diff options
author | Chris Wilson | 2016-10-24 14:42:18 +0200 |
---|---|---|
committer | Chris Wilson | 2016-10-24 14:45:38 +0200 |
commit | 7c108fd8feac9a4f380c22444ff8527e8d25bc75 (patch) | |
tree | 3be7c7d7b51c3820c86fb669554f1d1e5da91d7d /drivers/gpu/drm/i915/i915_gem_fence.c | |
parent | drm/i915: Remove RPM sequence checking (diff) | |
download | kernel-qcow2-linux-7c108fd8feac9a4f380c22444ff8527e8d25bc75.tar.gz kernel-qcow2-linux-7c108fd8feac9a4f380c22444ff8527e8d25bc75.tar.xz kernel-qcow2-linux-7c108fd8feac9a4f380c22444ff8527e8d25bc75.zip |
drm/i915: Move fence cancellation to runtime suspend
At the moment, we have dependency on the RPM as a barrier itself in both
i915_gem_release_all_mmaps() and i915_gem_restore_fences().
i915_gem_restore_fences() is also called along !runtime pm paths, but we
can move the markup of lost fences alongside releasing the mmaps into a
common i915_gem_runtime_suspend(). This has the advantage of locating
all the tricky barrier dependencies into one location.
v2: Just mark the fence as invalid (fence->dirty) so that upon waking we
will be sure to clear the fence after use, or restore it to the correct
value before use. This makes sure that if the fence is left intact
across the sleep, we do not leave it pointing to a region of GTT for the
next unsuspecting user.
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Imre Deak <imre.deak@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161024124218.18252-5-chris@chris-wilson.co.uk
Diffstat (limited to 'drivers/gpu/drm/i915/i915_gem_fence.c')
-rw-r--r-- | drivers/gpu/drm/i915/i915_gem_fence.c | 12 |
1 files changed, 5 insertions, 7 deletions
diff --git a/drivers/gpu/drm/i915/i915_gem_fence.c b/drivers/gpu/drm/i915/i915_gem_fence.c index 67013179b8ed..3c5a8082cac3 100644 --- a/drivers/gpu/drm/i915/i915_gem_fence.c +++ b/drivers/gpu/drm/i915/i915_gem_fence.c @@ -343,6 +343,9 @@ i915_vma_get_fence(struct i915_vma *vma) struct drm_i915_fence_reg *fence; struct i915_vma *set = i915_gem_object_is_tiled(vma->obj) ? vma : NULL; + /* Note that we revoke fences on runtime suspend. Therefore the user + * must keep the device awake whilst using the fence. + */ assert_rpm_wakelock_held(to_i915(vma->vm->dev)); /* Just update our place in the LRU if our fence is getting reused. */ @@ -368,19 +371,14 @@ i915_vma_get_fence(struct i915_vma *vma) * @dev: DRM device * * Restore the hw fence state to match the software tracking again, to be called - * after a gpu reset and on resume. + * after a gpu reset and on resume. Note that on runtime suspend we only cancel + * the fences, to be reacquired by the user later. */ void i915_gem_restore_fences(struct drm_device *dev) { struct drm_i915_private *dev_priv = to_i915(dev); int i; - /* Note that this may be called outside of struct_mutex, by - * runtime suspend/resume. The barrier we require is enforced by - * rpm itself - all access to fences/GTT are only within an rpm - * wakeref, and to acquire that wakeref you must pass through here. - */ - for (i = 0; i < dev_priv->num_fence_regs; i++) { struct drm_i915_fence_reg *reg = &dev_priv->fence_regs[i]; struct i915_vma *vma = reg->vma; |