unmappable vram V7

unmappable vram V7

[Jerome Glisse]

So here is i hope the last spawn of the patch serie,
i drop the needs ioremap flag and callback is responsible
for ioremapping memory and providing a valid virtual address
this simplify TTM code. I haven't change the value of enum
after removing dead flag, Thomas maybe you prefer to change
the enum values ?

Again only tested on radeon, just compile tested for vmwgfx
and nouveau.

Cheers,
Jerome


------------------------------------------------------------------------------
Download Intel® Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 01/13] drm/ttm: split no_wait argument in 2 GPU or reserve wait

[Jerome Glisse]

There is case where we want to be able to wait only for the
GPU while not waiting for other buffer to be unreserved. This
patch split the no_wait argument all the way down in the whole
ttm path so that upper level can decide on what to wait on or
not.

This patch break the API to other modules, update to others
driver are following in separate patches.

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c      |   57 ++++++++++++++++++++----------------
 drivers/gpu/drm/ttm/ttm_bo_util.c |    9 ++++--
 include/drm/ttm/ttm_bo_api.h      |    6 ++-
 include/drm/ttm/ttm_bo_driver.h   |   29 +++++++++++-------
 4 files changed, 60 insertions(+), 41 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 9db02bb..6f51b30 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -357,7 +357,8 @@ static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
 
 static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 				  struct ttm_mem_reg *mem,
-				  bool evict, bool interruptible, bool no_wait)
+				  bool evict, bool interruptible,
+				  bool no_wait_reserve, bool no_wait_gpu)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
 	bool old_is_pci = ttm_mem_reg_is_pci(bdev, &bo->mem);
@@ -402,12 +403,12 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 
 	if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
 	    !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
-		ret = ttm_bo_move_ttm(bo, evict, no_wait, mem);
+		ret = ttm_bo_move_ttm(bo, evict, no_wait_reserve, no_wait_gpu, mem);
 	else if (bdev->driver->move)
 		ret = bdev->driver->move(bo, evict, interruptible,
-					 no_wait, mem);
+					 no_wait_reserve, no_wait_gpu, mem);
 	else
-		ret = ttm_bo_move_memcpy(bo, evict, no_wait, mem);
+		ret = ttm_bo_move_memcpy(bo, evict, no_wait_reserve, no_wait_gpu, mem);
 
 	if (ret)
 		goto out_err;
@@ -606,7 +607,7 @@ void ttm_bo_unref(struct ttm_buffer_object **p_bo)
 EXPORT_SYMBOL(ttm_bo_unref);
 
 static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
-			bool no_wait)
+			bool no_wait_reserve, bool no_wait_gpu)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
 	struct ttm_bo_global *glob = bo->glob;
@@ -615,7 +616,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
 	int ret = 0;
 
 	spin_lock(&bo->lock);
-	ret = ttm_bo_wait(bo, false, interruptible, no_wait);
+	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
 	spin_unlock(&bo->lock);
 
 	if (unlikely(ret != 0)) {
@@ -638,7 +639,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
 	placement.num_busy_placement = 0;
 	bdev->driver->evict_flags(bo, &placement);
 	ret = ttm_bo_mem_space(bo, &placement, &evict_mem, interruptible,
-				no_wait);
+				no_wait_reserve, no_wait_gpu);
 	if (ret) {
 		if (ret != -ERESTARTSYS) {
 			printk(KERN_ERR TTM_PFX
@@ -650,7 +651,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
 	}
 
 	ret = ttm_bo_handle_move_mem(bo, &evict_mem, true, interruptible,
-				     no_wait);
+				     no_wait_reserve, no_wait_gpu);
 	if (ret) {
 		if (ret != -ERESTARTSYS)
 			printk(KERN_ERR TTM_PFX "Buffer eviction failed\n");
@@ -670,7 +671,8 @@ out:
 
 static int ttm_mem_evict_first(struct ttm_bo_device *bdev,
 				uint32_t mem_type,
-				bool interruptible, bool no_wait)
+				bool interruptible, bool no_wait_reserve,
+				bool no_wait_gpu)
 {
 	struct ttm_bo_global *glob = bdev->glob;
 	struct ttm_mem_type_manager *man = &bdev->man[mem_type];
@@ -687,11 +689,11 @@ retry:
 	bo = list_first_entry(&man->lru, struct ttm_buffer_object, lru);
 	kref_get(&bo->list_kref);
 
-	ret = ttm_bo_reserve_locked(bo, false, true, false, 0);
+	ret = ttm_bo_reserve_locked(bo, false, no_wait_reserve, false, 0);
 
 	if (unlikely(ret == -EBUSY)) {
 		spin_unlock(&glob->lru_lock);
-		if (likely(!no_wait))
+		if (likely(!no_wait_gpu))
 			ret = ttm_bo_wait_unreserved(bo, interruptible);
 
 		kref_put(&bo->list_kref, ttm_bo_release_list);
@@ -713,7 +715,7 @@ retry:
 	while (put_count--)
 		kref_put(&bo->list_kref, ttm_bo_ref_bug);
 
-	ret = ttm_bo_evict(bo, interruptible, no_wait);
+	ret = ttm_bo_evict(bo, interruptible, no_wait_reserve, no_wait_gpu);
 	ttm_bo_unreserve(bo);
 
 	kref_put(&bo->list_kref, ttm_bo_release_list);
@@ -764,7 +766,9 @@ static int ttm_bo_mem_force_space(struct ttm_buffer_object *bo,
 					uint32_t mem_type,
 					struct ttm_placement *placement,
 					struct ttm_mem_reg *mem,
-					bool interruptible, bool no_wait)
+					bool interruptible,
+					bool no_wait_reserve,
+					bool no_wait_gpu)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
 	struct ttm_bo_global *glob = bdev->glob;
@@ -785,7 +789,7 @@ static int ttm_bo_mem_force_space(struct ttm_buffer_object *bo,
 		}
 		spin_unlock(&glob->lru_lock);
 		ret = ttm_mem_evict_first(bdev, mem_type, interruptible,
-						no_wait);
+						no_wait_reserve, no_wait_gpu);
 		if (unlikely(ret != 0))
 			return ret;
 	} while (1);
@@ -855,7 +859,8 @@ static bool ttm_bo_mt_compatible(struct ttm_mem_type_manager *man,
 int ttm_bo_mem_space(struct ttm_buffer_object *bo,
 			struct ttm_placement *placement,
 			struct ttm_mem_reg *mem,
-			bool interruptible, bool no_wait)
+			bool interruptible, bool no_wait_reserve,
+			bool no_wait_gpu)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
 	struct ttm_mem_type_manager *man;
@@ -952,7 +957,7 @@ int ttm_bo_mem_space(struct ttm_buffer_object *bo,
 		}
 
 		ret = ttm_bo_mem_force_space(bo, mem_type, placement, mem,
-						interruptible, no_wait);
+						interruptible, no_wait_reserve, no_wait_gpu);
 		if (ret == 0 && mem->mm_node) {
 			mem->placement = cur_flags;
 			mem->mm_node->private = bo;
@@ -978,7 +983,8 @@ EXPORT_SYMBOL(ttm_bo_wait_cpu);
 
 int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 			struct ttm_placement *placement,
-			bool interruptible, bool no_wait)
+			bool interruptible, bool no_wait_reserve,
+			bool no_wait_gpu)
 {
 	struct ttm_bo_global *glob = bo->glob;
 	int ret = 0;
@@ -992,7 +998,7 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 	 * instead of doing it here.
 	 */
 	spin_lock(&bo->lock);
-	ret = ttm_bo_wait(bo, false, interruptible, no_wait);
+	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
 	spin_unlock(&bo->lock);
 	if (ret)
 		return ret;
@@ -1002,10 +1008,10 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 	/*
 	 * Determine where to move the buffer.
 	 */
-	ret = ttm_bo_mem_space(bo, placement, &mem, interruptible, no_wait);
+	ret = ttm_bo_mem_space(bo, placement, &mem, interruptible, no_wait_reserve, no_wait_gpu);
 	if (ret)
 		goto out_unlock;
-	ret = ttm_bo_handle_move_mem(bo, &mem, false, interruptible, no_wait);
+	ret = ttm_bo_handle_move_mem(bo, &mem, false, interruptible, no_wait_reserve, no_wait_gpu);
 out_unlock:
 	if (ret && mem.mm_node) {
 		spin_lock(&glob->lru_lock);
@@ -1039,7 +1045,8 @@ static int ttm_bo_mem_compat(struct ttm_placement *placement,
 
 int ttm_bo_validate(struct ttm_buffer_object *bo,
 			struct ttm_placement *placement,
-			bool interruptible, bool no_wait)
+			bool interruptible, bool no_wait_reserve,
+			bool no_wait_gpu)
 {
 	int ret;
 
@@ -1054,7 +1061,7 @@ int ttm_bo_validate(struct ttm_buffer_object *bo,
 	 */
 	ret = ttm_bo_mem_compat(placement, &bo->mem);
 	if (ret < 0) {
-		ret = ttm_bo_move_buffer(bo, placement, interruptible, no_wait);
+		ret = ttm_bo_move_buffer(bo, placement, interruptible, no_wait_reserve, no_wait_gpu);
 		if (ret)
 			return ret;
 	} else {
@@ -1175,7 +1182,7 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
 			goto out_err;
 	}
 
-	ret = ttm_bo_validate(bo, placement, interruptible, false);
+	ret = ttm_bo_validate(bo, placement, interruptible, false, false);
 	if (ret)
 		goto out_err;
 
@@ -1249,7 +1256,7 @@ static int ttm_bo_force_list_clean(struct ttm_bo_device *bdev,
 	spin_lock(&glob->lru_lock);
 	while (!list_empty(&man->lru)) {
 		spin_unlock(&glob->lru_lock);
-		ret = ttm_mem_evict_first(bdev, mem_type, false, false);
+		ret = ttm_mem_evict_first(bdev, mem_type, false, false, false);
 		if (ret) {
 			if (allow_errors) {
 				return ret;
@@ -1839,7 +1846,7 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
 		evict_mem.mem_type = TTM_PL_SYSTEM;
 
 		ret = ttm_bo_handle_move_mem(bo, &evict_mem, true,
-					     false, false);
+					     false, false, false);
 		if (unlikely(ret != 0))
 			goto out;
 	}
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 5ca37a5..865b2a8 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -49,7 +49,8 @@ void ttm_bo_free_old_node(struct ttm_buffer_object *bo)
 }
 
 int ttm_bo_move_ttm(struct ttm_buffer_object *bo,
-		    bool evict, bool no_wait, struct ttm_mem_reg *new_mem)
+		    bool evict, bool no_wait_reserve,
+		    bool no_wait_gpu, struct ttm_mem_reg *new_mem)
 {
 	struct ttm_tt *ttm = bo->ttm;
 	struct ttm_mem_reg *old_mem = &bo->mem;
@@ -207,7 +208,8 @@ static int ttm_copy_ttm_io_page(struct ttm_tt *ttm, void *dst,
 }
 
 int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
-		       bool evict, bool no_wait, struct ttm_mem_reg *new_mem)
+		       bool evict, bool no_wait_reserve, bool no_wait_gpu,
+		       struct ttm_mem_reg *new_mem)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
 	struct ttm_mem_type_manager *man = &bdev->man[new_mem->mem_type];
@@ -525,7 +527,8 @@ int ttm_bo_pfn_prot(struct ttm_buffer_object *bo,
 int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 			      void *sync_obj,
 			      void *sync_obj_arg,
-			      bool evict, bool no_wait,
+			      bool evict, bool no_wait_reserve,
+			      bool no_wait_gpu,
 			      struct ttm_mem_reg *new_mem)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index 81eb9f4..8c8005e 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -313,7 +313,8 @@ extern int ttm_bo_wait(struct ttm_buffer_object *bo, bool lazy,
  * @bo: The buffer object.
  * @placement: Proposed placement for the buffer object.
  * @interruptible: Sleep interruptible if sleeping.
- * @no_wait: Return immediately if the buffer is busy.
+ * @no_wait_reserve: Return immediately if other buffers are busy.
+ * @no_wait_gpu: Return immediately if the GPU is busy.
  *
  * Changes placement and caching policy of the buffer object
  * according proposed placement.
@@ -325,7 +326,8 @@ extern int ttm_bo_wait(struct ttm_buffer_object *bo, bool lazy,
  */
 extern int ttm_bo_validate(struct ttm_buffer_object *bo,
 				struct ttm_placement *placement,
-				bool interruptible, bool no_wait);
+				bool interruptible, bool no_wait_reserve,
+				bool no_wait_gpu);
 
