Allocate objects from stolen memory in preparation for BIOS takeover

Allocate objects from stolen memory in preparation for BIOS takeover

[Chris Wilson]

In comparison to last time, the use of generic objects backed by stolen
memory is fully supported. This has been used to test stolen objects by
allocating the ringbuffer, fbcon and other permanent objects out of
stolen memory and using them in anger.

The first several patches are in preparation for constructing new types
of objects by removing the complication of extraneous tracking lists and
disassociating the link between GTT space and pinned pages. We then
separate out the current stolen handling from dma, and begin the process
of making the core code stolen aware. Finally we start using stolen
memory to create objects, with the last patch suggesting how we might
use the stolen object to preserve the BIOS framebuffer (and mode
configuration) across GEM initialisation.
-Chris

[PATCH 01/28] drm/i915: Remove the list of pinned inactive objects

[Chris Wilson]

Simplify object tracking by removing the inactive but pinned list. The
only place where this was used is for counting the available memory,
which is just as easy performed by checking all objects on the rare
occasions it is required (application startup). For ease of debugging,
we keep the reporting of pinned objects through the error-state and
debugfs.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c   |   16 ++-----
 drivers/gpu/drm/i915/i915_drv.h       |    6 ---
 drivers/gpu/drm/i915/i915_gem.c       |   38 ++++-----------
 drivers/gpu/drm/i915/i915_gem_debug.c |   16 -------
 drivers/gpu/drm/i915/i915_gem_evict.c |   10 ++--
 drivers/gpu/drm/i915/i915_irq.c       |   82 ++++++++++++++++++++-------------
 6 files changed, 69 insertions(+), 99 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 54a1066..ecf7468 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -178,10 +178,6 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)
 		seq_printf(m, "Inactive:\n");
 		head = &dev_priv->mm.inactive_list;
 		break;
-	case PINNED_LIST:
-		seq_printf(m, "Pinned:\n");
-		head = &dev_priv->mm.pinned_list;
-		break;
 	case FLUSHING_LIST:
 		seq_printf(m, "Flushing:\n");
 		head = &dev_priv->mm.flushing_list;
@@ -252,11 +248,6 @@ static int i915_gem_object_info(struct seq_file *m, void* data)
 		   count, mappable_count, size, mappable_size);
 
 	size = count = mappable_size = mappable_count = 0;
-	count_objects(&dev_priv->mm.pinned_list, mm_list);
-	seq_printf(m, "  %u [%u] pinned objects, %zu [%zu] bytes\n",
-		   count, mappable_count, size, mappable_size);
-
-	size = count = mappable_size = mappable_count = 0;
 	count_objects(&dev_priv->mm.inactive_list, mm_list);
 	seq_printf(m, "  %u [%u] inactive objects, %zu [%zu] bytes\n",
 		   count, mappable_count, size, mappable_size);
@@ -294,6 +285,7 @@ static int i915_gem_gtt_info(struct seq_file *m, void* data)
 {
 	struct drm_info_node *node = (struct drm_info_node *) m->private;
 	struct drm_device *dev = node->minor->dev;
+	uintptr_t list = (uintptr_t) node->info_ent->data;
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct drm_i915_gem_object *obj;
 	size_t total_obj_size, total_gtt_size;
@@ -305,6 +297,9 @@ static int i915_gem_gtt_info(struct seq_file *m, void* data)
 
 	total_obj_size = total_gtt_size = count = 0;
 	list_for_each_entry(obj, &dev_priv->mm.gtt_list, gtt_list) {
+		if (list == PINNED_LIST && obj->pin_count == 0)
+			continue;
+
 		seq_printf(m, "   ");
 		describe_obj(m, obj);
 		seq_printf(m, "\n");
@@ -321,7 +316,6 @@ static int i915_gem_gtt_info(struct seq_file *m, void* data)
 	return 0;
 }
 
-
 static int i915_gem_pageflip_info(struct seq_file *m, void *data)
 {
 	struct drm_info_node *node = (struct drm_info_node *) m->private;
@@ -1842,10 +1836,10 @@ static struct drm_info_list i915_debugfs_list[] = {
 	{"i915_capabilities", i915_capabilities, 0},
 	{"i915_gem_objects", i915_gem_object_info, 0},
 	{"i915_gem_gtt", i915_gem_gtt_info, 0},
+	{"i915_gem_pinned", i915_gem_gtt_info, 0, (void *) PINNED_LIST},
 	{"i915_gem_active", i915_gem_object_list_info, 0, (void *) ACTIVE_LIST},
 	{"i915_gem_flushing", i915_gem_object_list_info, 0, (void *) FLUSHING_LIST},
 	{"i915_gem_inactive", i915_gem_object_list_info, 0, (void *) INACTIVE_LIST},
-	{"i915_gem_pinned", i915_gem_object_list_info, 0, (void *) PINNED_LIST},
 	{"i915_gem_deferred_free", i915_gem_object_list_info, 0, (void *) DEFERRED_FREE_LIST},
 	{"i915_gem_pageflip", i915_gem_pageflip_info, 0},
 	{"i915_gem_request", i915_gem_request_info, 0},
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 57f60fa..560ce7f 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -689,12 +689,6 @@ typedef struct drm_i915_private {
 		 */
 		struct list_head inactive_list;
 
-		/**
-		 * LRU list of objects which are not in the ringbuffer but
-		 * are still pinned in the GTT.
-		 */
-		struct list_head pinned_list;
-
 		/** LRU list of objects with fence regs on them. */
 		struct list_head fence_list;
 
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 2fc7c55..bf4683c 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -132,7 +132,7 @@ int i915_mutex_lock_interruptible(struct drm_device *dev)
 static inline bool
 i915_gem_object_is_inactive(struct drm_i915_gem_object *obj)
 {
-	return !obj->active && obj->pin_count == 0;
+	return !obj->active;
 }
 
 int
@@ -171,8 +171,9 @@ i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
 
 	pinned = 0;
 	mutex_lock(&dev->struct_mutex);
-	list_for_each_entry(obj, &dev_priv->mm.pinned_list, mm_list)
-		pinned += obj->gtt_space->size;
+	list_for_each_entry(obj, &dev_priv->mm.gtt_list, gtt_list)
+		if (obj->pin_count)
+			pinned += obj->gtt_space->size;
 	mutex_unlock(&dev->struct_mutex);
 
 	args->aper_size = dev_priv->mm.gtt_total;
@@ -1455,10 +1456,7 @@ i915_gem_object_move_to_inactive(struct drm_i915_gem_object *obj)
 	struct drm_device *dev = obj->base.dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
 
-	if (obj->pin_count != 0)
-		list_move_tail(&obj->mm_list, &dev_priv->mm.pinned_list);
-	else
-		list_move_tail(&obj->mm_list, &dev_priv->mm.inactive_list);
+	list_move_tail(&obj->mm_list, &dev_priv->mm.inactive_list);
 
 	BUG_ON(!list_empty(&obj->gpu_write_list));
 	BUG_ON(!obj->active);
@@ -3063,12 +3061,9 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj,
 		    uint32_t alignment,
 		    bool map_and_fenceable)
 {
-	struct drm_device *dev = obj->base.dev;
-	struct drm_i915_private *dev_priv = dev->dev_private;
 	int ret;
 
 	BUG_ON(obj->pin_count == DRM_I915_GEM_OBJECT_MAX_PIN_COUNT);
-	WARN_ON(i915_verify_lists(dev));
 
 	if (obj->gtt_space != NULL) {
 		if ((alignment && obj->gtt_offset & (alignment - 1)) ||
@@ -3096,34 +3091,20 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj,
 	if (!obj->has_global_gtt_mapping && map_and_fenceable)
 		i915_gem_gtt_bind_object(obj, obj->cache_level);
 
-	if (obj->pin_count++ == 0) {
-		if (!obj->active)
-			list_move_tail(&obj->mm_list,
-				       &dev_priv->mm.pinned_list);
-	}
+	obj->pin_count++;
 	obj->pin_mappable |= map_and_fenceable;
 
-	WARN_ON(i915_verify_lists(dev));
 	return 0;
 }
 
 void
 i915_gem_object_unpin(struct drm_i915_gem_object *obj)
 {
-	struct drm_device *dev = obj->base.dev;
-	drm_i915_private_t *dev_priv = dev->dev_private;
-
-	WARN_ON(i915_verify_lists(dev));
 	BUG_ON(obj->pin_count == 0);
 	BUG_ON(obj->gtt_space == NULL);
 
-	if (--obj->pin_count == 0) {
-		if (!obj->active)
-			list_move_tail(&obj->mm_list,
-				       &dev_priv->mm.inactive_list);
+	if (--obj->pin_count == 0)
 		obj->pin_mappable = false;
-	}
-	WARN_ON(i915_verify_lists(dev));
 }
 
 int
@@ -3426,12 +3407,10 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)
 	struct drm_i915_gem_object *obj = to_intel_bo(gem_obj);
 	struct drm_device *dev = obj->base.dev;
 
-	while (obj->pin_count > 0)
-		i915_gem_object_unpin(obj);
-
 	if (obj->phys_obj)
 		i915_gem_detach_phys_object(dev, obj);
 
+	obj->pin_count = 0;
 	i915_gem_free_object_tail(obj);
 }
 
@@ -3699,7 +3678,6 @@ i915_gem_load(struct drm_device *dev)
 	INIT_LIST_HEAD(&dev_priv->mm.active_list);
 	INIT_LIST_HEAD(&dev_priv->mm.flushing_list);
 	INIT_LIST_HEAD(&dev_priv->mm.inactive_list);
-	INIT_LIST_HEAD(&dev_priv->mm.pinned_list);
 	INIT_LIST_HEAD(&dev_priv->mm.fence_list);
 	INIT_LIST_HEAD(&dev_priv->mm.deferred_free_list);
 	INIT_LIST_HEAD(&dev_priv->mm.gtt_list);
diff --git a/drivers/gpu/drm/i915/i915_gem_debug.c b/drivers/gpu/drm/i915/i915_gem_debug.c
index cc93cac..a4f6aaa 100644
--- a/drivers/gpu/drm/i915/i915_gem_debug.c
+++ b/drivers/gpu/drm/i915/i915_gem_debug.c
@@ -114,22 +114,6 @@ i915_verify_lists(struct drm_device *dev)
 		}
 	}
 
-	list_for_each_entry(obj, &dev_priv->mm.pinned_list, list) {
-		if (obj->base.dev != dev ||
-		    !atomic_read(&obj->base.refcount.refcount)) {
-			DRM_ERROR("freed pinned %p\n", obj);
-			err++;
-			break;
-		} else if (!obj->pin_count || obj->active ||
-			   (obj->base.write_domain & I915_GEM_GPU_DOMAINS)) {
-			DRM_ERROR("invalid pinned %p (p %d a %d w %x)\n",
-				  obj,
-				  obj->pin_count, obj->active,
-				  obj->base.write_domain);
-			err++;
-		}
-	}
-
 	return warned = err;
 }
 #endif /* WATCH_INACTIVE */
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 399a3a8..91ebb94 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -35,6 +35,9 @@
 static bool
 mark_free(struct drm_i915_gem_object *obj, struct list_head *unwind)
 {
+	if (obj->pin_count)
+		return false;
+
 	list_add(&obj->exec_list, unwind);
 	return drm_mm_scan_add_block(obj->gtt_space);
 }
@@ -90,7 +93,7 @@ i915_gem_evict_something(struct drm_device *dev, int min_size,
 	/* Now merge in the soon-to-be-expired objects... */
 	list_for_each_entry(obj, &dev_priv->mm.active_list, mm_list) {
 		/* Does the object require an outstanding flush? */
-		if (obj->base.write_domain || obj->pin_count)
+		if (obj->base.write_domain)
 			continue;
 
 		if (mark_free(obj, &unwind_list))
@@ -99,14 +102,11 @@ i915_gem_evict_something(struct drm_device *dev, int min_size,
 
 	/* Finally add anything with a pending flush (in order of retirement) */
 	list_for_each_entry(obj, &dev_priv->mm.flushing_list, mm_list) {
-		if (obj->pin_count)
-			continue;
-
 		if (mark_free(obj, &unwind_list))
 			goto found;
 	}
 	list_for_each_entry(obj, &dev_priv->mm.active_list, mm_list) {
-		if (!obj->base.write_domain || obj->pin_count)
+		if (!obj->base.write_domain)
 			continue;
 
 		if (mark_free(obj, &unwind_list))
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 26172ee..24e1dd2 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -918,37 +918,56 @@ i915_error_state_free(struct drm_device *dev,
 	kfree(error->overlay);
 	kfree(error);
 }
-
-static u32 capture_bo_list(struct drm_i915_error_buffer *err,
-			   int count,
-			   struct list_head *head)
+static void capture_bo(struct drm_i915_error_buffer *err,
+		       struct drm_i915_gem_object *obj)
+{
+	err->size = obj->base.size;
+	err->name = obj->base.name;
+	err->seqno = obj->last_rendering_seqno;
+	err->gtt_offset = obj->gtt_offset;
+	err->read_domains = obj->base.read_domains;
+	err->write_domain = obj->base.write_domain;
+	err->fence_reg = obj->fence_reg;
+	err->pinned = 0;
+	if (obj->pin_count > 0)
+		err->pinned = 1;
+	if (obj->user_pin_count > 0)
+		err->pinned = -1;
+	err->tiling = obj->tiling_mode;
+	err->dirty = obj->dirty;
+	err->purgeable = obj->madv != I915_MADV_WILLNEED;
+	err->ring = obj->ring ? obj->ring->id : -1;
+	err->cache_level = obj->cache_level;
+}
+
+static u32 capture_active_bo(struct drm_i915_error_buffer *err,
+			     int count, struct list_head *head)
 {
 	struct drm_i915_gem_object *obj;
 	int i = 0;
 
 	list_for_each_entry(obj, head, mm_list) {
-		err->size = obj->base.size;
-		err->name = obj->base.name;
-		err->seqno = obj->last_rendering_seqno;
-		err->gtt_offset = obj->gtt_offset;
-		err->read_domains = obj->base.read_domains;
-		err->write_domain = obj->base.write_domain;
-		err->fence_reg = obj->fence_reg;
-		err->pinned = 0;
-		if (obj->pin_count > 0)
-			err->pinned = 1;
-		if (obj->user_pin_count > 0)
-			err->pinned = -1;
-		err->tiling = obj->tiling_mode;
-		err->dirty = obj->dirty;
-		err->purgeable = obj->madv != I915_MADV_WILLNEED;
-		err->ring = obj->ring ? obj->ring->id : -1;
-		err->cache_level = obj->cache_level;
-
+		capture_bo(err++, obj);
 		if (++i == count)
 			break;
+	}
+
+	return i;
+}
 
-		err++;
+static u32 capture_pinned_bo(struct drm_i915_error_buffer *err,
+			     int count, struct list_head *head)
+{
+	struct drm_i915_gem_object *obj;
+	int i = 0;
+
+	list_for_each_entry(obj, head, gtt_list) {
+		if (obj->pin_count == 0)
+			continue;
+
+		capture_bo(err++, obj);
+		if (++i == count)
+			break;
 	}
 
 	return i;
@@ -1155,8 +1174,9 @@ static void i915_capture_error_state(struct drm_device *dev)
 	list_for_each_entry(obj, &dev_priv->mm.active_list, mm_list)
 		i++;
 	error->active_bo_count = i;
-	list_for_each_entry(obj, &dev_priv->mm.pinned_list, mm_list)
-		i++;
+	list_for_each_entry(obj, &dev_priv->mm.gtt_list, gtt_list)
+		if (obj->pin_count)
+			i++;
 	error->pinned_bo_count = i - error->active_bo_count;
 
 	error->active_bo = NULL;
@@ -1171,15 +1191,15 @@ static void i915_capture_error_state(struct drm_device *dev)
 
 	if (error->active_bo)
 		error->active_bo_count =
-			capture_bo_list(error->active_bo,
-					error->active_bo_count,
-					&dev_priv->mm.active_list);
+			capture_active_bo(error->active_bo,
+					  error->active_bo_count,
+					  &dev_priv->mm.active_list);
 
 	if (error->pinned_bo)
 		error->pinned_bo_count =
-			capture_bo_list(error->pinned_bo,
-					error->pinned_bo_count,
-					&dev_priv->mm.pinned_list);
+			capture_pinned_bo(error->pinned_bo,
+					  error->pinned_bo_count,
+					  &dev_priv->mm.gtt_list);
 
 	do_gettimeofday(&error->time);
 
-- 
1.7.10

[PATCH 02/28] drm/i915: Remove the deferred-free list

[Chris Wilson]

The use of the mm_list by deferred-free breaks the following patches to
extend the range of objects tracked. We can simplify things if we just
make the unbind during free uninterrible.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c |   11 -------
 drivers/gpu/drm/i915/i915_drv.h     |    8 -----
 drivers/gpu/drm/i915/i915_gem.c     |   58 ++++++++++++-----------------------
 3 files changed, 19 insertions(+), 58 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index ecf7468..120db46 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -47,7 +47,6 @@ enum {
 	FLUSHING_LIST,
 	INACTIVE_LIST,
 	PINNED_LIST,
-	DEFERRED_FREE_LIST,
 };
 
 static const char *yesno(int v)
@@ -182,10 +181,6 @@ static int i915_gem_object_list_info(struct seq_file *m, void *data)
 		seq_printf(m, "Flushing:\n");
 		head = &dev_priv->mm.flushing_list;
 		break;
-	case DEFERRED_FREE_LIST:
-		seq_printf(m, "Deferred free:\n");
-		head = &dev_priv->mm.deferred_free_list;
-		break;
 	default:
 		mutex_unlock(&dev->struct_mutex);
 		return -EINVAL;
@@ -253,11 +248,6 @@ static int i915_gem_object_info(struct seq_file *m, void* data)
 		   count, mappable_count, size, mappable_size);
 
 	size = count = mappable_size = mappable_count = 0;
-	count_objects(&dev_priv->mm.deferred_free_list, mm_list);
-	seq_printf(m, "  %u [%u] freed objects, %zu [%zu] bytes\n",
-		   count, mappable_count, size, mappable_size);
-
-	size = count = mappable_size = mappable_count = 0;
 	list_for_each_entry(obj, &dev_priv->mm.gtt_list, gtt_list) {
 		if (obj->fault_mappable) {
 			size += obj->gtt_space->size;
@@ -1840,7 +1830,6 @@ static struct drm_info_list i915_debugfs_list[] = {
 	{"i915_gem_active", i915_gem_object_list_info, 0, (void *) ACTIVE_LIST},
 	{"i915_gem_flushing", i915_gem_object_list_info, 0, (void *) FLUSHING_LIST},
 	{"i915_gem_inactive", i915_gem_object_list_info, 0, (void *) INACTIVE_LIST},
-	{"i915_gem_deferred_free", i915_gem_object_list_info, 0, (void *) DEFERRED_FREE_LIST},
 	{"i915_gem_pageflip", i915_gem_pageflip_info, 0},
 	{"i915_gem_request", i915_gem_request_info, 0},
 	{"i915_gem_seqno", i915_gem_seqno_info, 0},
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 560ce7f..21127af 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -693,14 +693,6 @@ typedef struct drm_i915_private {
 		struct list_head fence_list;
 
 		/**
-		 * List of objects currently pending being freed.
-		 *
-		 * These objects are no longer in use, but due to a signal
-		 * we were prevented from freeing them at the appointed time.
-		 */
-		struct list_head deferred_free_list;
-
-		/**
 		 * We leave the user IRQ off as much as possible,
 		 * but this means that requests will finish and never
 		 * be retired once the system goes idle. Set a timer to
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index bf4683c..9bf11c8 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -46,7 +46,6 @@ static int i915_gem_phys_pwrite(struct drm_device *dev,
 				struct drm_i915_gem_object *obj,
 				struct drm_i915_gem_pwrite *args,
 				struct drm_file *file);
-static void i915_gem_free_object_tail(struct drm_i915_gem_object *obj);
 
 static void i915_gem_write_fence(struct drm_device *dev, int reg,
 				 struct drm_i915_gem_object *obj);
@@ -1782,20 +1781,6 @@ i915_gem_retire_requests(struct drm_device *dev)
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	int i;
 
-	if (!list_empty(&dev_priv->mm.deferred_free_list)) {
-	    struct drm_i915_gem_object *obj, *next;
-
-	    /* We must be careful that during unbind() we do not
-	     * accidentally infinitely recurse into retire requests.
-	     * Currently:
-	     *   retire -> free -> unbind -> wait -> retire_ring
-	     */
-	    list_for_each_entry_safe(obj, next,
-				     &dev_priv->mm.deferred_free_list,
-				     mm_list)
-		    i915_gem_free_object_tail(obj);
-	}
-
 	for (i = 0; i < I915_NUM_RINGS; i++)
 		i915_gem_retire_requests_ring(&dev_priv->ring[i]);
 }
@@ -2067,7 +2052,7 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj)
 	}
 
 	ret = i915_gem_object_finish_gpu(obj);
-	if (ret == -ERESTARTSYS)
+	if (ret)
 		return ret;
 	/* Continue on if we fail due to EIO, the GPU is hung so we
 	 * should be safe and we need to cleanup or else we might
@@ -2094,7 +2079,7 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj)
 
 	/* release the fence reg _after_ flushing */
 	ret = i915_gem_object_put_fence(obj);
-	if (ret == -ERESTARTSYS)
+	if (ret)
 		return ret;
 
 	trace_i915_gem_object_unbind(obj);
@@ -3377,21 +3362,29 @@ int i915_gem_init_object(struct drm_gem_object *obj)
 	return 0;
 }
 
-static void i915_gem_free_object_tail(struct drm_i915_gem_object *obj)
+void i915_gem_free_object(struct drm_gem_object *gem_obj)
 {
+	struct drm_i915_gem_object *obj = to_intel_bo(gem_obj);
 	struct drm_device *dev = obj->base.dev;
 	drm_i915_private_t *dev_priv = dev->dev_private;
-	int ret;
-
-	ret = i915_gem_object_unbind(obj);
-	if (ret == -ERESTARTSYS) {
-		list_move(&obj->mm_list,
-			  &dev_priv->mm.deferred_free_list);
-		return;
-	}
 
 	trace_i915_gem_object_destroy(obj);
 
+	if (obj->phys_obj)
+		i915_gem_detach_phys_object(dev, obj);
+
+	obj->pin_count = 0;
+	if (WARN_ON(i915_gem_object_unbind(obj) == -ERESTARTSYS)) {
+		bool was_interruptible;
+
+		was_interruptible = dev_priv->mm.interruptible;
+		dev_priv->mm.interruptible = false;
+
+		WARN_ON(i915_gem_object_unbind(obj));
+
+		dev_priv->mm.interruptible = was_interruptible;
+	}
+
 	if (obj->base.map_list.map)
 		drm_gem_free_mmap_offset(&obj->base);
 
@@ -3402,18 +3395,6 @@ static void i915_gem_free_object_tail(struct drm_i915_gem_object *obj)
 	kfree(obj);
 }
 
-void i915_gem_free_object(struct drm_gem_object *gem_obj)
-{
-	struct drm_i915_gem_object *obj = to_intel_bo(gem_obj);
-	struct drm_device *dev = obj->base.dev;
-
-	if (obj->phys_obj)
-		i915_gem_detach_phys_object(dev, obj);
-
-	obj->pin_count = 0;
-	i915_gem_free_object_tail(obj);
-}
-
 int
 i915_gem_idle(struct drm_device *dev)
 {
@@ -3679,7 +3660,6 @@ i915_gem_load(struct drm_device *dev)
 	INIT_LIST_HEAD(&dev_priv->mm.flushing_list);
 	INIT_LIST_HEAD(&dev_priv->mm.inactive_list);
 	INIT_LIST_HEAD(&dev_priv->mm.fence_list);
-	INIT_LIST_HEAD(&dev_priv->mm.deferred_free_list);
 	INIT_LIST_HEAD(&dev_priv->mm.gtt_list);
 	for (i = 0; i < I915_NUM_RINGS; i++)
 		init_ring_lists(&dev_priv->ring[i]);
-- 
1.7.10

[PATCH 03/28] drm/i915: Track unbound pages

[Chris Wilson]

When dealing with a working set larger than the GATT, or even the
mappable aperture when touching through the GTT, we end up with evicting
objects only to rebind them at a new offset again later. Moving an
object into and out of the GTT requires clflushing the pages, thus
causing a double-clflush penalty for rebinding.

