drm/radeon: introduce a sub allocator and convert ib pool to itHEADmaster

Somewhat specializaed sub-allocator designed to perform sub-allocation
for command buffer not only for current cs ioctl but for future command
submission ioctl as well. Patch also convert current ib pool to use
the sub allocator. Idea is that ib poll buffer can be share with other
command buffer submission not having 64K granularity.

Signed-off-by: Jerome Glisse <jglisse@redhat.com>

5 files changed, 345 insertions, 165 deletions

diff --git a/drivers/gpu/drm/radeon/Makefile b/drivers/gpu/drm/radeon/Makefile
index 94dcdc746041..2139fe893ec5 100644
--- a/drivers/gpu/drm/radeon/Makefile
+++ b/drivers/gpu/drm/radeon/Makefile
@@ -71,7 +71,7 @@ radeon-y += radeon_device.o radeon_asic.o radeon_kms.o \
 	r600_blit_kms.o radeon_pm.o atombios_dp.o r600_audio.o r600_hdmi.o \
 	evergreen.o evergreen_cs.o evergreen_blit_shaders.o evergreen_blit_kms.o \
 	radeon_trace_points.o ni.o cayman_blit_shaders.o atombios_encoders.o \
-	radeon_semaphore.o
+	radeon_semaphore.o radeon_sa.o
 
 radeon-$(CONFIG_COMPAT) += radeon_ioc32.o
 radeon-$(CONFIG_VGA_SWITCHEROO) += radeon_atpx_handler.o
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 8163ffa69c45..bd61fb182cee 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -317,6 +317,53 @@ struct radeon_bo_list {
 	u32			tiling_flags;
 };
 
+/* sub-allocation manager, it has to be protected by another lock.
+ * By conception this is an helper for other part of the driver
+ * like the indirect buffer or semaphore, which both have their
+ * locking.
+ *
+ * Principe is simple, we keep a list of sub allocation in offset
+ * order (first entry has offset == 0, last entry has the highest
+ * offset).
+ *
+ * When allocating new object we first check if there is room at
+ * the end total_size - (last_object_offset + last_object_size) >=
+ * alloc_size. If so we allocate new object there.
+ *
+ * When there is not enough room at the end, we start waiting for
+ * each sub object until we reach object_offset+object_size >=
+ * alloc_size, this object then become the sub object we return.
+ *
+ * Alignment can't be bigger than page size.
+ *
+ * Hole are not considered for allocation to keep things simple.
+ * Assumption is that there won't be hole (all object on same
+ * alignment).
+ */
+struct radeon_sa_manager {
+	struct radeon_bo	*bo;
+	struct list_head	sa_bo;
+	unsigned		size;
+	uint64_t		gpu_addr;
+	void			*cpu_ptr;
+};
+
+struct radeon_sa_bo;
+typedef void (*radeon_sa_bo_destroy_t)(struct radeon_device *rdev,
+				       struct radeon_sa_bo *sa_bo);
+typedef bool (*radeon_sa_bo_done_t)(struct radeon_device *rdev,
+				    struct radeon_sa_bo *sa_bo);
+
+/* sub-allocation buffer */
+struct radeon_sa_bo {
+	struct list_head		list;
+	struct radeon_sa_manager	*manager;
+	unsigned			offset;
+	unsigned			size;
+	radeon_sa_bo_destroy_t		destroy;
+	radeon_sa_bo_done_t		done;
+};
+
 /*
  * GEM objects.
  */
@@ -504,13 +551,12 @@ void radeon_irq_kms_pflip_irq_put(struct radeon_device *rdev, int crtc);
  */
 
 struct radeon_ib {
-	struct list_head	list;
+	struct radeon_sa_bo	sa_bo;
 	unsigned		idx;
+	uint32_t		length_dw;
 	uint64_t		gpu_addr;
-	struct radeon_fence	*fence;
 	uint32_t		*ptr;
-	uint32_t		length_dw;
-	bool			free;
+	struct radeon_fence	*fence;
 };
 
