signaling;
>
>   	 /** GEM sequence number associated with the previous request,
>   	  * when the HWS breadcrumb is equal to this the GPU is processing
> diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> index bc0e9e8b7871..661d6b415ac3 100644
> --- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
> +++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> @@ -351,35 +351,29 @@ out_unlock:
>   	spin_unlock(&b->lock);
>   }
>
> -struct signal {
> -	struct rb_node node;
> -	struct intel_wait wait;
> -	struct drm_i915_gem_request *request;
> -};
> -
> -static bool signal_complete(struct signal *signal)
> +static bool signal_complete(struct drm_i915_gem_request *request)
>   {
> -	if (signal == NULL)
> +	if (request == NULL)
>   		return false;
>
>   	/* If another process served as the bottom-half it may have already
>   	 * signalled that this wait is already completed.
>   	 */
> -	if (intel_wait_complete(&signal->wait))
> +	if (intel_wait_complete(&request->signaling.wait))
>   		return true;
>
>   	/* Carefully check if the request is complete, giving time for the
>   	 * seqno to be visible or if the GPU hung.
>   	 */
> -	if (__i915_request_irq_complete(signal->request))
> +	if (__i915_request_irq_complete(request))
>   		return true;
>
>   	return false;
>   }
>
> -static struct signal *to_signal(struct rb_node *rb)
> +static struct drm_i915_gem_request *to_signaler(struct rb_node *rb)
>   {
> -	return container_of(rb, struct signal, node);
> +	return container_of(rb, struct drm_i915_gem_request, signaling.node);
>   }
>
>   static void signaler_set_rtpriority(void)
> @@ -392,7 +386,7 @@ static int intel_breadcrumbs_signaler(void *arg)
>   {
>   	struct intel_engine_cs *engine = arg;
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
> -	struct signal *signal;
> +	struct drm_i915_gem_request *request;
>
>   	/* Install ourselves with high priority to reduce signalling latency */
>   	signaler_set_rtpriority();
> @@ -408,14 +402,13 @@ static int intel_breadcrumbs_signaler(void *arg)
>   		 * need to wait for a new interrupt from the GPU or for
>   		 * a new client.
>   		 */
> -		signal = READ_ONCE(b->first_signal);
> -		if (signal_complete(signal)) {
> +		request = READ_ONCE(b->first_signal);
> +		if (signal_complete(request)) {
>   			/* Wake up all other completed waiters and select the
>   			 * next bottom-half for the next user interrupt.
>   			 */
> -			intel_engine_remove_wait(engine, &signal->wait);
> -
> -			i915_gem_request_unreference(signal->request);
> +			intel_engine_remove_wait(engine,
> +						 &request->signaling.wait);
>
>   			/* Find the next oldest signal. Note that as we have
>   			 * not been holding the lock, another client may
> @@ -424,12 +417,15 @@ static int intel_breadcrumbs_signaler(void *arg)
>   			 * the oldest before picking the next one.
>   			 */
>   			spin_lock(&b->lock);
> -			if (signal == b->first_signal)
> -				b->first_signal = rb_next(&signal->node);
> -			rb_erase(&signal->node, &b->signals);
> +			if (request == b->first_signal) {
> +				struct rb_node *rb =
> +					rb_next(&request->signaling.node);
> +				b->first_signal = rb ? to_signaler(rb) : NULL;
> +			}
> +			rb_erase(&request->signaling.node, &b->signals);
>   			spin_unlock(&b->lock);
>
> -			kfree(signal);
> +			i915_gem_request_unreference(request);
>   		} else {
>   			if (kthread_should_stop())
>   				break;
> @@ -442,22 +438,25 @@ static int intel_breadcrumbs_signaler(void *arg)
>   	return 0;
>   }
>
> -int intel_engine_enable_signaling(struct drm_i915_gem_request *request)
> +void intel_engine_enable_signaling(struct drm_i915_gem_request *request)
>   {
>   	struct intel_engine_cs *engine = request->engine;
>   	struct intel_breadcrumbs *b = &engine->breadcrumbs;
>   	struct rb_node *parent, **p;
> -	struct signal *signal;
>   	bool first, wakeup;
>
> -	signal = kmalloc(sizeof(*signal), GFP_ATOMIC);
> -	if (unlikely(!signal))
> -		return -ENOMEM;
> +	if (unlikely(READ_ONCE(request->signaling.wait.tsk)))
> +		return;
>
> -	signal->wait.tsk = b->signaler;
> -	signal->wait.seqno = request->seqno;
> +	spin_lock(&b->lock);
> +	if (unlikely(request->signaling.wait.tsk)) {
> +		wakeup = false;
> +		goto unlock;
> +	}
>
> -	signal->request = i915_gem_request_reference(request);
> +	request->signaling.wait.tsk = b->signaler;
> +	request->signaling.wait.seqno = request->seqno;
> +	i915_gem_request_reference(request);
>
>   	/* First add ourselves into the list of waiters, but register our
>   	 * bottom-half as the signaller thread. As per usual, only the oldest
> @@ -467,35 +466,34 @@ int intel_engine_enable_signaling(struct drm_i915_gem_request *request)
>   	 * If we are the oldest waiter, enable the irq (after which we
>   	 * must double check that the seqno did not complete).
>   	 */
> -	wakeup = intel_engine_add_wait(engine, &signal->wait);
> +	wakeup = __intel_engine_add_wait(engine, &request->signaling.wait);
>
>   	/* Now insert ourselves into the retirement ordered list of signals
>   	 * on this engine. We track the oldest seqno as that will be the
>   	 * first signal to complete.
>   	 */
> -	spin_lock(&b->lock);
>   	parent = NULL;
>   	first = true;
>   	p = &b->signals.rb_node;
>   	while (*p) {
>   		parent = *p;
> -		if (i915_seqno_passed(signal->wait.seqno,
> -				      to_signal(parent)->wait.seqno)) {
> +		if (i915_seqno_passed(request->seqno,
> +				      to_signaler(parent)->seqno)) {
>   			p = &parent->rb_right;
>   			first = false;
>   		} else
>   			p = &parent->rb_left;
>   	}
> -	rb_link_node(&signal->node, parent, p);
> -	rb_insert_color(&signal->node, &b->signals);
> +	rb_link_node(&request->signaling.node, parent, p);
> +	rb_insert_color(&request->signaling.node, &b->signals);
>   	if (first)
> -		smp_store_mb(b->first_signal, signal);
> +		smp_store_mb(b->first_signal, request);
> +
> +unlock:
>   	spin_unlock(&b->lock);
>
>   	if (wakeup)
>   		wake_up_process(b->signaler);
> -
> -	return 0;
>   }
>
>   int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine)
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index c05b45727f7d..b55b9961d9f6 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -185,7 +185,7 @@ struct intel_engine_cs {
>   		struct intel_wait *first_wait; /* oldest waiter by retirement */
>   		struct task_struct *tasklet; /* bh for user interrupts */
>   		struct task_struct *signaler; /* used for fence signalling */
> -		void *first_signal;
> +		struct drm_i915_gem_request *first_signal;
>   		struct timer_list fake_irq; /* used after a missed interrupt */
>   		bool irq_enabled;
>   		bool rpm_wakelock;
> @@ -530,6 +530,12 @@ struct intel_wait {
>   	struct task_struct *tsk;
>   	u32 seqno;
>   };
> +
> +struct intel_signal_node {
> +	struct rb_node node;
> +	struct intel_wait wait;
> +};
> +
>   int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
>   static inline void intel_wait_init(struct intel_wait *wait, u32 seqno)
>   {
> @@ -544,7 +550,7 @@ bool intel_engine_add_wait(struct intel_engine_cs *engine,
>   			   struct intel_wait *wait);
>   void intel_engine_remove_wait(struct intel_engine_cs *engine,
>   			      struct intel_wait *wait);
> -int intel_engine_enable_signaling(struct drm_i915_gem_request *request);
> +void intel_engine_enable_signaling(struct drm_i915_gem_request *request);
>   static inline bool intel_engine_has_waiter(struct intel_engine_cs *engine)
>   {
>   	return READ_ONCE(engine->breadcrumbs.tasklet);
>

Looks OK.

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

[PATCH 23/25] drm/i915: Move the get/put irq locking into the caller

[Chris Wilson]

With only a single callsite for intel_engine_cs->irq_get and ->irq_put,
we can reduce the code size by moving the common preamble into the
caller, and we can also eliminate the reference counting.