To avoid having to clflush on rebinding, we can track the pages as they
are evicted from the GTT and only relinquish those pages on memory
pressure.

As usual, if it were not for the handling of out-of-memory condition and
having to manually shrink our own bo caches, it would be a net reduction
of code. Alas.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c        |    5 +
 drivers/gpu/drm/i915/i915_drv.h            |    9 +-
 drivers/gpu/drm/i915/i915_gem.c            |  444 ++++++++++++++++------------
 drivers/gpu/drm/i915/i915_gem_evict.c      |   13 +-
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |    9 +-
 drivers/gpu/drm/i915/i915_trace.h          |   10 +-
 6 files changed, 272 insertions(+), 218 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 120db46..de7c65e 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -247,6 +247,11 @@ static int i915_gem_object_info(struct seq_file *m, void* data)
 	seq_printf(m, "  %u [%u] inactive objects, %zu [%zu] bytes\n",
 		   count, mappable_count, size, mappable_size);
 
+	size = count = 0;
+	list_for_each_entry(obj, &dev_priv->mm.unbound_list, mm_list)
+		size += obj->base.size, ++count;
+	seq_printf(m, "%u unbound objects, %zu bytes\n", count, size);
+
 	size = count = mappable_size = mappable_count = 0;
 	list_for_each_entry(obj, &dev_priv->mm.gtt_list, gtt_list) {
 		if (obj->fault_mappable) {
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 21127af..866b280 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -689,6 +689,13 @@ typedef struct drm_i915_private {
 		 */
 		struct list_head inactive_list;
 
+		/**
+		 * List of objects which are not bound to the GTT (thus
+		 * are idle and not used by the GPU) but still have
+		 * (presumably uncached) pages still attached.
+		 */
+		struct list_head unbound_list;
+
 		/** LRU list of objects with fence regs on them. */
 		struct list_head fence_list;
 
@@ -1356,7 +1363,7 @@ void i915_gem_init_global_gtt(struct drm_device *dev,
 /* i915_gem_evict.c */
 int __must_check i915_gem_evict_something(struct drm_device *dev, int min_size,
 					  unsigned alignment, bool mappable);
-int i915_gem_evict_everything(struct drm_device *dev, bool purgeable_only);
+int i915_gem_evict_everything(struct drm_device *dev);
 
 /* i915_gem_tiling.c */
 void i915_gem_detect_bit_6_swizzle(struct drm_device *dev);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 9bf11c8..761af95 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -55,6 +55,8 @@ static void i915_gem_object_update_fence(struct drm_i915_gem_object *obj,
 
 static int i915_gem_inactive_shrink(struct shrinker *shrinker,
 				    struct shrink_control *sc);
+static long i915_gem_purge(struct drm_i915_private *dev_priv, long target);
+static void i915_gem_shrink_all(struct drm_i915_private *dev_priv);
 static void i915_gem_object_truncate(struct drm_i915_gem_object *obj);
 
 static inline void i915_gem_object_fence_lost(struct drm_i915_gem_object *obj)
@@ -413,9 +415,11 @@ i915_gem_shmem_pread(struct drm_device *dev,
 		 * anyway again before the next pread happens. */
 		if (obj->cache_level == I915_CACHE_NONE)
 			needs_clflush = 1;
-		ret = i915_gem_object_set_to_gtt_domain(obj, false);
-		if (ret)
-			return ret;
+		if (obj->gtt_space) {
+			ret = i915_gem_object_set_to_gtt_domain(obj, false);
+			if (ret)
+				return ret;
+		}
 	}
 
 	offset = args->offset;
@@ -733,9 +737,11 @@ i915_gem_shmem_pwrite(struct drm_device *dev,
 		 * right away and we therefore have to clflush anyway. */
 		if (obj->cache_level == I915_CACHE_NONE)
 			needs_clflush_after = 1;
-		ret = i915_gem_object_set_to_gtt_domain(obj, true);
-		if (ret)
-			return ret;
+		if (obj->gtt_space) {
+			ret = i915_gem_object_set_to_gtt_domain(obj, true);
+			if (ret)
+				return ret;
+		}
 	}
 	/* Same trick applies for invalidate partially written cachelines before
 	 * writing.  */
@@ -1318,59 +1324,56 @@ i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data,
 	return i915_gem_mmap_gtt(file, dev, args->handle, &args->offset);
 }
 
-
-static int
-i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj,
-			      gfp_t gfpmask)
+/* Immediately discard the backing storage */
+static void
+i915_gem_object_truncate(struct drm_i915_gem_object *obj)
 {
-	int page_count, i;
-	struct address_space *mapping;
 	struct inode *inode;
-	struct page *page;
 
-	/* Get the list of pages out of our struct file.  They'll be pinned
-	 * at this point until we release them.
+	/* Our goal here is to return as much of the memory as
+	 * is possible back to the system as we are called from OOM.
+	 * To do this we must instruct the shmfs to drop all of its
+	 * backing pages, *now*.
 	 */
-	page_count = obj->base.size / PAGE_SIZE;
-	BUG_ON(obj->pages != NULL);
-	obj->pages = drm_malloc_ab(page_count, sizeof(struct page *));
-	if (obj->pages == NULL)
-		return -ENOMEM;
-
 	inode = obj->base.filp->f_path.dentry->d_inode;
-	mapping = inode->i_mapping;
-	gfpmask |= mapping_gfp_mask(mapping);
-
-	for (i = 0; i < page_count; i++) {
-		page = shmem_read_mapping_page_gfp(mapping, i, gfpmask);
-		if (IS_ERR(page))
-			goto err_pages;
-
-		obj->pages[i] = page;
-	}
-
-	if (i915_gem_object_needs_bit17_swizzle(obj))
-		i915_gem_object_do_bit_17_swizzle(obj);
+	shmem_truncate_range(inode, 0, (loff_t)-1);
 
-	return 0;
+	if (obj->base.map_list.map)
+		drm_gem_free_mmap_offset(&obj->base);
 
-err_pages:
-	while (i--)
-		page_cache_release(obj->pages[i]);
+	obj->madv = __I915_MADV_PURGED;
+}
 
-	drm_free_large(obj->pages);
-	obj->pages = NULL;
-	return PTR_ERR(page);
+static inline int
+i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj)
+{
+	return obj->madv == I915_MADV_DONTNEED;
 }
 
-static void
+static int
 i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj)
 {
 	int page_count = obj->base.size / PAGE_SIZE;
-	int i;
+	int ret, i;
+
+	if (obj->pages == NULL)
+		return 0;
 
+	BUG_ON(obj->gtt_space);
 	BUG_ON(obj->madv == __I915_MADV_PURGED);
 
+	ret = i915_gem_object_set_to_cpu_domain(obj, 0);
+	if (ret && ret != -EIO)
+		return ret;
+
+	if (ret) {
+		/* In the event of a disaster, abandon all caches and
+		 * hope for the best.
+		 */
+		i915_gem_clflush_object(obj);
+		obj->base.read_domains = obj->base.write_domain = I915_GEM_DOMAIN_CPU;
+	}
+
 	if (i915_gem_object_needs_bit17_swizzle(obj))
 		i915_gem_object_save_bit_17_swizzle(obj);
 
@@ -1390,6 +1393,181 @@ i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj)
 
 	drm_free_large(obj->pages);
 	obj->pages = NULL;
+
+	list_del(&obj->mm_list);
+
+	if (i915_gem_object_is_purgeable(obj))
+		i915_gem_object_truncate(obj);
+
+	return 0;
+}
+
+static long
+i915_gem_purge(struct drm_i915_private *dev_priv, long target)
+{
+	struct drm_i915_gem_object *obj, *next;
+	long count = 0;
+
+	list_for_each_entry_safe(obj, next,
+				 &dev_priv->mm.unbound_list,
+				 mm_list) {
+		if (i915_gem_object_is_purgeable(obj) &&
+		    i915_gem_object_put_pages_gtt(obj) == 0) {
+			count += obj->base.size >> PAGE_SHIFT;
+			if (count >= target)
+				return count;
+		}
+	}
+
+	list_for_each_entry_safe(obj, next,
+				 &dev_priv->mm.inactive_list,
+				 mm_list) {
+		if (i915_gem_object_is_purgeable(obj) &&
+		    i915_gem_object_unbind(obj) == 0 &&
+		    i915_gem_object_put_pages_gtt(obj) == 0) {
+			count += obj->base.size >> PAGE_SHIFT;
+			if (count >= target)
+				return count;
+		}
+	}
+
+	return count;
+}
+
+inline static struct drm_i915_gem_object *
+first_unbound_bo(struct drm_i915_private *dev_priv)
+{
+	return list_first_entry(&dev_priv->mm.unbound_list,
+				struct drm_i915_gem_object,
+				mm_list);
+}
+
+static void
+i915_gem_shrink_all(struct drm_i915_private *dev_priv)
+{
+	i915_gem_evict_everything(dev_priv->dev);
+	while (!list_empty(&dev_priv->mm.unbound_list))
+		i915_gem_object_put_pages_gtt(first_unbound_bo(dev_priv));
+}
+
+/* Try to allocate some memory under the struct_mutex by purging some
+ * of our own buffers if necessary.
+ */
+static void *i915_malloc(struct drm_i915_private *dev_priv,
+			 unsigned long size)
+{
+	gfp_t gfp;
+	void *ptr;
+
+	gfp = GFP_KERNEL;
+	gfp |= __GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD;
+	gfp &= ~(__GFP_IO | __GFP_WAIT);
+
+	ptr = kmalloc(size, gfp);
+	if (ptr)
+		return ptr;
+
+	if (size <= 2*PAGE_SIZE) {
+		i915_gem_purge(dev_priv, (size >> PAGE_SHIFT) + 1);
+		ptr = kmalloc(size, gfp);
+		if (ptr)
+			return ptr;
+
+		i915_gem_shrink_all(dev_priv);
+
+		gfp &= ~(__GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD);
+		gfp |= __GFP_IO | __GFP_WAIT;
+		return kmalloc(size, gfp);
+	} else {
+		gfp |= __GFP_HIGHMEM;
+		ptr =  __vmalloc(size, gfp, PAGE_KERNEL);
+		if (ptr)
+			return ptr;
+
+		i915_gem_purge(dev_priv, (size >> PAGE_SHIFT) + 1);
+		ptr =  __vmalloc(size, gfp, PAGE_KERNEL);
+		if (ptr)
+			return ptr;
+
+		i915_gem_shrink_all(dev_priv);
+
+		gfp &= ~(__GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD);
+		gfp |= __GFP_IO | __GFP_WAIT;
+		return  __vmalloc(size, gfp, PAGE_KERNEL);
+	}
+}
+
+static int
+i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
+{
+	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
+	int page_count, i;
+	struct address_space *mapping;
+	struct page *page;
+	gfp_t gfp;
+
+	if (obj->pages)
+		return 0;
+
+	/* Assert that the object is not currently in any GPU domain. As it
+	 * wasn't in the GTT, there shouldn't be any way it could have been in
+	 * a GPU cache
+	 */
+	BUG_ON(obj->base.read_domains & I915_GEM_GPU_DOMAINS);
+	BUG_ON(obj->base.write_domain & I915_GEM_GPU_DOMAINS);
+
+	/* Get the list of pages out of our struct file.  They'll be pinned
+	 * at this point until we release them.
+	 */
+	page_count = obj->base.size / PAGE_SIZE;
+	obj->pages = i915_malloc(dev_priv, page_count*sizeof(struct page *));
+	if (obj->pages == NULL)
+		return -ENOMEM;
+
+	/* Fail silently without starting the shrinker */
+	mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping;
+	gfp = mapping_gfp_mask(mapping);
+	gfp |= __GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD;
+	gfp &= ~(__GFP_IO | __GFP_WAIT);
+	for (i = 0; i < page_count; i++) {
+		page = shmem_read_mapping_page_gfp(mapping, i, gfp);
+		if (IS_ERR(page)) {
+			i915_gem_purge(dev_priv, page_count);
+			page = shmem_read_mapping_page_gfp(mapping, i, gfp);
+		}
+		if (IS_ERR(page)) {
+			/* We've tried hard to allocate the memory by reaping
+			 * our own buffer, now let the real VM do its job and
+			 * go down in flames if truly OOM.
+			 */
+			gfp &= ~(__GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD);
+			gfp |= __GFP_IO | __GFP_WAIT;
+
+			i915_gem_shrink_all(dev_priv);
+			page = shmem_read_mapping_page_gfp(mapping, i, gfp);
+			if (IS_ERR(page))
+				goto err_pages;
+
+			gfp |= __GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD;
+			gfp &= ~(__GFP_IO | __GFP_WAIT);
+		}
+
+		obj->pages[i] = page;
+	}
+
+	if (i915_gem_object_needs_bit17_swizzle(obj))
+		i915_gem_object_do_bit_17_swizzle(obj);
+
+	list_add_tail(&obj->mm_list, &dev_priv->mm.unbound_list);
+	return 0;
+
+err_pages:
+	while (i--)
+		page_cache_release(obj->pages[i]);
+
+	drm_free_large(obj->pages);
+	obj->pages = NULL;
+	return PTR_ERR(page);
 }
 
 void
@@ -1471,32 +1649,6 @@ i915_gem_object_move_to_inactive(struct drm_i915_gem_object *obj)
 	WARN_ON(i915_verify_lists(dev));
 }
 
-/* Immediately discard the backing storage */
-static void
-i915_gem_object_truncate(struct drm_i915_gem_object *obj)
-{
-	struct inode *inode;
-
-	/* Our goal here is to return as much of the memory as
-	 * is possible back to the system as we are called from OOM.
-	 * To do this we must instruct the shmfs to drop all of its
-	 * backing pages, *now*.
-	 */
-	inode = obj->base.filp->f_path.dentry->d_inode;
-	shmem_truncate_range(inode, 0, (loff_t)-1);
-
-	if (obj->base.map_list.map)
-		drm_gem_free_mmap_offset(&obj->base);
-
-	obj->madv = __I915_MADV_PURGED;
-}
-
-static inline int
-i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj)
-{
-	return obj->madv == I915_MADV_DONTNEED;
-}
-
 static void
 i915_gem_process_flushing_list(struct intel_ring_buffer *ring,
 			       uint32_t flush_domains)
@@ -1700,6 +1852,9 @@ void i915_gem_reset(struct drm_device *dev)
 		obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS;
 	}
 
+	while (!list_empty(&dev_priv->mm.unbound_list))
+		i915_gem_object_put_pages_gtt(first_unbound_bo(dev_priv));
+
 	/* The fence registers are invalidated so clear them out */
 	i915_gem_reset_fences(dev);
 }
@@ -2046,10 +2201,8 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj)
 	if (obj->gtt_space == NULL)
 		return 0;
 
-	if (obj->pin_count != 0) {
-		DRM_ERROR("Attempting to unbind pinned buffer\n");
+	if (obj->pin_count != 0)
 		return -EINVAL;
-	}
 
 	ret = i915_gem_object_finish_gpu(obj);
 	if (ret)
@@ -2061,22 +2214,6 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj)
 
 	i915_gem_object_finish_gtt(obj);
 
-	/* Move the object to the CPU domain to ensure that
-	 * any possible CPU writes while it's not in the GTT
-	 * are flushed when we go to remap it.
-	 */
-	if (ret == 0)
-		ret = i915_gem_object_set_to_cpu_domain(obj, 1);
-	if (ret == -ERESTARTSYS)
-		return ret;
-	if (ret) {
-		/* In the event of a disaster, abandon all caches and
-		 * hope for the best.
-		 */
-		i915_gem_clflush_object(obj);
-		obj->base.read_domains = obj->base.write_domain = I915_GEM_DOMAIN_CPU;
-	}
-
 	/* release the fence reg _after_ flushing */
 	ret = i915_gem_object_put_fence(obj);
 	if (ret)
@@ -2092,10 +2229,8 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj)
 	}
 	i915_gem_gtt_finish_object(obj);
 
-	i915_gem_object_put_pages_gtt(obj);
-
 	list_del_init(&obj->gtt_list);
-	list_del_init(&obj->mm_list);
+	list_move_tail(&obj->mm_list, &dev_priv->mm.unbound_list);
 	/* Avoid an unnecessary call to unbind on rebind. */
 	obj->map_and_fenceable = true;
 
@@ -2103,10 +2238,10 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj)
 	obj->gtt_space = NULL;
 	obj->gtt_offset = 0;
 
-	if (i915_gem_object_is_purgeable(obj))
-		i915_gem_object_truncate(obj);
+	if (obj->base.read_domains & I915_GEM_DOMAIN_CPU)
+		i915_gem_object_put_pages_gtt(obj);
 
-	return ret;
+	return 0;
 }
 
 int
@@ -2490,7 +2625,6 @@ i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj,
 	struct drm_device *dev = obj->base.dev;
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct drm_mm_node *free_space;
-	gfp_t gfpmask = __GFP_NORETRY | __GFP_NOWARN;
 	u32 size, fence_size, fence_alignment, unfenced_alignment;
 	bool mappable, fenceable;
 	int ret;
@@ -2530,6 +2664,10 @@ i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj,
 		return -E2BIG;
 	}
 
+	ret = i915_gem_object_get_pages_gtt(obj);
+	if (ret)
+		return ret;
+
  search_free:
 	if (map_and_fenceable)
 		free_space =
@@ -2553,9 +2691,6 @@ i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj,
 				drm_mm_get_block(free_space, size, alignment);
 	}
 	if (obj->gtt_space == NULL) {
-		/* If the gtt is empty and we're still having trouble
-		 * fitting our object in, we're out of memory.
-		 */
 		ret = i915_gem_evict_something(dev, size, alignment,
 					       map_and_fenceable);
 		if (ret)
@@ -2564,54 +2699,19 @@ i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj,
 		goto search_free;
 	}
 
-	ret = i915_gem_object_get_pages_gtt(obj, gfpmask);
-	if (ret) {
-		drm_mm_put_block(obj->gtt_space);
-		obj->gtt_space = NULL;
-
-		if (ret == -ENOMEM) {
-			/* first try to reclaim some memory by clearing the GTT */
-			ret = i915_gem_evict_everything(dev, false);
-			if (ret) {
-				/* now try to shrink everyone else */
-				if (gfpmask) {
-					gfpmask = 0;
-					goto search_free;
-				}
-
-				return -ENOMEM;
-			}
-
-			goto search_free;
-		}
-
-		return ret;
-	}
 
 	ret = i915_gem_gtt_prepare_object(obj);
 	if (ret) {
-		i915_gem_object_put_pages_gtt(obj);
 		drm_mm_put_block(obj->gtt_space);
 		obj->gtt_space = NULL;
-
-		if (i915_gem_evict_everything(dev, false))
-			return ret;
-
-		goto search_free;
+		return ret;
 	}
 
 	if (!dev_priv->mm.aliasing_ppgtt)
 		i915_gem_gtt_bind_object(obj, obj->cache_level);
 
 	list_add_tail(&obj->gtt_list, &dev_priv->mm.gtt_list);
-	list_add_tail(&obj->mm_list, &dev_priv->mm.inactive_list);
-
-	/* Assert that the object is not currently in any GPU domain. As it
-	 * wasn't in the GTT, there shouldn't be any way it could have been in
-	 * a GPU cache
-	 */
-	BUG_ON(obj->base.read_domains & I915_GEM_GPU_DOMAINS);
-	BUG_ON(obj->base.write_domain & I915_GEM_GPU_DOMAINS);
+	list_move_tail(&obj->mm_list, &dev_priv->mm.inactive_list);
 
 	obj->gtt_offset = obj->gtt_space->start;
 