 /*
@@ -518,12 +564,11 @@ struct radeon_ib {
  * mutex protects scheduled_ibs, ready, alloc_bm
  */
 struct radeon_ib_pool {
-	struct mutex		mutex;
-	struct radeon_bo	*robj;
-	struct list_head	bogus_ib;
-	struct radeon_ib	ibs[RADEON_IB_POOL_SIZE];
-	bool			ready;
-	unsigned		head_id;
+	struct mutex			mutex;
+	struct radeon_sa_manager	sa_manager;
+	struct radeon_ib		ibs[RADEON_IB_POOL_SIZE];
+	bool				ready;
+	unsigned			head_id;
 };
 
 struct radeon_ring {
@@ -605,7 +650,6 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib);
 int radeon_ib_pool_init(struct radeon_device *rdev);
 void radeon_ib_pool_fini(struct radeon_device *rdev);
 int radeon_ib_test(struct radeon_device *rdev);
-extern void radeon_ib_bogus_add(struct radeon_device *rdev, struct radeon_ib *ib);
 /* Ring access between begin & end cannot sleep */
 int radeon_ring_index(struct radeon_device *rdev, struct radeon_ring *cp);
 void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *cp);
diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
index b07f0f9b8627..51e69671c7f9 100644
--- a/drivers/gpu/drm/radeon/radeon_object.h
+++ b/drivers/gpu/drm/radeon/radeon_object.h
@@ -128,4 +128,22 @@ extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
 					struct ttm_mem_reg *mem);
 extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
 extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
+
+/*
+ * sub allocation
+ */
+extern int radeon_sa_bo_manager_init(struct radeon_device *rdev,
+				     struct radeon_sa_manager *sa_manager,
+				     unsigned size, u32 domain);
+extern void radeon_sa_bo_manager_fini(struct radeon_device *rdev,
+				      struct radeon_sa_manager *sa_manager);
+extern int radeon_sa_bo_new(struct radeon_device *rdev,
+			    struct radeon_sa_manager *sa_manager,
+			    struct radeon_sa_bo *sa_bo,
+			    unsigned size, unsigned align,
+			    radeon_sa_bo_destroy_t destroy,
+			    radeon_sa_bo_done_t done);
+extern void radeon_sa_bo_free(struct radeon_device *rdev,
+			      struct radeon_sa_bo *sa_bo);
+
 #endif
diff --git a/drivers/gpu/drm/radeon/radeon_ring.c b/drivers/gpu/drm/radeon/radeon_ring.c
index 133e2636cea0..fc733d2701d2 100644
--- a/drivers/gpu/drm/radeon/radeon_ring.c
+++ b/drivers/gpu/drm/radeon/radeon_ring.c
@@ -74,92 +74,87 @@ void radeon_ring_write(struct radeon_ring *ring, uint32_t v)
 	ring->ring_free_dw--;
 }
 
-void radeon_ib_bogus_cleanup(struct radeon_device *rdev)
+/*
+ * IB.
+ */
+static void radeon_ib_destroy(struct radeon_device *rdev,
+			      struct radeon_sa_bo *sa_bo)
+{
+	/* noop no things to free */
+}
+
+static bool radeon_ib_done(struct radeon_device *rdev,
+			   struct radeon_sa_bo *sa_bo)
 {
-	struct radeon_ib *ib, *n;
+	struct radeon_ib *ib = container_of(sa_bo, struct radeon_ib, sa_bo);
+	bool done = false;
 
-	list_for_each_entry_safe(ib, n, &rdev->ib_pool.bogus_ib, list) {
-		list_del(&ib->list);
-		vfree(ib->ptr);
-		kfree(ib);
+	if (ib->fence && ib->fence->emitted) {
+		if (radeon_fence_signaled(ib->fence)) {
+			radeon_fence_unref(&ib->fence);
+			done = true;
+		}
 	}
+	return done;
 }
 