For completeness, as we are no longer doing reference counting on irq,
rename the get/put vfunctions to enable/disable respectively and are
able to review the use of posting reads. We only require the
serialisation with hardware when enabling the interrupt (i.e. so we
cannot miss an interrupt by going to sleep before the hardware truly
enables it).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_irq.c          |   8 +-
 drivers/gpu/drm/i915/intel_breadcrumbs.c |  10 +-
 drivers/gpu/drm/i915/intel_lrc.c         |  34 +---
 drivers/gpu/drm/i915/intel_ringbuffer.c  | 269 ++++++++++---------------------
 drivers/gpu/drm/i915/intel_ringbuffer.h  |   5 +-
 5 files changed, 108 insertions(+), 218 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 7724bae27bcf..be25b7bdacfe 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -259,12 +259,12 @@ static void ilk_update_gt_irq(struct drm_i915_private *dev_priv,
 	dev_priv->gt_irq_mask &= ~interrupt_mask;
 	dev_priv->gt_irq_mask |= (~enabled_irq_mask & interrupt_mask);
 	I915_WRITE(GTIMR, dev_priv->gt_irq_mask);
-	POSTING_READ(GTIMR);
 }
 
 void gen5_enable_gt_irq(struct drm_i915_private *dev_priv, uint32_t mask)
 {
 	ilk_update_gt_irq(dev_priv, mask, mask);
+	POSTING_READ_FW(GTIMR);
 }
 
 void gen5_disable_gt_irq(struct drm_i915_private *dev_priv, uint32_t mask)
@@ -2819,9 +2819,9 @@ ring_idle(struct intel_engine_cs *engine, u32 seqno)
 }
 
 static bool
-ipehr_is_semaphore_wait(struct drm_i915_private *dev_priv, u32 ipehr)
+ipehr_is_semaphore_wait(struct intel_engine_cs *engine, u32 ipehr)
 {
-	if (INTEL_GEN(dev_priv) >= 8) {
+	if (INTEL_GEN(engine->i915) >= 8) {
 		return (ipehr >> 23) == 0x1c;
 	} else {
 		ipehr &= ~MI_SEMAPHORE_SYNC_MASK;
@@ -2892,7 +2892,7 @@ semaphore_waits_for(struct intel_engine_cs *engine, u32 *seqno)
 		return NULL;
 
 	ipehr = I915_READ(RING_IPEHR(engine->mmio_base));
-	if (!ipehr_is_semaphore_wait(engine->i915, ipehr))
+	if (!ipehr_is_semaphore_wait(engine, ipehr))
 		return NULL;
 
 	/*
diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index 661d6b415ac3..3c2e53acd187 100644
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -50,12 +50,18 @@ static void irq_enable(struct intel_engine_cs *engine)
 	 * just in case.
 	 */
 	engine->irq_posted = true;
-	WARN_ON(!engine->irq_get(engine));
+
+	spin_lock_irq(&engine->i915->irq_lock);
+	engine->irq_enable(engine);
+	spin_unlock_irq(&engine->i915->irq_lock);
 }
 
 static void irq_disable(struct intel_engine_cs *engine)
 {
-	engine->irq_put(engine);
+	spin_lock_irq(&engine->i915->irq_lock);
+	engine->irq_disable(engine);
+	spin_unlock_irq(&engine->i915->irq_lock);
+
 	engine->irq_posted = false;
 }
 
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 526c71975f11..e8fecf051e2c 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1614,36 +1614,18 @@ static int gen8_emit_bb_start(struct drm_i915_gem_request *req,
 	return 0;
 }
 
-static bool gen8_logical_ring_get_irq(struct intel_engine_cs *engine)
+static void gen8_logical_ring_enable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
-
-	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
-		return false;
-
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (engine->irq_refcount++ == 0) {
-		I915_WRITE_IMR(engine,
-			       ~(engine->irq_enable_mask | engine->irq_keep_mask));
-		POSTING_READ(RING_IMR(engine->mmio_base));
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
-
-	return true;
+	I915_WRITE_IMR(engine,
+		       ~(engine->irq_enable_mask | engine->irq_keep_mask));
+	POSTING_READ_FW(RING_IMR(engine->mmio_base));
 }
 
-static void gen8_logical_ring_put_irq(struct intel_engine_cs *engine)
+static void gen8_logical_ring_disable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
-
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (--engine->irq_refcount == 0) {
-		I915_WRITE_IMR(engine, ~engine->irq_keep_mask);
-		POSTING_READ(RING_IMR(engine->mmio_base));
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
+	I915_WRITE_IMR(engine, ~engine->irq_keep_mask);
 }
 
 static int gen8_emit_flush(struct drm_i915_gem_request *request,
@@ -1963,8 +1945,8 @@ logical_ring_default_vfuncs(struct intel_engine_cs *engine)
 	engine->init_hw = gen8_init_common_ring;
 	engine->emit_request = gen8_emit_request;
 	engine->emit_flush = gen8_emit_flush;
-	engine->irq_get = gen8_logical_ring_get_irq;
-	engine->irq_put = gen8_logical_ring_put_irq;
+	engine->irq_enable = gen8_logical_ring_enable_irq;
+	engine->irq_disable = gen8_logical_ring_disable_irq;
 	engine->emit_bb_start = gen8_emit_bb_start;
 	if (IS_BXT_REVID(engine->i915, 0, BXT_REVID_A1))
 		engine->irq_seqno_barrier = bxt_a_seqno_barrier;
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index ad9b54071159..d7b5aae9cfd2 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -1632,103 +1632,54 @@ gen6_seqno_barrier(struct intel_engine_cs *engine)
 	spin_unlock_irq(&dev_priv->uncore.lock);
 }
 
-static bool
-gen5_ring_get_irq(struct intel_engine_cs *engine)
+static void
+gen5_ring_enable_irq(struct intel_engine_cs *engine)
 {
-	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
-
-	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
-		return false;
-
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (engine->irq_refcount++ == 0)
-		gen5_enable_gt_irq(dev_priv, engine->irq_enable_mask);
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
-
-	return true;
+	gen5_enable_gt_irq(engine->i915, engine->irq_enable_mask);
 }
 
 static void
-gen5_ring_put_irq(struct intel_engine_cs *engine)
+gen5_ring_disable_irq(struct intel_engine_cs *engine)
 {
-	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
-
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (--engine->irq_refcount == 0)
-		gen5_disable_gt_irq(dev_priv, engine->irq_enable_mask);
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
+	gen5_disable_gt_irq(engine->i915, engine->irq_enable_mask);
 }
 
-static bool
-i9xx_ring_get_irq(struct intel_engine_cs *engine)
+static void
+i9xx_ring_enable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
-
-	if (!intel_irqs_enabled(dev_priv))
-		return false;
-
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (engine->irq_refcount++ == 0) {
-		dev_priv->irq_mask &= ~engine->irq_enable_mask;
-		I915_WRITE(IMR, dev_priv->irq_mask);
-		POSTING_READ(IMR);
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
 
-	return true;
+	dev_priv->irq_mask &= ~engine->irq_enable_mask;
+	I915_WRITE(IMR, dev_priv->irq_mask);
+	POSTING_READ_FW(RING_IMR(engine->mmio_base));
 }
 
 static void
-i9xx_ring_put_irq(struct intel_engine_cs *engine)
+i9xx_ring_disable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
 
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (--engine->irq_refcount == 0) {
-		dev_priv->irq_mask |= engine->irq_enable_mask;
-		I915_WRITE(IMR, dev_priv->irq_mask);
-		POSTING_READ(IMR);
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
+	dev_priv->irq_mask |= engine->irq_enable_mask;
+	I915_WRITE(IMR, dev_priv->irq_mask);
 }
 
-static bool
-i8xx_ring_get_irq(struct intel_engine_cs *engine)
+static void
+i8xx_ring_enable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
 
-	if (!intel_irqs_enabled(dev_priv))
-		return false;
-
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (engine->irq_refcount++ == 0) {
-		dev_priv->irq_mask &= ~engine->irq_enable_mask;
-		I915_WRITE16(IMR, dev_priv->irq_mask);
-		POSTING_READ16(IMR);
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
-
-	return true;
+	dev_priv->irq_mask &= ~engine->irq_enable_mask;
+	I915_WRITE16(IMR, dev_priv->irq_mask);
+	POSTING_READ16(RING_IMR(engine->mmio_base));
 }
 
 static void
-i8xx_ring_put_irq(struct intel_engine_cs *engine)
+i8xx_ring_disable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
 