 /**
  * ttm_bo_unref
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index e929c27..69f70e4 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -311,7 +311,8 @@ struct ttm_bo_driver {
 	 */
 	int (*move) (struct ttm_buffer_object *bo,
 		     bool evict, bool interruptible,
-		     bool no_wait, struct ttm_mem_reg *new_mem);
+		     bool no_wait_reserve, bool no_wait_gpu,
+		     struct ttm_mem_reg *new_mem);
 
 	/**
 	 * struct ttm_bo_driver_member verify_access
@@ -633,7 +634,8 @@ extern bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev,
  * @proposed_placement: Proposed new placement for the buffer object.
  * @mem: A struct ttm_mem_reg.
  * @interruptible: Sleep interruptible when sliping.
- * @no_wait: Don't sleep waiting for space to become available.
+ * @no_wait_reserve: Return immediately if other buffers are busy.
+ * @no_wait_gpu: Return immediately if the GPU is busy.
  *
  * Allocate memory space for the buffer object pointed to by @bo, using
  * the placement flags in @mem, potentially evicting other idle buffer objects.
@@ -647,7 +649,8 @@ extern bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev,
 extern int ttm_bo_mem_space(struct ttm_buffer_object *bo,
 				struct ttm_placement *placement,
 				struct ttm_mem_reg *mem,
-				bool interruptible, bool no_wait);
+				bool interruptible,
+				bool no_wait_reserve, bool no_wait_gpu);
 /**
  * ttm_bo_wait_for_cpu
  *
@@ -826,7 +829,8 @@ extern void ttm_bo_unblock_reservation(struct ttm_buffer_object *bo);
  *
  * @bo: A pointer to a struct ttm_buffer_object.
  * @evict: 1: This is an eviction. Don't try to pipeline.
- * @no_wait: Never sleep, but rather return with -EBUSY.
+ * @no_wait_reserve: Return immediately if other buffers are busy.
+ * @no_wait_gpu: Return immediately if the GPU is busy.
  * @new_mem: struct ttm_mem_reg indicating where to move.
  *
  * Optimized move function for a buffer object with both old and
@@ -840,15 +844,16 @@ extern void ttm_bo_unblock_reservation(struct ttm_buffer_object *bo);
  */
 
 extern int ttm_bo_move_ttm(struct ttm_buffer_object *bo,
-			   bool evict, bool no_wait,
-			   struct ttm_mem_reg *new_mem);
+			   bool evict, bool no_wait_reserve,
+			   bool no_wait_gpu, struct ttm_mem_reg *new_mem);
 
 /**
  * ttm_bo_move_memcpy
  *
  * @bo: A pointer to a struct ttm_buffer_object.
  * @evict: 1: This is an eviction. Don't try to pipeline.
- * @no_wait: Never sleep, but rather return with -EBUSY.
+ * @no_wait_reserve: Return immediately if other buffers are busy.
+ * @no_wait_gpu: Return immediately if the GPU is busy.
  * @new_mem: struct ttm_mem_reg indicating where to move.
  *
  * Fallback move function for a mappable buffer object in mappable memory.
@@ -862,8 +867,8 @@ extern int ttm_bo_move_ttm(struct ttm_buffer_object *bo,
  */
 
 extern int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
-			      bool evict,
-			      bool no_wait, struct ttm_mem_reg *new_mem);
+			      bool evict, bool no_wait_reserve,
+			      bool no_wait_gpu, struct ttm_mem_reg *new_mem);
 
 /**
  * ttm_bo_free_old_node
@@ -882,7 +887,8 @@ extern void ttm_bo_free_old_node(struct ttm_buffer_object *bo);
  * @sync_obj_arg: An argument to pass to the sync object idle / wait
  * functions.
  * @evict: This is an evict move. Don't return until the buffer is idle.
- * @no_wait: Never sleep, but rather return with -EBUSY.
+ * @no_wait_reserve: Return immediately if other buffers are busy.
+ * @no_wait_gpu: Return immediately if the GPU is busy.
  * @new_mem: struct ttm_mem_reg indicating where to move.
  *
  * Accelerated move function to be called when an accelerated move
@@ -896,7 +902,8 @@ extern void ttm_bo_free_old_node(struct ttm_buffer_object *bo);
 extern int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 				     void *sync_obj,
 				     void *sync_obj_arg,
-				     bool evict, bool no_wait,
+				     bool evict, bool no_wait_reserve,
+				     bool no_wait_gpu,
 				     struct ttm_mem_reg *new_mem);
 /**
  * ttm_io_prot
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 02/13] drm/radeon/kms: update to TTM no_wait splitted argument

[Jerome Glisse]

This patch update radeon to the new no_wait splitted argument
TTM functionality.

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/radeon/radeon_object.c |    6 ++--
 drivers/gpu/drm/radeon/radeon_ttm.c    |   39 +++++++++++++++++--------------
 2 files changed, 24 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index fc9d00a..3adfa88 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -189,7 +189,7 @@ int radeon_bo_pin(struct radeon_bo *bo, u32 domain, u64 *gpu_addr)
 	bo->placement.lpfn = bo->rdev->mc.visible_vram_size >> PAGE_SHIFT;
 	for (i = 0; i < bo->placement.num_placement; i++)
 		bo->placements[i] |= TTM_PL_FLAG_NO_EVICT;
-	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
+	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
 	if (likely(r == 0)) {
 		bo->pin_count = 1;
 		if (gpu_addr != NULL)
@@ -213,7 +213,7 @@ int radeon_bo_unpin(struct radeon_bo *bo)
 		return 0;
 	for (i = 0; i < bo->placement.num_placement; i++)
 		bo->placements[i] &= ~TTM_PL_FLAG_NO_EVICT;
-	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
+	r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false, false);
 	if (unlikely(r != 0))
 		dev_err(bo->rdev->dev, "%p validate failed for unpin\n", bo);
 	return r;
@@ -328,7 +328,7 @@ int radeon_bo_list_validate(struct list_head *head)
 								lobj->rdomain);
 			}
 			r = ttm_bo_validate(&bo->tbo, &bo->placement,
-						true, false);
+						true, false, false);
 			if (unlikely(r))
 				return r;
 		}
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index 43c5ab3..ba4724c 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -243,9 +243,9 @@ static void radeon_move_null(struct ttm_buffer_object *bo,
 }
 
 static int radeon_move_blit(struct ttm_buffer_object *bo,
-			    bool evict, int no_wait,
-			    struct ttm_mem_reg *new_mem,
-			    struct ttm_mem_reg *old_mem)
+			bool evict, int no_wait_reserve, bool no_wait_gpu,
+			struct ttm_mem_reg *new_mem,
+			struct ttm_mem_reg *old_mem)
 {
 	struct radeon_device *rdev;
 	uint64_t old_start, new_start;
@@ -289,13 +289,14 @@ static int radeon_move_blit(struct ttm_buffer_object *bo,
 	r = radeon_copy(rdev, old_start, new_start, new_mem->num_pages, fence);
 	/* FIXME: handle copy error */
 	r = ttm_bo_move_accel_cleanup(bo, (void *)fence, NULL,
-				      evict, no_wait, new_mem);
+				      evict, no_wait_reserve, no_wait_gpu, new_mem);
 	radeon_fence_unref(&fence);
 	return r;
 }
 
 static int radeon_move_vram_ram(struct ttm_buffer_object *bo,
-				bool evict, bool interruptible, bool no_wait,
+				bool evict, bool interruptible,
+				bool no_wait_reserve, bool no_wait_gpu,
 				struct ttm_mem_reg *new_mem)
 {
 	struct radeon_device *rdev;
@@ -316,7 +317,7 @@ static int radeon_move_vram_ram(struct ttm_buffer_object *bo,
 	placement.busy_placement = &placements;
 	placements = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT;
 	r = ttm_bo_mem_space(bo, &placement, &tmp_mem,
-			     interruptible, no_wait);
+			     interruptible, no_wait_reserve, no_wait_gpu);
 	if (unlikely(r)) {
 		return r;
 	}
@@ -330,11 +331,11 @@ static int radeon_move_vram_ram(struct ttm_buffer_object *bo,
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
-	r = radeon_move_blit(bo, true, no_wait, &tmp_mem, old_mem);
+	r = radeon_move_blit(bo, true, no_wait_reserve, no_wait_gpu, &tmp_mem, old_mem);
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
-	r = ttm_bo_move_ttm(bo, true, no_wait, new_mem);
+	r = ttm_bo_move_ttm(bo, true, no_wait_reserve, no_wait_gpu, new_mem);
 out_cleanup:
 	if (tmp_mem.mm_node) {
 		struct ttm_bo_global *glob = rdev->mman.bdev.glob;
@@ -348,7 +349,8 @@ out_cleanup:
 }
 
 static int radeon_move_ram_vram(struct ttm_buffer_object *bo,
-				bool evict, bool interruptible, bool no_wait,
+				bool evict, bool interruptible,
+				bool no_wait_reserve, bool no_wait_gpu,
 				struct ttm_mem_reg *new_mem)
 {
 	struct radeon_device *rdev;
@@ -368,15 +370,15 @@ static int radeon_move_ram_vram(struct ttm_buffer_object *bo,
 	placement.num_busy_placement = 1;
 	placement.busy_placement = &placements;
 	placements = TTM_PL_MASK_CACHING | TTM_PL_FLAG_TT;
-	r = ttm_bo_mem_space(bo, &placement, &tmp_mem, interruptible, no_wait);
+	r = ttm_bo_mem_space(bo, &placement, &tmp_mem, interruptible, no_wait_reserve, no_wait_gpu);
 	if (unlikely(r)) {
 		return r;
 	}
-	r = ttm_bo_move_ttm(bo, true, no_wait, &tmp_mem);
+	r = ttm_bo_move_ttm(bo, true, no_wait_reserve, no_wait_gpu, &tmp_mem);
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
-	r = radeon_move_blit(bo, true, no_wait, new_mem, old_mem);
+	r = radeon_move_blit(bo, true, no_wait_reserve, no_wait_gpu, new_mem, old_mem);
 	if (unlikely(r)) {
 		goto out_cleanup;
 	}
@@ -393,8 +395,9 @@ out_cleanup:
 }
 