-void radeon_ib_bogus_add(struct radeon_device *rdev, struct radeon_ib *ib)
+void radeon_ib_try_free_locked(struct radeon_device *rdev, struct radeon_ib *ib)
 {
-	struct radeon_ib *bib;
-
-	bib = kmalloc(sizeof(*bib), GFP_KERNEL);
-	if (bib == NULL)
-		return;
-	bib->ptr = vmalloc(ib->length_dw * 4);
-	if (bib->ptr == NULL) {
-		kfree(bib);
-		return;
+	if (ib->fence && ib->fence->emitted) {
+		if (radeon_fence_signaled(ib->fence)) {
+			radeon_sa_bo_free(rdev, &ib->sa_bo);
+			radeon_fence_unref(&ib->fence);
+		}
 	}
-	memcpy(bib->ptr, ib->ptr, ib->length_dw * 4);
-	bib->length_dw = ib->length_dw;
-	mutex_lock(&rdev->ib_pool.mutex);
-	list_add_tail(&bib->list, &rdev->ib_pool.bogus_ib);
-	mutex_unlock(&rdev->ib_pool.mutex);
 }
 
-/*
- * IB.
- */
 int radeon_ib_get(struct radeon_device *rdev, int ring, struct radeon_ib **ib)
 {
 	struct radeon_fence *fence;
-	struct radeon_ib *nib;
-	int r = 0, i, c;
+	int r = 0, i, idx;
 
 	*ib = NULL;
+
 	r = radeon_fence_create(rdev, &fence, ring);
 	if (r) {
 		dev_err(rdev->dev, "failed to create fence for new IB\n");
 		return r;
 	}
+
 	mutex_lock(&rdev->ib_pool.mutex);
-	for (i = rdev->ib_pool.head_id, c = 0, nib = NULL; c < RADEON_IB_POOL_SIZE; c++, i++) {
-		i &= (RADEON_IB_POOL_SIZE - 1);
-		if (rdev->ib_pool.ibs[i].free) {
-			nib = &rdev->ib_pool.ibs[i];
-			break;
-		}
-	}
-	if (nib == NULL) {
-		/* This should never happen, it means we allocated all
-		 * IB and haven't scheduled one yet, return EBUSY to
-		 * userspace hoping that on ioctl recall we get better
-		 * luck
-		 */
-		dev_err(rdev->dev, "no free indirect buffer !\n");
-		mutex_unlock(&rdev->ib_pool.mutex);
-		radeon_fence_unref(&fence);
-		return -EBUSY;
-	}
-	rdev->ib_pool.head_id = (nib->idx + 1) & (RADEON_IB_POOL_SIZE - 1);
-	nib->free = false;
-	if (nib->fence) {
-		mutex_unlock(&rdev->ib_pool.mutex);
-		r = radeon_fence_wait(nib->fence, false);
-		if (r) {
-			dev_err(rdev->dev, "error waiting fence of IB(%u:0x%016lX:%u)\n",
-				nib->idx, (unsigned long)nib->gpu_addr, nib->length_dw);
-			mutex_lock(&rdev->ib_pool.mutex);
-			nib->free = true;
-			mutex_unlock(&rdev->ib_pool.mutex);
-			radeon_fence_unref(&fence);
-			return r;
+retry:
+	for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
+		idx = (i + rdev->ib_pool.head_id) & (RADEON_IB_POOL_SIZE - 1);
+		radeon_ib_try_free_locked(rdev, &rdev->ib_pool.ibs[idx]);
+		if (rdev->ib_pool.ibs[idx].fence == NULL) {
+			r = radeon_sa_bo_new(rdev, &rdev->ib_pool.sa_manager,
+					     &rdev->ib_pool.ibs[idx].sa_bo,
+					     64*1024, 64,
+					     &radeon_ib_destroy,
+					     &radeon_ib_done);
+			if (!r) {
+				*ib = &rdev->ib_pool.ibs[idx];
+				(*ib)->ptr = rdev->ib_pool.sa_manager.cpu_ptr;
+				(*ib)->ptr += ((*ib)->sa_bo.offset >> 2);
+				(*ib)->gpu_addr = rdev->ib_pool.sa_manager.gpu_addr;
+				(*ib)->gpu_addr += (*ib)->sa_bo.offset;
+				(*ib)->fence = fence;
+				rdev->ib_pool.head_id = idx;
+				mutex_unlock(&rdev->ib_pool.mutex);
+				return 0;
+			}
 		}
-		mutex_lock(&rdev->ib_pool.mutex);
 	}
-	radeon_fence_unref(&nib->fence);
-	nib->fence = fence;
-	nib->length_dw = 0;
+	/* this should be rare event, ie all ib scheduled none signaled
+	 * yet.
+	 */
+	r = radeon_fence_wait(rdev->ib_pool.ibs[rdev->ib_pool.head_id].fence, true);
+	if (!r)
+		goto retry;
 	mutex_unlock(&rdev->ib_pool.mutex);
-	*ib = nib;
-	return 0;
+	radeon_fence_unref(&fence);
+	DRM_INFO("wait for fence failed %d\n", r);
+	return r;
 }
 