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (--engine->irq_refcount == 0) {
-		dev_priv->irq_mask |= engine->irq_enable_mask;
-		I915_WRITE16(IMR, dev_priv->irq_mask);
-		POSTING_READ16(IMR);
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
+	dev_priv->irq_mask |= engine->irq_enable_mask;
+	I915_WRITE16(IMR, dev_priv->irq_mask);
 }
 
 static int
@@ -1769,122 +1720,74 @@ i9xx_add_request(struct drm_i915_gem_request *req)
 	return 0;
 }
 
-static bool
-gen6_ring_get_irq(struct intel_engine_cs *engine)
+static void
+gen6_ring_enable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
-
-	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
-		return false;
 
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (engine->irq_refcount++ == 0) {
-		if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
-			I915_WRITE_IMR(engine,
-				       ~(engine->irq_enable_mask |
-					 GT_PARITY_ERROR(dev_priv)));
-		else
-			I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
-		gen5_enable_gt_irq(dev_priv, engine->irq_enable_mask);
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
-
-	return true;
+	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
+		I915_WRITE_IMR(engine,
+			       ~(engine->irq_enable_mask |
+				 GT_PARITY_ERROR(dev_priv)));
+	else
+		I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
+	gen5_enable_gt_irq(dev_priv, engine->irq_enable_mask);
 }
 
 static void
-gen6_ring_put_irq(struct intel_engine_cs *engine)
+gen6_ring_disable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
 
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (--engine->irq_refcount == 0) {
-		if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
-			I915_WRITE_IMR(engine, ~GT_PARITY_ERROR(dev_priv));
-		else
-			I915_WRITE_IMR(engine, ~0);
-		gen5_disable_gt_irq(dev_priv, engine->irq_enable_mask);
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
+	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
+		I915_WRITE_IMR(engine, ~GT_PARITY_ERROR(dev_priv));
+	else
+		I915_WRITE_IMR(engine, ~0);
+	gen5_disable_gt_irq(dev_priv, engine->irq_enable_mask);
 }
 
-static bool
-hsw_vebox_get_irq(struct intel_engine_cs *engine)
+static void
+hsw_vebox_enable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
-
-	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
-		return false;
 
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (engine->irq_refcount++ == 0) {
-		I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
-		gen6_enable_pm_irq(dev_priv, engine->irq_enable_mask);
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
-
-	return true;
+	I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
+	gen6_enable_pm_irq(dev_priv, engine->irq_enable_mask);
 }
 
 static void
-hsw_vebox_put_irq(struct intel_engine_cs *engine)
+hsw_vebox_disable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
 
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (--engine->irq_refcount == 0) {
-		I915_WRITE_IMR(engine, ~0);
-		gen6_disable_pm_irq(dev_priv, engine->irq_enable_mask);
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
+	I915_WRITE_IMR(engine, ~0);
+	gen6_disable_pm_irq(dev_priv, engine->irq_enable_mask);
 }
 
-static bool
-gen8_ring_get_irq(struct intel_engine_cs *engine)
+static void
+gen8_ring_enable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
 
-	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
-		return false;
-
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (engine->irq_refcount++ == 0) {
-		if (HAS_L3_DPF(dev_priv) && engine->id == RCS) {
-			I915_WRITE_IMR(engine,
-				       ~(engine->irq_enable_mask |
-					 GT_RENDER_L3_PARITY_ERROR_INTERRUPT));
-		} else {
-			I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
-		}
-		POSTING_READ(RING_IMR(engine->mmio_base));
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
-
-	return true;
+	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
+		I915_WRITE_IMR(engine,
+			       ~(engine->irq_enable_mask |
+				 GT_RENDER_L3_PARITY_ERROR_INTERRUPT));
+	else
+		I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
+	POSTING_READ_FW(RING_IMR(engine->mmio_base));
 }
 
 static void
-gen8_ring_put_irq(struct intel_engine_cs *engine)
+gen8_ring_disable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
-	unsigned long flags;
 
-	spin_lock_irqsave(&dev_priv->irq_lock, flags);
-	if (--engine->irq_refcount == 0) {
-		if (HAS_L3_DPF(dev_priv) && engine->id == RCS) {
-			I915_WRITE_IMR(engine,
-				       ~GT_RENDER_L3_PARITY_ERROR_INTERRUPT);
-		} else {
-			I915_WRITE_IMR(engine, ~0);
-		}
-		POSTING_READ(RING_IMR(engine->mmio_base));
-	}
-	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
+	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
+		I915_WRITE_IMR(engine,
+			       ~GT_RENDER_L3_PARITY_ERROR_INTERRUPT);
+	else
+		I915_WRITE_IMR(engine, ~0);
 }
 