@@ -3286,9 +3386,8 @@ i915_gem_madvise_ioctl(struct drm_device *dev, void *data,
 	if (obj->madv != __I915_MADV_PURGED)
 		obj->madv = args->madv;
 
-	/* if the object is no longer bound, discard its backing storage */
-	if (i915_gem_object_is_purgeable(obj) &&
-	    obj->gtt_space == NULL)
+	/* if the object is no longer attached, discard its backing storage */
+	if (i915_gem_object_is_purgeable(obj) && obj->pages == NULL)
 		i915_gem_object_truncate(obj);
 
 	args->retained = obj->madv != __I915_MADV_PURGED;
@@ -3385,6 +3484,7 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)
 		dev_priv->mm.interruptible = was_interruptible;
 	}
 
+	i915_gem_object_put_pages_gtt(obj);
 	if (obj->base.map_list.map)
 		drm_gem_free_mmap_offset(&obj->base);
 
@@ -3416,7 +3516,7 @@ i915_gem_idle(struct drm_device *dev)
 
 	/* Under UMS, be paranoid and evict. */
 	if (!drm_core_check_feature(dev, DRIVER_MODESET))
-		i915_gem_evict_everything(dev, false);
+		i915_gem_evict_everything(dev);
 
 	i915_gem_reset_fences(dev);
 
@@ -3659,6 +3759,7 @@ i915_gem_load(struct drm_device *dev)
 	INIT_LIST_HEAD(&dev_priv->mm.active_list);
 	INIT_LIST_HEAD(&dev_priv->mm.flushing_list);
 	INIT_LIST_HEAD(&dev_priv->mm.inactive_list);
+	INIT_LIST_HEAD(&dev_priv->mm.unbound_list);
 	INIT_LIST_HEAD(&dev_priv->mm.fence_list);
 	INIT_LIST_HEAD(&dev_priv->mm.gtt_list);
 	for (i = 0; i < I915_NUM_RINGS; i++)
@@ -3909,18 +4010,6 @@ void i915_gem_release(struct drm_device *dev, struct drm_file *file)
 }
 
 static int
-i915_gpu_is_active(struct drm_device *dev)
-{
-	drm_i915_private_t *dev_priv = dev->dev_private;
-	int lists_empty;
-
-	lists_empty = list_empty(&dev_priv->mm.flushing_list) &&
-		      list_empty(&dev_priv->mm.active_list);
-
-	return !lists_empty;
-}
-
-static int
 i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc)
 {
 	struct drm_i915_private *dev_priv =
@@ -3928,60 +4017,25 @@ i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc)
 			     struct drm_i915_private,
 			     mm.inactive_shrinker);
 	struct drm_device *dev = dev_priv->dev;
-	struct drm_i915_gem_object *obj, *next;
+	struct drm_i915_gem_object *obj;
 	int nr_to_scan = sc->nr_to_scan;
 	int cnt;
 
 	if (!mutex_trylock(&dev->struct_mutex))
 		return 0;
 
-	/* "fast-path" to count number of available objects */
-	if (nr_to_scan == 0) {
-		cnt = 0;
-		list_for_each_entry(obj,
-				    &dev_priv->mm.inactive_list,
-				    mm_list)
-			cnt++;
-		mutex_unlock(&dev->struct_mutex);
-		return cnt / 100 * sysctl_vfs_cache_pressure;
-	}
-
-rescan:
-	/* first scan for clean buffers */
-	i915_gem_retire_requests(dev);
-
-	list_for_each_entry_safe(obj, next,
-				 &dev_priv->mm.inactive_list,
-				 mm_list) {
-		if (i915_gem_object_is_purgeable(obj)) {
-			if (i915_gem_object_unbind(obj) == 0 &&
-			    --nr_to_scan == 0)
-				break;
-		}
+	if (nr_to_scan) {
+		nr_to_scan -= i915_gem_purge(dev_priv, nr_to_scan);
+		if (nr_to_scan > 0)
+			i915_gem_shrink_all(dev_priv);
 	}
 
-	/* second pass, evict/count anything still on the inactive list */
 	cnt = 0;
-	list_for_each_entry_safe(obj, next,
-				 &dev_priv->mm.inactive_list,
-				 mm_list) {
-		if (nr_to_scan &&
-		    i915_gem_object_unbind(obj) == 0)
-			nr_to_scan--;
-		else
-			cnt++;
-	}
+	list_for_each_entry(obj, &dev_priv->mm.unbound_list, mm_list)
+		cnt += obj->base.size >> PAGE_SHIFT;
+	list_for_each_entry(obj, &dev_priv->mm.inactive_list, mm_list)
+		cnt += obj->base.size >> PAGE_SHIFT;
 
-	if (nr_to_scan && i915_gpu_is_active(dev)) {
-		/*
-		 * We are desperate for pages, so as a last resort, wait
-		 * for the GPU to finish and discard whatever we can.
-		 * This has a dramatic impact to reduce the number of
-		 * OOM-killer events whilst running the GPU aggressively.
-		 */
-		if (i915_gpu_idle(dev, true) == 0)
-			goto rescan;
-	}
 	mutex_unlock(&dev->struct_mutex);
 	return cnt / 100 * sysctl_vfs_cache_pressure;
 }
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 91ebb94..99e7696 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -163,7 +163,7 @@ found:
 }
 
 int
-i915_gem_evict_everything(struct drm_device *dev, bool purgeable_only)
+i915_gem_evict_everything(struct drm_device *dev)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct drm_i915_gem_object *obj, *next;
@@ -176,7 +176,7 @@ i915_gem_evict_everything(struct drm_device *dev, bool purgeable_only)
 	if (lists_empty)
 		return -ENOSPC;
 
-	trace_i915_gem_evict_everything(dev, purgeable_only);
+	trace_i915_gem_evict_everything(dev);
 
 	/* Flush everything (on to the inactive lists) and evict */
 	ret = i915_gpu_idle(dev, true);
@@ -187,12 +187,9 @@ i915_gem_evict_everything(struct drm_device *dev, bool purgeable_only)
 
 	/* Having flushed everything, unbind() should never raise an error */
 	list_for_each_entry_safe(obj, next,
-				 &dev_priv->mm.inactive_list, mm_list) {
-		if (!purgeable_only || obj->madv != I915_MADV_WILLNEED) {
-			if (obj->pin_count == 0)
-				WARN_ON(i915_gem_object_unbind(obj));
-		}
-	}
+				 &dev_priv->mm.inactive_list, mm_list)
+		if (obj->pin_count == 0)
+			WARN_ON(i915_gem_object_unbind(obj));
 
 	return ret;
 }
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 68ec013..2091c86 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -675,17 +675,12 @@ i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,
 			}
 		}
 
-		if (ret != -ENOSPC || retry > 1)
+		if (ret != -ENOSPC || retry++)
 			return ret;
 
-		/* First attempt, just clear anything that is purgeable.
-		 * Second attempt, clear the entire GTT.
-		 */
-		ret = i915_gem_evict_everything(ring->dev, retry == 0);
+		ret = i915_gem_evict_everything(ring->dev);
 		if (ret)
 			return ret;
-
-		retry++;
 	} while (1);
 
 err:
diff --git a/drivers/gpu/drm/i915/i915_trace.h b/drivers/gpu/drm/i915/i915_trace.h
index dac7bba..b6ce36a 100644
--- a/drivers/gpu/drm/i915/i915_trace.h
+++ b/drivers/gpu/drm/i915/i915_trace.h
@@ -214,22 +214,18 @@ TRACE_EVENT(i915_gem_evict,
 );
 
 TRACE_EVENT(i915_gem_evict_everything,
-	    TP_PROTO(struct drm_device *dev, bool purgeable),
-	    TP_ARGS(dev, purgeable),
+	    TP_PROTO(struct drm_device *dev),
+	    TP_ARGS(dev),
 
 	    TP_STRUCT__entry(
 			     __field(u32, dev)
-			     __field(bool, purgeable)
 			    ),
 
 	    TP_fast_assign(
 			   __entry->dev = dev->primary->index;
-			   __entry->purgeable = purgeable;
 			  ),
 
-	    TP_printk("dev=%d%s",
-		      __entry->dev,
-		      __entry->purgeable ? ", purgeable only" : "")
+	    TP_printk("dev=%d", __entry->dev)
 );
 
 TRACE_EVENT(i915_gem_ring_dispatch,
-- 
1.7.10

[PATCH 04/28] drm/i915: Only pwrite through the GTT if there is space in the aperture

[Chris Wilson]

Avoid stalling and waiting for the GPU by checking to see if there is
sufficient inactive space in the aperture for us to bind the buffer
prior to writing through the GTT. If there is inadequate space we will
have to stall waiting for the GPU, and incur overheads moving objects
about. Instead, only incur the clflush overhead on the target object by
writing through shmem.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h            |    6 ++++--
 drivers/gpu/drm/i915/i915_gem.c            |   29 +++++++++++++++-------------
 drivers/gpu/drm/i915/i915_gem_evict.c      |    7 ++++++-
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |    2 +-
 drivers/gpu/drm/i915/intel_overlay.c       |    2 +-
 drivers/gpu/drm/i915/intel_pm.c            |    2 +-
 drivers/gpu/drm/i915/intel_ringbuffer.c    |    6 +++---
 7 files changed, 32 insertions(+), 22 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 866b280..0cac7a0 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1237,7 +1237,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
 void i915_gem_free_object(struct drm_gem_object *obj);
 int __must_check i915_gem_object_pin(struct drm_i915_gem_object *obj,
 				     uint32_t alignment,
-				     bool map_and_fenceable);
+				     bool map_and_fenceable,
+				     bool nonblocking);
 void i915_gem_object_unpin(struct drm_i915_gem_object *obj);
 int __must_check i915_gem_object_unbind(struct drm_i915_gem_object *obj);
 void i915_gem_release_mmap(struct drm_i915_gem_object *obj);
@@ -1362,7 +1363,8 @@ void i915_gem_init_global_gtt(struct drm_device *dev,
 
 /* i915_gem_evict.c */
 int __must_check i915_gem_evict_something(struct drm_device *dev, int min_size,
-					  unsigned alignment, bool mappable);
+					  unsigned alignment, bool mappable,
+					  bool nonblock);
 int i915_gem_evict_everything(struct drm_device *dev);
 
 /* i915_gem_tiling.c */
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 761af95..31a601d 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -41,7 +41,8 @@ static void i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *o
 static void i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj);
 static __must_check int i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj,
 						    unsigned alignment,
-						    bool map_and_fenceable);
+						    bool map_and_fenceable,
+						    bool nonblocking);
 static int i915_gem_phys_pwrite(struct drm_device *dev,
 				struct drm_i915_gem_object *obj,
 				struct drm_i915_gem_pwrite *args,
@@ -591,7 +592,7 @@ i915_gem_gtt_pwrite_fast(struct drm_device *dev,
 	char __user *user_data;
 	int page_offset, page_length, ret;
 
-	ret = i915_gem_object_pin(obj, 0, true);
+	ret = i915_gem_object_pin(obj, 0, true, true);
 	if (ret)
 		goto out;
 
@@ -899,10 +900,8 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
 		goto out;
 	}
 
-	if (obj->gtt_space &&
-	    obj->cache_level == I915_CACHE_NONE &&
+	if (obj->cache_level == I915_CACHE_NONE &&
 	    obj->tiling_mode == I915_TILING_NONE &&
-	    obj->map_and_fenceable &&
 	    obj->base.write_domain != I915_GEM_DOMAIN_CPU) {
 		ret = i915_gem_gtt_pwrite_fast(dev, obj, args, file);
 		/* Note that the gtt paths might fail with non-page-backed user
@@ -910,7 +909,7 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data,
 		 * textures). Fallback to the shmem path in that case. */
 	}
 
-	if (ret == -EFAULT)
+	if (ret == -EFAULT || ret == -ENOSPC)
 		ret = i915_gem_shmem_pwrite(dev, obj, args, file);
 
 out:
@@ -1092,7 +1091,7 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 			goto unlock;
 	}
 	if (!obj->gtt_space) {
-		ret = i915_gem_object_bind_to_gtt(obj, 0, true);
+		ret = i915_gem_object_bind_to_gtt(obj, 0, true, false);
 		if (ret)
 			goto unlock;
 
@@ -2620,7 +2619,8 @@ i915_gem_object_get_fence(struct drm_i915_gem_object *obj)
 static int
 i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj,
 			    unsigned alignment,
-			    bool map_and_fenceable)
+			    bool map_and_fenceable,
+			    bool nonblocking)
 {
 	struct drm_device *dev = obj->base.dev;
 	drm_i915_private_t *dev_priv = dev->dev_private;
@@ -2692,7 +2692,8 @@ i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj,
 	}
 	if (obj->gtt_space == NULL) {
 		ret = i915_gem_evict_something(dev, size, alignment,
-					       map_and_fenceable);
+					       map_and_fenceable,
+					       nonblocking);
 		if (ret)
 			return ret;
 
@@ -2974,7 +2975,7 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj,
 	 * (e.g. libkms for the bootup splash), we have to ensure that we
 	 * always use map_and_fenceable for all scanout buffers.
 	 */
-	ret = i915_gem_object_pin(obj, alignment, true);
+	ret = i915_gem_object_pin(obj, alignment, true, false);
 	if (ret)
 		return ret;
 
@@ -3144,7 +3145,8 @@ i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file)
 int
 i915_gem_object_pin(struct drm_i915_gem_object *obj,
 		    uint32_t alignment,
-		    bool map_and_fenceable)
+		    bool map_and_fenceable,
+		    bool nonblocking)
 {
 	int ret;
 
@@ -3168,7 +3170,8 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj,
 
 	if (obj->gtt_space == NULL) {
 		ret = i915_gem_object_bind_to_gtt(obj, alignment,
-						  map_and_fenceable);
+						  map_and_fenceable,
+						  nonblocking);
 		if (ret)
 			return ret;
 	}
@@ -3226,7 +3229,7 @@ i915_gem_pin_ioctl(struct drm_device *dev, void *data,
 	obj->user_pin_count++;
 	obj->pin_filp = file;
 	if (obj->user_pin_count == 1) {
-		ret = i915_gem_object_pin(obj, args->alignment, true);
+		ret = i915_gem_object_pin(obj, args->alignment, true, false);
 		if (ret)
 			goto out;
 	}
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 99e7696..96fd8a6 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -44,7 +44,8 @@ mark_free(struct drm_i915_gem_object *obj, struct list_head *unwind)
 
 int
 i915_gem_evict_something(struct drm_device *dev, int min_size,
-			 unsigned alignment, bool mappable)
+			 unsigned alignment, bool mappable,
+			 bool nonblocking)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct list_head eviction_list, unwind_list;
@@ -90,6 +91,9 @@ i915_gem_evict_something(struct drm_device *dev, int min_size,
 			goto found;
 	}
 
+	if (nonblocking)
+		goto none;
+
 	/* Now merge in the soon-to-be-expired objects... */
 	list_for_each_entry(obj, &dev_priv->mm.active_list, mm_list) {
 		/* Does the object require an outstanding flush? */
@@ -113,6 +117,7 @@ i915_gem_evict_something(struct drm_device *dev, int min_size,
 			goto found;
 	}
 
+none:
 	/* Nothing found, clean up and bail out! */
 	while (!list_empty(&unwind_list)) {
 		obj = list_first_entry(&unwind_list,
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 2091c86..017190e 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -528,7 +528,7 @@ pin_and_fence_object(struct drm_i915_gem_object *obj,
 		obj->tiling_mode != I915_TILING_NONE;
 	need_mappable = need_fence || need_reloc_mappable(obj);
 
-	ret = i915_gem_object_pin(obj, entry->alignment, need_mappable);
+	ret = i915_gem_object_pin(obj, entry->alignment, need_mappable, false);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/i915/intel_overlay.c b/drivers/gpu/drm/i915/intel_overlay.c
index 0f0fe31..9845cfc 100644
--- a/drivers/gpu/drm/i915/intel_overlay.c
+++ b/drivers/gpu/drm/i915/intel_overlay.c
@@ -1441,7 +1441,7 @@ void intel_setup_overlay(struct drm_device *dev)
 		}
 		overlay->flip_addr = reg_bo->phys_obj->handle->busaddr;
 	} else {
-		ret = i915_gem_object_pin(reg_bo, PAGE_SIZE, true);
+		ret = i915_gem_object_pin(reg_bo, PAGE_SIZE, true, false);
 		if (ret) {
 			DRM_ERROR("failed to pin overlay register bo\n");
 			goto out_free_bo;
diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index a26bf49..cbb7217 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -1994,7 +1994,7 @@ intel_alloc_context_page(struct drm_device *dev)
 		return NULL;
 	}
 
-	ret = i915_gem_object_pin(ctx, 4096, true);
+	ret = i915_gem_object_pin(ctx, 4096, true, false);
 	if (ret) {
 		DRM_ERROR("failed to pin power context: %d\n", ret);
 		goto err_unref;
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index f797613..2a3fcd6 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -354,7 +354,7 @@ init_pipe_control(struct intel_ring_buffer *ring)
 
 	i915_gem_object_set_cache_level(obj, I915_CACHE_LLC);
 
-	ret = i915_gem_object_pin(obj, 4096, true);
+	ret = i915_gem_object_pin(obj, 4096, true, false);
 	if (ret)
 		goto err_unref;
 
@@ -913,7 +913,7 @@ static int init_status_page(struct intel_ring_buffer *ring)
 
 	i915_gem_object_set_cache_level(obj, I915_CACHE_LLC);
 
-	ret = i915_gem_object_pin(obj, 4096, true);
+	ret = i915_gem_object_pin(obj, 4096, true, false);
 	if (ret != 0) {
 		goto err_unref;
 	}
@@ -971,7 +971,7 @@ static int intel_init_ring_buffer(struct drm_device *dev,
 
 	ring->obj = obj;
 
-	ret = i915_gem_object_pin(obj, PAGE_SIZE, true);
+	ret = i915_gem_object_pin(obj, PAGE_SIZE, true, false);
 	if (ret)
 		goto err_unref;
 
-- 
1.7.10

[PATCH 05/28] drm/i915: Try harder to allocate an mmap_offset

[Chris Wilson]

Given the persistence of an offset for the lifetime of an object, itis
easy to contemplate how the mmap space becomes badly fragmented to the
point that further allocations fail with ENOSPC. Our only recourse at
this point is to try to purge the objects to release some space and
reattempt the allocation.