 void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib **ib)
@@ -170,10 +165,11 @@ void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib **ib)
 	if (tmp == NULL) {
 		return;
 	}
-	if (!tmp->fence->emitted)
-		radeon_fence_unref(&tmp->fence);
 	mutex_lock(&rdev->ib_pool.mutex);
-	tmp->free = true;
+	if (tmp->fence && !tmp->fence->emitted) {
+		radeon_sa_bo_free(rdev, &tmp->sa_bo);
+		radeon_fence_unref(&tmp->fence);
+	}
 	mutex_unlock(&rdev->ib_pool.mutex);
 }
 
@@ -196,92 +192,62 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib)
 	}
 	radeon_ring_ib_execute(rdev, ib->fence->ring, ib);
 	radeon_fence_emit(rdev, ib->fence);
-	mutex_lock(&rdev->ib_pool.mutex);
-	/* once scheduled IB is considered free and protected by the fence */
-	ib->free = true;
-	mutex_unlock(&rdev->ib_pool.mutex);
 	radeon_ring_unlock_commit(rdev, ring);
 	return 0;
 }
 
 int radeon_ib_pool_init(struct radeon_device *rdev)
 {
-	void *ptr;
-	uint64_t gpu_addr;
-	int i;
-	int r = 0;
+	int i, r;
 
-	if (rdev->ib_pool.robj)
+	mutex_lock(&rdev->ib_pool.mutex);
+	if (rdev->ib_pool.ready) {
+		mutex_unlock(&rdev->ib_pool.mutex);
 		return 0;
-	INIT_LIST_HEAD(&rdev->ib_pool.bogus_ib);
-	/* Allocate 1M object buffer */
-	r = radeon_bo_create(rdev, RADEON_IB_POOL_SIZE*64*1024,
-			     PAGE_SIZE, true, RADEON_GEM_DOMAIN_GTT,
-			     &rdev->ib_pool.robj);
-	if (r) {
-		DRM_ERROR("radeon: failed to ib pool (%d).\n", r);
-		return r;
 	}
-	r = radeon_bo_reserve(rdev->ib_pool.robj, false);
-	if (unlikely(r != 0))
-		return r;
-	r = radeon_bo_pin(rdev->ib_pool.robj, RADEON_GEM_DOMAIN_GTT, &gpu_addr);
-	if (r) {
-		radeon_bo_unreserve(rdev->ib_pool.robj);
-		DRM_ERROR("radeon: failed to pin ib pool (%d).\n", r);
-		return r;
-	}
-	r = radeon_bo_kmap(rdev->ib_pool.robj, &ptr);
-	radeon_bo_unreserve(rdev->ib_pool.robj);
+
+	r = radeon_sa_bo_manager_init(rdev, &rdev->ib_pool.sa_manager,
+				      RADEON_IB_POOL_SIZE*64*1024,
+				      RADEON_GEM_DOMAIN_GTT);
 	if (r) {
-		DRM_ERROR("radeon: failed to map ib pool (%d).\n", r);
+		mutex_unlock(&rdev->ib_pool.mutex);
 		return r;
 	}
-	for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
-		unsigned offset;
 