 static int
@@ -2829,8 +2732,8 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 		engine->init_context = intel_rcs_ctx_init;
 		engine->add_request = gen8_render_add_request;
 		engine->flush = gen8_render_ring_flush;
-		engine->irq_get = gen8_ring_get_irq;
-		engine->irq_put = gen8_ring_put_irq;
+		engine->irq_enable = gen8_ring_enable_irq;
+		engine->irq_disable = gen8_ring_disable_irq;
 		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT;
 		if (i915_semaphore_is_enabled(dev_priv)) {
 			WARN_ON(!dev_priv->semaphore_obj);
@@ -2844,8 +2747,8 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 		engine->flush = gen7_render_ring_flush;
 		if (IS_GEN6(dev_priv))
 			engine->flush = gen6_render_ring_flush;
-		engine->irq_get = gen6_ring_get_irq;
-		engine->irq_put = gen6_ring_put_irq;
+		engine->irq_enable = gen6_ring_enable_irq;
+		engine->irq_disable = gen6_ring_disable_irq;
 		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT;
 		engine->irq_seqno_barrier = gen6_seqno_barrier;
 		if (i915_semaphore_is_enabled(dev_priv)) {
@@ -2872,8 +2775,8 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 	} else if (IS_GEN5(dev_priv)) {
 		engine->add_request = i9xx_add_request;
 		engine->flush = gen4_render_ring_flush;
-		engine->irq_get = gen5_ring_get_irq;
-		engine->irq_put = gen5_ring_put_irq;
+		engine->irq_enable = gen5_ring_enable_irq;
+		engine->irq_disable = gen5_ring_disable_irq;
 		engine->irq_seqno_barrier = gen5_seqno_barrier;
 		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT;
 	} else {
@@ -2883,11 +2786,11 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 		else
 			engine->flush = gen4_render_ring_flush;
 		if (IS_GEN2(dev_priv)) {
-			engine->irq_get = i8xx_ring_get_irq;
-			engine->irq_put = i8xx_ring_put_irq;
+			engine->irq_enable = i8xx_ring_enable_irq;
+			engine->irq_disable = i8xx_ring_disable_irq;
 		} else {
-			engine->irq_get = i9xx_ring_get_irq;
-			engine->irq_put = i9xx_ring_put_irq;
+			engine->irq_enable = i9xx_ring_enable_irq;
+			engine->irq_disable = i9xx_ring_disable_irq;
 		}
 		engine->irq_enable_mask = I915_USER_INTERRUPT;
 	}
@@ -2947,8 +2850,8 @@ int intel_init_bsd_ring_buffer(struct drm_device *dev)
 		if (INTEL_GEN(dev_priv) >= 8) {
 			engine->irq_enable_mask =
 				GT_RENDER_USER_INTERRUPT << GEN8_VCS1_IRQ_SHIFT;
-			engine->irq_get = gen8_ring_get_irq;
-			engine->irq_put = gen8_ring_put_irq;
+			engine->irq_enable = gen8_ring_enable_irq;
+			engine->irq_disable = gen8_ring_disable_irq;
 			engine->dispatch_execbuffer =
 				gen8_ring_dispatch_execbuffer;
 			if (i915_semaphore_is_enabled(dev_priv)) {
@@ -2958,8 +2861,8 @@ int intel_init_bsd_ring_buffer(struct drm_device *dev)
 			}
 		} else {
 			engine->irq_enable_mask = GT_BSD_USER_INTERRUPT;
-			engine->irq_get = gen6_ring_get_irq;
-			engine->irq_put = gen6_ring_put_irq;
+			engine->irq_enable = gen6_ring_enable_irq;
+			engine->irq_disable = gen6_ring_disable_irq;
 			engine->dispatch_execbuffer =
 				gen6_ring_dispatch_execbuffer;
 			if (i915_semaphore_is_enabled(dev_priv)) {
@@ -2983,13 +2886,13 @@ int intel_init_bsd_ring_buffer(struct drm_device *dev)
 		engine->add_request = i9xx_add_request;
 		if (IS_GEN5(dev_priv)) {
 			engine->irq_enable_mask = ILK_BSD_USER_INTERRUPT;
-			engine->irq_get = gen5_ring_get_irq;
-			engine->irq_put = gen5_ring_put_irq;
+			engine->irq_enable = gen5_ring_enable_irq;
+			engine->irq_disable = gen5_ring_disable_irq;
 			engine->irq_seqno_barrier = gen5_seqno_barrier;
 		} else {
 			engine->irq_enable_mask = I915_BSD_USER_INTERRUPT;
-			engine->irq_get = i9xx_ring_get_irq;
-			engine->irq_put = i9xx_ring_put_irq;
+			engine->irq_enable = i9xx_ring_enable_irq;
+			engine->irq_disable = i9xx_ring_disable_irq;
 		}
 		engine->dispatch_execbuffer = i965_dispatch_execbuffer;
 	}
@@ -3018,8 +2921,8 @@ int intel_init_bsd2_ring_buffer(struct drm_device *dev)
 	engine->irq_seqno_barrier = gen6_seqno_barrier;
 	engine->irq_enable_mask =
 			GT_RENDER_USER_INTERRUPT << GEN8_VCS2_IRQ_SHIFT;
-	engine->irq_get = gen8_ring_get_irq;
-	engine->irq_put = gen8_ring_put_irq;
+	engine->irq_enable = gen8_ring_enable_irq;
+	engine->irq_disable = gen8_ring_disable_irq;
 	engine->dispatch_execbuffer =
 			gen8_ring_dispatch_execbuffer;
 	if (i915_semaphore_is_enabled(dev_priv)) {
@@ -3050,8 +2953,8 @@ int intel_init_blt_ring_buffer(struct drm_device *dev)
 	if (INTEL_GEN(dev_priv) >= 8) {
 		engine->irq_enable_mask =
 			GT_RENDER_USER_INTERRUPT << GEN8_BCS_IRQ_SHIFT;
-		engine->irq_get = gen8_ring_get_irq;
-		engine->irq_put = gen8_ring_put_irq;
+		engine->irq_enable = gen8_ring_enable_irq;
+		engine->irq_disable = gen8_ring_disable_irq;
 		engine->dispatch_execbuffer = gen8_ring_dispatch_execbuffer;
 		if (i915_semaphore_is_enabled(dev_priv)) {
 			engine->semaphore.sync_to = gen8_ring_sync;
@@ -3060,8 +2963,8 @@ int intel_init_blt_ring_buffer(struct drm_device *dev)
 		}
 	} else {
 		engine->irq_enable_mask = GT_BLT_USER_INTERRUPT;
-		engine->irq_get = gen6_ring_get_irq;
-		engine->irq_put = gen6_ring_put_irq;
+		engine->irq_enable = gen6_ring_enable_irq;
+		engine->irq_disable = gen6_ring_disable_irq;
 		engine->dispatch_execbuffer = gen6_ring_dispatch_execbuffer;
 		if (i915_semaphore_is_enabled(dev_priv)) {
 			engine->semaphore.signal = gen6_signal;
@@ -3109,8 +3012,8 @@ int intel_init_vebox_ring_buffer(struct drm_device *dev)
 	if (INTEL_GEN(dev_priv) >= 8) {
 		engine->irq_enable_mask =
 			GT_RENDER_USER_INTERRUPT << GEN8_VECS_IRQ_SHIFT;
-		engine->irq_get = gen8_ring_get_irq;
-		engine->irq_put = gen8_ring_put_irq;
+		engine->irq_enable = gen8_ring_enable_irq;
+		engine->irq_disable = gen8_ring_disable_irq;
 		engine->dispatch_execbuffer = gen8_ring_dispatch_execbuffer;
 		if (i915_semaphore_is_enabled(dev_priv)) {
 			engine->semaphore.sync_to = gen8_ring_sync;
@@ -3119,8 +3022,8 @@ int intel_init_vebox_ring_buffer(struct drm_device *dev)
 		}
 	} else {
 		engine->irq_enable_mask = PM_VEBOX_USER_INTERRUPT;
-		engine->irq_get = hsw_vebox_get_irq;
-		engine->irq_put = hsw_vebox_put_irq;
+		engine->irq_enable = hsw_vebox_enable_irq;
+		engine->irq_disable = hsw_vebox_disable_irq;
 		engine->dispatch_execbuffer = gen6_ring_dispatch_execbuffer;
 		if (i915_semaphore_is_enabled(dev_priv)) {
 			engine->semaphore.sync_to = gen6_ring_sync;
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index b55b9961d9f6..59e7234ab178 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -201,11 +201,10 @@ struct intel_engine_cs {
 	struct intel_hw_status_page status_page;
 	struct i915_ctx_workarounds wa_ctx;
 
-	unsigned irq_refcount; /* protected by dev_priv->irq_lock */
 	bool		irq_posted;
 	u32		irq_enable_mask;	/* bitmask to enable ring interrupt */
-	bool __must_check (*irq_get)(struct intel_engine_cs *ring);
-	void		(*irq_put)(struct intel_engine_cs *ring);
+	void		(*irq_enable)(struct intel_engine_cs *ring);
+	void		(*irq_disable)(struct intel_engine_cs *ring);
 
 	int		(*init_hw)(struct intel_engine_cs *ring);
 
-- 
2.8.1

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

Re: [PATCH 23/25] drm/i915: Move the get/put irq locking into the caller

[Tvrtko Ursulin]

On 25/06/16 11:13, Chris Wilson wrote:
> With only a single callsite for intel_engine_cs->irq_get and ->irq_put,
> we can reduce the code size by moving the common preamble into the
> caller, and we can also eliminate the reference counting.
>
> For completeness, as we are no longer doing reference counting on irq,
> rename the get/put vfunctions to enable/disable respectively and are
> able to review the use of posting reads. We only require the
> serialisation with hardware when enabling the interrupt (i.e. so we
> cannot miss an interrupt by going to sleep before the hardware truly
> enables it).
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_irq.c          |   8 +-
>   drivers/gpu/drm/i915/intel_breadcrumbs.c |  10 +-
>   drivers/gpu/drm/i915/intel_lrc.c         |  34 +---
>   drivers/gpu/drm/i915/intel_ringbuffer.c  | 269 ++++++++++---------------------
>   drivers/gpu/drm/i915/intel_ringbuffer.h  |   5 +-
>   5 files changed, 108 insertions(+), 218 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> index 7724bae27bcf..be25b7bdacfe 100644
> --- a/drivers/gpu/drm/i915/i915_irq.c
> +++ b/drivers/gpu/drm/i915/i915_irq.c
> @@ -259,12 +259,12 @@ static void ilk_update_gt_irq(struct drm_i915_private *dev_priv,
>   	dev_priv->gt_irq_mask &= ~interrupt_mask;
>   	dev_priv->gt_irq_mask |= (~enabled_irq_mask & interrupt_mask);
>   	I915_WRITE(GTIMR, dev_priv->gt_irq_mask);
> -	POSTING_READ(GTIMR);
>   }
>
>   void gen5_enable_gt_irq(struct drm_i915_private *dev_priv, uint32_t mask)
>   {
>   	ilk_update_gt_irq(dev_priv, mask, mask);
> +	POSTING_READ_FW(GTIMR);
>   }
>
>   void gen5_disable_gt_irq(struct drm_i915_private *dev_priv, uint32_t mask)
> @@ -2819,9 +2819,9 @@ ring_idle(struct intel_engine_cs *engine, u32 seqno)
>   }
>
>   static bool
> -ipehr_is_semaphore_wait(struct drm_i915_private *dev_priv, u32 ipehr)
> +ipehr_is_semaphore_wait(struct intel_engine_cs *engine, u32 ipehr)
>   {
> -	if (INTEL_GEN(dev_priv) >= 8) {
> +	if (INTEL_GEN(engine->i915) >= 8) {
>   		return (ipehr >> 23) == 0x1c;
>   	} else {
>   		ipehr &= ~MI_SEMAPHORE_SYNC_MASK;
> @@ -2892,7 +2892,7 @@ semaphore_waits_for(struct intel_engine_cs *engine, u32 *seqno)
>   		return NULL;
>
>   	ipehr = I915_READ(RING_IPEHR(engine->mmio_base));
> -	if (!ipehr_is_semaphore_wait(engine->i915, ipehr))
> +	if (!ipehr_is_semaphore_wait(engine, ipehr))
>   		return NULL;
>
>   	/*
> diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> index 661d6b415ac3..3c2e53acd187 100644
> --- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
> +++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
> @@ -50,12 +50,18 @@ static void irq_enable(struct intel_engine_cs *engine)
>   	 * just in case.
>   	 */
>   	engine->irq_posted = true;
> -	WARN_ON(!engine->irq_get(engine));
> +
> +	spin_lock_irq(&engine->i915->irq_lock);
> +	engine->irq_enable(engine);
> +	spin_unlock_irq(&engine->i915->irq_lock);
>   }
>
>   static void irq_disable(struct intel_engine_cs *engine)
>   {
> -	engine->irq_put(engine);
> +	spin_lock_irq(&engine->i915->irq_lock);
> +	engine->irq_disable(engine);
> +	spin_unlock_irq(&engine->i915->irq_lock);
> +
>   	engine->irq_posted = false;
>   }