References: https://bugs.freedesktop.org/show_bug.cgi?id=39552
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c |   50 ++++++++++++++++++++++++++++++++-------
 1 file changed, 41 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 31a601d..af89ba4 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1247,6 +1247,42 @@ i915_gem_get_unfenced_gtt_alignment(struct drm_device *dev,
 	return i915_gem_get_gtt_size(dev, size, tiling_mode);
 }
 
+static int i915_gem_object_create_mmap_offset(struct drm_i915_gem_object *obj)
+{
+	struct drm_i915_private *dev_priv = obj->base.dev->dev_private;
+	int ret;
+
+	if (obj->base.map_list.map)
+		return 0;
+
+	ret = drm_gem_create_mmap_offset(&obj->base);
+	if (ret != -ENOSPC)
+		return ret;
+
+	/* Badly fragmented mmap space? The only way we can recover
+	 * space is by destroying unwanted objects. We can't randomly release
+	 * mmap_offsets as userspace expects them to be persistent for the
+	 * lifetime of the objects. The closest we can is to release the
+	 * offsets on purgeable objects by truncating it and marking it purged,
+	 * which prevents userspace from ever using that object again.
+	 */
+	i915_gem_purge(dev_priv, obj->base.size >> PAGE_SHIFT);
+	ret = drm_gem_create_mmap_offset(&obj->base);
+	if (ret != -ENOSPC)
+		return ret;
+
+	i915_gem_shrink_all(dev_priv);
+	return drm_gem_create_mmap_offset(&obj->base);
+}
+
+static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj)
+{
+	if (!obj->base.map_list.map)
+		return;
+
+	drm_gem_free_mmap_offset(&obj->base);
+}
+
 int
 i915_gem_mmap_gtt(struct drm_file *file,
 		  struct drm_device *dev,
@@ -1281,11 +1317,9 @@ i915_gem_mmap_gtt(struct drm_file *file,
 		goto out;
 	}
 
-	if (!obj->base.map_list.map) {
-		ret = drm_gem_create_mmap_offset(&obj->base);
-		if (ret)
-			goto out;
-	}
+	ret = i915_gem_object_create_mmap_offset(obj);
+	if (ret)
+		goto out;
 
 	*offset = (u64)obj->base.map_list.hash.key << PAGE_SHIFT;
 
@@ -1337,8 +1371,7 @@ i915_gem_object_truncate(struct drm_i915_gem_object *obj)
 	inode = obj->base.filp->f_path.dentry->d_inode;
 	shmem_truncate_range(inode, 0, (loff_t)-1);
 
-	if (obj->base.map_list.map)
-		drm_gem_free_mmap_offset(&obj->base);
+	i915_gem_object_free_mmap_offset(obj);
 
 	obj->madv = __I915_MADV_PURGED;
 }
@@ -3488,8 +3521,7 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)
 	}
 
 	i915_gem_object_put_pages_gtt(obj);
-	if (obj->base.map_list.map)
-		drm_gem_free_mmap_offset(&obj->base);
+	i915_gem_object_free_mmap_offset(obj);
 
 	drm_gem_object_release(&obj->base);
 	i915_gem_info_remove_obj(dev_priv, obj->base.size);
-- 
1.7.10

[PATCH 06/28] drm/i915: Pin backing pages for pwrite

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c |   44 ++++++++++++++++++---------------------
 1 file changed, 20 insertions(+), 24 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index af89ba4..c81962b 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -60,6 +60,8 @@ static long i915_gem_purge(struct drm_i915_private *dev_priv, long target);
 static void i915_gem_shrink_all(struct drm_i915_private *dev_priv);
 static void i915_gem_object_truncate(struct drm_i915_gem_object *obj);
 
+static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj);
+
 static inline void i915_gem_object_fence_lost(struct drm_i915_gem_object *obj)
 {
 	if (obj->tiling_mode)
@@ -672,7 +674,7 @@ shmem_pwrite_fast(struct page *page, int shmem_page_offset, int page_length,
 				       page_length);
 	kunmap_atomic(vaddr);
 
-	return ret;
+	return ret ? -EFAULT : 0;
 }
 
 /* Only difference to the fast-path function is that this can handle bit17
@@ -706,7 +708,7 @@ shmem_pwrite_slow(struct page *page, int shmem_page_offset, int page_length,
 					     page_do_bit17_swizzling);
 	kunmap(page);
 
-	return ret;
+	return ret ? -EFAULT : 0;
 }
 
 static int
@@ -715,7 +717,6 @@ i915_gem_shmem_pwrite(struct drm_device *dev,
 		      struct drm_i915_gem_pwrite *args,
 		      struct drm_file *file)
 {
-	struct address_space *mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping;
 	ssize_t remain;
 	loff_t offset;
 	char __user *user_data;
@@ -724,7 +725,6 @@ i915_gem_shmem_pwrite(struct drm_device *dev,
 	int hit_slowpath = 0;
 	int needs_clflush_after = 0;
 	int needs_clflush_before = 0;
-	int release_page;
 
 	user_data = (char __user *) (uintptr_t) args->data_ptr;
 	remain = args->size;
@@ -750,6 +750,10 @@ i915_gem_shmem_pwrite(struct drm_device *dev,
 	    && obj->cache_level == I915_CACHE_NONE)
 		needs_clflush_before = 1;
 
+	ret = i915_gem_object_get_pages_gtt(obj);
+	if (ret)
+		return ret;
+
 	offset = args->offset;
 	obj->dirty = 1;
 
@@ -775,17 +779,8 @@ i915_gem_shmem_pwrite(struct drm_device *dev,
 			((shmem_page_offset | page_length)
 				& (boot_cpu_data.x86_clflush_size - 1));
 
-		if (obj->pages) {
-			page = obj->pages[offset >> PAGE_SHIFT];
-			release_page = 0;
-		} else {
-			page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
-			if (IS_ERR(page)) {
-				ret = PTR_ERR(page);
-				goto out;
-			}
-			release_page = 1;
-		}
+		page = obj->pages[offset >> PAGE_SHIFT];
+		page_cache_get(page);
 
 		page_do_bit17_swizzling = obj_do_bit17_swizzling &&
 			(page_to_phys(page) & (1 << 17)) != 0;
@@ -797,27 +792,25 @@ i915_gem_shmem_pwrite(struct drm_device *dev,
 		if (ret == 0)
 			goto next_page;
 
-		hit_slowpath = 1;
-		page_cache_get(page);
 		mutex_unlock(&dev->struct_mutex);
-
 		ret = shmem_pwrite_slow(page, shmem_page_offset, page_length,
 					user_data, page_do_bit17_swizzling,
 					partial_cacheline_write,
 					needs_clflush_after);
 
 		mutex_lock(&dev->struct_mutex);
-		page_cache_release(page);
+		hit_slowpath = 1;
+
+		if (ret == 0)
+			ret = i915_gem_object_get_pages_gtt(obj);
+
 next_page:
 		set_page_dirty(page);
 		mark_page_accessed(page);
-		if (release_page)
-			page_cache_release(page);
+		page_cache_release(page);
 
-		if (ret) {
-			ret = -EFAULT;
+		if (ret)
 			goto out;
-		}
 
 		remain -= page_length;
 		user_data += page_length;
@@ -1541,6 +1534,9 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
 	if (obj->pages)
 		return 0;
 
+	if (obj->madv != I915_MADV_WILLNEED)
+		return -EINVAL;
+
 	/* Assert that the object is not currently in any GPU domain. As it
 	 * wasn't in the GTT, there shouldn't be any way it could have been in
 	 * a GPU cache
-- 
1.7.10

[PATCH 07/28] drm/i915: Pin pages for pread

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c |   38 +++++++++++++++-----------------------
 1 file changed, 15 insertions(+), 23 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index c81962b..e7aa44b 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -335,7 +335,7 @@ shmem_pread_fast(struct page *page, int shmem_page_offset, int page_length,
 				      page_length);
 	kunmap_atomic(vaddr);
 
-	return ret;
+	return ret ? -EFAULT : 0;
 }
 
 static void
@@ -386,7 +386,7 @@ shmem_pread_slow(struct page *page, int shmem_page_offset, int page_length,
 				     page_length);
 	kunmap(page);
 
-	return ret;
+	return ret ? - EFAULT : 0;
 }
 
 static int
@@ -395,7 +395,6 @@ i915_gem_shmem_pread(struct drm_device *dev,
 		     struct drm_i915_gem_pread *args,
 		     struct drm_file *file)
 {
-	struct address_space *mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping;
 	char __user *user_data;
 	ssize_t remain;
 	loff_t offset;
@@ -404,7 +403,6 @@ i915_gem_shmem_pread(struct drm_device *dev,
 	int hit_slowpath = 0;
 	int prefaulted = 0;
 	int needs_clflush = 0;
-	int release_page;
 
 	user_data = (char __user *) (uintptr_t) args->data_ptr;
 	remain = args->size;
@@ -425,6 +423,10 @@ i915_gem_shmem_pread(struct drm_device *dev,
 		}
 	}
 
+	ret = i915_gem_object_get_pages_gtt(obj);
+	if (ret)
+		return ret;
+
 	offset = args->offset;
 
 	while (remain > 0) {
@@ -440,17 +442,8 @@ i915_gem_shmem_pread(struct drm_device *dev,
 		if ((shmem_page_offset + page_length) > PAGE_SIZE)
 			page_length = PAGE_SIZE - shmem_page_offset;
 
-		if (obj->pages) {
-			page = obj->pages[offset >> PAGE_SHIFT];
-			release_page = 0;
-		} else {
-			page = shmem_read_mapping_page(mapping, offset >> PAGE_SHIFT);
-			if (IS_ERR(page)) {
-				ret = PTR_ERR(page);
-				goto out;
-			}
-			release_page = 1;
-		}
+		page = obj->pages[offset >> PAGE_SHIFT];
+		page_cache_get(page);
 
 		page_do_bit17_swizzling = obj_do_bit17_swizzling &&
 			(page_to_phys(page) & (1 << 17)) != 0;
@@ -461,8 +454,6 @@ i915_gem_shmem_pread(struct drm_device *dev,
 		if (ret == 0)
 			goto next_page;
 
-		hit_slowpath = 1;
-		page_cache_get(page);
 		mutex_unlock(&dev->struct_mutex);
 
 		if (!prefaulted) {
@@ -479,17 +470,18 @@ i915_gem_shmem_pread(struct drm_device *dev,
 				       user_data, page_do_bit17_swizzling,
 				       needs_clflush);
 
+		hit_slowpath = 1;
 		mutex_lock(&dev->struct_mutex);
-		page_cache_release(page);
+
+		if (ret == 0)
+			ret = i915_gem_object_get_pages_gtt(obj);
+
 next_page:
 		mark_page_accessed(page);
-		if (release_page)
-			page_cache_release(page);
+		page_cache_release(page);
 
-		if (ret) {
-			ret = -EFAULT;
+		if (ret)
 			goto out;
-		}
 
 		remain -= page_length;
 		user_data += page_length;
-- 
1.7.10

[PATCH 08/28] drm/i915/debugfs: Show (count, size) of purgeable objects in i915_gem_objects

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c |   19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index de7c65e..1e12c72 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -218,8 +218,8 @@ static int i915_gem_object_info(struct seq_file *m, void* data)
 	struct drm_info_node *node = (struct drm_info_node *) m->private;
 	struct drm_device *dev = node->minor->dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
-	u32 count, mappable_count;
-	size_t size, mappable_size;
+	u32 count, mappable_count, purgeable_count;
+	size_t size, mappable_size, purgeable_size;
 	struct drm_i915_gem_object *obj;
 	int ret;
 
@@ -247,9 +247,12 @@ static int i915_gem_object_info(struct seq_file *m, void* data)
 	seq_printf(m, "  %u [%u] inactive objects, %zu [%zu] bytes\n",
 		   count, mappable_count, size, mappable_size);
 
-	size = count = 0;
-	list_for_each_entry(obj, &dev_priv->mm.unbound_list, mm_list)
+	size = count = purgeable_size = purgeable_count = 0;
+	list_for_each_entry(obj, &dev_priv->mm.unbound_list, mm_list) {
 		size += obj->base.size, ++count;
+		if (obj->madv == I915_MADV_DONTNEED)
+			purgeable_size += obj->base.size, ++purgeable_count;
+	}
 	seq_printf(m, "%u unbound objects, %zu bytes\n", count, size);
 
 	size = count = mappable_size = mappable_count = 0;
@@ -259,10 +262,16 @@ static int i915_gem_object_info(struct seq_file *m, void* data)
 			++count;
 		}
 		if (obj->pin_mappable) {
-			mappable_size += obj->gtt_space->size;
+			mappable_size += obj->gtt_space->size,
 			++mappable_count;
 		}
+		if (obj->madv == I915_MADV_DONTNEED) {
+			purgeable_size += obj->base.size;
+			++purgeable_count;
+		}
 	}
+	seq_printf(m, "%u purgeable objects, %zu bytes\n",
+		   purgeable_count, purgeable_size);
 	seq_printf(m, "%u pinned mappable objects, %zu bytes\n",
 		   mappable_count, mappable_size);
 	seq_printf(m, "%u fault mappable objects, %zu bytes\n",
-- 
1.7.10

[PATCH 09/28] drm: Introduce drm_mm_create_block()

[Chris Wilson]

To be used later by i915 to preallocate exact blocks of space from the
range manager.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Dave Airlie <airlied@redhat.com>
---
 drivers/gpu/drm/drm_mm.c |   49 ++++++++++++++++++++++++++++++++++++++++++++++
 include/drm/drm_mm.h     |    4 ++++
 2 files changed, 53 insertions(+)

diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c
index 961fb54..76df409 100644
--- a/drivers/gpu/drm/drm_mm.c
+++ b/drivers/gpu/drm/drm_mm.c
@@ -154,6 +154,55 @@ static void drm_mm_insert_helper(struct drm_mm_node *hole_node,
 	}
 }
 
+struct drm_mm_node *drm_mm_create_block(struct drm_mm *mm,
+					unsigned long start,
+					unsigned long size,
+					bool atomic)
+{
+	struct drm_mm_node *hole, *node;
+	unsigned long end = start + size;
+
+	list_for_each_entry(hole, &mm->hole_stack, hole_stack) {
+		unsigned long hole_start;
+		unsigned long hole_end;
+
+		BUG_ON(!hole->hole_follows);
+		hole_start = drm_mm_hole_node_start(hole);
+		hole_end = drm_mm_hole_node_end(hole);
+
+		if (hole_start > start || hole_end < end)
+			continue;
+
+		node = drm_mm_kmalloc(mm, atomic);
+		if (unlikely(node == NULL))
+			return NULL;
+
+		node->start = start;
+		node->size = size;
+		node->mm = mm;
+		node->allocated = 1;
+
+		INIT_LIST_HEAD(&node->hole_stack);
+		list_add(&node->node_list, &hole->node_list);
+
+		if (start == hole_start) {
+			hole->hole_follows = 0;
+			list_del_init(&hole->hole_stack);
+		}
+
+		node->hole_follows = 0;
+		if (end != hole_end) {
+			list_add(&node->hole_stack, &mm->hole_stack);
+			node->hole_follows = 1;
+		}
+
+		return node;
+	}
+
+	return NULL;
+}
+EXPORT_SYMBOL(drm_mm_create_block);
+
 struct drm_mm_node *drm_mm_get_block_generic(struct drm_mm_node *hole_node,
 					     unsigned long size,
 					     unsigned alignment,
diff --git a/include/drm/drm_mm.h b/include/drm/drm_mm.h
index 564b14a..1775fa4 100644
--- a/include/drm/drm_mm.h
+++ b/include/drm/drm_mm.h
@@ -97,6 +97,10 @@ static inline bool drm_mm_initialized(struct drm_mm *mm)
 /*
  * Basic range manager support (drm_mm.c)
  */
+extern struct drm_mm_node *drm_mm_create_block(struct drm_mm *mm,
+					       unsigned long start,
+					       unsigned long size,
+					       bool atomic);
 extern struct drm_mm_node *drm_mm_get_block_generic(struct drm_mm_node *node,
 						    unsigned long size,
 						    unsigned alignment,
-- 
1.7.10

[PATCH 10/28] drm/i915: Split the stolen handling for GEM out of i915_dma.c

[Chris Wilson]

We slightly modify the initialisation sequence to move the
initialisation of the memory managers earlier and in particular before
probing outputs and detecting any existing output configuration. This is
essential if we wish to track preallocated objects and preserve them
whilst initialising GEM.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/Makefile          |    1 +
 drivers/gpu/drm/i915/i915_dma.c        |  161 ++-----------------------
 drivers/gpu/drm/i915/i915_drv.h        |    4 +
 drivers/gpu/drm/i915/i915_gem_stolen.c |  202 ++++++++++++++++++++++++++++++++
 4 files changed, 219 insertions(+), 149 deletions(-)
 create mode 100644 drivers/gpu/drm/i915/i915_gem_stolen.c

diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile
index b65c06f..8b8bbc7 100644
--- a/drivers/gpu/drm/i915/Makefile
+++ b/drivers/gpu/drm/i915/Makefile
@@ -11,6 +11,7 @@ i915-y := i915_drv.o i915_dma.o i915_irq.o \
 	  i915_gem_evict.o \
 	  i915_gem_execbuffer.o \
 	  i915_gem_gtt.o \
+	  i915_gem_stolen.o \
 	  i915_gem_tiling.o \
 	  i915_sysfs.o \
 	  i915_trace_points.o \
diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
index a813f65..e3f1488 100644
--- a/drivers/gpu/drm/i915/i915_dma.c
+++ b/drivers/gpu/drm/i915/i915_dma.c
@@ -1019,133 +1019,6 @@ intel_teardown_mchbar(struct drm_device *dev)
 		release_resource(&dev_priv->mch_res);
 }
 
-#define PTE_ADDRESS_MASK		0xfffff000
-#define PTE_ADDRESS_MASK_HIGH		0x000000f0 /* i915+ */
-#define PTE_MAPPING_TYPE_UNCACHED	(0 << 1)
-#define PTE_MAPPING_TYPE_DCACHE		(1 << 1) /* i830 only */
-#define PTE_MAPPING_TYPE_CACHED		(3 << 1)
-#define PTE_MAPPING_TYPE_MASK		(3 << 1)
-#define PTE_VALID			(1 << 0)
-
-/**
- * i915_stolen_to_phys - take an offset into stolen memory and turn it into
- *                       a physical one
- * @dev: drm device
- * @offset: address to translate
- *
- * Some chip functions require allocations from stolen space and need the
- * physical address of the memory in question.
- */
-static unsigned long i915_stolen_to_phys(struct drm_device *dev, u32 offset)
-{
-	struct drm_i915_private *dev_priv = dev->dev_private;
-	struct pci_dev *pdev = dev_priv->bridge_dev;
-	u32 base;
-
-#if 0
-	/* On the machines I have tested the Graphics Base of Stolen Memory
-	 * is unreliable, so compute the base by subtracting the stolen memory
-	 * from the Top of Low Usable DRAM which is where the BIOS places
-	 * the graphics stolen memory.
-	 */
-	if (INTEL_INFO(dev)->gen > 3 || IS_G33(dev)) {
-		/* top 32bits are reserved = 0 */
-		pci_read_config_dword(pdev, 0xA4, &base);
-	} else {
-		/* XXX presume 8xx is the same as i915 */
-		pci_bus_read_config_dword(pdev->bus, 2, 0x5C, &base);
-	}
-#else
-	if (INTEL_INFO(dev)->gen > 3 || IS_G33(dev)) {
-		u16 val;
-		pci_read_config_word(pdev, 0xb0, &val);
-		base = val >> 4 << 20;
-	} else {
-		u8 val;
-		pci_read_config_byte(pdev, 0x9c, &val);
-		base = val >> 3 << 27;
-	}
-	base -= dev_priv->mm.gtt->stolen_size;
-#endif
-
-	return base + offset;
-}
-
-static void i915_warn_stolen(struct drm_device *dev)
-{
-	DRM_ERROR("not enough stolen space for compressed buffer, disabling\n");
-	DRM_ERROR("hint: you may be able to increase stolen memory size in the BIOS to avoid this\n");
-}
-
-static void i915_setup_compression(struct drm_device *dev, int size)
-{
-	struct drm_i915_private *dev_priv = dev->dev_private;
-	struct drm_mm_node *compressed_fb, *uninitialized_var(compressed_llb);
-	unsigned long cfb_base;
-	unsigned long ll_base = 0;
-
-	/* Just in case the BIOS is doing something questionable. */
-	intel_disable_fbc(dev);
-
-	compressed_fb = drm_mm_search_free(&dev_priv->mm.stolen, size, 4096, 0);
-	if (compressed_fb)
-		compressed_fb = drm_mm_get_block(compressed_fb, size, 4096);
-	if (!compressed_fb)
-		goto err;
-
-	cfb_base = i915_stolen_to_phys(dev, compressed_fb->start);
-	if (!cfb_base)
-		goto err_fb;
-
-	if (!(IS_GM45(dev) || HAS_PCH_SPLIT(dev))) {
-		compressed_llb = drm_mm_search_free(&dev_priv->mm.stolen,
-						    4096, 4096, 0);
-		if (compressed_llb)
-			compressed_llb = drm_mm_get_block(compressed_llb,
-							  4096, 4096);
-		if (!compressed_llb)
-			goto err_fb;
-
-		ll_base = i915_stolen_to_phys(dev, compressed_llb->start);
-		if (!ll_base)
-			goto err_llb;
-	}
-
-	dev_priv->cfb_size = size;
-
-	dev_priv->compressed_fb = compressed_fb;
-	if (HAS_PCH_SPLIT(dev))
-		I915_WRITE(ILK_DPFC_CB_BASE, compressed_fb->start);
-	else if (IS_GM45(dev)) {
-		I915_WRITE(DPFC_CB_BASE, compressed_fb->start);
-	} else {
-		I915_WRITE(FBC_CFB_BASE, cfb_base);
-		I915_WRITE(FBC_LL_BASE, ll_base);
-		dev_priv->compressed_llb = compressed_llb;
-	}
-
-	DRM_DEBUG_KMS("FBC base 0x%08lx, ll base 0x%08lx, size %dM\n",
-		      cfb_base, ll_base, size >> 20);
-	return;
-
-err_llb:
-	drm_mm_put_block(compressed_llb);
-err_fb:
-	drm_mm_put_block(compressed_fb);
-err:
-	dev_priv->no_fbc_reason = FBC_STOLEN_TOO_SMALL;
-	i915_warn_stolen(dev);
-}
-
-static void i915_cleanup_compression(struct drm_device *dev)
-{
-	struct drm_i915_private *dev_priv = dev->dev_private;
-
-	drm_mm_put_block(dev_priv->compressed_fb);
-	if (dev_priv->compressed_llb)
-		drm_mm_put_block(dev_priv->compressed_llb);
-}
-
 /* true = enable decode, false = disable decoder */
 static unsigned int i915_vga_set_decode(void *cookie, bool state)
 {
@@ -1207,16 +1080,12 @@ intel_enable_ppgtt(struct drm_device *dev)
 static int i915_load_gem_init(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
-	unsigned long prealloc_size, gtt_size, mappable_size;
+	unsigned long gtt_size, mappable_size;
 	int ret;
 
-	prealloc_size = dev_priv->mm.gtt->stolen_size;
 	gtt_size = dev_priv->mm.gtt->gtt_total_entries << PAGE_SHIFT;
 	mappable_size = dev_priv->mm.gtt->gtt_mappable_entries << PAGE_SHIFT;
 
-	/* Basic memrange allocator for stolen space */
-	drm_mm_init(&dev_priv->mm.stolen, 0, prealloc_size);
-
 	mutex_lock(&dev->struct_mutex);
 	if (intel_enable_ppgtt(dev) && HAS_ALIASING_PPGTT(dev)) {
 		/* PPGTT pdes are stolen from global gtt ptes, so shrink the
@@ -1252,20 +1121,6 @@ static int i915_load_gem_init(struct drm_device *dev)
 		return ret;
 	}
 
-	/* Try to set up FBC with a reasonable compressed buffer size */
-	if (I915_HAS_FBC(dev) && i915_powersave) {
-		int cfb_size;
-
-		/* Leave 1M for line length buffer & misc. */
-
-		/* Try to get a 32M buffer... */
-		if (prealloc_size > (36*1024*1024))
-			cfb_size = 32*1024*1024;
-		else /* fall back to 7/8 of the stolen space */
-			cfb_size = prealloc_size * 7 / 8;
-		i915_setup_compression(dev, cfb_size);
-	}
-
 	/* Allow hardware batchbuffers unless told otherwise. */
 	dev_priv->allow_batchbuffer = 1;
 	return 0;
@@ -1304,11 +1159,18 @@ static int i915_load_modeset_init(struct drm_device *dev)
 	if (IS_GEN3(dev) && (I915_READ(ECOSKPD) & ECO_FLIP_DONE))
 		dev_priv->flip_pending_is_done = true;
 