-		offset = i * 64 * 1024;
-		rdev->ib_pool.ibs[i].gpu_addr = gpu_addr + offset;
-		rdev->ib_pool.ibs[i].ptr = ptr + offset;
+	for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
+		rdev->ib_pool.ibs[i].fence = NULL;
 		rdev->ib_pool.ibs[i].idx = i;
 		rdev->ib_pool.ibs[i].length_dw = 0;
-		rdev->ib_pool.ibs[i].free = true;
+		INIT_LIST_HEAD(&rdev->ib_pool.ibs[i].sa_bo.list);
 	}
 	rdev->ib_pool.head_id = 0;
 	rdev->ib_pool.ready = true;
 	DRM_INFO("radeon: ib pool ready.\n");
+
 	if (radeon_debugfs_ib_init(rdev)) {
 		DRM_ERROR("Failed to register debugfs file for IB !\n");
 	}
 	if (radeon_debugfs_ring_init(rdev)) {
 		DRM_ERROR("Failed to register debugfs file for rings !\n");
 	}
-	return r;
+	mutex_unlock(&rdev->ib_pool.mutex);
+	return 0;
 }
 
 void radeon_ib_pool_fini(struct radeon_device *rdev)
 {
-	int r;
-	struct radeon_bo *robj;
+	unsigned i;
 
-	if (!rdev->ib_pool.ready) {
-		return;
-	}
 	mutex_lock(&rdev->ib_pool.mutex);
-	radeon_ib_bogus_cleanup(rdev);
-	robj = rdev->ib_pool.robj;
-	rdev->ib_pool.robj = NULL;
-	mutex_unlock(&rdev->ib_pool.mutex);
-
-	if (robj) {
-		r = radeon_bo_reserve(robj, false);
-		if (likely(r == 0)) {
-			radeon_bo_kunmap(robj);
-			radeon_bo_unpin(robj);
-			radeon_bo_unreserve(robj);
+	if (rdev->ib_pool.ready) {
+		for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
+			radeon_sa_bo_free(rdev, &rdev->ib_pool.ibs[i].sa_bo);
+			radeon_fence_unref(&rdev->ib_pool.ibs[i].fence);
 		}
-		radeon_bo_unref(&robj);
+		radeon_sa_bo_manager_fini(rdev, &rdev->ib_pool.sa_manager);
+		rdev->ib_pool.ready = false;
 	}
+	mutex_unlock(&rdev->ib_pool.mutex);
 }
 
 
@@ -509,37 +475,8 @@ static int radeon_debugfs_ib_info(struct seq_file *m, void *data)
 	return 0;
 }
 
