drm/i915: Clarify event_lock locking, process context
authorDaniel Vetter <daniel.vetter@ffwll.ch>
Mon, 15 Sep 2014 12:55:22 +0000 (14:55 +0200)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Fri, 19 Sep 2014 12:43:17 +0000 (14:43 +0200)
It's good practice to use the more specific versions for irq save
spinlocks both as executable documentation and to enforce saner
design. The _irqsave version really should only be used if the calling
context is unknown and there's a good reason to call a function from
all kinds of places.

This is the first step whice replaces all occurances of _irqsave in
process context with the simpler irq disable/enable variants. We don't
have any funky spinlock nesting going on, especially since the
event_lock is the outermost of the irq/vblank related spinlocks.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_debugfs.c
drivers/gpu/drm/i915/intel_display.c

index 063b44817e083649088f59f677fcc300679bd5ec..0ba5c7145240c85e2a770ad683e8cfdba0607c3d 100644 (file)
@@ -516,7 +516,6 @@ static int i915_gem_pageflip_info(struct seq_file *m, void *data)
        struct drm_info_node *node = m->private;
        struct drm_device *dev = node->minor->dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
-       unsigned long flags;
        struct intel_crtc *crtc;
        int ret;
 
@@ -529,7 +528,7 @@ static int i915_gem_pageflip_info(struct seq_file *m, void *data)
                const char plane = plane_name(crtc->plane);
                struct intel_unpin_work *work;
 
-               spin_lock_irqsave(&dev->event_lock, flags);
+               spin_lock_irq(&dev->event_lock);
                work = crtc->unpin_work;
                if (work == NULL) {
                        seq_printf(m, "No flip due on pipe %c (plane %c)\n",
@@ -575,7 +574,7 @@ static int i915_gem_pageflip_info(struct seq_file *m, void *data)
                                seq_printf(m, "MMIO update completed? %d\n",  addr == work->gtt_offset);
                        }
                }
-               spin_unlock_irqrestore(&dev->event_lock, flags);
+               spin_unlock_irq(&dev->event_lock);
        }
 
        mutex_unlock(&dev->struct_mutex);
index 365e04ca9120e75e444d41379950587e014d910d..624d9bfc5c0bae41e1ef4714e30e76fe7d1a67d8 100644 (file)
@@ -2711,16 +2711,15 @@ static bool intel_crtc_has_pending_flip(struct drm_crtc *crtc)
        struct drm_device *dev = crtc->dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
        struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       unsigned long flags;
        bool pending;
 
        if (i915_reset_in_progress(&dev_priv->gpu_error) ||
            intel_crtc->reset_counter != atomic_read(&dev_priv->gpu_error.reset_counter))
                return false;
 
-       spin_lock_irqsave(&dev->event_lock, flags);
+       spin_lock_irq(&dev->event_lock);
        pending = to_intel_crtc(crtc)->unpin_work != NULL;
-       spin_unlock_irqrestore(&dev->event_lock, flags);
+       spin_unlock_irq(&dev->event_lock);
 
        return pending;
 }
@@ -3431,14 +3430,13 @@ void intel_crtc_wait_for_pending_flips(struct drm_crtc *crtc)
                                       !intel_crtc_has_pending_flip(crtc),
                                       60*HZ) == 0)) {
                struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-               unsigned long flags;
 
-               spin_lock_irqsave(&dev->event_lock, flags);
+               spin_lock_irq(&dev->event_lock);
                if (intel_crtc->unpin_work) {
                        WARN_ONCE(1, "Removing stuck page flip\n");
                        page_flip_completed(intel_crtc);
                }
-               spin_unlock_irqrestore(&dev->event_lock, flags);
+               spin_unlock_irq(&dev->event_lock);
        }
 
        if (crtc->primary->fb) {
@@ -9280,12 +9278,11 @@ static void intel_crtc_destroy(struct drm_crtc *crtc)
        struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
        struct drm_device *dev = crtc->dev;
        struct intel_unpin_work *work;
-       unsigned long flags;
 
-       spin_lock_irqsave(&dev->event_lock, flags);
+       spin_lock_irq(&dev->event_lock);
        work = intel_crtc->unpin_work;
        intel_crtc->unpin_work = NULL;
-       spin_unlock_irqrestore(&dev->event_lock, flags);
+       spin_unlock_irq(&dev->event_lock);
 
        if (work) {
                cancel_work_sync(&work->work);
@@ -9896,7 +9893,6 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
        enum pipe pipe = intel_crtc->pipe;
        struct intel_unpin_work *work;
        struct intel_engine_cs *ring;
-       unsigned long flags;
        int ret;
 
        //trigger software GT busyness calculation
@@ -9940,7 +9936,7 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
                goto free_work;
 
        /* We borrow the event spin lock for protecting unpin_work */
-       spin_lock_irqsave(&dev->event_lock, flags);
+       spin_lock_irq(&dev->event_lock);
        if (intel_crtc->unpin_work) {
                /* Before declaring the flip queue wedged, check if
                 * the hardware completed the operation behind our backs.
@@ -9950,7 +9946,7 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
                        page_flip_completed(intel_crtc);
                } else {
                        DRM_DEBUG_DRIVER("flip queue: crtc already busy\n");
-                       spin_unlock_irqrestore(&dev->event_lock, flags);
+                       spin_unlock_irq(&dev->event_lock);
 
                        drm_crtc_vblank_put(crtc);
                        kfree(work);
@@ -9958,7 +9954,7 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
                }
        }
        intel_crtc->unpin_work = work;
-       spin_unlock_irqrestore(&dev->event_lock, flags);
+       spin_unlock_irq(&dev->event_lock);
 
        if (atomic_read(&intel_crtc->unpin_work_count) >= 2)
                flush_workqueue(dev_priv->wq);
@@ -10045,9 +10041,9 @@ cleanup_pending:
        mutex_unlock(&dev->struct_mutex);
 
 cleanup:
-       spin_lock_irqsave(&dev->event_lock, flags);
+       spin_lock_irq(&dev->event_lock);
        intel_crtc->unpin_work = NULL;
-       spin_unlock_irqrestore(&dev->event_lock, flags);
+       spin_unlock_irq(&dev->event_lock);
 
        drm_crtc_vblank_put(crtc);
 free_work:
@@ -10058,9 +10054,9 @@ out_hang:
                intel_crtc_wait_for_pending_flips(crtc);
                ret = intel_pipe_set_base(crtc, crtc->x, crtc->y, fb);
                if (ret == 0 && event) {
-                       spin_lock_irqsave(&dev->event_lock, flags);
+                       spin_lock_irq(&dev->event_lock);
                        drm_send_vblank_event(dev, pipe, event);
-                       spin_unlock_irqrestore(&dev->event_lock, flags);
+                       spin_unlock_irq(&dev->event_lock);
                }
        }
        return ret;
@@ -13769,9 +13765,8 @@ void intel_modeset_preclose(struct drm_device *dev, struct drm_file *file)
 
        for_each_intel_crtc(dev, crtc) {
                struct intel_unpin_work *work;
-               unsigned long irqflags;
 
-               spin_lock_irqsave(&dev->event_lock, irqflags);
+               spin_lock_irq(&dev->event_lock);
 
                work = crtc->unpin_work;
 
@@ -13781,6 +13776,6 @@ void intel_modeset_preclose(struct drm_device *dev, struct drm_file *file)
                        work->event = NULL;
                }
 
-               spin_unlock_irqrestore(&dev->event_lock, irqflags);
+               spin_unlock_irq(&dev->event_lock);
        }
 }