+	/* Initialise stolen first so that we may reserve preallocated
+	 * objects for the BIOS to KMS transition.
+	 */
+	ret = i915_gem_init_stolen(dev);
+	if (ret)
+		goto cleanup_vga_switcheroo;
+
 	intel_modeset_init(dev);
 
 	ret = i915_load_gem_init(dev);
 	if (ret)
-		goto cleanup_vga_switcheroo;
+		goto cleanup_gem_stolen;
 
 	intel_modeset_gem_init(dev);
 
@@ -1338,6 +1200,8 @@ cleanup_gem:
 	i915_gem_cleanup_ringbuffer(dev);
 	mutex_unlock(&dev->struct_mutex);
 	i915_gem_cleanup_aliasing_ppgtt(dev);
+cleanup_gem_stolen:
+	i915_gem_cleanup_stolen(dev);
 cleanup_vga_switcheroo:
 	vga_switcheroo_unregister_client(dev->pdev);
 cleanup_vga_client:
@@ -2252,8 +2116,7 @@ int i915_driver_unload(struct drm_device *dev)
 		i915_gem_cleanup_ringbuffer(dev);
 		mutex_unlock(&dev->struct_mutex);
 		i915_gem_cleanup_aliasing_ppgtt(dev);
-		if (I915_HAS_FBC(dev) && i915_powersave)
-			i915_cleanup_compression(dev);
+		i915_gem_cleanup_stolen(dev);
 		drm_mm_takedown(&dev_priv->mm.stolen);
 
 		intel_cleanup_overlay(dev);
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 0cac7a0..1147d28 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1367,6 +1367,10 @@ int __must_check i915_gem_evict_something(struct drm_device *dev, int min_size,
 					  bool nonblock);
 int i915_gem_evict_everything(struct drm_device *dev);
 
+/* i915_gem_stolen.c */
+int i915_gem_init_stolen(struct drm_device *dev);
+void i915_gem_cleanup_stolen(struct drm_device *dev);
+
 /* i915_gem_tiling.c */
 void i915_gem_detect_bit_6_swizzle(struct drm_device *dev);
 void i915_gem_object_do_bit_17_swizzle(struct drm_i915_gem_object *obj);
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
new file mode 100644
index 0000000..ada2e90
--- /dev/null
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -0,0 +1,202 @@
+/*
+ * Copyright © 2008-2012 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Eric Anholt <eric@anholt.net>
+ *    Chris Wilson <chris@chris-wilson.co.uk>
+ *
+ */
+
+#include "drmP.h"
+#include "drm.h"
+#include "i915_drm.h"
+#include "i915_drv.h"
+
+/*
+ * The BIOS typically reserves some of the system's memory for the exclusive
+ * use of the integrated graphics. This memory is no longer available for
+ * use by the OS and so the user finds that his system has less memory
+ * available than he put in. We refer to this memory as stolen.
+ *
+ * The BIOS will allocate its framebuffer from the stolen memory. Our
+ * goal is try to reuse that object for our own fbcon which must always
+ * be available for panics. Anything else we can reuse the stolen memory
+ * for is a boon.
+ */
+
+#define PTE_ADDRESS_MASK		0xfffff000
+#define PTE_ADDRESS_MASK_HIGH		0x000000f0 /* i915+ */
+#define PTE_MAPPING_TYPE_UNCACHED	(0 << 1)
+#define PTE_MAPPING_TYPE_DCACHE		(1 << 1) /* i830 only */
+#define PTE_MAPPING_TYPE_CACHED		(3 << 1)
+#define PTE_MAPPING_TYPE_MASK		(3 << 1)
+#define PTE_VALID			(1 << 0)
+
+/**
+ * i915_stolen_to_phys - take an offset into stolen memory and turn it into
+ *                       a physical one
+ * @dev: drm device
+ * @offset: address to translate
+ *
+ * Some chip functions require allocations from stolen space and need the
+ * physical address of the memory in question.
+ */
+static unsigned long i915_stolen_to_phys(struct drm_device *dev, u32 offset)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct pci_dev *pdev = dev_priv->bridge_dev;
+	u32 base;
+
+#if 0
+	/* On the machines I have tested the Graphics Base of Stolen Memory
+	 * is unreliable, so compute the base by subtracting the stolen memory
+	 * from the Top of Low Usable DRAM which is where the BIOS places
+	 * the graphics stolen memory.
+	 */
+	if (INTEL_INFO(dev)->gen > 3 || IS_G33(dev)) {
+		/* top 32bits are reserved = 0 */
+		pci_read_config_dword(pdev, 0xA4, &base);
+	} else {
+		/* XXX presume 8xx is the same as i915 */
+		pci_bus_read_config_dword(pdev->bus, 2, 0x5C, &base);
+	}
+#else
+	if (INTEL_INFO(dev)->gen > 3 || IS_G33(dev)) {
+		u16 val;
+		pci_read_config_word(pdev, 0xb0, &val);
+		base = val >> 4 << 20;
+	} else {
+		u8 val;
+		pci_read_config_byte(pdev, 0x9c, &val);
+		base = val >> 3 << 27;
+	}
+	base -= dev_priv->mm.gtt->stolen_size;
+#endif
+
+	return base + offset;
+}
+
+static void i915_warn_stolen(struct drm_device *dev)
+{
+	DRM_INFO("not enough stolen space for compressed buffer, disabling\n");
+	DRM_INFO("hint: you may be able to increase stolen memory size in the BIOS to avoid this\n");
+}
+
+static void i915_setup_compression(struct drm_device *dev, int size)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct drm_mm_node *compressed_fb, *uninitialized_var(compressed_llb);
+	unsigned long cfb_base;
+	unsigned long ll_base = 0;
+
+	/* Just in case the BIOS is doing something questionable. */
+	intel_disable_fbc(dev);
+
+	compressed_fb = drm_mm_search_free(&dev_priv->mm.stolen, size, 4096, 0);
+	if (compressed_fb)
+		compressed_fb = drm_mm_get_block(compressed_fb, size, 4096);
+	if (!compressed_fb)
+		goto err;
+
+	cfb_base = i915_stolen_to_phys(dev, compressed_fb->start);
+	if (!cfb_base)
+		goto err_fb;
+
+	if (!(IS_GM45(dev) || HAS_PCH_SPLIT(dev))) {
+		compressed_llb = drm_mm_search_free(&dev_priv->mm.stolen,
+						    4096, 4096, 0);
+		if (compressed_llb)
+			compressed_llb = drm_mm_get_block(compressed_llb,
+							  4096, 4096);
+		if (!compressed_llb)
+			goto err_fb;
+
+		ll_base = i915_stolen_to_phys(dev, compressed_llb->start);
+		if (!ll_base)
+			goto err_llb;
+	}
+
+	dev_priv->cfb_size = size;
+
+	dev_priv->compressed_fb = compressed_fb;
+	if (HAS_PCH_SPLIT(dev))
+		I915_WRITE(ILK_DPFC_CB_BASE, compressed_fb->start);
+	else if (IS_GM45(dev)) {
+		I915_WRITE(DPFC_CB_BASE, compressed_fb->start);
+	} else {
+		I915_WRITE(FBC_CFB_BASE, cfb_base);
+		I915_WRITE(FBC_LL_BASE, ll_base);
+		dev_priv->compressed_llb = compressed_llb;
+	}
+
+	DRM_DEBUG_KMS("FBC base 0x%08lx, ll base 0x%08lx, size %dM\n",
+		      cfb_base, ll_base, size >> 20);
+	return;
+
+err_llb:
+	drm_mm_put_block(compressed_llb);
+err_fb:
+	drm_mm_put_block(compressed_fb);
+err:
+	dev_priv->no_fbc_reason = FBC_STOLEN_TOO_SMALL;
+	i915_warn_stolen(dev);
+}
+
+static void i915_cleanup_compression(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+
+	drm_mm_put_block(dev_priv->compressed_fb);
+	if (dev_priv->compressed_llb)
+		drm_mm_put_block(dev_priv->compressed_llb);
+}
+
+void i915_gem_cleanup_stolen(struct drm_device *dev)
+{
+	if (I915_HAS_FBC(dev) && i915_powersave)
+		i915_cleanup_compression(dev);
+}
+
+int i915_gem_init_stolen(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	unsigned long prealloc_size = dev_priv->mm.gtt->stolen_size;
+
+	/* Basic memrange allocator for stolen space */
+	drm_mm_init(&dev_priv->mm.stolen, 0, prealloc_size);
+
+	/* Try to set up FBC with a reasonable compressed buffer size */
+	if (I915_HAS_FBC(dev) && i915_powersave) {
+		int cfb_size;
+
+		/* Leave 1M for line length buffer & misc. */
+
+		/* Try to get a 32M buffer... */
+		if (prealloc_size > (36*1024*1024))
+			cfb_size = 32*1024*1024;
+		else /* fall back to 7/8 of the stolen space */
+			cfb_size = prealloc_size * 7 / 8;
+		i915_setup_compression(dev, cfb_size);
+	}
+
+	return 0;
+}
-- 
1.7.10

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

[PATCH 11/28] drm/i915: Fix detection of stolen base for gen2

[Chris Wilson]

It was not until the G33 refresh, that a PCI config register was
introduced that explicitly said where the stolen memory was. Prior to
865G there was not even a register that said where the end of usable
low memory was and where the stolen memory began (or ended depending
upon chipset). Before then, one has to look at the BIOS memory maps to
find the Top of Memory. Alas that is not exported by arch/x86 and so we
have to resort to disabling stolen memory on gen2 for the time being.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h        |    1 +
 drivers/gpu/drm/i915/i915_gem_stolen.c |   69 ++++++++++++++------------------
 2 files changed, 31 insertions(+), 39 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 1147d28..1ae807d 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -646,6 +646,7 @@ typedef struct drm_i915_private {
 		unsigned long gtt_start;
 		unsigned long gtt_mappable_end;
 		unsigned long gtt_end;
+		unsigned long stolen_base; /* limited to low memory (32-bit) */
 
 		struct io_mapping *gtt_mapping;
 		int gtt_mtrr;
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index ada2e90..a01ff74 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -43,56 +43,43 @@
  * for is a boon.
  */
 
-#define PTE_ADDRESS_MASK		0xfffff000
-#define PTE_ADDRESS_MASK_HIGH		0x000000f0 /* i915+ */
-#define PTE_MAPPING_TYPE_UNCACHED	(0 << 1)
-#define PTE_MAPPING_TYPE_DCACHE		(1 << 1) /* i830 only */
-#define PTE_MAPPING_TYPE_CACHED		(3 << 1)
-#define PTE_MAPPING_TYPE_MASK		(3 << 1)
-#define PTE_VALID			(1 << 0)
-
-/**
- * i915_stolen_to_phys - take an offset into stolen memory and turn it into
- *                       a physical one
- * @dev: drm device
- * @offset: address to translate
- *
- * Some chip functions require allocations from stolen space and need the
- * physical address of the memory in question.
- */
-static unsigned long i915_stolen_to_phys(struct drm_device *dev, u32 offset)
+static unsigned long i915_stolen_to_physical(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct pci_dev *pdev = dev_priv->bridge_dev;
 	u32 base;
 
-#if 0
 	/* On the machines I have tested the Graphics Base of Stolen Memory
-	 * is unreliable, so compute the base by subtracting the stolen memory
-	 * from the Top of Low Usable DRAM which is where the BIOS places
-	 * the graphics stolen memory.
+	 * is unreliable, so on those compute the base by subtracting the
+	 * stolen memory from the Top of Low Usable DRAM which is where the
+	 * BIOS places the graphics stolen memory.
+	 *
+	 * On gen2, the layout is slightly different with the Graphics Segment
+	 * immediately following Top of Memory (or Top of Usable DRAM). Note
+	 * it appears that TOUD is only reported by 865g, so we just use the
+	 * top of memory as determined by the e820 probe.
+	 *
+	 * XXX gen2 requires an unavailable symbol and 945gm fails with
+	 * its value of TOLUD.
 	 */
+	base = 0;
 	if (INTEL_INFO(dev)->gen > 3 || IS_G33(dev)) {
-		/* top 32bits are reserved = 0 */
+		/* Read Graphics Base of Stolen Memory directly */
 		pci_read_config_dword(pdev, 0xA4, &base);
-	} else {
-		/* XXX presume 8xx is the same as i915 */
-		pci_bus_read_config_dword(pdev->bus, 2, 0x5C, &base);
-	}
-#else
-	if (INTEL_INFO(dev)->gen > 3 || IS_G33(dev)) {
-		u16 val;
-		pci_read_config_word(pdev, 0xb0, &val);
-		base = val >> 4 << 20;
-	} else {
+#if 0
+	} else if (IS_GEN3(dev)) {
 		u8 val;
+		/* Stolen is immediately below Top of Low Usable DRAM */
 		pci_read_config_byte(pdev, 0x9c, &val);
 		base = val >> 3 << 27;
-	}
-	base -= dev_priv->mm.gtt->stolen_size;
+		base -= dev_priv->mm.gtt->stolen_size;
+	} else {
+		/* Stolen is immediately above Top of Memory */
+		base = max_low_pfn_mapped << PAGE_SHIFT;
 #endif
+	}
 
-	return base + offset;
+	return base;
 }
 
 static void i915_warn_stolen(struct drm_device *dev)
@@ -117,7 +104,7 @@ static void i915_setup_compression(struct drm_device *dev, int size)
 	if (!compressed_fb)
 		goto err;
 
-	cfb_base = i915_stolen_to_phys(dev, compressed_fb->start);
+	cfb_base = dev_priv->mm.stolen_base + compressed_fb->start;
 	if (!cfb_base)
 		goto err_fb;
 
@@ -130,7 +117,7 @@ static void i915_setup_compression(struct drm_device *dev, int size)
 		if (!compressed_llb)
 			goto err_fb;
 
-		ll_base = i915_stolen_to_phys(dev, compressed_llb->start);
+		ll_base = dev_priv->mm.stolen_base + compressed_llb->start;
 		if (!ll_base)
 			goto err_llb;
 	}
@@ -149,7 +136,7 @@ static void i915_setup_compression(struct drm_device *dev, int size)
 	}
 
 	DRM_DEBUG_KMS("FBC base 0x%08lx, ll base 0x%08lx, size %dM\n",
-		      cfb_base, ll_base, size >> 20);
+		      (long)cfb_base, (long)ll_base, size >> 20);
 	return;
 
 err_llb:
@@ -181,6 +168,10 @@ int i915_gem_init_stolen(struct drm_device *dev)
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	unsigned long prealloc_size = dev_priv->mm.gtt->stolen_size;
 
+	dev_priv->mm.stolen_base = i915_stolen_to_physical(dev);
+	if (dev_priv->mm.stolen_base == 0)
+		return 0;
+
 	/* Basic memrange allocator for stolen space */
 	drm_mm_init(&dev_priv->mm.stolen, 0, prealloc_size);
 
-- 
1.7.10

[PATCH 12/28] drm/i915: Move GEM initialisation from i915_dma.c to i915_gem.c

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_dma.c |   66 +--------------------------------------
 drivers/gpu/drm/i915/i915_drv.h |    1 +
 drivers/gpu/drm/i915/i915_gem.c |   64 +++++++++++++++++++++++++++++++++++++
 3 files changed, 66 insertions(+), 65 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
index e3f1488..9e80d5a 100644
--- a/drivers/gpu/drm/i915/i915_dma.c
+++ b/drivers/gpu/drm/i915/i915_dma.c
@@ -1062,70 +1062,6 @@ static bool i915_switcheroo_can_switch(struct pci_dev *pdev)
 	return can_switch;
 }
 
-static bool
-intel_enable_ppgtt(struct drm_device *dev)
-{
-	if (i915_enable_ppgtt >= 0)
-		return i915_enable_ppgtt;
-
-#ifdef CONFIG_INTEL_IOMMU
-	/* Disable ppgtt on SNB if VT-d is on. */
-	if (INTEL_INFO(dev)->gen == 6 && intel_iommu_gfx_mapped)
-		return false;
-#endif
-
-	return true;
-}
-
-static int i915_load_gem_init(struct drm_device *dev)
-{
-	struct drm_i915_private *dev_priv = dev->dev_private;
-	unsigned long gtt_size, mappable_size;
-	int ret;
-
-	gtt_size = dev_priv->mm.gtt->gtt_total_entries << PAGE_SHIFT;
-	mappable_size = dev_priv->mm.gtt->gtt_mappable_entries << PAGE_SHIFT;
-
-	mutex_lock(&dev->struct_mutex);
-	if (intel_enable_ppgtt(dev) && HAS_ALIASING_PPGTT(dev)) {
-		/* PPGTT pdes are stolen from global gtt ptes, so shrink the
-		 * aperture accordingly when using aliasing ppgtt. */
-		gtt_size -= I915_PPGTT_PD_ENTRIES*PAGE_SIZE;
-
-		i915_gem_init_global_gtt(dev, 0, mappable_size, gtt_size);
-
-		ret = i915_gem_init_aliasing_ppgtt(dev);
-		if (ret) {
-			mutex_unlock(&dev->struct_mutex);
-			return ret;
-		}
-	} else {
-		/* Let GEM Manage all of the aperture.
-		 *
-		 * However, leave one page at the end still bound to the scratch
-		 * page.  There are a number of places where the hardware
-		 * apparently prefetches past the end of the object, and we've
-		 * seen multiple hangs with the GPU head pointer stuck in a
-		 * batchbuffer bound at the last page of the aperture.  One page
-		 * should be enough to keep any prefetching inside of the
-		 * aperture.
-		 */
-		i915_gem_init_global_gtt(dev, 0, mappable_size,
-					 gtt_size);
-	}
-
-	ret = i915_gem_init_hw(dev);
-	mutex_unlock(&dev->struct_mutex);
-	if (ret) {
-		i915_gem_cleanup_aliasing_ppgtt(dev);
-		return ret;
-	}
-
-	/* Allow hardware batchbuffers unless told otherwise. */
-	dev_priv->allow_batchbuffer = 1;
-	return 0;
-}
-
 static int i915_load_modeset_init(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
@@ -1168,7 +1104,7 @@ static int i915_load_modeset_init(struct drm_device *dev)
 
 	intel_modeset_init(dev);
 
-	ret = i915_load_gem_init(dev);
+	ret = i915_gem_init(dev);
 	if (ret)
 		goto cleanup_gem_stolen;
 
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 1ae807d..32e6c4d 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1303,6 +1303,7 @@ int __must_check i915_gem_object_set_domain(struct drm_i915_gem_object *obj,
 					    uint32_t read_domains,
 					    uint32_t write_domain);
 int __must_check i915_gem_object_finish_gpu(struct drm_i915_gem_object *obj);
+int __must_check i915_gem_init(struct drm_device *dev);
 int __must_check i915_gem_init_hw(struct drm_device *dev);
 void i915_gem_init_swizzling(struct drm_device *dev);
 void i915_gem_init_ppgtt(struct drm_device *dev);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index e7aa44b..f85514b 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3683,6 +3683,70 @@ cleanup_render_ring:
 	return ret;
 }
 
+static bool
+intel_enable_ppgtt(struct drm_device *dev)
+{
+	if (i915_enable_ppgtt >= 0)
+		return i915_enable_ppgtt;
+
+#ifdef CONFIG_INTEL_IOMMU
+	/* Disable ppgtt on SNB if VT-d is on. */
+	if (INTEL_INFO(dev)->gen == 6 && intel_iommu_gfx_mapped)
+		return false;
+#endif
+
+	return true;
+}
+
+int i915_gem_init(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	unsigned long gtt_size, mappable_size;
+	int ret;
+
+	gtt_size = dev_priv->mm.gtt->gtt_total_entries << PAGE_SHIFT;
+	mappable_size = dev_priv->mm.gtt->gtt_mappable_entries << PAGE_SHIFT;
+
+	mutex_lock(&dev->struct_mutex);
+	if (intel_enable_ppgtt(dev) && HAS_ALIASING_PPGTT(dev)) {
+		/* PPGTT pdes are stolen from global gtt ptes, so shrink the
+		 * aperture accordingly when using aliasing ppgtt. */
+		gtt_size -= I915_PPGTT_PD_ENTRIES*PAGE_SIZE;
+
+		i915_gem_init_global_gtt(dev, 0, mappable_size, gtt_size);
+
+		ret = i915_gem_init_aliasing_ppgtt(dev);
+		if (ret) {
+			mutex_unlock(&dev->struct_mutex);
+			return ret;
+		}
+	} else {
+		/* Let GEM Manage all of the aperture.
+		 *
+		 * However, leave one page at the end still bound to the scratch
+		 * page.  There are a number of places where the hardware
+		 * apparently prefetches past the end of the object, and we've
+		 * seen multiple hangs with the GPU head pointer stuck in a
+		 * batchbuffer bound at the last page of the aperture.  One page
+		 * should be enough to keep any prefetching inside of the
+		 * aperture.
+		 */
+		i915_gem_init_global_gtt(dev, 0, mappable_size,
+					 gtt_size);
+	}
+
+	ret = i915_gem_init_hw(dev);
+	mutex_unlock(&dev->struct_mutex);
+	if (ret) {
+		i915_gem_cleanup_aliasing_ppgtt(dev);
+		return ret;
+	}
+
+	/* Allow hardware batchbuffers unless told otherwise. */
+	dev_priv->allow_batchbuffer = 1;
+	return 0;
+}
+
 void
 i915_gem_cleanup_ringbuffer(struct drm_device *dev)
 {
-- 
1.7.10

[PATCH 13/28] drm/i915: Avoid clearing preallocated regions from the GTT

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h     |    2 ++
 drivers/gpu/drm/i915/i915_gem_gtt.c |   32 +++++++++++++++++++++++++++++---
 2 files changed, 31 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 32e6c4d..3492e20 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -827,6 +827,8 @@ enum i915_cache_level {
 	I915_CACHE_LLC_MLC, /* gen6+ */
 };
 