Wouldn't it be better if locks stayed in the vfuncs? I can't figure out 
why would you want to pull that out like this.

>
> diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
> index 526c71975f11..e8fecf051e2c 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/intel_lrc.c
> @@ -1614,36 +1614,18 @@ static int gen8_emit_bb_start(struct drm_i915_gem_request *req,
>   	return 0;
>   }
>
> -static bool gen8_logical_ring_get_irq(struct intel_engine_cs *engine)
> +static void gen8_logical_ring_enable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
> -
> -	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
> -		return false;
> -
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (engine->irq_refcount++ == 0) {
> -		I915_WRITE_IMR(engine,
> -			       ~(engine->irq_enable_mask | engine->irq_keep_mask));
> -		POSTING_READ(RING_IMR(engine->mmio_base));
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> -
> -	return true;
> +	I915_WRITE_IMR(engine,
> +		       ~(engine->irq_enable_mask | engine->irq_keep_mask));
> +	POSTING_READ_FW(RING_IMR(engine->mmio_base));
>   }
>
> -static void gen8_logical_ring_put_irq(struct intel_engine_cs *engine)
> +static void gen8_logical_ring_disable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
> -
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (--engine->irq_refcount == 0) {
> -		I915_WRITE_IMR(engine, ~engine->irq_keep_mask);
> -		POSTING_READ(RING_IMR(engine->mmio_base));
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> +	I915_WRITE_IMR(engine, ~engine->irq_keep_mask);
>   }
>
>   static int gen8_emit_flush(struct drm_i915_gem_request *request,
> @@ -1963,8 +1945,8 @@ logical_ring_default_vfuncs(struct intel_engine_cs *engine)
>   	engine->init_hw = gen8_init_common_ring;
>   	engine->emit_request = gen8_emit_request;
>   	engine->emit_flush = gen8_emit_flush;
> -	engine->irq_get = gen8_logical_ring_get_irq;
> -	engine->irq_put = gen8_logical_ring_put_irq;
> +	engine->irq_enable = gen8_logical_ring_enable_irq;
> +	engine->irq_disable = gen8_logical_ring_disable_irq;
>   	engine->emit_bb_start = gen8_emit_bb_start;
>   	if (IS_BXT_REVID(engine->i915, 0, BXT_REVID_A1))
>   		engine->irq_seqno_barrier = bxt_a_seqno_barrier;
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
> index ad9b54071159..d7b5aae9cfd2 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> @@ -1632,103 +1632,54 @@ gen6_seqno_barrier(struct intel_engine_cs *engine)
>   	spin_unlock_irq(&dev_priv->uncore.lock);
>   }
>
> -static bool
> -gen5_ring_get_irq(struct intel_engine_cs *engine)
> +static void
> +gen5_ring_enable_irq(struct intel_engine_cs *engine)
>   {
> -	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
> -
> -	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
> -		return false;
> -
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (engine->irq_refcount++ == 0)
> -		gen5_enable_gt_irq(dev_priv, engine->irq_enable_mask);
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> -
> -	return true;
> +	gen5_enable_gt_irq(engine->i915, engine->irq_enable_mask);
>   }
>
>   static void
> -gen5_ring_put_irq(struct intel_engine_cs *engine)
> +gen5_ring_disable_irq(struct intel_engine_cs *engine)
>   {
> -	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
> -
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (--engine->irq_refcount == 0)
> -		gen5_disable_gt_irq(dev_priv, engine->irq_enable_mask);
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> +	gen5_disable_gt_irq(engine->i915, engine->irq_enable_mask);
>   }
>
> -static bool
> -i9xx_ring_get_irq(struct intel_engine_cs *engine)
> +static void
> +i9xx_ring_enable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
> -
> -	if (!intel_irqs_enabled(dev_priv))
> -		return false;
> -
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (engine->irq_refcount++ == 0) {
> -		dev_priv->irq_mask &= ~engine->irq_enable_mask;
> -		I915_WRITE(IMR, dev_priv->irq_mask);
> -		POSTING_READ(IMR);
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
>
> -	return true;
> +	dev_priv->irq_mask &= ~engine->irq_enable_mask;
> +	I915_WRITE(IMR, dev_priv->irq_mask);
> +	POSTING_READ_FW(RING_IMR(engine->mmio_base));
>   }
>
>   static void
> -i9xx_ring_put_irq(struct intel_engine_cs *engine)
> +i9xx_ring_disable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
>
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (--engine->irq_refcount == 0) {
> -		dev_priv->irq_mask |= engine->irq_enable_mask;
> -		I915_WRITE(IMR, dev_priv->irq_mask);
> -		POSTING_READ(IMR);
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> +	dev_priv->irq_mask |= engine->irq_enable_mask;
> +	I915_WRITE(IMR, dev_priv->irq_mask);
>   }
>
> -static bool
> -i8xx_ring_get_irq(struct intel_engine_cs *engine)
> +static void
> +i8xx_ring_enable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
>
> -	if (!intel_irqs_enabled(dev_priv))
> -		return false;
> -
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (engine->irq_refcount++ == 0) {
> -		dev_priv->irq_mask &= ~engine->irq_enable_mask;
> -		I915_WRITE16(IMR, dev_priv->irq_mask);
> -		POSTING_READ16(IMR);
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> -
> -	return true;
> +	dev_priv->irq_mask &= ~engine->irq_enable_mask;
> +	I915_WRITE16(IMR, dev_priv->irq_mask);
> +	POSTING_READ16(RING_IMR(engine->mmio_base));
>   }
>
>   static void
> -i8xx_ring_put_irq(struct intel_engine_cs *engine)
> +i8xx_ring_disable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
>
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (--engine->irq_refcount == 0) {
> -		dev_priv->irq_mask |= engine->irq_enable_mask;
> -		I915_WRITE16(IMR, dev_priv->irq_mask);
> -		POSTING_READ16(IMR);
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> +	dev_priv->irq_mask |= engine->irq_enable_mask;
> +	I915_WRITE16(IMR, dev_priv->irq_mask);
>   }
>
>   static int
> @@ -1769,122 +1720,74 @@ i9xx_add_request(struct drm_i915_gem_request *req)
>   	return 0;
>   }
>
> -static bool
> -gen6_ring_get_irq(struct intel_engine_cs *engine)
> +static void
> +gen6_ring_enable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
> -
> -	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
> -		return false;
>
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (engine->irq_refcount++ == 0) {
> -		if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
> -			I915_WRITE_IMR(engine,
> -				       ~(engine->irq_enable_mask |
> -					 GT_PARITY_ERROR(dev_priv)));
> -		else
> -			I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
> -		gen5_enable_gt_irq(dev_priv, engine->irq_enable_mask);
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> -
> -	return true;
> +	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
> +		I915_WRITE_IMR(engine,
> +			       ~(engine->irq_enable_mask |
> +				 GT_PARITY_ERROR(dev_priv)));
> +	else
> +		I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
> +	gen5_enable_gt_irq(dev_priv, engine->irq_enable_mask);
>   }
>
>   static void
> -gen6_ring_put_irq(struct intel_engine_cs *engine)
> +gen6_ring_disable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
>
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (--engine->irq_refcount == 0) {
> -		if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
> -			I915_WRITE_IMR(engine, ~GT_PARITY_ERROR(dev_priv));
> -		else
> -			I915_WRITE_IMR(engine, ~0);
> -		gen5_disable_gt_irq(dev_priv, engine->irq_enable_mask);
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> +	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
> +		I915_WRITE_IMR(engine, ~GT_PARITY_ERROR(dev_priv));
> +	else
> +		I915_WRITE_IMR(engine, ~0);
> +	gen5_disable_gt_irq(dev_priv, engine->irq_enable_mask);
>   }
>
> -static bool
> -hsw_vebox_get_irq(struct intel_engine_cs *engine)
> +static void
> +hsw_vebox_enable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
> -
> -	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
> -		return false;
>
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (engine->irq_refcount++ == 0) {
> -		I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
> -		gen6_enable_pm_irq(dev_priv, engine->irq_enable_mask);
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> -
> -	return true;
> +	I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
> +	gen6_enable_pm_irq(dev_priv, engine->irq_enable_mask);
>   }
>
>   static void
> -hsw_vebox_put_irq(struct intel_engine_cs *engine)
> +hsw_vebox_disable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
>
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (--engine->irq_refcount == 0) {
> -		I915_WRITE_IMR(engine, ~0);
> -		gen6_disable_pm_irq(dev_priv, engine->irq_enable_mask);
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> +	I915_WRITE_IMR(engine, ~0);
> +	gen6_disable_pm_irq(dev_priv, engine->irq_enable_mask);
>   }
>
> -static bool
> -gen8_ring_get_irq(struct intel_engine_cs *engine)
> +static void
> +gen8_ring_enable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
>
> -	if (WARN_ON(!intel_irqs_enabled(dev_priv)))