-static int radeon_debugfs_ib_bogus_info(struct seq_file *m, void *data)
-{
-	struct drm_info_node *node = (struct drm_info_node *) m->private;
-	struct radeon_device *rdev = node->info_ent->data;
-	struct radeon_ib *ib;
-	unsigned i;
-
-	mutex_lock(&rdev->ib_pool.mutex);
-	if (list_empty(&rdev->ib_pool.bogus_ib)) {
-		mutex_unlock(&rdev->ib_pool.mutex);
-		seq_printf(m, "no bogus IB recorded\n");
-		return 0;
-	}
-	ib = list_first_entry(&rdev->ib_pool.bogus_ib, struct radeon_ib, list);
-	list_del_init(&ib->list);
-	mutex_unlock(&rdev->ib_pool.mutex);
-	seq_printf(m, "IB size %05u dwords\n", ib->length_dw);
-	for (i = 0; i < ib->length_dw; i++) {
-		seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]);
-	}
-	vfree(ib->ptr);
-	kfree(ib);
-	return 0;
-}
-
 static struct drm_info_list radeon_debugfs_ib_list[RADEON_IB_POOL_SIZE];
 static char radeon_debugfs_ib_names[RADEON_IB_POOL_SIZE][32];
-
-static struct drm_info_list radeon_debugfs_ib_bogus_info_list[] = {
-	{"radeon_ib_bogus", radeon_debugfs_ib_bogus_info, 0, NULL},
-};
 #endif
 
 int radeon_debugfs_ring_init(struct radeon_device *rdev)
@@ -556,12 +493,7 @@ int radeon_debugfs_ib_init(struct radeon_device *rdev)
 {
 #if defined(CONFIG_DEBUG_FS)
 	unsigned i;
-	int r;
 
