All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH] drm/i915: Defer removing fence register tracking to rpm wakeup
@ 2019-02-06 15:57 Chris Wilson
  2019-02-06 17:09 ` ✓ Fi.CI.BAT: success for " Patchwork
  2019-02-06 18:47 ` ✓ Fi.CI.IGT: " Patchwork
  0 siblings, 2 replies; 3+ messages in thread
From: Chris Wilson @ 2019-02-06 15:57 UTC (permalink / raw)
  To: intel-gfx

Currently, we may simultaneously release the fence register from both
fence_update() and i915_gem_restore_fences(). This is dangerous, so
defer the bookkeeping entirely to i915_gem_restore_fences() when the
device is asleep.

Reported-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
---
 drivers/gpu/drm/i915/i915_gem_fence_reg.c | 27 +++++++++++++++--------
 1 file changed, 18 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_fence_reg.c b/drivers/gpu/drm/i915/i915_gem_fence_reg.c
index e037e94792f3..4331dadc7000 100644
--- a/drivers/gpu/drm/i915/i915_gem_fence_reg.c
+++ b/drivers/gpu/drm/i915/i915_gem_fence_reg.c
@@ -240,6 +240,22 @@ static int fence_update(struct drm_i915_fence_reg *fence,
 		i915_vma_flush_writes(old);
 	}
 
+	/*
+	 * We only need to update the register itself if the device is awake.
+	 * If the device is currently powered down, we will defer the write
+	 * to the runtime resume, see i915_gem_restore_fences().
+	 *
+	 * This only works for removing the fence register, on acquisition
+	 * the caller must hold the rpm wakeref. The fence register must
+	 * be cleared before we can use any other fences to ensure that
+	 * the new fences do not overlap the elided clears, confusing HW.
+	 */
+	wakeref = intel_runtime_pm_get_if_in_use(fence->i915);
+	if (!wakeref) {
+		GEM_BUG_ON(vma);
+		return 0;
+	}
+
 	if (fence->vma && fence->vma != vma) {
 		/* Ensure that all userspace CPU access is completed before
 		 * stealing the fence.
@@ -253,15 +269,7 @@ static int fence_update(struct drm_i915_fence_reg *fence,
 		list_move(&fence->link, &fence->i915->mm.fence_list);
 	}
 
-	/* We only need to update the register itself if the device is awake.
-	 * If the device is currently powered down, we will defer the write
-	 * to the runtime resume, see i915_gem_restore_fences().
-	 */
-	wakeref = intel_runtime_pm_get_if_in_use(fence->i915);
-	if (wakeref) {
-		fence_write(fence, vma);
-		intel_runtime_pm_put(fence->i915, wakeref);
-	}
+	fence_write(fence, vma);
 
 	if (vma) {
 		if (fence->vma != vma) {
@@ -272,6 +280,7 @@ static int fence_update(struct drm_i915_fence_reg *fence,
 		list_move_tail(&fence->link, &fence->i915->mm.fence_list);
 	}
 
+	intel_runtime_pm_put(fence->i915, wakeref);
 	return 0;
 }
 
-- 
2.20.1

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

^ permalink raw reply related	[flat|nested] 3+ messages in thread
end of thread, other threads:[~2019-02-06 18:47 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-02-06 15:57 [PATCH] drm/i915: Defer removing fence register tracking to rpm wakeup Chris Wilson
2019-02-06 17:09 ` ✓ Fi.CI.BAT: success for " Patchwork
2019-02-06 18:47 ` ✓ Fi.CI.IGT: " Patchwork

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.