> -		return false;
> -
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (engine->irq_refcount++ == 0) {
> -		if (HAS_L3_DPF(dev_priv) && engine->id == RCS) {
> -			I915_WRITE_IMR(engine,
> -				       ~(engine->irq_enable_mask |
> -					 GT_RENDER_L3_PARITY_ERROR_INTERRUPT));
> -		} else {
> -			I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
> -		}
> -		POSTING_READ(RING_IMR(engine->mmio_base));
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> -
> -	return true;
> +	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
> +		I915_WRITE_IMR(engine,
> +			       ~(engine->irq_enable_mask |
> +				 GT_RENDER_L3_PARITY_ERROR_INTERRUPT));
> +	else
> +		I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
> +	POSTING_READ_FW(RING_IMR(engine->mmio_base));
>   }
>
>   static void
> -gen8_ring_put_irq(struct intel_engine_cs *engine)
> +gen8_ring_disable_irq(struct intel_engine_cs *engine)
>   {
>   	struct drm_i915_private *dev_priv = engine->i915;
> -	unsigned long flags;
>
> -	spin_lock_irqsave(&dev_priv->irq_lock, flags);
> -	if (--engine->irq_refcount == 0) {
> -		if (HAS_L3_DPF(dev_priv) && engine->id == RCS) {
> -			I915_WRITE_IMR(engine,
> -				       ~GT_RENDER_L3_PARITY_ERROR_INTERRUPT);
> -		} else {
> -			I915_WRITE_IMR(engine, ~0);
> -		}
> -		POSTING_READ(RING_IMR(engine->mmio_base));
> -	}
> -	spin_unlock_irqrestore(&dev_priv->irq_lock, flags);
> +	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
> +		I915_WRITE_IMR(engine,
> +			       ~GT_RENDER_L3_PARITY_ERROR_INTERRUPT);
> +	else
> +		I915_WRITE_IMR(engine, ~0);
>   }
>
>   static int
> @@ -2829,8 +2732,8 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
>   		engine->init_context = intel_rcs_ctx_init;
>   		engine->add_request = gen8_render_add_request;
>   		engine->flush = gen8_render_ring_flush;
> -		engine->irq_get = gen8_ring_get_irq;
> -		engine->irq_put = gen8_ring_put_irq;
> +		engine->irq_enable = gen8_ring_enable_irq;
> +		engine->irq_disable = gen8_ring_disable_irq;
>   		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT;
>   		if (i915_semaphore_is_enabled(dev_priv)) {
>   			WARN_ON(!dev_priv->semaphore_obj);
> @@ -2844,8 +2747,8 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
>   		engine->flush = gen7_render_ring_flush;
>   		if (IS_GEN6(dev_priv))
>   			engine->flush = gen6_render_ring_flush;
> -		engine->irq_get = gen6_ring_get_irq;
> -		engine->irq_put = gen6_ring_put_irq;
> +		engine->irq_enable = gen6_ring_enable_irq;
> +		engine->irq_disable = gen6_ring_disable_irq;
>   		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT;
>   		engine->irq_seqno_barrier = gen6_seqno_barrier;
>   		if (i915_semaphore_is_enabled(dev_priv)) {
> @@ -2872,8 +2775,8 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
>   	} else if (IS_GEN5(dev_priv)) {
>   		engine->add_request = i9xx_add_request;
>   		engine->flush = gen4_render_ring_flush;
> -		engine->irq_get = gen5_ring_get_irq;
> -		engine->irq_put = gen5_ring_put_irq;
> +		engine->irq_enable = gen5_ring_enable_irq;
> +		engine->irq_disable = gen5_ring_disable_irq;
>   		engine->irq_seqno_barrier = gen5_seqno_barrier;
>   		engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT;
>   	} else {
> @@ -2883,11 +2786,11 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
>   		else
>   			engine->flush = gen4_render_ring_flush;
>   		if (IS_GEN2(dev_priv)) {
> -			engine->irq_get = i8xx_ring_get_irq;
> -			engine->irq_put = i8xx_ring_put_irq;
> +			engine->irq_enable = i8xx_ring_enable_irq;
> +			engine->irq_disable = i8xx_ring_disable_irq;
>   		} else {
> -			engine->irq_get = i9xx_ring_get_irq;
> -			engine->irq_put = i9xx_ring_put_irq;
> +			engine->irq_enable = i9xx_ring_enable_irq;
> +			engine->irq_disable = i9xx_ring_disable_irq;
>   		}
>   		engine->irq_enable_mask = I915_USER_INTERRUPT;
>   	}
> @@ -2947,8 +2850,8 @@ int intel_init_bsd_ring_buffer(struct drm_device *dev)
>   		if (INTEL_GEN(dev_priv) >= 8) {
>   			engine->irq_enable_mask =
>   				GT_RENDER_USER_INTERRUPT << GEN8_VCS1_IRQ_SHIFT;
> -			engine->irq_get = gen8_ring_get_irq;
> -			engine->irq_put = gen8_ring_put_irq;
> +			engine->irq_enable = gen8_ring_enable_irq;
> +			engine->irq_disable = gen8_ring_disable_irq;
>   			engine->dispatch_execbuffer =
>   				gen8_ring_dispatch_execbuffer;
>   			if (i915_semaphore_is_enabled(dev_priv)) {
> @@ -2958,8 +2861,8 @@ int intel_init_bsd_ring_buffer(struct drm_device *dev)
>   			}
>   		} else {
>   			engine->irq_enable_mask = GT_BSD_USER_INTERRUPT;
> -			engine->irq_get = gen6_ring_get_irq;
> -			engine->irq_put = gen6_ring_put_irq;
> +			engine->irq_enable = gen6_ring_enable_irq;
> +			engine->irq_disable = gen6_ring_disable_irq;
>   			engine->dispatch_execbuffer =
>   				gen6_ring_dispatch_execbuffer;
>   			if (i915_semaphore_is_enabled(dev_priv)) {
> @@ -2983,13 +2886,13 @@ int intel_init_bsd_ring_buffer(struct drm_device *dev)
>   		engine->add_request = i9xx_add_request;
>   		if (IS_GEN5(dev_priv)) {
>   			engine->irq_enable_mask = ILK_BSD_USER_INTERRUPT;
> -			engine->irq_get = gen5_ring_get_irq;
> -			engine->irq_put = gen5_ring_put_irq;
> +			engine->irq_enable = gen5_ring_enable_irq;
> +			engine->irq_disable = gen5_ring_disable_irq;
>   			engine->irq_seqno_barrier = gen5_seqno_barrier;
>   		} else {
>   			engine->irq_enable_mask = I915_BSD_USER_INTERRUPT;
> -			engine->irq_get = i9xx_ring_get_irq;
> -			engine->irq_put = i9xx_ring_put_irq;
> +			engine->irq_enable = i9xx_ring_enable_irq;
> +			engine->irq_disable = i9xx_ring_disable_irq;
>   		}
>   		engine->dispatch_execbuffer = i965_dispatch_execbuffer;
>   	}
> @@ -3018,8 +2921,8 @@ int intel_init_bsd2_ring_buffer(struct drm_device *dev)
>   	engine->irq_seqno_barrier = gen6_seqno_barrier;
>   	engine->irq_enable_mask =
>   			GT_RENDER_USER_INTERRUPT << GEN8_VCS2_IRQ_SHIFT;
> -	engine->irq_get = gen8_ring_get_irq;
> -	engine->irq_put = gen8_ring_put_irq;
> +	engine->irq_enable = gen8_ring_enable_irq;
> +	engine->irq_disable = gen8_ring_disable_irq;
>   	engine->dispatch_execbuffer =
>   			gen8_ring_dispatch_execbuffer;
>   	if (i915_semaphore_is_enabled(dev_priv)) {
> @@ -3050,8 +2953,8 @@ int intel_init_blt_ring_buffer(struct drm_device *dev)
>   	if (INTEL_GEN(dev_priv) >= 8) {
>   		engine->irq_enable_mask =
>   			GT_RENDER_USER_INTERRUPT << GEN8_BCS_IRQ_SHIFT;
> -		engine->irq_get = gen8_ring_get_irq;
> -		engine->irq_put = gen8_ring_put_irq;
> +		engine->irq_enable = gen8_ring_enable_irq;
> +		engine->irq_disable = gen8_ring_disable_irq;
>   		engine->dispatch_execbuffer = gen8_ring_dispatch_execbuffer;
>   		if (i915_semaphore_is_enabled(dev_priv)) {
>   			engine->semaphore.sync_to = gen8_ring_sync;
> @@ -3060,8 +2963,8 @@ int intel_init_blt_ring_buffer(struct drm_device *dev)
>   		}
>   	} else {
>   		engine->irq_enable_mask = GT_BLT_USER_INTERRUPT;
> -		engine->irq_get = gen6_ring_get_irq;
> -		engine->irq_put = gen6_ring_put_irq;
> +		engine->irq_enable = gen6_ring_enable_irq;
> +		engine->irq_disable = gen6_ring_disable_irq;
>   		engine->dispatch_execbuffer = gen6_ring_dispatch_execbuffer;
>   		if (i915_semaphore_is_enabled(dev_priv)) {
>   			engine->semaphore.signal = gen6_signal;
> @@ -3109,8 +3012,8 @@ int intel_init_vebox_ring_buffer(struct drm_device *dev)
>   	if (INTEL_GEN(dev_priv) >= 8) {
>   		engine->irq_enable_mask =
>   			GT_RENDER_USER_INTERRUPT << GEN8_VECS_IRQ_SHIFT;
> -		engine->irq_get = gen8_ring_get_irq;
> -		engine->irq_put = gen8_ring_put_irq;
> +		engine->irq_enable = gen8_ring_enable_irq;
> +		engine->irq_disable = gen8_ring_disable_irq;
>   		engine->dispatch_execbuffer = gen8_ring_dispatch_execbuffer;
>   		if (i915_semaphore_is_enabled(dev_priv)) {
>   			engine->semaphore.sync_to = gen8_ring_sync;
> @@ -3119,8 +3022,8 @@ int intel_init_vebox_ring_buffer(struct drm_device *dev)
>   		}
>   	} else {
>   		engine->irq_enable_mask = PM_VEBOX_USER_INTERRUPT;
> -		engine->irq_get = hsw_vebox_get_irq;
> -		engine->irq_put = hsw_vebox_put_irq;
> +		engine->irq_enable = hsw_vebox_enable_irq;
> +		engine->irq_disable = hsw_vebox_disable_irq;
>   		engine->dispatch_execbuffer = gen6_ring_dispatch_execbuffer;
>   		if (i915_semaphore_is_enabled(dev_priv)) {
>   			engine->semaphore.sync_to = gen6_ring_sync;
> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
> index b55b9961d9f6..59e7234ab178 100644
> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> @@ -201,11 +201,10 @@ struct intel_engine_cs {
>   	struct intel_hw_status_page status_page;
>   	struct i915_ctx_workarounds wa_ctx;
>
> -	unsigned irq_refcount; /* protected by dev_priv->irq_lock */
>   	bool		irq_posted;
>   	u32		irq_enable_mask;	/* bitmask to enable ring interrupt */
> -	bool __must_check (*irq_get)(struct intel_engine_cs *ring);
> -	void		(*irq_put)(struct intel_engine_cs *ring);
> +	void		(*irq_enable)(struct intel_engine_cs *ring);
> +	void		(*irq_disable)(struct intel_engine_cs *ring);
>
>   	int		(*init_hw)(struct intel_engine_cs *ring);
>
>