-	radeon_debugfs_ib_bogus_info_list[0].data = rdev;
-	r = radeon_debugfs_add_files(rdev, radeon_debugfs_ib_bogus_info_list, 1);
-	if (r)
-		return r;
 	for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
 		sprintf(radeon_debugfs_ib_names[i], "radeon_ib_%04u", i);
 		radeon_debugfs_ib_list[i].name = radeon_debugfs_ib_names[i];
diff --git a/drivers/gpu/drm/radeon/radeon_sa.c b/drivers/gpu/drm/radeon/radeon_sa.c
new file mode 100644
index 000000000000..5d78b466441e
--- /dev/null
+++ b/drivers/gpu/drm/radeon/radeon_sa.c
@@ -0,0 +1,186 @@
+/*
+ * Copyright 2011 Red Hat Inc.
+ * All Rights Reserved.
+ *
+ * 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, sub license, 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 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ */
+/*
+ * Authors:
+ *    Jerome Glisse <glisse@freedesktop.org>
+ */
+#include "drmP.h"
+#include "drm.h"
+#include "radeon.h"
+
+int radeon_sa_bo_manager_init(struct radeon_device *rdev,
+			      struct radeon_sa_manager *sa_manager,
+			      unsigned size, u32 domain)
+{
+	int r;
+
+	sa_manager->bo = NULL;
+	sa_manager->size = size;
+	INIT_LIST_HEAD(&sa_manager->sa_bo);
+
+	r = radeon_bo_create(rdev, size, RADEON_GPU_PAGE_SIZE, true,
+			     domain, &sa_manager->bo);
+	if (r) {
+		dev_err(rdev->dev, "(%d) failed to allocate bo for manager\n", r);
+		return r;
+	}
+
+	/* map the buffer */
+	r = radeon_bo_reserve(sa_manager->bo, false);
+	if (r) {
+		dev_err(rdev->dev, "(%d) failed to reserve manager bo\n", r);
+		return r;
+	}
+	r = radeon_bo_pin(sa_manager->bo, domain, &sa_manager->gpu_addr);
+	if (r) {
+		radeon_bo_unreserve(sa_manager->bo);
+		radeon_sa_bo_manager_fini(rdev, sa_manager);
+		dev_err(rdev->dev, "(%d) failed to pin manager bo\n", r);
+		return r;
+	}
+	r = radeon_bo_kmap(sa_manager->bo, &sa_manager->cpu_ptr);
+	radeon_bo_unreserve(sa_manager->bo);
+	if (r) {
+		radeon_sa_bo_manager_fini(rdev, sa_manager);
+	}
+
+	return r;
+}
+
+void radeon_sa_bo_manager_fini(struct radeon_device *rdev,
+			       struct radeon_sa_manager *sa_manager)
+{
+	struct radeon_sa_bo *sa_bo, *tmp;
+	int r;
+
+	list_for_each_entry_safe(sa_bo, tmp, &sa_manager->sa_bo, list) {
+		list_del_init(&sa_bo->list);
+		sa_bo->destroy(rdev, sa_bo);
+	}
+
+	r = radeon_bo_reserve(sa_manager->bo, false);
+	if (!r) {
+		radeon_bo_kunmap(sa_manager->bo);
+		radeon_bo_unpin(sa_manager->bo);
+		radeon_bo_unreserve(sa_manager->bo);
+	}
+	radeon_bo_unref(&sa_manager->bo);
+	sa_manager->size = 0;
+}
+
+/*
+ * Principe is simple, we keep a list of sub allocation in offset
+ * order (first entry has offset == 0, last entry has the highest
+ * offset).
+ *
+ * When allocating new object we first check if there is room at
+ * the end total_size - (last_object_offset + last_object_size) >=
+ * alloc_size. If so we allocate new object there.
+ *
+ * When there is not enough room at the end, we start waiting for
+ * each sub object until we reach object_offset+object_size >=
+ * alloc_size, this object then become the sub object we return.
+ *
+ * Alignment can't be bigger than page size
+ */
+int radeon_sa_bo_new(struct radeon_device *rdev,
+		     struct radeon_sa_manager *sa_manager,
+		     struct radeon_sa_bo *sa_bo,
+		     unsigned size, unsigned align,
+		     radeon_sa_bo_destroy_t destroy,
+		     radeon_sa_bo_done_t done)
+{
+	struct radeon_sa_bo *tmp, *next;
+	struct list_head *head;
+	unsigned offset = 0, wasted = 0;
+
+	BUG_ON(align > RADEON_GPU_PAGE_SIZE);
+	BUG_ON(size > sa_manager->size);
+
+	/* no one ? */
+	head = sa_manager->sa_bo.prev;
+	if (list_empty(&sa_manager->sa_bo)) {
+		goto out;
+	}
+
+	/* room at the end ? */
+	tmp = list_entry(sa_manager->sa_bo.prev, struct radeon_sa_bo, list);
+	offset = tmp->offset + tmp->size;
+	wasted = offset % align;
+	if (wasted) {
+		wasted = align - wasted;
+	}
+	offset += wasted;
+	if ((sa_manager->size - offset) >= size) {
+		goto out;
+	}
+
+	/* have to wait */
+	offset = 0;
+	list_for_each_entry_safe(tmp, next, &sa_manager->sa_bo, list) {
+		/* room before this object */
+		if ((tmp->offset - offset) >= size) {
+			head = tmp->list.prev;
+			goto out;
+		}
+		/* sub object that have not been scheduled are not
+		 * considered
+		 */
+		if (tmp->done(rdev, tmp)) {
+			if (((tmp->offset + tmp->size) - offset) >= size) {
+				head = tmp->list.prev;
+				list_del_init(&tmp->list);
+				tmp->destroy(rdev, tmp);
+				goto out;
+			}
+			list_del_init(&tmp->list);
+			tmp->destroy(rdev, tmp);
+		} else {
+			offset = tmp->offset + tmp->size;
+			wasted = offset % align;
+			if (wasted) {
+				wasted = align - wasted;
+			}
+			offset += wasted;
+		}
+	}
+	/* failed to find somethings big enough to wait for */
+	return -EBUSY;
+
+out:
+	sa_bo->manager = sa_manager;
+	sa_bo->offset = offset;
+	sa_bo->size = size;
+	sa_bo->destroy = destroy;
+	sa_bo->done = done;
+	list_add(&sa_bo->list, head);
+	return 0;
+}
+
+void radeon_sa_bo_free(struct radeon_device *rdev, struct radeon_sa_bo *sa_bo)
+{
+	list_del_init(&sa_bo->list);
+}