 static int radeon_bo_move(struct ttm_buffer_object *bo,
-			  bool evict, bool interruptible, bool no_wait,
-			  struct ttm_mem_reg *new_mem)
+			bool evict, bool interruptible,
+			bool no_wait_reserve, bool no_wait_gpu,
+			struct ttm_mem_reg *new_mem)
 {
 	struct radeon_device *rdev;
 	struct ttm_mem_reg *old_mem = &bo->mem;
@@ -421,18 +424,18 @@ static int radeon_bo_move(struct ttm_buffer_object *bo,
 	if (old_mem->mem_type == TTM_PL_VRAM &&
 	    new_mem->mem_type == TTM_PL_SYSTEM) {
 		r = radeon_move_vram_ram(bo, evict, interruptible,
-					    no_wait, new_mem);
+					no_wait_reserve, no_wait_gpu, new_mem);
 	} else if (old_mem->mem_type == TTM_PL_SYSTEM &&
 		   new_mem->mem_type == TTM_PL_VRAM) {
 		r = radeon_move_ram_vram(bo, evict, interruptible,
-					    no_wait, new_mem);
+					    no_wait_reserve, no_wait_gpu, new_mem);
 	} else {
-		r = radeon_move_blit(bo, evict, no_wait, new_mem, old_mem);
+		r = radeon_move_blit(bo, evict, no_wait_reserve, no_wait_gpu, new_mem, old_mem);
 	}
 
 	if (r) {
 memcpy:
-		r = ttm_bo_move_memcpy(bo, evict, no_wait, new_mem);
+		r = ttm_bo_move_memcpy(bo, evict, no_wait_reserve, no_wait_gpu, new_mem);
 	}
 
 	return r;
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 03/13] drm/nouveau: update to TTM no_wait splitted argument

[Jerome Glisse]

This patch update radeon to the new no_wait splitted argument
TTM functionality.

Compile tested only (but thing should run as there is no
operating change from driver point of view)

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/nouveau/nouveau_bo.c  |   45 ++++++++++++++++++--------------
 drivers/gpu/drm/nouveau/nouveau_gem.c |    2 +-
 2 files changed, 26 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 0266124..5a167de 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -219,7 +219,7 @@ nouveau_bo_pin(struct nouveau_bo *nvbo, uint32_t memtype)
 	for (i = 0; i < nvbo->placement.num_placement; i++)
 		nvbo->placements[i] |= TTM_PL_FLAG_NO_EVICT;
 
-	ret = ttm_bo_validate(bo, &nvbo->placement, false, false);
+	ret = ttm_bo_validate(bo, &nvbo->placement, false, false, false);
 	if (ret == 0) {
 		switch (bo->mem.mem_type) {
 		case TTM_PL_VRAM:
@@ -256,7 +256,7 @@ nouveau_bo_unpin(struct nouveau_bo *nvbo)
 	for (i = 0; i < nvbo->placement.num_placement; i++)
 		nvbo->placements[i] &= ~TTM_PL_FLAG_NO_EVICT;
 
-	ret = ttm_bo_validate(bo, &nvbo->placement, false, false);
+	ret = ttm_bo_validate(bo, &nvbo->placement, false, false, false);
 	if (ret == 0) {
 		switch (bo->mem.mem_type) {
 		case TTM_PL_VRAM:
@@ -456,7 +456,8 @@ nouveau_bo_evict_flags(struct ttm_buffer_object *bo, struct ttm_placement *pl)
 
 static int
 nouveau_bo_move_accel_cleanup(struct nouveau_channel *chan,
-			      struct nouveau_bo *nvbo, bool evict, bool no_wait,
+			      struct nouveau_bo *nvbo, bool evict,
+			      bool no_wait_reserve, bool no_wait_gpu,
 			      struct ttm_mem_reg *new_mem)
 {
 	struct nouveau_fence *fence = NULL;
@@ -467,7 +468,7 @@ nouveau_bo_move_accel_cleanup(struct nouveau_channel *chan,
 		return ret;
 
 	ret = ttm_bo_move_accel_cleanup(&nvbo->bo, fence, NULL,
-					evict, no_wait, new_mem);
+					evict, no_wait_reserve, no_wait_gpu, new_mem);
 	if (nvbo->channel && nvbo->channel != chan)
 		ret = nouveau_fence_wait(fence, NULL, false, false);
 	nouveau_fence_unref((void *)&fence);
@@ -491,7 +492,8 @@ nouveau_bo_mem_ctxdma(struct nouveau_bo *nvbo, struct nouveau_channel *chan,
 
 static int
 nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
-		     int no_wait, struct ttm_mem_reg *new_mem)
+		     bool no_wait_reserve, bool no_wait_gpu,
+		     struct ttm_mem_reg *new_mem)
 {
 	struct nouveau_bo *nvbo = nouveau_bo(bo);
 	struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
@@ -569,12 +571,13 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
 		dst_offset += (PAGE_SIZE * line_count);
 	}
 
-	return nouveau_bo_move_accel_cleanup(chan, nvbo, evict, no_wait, new_mem);
+	return nouveau_bo_move_accel_cleanup(chan, nvbo, evict, no_wait_reserve, no_wait_gpu, new_mem);
 }
 
 static int
 nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
-		      bool no_wait, struct ttm_mem_reg *new_mem)
+		      bool no_wait_reserve, bool no_wait_gpu,
+		      struct ttm_mem_reg *new_mem)
 {
 	u32 placement_memtype = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING;
 	struct ttm_placement placement;
@@ -587,7 +590,7 @@ nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
 
 	tmp_mem = *new_mem;
 	tmp_mem.mm_node = NULL;
-	ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait);
+	ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait_reserve, no_wait_gpu);
 	if (ret)
 		return ret;
 
@@ -595,11 +598,11 @@ nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
 	if (ret)
 		goto out;
 
-	ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait, &tmp_mem);
+	ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_reserve, no_wait_gpu, &tmp_mem);
 	if (ret)
 		goto out;
 
-	ret = ttm_bo_move_ttm(bo, evict, no_wait, new_mem);
+	ret = ttm_bo_move_ttm(bo, evict, no_wait_reserve, no_wait_gpu, new_mem);
 out:
 	if (tmp_mem.mm_node) {
 		spin_lock(&bo->bdev->glob->lru_lock);
@@ -612,7 +615,8 @@ out:
 
 static int
 nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
-		      bool no_wait, struct ttm_mem_reg *new_mem)
+		      bool no_wait_reserve, bool no_wait_gpu,
+		      struct ttm_mem_reg *new_mem)
 {
 	u32 placement_memtype = TTM_PL_FLAG_TT | TTM_PL_MASK_CACHING;
 	struct ttm_placement placement;
@@ -625,15 +629,15 @@ nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
 
 	tmp_mem = *new_mem;
 	tmp_mem.mm_node = NULL;
-	ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait);
+	ret = ttm_bo_mem_space(bo, &placement, &tmp_mem, intr, no_wait_reserve, no_wait_gpu);
 	if (ret)
 		return ret;
 
-	ret = ttm_bo_move_ttm(bo, evict, no_wait, &tmp_mem);
+	ret = ttm_bo_move_ttm(bo, evict, no_wait_reserve, no_wait_gpu, &tmp_mem);
 	if (ret)
 		goto out;
 
-	ret = nouveau_bo_move_m2mf(bo, evict, intr, no_wait, new_mem);
+	ret = nouveau_bo_move_m2mf(bo, evict, intr, no_wait_reserve, no_wait_gpu, new_mem);
 	if (ret)
 		goto out;
 
@@ -700,7 +704,8 @@ nouveau_bo_vm_cleanup(struct ttm_buffer_object *bo,
 
 static int
 nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
-		bool no_wait, struct ttm_mem_reg *new_mem)
+		bool no_wait_reserve, bool no_wait_gpu,
+		struct ttm_mem_reg *new_mem)
 {
 	struct drm_nouveau_private *dev_priv = nouveau_bdev(bo->bdev);
 	struct nouveau_bo *nvbo = nouveau_bo(bo);
@@ -715,7 +720,7 @@ nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
 	/* Software copy if the card isn't up and running yet. */
 	if (dev_priv->init_state != NOUVEAU_CARD_INIT_DONE ||
 	    !dev_priv->channel) {
-		ret = ttm_bo_move_memcpy(bo, evict, no_wait, new_mem);
+		ret = ttm_bo_move_memcpy(bo, evict, no_wait_reserve, no_wait_gpu, new_mem);
 		goto out;
 	}
 
@@ -729,17 +734,17 @@ nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
 
 	/* Hardware assisted copy. */
 	if (new_mem->mem_type == TTM_PL_SYSTEM)
-		ret = nouveau_bo_move_flipd(bo, evict, intr, no_wait, new_mem);
+		ret = nouveau_bo_move_flipd(bo, evict, intr, no_wait_reserve, no_wait_gpu, new_mem);
 	else if (old_mem->mem_type == TTM_PL_SYSTEM)
-		ret = nouveau_bo_move_flips(bo, evict, intr, no_wait, new_mem);
+		ret = nouveau_bo_move_flips(bo, evict, intr, no_wait_reserve, no_wait_gpu, new_mem);
 	else
-		ret = nouveau_bo_move_m2mf(bo, evict, intr, no_wait, new_mem);
+		ret = nouveau_bo_move_m2mf(bo, evict, intr, no_wait_reserve, no_wait_gpu, new_mem);
 
 	if (!ret)
 		goto out;
 
 	/* Fallback to software copy. */
-	ret = ttm_bo_move_memcpy(bo, evict, no_wait, new_mem);
+	ret = ttm_bo_move_memcpy(bo, evict, no_wait_reserve, no_wait_gpu, new_mem);
 
 out:
 	if (ret)
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 0d22f66..1f50403 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -387,7 +387,7 @@ validate_list(struct nouveau_channel *chan, struct list_head *list,
 
 		nvbo->channel = chan;
 		ret = ttm_bo_validate(&nvbo->bo, &nvbo->placement,
-				      false, false);
+				      false, false, false);
 		nvbo->channel = NULL;
 		if (unlikely(ret)) {
 			NV_ERROR(dev, "fail ttm_validate\n");
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 04/13] drm/vmwgfx: update to TTM no_wait splitted argument

[Jerome Glisse]

This patch update radeon to the new no_wait splitted argument
TTM functionality.