The rest looks OK to me.

Regards,

Tvrtko

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

Re: [PATCH 23/25] drm/i915: Move the get/put irq locking into the caller

[Chris Wilson]

On Mon, Jun 27, 2016 at 01:11:38PM +0100, Tvrtko Ursulin wrote:
> >  static void irq_disable(struct intel_engine_cs *engine)
> >  {
> >-	engine->irq_put(engine);
> >+	spin_lock_irq(&engine->i915->irq_lock);
> >+	engine->irq_disable(engine);
> >+	spin_unlock_irq(&engine->i915->irq_lock);
> >+
> >  	engine->irq_posted = false;
> >  }
> 
> Wouldn't it be better if locks stayed in the vfuncs? I can't figure
> out why would you want to pull that out like this.

Because removing the recfounting and locking to the single place where
it is required is simpler.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 24/25] drm/i915: Simplify enabling user-interrupts with L3-remapping

[Chris Wilson]

Borrow the idea from intel_lrc.c to precompute the mask of interrupts we
wish to always enable to avoid having lots of conditionals inside the
interrupt enabling.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/intel_ringbuffer.c | 35 +++++++++++----------------------
 drivers/gpu/drm/i915/intel_ringbuffer.h |  4 ++--
 2 files changed, 14 insertions(+), 25 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index d7b5aae9cfd2..d570625189ee 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -1308,8 +1308,7 @@ static int init_render_ring(struct intel_engine_cs *engine)
 	if (IS_GEN(dev_priv, 6, 7))
 		I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_FORCE_ORDERING));
 
-	if (HAS_L3_DPF(dev_priv))
-		I915_WRITE_IMR(engine, ~GT_PARITY_ERROR(dev_priv));
+	I915_WRITE_IMR(engine, ~engine->irq_keep_mask);
 
 	return init_workarounds_ring(engine);
 }
@@ -1725,12 +1724,9 @@ gen6_ring_enable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
 
-	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
-		I915_WRITE_IMR(engine,
-			       ~(engine->irq_enable_mask |
-				 GT_PARITY_ERROR(dev_priv)));
-	else
-		I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
+	I915_WRITE_IMR(engine,
+		       ~(engine->irq_enable_mask |
+			 engine->irq_keep_mask));
 	gen5_enable_gt_irq(dev_priv, engine->irq_enable_mask);
 }
 
@@ -1739,10 +1735,7 @@ gen6_ring_disable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
 
-	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
-		I915_WRITE_IMR(engine, ~GT_PARITY_ERROR(dev_priv));
-	else
-		I915_WRITE_IMR(engine, ~0);
+	I915_WRITE_IMR(engine, ~engine->irq_keep_mask);
 	gen5_disable_gt_irq(dev_priv, engine->irq_enable_mask);
 }
 
@@ -1769,12 +1762,9 @@ gen8_ring_enable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
 
-	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
-		I915_WRITE_IMR(engine,
-			       ~(engine->irq_enable_mask |
-				 GT_RENDER_L3_PARITY_ERROR_INTERRUPT));
-	else
-		I915_WRITE_IMR(engine, ~engine->irq_enable_mask);
+	I915_WRITE_IMR(engine,
+		       ~(engine->irq_enable_mask |
+			 engine->irq_keep_mask));
 	POSTING_READ_FW(RING_IMR(engine->mmio_base));
 }
 