+#define I915_GTT_RESERVED ((struct drm_mm_node *)0x1)
+
 struct drm_i915_gem_object {
 	struct drm_gem_object base;
 
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 25c8bf9..b0c20a5 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -417,16 +417,42 @@ void i915_gem_init_global_gtt(struct drm_device *dev,
 			      unsigned long end)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
+	struct drm_mm_node *entry;
+	struct drm_i915_gem_object *obj;
 
-	/* Substract the guard page ... */
+	/* Subtract the guard page ... */
 	drm_mm_init(&dev_priv->mm.gtt_space, start, end - start - PAGE_SIZE);
 
+	/* Mark any preallocated objects as occupied */
+	list_for_each_entry(obj, &dev_priv->mm.gtt_list, gtt_list) {
+		DRM_DEBUG_KMS("reserving preallocated space: %x + %zx\n",
+			      obj->gtt_offset, obj->base.size);
+
+		BUG_ON(obj->gtt_space != I915_GTT_RESERVED);
+		obj->gtt_space = drm_mm_create_block(&dev_priv->mm.gtt_space,
+						     obj->gtt_offset,
+						     obj->base.size,
+						     false);
+		obj->has_global_gtt_mapping = 1;
+	}
+
 	dev_priv->mm.gtt_start = start;
 	dev_priv->mm.gtt_mappable_end = mappable_end;
 	dev_priv->mm.gtt_end = end;
 	dev_priv->mm.gtt_total = end - start;
 	dev_priv->mm.mappable_gtt_total = min(end, mappable_end) - start;
 
-	/* ... but ensure that we clear the entire range. */
-	intel_gtt_clear_range(start / PAGE_SIZE, (end-start) / PAGE_SIZE);
+	/* Clear any non-preallocated blocks */
+	list_for_each_entry(entry, &dev_priv->mm.gtt_space.hole_stack, hole_stack) {
+		unsigned long hole_start = entry->start + entry->size;
+		unsigned long hole_end = list_entry(entry->node_list.next,
+						    struct drm_mm_node,
+						    node_list)->start;
+
+		DRM_DEBUG_KMS("clearing unused GTT space: [%lx, %lx]\n",
+			      hole_start, hole_end);
+
+		intel_gtt_clear_range(hole_start / PAGE_SIZE,
+				      (hole_end-hole_start) / PAGE_SIZE);
+	}
 }
-- 
1.7.10

[PATCH 14/28] drm/i915: Always use the scatterlists if available for GTT insertion

[Chris Wilson]

If we have created a scatterlist for the physical mapping of the object,
simply use it. This facilitates the later insertion of stolen objects
into the GATT which are not backed by struct page.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c |   19 ++++++++-----------
 1 file changed, 8 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index b0c20a5..7afd631 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -249,8 +249,6 @@ void i915_ppgtt_bind_object(struct i915_hw_ppgtt *ppgtt,
 			    struct drm_i915_gem_object *obj,
 			    enum i915_cache_level cache_level)
 {
-	struct drm_device *dev = obj->base.dev;
-	struct drm_i915_private *dev_priv = dev->dev_private;
 	uint32_t pte_flags = GEN6_PTE_VALID;
 
 	switch (cache_level) {
@@ -267,20 +265,20 @@ void i915_ppgtt_bind_object(struct i915_hw_ppgtt *ppgtt,
 		BUG();
 	}
 
-	if (dev_priv->mm.gtt->needs_dmar) {
-		BUG_ON(!obj->sg_list);
-
+	if (obj->sg_list) {
 		i915_ppgtt_insert_sg_entries(ppgtt,
 					     obj->sg_list,
 					     obj->num_sg,
 					     obj->gtt_space->start >> PAGE_SHIFT,
 					     pte_flags);
-	} else
+	} else {
+		BUG_ON(!obj->pages);
 		i915_ppgtt_insert_pages(ppgtt,
 					obj->gtt_space->start >> PAGE_SHIFT,
 					obj->base.size >> PAGE_SHIFT,
 					obj->pages,
 					pte_flags);
+	}
 }
 
 void i915_ppgtt_unbind_object(struct i915_hw_ppgtt *ppgtt,
@@ -368,21 +366,20 @@ void i915_gem_gtt_bind_object(struct drm_i915_gem_object *obj,
 			      enum i915_cache_level cache_level)
 {
 	struct drm_device *dev = obj->base.dev;
-	struct drm_i915_private *dev_priv = dev->dev_private;
 	unsigned int agp_type = cache_level_to_agp_type(dev, cache_level);
 
-	if (dev_priv->mm.gtt->needs_dmar) {
-		BUG_ON(!obj->sg_list);
-
+	if (obj->sg_list) {
 		intel_gtt_insert_sg_entries(obj->sg_list,
 					    obj->num_sg,
 					    obj->gtt_space->start >> PAGE_SHIFT,
 					    agp_type);
-	} else
+	} else {
+		BUG_ON(!obj->pages);
 		intel_gtt_insert_pages(obj->gtt_space->start >> PAGE_SHIFT,
 				       obj->base.size >> PAGE_SHIFT,
 				       obj->pages,
 				       agp_type);
+	}
 
 	obj->has_global_gtt_mapping = 1;
 }
-- 
1.7.10

[PATCH 15/28] drm/i915: Tidy handling of the scatterlist by ppgtt

[Chris Wilson]

Check to see if we've reached the end before dereferencing to get the
next scatterlist. This helps when creating scatterlists by hand.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c |    8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 7afd631..fbdaf5a 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -194,13 +194,11 @@ static void i915_ppgtt_insert_sg_entries(struct i915_hw_ppgtt *ppgtt,
 			pt_vaddr[j] = pte | pte_flags;
 
 			/* grab the next page */
-			m++;
-			if (m == segment_len) {
-				sg = sg_next(sg);
-				i++;
-				if (i == sg_len)
+			if (++m == segment_len) {
+				if (++i == sg_len)
 					break;
 
+				sg = sg_next(sg);
 				segment_len = sg_dma_len(sg) >> PAGE_SHIFT;
 				m = 0;
 			}
-- 
1.7.10

[PATCH 16/28] drm/i915: Delay allocation of stolen space for FBC

[Chris Wilson]

As we may wish to wrap regions preallocated by the BIOS, we need to do
that before carving out contiguous chunks of stolen space for FBC.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h        |    1 +
 drivers/gpu/drm/i915/i915_gem_stolen.c |   68 +++++++++++++++++++++-----------
 drivers/gpu/drm/i915/intel_display.c   |    3 ++
 drivers/gpu/drm/i915/intel_pm.c        |   13 +++---
 4 files changed, 55 insertions(+), 30 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 3492e20..69efb8e 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1373,6 +1373,7 @@ int i915_gem_evict_everything(struct drm_device *dev);
 
 /* i915_gem_stolen.c */
 int i915_gem_init_stolen(struct drm_device *dev);
+int i915_gem_stolen_setup_compression(struct drm_device *dev);
 void i915_gem_cleanup_stolen(struct drm_device *dev);
 
 /* i915_gem_tiling.c */
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index a01ff74..5e10a23 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -88,15 +88,15 @@ static void i915_warn_stolen(struct drm_device *dev)
 	DRM_INFO("hint: you may be able to increase stolen memory size in the BIOS to avoid this\n");
 }
 
-static void i915_setup_compression(struct drm_device *dev, int size)
+static int i915_setup_compression(struct drm_device *dev, int size)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct drm_mm_node *compressed_fb, *uninitialized_var(compressed_llb);
 	unsigned long cfb_base;
 	unsigned long ll_base = 0;
 
-	/* Just in case the BIOS is doing something questionable. */
-	intel_disable_fbc(dev);
+	DRM_DEBUG_KMS("reserving %d bytes of contiguous stolen space for FBC\n",
+		      size);
 
 	compressed_fb = drm_mm_search_free(&dev_priv->mm.stolen, size, 4096, 0);
 	if (compressed_fb)
@@ -123,8 +123,8 @@ static void i915_setup_compression(struct drm_device *dev, int size)
 	}
 
 	dev_priv->cfb_size = size;
-
 	dev_priv->compressed_fb = compressed_fb;
+
 	if (HAS_PCH_SPLIT(dev))
 		I915_WRITE(ILK_DPFC_CB_BASE, compressed_fb->start);
 	else if (IS_GM45(dev)) {
@@ -137,7 +137,7 @@ static void i915_setup_compression(struct drm_device *dev, int size)
 
 	DRM_DEBUG_KMS("FBC base 0x%08lx, ll base 0x%08lx, size %dM\n",
 		      (long)cfb_base, (long)ll_base, size >> 20);
-	return;
+	return size;
 
 err_llb:
 	drm_mm_put_block(compressed_llb);
@@ -146,48 +146,68 @@ err_fb:
 err:
 	dev_priv->no_fbc_reason = FBC_STOLEN_TOO_SMALL;
 	i915_warn_stolen(dev);
+	return 0;
+}
+
+int i915_gem_stolen_setup_compression(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct drm_mm_node *entry;
+	unsigned long size;
+
+	if (dev_priv->mm.stolen_base == 0)
+		return 0;
+
+	if (dev_priv->cfb_size)
+		return dev_priv->cfb_size;
+
+	/* Try to set up FBC with a reasonable compressed buffer size */
+	size = 0;
+	list_for_each_entry(entry, &dev_priv->mm.gtt_space.hole_stack, hole_stack) {
+		unsigned long hole_start = entry->start + entry->size;
+		unsigned long hole_end = list_entry(entry->node_list.next,
+						    struct drm_mm_node,
+						    node_list)->start;
+		unsigned long hole_size = hole_end - hole_start;
+		if (hole_size > size)
+			size = hole_size;
+	}
+
+	/* Try to get a 32M buffer... */
+	if (size > (36*1024*1024))
+		size = 32*1024*1024;
+	else /* fall back to 7/8 of the stolen space */
+		size = size * 7 / 8;
+
+	return i915_setup_compression(dev, size);
 }
 
 static void i915_cleanup_compression(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
 
-	drm_mm_put_block(dev_priv->compressed_fb);
+	if (dev_priv->compressed_fb)
+		drm_mm_put_block(dev_priv->compressed_fb);
+
 	if (dev_priv->compressed_llb)
 		drm_mm_put_block(dev_priv->compressed_llb);
 }
 
 void i915_gem_cleanup_stolen(struct drm_device *dev)
 {
-	if (I915_HAS_FBC(dev) && i915_powersave)
-		i915_cleanup_compression(dev);
+	i915_cleanup_compression(dev);
 }
 
 int i915_gem_init_stolen(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
-	unsigned long prealloc_size = dev_priv->mm.gtt->stolen_size;
 
 	dev_priv->mm.stolen_base = i915_stolen_to_physical(dev);
 	if (dev_priv->mm.stolen_base == 0)
 		return 0;
 
 	/* Basic memrange allocator for stolen space */
-	drm_mm_init(&dev_priv->mm.stolen, 0, prealloc_size);
-
-	/* Try to set up FBC with a reasonable compressed buffer size */
-	if (I915_HAS_FBC(dev) && i915_powersave) {
-		int cfb_size;
-
-		/* Leave 1M for line length buffer & misc. */
-
-		/* Try to get a 32M buffer... */
-		if (prealloc_size > (36*1024*1024))
-			cfb_size = 32*1024*1024;
-		else /* fall back to 7/8 of the stolen space */
-			cfb_size = prealloc_size * 7 / 8;
-		i915_setup_compression(dev, cfb_size);
-	}
+	drm_mm_init(&dev_priv->mm.stolen, 0, dev_priv->mm.gtt->stolen_size);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index 2e41eaf..69434b6 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -6739,6 +6739,9 @@ void intel_modeset_init(struct drm_device *dev)
 	INIT_WORK(&dev_priv->idle_work, intel_idle_update);
 	setup_timer(&dev_priv->idle_timer, intel_gpu_idle_timer,
 		    (unsigned long)dev);
+
+	/* Just in case the BIOS is doing something questionable. */
+	intel_disable_fbc(dev);
 }
 
 void intel_modeset_gem_init(struct drm_device *dev)
diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index cbb7217..98756d2 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -436,12 +436,6 @@ void intel_update_fbc(struct drm_device *dev)
 		dev_priv->no_fbc_reason = FBC_MODULE_PARAM;
 		goto out_disable;
 	}
-	if (intel_fb->obj->base.size > dev_priv->cfb_size) {
-		DRM_DEBUG_KMS("framebuffer too large, disabling "
-			      "compression\n");
-		dev_priv->no_fbc_reason = FBC_STOLEN_TOO_SMALL;
-		goto out_disable;
-	}
 	if ((crtc->mode.flags & DRM_MODE_FLAG_INTERLACE) ||
 	    (crtc->mode.flags & DRM_MODE_FLAG_DBLSCAN)) {
 		DRM_DEBUG_KMS("mode incompatible with compression, "
@@ -475,6 +469,13 @@ void intel_update_fbc(struct drm_device *dev)
 	if (in_dbg_master())
 		goto out_disable;
 
+	if (intel_fb->obj->base.size > i915_gem_stolen_setup_compression(dev)) {
+		DRM_DEBUG_KMS("framebuffer too large, disabling "
+			      "compression\n");
+		dev_priv->no_fbc_reason = FBC_STOLEN_TOO_SMALL;
+		goto out_disable;
+	}
+
 	/* If the scanout has not changed, don't modify the FBC settings.
 	 * Note that we make the fundamental assumption that the fb->obj
 	 * cannot be unpinned (and have its GTT offset and fence revoked)
-- 
1.7.10

[PATCH 17/28] drm/i915: Extract general object init routine

[Chris Wilson]

As we wish to create specialised object constructions in the near
future that share the same basic GEM object struct, export the default
initializer.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h |    1 +
 drivers/gpu/drm/i915/i915_gem.c |   32 +++++++++++++++++++-------------
 2 files changed, 20 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 69efb8e..f1dcca2 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1235,6 +1235,7 @@ int i915_gem_init_object(struct drm_gem_object *obj);
 int __must_check i915_gem_flush_ring(struct intel_ring_buffer *ring,
 				     uint32_t invalidate_domains,
 				     uint32_t flush_domains);
+void i915_gem_object_init(struct drm_i915_gem_object *obj);
 struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
 						  size_t size);
 void i915_gem_free_object(struct drm_gem_object *obj);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index f85514b..6d7e312 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3423,10 +3423,27 @@ unlock:
 	return ret;
 }
 
+void i915_gem_object_init(struct drm_i915_gem_object *obj)
+{
+	obj->base.driver_private = NULL;
+	obj->fence_reg = I915_FENCE_REG_NONE;
+
+	INIT_LIST_HEAD(&obj->mm_list);
+	INIT_LIST_HEAD(&obj->gtt_list);
+	INIT_LIST_HEAD(&obj->ring_list);
+	INIT_LIST_HEAD(&obj->exec_list);
+	INIT_LIST_HEAD(&obj->gpu_write_list);
+
+	obj->madv = I915_MADV_WILLNEED;
+	/* Avoid an unnecessary call to unbind on the first bind. */
+	obj->map_and_fenceable = true;
+
+	i915_gem_info_add_obj(obj->base.dev->dev_private, obj->base.size);
+}
+
 struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
 						  size_t size)
 {
-	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct drm_i915_gem_object *obj;
 	struct address_space *mapping;
 
@@ -3442,7 +3459,7 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
 	mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping;
 	mapping_set_gfp_mask(mapping, GFP_HIGHUSER | __GFP_RECLAIMABLE);
 
-	i915_gem_info_add_obj(dev_priv, size);
+	i915_gem_object_init(obj);
 
 	obj->base.write_domain = I915_GEM_DOMAIN_CPU;
 	obj->base.read_domains = I915_GEM_DOMAIN_CPU;
@@ -3464,17 +3481,6 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
 	} else
 		obj->cache_level = I915_CACHE_NONE;
 
-	obj->base.driver_private = NULL;
-	obj->fence_reg = I915_FENCE_REG_NONE;
-	INIT_LIST_HEAD(&obj->mm_list);
-	INIT_LIST_HEAD(&obj->gtt_list);
-	INIT_LIST_HEAD(&obj->ring_list);
-	INIT_LIST_HEAD(&obj->exec_list);
-	INIT_LIST_HEAD(&obj->gpu_write_list);
-	obj->madv = I915_MADV_WILLNEED;
-	/* Avoid an unnecessary call to unbind on the first bind. */
-	obj->map_and_fenceable = true;
-
 	return obj;
 }
 
-- 
1.7.10

[PATCH 18/28] drm/i915: Allow objects to be created with no backing pages, but stolen space

[Chris Wilson]

In order to accommodate objects that are not backed by struct pages, but
instead point into a contiguous region of stolen space, we need to make
various changes to avoid dereferencing obj->pages or obj->base.filp.

First introduce a marker for the stolen object, that specifies its
offset into the stolen region and implies that it has no backing pages.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c |    2 ++
 drivers/gpu/drm/i915/i915_drv.h     |    2 ++
 drivers/gpu/drm/i915/i915_gem.c     |    2 +-
 3 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 1e12c72..3a1cbe2 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -140,6 +140,8 @@ describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)
 	if (obj->gtt_space != NULL)
 		seq_printf(m, " (gtt offset: %08x, size: %08x)",
 			   obj->gtt_offset, (unsigned int)obj->gtt_space->size);
+	if (obj->stolen)
+		seq_printf(m, " (stolen: %08lx)", obj->stolen->start);
 	if (obj->pin_mappable || obj->fault_mappable) {
 		char s[3], *t = s;
 		if (obj->pin_mappable)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index f1dcca2..8850b19 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -834,6 +834,8 @@ struct drm_i915_gem_object {
 
 	/** Current space allocated to this object in the GTT, if any. */
 	struct drm_mm_node *gtt_space;
+	/** Stolen memory for this object, instead of being backed by shmem. */
+	struct drm_mm_node *stolen;
 	struct list_head gtt_list;
 
 	/** This object's place on the active/flushing/inactive lists */
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 6d7e312..a0b6375 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1523,7 +1523,7 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
 	struct page *page;
 	gfp_t gfp;
 
-	if (obj->pages)
+	if (obj->pages || obj->stolen)
 		return 0;
 
 	if (obj->madv != I915_MADV_WILLNEED)
-- 
1.7.10

[PATCH 19/28] drm/i915: Support readback of stolen objects upon error

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_irq.c |   13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 24e1dd2..033f9be 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -827,7 +827,10 @@ i915_error_object_create(struct drm_i915_private *dev_priv,
 	int page, page_count;
 	u32 reloc_offset;
 
-	if (src == NULL || src->pages == NULL)
+	if (src == NULL)
+		return NULL;
+
+	if (src->pages == NULL && !src->stolen)
 		return NULL;
 
 	page_count = src->base.size / PAGE_SIZE;
@@ -859,6 +862,14 @@ i915_error_object_create(struct drm_i915_private *dev_priv,
 						     reloc_offset);
 			memcpy_fromio(d, s, PAGE_SIZE);
 			io_mapping_unmap_atomic(s);
+		} else if (src->stolen) {
+			unsigned long offset;
+
+			offset = dev_priv->mm.stolen_base;
+			offset += src->stolen->start;
+			offset += page << PAGE_SHIFT;
+
+			memcpy_fromio(d, (void *)offset, PAGE_SIZE);
 		} else {
 			void *s;
 
-- 
1.7.10

[PATCH 20/28] drm/i915: Handle stolen objects in pwrite

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c |   46 ++++++++++++++++++++++++++-------------
 1 file changed, 31 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index a0b6375..45f215f 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -642,19 +642,17 @@ out:
  * needs_clflush_before is set and flushes out any written cachelines after
  * writing if needs_clflush is set. */
 static int
-shmem_pwrite_fast(struct page *page, int shmem_page_offset, int page_length,
+shmem_pwrite_fast(char *vaddr, int shmem_page_offset, int page_length,
 		  char __user *user_data,
 		  bool page_do_bit17_swizzling,
 		  bool needs_clflush_before,
 		  bool needs_clflush_after)
 {
-	char *vaddr;
 	int ret;
 
 	if (unlikely(page_do_bit17_swizzling))
 		return -EINVAL;
 
-	vaddr = kmap_atomic(page);
 	if (needs_clflush_before)
 		drm_clflush_virt_range(vaddr + shmem_page_offset,
 				       page_length);
@@ -664,7 +662,6 @@ shmem_pwrite_fast(struct page *page, int shmem_page_offset, int page_length,
 	if (needs_clflush_after)
 		drm_clflush_virt_range(vaddr + shmem_page_offset,
 				       page_length);
-	kunmap_atomic(vaddr);
 
 	return ret ? -EFAULT : 0;
 }
@@ -672,16 +669,14 @@ shmem_pwrite_fast(struct page *page, int shmem_page_offset, int page_length,
 /* Only difference to the fast-path function is that this can handle bit17
  * and uses non-atomic copy and kmap functions. */
 static int
-shmem_pwrite_slow(struct page *page, int shmem_page_offset, int page_length,
+shmem_pwrite_slow(char *vaddr, int shmem_page_offset, int page_length,
 		  char __user *user_data,
 		  bool page_do_bit17_swizzling,
 		  bool needs_clflush_before,
 		  bool needs_clflush_after)
 {
-	char *vaddr;
 	int ret;
 
-	vaddr = kmap(page);
 	if (unlikely(needs_clflush_before || page_do_bit17_swizzling))
 		shmem_clflush_swizzled_range(vaddr + shmem_page_offset,
 					     page_length,
@@ -698,7 +693,6 @@ shmem_pwrite_slow(struct page *page, int shmem_page_offset, int page_length,
 		shmem_clflush_swizzled_range(vaddr + shmem_page_offset,
 					     page_length,
 					     page_do_bit17_swizzling);
-	kunmap(page);
 
 	return ret ? -EFAULT : 0;
 }
@@ -750,6 +744,7 @@ i915_gem_shmem_pwrite(struct drm_device *dev,
 	obj->dirty = 1;
 
 	while (remain > 0) {
+		char *uninitialized_var(vaddr);
 		struct page *page;
 		int partial_cacheline_write;
 