Compile tested only (but thing should run as there is no
operating change from driver point of view)

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c |    4 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_fb.c      |    4 ++--
 drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c |    2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 0897359..dbd36b8 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -570,7 +570,7 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
 	 * Put BO in VRAM, only if there is space.
 	 */
 
-	ret = ttm_bo_validate(bo, &vmw_vram_sys_placement, true, false);
+	ret = ttm_bo_validate(bo, &vmw_vram_sys_placement, true, false, false);
 	if (unlikely(ret == -ERESTARTSYS))
 		return ret;
 
@@ -590,7 +590,7 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
 	 * previous contents.
 	 */
 
-	ret = ttm_bo_validate(bo, &vmw_vram_placement, true, false);
+	ret = ttm_bo_validate(bo, &vmw_vram_placement, true, false, false);
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
index a933670..80125ff 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
@@ -628,7 +628,7 @@ int vmw_dmabuf_from_vram(struct vmw_private *vmw_priv,
 	if (unlikely(ret != 0))
 		return ret;
 
-	ret = ttm_bo_validate(bo, &vmw_sys_placement, false, false);
+	ret = ttm_bo_validate(bo, &vmw_sys_placement, false, false, false);
 	ttm_bo_unreserve(bo);
 
 	return ret;
@@ -652,7 +652,7 @@ int vmw_dmabuf_to_start_of_vram(struct vmw_private *vmw_priv,
 	if (unlikely(ret != 0))
 		goto err_unlock;
 
-	ret = ttm_bo_validate(bo, &ne_placement, false, false);
+	ret = ttm_bo_validate(bo, &ne_placement, false, false, false);
 	ttm_bo_unreserve(bo);
 err_unlock:
 	ttm_write_unlock(&vmw_priv->active_master->lock);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
index 5b6eabe..ad566c8 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
@@ -118,7 +118,7 @@ static int vmw_dmabuf_pin_in_vram(struct vmw_private *dev_priv,
 	if (pin)
 		overlay_placement = &vmw_vram_ne_placement;
 
-	ret = ttm_bo_validate(bo, overlay_placement, interruptible, false);
+	ret = ttm_bo_validate(bo, overlay_placement, interruptible, false, false);
 
 	ttm_bo_unreserve(bo);
 
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 05/13] drm/ttm: ttm_fault callback to allow driver to handle bo placement V5

[Jerome Glisse]

On fault the driver is given the opportunity to perform any operation
it sees fit in order to place the buffer into a CPU visible area of
memory. This patch doesn't break TTM users, nouveau, vmwgfx and radeon
should keep working properly. Future patch will take advantage of this
infrastructure and remove the old path from TTM once driver are
converted.

V2 return VM_FAULT_NOPAGE if callback return -EBUSY or -ERESTARTSYS
V3 balance io_mem_reserve and io_mem_free call, fault_reserve_notify
   is responsible to perform any necessary task for mapping to succeed
V4 minor cleanup, atomic_t -> bool as member is protected by reserve
   mecanism from concurent access
V5 the callback is now responsible for iomapping the bo and providing
   a virtual address this simplify TTM and will allow to get rid of
   TTM_MEMTYPE_FLAG_NEEDS_IOREMAP

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c      |    7 ++-
 drivers/gpu/drm/ttm/ttm_bo_util.c |  115 +++++++++++++++++++------------------
 drivers/gpu/drm/ttm/ttm_bo_vm.c   |   41 +++++++-------
 include/drm/ttm/ttm_bo_api.h      |   23 +++++++
 include/drm/ttm/ttm_bo_driver.h   |   16 +++++-
 5 files changed, 122 insertions(+), 80 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 6f51b30..2171f86 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -632,6 +632,7 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
 
 	evict_mem = bo->mem;
 	evict_mem.mm_node = NULL;
+	evict_mem.bus.io_reserved = false;
 
 	placement.fpfn = 0;
 	placement.lpfn = 0;
@@ -1005,6 +1006,7 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 	mem.num_pages = bo->num_pages;
 	mem.size = mem.num_pages << PAGE_SHIFT;
 	mem.page_alignment = bo->mem.page_alignment;
+	mem.bus.io_reserved = false;
 	/*
 	 * Determine where to move the buffer.
 	 */
@@ -1160,6 +1162,7 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
 	bo->mem.num_pages = bo->num_pages;
 	bo->mem.mm_node = NULL;
 	bo->mem.page_alignment = page_alignment;
+	bo->mem.bus.io_reserved = false;
 	bo->buffer_start = buffer_start & PAGE_MASK;
 	bo->priv_flags = 0;
 	bo->mem.placement = (TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED);
@@ -1574,7 +1577,7 @@ int ttm_bo_pci_offset(struct ttm_bo_device *bdev,
 	if (ttm_mem_reg_is_pci(bdev, mem)) {
 		*bus_offset = mem->mm_node->start << PAGE_SHIFT;
 		*bus_size = mem->num_pages << PAGE_SHIFT;
-		*bus_base = man->io_offset;
+		*bus_base = man->io_offset + (uintptr_t)man->io_addr;
 	}
 
 	return 0;
@@ -1588,8 +1591,8 @@ void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
 
 	if (!bdev->dev_mapping)
 		return;
-
 	unmap_mapping_range(bdev->dev_mapping, offset, holelen, 1);
+	ttm_mem_io_free(bdev, &bo->mem);
 }
 EXPORT_SYMBOL(ttm_bo_unmap_virtual);
 
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 865b2a8..8493f39 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -81,30 +81,61 @@ int ttm_bo_move_ttm(struct ttm_buffer_object *bo,
 }
 EXPORT_SYMBOL(ttm_bo_move_ttm);
 
+int ttm_mem_io_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
+	int ret;
+
+	if (bdev->driver->io_mem_reserve) {
+		if (!mem->bus.io_reserved) {
+			mem->bus.io_reserved = true;
+			ret = bdev->driver->io_mem_reserve(bdev, mem);
+			if (unlikely(ret != 0))
+				return ret;
+		}
+	} else {
+		ret = ttm_bo_pci_offset(bdev, mem, &mem->bus.base, &mem->bus.offset, &mem->bus.size);
+		if (unlikely(ret != 0))
+			return ret;
+		mem->bus.addr = man->io_addr;
+		mem->bus.is_iomem = (mem->bus.size > 0) ? 1 : 0;
+	}
+	return 0;
+}
+
+void ttm_mem_io_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+	if (bdev->driver->io_mem_reserve) {
+		if (mem->bus.io_reserved) {
+			mem->bus.io_reserved = false;
+			bdev->driver->io_mem_free(bdev, mem);
+		}
+	}
+}
+
 int ttm_mem_reg_ioremap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem,
 			void **virtual)
 {
 	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
-	unsigned long bus_offset;
-	unsigned long bus_size;
-	unsigned long bus_base;
 	int ret;
 	void *addr;
 
 	*virtual = NULL;
-	ret = ttm_bo_pci_offset(bdev, mem, &bus_base, &bus_offset, &bus_size);
-	if (ret || bus_size == 0)
+	ret = ttm_mem_io_reserve(bdev, mem);
+	if (ret)
 		return ret;
 
-	if (!(man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP))
-		addr = (void *)(((u8 *) man->io_addr) + bus_offset);
-	else {
+	if (!(man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP)) {
+		addr = mem->bus.addr;
+	} else {
 		if (mem->placement & TTM_PL_FLAG_WC)
-			addr = ioremap_wc(bus_base + bus_offset, bus_size);
+			addr = ioremap_wc(mem->bus.base + mem->bus.offset, mem->bus.size);
 		else
-			addr = ioremap_nocache(bus_base + bus_offset, bus_size);
-		if (!addr)
+			addr = ioremap_nocache(mem->bus.base + mem->bus.offset, mem->bus.size);
+		if (!addr) {
+			ttm_mem_io_free(bdev, mem);
 			return -ENOMEM;
+		}
 	}
 	*virtual = addr;
 	return 0;
@@ -119,6 +150,7 @@ void ttm_mem_reg_iounmap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem,
 
 	if (virtual && (man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP))
 		iounmap(virtual);
+	ttm_mem_io_free(bdev, mem);
 }
 
 static int ttm_copy_io_page(void *dst, void *src, unsigned long page)
@@ -370,9 +402,8 @@ pgprot_t ttm_io_prot(uint32_t caching_flags, pgprot_t tmp)
 EXPORT_SYMBOL(ttm_io_prot);
 
 static int ttm_bo_ioremap(struct ttm_buffer_object *bo,
-			  unsigned long bus_base,
-			  unsigned long bus_offset,
-			  unsigned long bus_size,
+			  unsigned long offset,
+			  unsigned long size,
 			  struct ttm_bo_kmap_obj *map)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
@@ -381,15 +412,15 @@ static int ttm_bo_ioremap(struct ttm_buffer_object *bo,
 
 	if (!(man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP)) {
 		map->bo_kmap_type = ttm_bo_map_premapped;
-		map->virtual = (void *)(((u8 *) man->io_addr) + bus_offset);
+		map->virtual = (void *)(((u8 *) bo->mem.bus.addr) + offset);
 	} else {
 		map->bo_kmap_type = ttm_bo_map_iomap;
 		if (mem->placement & TTM_PL_FLAG_WC)
-			map->virtual = ioremap_wc(bus_base + bus_offset,
-						  bus_size);
+			map->virtual = ioremap_wc(bo->mem.bus.base + bo->mem.bus.offset + offset,
+						  size);
 		else
-			map->virtual = ioremap_nocache(bus_base + bus_offset,
-						       bus_size);
+			map->virtual = ioremap_nocache(bo->mem.bus.base + bo->mem.bus.offset + offset,
+						       size);
 	}
 	return (!map->virtual) ? -ENOMEM : 0;
 }
@@ -442,13 +473,12 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo,
 		unsigned long start_page, unsigned long num_pages,
 		struct ttm_bo_kmap_obj *map)
 {
+	unsigned long offset, size;
 	int ret;
-	unsigned long bus_base;
-	unsigned long bus_offset;
-	unsigned long bus_size;
 
 	BUG_ON(!list_empty(&bo->swap));
 	map->virtual = NULL;
+	map->bo = bo;
 	if (num_pages > bo->num_pages)
 		return -EINVAL;
 	if (start_page > bo->num_pages)
@@ -457,16 +487,15 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo,
 	if (num_pages > 1 && !DRM_SUSER(DRM_CURPROC))
 		return -EPERM;
 #endif
-	ret = ttm_bo_pci_offset(bo->bdev, &bo->mem, &bus_base,
-				&bus_offset, &bus_size);
+	ret = ttm_mem_io_reserve(bo->bdev, &bo->mem);
 	if (ret)
 		return ret;
-	if (bus_size == 0) {
+	if (!bo->mem.bus.is_iomem) {
 		return ttm_bo_kmap_ttm(bo, start_page, num_pages, map);
 	} else {
-		bus_offset += start_page << PAGE_SHIFT;
-		bus_size = num_pages << PAGE_SHIFT;
-		return ttm_bo_ioremap(bo, bus_base, bus_offset, bus_size, map);
+		offset = start_page << PAGE_SHIFT;
+		size = num_pages << PAGE_SHIFT;
+		return ttm_bo_ioremap(bo, offset, size, map);
 	}
 }
 EXPORT_SYMBOL(ttm_bo_kmap);
@@ -478,6 +507,7 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map)
 	switch (map->bo_kmap_type) {
 	case ttm_bo_map_iomap:
 		iounmap(map->virtual);
+		ttm_mem_io_free(map->bo->bdev, &map->bo->mem);
 		break;
 	case ttm_bo_map_vmap:
 		vunmap(map->virtual);
@@ -495,35 +525,6 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map)
 }
 EXPORT_SYMBOL(ttm_bo_kunmap);
 
