From ed73651d47a5f95c3436207144b70811366e4edd Mon Sep 17 00:00:00 2001
From: Keith Packard <keithp@keithp.com>
Date: Mon, 23 Jun 2008 10:16:35 -0700
Subject: [intel-gem] Recover resources from wedged hardware.

Clean up queues, free objects. On the next entervt, unmark the hardware to
let the user try again (presumably after resetting the chip). Someday we'll
automatically recover...
---
 linux-core/i915_gem.c  | 23 +++++++++++++++++++++--
 shared-core/i915_drv.h |  9 +++++++++
 2 files changed, 30 insertions(+), 2 deletions(-)

diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c
index 4dfbd2a1..16a1b07e 100644
--- a/linux-core/i915_gem.c
+++ b/linux-core/i915_gem.c
@@ -653,7 +653,7 @@ i915_gem_retire_requests(struct drm_device *dev)
 					   list);
 		retiring_seqno = request->seqno;
 
-		if (i915_seqno_passed(seqno, retiring_seqno)) {
+		if (i915_seqno_passed(seqno, retiring_seqno) || dev_priv->mm.wedged) {
 			i915_gem_retire_request(dev, request);
 
 			list_del(&request->list);
@@ -697,10 +697,13 @@ i915_wait_request(struct drm_device *dev, uint32_t seqno)
 		i915_user_irq_on(dev_priv);
 		ret = wait_event_interruptible(dev_priv->irq_queue,
 					       i915_seqno_passed(i915_get_gem_seqno(dev),
-								 seqno));
+								 seqno) || dev_priv->mm.wedged);
 		i915_user_irq_off(dev_priv);
 		dev_priv->mm.waiting_gem_seqno = 0;
 	}
+	if (dev_priv->mm.wedged)
+		ret = -EIO;
+
 	if (ret)
 		DRM_ERROR("%s returns %d (awaiting %d at %d)\n",
 			  __func__, ret, seqno, i915_get_gem_seqno(dev));
@@ -1019,6 +1022,8 @@ i915_gem_evict_something(struct drm_device *dev)
 						   list);
 
 			ret = i915_wait_request(dev, request->seqno);
+			if (ret)
+				break;
 
 			/* if waiting caused an object to become inactive,
 			 * then loop around and wait for it. Otherwise, we
@@ -1822,6 +1827,13 @@ i915_gem_execbuffer(struct drm_device *dev, void *data,
 	mutex_lock(&dev->struct_mutex);
 
 	i915_verify_inactive(dev, __FILE__, __LINE__);
+
+	if (dev_priv->mm.wedged) {
+		DRM_ERROR("Execbuf while wedged\n");
+		mutex_unlock(&dev->struct_mutex);
+		return -EIO;
+	}
+		
 	if (dev_priv->mm.suspended) {
 		DRM_ERROR("Execbuf while VT-switched.\n");
 		mutex_unlock(&dev->struct_mutex);
@@ -2264,6 +2276,8 @@ i915_gem_idle(struct drm_device *dev)
 		if (last_seqno == cur_seqno) {
 			if (stuck++ > 100) {
 				DRM_ERROR("hardware wedged\n");
+				dev_priv->mm.wedged = 1;
+				DRM_WAKEUP(&dev_priv->irq_queue);
 				break;
 			}
 		}
@@ -2378,6 +2392,11 @@ i915_gem_entervt_ioctl(struct drm_device *dev, void *data,
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	int ret;
 
+	if (dev_priv->mm.wedged) {
+		DRM_ERROR("Renabling wedged hardware, good luck\n");
+		dev_priv->mm.wedged = 0;
+	}
+
 	ret = i915_gem_init_ringbuffer(dev);
 	if (ret != 0)
 		return ret;
diff --git a/shared-core/i915_drv.h b/shared-core/i915_drv.h
index 9f500330..0df5af9c 100644
--- a/shared-core/i915_drv.h
+++ b/shared-core/i915_drv.h
@@ -307,6 +307,15 @@ typedef struct drm_i915_private {
 		 * transitioned away from for kernel modesetting.
 		 */
 		int suspended;
+
+		/**
+		 * Flag if the hardware appears to be wedged.
+		 *
+		 * This is set when attempts to idle the device timeout.
+		 * It prevents command submission from occuring and makes
+		 * every pending request fail
+		 */
+		int wedged;
 	} mm;
 } drm_i915_private_t;
 
-- 
cgit v1.2.3