@@ -771,25 +766,44 @@ i915_gem_shmem_pwrite(struct drm_device *dev,
 			((shmem_page_offset | page_length)
 				& (boot_cpu_data.x86_clflush_size - 1));
 
-		page = obj->pages[offset >> PAGE_SHIFT];
-		page_cache_get(page);
+		if (obj->stolen) {
+			page = NULL;
+			vaddr = (char *)(dev_priv->mm.stolen_base + obj->stolen->start + offset);
+		} else {
+			page = obj->pages[offset >> PAGE_SHIFT];
+			page_cache_get(page);
+		}
 
 		page_do_bit17_swizzling = obj_do_bit17_swizzling &&
 			(page_to_phys(page) & (1 << 17)) != 0;
 
-		ret = shmem_pwrite_fast(page, shmem_page_offset, page_length,
+		if (page)
+			vaddr = kmap_atomic(page);
+
+		ret = shmem_pwrite_fast(vaddr, shmem_page_offset, page_length,
 					user_data, page_do_bit17_swizzling,
 					partial_cacheline_write,
 					needs_clflush_after);
+
+		if (page)
+			kunmap_atomic(vaddr);
+
 		if (ret == 0)
 			goto next_page;
 
 		mutex_unlock(&dev->struct_mutex);
-		ret = shmem_pwrite_slow(page, shmem_page_offset, page_length,
+
+		if (page)
+			vaddr = kmap(page);
+
+		ret = shmem_pwrite_slow(vaddr, shmem_page_offset, page_length,
 					user_data, page_do_bit17_swizzling,
 					partial_cacheline_write,
 					needs_clflush_after);
 
+		if (page)
+			kunmap(page);
+
 		mutex_lock(&dev->struct_mutex);
 		hit_slowpath = 1;
 
@@ -797,9 +811,11 @@ i915_gem_shmem_pwrite(struct drm_device *dev,
 			ret = i915_gem_object_get_pages_gtt(obj);
 
 next_page:
-		set_page_dirty(page);
-		mark_page_accessed(page);
-		page_cache_release(page);
+		if (page) {
+			set_page_dirty(page);
+			mark_page_accessed(page);
+			page_cache_release(page);
+		}
 
 		if (ret)
 			goto out;
-- 
1.7.10

[PATCH 21/28] drm/i915: Handle stolen objects for pread

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c |   44 ++++++++++++++++++++++++++-------------
 1 file changed, 30 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 45f215f..e72f661 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -316,24 +316,21 @@ __copy_from_user_swizzled(char *gpu_vaddr, int gpu_offset,
  * Flushes invalid cachelines before reading the target if
  * needs_clflush is set. */
 static int
-shmem_pread_fast(struct page *page, int shmem_page_offset, int page_length,
+shmem_pread_fast(char *vaddr, int shmem_page_offset, int page_length,
 		 char __user *user_data,
 		 bool page_do_bit17_swizzling, bool needs_clflush)
 {
-	char *vaddr;
 	int ret;
 
 	if (unlikely(page_do_bit17_swizzling))
 		return -EINVAL;
 
-	vaddr = kmap_atomic(page);
 	if (needs_clflush)
 		drm_clflush_virt_range(vaddr + shmem_page_offset,
 				       page_length);
 	ret = __copy_to_user_inatomic(user_data,
 				      vaddr + shmem_page_offset,
 				      page_length);
-	kunmap_atomic(vaddr);
 
 	return ret ? -EFAULT : 0;
 }
@@ -363,14 +360,12 @@ shmem_clflush_swizzled_range(char *addr, unsigned long length,
 /* Only difference to the fast-path function is that this can handle bit17
  * and uses non-atomic copy and kmap functions. */
 static int
-shmem_pread_slow(struct page *page, int shmem_page_offset, int page_length,
+shmem_pread_slow(char *vaddr, int shmem_page_offset, int page_length,
 		 char __user *user_data,
 		 bool page_do_bit17_swizzling, bool needs_clflush)
 {
-	char *vaddr;
 	int ret;
 
-	vaddr = kmap(page);
 	if (needs_clflush)
 		shmem_clflush_swizzled_range(vaddr + shmem_page_offset,
 					     page_length,
@@ -384,7 +379,6 @@ shmem_pread_slow(struct page *page, int shmem_page_offset, int page_length,
 		ret = __copy_to_user(user_data,
 				     vaddr + shmem_page_offset,
 				     page_length);
-	kunmap(page);
 
 	return ret ? - EFAULT : 0;
 }
@@ -395,6 +389,7 @@ i915_gem_shmem_pread(struct drm_device *dev,
 		     struct drm_i915_gem_pread *args,
 		     struct drm_file *file)
 {
+	struct drm_i915_private *dev_priv = dev->dev_private;
 	char __user *user_data;
 	ssize_t remain;
 	loff_t offset;
@@ -430,6 +425,7 @@ i915_gem_shmem_pread(struct drm_device *dev,
 	offset = args->offset;
 
 	while (remain > 0) {
+		char *uninitialized_var(vaddr);
 		struct page *page;
 
 		/* Operation in this page
@@ -442,15 +438,27 @@ i915_gem_shmem_pread(struct drm_device *dev,
 		if ((shmem_page_offset + page_length) > PAGE_SIZE)
 			page_length = PAGE_SIZE - shmem_page_offset;
 
-		page = obj->pages[offset >> PAGE_SHIFT];
-		page_cache_get(page);
+		if (obj->stolen) {
+			page = NULL;
+			vaddr = (char *)(dev_priv->mm.stolen_base + obj->stolen->start + offset);
+		} else {
+			page = obj->pages[offset >> PAGE_SHIFT];
+			page_cache_get(page);
+		}
 
 		page_do_bit17_swizzling = obj_do_bit17_swizzling &&
 			(page_to_phys(page) & (1 << 17)) != 0;
 
-		ret = shmem_pread_fast(page, shmem_page_offset, page_length,
+		if (page)
+			vaddr = kmap_atomic(page);
+
+		ret = shmem_pread_fast(vaddr, shmem_page_offset, page_length,
 				       user_data, page_do_bit17_swizzling,
 				       needs_clflush);
+
+		if (page)
+			kunmap_atomic(vaddr);
+
 		if (ret == 0)
 			goto next_page;
 
@@ -466,10 +474,16 @@ i915_gem_shmem_pread(struct drm_device *dev,
 			prefaulted = 1;
 		}
 
-		ret = shmem_pread_slow(page, shmem_page_offset, page_length,
+		if (page)
+			vaddr = kmap(page);
+
+		ret = shmem_pread_slow(vaddr, shmem_page_offset, page_length,
 				       user_data, page_do_bit17_swizzling,
 				       needs_clflush);
 
+		if (page)
+			kunmap(page);
+
 		hit_slowpath = 1;
 		mutex_lock(&dev->struct_mutex);
 
@@ -477,8 +491,10 @@ i915_gem_shmem_pread(struct drm_device *dev,
 			ret = i915_gem_object_get_pages_gtt(obj);
 
 next_page:
-		mark_page_accessed(page);
-		page_cache_release(page);
+		if (page) {
+			mark_page_accessed(page);
+			page_cache_release(page);
+		}
 
 		if (ret)
 			goto out;
-- 
1.7.10

[PATCH 22/28] drm/i915: Introduce i915_gem_object_create_stolen()

[Chris Wilson]

Allow for the creation of GEM objects backed by stolen memory. As these
are not backed by ordinary pages, we create a fake dma mapping and store
the address in the scatterlist rather than obj->pages.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h        |    3 +
 drivers/gpu/drm/i915/i915_gem.c        |    2 +
 drivers/gpu/drm/i915/i915_gem_gtt.c    |    2 +-
 drivers/gpu/drm/i915/i915_gem_stolen.c |  110 ++++++++++++++++++++++++++++++++
 4 files changed, 116 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 8850b19..d487301 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1378,6 +1378,9 @@ int i915_gem_evict_everything(struct drm_device *dev);
 int i915_gem_init_stolen(struct drm_device *dev);
 int i915_gem_stolen_setup_compression(struct drm_device *dev);
 void i915_gem_cleanup_stolen(struct drm_device *dev);
+struct drm_i915_gem_object *
+i915_gem_object_create_stolen(struct drm_device *dev, u32 size);
+void i915_gem_object_release_stolen(struct drm_i915_gem_object *obj);
 
 /* i915_gem_tiling.c */
 void i915_gem_detect_bit_6_swizzle(struct drm_device *dev);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index e72f661..2a642f5 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3549,6 +3549,8 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)
 	i915_gem_object_put_pages_gtt(obj);
 	i915_gem_object_free_mmap_offset(obj);
 
+	i915_gem_object_release_stolen(obj);
+
 	drm_gem_object_release(&obj->base);
 	i915_gem_info_remove_obj(dev_priv, obj->base.size);
 
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index fbdaf5a..e847caec 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -398,7 +398,7 @@ void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj)
 
 	interruptible = do_idling(dev_priv);
 
-	if (obj->sg_list) {
+	if (obj->sg_list && !obj->stolen) {
 		intel_gtt_unmap_memory(obj->sg_list, obj->num_sg);
 		obj->sg_list = NULL;
 	}
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index 5e10a23..0011006 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -211,3 +211,113 @@ int i915_gem_init_stolen(struct drm_device *dev)
 
 	return 0;
 }
+
+struct scatterlist *
+i915_sg_create_for_stolen(struct drm_device *dev, u32 offset, u32 size)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct scatterlist *sg;
+
+	/* We hide that we have no struct page backing our stolen object
+	 * by wrapping the contiguous physical allocation with a fake
+	 * dma mapping in a single scatterlist.
+	 */
+
+	sg = kmalloc(sizeof(*sg), GFP_KERNEL);
+	if (sg == NULL)
+		return NULL;
+
+	sg_init_table(sg, 1);
+
+	sg->offset = offset;
+	sg->length = size;
+
+	sg_dma_address(sg) = dev_priv->mm.stolen_base + offset;
+	sg_dma_len(sg) = size;
+
+	return sg;
+}
+
+struct drm_i915_gem_object *
+_i915_gem_object_create_stolen(struct drm_device *dev,
+			       struct drm_mm_node *stolen)
+{
+	struct drm_i915_gem_object *obj;
+	struct scatterlist *sg;
+
+	obj = kzalloc(sizeof(*obj), GFP_KERNEL);
+	if (obj == NULL)
+		return NULL;
+
+	if (drm_gem_private_object_init(dev, &obj->base, stolen->size))
+		goto cleanup;
+
+	sg = i915_sg_create_for_stolen(dev, stolen->start, stolen->size);
+	if (sg == NULL)
+		goto cleanup;
+
+	i915_gem_object_init(obj);
+
+	obj->stolen = stolen;
+
+	obj->base.write_domain = I915_GEM_DOMAIN_GTT;
+	obj->base.read_domains = I915_GEM_DOMAIN_GTT;
+	obj->cache_level = I915_CACHE_NONE;
+
+	obj->sg_list = sg;
+	obj->num_sg = 1;
+
+	return obj;
+
+cleanup:
+	kfree(obj);
+	return NULL;
+}
+
+struct drm_i915_gem_object *
+i915_gem_object_create_stolen(struct drm_device *dev, u32 size)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct drm_i915_gem_object *obj;
+	struct drm_mm_node *stolen;
+
+	if (dev_priv->mm.stolen_base == 0)
+		return 0;
+
+	DRM_DEBUG_KMS("creating stolen object: size=%x\n", size);
+	if (size == 0)
+		return NULL;
+
+	stolen = drm_mm_search_free(&dev_priv->mm.stolen, size, 4096, 0);
+	if (stolen)
+		stolen = drm_mm_get_block(stolen, size, 4096);
+	if (stolen == NULL)
+		return NULL;
+
+	obj = _i915_gem_object_create_stolen(dev, stolen);
+	if (obj)
+		return obj;
+
+	drm_mm_put_block(stolen);
+	return NULL;
+}
+
+void
+i915_gem_object_release_stolen(struct drm_i915_gem_object *obj)
+{
+	if (obj->stolen) {
+		drm_mm_put_block(obj->stolen);
+		obj->stolen = NULL;
+	}
+
+	if (obj->sg_list) {
+		struct sg_table st;
+
+		st.sgl = obj->sg_list;
+		st.orig_nents = st.nents = obj->num_sg;
+
+		sg_free_table(&st);
+
+		obj->sg_list = NULL;
+	}
+}
-- 
1.7.10

[PATCH 23/28] drm/i915: Allocate fbcon from stolen memory

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_fb.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/intel_fb.c b/drivers/gpu/drm/i915/intel_fb.c
index 71ef289..eb65e3c 100644
--- a/drivers/gpu/drm/i915/intel_fb.c
+++ b/drivers/gpu/drm/i915/intel_fb.c
@@ -84,7 +84,9 @@ static int intelfb_create(struct intel_fbdev *ifbdev,
 
 	size = mode_cmd.pitches[0] * mode_cmd.height;
 	size = ALIGN(size, PAGE_SIZE);
-	obj = i915_gem_alloc_object(dev, size);
+	obj = i915_gem_object_create_stolen(dev, size);
+	if (obj == NULL)
+		obj = i915_gem_alloc_object(dev, size);
 	if (!obj) {
 		DRM_ERROR("failed to allocate framebuffer\n");
 		ret = -ENOMEM;
-- 
1.7.10

[PATCH 24/28] drm/i915: Allocate ringbuffers from stolen memory

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_ringbuffer.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 2a3fcd6..a8b7605 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -962,7 +962,11 @@ static int intel_init_ring_buffer(struct drm_device *dev,
 			return ret;
 	}
 
-	obj = i915_gem_alloc_object(dev, ring->size);
+	obj = NULL;
+	if (!HAS_LLC(dev))
+		obj = i915_gem_object_create_stolen(dev, ring->size);
+	if (obj == NULL)
+		obj = i915_gem_alloc_object(dev, ring->size);
 	if (obj == NULL) {
 		DRM_ERROR("Failed to allocate ringbuffer\n");
 		ret = -ENOMEM;
-- 
1.7.10

[PATCH 25/28] drm/i915: Allocate overlay registers from stolen memory

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_overlay.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_overlay.c b/drivers/gpu/drm/i915/intel_overlay.c
index 9845cfc..22bb856 100644
--- a/drivers/gpu/drm/i915/intel_overlay.c
+++ b/drivers/gpu/drm/i915/intel_overlay.c
@@ -1426,8 +1426,10 @@ void intel_setup_overlay(struct drm_device *dev)
 
 	overlay->dev = dev;
 
-	reg_bo = i915_gem_alloc_object(dev, PAGE_SIZE);
-	if (!reg_bo)
+	reg_bo = i915_gem_object_create_stolen(dev, PAGE_SIZE);
+	if (reg_bo == NULL)
+		reg_bo = i915_gem_alloc_object(dev, PAGE_SIZE);
+	if (reg_bo == NULL)
 		goto out_free;
 	overlay->reg_bo = reg_bo;
 
-- 
1.7.10

[PATCH 26/28] drm/i915: Introduce i915_gem_object_create_stolen_for_preallocated

[Chris Wilson]

Wrap a preallocated region of stolen memory within an ordinary GEM
object, for example the BIOS framebuffer.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h        |    5 +++
 drivers/gpu/drm/i915/i915_gem_stolen.c |   58 ++++++++++++++++++++++++++++++++
 2 files changed, 63 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index d487301..4190210 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1380,6 +1380,11 @@ int i915_gem_stolen_setup_compression(struct drm_device *dev);
 void i915_gem_cleanup_stolen(struct drm_device *dev);
 struct drm_i915_gem_object *
 i915_gem_object_create_stolen(struct drm_device *dev, u32 size);
+struct drm_i915_gem_object *
+i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev,
+					       u32 stolen_offset,
+					       u32 gtt_offset,
+					       u32 size);
 void i915_gem_object_release_stolen(struct drm_i915_gem_object *obj);
 
 /* i915_gem_tiling.c */
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index 0011006..0cd2671 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -302,6 +302,64 @@ i915_gem_object_create_stolen(struct drm_device *dev, u32 size)
 	return NULL;
 }
 
+struct drm_i915_gem_object *
+i915_gem_object_create_stolen_for_preallocated(struct drm_device *dev,
+					       u32 stolen_offset,
+					       u32 gtt_offset,
+					       u32 size)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct drm_i915_gem_object *obj;
+	struct drm_mm_node *stolen;
+
+	if (dev_priv->mm.stolen_base == 0)
+		return 0;
+
+	DRM_DEBUG_KMS("creating preallocated stolen object: stolen_offset=%x, gtt_offset=%x, size=%x\n",
+			stolen_offset, gtt_offset, size);
+	if (size == 0)
+		return NULL;
+
+	stolen = drm_mm_create_block(&dev_priv->mm.stolen,
+				     stolen_offset, size,
+				     false);
+	if (stolen == NULL)
+		return NULL;
+
+	obj = _i915_gem_object_create_stolen(dev, stolen);
+	if (obj == NULL)
+		goto cleanup;
+
+	/* To simplify the initialisation sequence between KMS and GTT,
+	 * we allow construction of the stolen object prior to
+	 * setting up the GTT space. The actual reservation will occur
+	 * later.
+	 */
+	if (drm_mm_initialized(&dev_priv->mm.gtt_space)) {
+		obj->gtt_space = drm_mm_create_block(&dev_priv->mm.gtt_space,
+						     gtt_offset, size,
+						     false);
+		if (obj->gtt_space == NULL)
+			goto cleanup_obj;
+	} else
+		obj->gtt_space = I915_GTT_RESERVED;
+
+	obj->gtt_offset = gtt_offset;
+	obj->has_global_gtt_mapping = 1;
+
+	list_add_tail(&obj->gtt_list, &dev_priv->mm.gtt_list);
+	list_add_tail(&obj->mm_list, &dev_priv->mm.inactive_list);
+
+	return obj;
+
+cleanup_obj:
+	kfree(obj->sg_list);
+	kfree(obj);
+cleanup:
+	drm_mm_put_block(stolen);
+	return NULL;
+}
+
 void
 i915_gem_object_release_stolen(struct drm_i915_gem_object *obj)
 {
-- 
1.7.10

[PATCH 27/28] drm/i915: Split the framebuffer_info creation into a separate routine

[Chris Wilson]

This will be shared with wrapping the BIOS framebuffer into the fbdev
later. In the meantime, we can tidy the code slightly and improve the
error path handling.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_fb.c |  125 +++++++++++++++++++++------------------
 1 file changed, 69 insertions(+), 56 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_fb.c b/drivers/gpu/drm/i915/intel_fb.c
index eb65e3c..c4d6cff 100644
--- a/drivers/gpu/drm/i915/intel_fb.c
+++ b/drivers/gpu/drm/i915/intel_fb.c
@@ -58,16 +58,74 @@ static struct fb_ops intelfb_ops = {
 	.fb_debug_leave = drm_fb_helper_debug_leave,
 };
 
+static struct fb_info *intelfb_create_info(struct intel_fbdev *ifbdev)
+{
+	struct drm_framebuffer *fb = &ifbdev->ifb.base;
+	struct drm_device *dev = fb->dev;
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct fb_info *info;
+	u32 gtt_offset, size;
+	int ret;
+
+	info = framebuffer_alloc(0, &dev->pdev->dev);
+	if (!info)
+		return NULL;
+
+	info->par = ifbdev;
+	ifbdev->helper.fb = fb;
+	ifbdev->helper.fbdev = info;
+
+	strcpy(info->fix.id, "inteldrmfb");
+
+	info->flags = FBINFO_DEFAULT | FBINFO_CAN_FORCE_OUTPUT;
+	info->fbops = &intelfb_ops;
+
+	ret = fb_alloc_cmap(&info->cmap, 256, 0);
+	if (ret)
+		goto err_info;
+
+	/* setup aperture base/size for vesafb takeover */
+	info->apertures = alloc_apertures(1);
+	if (!info->apertures)
+		goto err_cmap;
+
+	info->apertures->ranges[0].base = dev->mode_config.fb_base;
+	info->apertures->ranges[0].size =
+		dev_priv->mm.gtt->gtt_mappable_entries << PAGE_SHIFT;
+
+	gtt_offset = ifbdev->ifb.obj->gtt_offset;
+	size = ifbdev->ifb.obj->base.size;
+
+	info->fix.smem_start = dev->mode_config.fb_base + gtt_offset;
+	info->fix.smem_len = size;
+
+	info->screen_size = size;
+	info->screen_base = ioremap_wc(dev->agp->base + gtt_offset, size);
+	if (!info->screen_base)
+		goto err_cmap;
+
+	/* Use default scratch pixmap (info->pixmap.flags = FB_PIXMAP_SYSTEM) */
+
+	drm_fb_helper_fill_fix(info, fb->pitches[0], fb->depth);
+	drm_fb_helper_fill_var(info, &ifbdev->helper, fb->width, fb->height);
+
+	return info;
+
+err_cmap:
+	if (info->cmap.len)
+		fb_dealloc_cmap(&info->cmap);
+err_info:
+	framebuffer_release(info);
+	return NULL;
+}
+
 static int intelfb_create(struct intel_fbdev *ifbdev,
 			  struct drm_fb_helper_surface_size *sizes)
 {
 	struct drm_device *dev = ifbdev->helper.dev;
-	struct drm_i915_private *dev_priv = dev->dev_private;
-	struct fb_info *info;
-	struct drm_framebuffer *fb;
 	struct drm_mode_fb_cmd2 mode_cmd;
 	struct drm_i915_gem_object *obj;
-	struct device *device = &dev->pdev->dev;
+	struct fb_info *info;
 	int size, ret;
 