-int ttm_bo_pfn_prot(struct ttm_buffer_object *bo,
-		    unsigned long dst_offset,
-		    unsigned long *pfn, pgprot_t *prot)
-{
-	struct ttm_mem_reg *mem = &bo->mem;
-	struct ttm_bo_device *bdev = bo->bdev;
-	unsigned long bus_offset;
-	unsigned long bus_size;
-	unsigned long bus_base;
-	int ret;
-	ret = ttm_bo_pci_offset(bdev, mem, &bus_base, &bus_offset,
-			&bus_size);
-	if (ret)
-		return -EINVAL;
-	if (bus_size != 0)
-		*pfn = (bus_base + bus_offset + dst_offset) >> PAGE_SHIFT;
-	else
-		if (!bo->ttm)
-			return -EINVAL;
-		else
-			*pfn = page_to_pfn(ttm_tt_get_page(bo->ttm,
-							   dst_offset >>
-							   PAGE_SHIFT));
-	*prot = (mem->placement & TTM_PL_FLAG_CACHED) ?
-		PAGE_KERNEL : ttm_io_prot(mem->placement, PAGE_KERNEL);
-
-	return 0;
-}
-
 int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 			      void *sync_obj,
 			      void *sync_obj_arg,
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index 668dbe8..fe6cb77 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -74,9 +74,6 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	struct ttm_buffer_object *bo = (struct ttm_buffer_object *)
 	    vma->vm_private_data;
 	struct ttm_bo_device *bdev = bo->bdev;
-	unsigned long bus_base;
-	unsigned long bus_offset;
-	unsigned long bus_size;
 	unsigned long page_offset;
 	unsigned long page_last;
 	unsigned long pfn;
@@ -84,7 +81,6 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	struct page *page;
 	int ret;
 	int i;
-	bool is_iomem;
 	unsigned long address = (unsigned long)vmf->virtual_address;
 	int retval = VM_FAULT_NOPAGE;
 
@@ -101,8 +97,21 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 		return VM_FAULT_NOPAGE;
 	}
 
-	if (bdev->driver->fault_reserve_notify)
-		bdev->driver->fault_reserve_notify(bo);
+	if (bdev->driver->fault_reserve_notify) {
+		ret = bdev->driver->fault_reserve_notify(bo);
+		switch (ret) {
+		case 0:
+			break;
+		case -EBUSY:
+			set_need_resched();
+		case -ERESTARTSYS:
+			retval = VM_FAULT_NOPAGE;
+			goto out_unlock;
+		default:
+			retval = VM_FAULT_SIGBUS;
+			goto out_unlock;
+		}
+	}
 
 	/*
 	 * Wait for buffer data in transit, due to a pipelined
@@ -122,15 +131,12 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 		spin_unlock(&bo->lock);
 
 
-	ret = ttm_bo_pci_offset(bdev, &bo->mem, &bus_base, &bus_offset,
-				&bus_size);
-	if (unlikely(ret != 0)) {
+	ret = ttm_mem_io_reserve(bdev, &bo->mem);
+	if (ret) {
 		retval = VM_FAULT_SIGBUS;
 		goto out_unlock;
 	}
 
-	is_iomem = (bus_size != 0);
-
 	page_offset = ((address - vma->vm_start) >> PAGE_SHIFT) +
 	    bo->vm_node->start - vma->vm_pgoff;
 	page_last = ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) +
@@ -154,8 +160,7 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	 * vma->vm_page_prot when the object changes caching policy, with
 	 * the correct locks held.
 	 */
-
-	if (is_iomem) {
+	if (bo->mem.bus.is_iomem) {
 		vma->vm_page_prot = ttm_io_prot(bo->mem.placement,
 						vma->vm_page_prot);
 	} else {
@@ -171,10 +176,8 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	 */
 
 	for (i = 0; i < TTM_BO_VM_NUM_PREFAULT; ++i) {
-
-		if (is_iomem)
-			pfn = ((bus_base + bus_offset) >> PAGE_SHIFT) +
-			    page_offset;
+		if (bo->mem.bus.is_iomem)
+			pfn = ((bo->mem.bus.base + bo->mem.bus.offset) >> PAGE_SHIFT) + page_offset;
 		else {
 			page = ttm_tt_get_page(ttm, page_offset);
 			if (unlikely(!page && i == 0)) {
@@ -198,7 +201,6 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 			retval =
 			    (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
 			goto out_unlock;
-
 		}
 
 		address += PAGE_SIZE;
@@ -221,8 +223,7 @@ static void ttm_bo_vm_open(struct vm_area_struct *vma)
 
 static void ttm_bo_vm_close(struct vm_area_struct *vma)
 {
-	struct ttm_buffer_object *bo =
-	    (struct ttm_buffer_object *)vma->vm_private_data;
+	struct ttm_buffer_object *bo = (struct ttm_buffer_object *)vma->vm_private_data;
 
 	ttm_bo_unref(&bo);
 	vma->vm_private_data = NULL;
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index 8c8005e..3e273e0 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -66,6 +66,26 @@ struct ttm_placement {
 	const uint32_t	*busy_placement;
 };
 
+/**
+ * struct ttm_bus_placement
+ *
+ * @addr:		mapped virtual address
+ * @base:		bus base address
+ * @is_iomem:		is this io memory ?
+ * @size:		size in byte
+ * @offset:		offset from the base address
+ *
+ * Structure indicating the bus placement of an object.
+ */
+struct ttm_bus_placement {
+	void		*addr;
+	unsigned long	base;
+	unsigned long	size;
+	unsigned long	offset;
+	bool		is_iomem;
+	bool		io_reserved;
+};
+
 
 /**
  * struct ttm_mem_reg
@@ -75,6 +95,7 @@ struct ttm_placement {
  * @num_pages: Actual size of memory region in pages.
  * @page_alignment: Page alignment.
  * @placement: Placement flags.
+ * @bus: Placement on io bus accessible to the CPU
  *
  * Structure indicating the placement and space resources used by a
  * buffer object.
@@ -87,6 +108,7 @@ struct ttm_mem_reg {
 	uint32_t page_alignment;
 	uint32_t mem_type;
 	uint32_t placement;
+	struct ttm_bus_placement bus;
 };
 
 /**
@@ -274,6 +296,7 @@ struct ttm_bo_kmap_obj {
 		ttm_bo_map_kmap         = 3,
 		ttm_bo_map_premapped    = 4 | TTM_BO_MAP_IOMEM_MASK,
 	} bo_kmap_type;
+	struct ttm_buffer_object *bo;
 };
 
 /**
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index 69f70e4..da39865 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -352,12 +352,21 @@ struct ttm_bo_driver {
 			    struct ttm_mem_reg *new_mem);
 	/* notify the driver we are taking a fault on this BO
 	 * and have reserved it */
-	void (*fault_reserve_notify)(struct ttm_buffer_object *bo);
+	int (*fault_reserve_notify)(struct ttm_buffer_object *bo);
 
 	/**
 	 * notify the driver that we're about to swap out this bo
 	 */
 	void (*swap_notify) (struct ttm_buffer_object *bo);
+
+	/**
+	 * Driver callback on when mapping io memory (for bo_move_memcpy
+	 * for instance). TTM will take care to call io_mem_free whenever
+	 * the mapping is not use anymore. io_mem_reserve & io_mem_free
+	 * are balanced.
+	 */
+	int (*io_mem_reserve)(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem);
+	void (*io_mem_free)(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem);
 };
 
 /**
@@ -685,6 +694,11 @@ extern int ttm_bo_pci_offset(struct ttm_bo_device *bdev,
 			     unsigned long *bus_offset,
 			     unsigned long *bus_size);
 
+extern int ttm_mem_io_reserve(struct ttm_bo_device *bdev,
+				struct ttm_mem_reg *mem);
+extern void ttm_mem_io_free(struct ttm_bo_device *bdev,
+				struct ttm_mem_reg *mem);
+
 extern void ttm_bo_global_release(struct ttm_global_reference *ref);
 extern int ttm_bo_global_init(struct ttm_global_reference *ref);
 
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 06/13] drm/radeon/kms: add support for new fault callback V6

[Jerome Glisse]

This add the support for the new fault callback and also the
infrastructure for supporting unmappable VRAM.

V2 validate BO with no_wait = true
V3 don't derefence bo->mem.mm_node as it's not NULL only for
   VRAM or GTT
V4 update to splitted no_wait ttm change
V5 update to new balanced io_mem_reserve/free change
V6 callback is responsible for iomapping memory

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/radeon/radeon_object.c |   26 +++++++++++++-
 drivers/gpu/drm/radeon/radeon_object.h |    2 +-
 drivers/gpu/drm/radeon/radeon_ttm.c    |   59 +++++++++++++++++++++++++++++--
 3 files changed, 80 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index 3adfa88..aef44f3 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -496,11 +496,33 @@ void radeon_bo_move_notify(struct ttm_buffer_object *bo,
 	radeon_bo_check_tiling(rbo, 0, 1);
 }
 
-void radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
+int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
 {
+	struct radeon_device *rdev;
 	struct radeon_bo *rbo;
+	unsigned long offset, size;
+	int r;
+
 	if (!radeon_ttm_bo_is_radeon_bo(bo))
-		return;
+		return 0;
 	rbo = container_of(bo, struct radeon_bo, tbo);
 	radeon_bo_check_tiling(rbo, 0, 0);
+	rdev = rbo->rdev;
+	if (bo->mem.mem_type == TTM_PL_VRAM) {
+		size = bo->mem.num_pages << PAGE_SHIFT;
+		offset = bo->mem.mm_node->start << PAGE_SHIFT;
+		if ((offset + size) > rdev->mc.visible_vram_size) {
+			/* hurrah the memory is not visible ! */
+			radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_VRAM);
+			rbo->placement.lpfn = rdev->mc.visible_vram_size >> PAGE_SHIFT;
+			r = ttm_bo_validate(bo, &rbo->placement, false, true, false);
+			if (unlikely(r != 0))
+				return r;
+			offset = bo->mem.mm_node->start << PAGE_SHIFT;
+			/* this should not happen */
+			if ((offset + size) > rdev->mc.visible_vram_size)
+				return -EINVAL;
+		}
+	}
+	return 0;
 }
diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
index 7ab43de..353998d 100644
--- a/drivers/gpu/drm/radeon/radeon_object.h
+++ b/drivers/gpu/drm/radeon/radeon_object.h
@@ -168,6 +168,6 @@ extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
 				bool force_drop);
 extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
 					struct ttm_mem_reg *mem);
-extern void radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
+extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
 extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
 #endif
diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index ba4724c..1fc6cad 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -165,8 +165,7 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 			man->io_size = rdev->mc.gtt_size;
 			man->io_addr = NULL;
 			if (!rdev->ddev->agp->cant_use_aperture)
-				man->flags = TTM_MEMTYPE_FLAG_NEEDS_IOREMAP |
-					     TTM_MEMTYPE_FLAG_MAPPABLE;
+				man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
 			man->available_caching = TTM_PL_FLAG_UNCACHED |
 						 TTM_PL_FLAG_WC;
 			man->default_caching = TTM_PL_FLAG_WC;
@@ -182,7 +181,6 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		/* "On-card" video ram */
 		man->gpu_offset = rdev->mc.vram_start;
 		man->flags = TTM_MEMTYPE_FLAG_FIXED |
-			     TTM_MEMTYPE_FLAG_NEEDS_IOREMAP |
 			     TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC;
 		man->default_caching = TTM_PL_FLAG_WC;
@@ -437,10 +435,61 @@ static int radeon_bo_move(struct ttm_buffer_object *bo,
 memcpy:
 		r = ttm_bo_move_memcpy(bo, evict, no_wait_reserve, no_wait_gpu, new_mem);
 	}
-
 	return r;
 }
 
