drm/i915: use wake_up instead of drm wrapper.
Subsystems (2)
Dave Airlie
maintainerPATCH 1/1 drm/i915: use wake_up instead of drm wrapper.
From: Dave Airlie <[email protected]>
Sinceb9241ea31fae4887104e5d1b3b18f4009c25a0c4 drm/i915: Don't wait interruptible for possible plane buffer flush and previous to that48764bf43f746113fc77877d7e80f2df23ca4cbb drm/i915: add i915_lp_ring_sync helper
The irq wait queue has uninterruptible waiters, however the DRM_WAKEUP macro wraps wake_up_interruptible which isn't going to do what you want in this case, just use the normal wake_up variant which does both types fine.
This fixes a regression in the copy fb code we use to get seamless startup on Fedora/RHEL, without this fix, we block forever in the wait when starting X.
--- drivers/gpu/drm/i915/i915_debugfs.c | 2 +- drivers/gpu/drm/i915/i915_irq.c | 6 +++--- 2 files changed, 4 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_debugfs.c +1 -1
@@ -673,7 +673,7 @@ i915_wedged_write(struct file *filp,atomic_set (&dev_priv->mm.wedged, val);if (val) { -DRM_WAKEUP (&dev_priv->irq_queue); +wake_up (&dev_priv->irq_queue);queue_work (dev_priv->wq, &dev_priv->error_work); }
drivers/gpu/drm/i915/i915_irq.c +3 -3
@@ -357,7 +357,7 @@ irqreturn_t ironlake_irq_handler(struct drm_device *dev)u32 seqno =i915_get_gem_seqno (dev); dev_priv->mm.irq_gem_seqno = seqno;trace_i915_gem_request_complete (dev, seqno); -DRM_WAKEUP (&dev_priv->irq_queue); +wake_up (&dev_priv->irq_queue); dev_priv->hangcheck_count =0 ;mod_timer (&dev_priv->hangcheck_timer, jiffies +DRM_I915_HANGCHECK_PERIOD ); }
@@ -829,7 +829,7 @@ static void i915_handle_error(struct drm_device *dev, bool wedged)/* * Wakeup waiting processes so they don't hang */ -DRM_WAKEUP (&dev_priv->irq_queue); +wake_up (&dev_priv->irq_queue); }queue_work (dev_priv->wq, &dev_priv->error_work);
@@ -931,7 +931,7 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS)u32 seqno =i915_get_gem_seqno (dev); dev_priv->mm.irq_gem_seqno = seqno;trace_i915_gem_request_complete (dev, seqno); -DRM_WAKEUP (&dev_priv->irq_queue); +wake_up (&dev_priv->irq_queue); dev_priv->hangcheck_count =0 ;mod_timer (&dev_priv->hangcheck_timer, jiffies +DRM_I915_HANGCHECK_PERIOD ); }
Signature · 1 line
1.6.5.2
Quoted 3 lines · Dave Airlie
This fixes a regression in the copy fb code we use to get seamless startup on Fedora/RHEL, without this fix, we block forever in the wait when starting X.
I've just checked latest -linus and found in include/drm/drm_os_linux.h:
#define DRM_WAKEUP( queue ) wake_up( queue )
It's been like that since I've originally wrote the uninterruptible sleep stuff for the intel overlay code (I've checked way back then to make sure it doesn't hang). Care to explain how this fixes a regression?
Cheers, Daniel
Signature · 3 lines
Daniel Vetter Mail: [email protected] Mobile: +41 (0)79 365 57 48
Dave Airlie
maintainerQuoted 5 lines · Daniel Vetter
Quoted 3 lines · Dave Airlie
This fixes a regression in the copy fb code we use to get seamless startup on Fedora/RHEL, without this fix, we block forever in the wait when starting X.
Quoted 8 lines
I've just checked latest -linus and found in
include/drm/drm_os_linux.h:#define
DRM_WAKEUP( queue )wake_up( queue )It's been like that since I've originally wrote the uninterruptible sleep stuff for the intel overlay code (I've checked way back then to make sure it doesn't hang). Care to explain how this fixes a regression?
doh, you fixed that in the patch, and for some reason we don't have that in our Fedora or RHEL backports wierd.
Ignore this.
I don't remember much discussion on changing drm_os_linux, but it should have been fine.
Dave.
Quoted 15 lines · On Fri, 28 May 2010 19:55:24 +1000 Dave Airlie <[email protected]> wrote:
Quoted 5 lines · Daniel Vetter
Quoted 3 lines · Dave Airlie
This fixes a regression in the copy fb code we use to get seamless startup on Fedora/RHEL, without this fix, we block forever in the wait when starting X.
Quoted 8 lines
I've just checked latest -linus and found in
include/drm/drm_os_linux.h:#define
DRM_WAKEUP( queue )wake_up( queue )It's been like that since I've originally wrote the uninterruptible sleep stuff for the intel overlay code (I've checked way back then to make sure it doesn't hang). Care to explain how this fixes a regression?
Quoted 7 lines
doh, you fixed that in the patch, and for some reason we don't have that in our Fedora or RHEL backports wierd.
Ignore this.
I don't remember much discussion on changing
drm_os_linux, but it should have been fine.
We should still kill the drm_os_linux bits where possible...
Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center