@@ -1783,11 +1773,7 @@ gen8_ring_disable_irq(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *dev_priv = engine->i915;
 
-	if (HAS_L3_DPF(dev_priv) && engine->id == RCS)
-		I915_WRITE_IMR(engine,
-			       ~GT_RENDER_L3_PARITY_ERROR_INTERRUPT);
-	else
-		I915_WRITE_IMR(engine, ~0);
+	I915_WRITE_IMR(engine, ~engine->irq_keep_mask);
 }
 
 static int
@@ -2711,6 +2697,9 @@ int intel_init_render_ring_buffer(struct drm_device *dev)
 	engine->hw_id = 0;
 	engine->mmio_base = RENDER_RING_BASE;
 
+	if (HAS_L3_DPF(dev_priv))
+		engine->irq_keep_mask = GT_RENDER_L3_PARITY_ERROR_INTERRUPT;
+
 	if (INTEL_GEN(dev_priv) >= 8) {
 		if (i915_semaphore_is_enabled(dev_priv)) {
 			obj = i915_gem_object_create(dev, 4096);
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 59e7234ab178..d804771e9110 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -202,7 +202,8 @@ struct intel_engine_cs {
 	struct i915_ctx_workarounds wa_ctx;
 
 	bool		irq_posted;
-	u32		irq_enable_mask;	/* bitmask to enable ring interrupt */
+	u32             irq_keep_mask; /* bitmask for interrupts that should not be masked */
+	u32		irq_enable_mask;/* bitmask to enable ring interrupt */
 	void		(*irq_enable)(struct intel_engine_cs *ring);
 	void		(*irq_disable)(struct intel_engine_cs *ring);
 
@@ -299,7 +300,6 @@ struct intel_engine_cs {
 	unsigned int idle_lite_restore_wa;
 	bool disable_lite_restore_wa;
 	u32 ctx_desc_template;
-	u32             irq_keep_mask; /* bitmask for interrupts that should not be masked */
 	int		(*emit_request)(struct drm_i915_gem_request *request);
 	int		(*emit_flush)(struct drm_i915_gem_request *request,
 				      u32 invalidate_domains,
-- 
2.8.1

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

[PATCH 25/25] drm/i915: Remove debug noise on detecting fault-injection of missed interrupts

[Chris Wilson]

Since the tests can and do explicitly check debugfs/i915_ring_missed_irqs
for the handling of a "missed interrupt", adding it to the dmesg at INFO
is just noise. When it happens for real, we still class it as an ERROR.

Note that I have chose to remove it entirely because when we detect the
"missed interrupt" is irrelevant and the message contains no more
information than we glean from looking in debugfs.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_irq.c | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index be25b7bdacfe..53d4c8088f28 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -3072,9 +3072,6 @@ static unsigned kick_waiters(struct intel_engine_cs *engine)
 		if (!test_bit(engine->id, &i915->gpu_error.test_irq_rings))
 			DRM_ERROR("Hangcheck timer elapsed... %s idle\n",
 				  engine->name);
-		else
-			DRM_INFO("Fake missed irq on %s\n",
-				 engine->name);
 
 		intel_engine_enable_fake_irq(engine);
 	}
-- 
2.8.1

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

✗ Ro.CI.BAT: warning for series starting with [01/25] drm/i915: Preserve current RPS frequency across init

[Patchwork]

== Series Details ==

Series: series starting with [01/25] drm/i915: Preserve current RPS frequency across init
URL   : https://patchwork.freedesktop.org/series/9161/
State : warning

== Summary ==

Series 9161v1 Series without cover letter
http://patchwork.freedesktop.org/api/1.0/series/9161/revisions/1/mbox

Test drv_module_reload_basic:
                dmesg-warn -> PASS       (ro-byt-n2820)
Test gem_exec_flush:
        Subgroup basic-batch-kernel-default-cmd:
                fail       -> PASS       (ro-byt-n2820)
        Subgroup basic-uc-pro-default:
                fail       -> PASS       (ro-byt-n2820)
Test gem_ringfill:
        Subgroup basic-default-forked:
                pass       -> DMESG-WARN (ro-byt-n2820)
Test kms_pipe_crc_basic:
        Subgroup suspend-read-crc-pipe-a:
                dmesg-warn -> SKIP       (ro-bdw-i5-5250u)
        Subgroup suspend-read-crc-pipe-c:
                dmesg-warn -> SKIP       (ro-bdw-i5-5250u)
Test kms_setmode:
        Subgroup basic-clone-single-crtc:
                pass       -> DMESG-WARN (ro-byt-n2820)

fi-hsw-i7-4770k  total:229  pass:194  dwarn:0   dfail:0   fail:2   skip:33 
fi-skl-i5-6260u  total:229  pass:202  dwarn:0   dfail:0   fail:2   skip:25 
fi-skl-i7-6700k  total:229  pass:188  dwarn:0   dfail:0   fail:2   skip:39 
ro-bdw-i5-5250u  total:227  pass:201  dwarn:2   dfail:1   fail:1   skip:22 
ro-bdw-i7-5557U  total:227  pass:201  dwarn:3   dfail:2   fail:1   skip:20 
ro-bdw-i7-5600u  total:227  pass:189  dwarn:0   dfail:1   fail:0   skip:37 
ro-bsw-n3050     total:227  pass:177  dwarn:0   dfail:1   fail:1   skip:48 
ro-byt-n2820     total:227  pass:177  dwarn:2   dfail:1   fail:2   skip:45 
ro-hsw-i3-4010u  total:227  pass:194  dwarn:0   dfail:1   fail:1   skip:31 
ro-hsw-i7-4770r  total:227  pass:194  dwarn:0   dfail:1   fail:1   skip:31 
ro-ilk-i7-620lm  total:227  pass:154  dwarn:0   dfail:1   fail:2   skip:70 
ro-ilk1-i5-650   total:222  pass:154  dwarn:0   dfail:1   fail:2   skip:65 
ro-ivb-i7-3770   total:227  pass:185  dwarn:0   dfail:1   fail:1   skip:40 
ro-ivb2-i7-3770  total:227  pass:189  dwarn:0   dfail:1   fail:1   skip:36 
ro-skl3-i5-6260u total:227  pass:205  dwarn:1   dfail:1   fail:1   skip:19 
ro-snb-i7-2620M  total:227  pass:178  dwarn:0   dfail:1   fail:1   skip:47 

Results at /archive/results/CI_IGT_test/RO_Patchwork_1308/

5c244f4 drm-intel-nightly: 2016y-06m-24d-15h-17m-32s UTC integration manifest
fd30c87 drm/i915: Remove debug noise on detecting fault-injection of missed interrupts
65c49ad drm/i915: Simplify enabling user-interrupts with L3-remapping
8f020ca drm/i915: Move the get/put irq locking into the caller
d186bfb drm/i915: Embed signaling node into the GEM request
3ab50f4 drm/i915: Convert trace-irq to the breadcrumb waiter
0c55bd2 drm/i915: Only query timestamp when measuring elapsed time
b977ad5 drm/i915: Stop setting wraparound seqno on initialisation
5eddd9e drm/i915: Only apply one barrier after a breadcrumb interrupt is posted
614716f drm/i915: Check the CPU cached value in HWS of seqno after waking the waiter
13fd8c0 drm/i915: Add a delay between interrupt and inspecting the final seqno (ilk)
42f838d drm/i915: Refactor scratch object allocation for gen2 w/a buffer
8934d46 drm/i915: Allocate scratch page from stolen
67a8b4a drm/i915: Stop mapping the scratch page into CPU space
0794a65 drm/i915: Use HWS for seqno tracking everywhere
e44268f drm/i915: Spin after waking up for an interrupt
ead664a drm/i915: Slaughter the thundering i915_wait_request herd
4669707 drm/i915: Separate GPU hang waitqueue from advance
6f0e5b3 drm/i915: Make queueing the hangcheck work inline
d73cfe9 drm/i915: Remove the dedicated hangcheck workqueue
69ce6c7 drm/i915: Delay queuing hangcheck to wait-request
8067d88 drm/i915/shrinker: Flush active on objects before counting
c57a016 drm: Restore double clflush on the last partial cacheline
b61d2d6 drm/i915: Defer enabling rc6 til after we submit the first batch/context
440bbf8 drm/i915: Remove superfluous powersave work flushing
3bcef74 drm/i915: Preserve current RPS frequency across init

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