+static int radeon_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
+	struct radeon_device *rdev = radeon_get_rdev(bdev);
+
+	mem->bus.addr = NULL;
+	mem->bus.offset = 0;
+	mem->bus.size = mem->num_pages << PAGE_SHIFT;
+	mem->bus.base = 0;
+	mem->bus.is_iomem = false;
+	if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
+		return -EINVAL;
+	switch (mem->mem_type) {
+	case TTM_PL_SYSTEM:
+		/* system memory */
+		return 0;
+	case TTM_PL_TT:
+#if __OS_HAS_AGP
+		if (rdev->flags & RADEON_IS_AGP) {
+			/* RADEON_IS_AGP is set only if AGP is active */
+			mem->bus.offset = mem->mm_node->start << PAGE_SHIFT;
+			mem->bus.base = rdev->mc.agp_base;
+			mem->bus.is_iomem = true;
+			mem->bus.addr = ioremap_wc(mem->bus.base + mem->bus.offset, mem->bus.size);
+			if (mem->bus.addr == NULL)
+				return -ENOMEM;
+		}
+#endif
+		break;
+	case TTM_PL_VRAM:
+		mem->bus.offset = mem->mm_node->start << PAGE_SHIFT;
+		/* check if it's visible */
+		if ((mem->bus.offset + mem->bus.size) > rdev->mc.visible_vram_size)
+			return -EINVAL;
+		mem->bus.base = rdev->mc.aper_base;
+		mem->bus.is_iomem = true;
+		mem->bus.addr = ioremap_wc(mem->bus.base + mem->bus.offset, mem->bus.size);
+		if (mem->bus.addr == NULL)
+			return -ENOMEM;
+		break;
+	default:
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static void radeon_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+	if (mem->bus.addr)
+		iounmap(mem->bus.addr);
+}
+
 static int radeon_sync_obj_wait(void *sync_obj, void *sync_arg,
 				bool lazy, bool interruptible)
 {
@@ -481,6 +530,8 @@ static struct ttm_bo_driver radeon_bo_driver = {
 	.sync_obj_ref = &radeon_sync_obj_ref,
 	.move_notify = &radeon_bo_move_notify,
 	.fault_reserve_notify = &radeon_bo_fault_reserve_notify,
+	.io_mem_reserve = &radeon_ttm_io_mem_reserve,
+	.io_mem_free = &radeon_ttm_io_mem_free,
 };
 
 int radeon_ttm_init(struct radeon_device *rdev)
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 07/13] drm/nouveau/kms: add support for new TTM fault callback V4

[Jerome Glisse]

This add the support for the new fault callback, does change anything
from driver point of view, thought it should allow nouveau to add
support for unmappable VRAM.

Improvement: store the aperture base in a variable so that we don't
call a function to get it on each fault.

Patch hasn't been tested on any hw.

V2 don't derefence bo->mem.mm_node as it's not NULL only for
   VRAM or GTT
V3 update after io_mem_reserve/io_mem_free callback balancing
V4 callback has to ioremap

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/nouveau/nouveau_bo.c |   66 +++++++++++++++++++++++++++++++--
 1 files changed, 62 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 5a167de..efc124d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -386,8 +386,7 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		break;
 	case TTM_PL_VRAM:
 		man->flags = TTM_MEMTYPE_FLAG_FIXED |
-			     TTM_MEMTYPE_FLAG_MAPPABLE |
-			     TTM_MEMTYPE_FLAG_NEEDS_IOREMAP;
+			     TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->available_caching = TTM_PL_FLAG_UNCACHED |
 					 TTM_PL_FLAG_WC;
 		man->default_caching = TTM_PL_FLAG_WC;
@@ -403,8 +402,7 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 	case TTM_PL_TT:
 		switch (dev_priv->gart_info.type) {
 		case NOUVEAU_GART_AGP:
-			man->flags = TTM_MEMTYPE_FLAG_MAPPABLE |
-				     TTM_MEMTYPE_FLAG_NEEDS_IOREMAP;
+			man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
 			man->available_caching = TTM_PL_FLAG_UNCACHED;
 			man->default_caching = TTM_PL_FLAG_UNCACHED;
 			break;
@@ -761,6 +759,63 @@ nouveau_bo_verify_access(struct ttm_buffer_object *bo, struct file *filp)
 	return 0;
 }
 
+static int
+nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
+	struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev);
+	struct drm_device *dev = dev_priv->dev;
+
+	mem->bus.addr = NULL;
+	mem->bus.offset = 0;
+	mem->bus.size = mem->num_pages << PAGE_SHIFT;
+	mem->bus.base = 0;
+	mem->bus.is_iomem = false;
+	if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
+		return -EINVAL;
+	switch (mem->mem_type) {
+	case TTM_PL_SYSTEM:
+		/* System memory */
+		return 0;
+	case TTM_PL_TT:
+#if __OS_HAS_AGP
+		if (dev_priv->gart_info.type == NOUVEAU_GART_AGP) {
+			mem->bus.offset = mem->mm_node->start << PAGE_SHIFT;
+			mem->bus.base = dev_priv->gart_info.aper_base;
+			mem->bus.is_iomem = true;
+			mem->bus.addr = ioremap_wc(mem->bus.base + mem->bus.offset, mem->bus.size);
+			if (mem->bus.addr == NULL)
+				return -ENOMEM;
+		}
+#endif
+		break;
+	case TTM_PL_VRAM:
+		mem->bus.offset = mem->mm_node->start << PAGE_SHIFT;
+		mem->bus.base = drm_get_resource_start(dev, 1);
+		mem->bus.is_iomem = true;
+		mem->bus.addr = ioremap_wc(mem->bus.base + mem->bus.offset, mem->bus.size);
+		if (mem->bus.addr == NULL)
+			return -ENOMEM;
+		break;
+	default:
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static void
+nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+	if (mem->bus.addr)
+		iounmap(mem->bus.addr);
+}
+
+static int
+nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
+{
+	return 0;
+}
+
 struct ttm_bo_driver nouveau_bo_driver = {
 	.create_ttm_backend_entry = nouveau_bo_create_ttm_backend_entry,
 	.invalidate_caches = nouveau_bo_invalidate_caches,
@@ -773,5 +828,8 @@ struct ttm_bo_driver nouveau_bo_driver = {
 	.sync_obj_flush = nouveau_fence_flush,
 	.sync_obj_unref = nouveau_fence_unref,
 	.sync_obj_ref = nouveau_fence_ref,
+	.fault_reserve_notify = &nouveau_ttm_fault_reserve_notify,
+	.io_mem_reserve = &nouveau_ttm_io_mem_reserve,
+	.io_mem_free = &nouveau_ttm_io_mem_free,
 };
 
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 08/13] drm/vmwgfx: add support for new TTM fault callback V4

[Jerome Glisse]

This add the support for the new fault callback, does change anything
from driver point of view.

Improvement: store the aperture base in a variable so that we don't
call a function to get it on each fault.

Patch hasn't been tested.

V2 don't derefence bo->mem.mm_node as it's not NULL only for
   VRAM or GTT
V3 update after io_mem_reserve/io_mem_free callback balancing
V4 callback has to ioremap

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c |   49 ++++++++++++++++++++++++++++++--
 1 files changed, 46 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index 825ebe3..9d5a08d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -153,8 +153,7 @@ int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		man->gpu_offset = 0;
 		man->io_offset = dev_priv->vram_start;
 		man->io_size = dev_priv->vram_size;
-		man->flags = TTM_MEMTYPE_FLAG_FIXED |
-		    TTM_MEMTYPE_FLAG_NEEDS_IOREMAP | TTM_MEMTYPE_FLAG_MAPPABLE;
+		man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->io_addr = NULL;
 		man->available_caching = TTM_PL_MASK_CACHING;
 		man->default_caching = TTM_PL_FLAG_WC;
@@ -193,6 +192,47 @@ static void vmw_swap_notify(struct ttm_buffer_object *bo)
 	vmw_dmabuf_gmr_unbind(bo);
 }
 
+static int vmw_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
+	struct vmw_private *dev_priv = container_of(bdev, struct vmw_private, bdev);
+
+	mem->bus.addr = NULL;
+	mem->bus.is_iomem = false;
+	mem->bus.offset = 0;
+	mem->bus.size = mem->num_pages << PAGE_SHIFT;
+	mem->bus.base = 0;
+	if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
+		return -EINVAL;
+	switch (mem->mem_type) {
+	case TTM_PL_SYSTEM:
+		/* System memory */
+		return 0;
+	case TTM_PL_VRAM:
+		mem->bus.offset = mem->mm_node->start << PAGE_SHIFT;
+		mem->bus.base = dev_priv->vram_start;
+		mem->bus.is_iomem = true;
+		mem->bus.addr = ioremap_wc(mem->bus.base + mem->bus.offset, mem->bus.size);
+		if (mem->bus.addr == NULL)
+			return -ENOMEM;
+		break;
+	default:
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static void vmw_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+{
+	if (mem->bus.addr)
+		iounmap(mem->bus.addr);
+}
+
+static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
+{
+	return 0;
+}
+
 /**
  * FIXME: We're using the old vmware polling method to sync.
  * Do this with fences instead.
@@ -248,5 +288,8 @@ struct ttm_bo_driver vmw_bo_driver = {
 	.sync_obj_unref = vmw_sync_obj_unref,
 	.sync_obj_ref = vmw_sync_obj_ref,
 	.move_notify = vmw_move_notify,
-	.swap_notify = vmw_swap_notify
+	.swap_notify = vmw_swap_notify,
+	.fault_reserve_notify = &vmw_ttm_fault_reserve_notify,
+	.io_mem_reserve = &vmw_ttm_io_mem_reserve,
+	.io_mem_free = &vmw_ttm_io_mem_free,
 };
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 09/13] drm/radeon/kms: don't initialize TTM io memory manager field

[Jerome Glisse]

This isn't needed anymore with the new TTM fault callback

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/radeon/radeon_ttm.c |   13 +------------
 1 files changed, 1 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
index 1fc6cad..14da488 100644
--- a/drivers/gpu/drm/radeon/radeon_ttm.c
+++ b/drivers/gpu/drm/radeon/radeon_ttm.c
@@ -161,21 +161,13 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 					  (unsigned)type);
 				return -EINVAL;
 			}
-			man->io_offset = rdev->mc.agp_base;
-			man->io_size = rdev->mc.gtt_size;
-			man->io_addr = NULL;
 			if (!rdev->ddev->agp->cant_use_aperture)
 				man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
 			man->available_caching = TTM_PL_FLAG_UNCACHED |
 						 TTM_PL_FLAG_WC;
 			man->default_caching = TTM_PL_FLAG_WC;
-		} else
-#endif
-		{
-			man->io_offset = 0;
-			man->io_size = 0;
-			man->io_addr = NULL;
 		}
+#endif
 		break;
 	case TTM_PL_VRAM:
 		/* "On-card" video ram */
@@ -184,9 +176,6 @@ static int radeon_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 			     TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->available_caching = TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_WC;
 		man->default_caching = TTM_PL_FLAG_WC;
-		man->io_addr = NULL;
-		man->io_offset = rdev->mc.aper_base;
-		man->io_size = rdev->mc.aper_size;
 		break;
 	default:
 		DRM_ERROR("Unsupported memory type %u\n", (unsigned)type);
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 10/13] drm/nouveau/kms: don't initialize TTM io memory manager field

[Jerome Glisse]

This isn't needed anymore with the new TTM fault callback

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/nouveau/nouveau_bo.c |   11 -----------
 1 files changed, 0 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index efc124d..5ee48ec 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -390,13 +390,6 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		man->available_caching = TTM_PL_FLAG_UNCACHED |
 					 TTM_PL_FLAG_WC;
 		man->default_caching = TTM_PL_FLAG_WC;
-
-		man->io_addr = NULL;
-		man->io_offset = drm_get_resource_start(dev, 1);
-		man->io_size = drm_get_resource_len(dev, 1);
-		if (man->io_size > nouveau_mem_fb_amount(dev))
-			man->io_size = nouveau_mem_fb_amount(dev);
-
 		man->gpu_offset = dev_priv->vm_vram_base;
 		break;
 	case TTM_PL_TT:
@@ -417,10 +410,6 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 				 dev_priv->gart_info.type);
 			return -EINVAL;
 		}
-
-		man->io_offset  = dev_priv->gart_info.aper_base;
-		man->io_size    = dev_priv->gart_info.aper_size;
-		man->io_addr   = NULL;
 		man->gpu_offset = dev_priv->vm_gart_base;
 		break;
 	default:
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 11/13] drm/vmwgfx: don't initialize TTM io memory manager field

[Jerome Glisse]

This isn't needed anymore with the new TTM fault callback

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c |    6 ------
 1 files changed, 0 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index 9d5a08d..25d332a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -137,9 +137,6 @@ int vmw_invalidate_caches(struct ttm_bo_device *bdev, uint32_t flags)
 int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		      struct ttm_mem_type_manager *man)
 {
-	struct vmw_private *dev_priv =
-	    container_of(bdev, struct vmw_private, bdev);
-
 	switch (type) {
 	case TTM_PL_SYSTEM:
 		/* System memory */
@@ -151,10 +148,7 @@ int vmw_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 	case TTM_PL_VRAM:
 		/* "On-card" video ram */
 		man->gpu_offset = 0;
-		man->io_offset = dev_priv->vram_start;
-		man->io_size = dev_priv->vram_size;
 		man->flags = TTM_MEMTYPE_FLAG_FIXED | TTM_MEMTYPE_FLAG_MAPPABLE;
-		man->io_addr = NULL;
 		man->available_caching = TTM_PL_MASK_CACHING;
 		man->default_caching = TTM_PL_FLAG_WC;
 		break;
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 12/13] drm/ttm: remove io_ field from TTM V5

[Jerome Glisse]

All TTM driver have been converted to new io_mem_reserve/free
interface which allow driver to choose and return proper io
base, offset to core TTM for ioremapping if necessary. This
patch remove what is now deadcode.

V2 adapt to match with change in first patch of the patchset
V3 update after io_mem_reserve/io_mem_free callback balancing
V4 adjust to minor cleanup
V5 remove the needs ioremap flag

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c      |   22 ----------
 drivers/gpu/drm/ttm/ttm_bo_util.c |   80 ++++++-------------------------------
 include/drm/ttm/ttm_bo_api.h      |    4 +-
 include/drm/ttm/ttm_bo_driver.h   |   12 ------
 4 files changed, 13 insertions(+), 105 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 2171f86..1f27cf2 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -79,8 +79,6 @@ static void ttm_mem_type_debug(struct ttm_bo_device *bdev, int mem_type)
 	printk(KERN_ERR TTM_PFX "    use_type: %d\n", man->use_type);
 	printk(KERN_ERR TTM_PFX "    flags: 0x%08X\n", man->flags);
 	printk(KERN_ERR TTM_PFX "    gpu_offset: 0x%08lX\n", man->gpu_offset);
-	printk(KERN_ERR TTM_PFX "    io_offset: 0x%08lX\n", man->io_offset);
-	printk(KERN_ERR TTM_PFX "    io_size: %ld\n", man->io_size);
 	printk(KERN_ERR TTM_PFX "    size: %llu\n", man->size);
 	printk(KERN_ERR TTM_PFX "    available_caching: 0x%08X\n",
 		man->available_caching);
@@ -1563,26 +1561,6 @@ bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 	return true;
 }
 
-int ttm_bo_pci_offset(struct ttm_bo_device *bdev,
-		      struct ttm_mem_reg *mem,
-		      unsigned long *bus_base,
-		      unsigned long *bus_offset, unsigned long *bus_size)
-{
-	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
-
-	*bus_size = 0;
-	if (!(man->flags & TTM_MEMTYPE_FLAG_MAPPABLE))
-		return -EINVAL;
-
-	if (ttm_mem_reg_is_pci(bdev, mem)) {
-		*bus_offset = mem->mm_node->start << PAGE_SHIFT;
-		*bus_size = mem->num_pages << PAGE_SHIFT;
-		*bus_base = man->io_offset + (uintptr_t)man->io_addr;
-	}
-
-	return 0;
-}
-
 void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 8493f39..15d1bf3 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -83,61 +83,33 @@ EXPORT_SYMBOL(ttm_bo_move_ttm);
 
 int ttm_mem_io_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 {
-	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
-	int ret;
+	int ret = 0;
 
-	if (bdev->driver->io_mem_reserve) {
-		if (!mem->bus.io_reserved) {
-			mem->bus.io_reserved = true;
-			ret = bdev->driver->io_mem_reserve(bdev, mem);
-			if (unlikely(ret != 0))
-				return ret;
-		}
-	} else {
-		ret = ttm_bo_pci_offset(bdev, mem, &mem->bus.base, &mem->bus.offset, &mem->bus.size);
-		if (unlikely(ret != 0))
-			return ret;
-		mem->bus.addr = man->io_addr;
-		mem->bus.is_iomem = (mem->bus.size > 0) ? 1 : 0;
+	if (!mem->bus.io_reserved) {
+		mem->bus.io_reserved = true;
+		ret = bdev->driver->io_mem_reserve(bdev, mem);
 	}
-	return 0;
+	return ret;
 }
 
 void ttm_mem_io_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 {
-	if (bdev->driver->io_mem_reserve) {
-		if (mem->bus.io_reserved) {
-			mem->bus.io_reserved = false;
-			bdev->driver->io_mem_free(bdev, mem);
-		}
+	if (mem->bus.io_reserved) {
+		mem->bus.io_reserved = false;
+		bdev->driver->io_mem_free(bdev, mem);
 	}
 }
 
 int ttm_mem_reg_ioremap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem,
 			void **virtual)
 {
-	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
 	int ret;
-	void *addr;
 
 	*virtual = NULL;
 	ret = ttm_mem_io_reserve(bdev, mem);
 	if (ret)
 		return ret;
-
-	if (!(man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP)) {
-		addr = mem->bus.addr;
-	} else {
-		if (mem->placement & TTM_PL_FLAG_WC)
-			addr = ioremap_wc(mem->bus.base + mem->bus.offset, mem->bus.size);
-		else
-			addr = ioremap_nocache(mem->bus.base + mem->bus.offset, mem->bus.size);
-		if (!addr) {
-			ttm_mem_io_free(bdev, mem);
-			return -ENOMEM;
-		}
-	}
-	*virtual = addr;
+	*virtual = mem->bus.addr;
 	return 0;
 }
 
@@ -147,9 +119,6 @@ void ttm_mem_reg_iounmap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem,
 	struct ttm_mem_type_manager *man;
 
 	man = &bdev->man[mem->mem_type];
-
-	if (virtual && (man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP))
-		iounmap(virtual);
 	ttm_mem_io_free(bdev, mem);
 }
 
@@ -401,30 +370,6 @@ pgprot_t ttm_io_prot(uint32_t caching_flags, pgprot_t tmp)
 }
 EXPORT_SYMBOL(ttm_io_prot);
 
-static int ttm_bo_ioremap(struct ttm_buffer_object *bo,
-			  unsigned long offset,
-			  unsigned long size,
-			  struct ttm_bo_kmap_obj *map)
-{
-	struct ttm_bo_device *bdev = bo->bdev;
-	struct ttm_mem_reg *mem = &bo->mem;
-	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
-
-	if (!(man->flags & TTM_MEMTYPE_FLAG_NEEDS_IOREMAP)) {
-		map->bo_kmap_type = ttm_bo_map_premapped;
-		map->virtual = (void *)(((u8 *) bo->mem.bus.addr) + offset);
-	} else {
-		map->bo_kmap_type = ttm_bo_map_iomap;
-		if (mem->placement & TTM_PL_FLAG_WC)
-			map->virtual = ioremap_wc(bo->mem.bus.base + bo->mem.bus.offset + offset,
-						  size);
-		else
-			map->virtual = ioremap_nocache(bo->mem.bus.base + bo->mem.bus.offset + offset,
-						       size);
-	}
-	return (!map->virtual) ? -ENOMEM : 0;
-}
-
 static int ttm_bo_kmap_ttm(struct ttm_buffer_object *bo,
 			   unsigned long start_page,
 			   unsigned long num_pages,
@@ -495,7 +440,9 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo,
 	} else {
 		offset = start_page << PAGE_SHIFT;
 		size = num_pages << PAGE_SHIFT;
-		return ttm_bo_ioremap(bo, offset, size, map);
+		map->bo_kmap_type = ttm_bo_map_iomap;
+		map->virtual = (void *)(((u8 *) bo->mem.bus.addr) + offset);
+		return 0;
 	}
 }
 EXPORT_SYMBOL(ttm_bo_kmap);
@@ -506,7 +453,6 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map)
 		return;
 	switch (map->bo_kmap_type) {
 	case ttm_bo_map_iomap:
-		iounmap(map->virtual);
 		ttm_mem_io_free(map->bo->bdev, &map->bo->mem);
 		break;
 	case ttm_bo_map_vmap:
@@ -515,8 +461,6 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map)
 	case ttm_bo_map_kmap:
 		kunmap(map->page);
 		break;
-	case ttm_bo_map_premapped:
-		break;
 	default:
 		BUG();
 	}
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index 3e273e0..f3171a2 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -282,8 +282,7 @@ struct ttm_buffer_object {
  *
  * Object describing a kernel mapping. Since a TTM bo may be located
  * in various memory types with various caching policies, the
- * mapping can either be an ioremap, a vmap, a kmap or part of a
- * premapped region.
+ * mapping can either be an ioremap, a vmap, a kmap.
  */
 
 #define TTM_BO_MAP_IOMEM_MASK 0x80
@@ -294,7 +293,6 @@ struct ttm_bo_kmap_obj {
 		ttm_bo_map_iomap        = 1 | TTM_BO_MAP_IOMEM_MASK,
 		ttm_bo_map_vmap         = 2,
 		ttm_bo_map_kmap         = 3,
-		ttm_bo_map_premapped    = 4 | TTM_BO_MAP_IOMEM_MASK,
 	} bo_kmap_type;
 	struct ttm_buffer_object *bo;
 };
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index da39865..7720b17 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -176,8 +176,6 @@ struct ttm_tt {
 
 #define TTM_MEMTYPE_FLAG_FIXED         (1 << 0)	/* Fixed (on-card) PCI memory */
 #define TTM_MEMTYPE_FLAG_MAPPABLE      (1 << 1)	/* Memory mappable */
-#define TTM_MEMTYPE_FLAG_NEEDS_IOREMAP (1 << 2)	/* Fixed memory needs ioremap
-						   before kernel access. */
 #define TTM_MEMTYPE_FLAG_CMA           (1 << 3)	/* Can't map aperture */
 
 /**
@@ -189,13 +187,6 @@ struct ttm_tt {
  * managed by this memory type.
  * @gpu_offset: If used, the GPU offset of the first managed page of
  * fixed memory or the first managed location in an aperture.
- * @io_offset: The io_offset of the first managed page of IO memory or
- * the first managed location in an aperture. For TTM_MEMTYPE_FLAG_CMA
- * memory, this should be set to NULL.
- * @io_size: The size of a managed IO region (fixed memory or aperture).
- * @io_addr: Virtual kernel address if the io region is pre-mapped. For
- * TTM_MEMTYPE_FLAG_NEEDS_IOREMAP there is no pre-mapped io map and
- * @io_addr should be set to NULL.
  * @size: Size of the managed region.
  * @available_caching: A mask of available caching types, TTM_PL_FLAG_XX,
  * as defined in ttm_placement_common.h
@@ -221,9 +212,6 @@ struct ttm_mem_type_manager {
 	bool use_type;
 	uint32_t flags;
 	unsigned long gpu_offset;
-	unsigned long io_offset;
-	unsigned long io_size;
-	void *io_addr;
 	uint64_t size;
 	uint32_t available_caching;
 	uint32_t default_caching;
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

[PATCH 13/13] drm/radeon/kms: enable use of unmappable VRAM V2

[Jerome Glisse]

This patch enable the use of unmappable VRAM thanks to
previous TTM infrastructure change.

V2 update after io_mem_reserve/io_mem_free callback balancing

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/radeon/evergreen.c |    5 -----
 drivers/gpu/drm/radeon/r100.c      |    5 -----
 drivers/gpu/drm/radeon/r600.c      |    5 -----
 drivers/gpu/drm/radeon/rv770.c     |    5 -----
 4 files changed, 0 insertions(+), 20 deletions(-)

diff --git a/drivers/gpu/drm/radeon/evergreen.c b/drivers/gpu/drm/radeon/evergreen.c
index 7672f11..48e0dfb 100644
--- a/drivers/gpu/drm/radeon/evergreen.c
+++ b/drivers/gpu/drm/radeon/evergreen.c
@@ -475,11 +475,6 @@ int evergreen_mc_init(struct radeon_device *rdev)
 	rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
 	rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
 	rdev->mc.visible_vram_size = rdev->mc.aper_size;
-	/* FIXME remove this once we support unmappable VRAM */
-	if (rdev->mc.mc_vram_size > rdev->mc.aper_size) {
-		rdev->mc.mc_vram_size = rdev->mc.aper_size;
-		rdev->mc.real_vram_size = rdev->mc.aper_size;
-	}
 	r600_vram_gtt_location(rdev, &rdev->mc);
 	radeon_update_bandwidth_info(rdev);
 
diff --git a/drivers/gpu/drm/radeon/r100.c b/drivers/gpu/drm/radeon/r100.c
index 7a1180d..fc95907 100644
--- a/drivers/gpu/drm/radeon/r100.c
+++ b/drivers/gpu/drm/radeon/r100.c
@@ -2026,11 +2026,6 @@ void r100_vram_init_sizes(struct radeon_device *rdev)
 		else
 			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
 	}
-	/* FIXME remove this once we support unmappable VRAM */
-	if (rdev->mc.mc_vram_size > rdev->mc.aper_size) {
-		rdev->mc.mc_vram_size = rdev->mc.aper_size;
-		rdev->mc.real_vram_size = rdev->mc.aper_size;
-	}
 }
 
 void r100_vga_set_state(struct radeon_device *rdev, bool state)
diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
index a948947..4206c8d 100644
--- a/drivers/gpu/drm/radeon/r600.c
+++ b/drivers/gpu/drm/radeon/r600.c
@@ -713,11 +713,6 @@ int r600_mc_init(struct radeon_device *rdev)
 	rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
 	rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
 	rdev->mc.visible_vram_size = rdev->mc.aper_size;
-	/* FIXME remove this once we support unmappable VRAM */
-	if (rdev->mc.mc_vram_size > rdev->mc.aper_size) {
-		rdev->mc.mc_vram_size = rdev->mc.aper_size;
-		rdev->mc.real_vram_size = rdev->mc.aper_size;
-	}
 	r600_vram_gtt_location(rdev, &rdev->mc);
 
 	if (rdev->flags & RADEON_IS_IGP)
diff --git a/drivers/gpu/drm/radeon/rv770.c b/drivers/gpu/drm/radeon/rv770.c
index 2b8a4e1..599121b 100644
--- a/drivers/gpu/drm/radeon/rv770.c
+++ b/drivers/gpu/drm/radeon/rv770.c
@@ -905,11 +905,6 @@ int rv770_mc_init(struct radeon_device *rdev)
 	rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
 	rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
 	rdev->mc.visible_vram_size = rdev->mc.aper_size;
-	/* FIXME remove this once we support unmappable VRAM */
-	if (rdev->mc.mc_vram_size > rdev->mc.aper_size) {
-		rdev->mc.mc_vram_size = rdev->mc.aper_size;
-		rdev->mc.real_vram_size = rdev->mc.aper_size;
-	}
 	r600_vram_gtt_location(rdev, &rdev->mc);
 	radeon_update_bandwidth_info(rdev);
 
-- 
1.6.6.1


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

Re: [PATCH 05/13] drm/ttm: ttm_fault callback to allow driver to handle bo placement V5

[Dave Airlie]

On Wed, Apr 7, 2010 at 8:21 PM, Jerome Glisse <jglisse@redhat.com> wrote:
> On fault the driver is given the opportunity to perform any operation
> it sees fit in order to place the buffer into a CPU visible area of
> memory. This patch doesn't break TTM users, nouveau, vmwgfx and radeon
> should keep working properly. Future patch will take advantage of this
> infrastructure and remove the old path from TTM once driver are
> converted.
>
> V2 return VM_FAULT_NOPAGE if callback return -EBUSY or -ERESTARTSYS
> V3 balance io_mem_reserve and io_mem_free call, fault_reserve_notify
>   is responsible to perform any necessary task for mapping to succeed
> V4 minor cleanup, atomic_t -> bool as member is protected by reserve
>   mecanism from concurent access
> V5 the callback is now responsible for iomapping the bo and providing
>   a virtual address this simplify TTM and will allow to get rid of
>   TTM_MEMTYPE_FLAG_NEEDS_IOREMAP

Okay I've applied all these to drm-next and it fails to run X for longer than
5 mins on my 32-bit machine.

The whole IO reserve thing looks like a bad idea for anything but kmap,
since it ioremap's the pages, but we have limited ioremap space on 32-bit,
and you hit that and vmallocs all start failing soon after.

>
> -       ret = ttm_bo_pci_offset(bdev, &bo->mem, &bus_base, &bus_offset,
> -                               &bus_size);
> -       if (unlikely(ret != 0)) {
> +       ret = ttm_mem_io_reserve(bdev, &bo->mem);
> +       if (ret) {
>                retval = VM_FAULT_SIGBUS;
>                goto out_unlock;
>        }

This is the start of the insanity, since ever bo that takes a
userspace pagefault
ends up using ioremap space for *no* reason whatsoever at all. You only need
to use ioremap space if the *kernel* is accessing the bo not if userspace is,
so really only in the kmap space.

I've dropped it all from drm-next until you can resolve this.

Dave.

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

Re: [PATCH 05/13] drm/ttm: ttm_fault callback to allow driver to handle bo placement V5

[Thomas Hellstrom]

Dave Airlie wrote:
> On Wed, Apr 7, 2010 at 8:21 PM, Jerome Glisse <jglisse@redhat.com> wrote:
>   
>> On fault the driver is given the opportunity to perform any operation
>> it sees fit in order to place the buffer into a CPU visible area of
>> memory. This patch doesn't break TTM users, nouveau, vmwgfx and radeon
>> should keep working properly. Future patch will take advantage of this
>> infrastructure and remove the old path from TTM once driver are
>> converted.
>>
>> V2 return VM_FAULT_NOPAGE if callback return -EBUSY or -ERESTARTSYS
>> V3 balance io_mem_reserve and io_mem_free call, fault_reserve_notify
>>   is responsible to perform any necessary task for mapping to succeed
>> V4 minor cleanup, atomic_t -> bool as member is protected by reserve
>>   mecanism from concurent access
>> V5 the callback is now responsible for iomapping the bo and providing
>>   a virtual address this simplify TTM and will allow to get rid of
>>   TTM_MEMTYPE_FLAG_NEEDS_IOREMAP
>>     
>
> Okay I've applied all these to drm-next and it fails to run X for longer than
> 5 mins on my 32-bit machine.
>
> The whole IO reserve thing looks like a bad idea for anything but kmap,
> since it ioremap's the pages, but we have limited ioremap space on 32-bit,
> and you hit that and vmallocs all start failing soon after.
>
>   
>> -       ret = ttm_bo_pci_offset(bdev, &bo->mem, &bus_base, &bus_offset,
>> -                               &bus_size);
>> -       if (unlikely(ret != 0)) {
>> +       ret = ttm_mem_io_reserve(bdev, &bo->mem);
>> +       if (ret) {
>>                retval = VM_FAULT_SIGBUS;
>>                goto out_unlock;
>>        }
>>     
>
> This is the start of the insanity, since ever bo that takes a
> userspace pagefault
> ends up using ioremap space for *no* reason whatsoever at all. You only need
> to use ioremap space if the *kernel* is accessing the bo not if userspace is,
> so really only in the kmap space.
>
> I've dropped it all from drm-next until you can resolve this.
>
> Dave.
>   

Indeed, ttm_mem_io_reserve() shouldn't do anything like that. It should 
place the bo data in a mappable region. If there is a copy involving 
ioremap() to do so, it should be iounmapped() before the operation is 
finished.

Thomas



------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--