 	/* we don't do packed 24bpp */
@@ -102,64 +160,19 @@ static int intelfb_create(struct intel_fbdev *ifbdev,
 		goto out_unref;
 	}
 
-	info = framebuffer_alloc(0, device);
-	if (!info) {
-		ret = -ENOMEM;
-		goto out_unpin;
-	}
-
-	info->par = ifbdev;
-
 	ret = intel_framebuffer_init(dev, &ifbdev->ifb, &mode_cmd, obj);
 	if (ret)
 		goto out_unpin;
 
-	fb = &ifbdev->ifb.base;
-
-	ifbdev->helper.fb = fb;
-	ifbdev->helper.fbdev = info;
-
-	strcpy(info->fix.id, "inteldrmfb");
-
-	info->flags = FBINFO_DEFAULT | FBINFO_CAN_FORCE_OUTPUT;
-	info->fbops = &intelfb_ops;
+	DRM_DEBUG_KMS("allocated %dx%d fb: 0x%08x, bo %p\n",
+		      mode_cmd.width, mode_cmd.height,
+		      obj->gtt_offset, obj);
 
-	ret = fb_alloc_cmap(&info->cmap, 256, 0);
-	if (ret) {
-		ret = -ENOMEM;
-		goto out_unpin;
-	}
-	/* setup aperture base/size for vesafb takeover */
-	info->apertures = alloc_apertures(1);
-	if (!info->apertures) {
+	info = intelfb_create_info(ifbdev);
+	if (info == NULL) {
 		ret = -ENOMEM;
 		goto out_unpin;
 	}
-	info->apertures->ranges[0].base = dev->mode_config.fb_base;
-	info->apertures->ranges[0].size =
-		dev_priv->mm.gtt->gtt_mappable_entries << PAGE_SHIFT;
-
-	info->fix.smem_start = dev->mode_config.fb_base + obj->gtt_offset;
-	info->fix.smem_len = size;
-
-	info->screen_base = ioremap_wc(dev->agp->base + obj->gtt_offset, size);
-	if (!info->screen_base) {
-		ret = -ENOSPC;
-		goto out_unpin;
-	}
-	info->screen_size = size;
-
-//	memset(info->screen_base, 0, size);
-
-	drm_fb_helper_fill_fix(info, fb->pitches[0], fb->depth);
-	drm_fb_helper_fill_var(info, &ifbdev->helper, sizes->fb_width, sizes->fb_height);
-
-	/* Use default scratch pixmap (info->pixmap.flags = FB_PIXMAP_SYSTEM) */
-
-	DRM_DEBUG_KMS("allocated %dx%d fb: 0x%08x, bo %p\n",
-		      fb->width, fb->height,
-		      obj->gtt_offset, obj);
-
 
 	mutex_unlock(&dev->struct_mutex);
 	vga_switcheroo_client_fb_set(dev->pdev, info);
@@ -199,11 +212,11 @@ static struct drm_fb_helper_funcs intel_fb_helper_funcs = {
 static void intel_fbdev_destroy(struct drm_device *dev,
 				struct intel_fbdev *ifbdev)
 {
-	struct fb_info *info;
 	struct intel_framebuffer *ifb = &ifbdev->ifb;
 
 	if (ifbdev->helper.fbdev) {
-		info = ifbdev->helper.fbdev;
+		struct fb_info *info = ifbdev->helper.fbdev;
+
 		unregister_framebuffer(info);
 		iounmap(info->screen_base);
 		if (info->cmap.len)
-- 
1.7.10

[PATCH 28/28] drm/i915: Wrap the preallocated BIOS framebuffer and preserve for KMS fbcon

[Chris Wilson]

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_dma.c      |    4 ++
 drivers/gpu/drm/i915/intel_display.c |    3 -
 drivers/gpu/drm/i915/intel_drv.h     |    1 +
 drivers/gpu/drm/i915/intel_fb.c      |  118 ++++++++++++++++++++++++++++++++--
 4 files changed, 119 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
index 9e80d5a..591b8a6 100644
--- a/drivers/gpu/drm/i915/i915_dma.c
+++ b/drivers/gpu/drm/i915/i915_dma.c
@@ -1104,6 +1104,9 @@ static int i915_load_modeset_init(struct drm_device *dev)
 
 	intel_modeset_init(dev);
 
+	/* Wrap existing BIOS mode configuration prior to GEM takeover */
+	intel_fbdev_init_bios(dev);
+
 	ret = i915_gem_init(dev);
 	if (ret)
 		goto cleanup_gem_stolen;
@@ -1118,6 +1121,7 @@ static int i915_load_modeset_init(struct drm_device *dev)
 	/* FIXME: do pre/post-mode set stuff in core KMS code */
 	dev->vblank_disable_allowed = 1;
 
+	/* Install a default KMS/GEM fbcon if we failed to wrap the BIOS fb */
 	ret = intel_fbdev_init(dev);
 	if (ret)
 		goto cleanup_irq;
diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index 69434b6..8f2b1d4 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -6354,9 +6354,6 @@ static void intel_setup_outputs(struct drm_device *dev)
 			intel_encoder_clones(dev, encoder->clone_mask);
 	}
 
-	/* disable all the possible outputs/crtcs before entering KMS mode */
-	drm_helper_disable_unused_functions(dev);
-
 	if (HAS_PCH_SPLIT(dev))
 		ironlake_init_pch_refclk(dev);
 }
diff --git a/drivers/gpu/drm/i915/intel_drv.h b/drivers/gpu/drm/i915/intel_drv.h
index 4b7ec44..f64662a 100644
--- a/drivers/gpu/drm/i915/intel_drv.h
+++ b/drivers/gpu/drm/i915/intel_drv.h
@@ -415,6 +415,7 @@ extern int intel_framebuffer_init(struct drm_device *dev,
 				  struct intel_framebuffer *ifb,
 				  struct drm_mode_fb_cmd2 *mode_cmd,
 				  struct drm_i915_gem_object *obj);
+extern void intel_fbdev_init_bios(struct drm_device *dev);
 extern int intel_fbdev_init(struct drm_device *dev);
 extern void intel_fbdev_fini(struct drm_device *dev);
 extern void intel_fbdev_set_suspend(struct drm_device *dev, int state);
diff --git a/drivers/gpu/drm/i915/intel_fb.c b/drivers/gpu/drm/i915/intel_fb.c
index c4d6cff..f09ce31 100644
--- a/drivers/gpu/drm/i915/intel_fb.c
+++ b/drivers/gpu/drm/i915/intel_fb.c
@@ -233,19 +233,127 @@ static void intel_fbdev_destroy(struct drm_device *dev,
 	}
 }
 
-int intel_fbdev_init(struct drm_device *dev)
+void intel_fbdev_init_bios(struct drm_device *dev)
 {
+	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct intel_fbdev *ifbdev;
+	int pipe, ret;
+
+	for_each_pipe(pipe) {
+		struct intel_crtc *crtc = to_intel_crtc(dev_priv->pipe_to_crtc_mapping[pipe]);
+		struct drm_mode_fb_cmd2 mode_cmd;
+		struct drm_i915_gem_object *obj;
+		struct fb_info *info;
+		u32 val, bpp, depth;
+		u32 size, offset;
+
+		val = I915_READ(DSPCNTR(crtc->plane));
+		if ((val & DISPLAY_PLANE_ENABLE) == 0)
+			continue;
+
+		if (INTEL_INFO(dev)->gen >= 4) {
+			if (val & DISPPLANE_TILED)
+				continue; /* unexpected! */
+		}
+
+		switch (val & DISPPLANE_PIXFORMAT_MASK) {
+		default:
+		case DISPPLANE_8BPP:
+			continue; /* ignore palettes */
+		case DISPPLANE_15_16BPP:
+			depth = 15; bpp = 16;
+			break;
+		case DISPPLANE_16BPP:
+			depth = bpp = 16;
+			break;
+		case DISPPLANE_32BPP_NO_ALPHA:
+			depth = 24; bpp = 32;
+			break;
+		}
+
+		if (INTEL_INFO(dev)->gen >= 4)
+			offset = I915_READ(DSPSURF(crtc->plane));
+		else
+			offset = I915_READ(DSPADDR(crtc->plane));
+		mode_cmd.pitches[0] = I915_READ(DSPSTRIDE(crtc->plane));
+
+		val = I915_READ(PIPESRC(crtc->pipe));
+		mode_cmd.width = ((val >> 16) & 0xfff) + 1;
+		mode_cmd.height = ((val >> 0) & 0xfff) + 1;
+		mode_cmd.pixel_format = drm_mode_legacy_fb_format(bpp, depth);
+
+		DRM_DEBUG_KMS("Found active pipe [%d/%d]: size=%dx%d@%d, offset=%x\n",
+			      crtc->pipe, crtc->plane,
+			      mode_cmd.width, mode_cmd.height, depth, offset);
+
+		size = mode_cmd.pitches[0] * mode_cmd.height;
+		size = ALIGN(size, PAGE_SIZE);
+
+		ifbdev = kzalloc(sizeof(struct intel_fbdev), GFP_KERNEL);
+		if (!ifbdev)
+			continue;
+
+		ifbdev->helper.funcs = &intel_fb_helper_funcs;
+		ret = drm_fb_helper_init(dev, &ifbdev->helper,
+					 dev_priv->num_pipe,
+					 INTELFB_CONN_LIMIT);
+		if (ret) {
+			kfree(ifbdev);
+			continue;
+		}
+
+		/* assume a 1:1 linear mapping between stolen and GTT */
+		obj = i915_gem_object_create_stolen_for_preallocated
+			(dev, offset, offset, size);
+		if (obj == NULL) {
+			kfree(ifbdev);
+			continue;
+		}
+
+		ret = intel_framebuffer_init(dev, &ifbdev->ifb, &mode_cmd, obj);
+		if (ret) {
+			drm_gem_object_unreference(&obj->base);
+			kfree(ifbdev);
+			continue;
+		}
+
+		info = intelfb_create_info(ifbdev);
+		if (info == NULL) {
+			drm_gem_object_unreference(&obj->base);
+			kfree(ifbdev);
+			continue;
+		}
+
+		crtc->base.fb = &ifbdev->ifb.base;
+		obj->pin_count++;
+
+		drm_fb_helper_single_add_all_connectors(&ifbdev->helper);
+		drm_fb_helper_hotplug_event(&ifbdev->helper);
+
+		vga_switcheroo_client_fb_set(dev->pdev, info);
+		dev_priv->fbdev = ifbdev;
+		return;
+	}
+
+	/* otherwise disable all the possible crtcs before KMS */
+	DRM_DEBUG_KMS("No active planes found\n");
+	drm_helper_disable_unused_functions(dev);
+}
+
+int intel_fbdev_init(struct drm_device *dev)
+{
 	drm_i915_private_t *dev_priv = dev->dev_private;
+	struct intel_fbdev *ifbdev;
 	int ret;
 
+	if (dev_priv->fbdev)
+		return 0;
+
 	ifbdev = kzalloc(sizeof(struct intel_fbdev), GFP_KERNEL);
-	if (!ifbdev)
+	if (ifbdev == NULL)
 		return -ENOMEM;
 
-	dev_priv->fbdev = ifbdev;
 	ifbdev->helper.funcs = &intel_fb_helper_funcs;
-
 	ret = drm_fb_helper_init(dev, &ifbdev->helper,
 				 dev_priv->num_pipe,
 				 INTELFB_CONN_LIMIT);
@@ -254,6 +362,8 @@ int intel_fbdev_init(struct drm_device *dev)
 		return ret;
 	}
 
+	dev_priv->fbdev = ifbdev;
+
 	drm_fb_helper_single_add_all_connectors(&ifbdev->helper);
 	drm_fb_helper_initial_config(&ifbdev->helper, 32);
 	return 0;
-- 
1.7.10

[PATCH] [-1/28] drm/i915: Remove i915_gem_evict_inactive()

[Chris Wilson]

This was only used by one external caller who would just be as happy
with evict-everything, so perform the replacement and make the function
private.

In the process we note that unbinding the inactive list should not fail,
and make it a warning instead.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
Another early patch that is used later in the stolen series.
-Chris
---
 drivers/gpu/drm/i915/i915_drv.h       |    5 +----
 drivers/gpu/drm/i915/i915_gem.c       |    9 ++-------
 drivers/gpu/drm/i915/i915_gem_evict.c |   21 ++++++---------------
 3 files changed, 9 insertions(+), 26 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index c144013..57f60fa 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1370,10 +1370,7 @@ void i915_gem_init_global_gtt(struct drm_device *dev,
 /* i915_gem_evict.c */
 int __must_check i915_gem_evict_something(struct drm_device *dev, int min_size,
 					  unsigned alignment, bool mappable);
-int __must_check i915_gem_evict_everything(struct drm_device *dev,
-					   bool purgeable_only);
-int __must_check i915_gem_evict_inactive(struct drm_device *dev,
-					 bool purgeable_only);
+int i915_gem_evict_everything(struct drm_device *dev, bool purgeable_only);
 
 /* i915_gem_tiling.c */
 void i915_gem_detect_bit_6_swizzle(struct drm_device *dev);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index cfbcf7e..1cff8b1 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3450,13 +3450,8 @@ i915_gem_idle(struct drm_device *dev)
 	}
 
 	/* Under UMS, be paranoid and evict. */
-	if (!drm_core_check_feature(dev, DRIVER_MODESET)) {
-		ret = i915_gem_evict_inactive(dev, false);
-		if (ret) {
-			mutex_unlock(&dev->struct_mutex);
-			return ret;
-		}
-	}
+	if (!drm_core_check_feature(dev, DRIVER_MODESET))
+		i915_gem_evict_everything(dev, false);
 
 	i915_gem_reset_fences(dev);
 
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 21a8271..399a3a8 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -166,8 +166,9 @@ int
 i915_gem_evict_everything(struct drm_device *dev, bool purgeable_only)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
-	int ret;
+	struct drm_i915_gem_object *obj, *next;
 	bool lists_empty;
+	int ret;
 
 	lists_empty = (list_empty(&dev_priv->mm.inactive_list) &&
 		       list_empty(&dev_priv->mm.flushing_list) &&
@@ -184,24 +185,14 @@ i915_gem_evict_everything(struct drm_device *dev, bool purgeable_only)
 
 	BUG_ON(!list_empty(&dev_priv->mm.flushing_list));
 
-	return i915_gem_evict_inactive(dev, purgeable_only);
-}
-
-/** Unbinds all inactive objects. */
-int
-i915_gem_evict_inactive(struct drm_device *dev, bool purgeable_only)
-{
-	drm_i915_private_t *dev_priv = dev->dev_private;
-	struct drm_i915_gem_object *obj, *next;
-
+	/* Having flushed everything, unbind() should never raise an error */
 	list_for_each_entry_safe(obj, next,
 				 &dev_priv->mm.inactive_list, mm_list) {
 		if (!purgeable_only || obj->madv != I915_MADV_WILLNEED) {
-			int ret = i915_gem_object_unbind(obj);
-			if (ret)
-				return ret;
+			if (obj->pin_count == 0)
+				WARN_ON(i915_gem_object_unbind(obj));
 		}
 	}
 
-	return 0;
+	return ret;
 }
-- 
1.7.10

Re: [PATCH] [-1/28] drm/i915: Remove i915_gem_evict_inactive()

[Daniel Vetter]

On Tue, Apr 24, 2012 at 06:22:52PM +0100, Chris Wilson wrote:
> This was only used by one external caller who would just be as happy
> with evict-everything, so perform the replacement and make the function
> private.
> 
> In the process we note that unbinding the inactive list should not fail,
> and make it a warning instead.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
I've queued patches -1,1,2,10&12 from this series for next, with some
bikeshed added to the commit message of patch 2.

Thanks, Daniel
-- 
Daniel Vetter
Mail: daniel@ffwll.ch
Mobile: +41 (0)79 365 57 48

[PATCH] drm/i915: Fix location of stolen memory register for SandyBridge+

[Chris Wilson]

A few of the earlier registers where enlarged and so the Base Data of
Stolem Memory Register (BDSM) was pushed to 0xb0.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_stolen.c |    8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index f9aea38..f77b2a8 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -63,7 +63,10 @@ static unsigned long i915_stolen_to_physical(struct drm_device *dev)
 	 * its value of TOLUD.
 	 */
 	base = 0;
-	if (INTEL_INFO(dev)->gen > 3 || IS_G33(dev)) {
+	if (INTEL_INFO(dev)->gen >= 6) {
+		/* Read Base Data of Stolen Memory Register (BDSM) directly */
+		pci_read_config_dword(pdev, 0xB0, &base);
+	} else if (INTEL_INFO(dev)->gen > 3 || IS_G33(dev)) {
 		/* Read Graphics Base of Stolen Memory directly */
 		pci_read_config_dword(pdev, 0xA4, &base);
 #if 0
@@ -207,6 +210,9 @@ int i915_gem_init_stolen(struct drm_device *dev)
 	if (dev_priv->mm.stolen_base == 0)
 		return 0;
 
+	DRM_DEBUG_KMS("found %d bytes of stolen memory at %08lx\n",
+		      dev_priv->mm.gtt->stolen_size, dev_priv->mm.stolen_base);
+
 	/* Basic memrange allocator for stolen space */
 	drm_mm_init(&dev_priv->mm.stolen, 0, dev_priv->mm.gtt->stolen_size);
 
-- 
1.7.10

Re: [PATCH 13/28] drm/i915: Avoid clearing preallocated regions from the GTT

[Daniel Vetter]

On Tue, Apr 24, 2012 at 03:47:42PM +0100, Chris Wilson wrote:
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/i915/i915_drv.h     |    2 ++
>  drivers/gpu/drm/i915/i915_gem_gtt.c |   32 +++++++++++++++++++++++++++++---
>  2 files changed, 31 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 32e6c4d..3492e20 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -827,6 +827,8 @@ enum i915_cache_level {
>  	I915_CACHE_LLC_MLC, /* gen6+ */
>  };
>  
> +#define I915_GTT_RESERVED ((struct drm_mm_node *)0x1)
> +
>  struct drm_i915_gem_object {
>  	struct drm_gem_object base;
>  
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
> index 25c8bf9..b0c20a5 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
> @@ -417,16 +417,42 @@ void i915_gem_init_global_gtt(struct drm_device *dev,
>  			      unsigned long end)
>  {
>  	drm_i915_private_t *dev_priv = dev->dev_private;
> +	struct drm_mm_node *entry;
> +	struct drm_i915_gem_object *obj;
>  
> -	/* Substract the guard page ... */
> +	/* Subtract the guard page ... */
>  	drm_mm_init(&dev_priv->mm.gtt_space, start, end - start - PAGE_SIZE);
>  
> +	/* Mark any preallocated objects as occupied */
> +	list_for_each_entry(obj, &dev_priv->mm.gtt_list, gtt_list) {
> +		DRM_DEBUG_KMS("reserving preallocated space: %x + %zx\n",
> +			      obj->gtt_offset, obj->base.size);
> +
> +		BUG_ON(obj->gtt_space != I915_GTT_RESERVED);
> +		obj->gtt_space = drm_mm_create_block(&dev_priv->mm.gtt_space,
> +						     obj->gtt_offset,
> +						     obj->base.size,
> +						     false);
> +		obj->has_global_gtt_mapping = 1;
> +	}
> +
>  	dev_priv->mm.gtt_start = start;
>  	dev_priv->mm.gtt_mappable_end = mappable_end;
>  	dev_priv->mm.gtt_end = end;
>  	dev_priv->mm.gtt_total = end - start;
>  	dev_priv->mm.mappable_gtt_total = min(end, mappable_end) - start;
>  
> -	/* ... but ensure that we clear the entire range. */
> -	intel_gtt_clear_range(start / PAGE_SIZE, (end-start) / PAGE_SIZE);
> +	/* Clear any non-preallocated blocks */
> +	list_for_each_entry(entry, &dev_priv->mm.gtt_space.hole_stack, hole_stack) {
> +		unsigned long hole_start = entry->start + entry->size;
> +		unsigned long hole_end = list_entry(entry->node_list.next,
> +						    struct drm_mm_node,
> +						    node_list)->start;
> +
> +		DRM_DEBUG_KMS("clearing unused GTT space: [%lx, %lx]\n",
> +			      hole_start, hole_end);
> +
> +		intel_gtt_clear_range(hole_start / PAGE_SIZE,
> +				      (hole_end-hole_start) / PAGE_SIZE);
> +	}

Assuming I'm reading the code correctly this does not clear the unmanaged
guard page at the end of the gtt.
-Daniel
-- 
Daniel Vetter
Mail: daniel@ffwll.ch
Mobile: +41 (0)79 365 57 48

Re: [PATCH 13/28] drm/i915: Avoid clearing preallocated regions from the GTT

[Chris Wilson]

On Tue, 1 May 2012 19:30:10 +0200, Daniel Vetter <daniel@ffwll.ch> wrote:
> On Tue, Apr 24, 2012 at 03:47:42PM +0100, Chris Wilson wrote:
> > +	/* Clear any non-preallocated blocks */
> > +	list_for_each_entry(entry, &dev_priv->mm.gtt_space.hole_stack, hole_stack) {
> > +		unsigned long hole_start = entry->start + entry->size;
> > +		unsigned long hole_end = list_entry(entry->node_list.next,
> > +						    struct drm_mm_node,
> > +						    node_list)->start;
> > +
> > +		DRM_DEBUG_KMS("clearing unused GTT space: [%lx, %lx]\n",
> > +			      hole_start, hole_end);
> > +
> > +		intel_gtt_clear_range(hole_start / PAGE_SIZE,
> > +				      (hole_end-hole_start) / PAGE_SIZE);
> > +	}
> 
> Assuming I'm reading the code correctly this does not clear the unmanaged
> guard page at the end of the gtt.

Right, missed. The simplest fix would seem to be do that as an explicit
step afterwards. Bonus marks for commentary.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre