Merge branch 'intel-post-reloc'

Conflicts:

	linux-core/drm_compat.c
	linux-core/drm_compat.h
	linux-core/drm_ttm.c
	shared-core/i915_dma.c

Bump driver minor to 13 due to introduction of new
relocation type.

3 files changed, 27 insertions, 562 deletions

diff --git a/shared-core/i915_dma.c b/shared-core/i915_dma.c
index f6df74a7..407befcc 100644
--- a/shared-core/i915_dma.c
+++ b/shared-core/i915_dma.c
@@ -583,8 +583,8 @@ static int i915_dispatch_cmdbuffer(struct drm_device * dev,
 	return 0;
 }
 
-static int i915_dispatch_batchbuffer(struct drm_device * dev,
-				     drm_i915_batchbuffer_t * batch)
+int i915_dispatch_batchbuffer(struct drm_device * dev,
+			      drm_i915_batchbuffer_t * batch)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct drm_clip_rect __user *boxes = batch->cliprects;
@@ -711,7 +711,7 @@ void i915_dispatch_flip(struct drm_device * dev, int planes, int sync)
 #endif
 }
 
-static int i915_quiescent(struct drm_device *dev)
+int i915_quiescent(struct drm_device *dev)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
 
@@ -796,563 +796,6 @@ static int i915_cmdbuffer(struct drm_device *dev, void *data,
 #define DRM_DEBUG_RELOCATION	0
 #endif
 
-#ifdef I915_HAVE_BUFFER
-
-struct i915_relocatee_info {
-	struct drm_buffer_object *buf;
-	unsigned long offset;
-	u32 *data_page;
-	unsigned page_offset;
-	struct drm_bo_kmap_obj kmap;
-	int is_iomem;
-	int idle;
-	int evicted;
-};
-
-struct drm_i915_validate_buffer {
-	struct drm_buffer_object *buffer;
-	struct drm_bo_info_rep rep;
-	int presumed_offset_correct;
-	void __user *data;
-	int ret;
-};
-
-static void i915_dereference_buffers_locked(struct drm_i915_validate_buffer *buffers,
-					    unsigned num_buffers)
-{
-	while (num_buffers--)
-		drm_bo_usage_deref_locked(&buffers[num_buffers].buffer);
-}
-
-int i915_apply_reloc(struct drm_file *file_priv, int num_buffers,
-		     struct drm_i915_validate_buffer *buffers,
-		     struct i915_relocatee_info *relocatee,
-		     uint32_t *reloc)
-{
-	unsigned index;
-	unsigned long new_cmd_offset;
-	u32 val;
-	int ret, i;
-	int buf_index = -1;
-
-	/*
-	 * FIXME: O(relocs * buffers) complexity.
-	 */
-
-	for (i = 0; i <= num_buffers; i++)
-		if (buffers[i].buffer)
-			if (reloc[2] == buffers[i].buffer->base.hash.key)
-				buf_index = i;
-
-	if (buf_index == -1) {
-		DRM_ERROR("Illegal relocation buffer %08X\n", reloc[2]);
-		return -EINVAL;
-	}
-
-	/*
-	 * Short-circuit relocations that were correctly
-	 * guessed by the client
-	 */
-	if (buffers[buf_index].presumed_offset_correct && !DRM_DEBUG_RELOCATION)
-		return 0;
-
-	new_cmd_offset = reloc[0];
-	if (!relocatee->data_page ||
-	    !drm_bo_same_page(relocatee->offset, new_cmd_offset)) {
-		drm_bo_kunmap(&relocatee->kmap);
-		relocatee->data_page = NULL;
-		relocatee->offset = new_cmd_offset;
-		
-		if (unlikely(!relocatee->idle)) {
-			ret = drm_bo_wait(relocatee->buf, 0, 0, 0);
-			if (ret)
-				return ret;
-			relocatee->idle = 1;
-		}
-
-		ret = drm_bo_kmap(relocatee->buf, new_cmd_offset >> PAGE_SHIFT,
-				  1, &relocatee->kmap);
-		if (ret) {
-			DRM_ERROR("Could not map command buffer to apply relocs\n %08lx", new_cmd_offset);
-			return ret;
-		}
-		relocatee->data_page = drm_bmo_virtual(&relocatee->kmap,
-						       &relocatee->is_iomem);
-		relocatee->page_offset = (relocatee->offset & PAGE_MASK);
-		
-		if (!relocatee->evicted && 
-		    relocatee->buf->mem.flags & DRM_BO_FLAG_CACHED_MAPPED) {
-		    drm_bo_evict_cached(relocatee->buf);
-		    relocatee->evicted = 1;
-		}
-	}
-
-	val = buffers[buf_index].buffer->offset;
-	index = (reloc[0] - relocatee->page_offset) >> 2;
-
-	/* add in validate */
-	val = val + reloc[1];
-
-	if (DRM_DEBUG_RELOCATION) {
-		if (buffers[buf_index].presumed_offset_correct &&
-		    relocatee->data_page[index] != val) {
-			DRM_DEBUG ("Relocation mismatch source %d target %d buffer %d user %08x kernel %08x\n",
-				   reloc[0], reloc[1], buf_index, relocatee->data_page[index], val);
-		}
-	}
-
-	if (relocatee->is_iomem)
-		iowrite32(val, relocatee->data_page + index);
-	else
-		relocatee->data_page[index] = val;
-	return 0;
-}
-
-int i915_process_relocs(struct drm_file *file_priv,
-			uint32_t buf_handle,
-			uint32_t __user **reloc_user_ptr,
-			struct i915_relocatee_info *relocatee,
-			struct drm_i915_validate_buffer *buffers,
-			uint32_t num_buffers)
-{
-	int ret, reloc_stride;
-	uint32_t cur_offset;
-	uint32_t reloc_count;
-	uint32_t reloc_type;
-	uint32_t reloc_buf_size;
-	uint32_t *reloc_buf = NULL;
-	int i;
-
-	/* do a copy from user from the user ptr */
-	ret = get_user(reloc_count, *reloc_user_ptr);
-	if (ret) {
-		DRM_ERROR("Could not map relocation buffer.\n");
-		goto out;
-	}
-
-	ret = get_user(reloc_type, (*reloc_user_ptr)+1);
-	if (ret) {
-		DRM_ERROR("Could not map relocation buffer.\n");
-		goto out;
-	}
-
-	if (reloc_type != 0) {
-		DRM_ERROR("Unsupported relocation type requested\n");
-		ret = -EINVAL;
-		goto out;
-	}
-
-	reloc_buf_size = (I915_RELOC_HEADER + (reloc_count * I915_RELOC0_STRIDE)) * sizeof(uint32_t);
-	reloc_buf = kmalloc(reloc_buf_size, GFP_KERNEL);
-	if (!reloc_buf) {
-		DRM_ERROR("Out of memory for reloc buffer\n");
-		ret = -ENOMEM;
-		goto out;
-	}
-
-	if (copy_from_user(reloc_buf, *reloc_user_ptr, reloc_buf_size)) {
-		ret = -EFAULT;
-		goto out;
-	}
-
-	/* get next relocate buffer handle */
-	*reloc_user_ptr = (uint32_t *)*(unsigned long *)&reloc_buf[2];
-
-	reloc_stride = I915_RELOC0_STRIDE * sizeof(uint32_t); /* may be different for other types of relocs */
-
-	DRM_DEBUG("num relocs is %d, next is %p\n", reloc_count, *reloc_user_ptr);
-
-	for (i = 0; i < reloc_count; i++) {
-		cur_offset = I915_RELOC_HEADER + (i * I915_RELOC0_STRIDE);
-		  
-		ret = i915_apply_reloc(file_priv, num_buffers, buffers,
-				       relocatee, reloc_buf + cur_offset);
-		if (ret)
-			goto out;
-	}
-
-out:
-	if (reloc_buf)
-		kfree(reloc_buf);
-
-	if (relocatee->data_page) {		
-		drm_bo_kunmap(&relocatee->kmap);
-		relocatee->data_page = NULL;
-	}
-
-	return ret;
-}
-
-static int i915_exec_reloc(struct drm_file *file_priv, drm_handle_t buf_handle,
-			   uint32_t __user *reloc_user_ptr,
-			   struct drm_i915_validate_buffer *buffers,
-			   uint32_t buf_count)
-{
-	struct drm_device *dev = file_priv->minor->dev;
-	struct i915_relocatee_info relocatee;
-	int ret = 0;
-	int b;
-
-	/*
-	 * Short circuit relocations when all previous
-	 * buffers offsets were correctly guessed by
-	 * the client
-	 */
-	if (!DRM_DEBUG_RELOCATION) {
-		for (b = 0; b < buf_count; b++)
-			if (!buffers[b].presumed_offset_correct)
-				break;
-	
-		if (b == buf_count)
-			return 0;
-	}
-
-	memset(&relocatee, 0, sizeof(relocatee));
-
-	mutex_lock(&dev->struct_mutex);
-	relocatee.buf = drm_lookup_buffer_object(file_priv, buf_handle, 1);
-	mutex_unlock(&dev->struct_mutex);
-	if (!relocatee.buf) {
-		DRM_DEBUG("relocatee buffer invalid %08x\n", buf_handle);
-		ret = -EINVAL;
-		goto out_err;
-	}
-
-	mutex_lock (&relocatee.buf->mutex);
-	while (reloc_user_ptr) {
-		ret = i915_process_relocs(file_priv, buf_handle, &reloc_user_ptr, &relocatee, buffers, buf_count);
-		if (ret) {
-			DRM_ERROR("process relocs failed\n");
-			goto out_err1;
-		}
-	}
-
-out_err1:
-	mutex_unlock (&relocatee.buf->mutex);
-	drm_bo_usage_deref_unlocked(&relocatee.buf);
-out_err:
-	return ret;
-}
-
-static int i915_check_presumed(struct drm_i915_op_arg *arg,
-			       struct drm_buffer_object *bo,
-			       uint32_t __user *data,
-			       int *presumed_ok)
-{
-	struct drm_bo_op_req *req = &arg->d.req;
-	uint32_t hint_offset;
-	uint32_t hint = req->bo_req.hint;
-
-	*presumed_ok = 0;
-
-	if (!(hint & DRM_BO_HINT_PRESUMED_OFFSET))
-		return 0;
-	if (bo->offset == req->bo_req.presumed_offset) {
-		*presumed_ok = 1;
-		return 0;
-	}
-
-	/*
-	 * We need to turn off the HINT_PRESUMED_OFFSET for this buffer in
-	 * the user-space IOCTL argument list, since the buffer has moved,
-	 * we're about to apply relocations and we might subsequently
-	 * hit an -EAGAIN. In that case the argument list will be reused by
-	 * user-space, but the presumed offset is no longer valid.
-	 *
-	 * Needless to say, this is a bit ugly.
-	 */
-
-       	hint_offset = (uint32_t *)&req->bo_req.hint - (uint32_t *)arg;
-	hint &= ~DRM_BO_HINT_PRESUMED_OFFSET;
-	return __put_user(hint, data + hint_offset);
-}
-
-
-/*
- * Validate, add fence and relocate a block of bos from a userspace list
- */
-int i915_validate_buffer_list(struct drm_file *file_priv,
-			      unsigned int fence_class, uint64_t data,
-			      struct drm_i915_validate_buffer *buffers,
-			      uint32_t *num_buffers)
-{
-	struct drm_i915_op_arg arg;
-	struct drm_bo_op_req *req = &arg.d.req;
-	int ret = 0;
-	unsigned buf_count = 0;
-	uint32_t buf_handle;
-	uint32_t __user *reloc_user_ptr;
-	struct drm_i915_validate_buffer *item = buffers;
-
-	do {
-		if (buf_count >= *num_buffers) {
-			DRM_ERROR("Buffer count exceeded %d\n.", *num_buffers);
-			ret = -EINVAL;
-			goto out_err;
-		}
-		item = buffers + buf_count;
-		item->buffer = NULL;
-		item->presumed_offset_correct = 0;
-
-		buffers[buf_count].buffer = NULL;
-
-		if (copy_from_user(&arg, (void __user *)(unsigned long)data, sizeof(arg))) {
-			ret = -EFAULT;
-			goto out_err;
-		}
-
-		ret = 0;
-		if (req->op != drm_bo_validate) {
-			DRM_ERROR
-			    ("Buffer object operation wasn't \"validate\".\n");
-			ret = -EINVAL;
-			goto out_err;
-		}
-		item->ret = 0;
-		item->data = (void __user *) (unsigned long) data;
-
-		buf_handle = req->bo_req.handle;
-		reloc_user_ptr = (uint32_t *)(unsigned long)arg.reloc_ptr;
-
-		if (reloc_user_ptr) {
-			ret = i915_exec_reloc(file_priv, buf_handle, reloc_user_ptr, buffers, buf_count);
-			if (ret)
-				goto out_err;
-			DRM_MEMORYBARRIER();
-		}
-
-		ret = drm_bo_handle_validate(file_priv, req->bo_req.handle,
-					     req->bo_req.flags, req->bo_req.mask,
-					     req->bo_req.hint,
-					     req->bo_req.fence_class, 0,
-					     &item->rep,
-					     &item->buffer);
-
-		if (ret) {
-			DRM_ERROR("error on handle validate %d\n", ret);
-			goto out_err;
-		}
-
-		buf_count++;
-
-		ret = i915_check_presumed(&arg, item->buffer,
-					  (uint32_t __user *)
-					  (unsigned long) data,
-					  &item->presumed_offset_correct);
-		if (ret)
-			goto out_err;
-
-		data = arg.next;
-	} while (data != 0);
-out_err:
-	*num_buffers = buf_count;
-	item->ret = (ret != -EAGAIN) ? ret : 0;
-	return ret;
-}
-
-
-/*
- * Remove all buffers from the unfenced list.
- * If the execbuffer operation was aborted, for example due to a signal,
- * this also make sure that buffers retain their original state and
- * fence pointers.
- * Copy back buffer information to user-space unless we were interrupted
- * by a signal. In which case the IOCTL must be rerun.
- */
-
-static int i915_handle_copyback(struct drm_device *dev,
-				struct drm_i915_validate_buffer *buffers,
-				unsigned int num_buffers, int ret)
-{
-	int err = ret;
-	int i;
-	struct drm_i915_op_arg arg;
-
-	if (ret)
-		drm_putback_buffer_objects(dev);
-
-	if (ret != -EAGAIN) {
-		for (i = 0; i < num_buffers; ++i) {
-			arg.handled = 1;
-			arg.d.rep.ret = buffers->ret;
-			arg.d.rep.bo_info = buffers->rep;
-			if (__copy_to_user(buffers->data, &arg, sizeof(arg)))
-				err = -EFAULT;
-			buffers++;
-		}
-	}
-	return err;
-}
-
-/*
- * Create a fence object, and if that fails, pretend that everything is
- * OK and just idle the GPU.
- */
-
-void i915_fence_or_sync(struct drm_file *file_priv,
-			uint32_t fence_flags,
-			struct drm_fence_arg *fence_arg,
-			struct drm_fence_object **fence_p)
-{
-	struct drm_device *dev = file_priv->minor->dev;
-	int ret;
-	struct drm_fence_object *fence;
-
-	ret = drm_fence_buffer_objects(dev, NULL, fence_flags,
-			 NULL, &fence);
-
-	if (ret) {
-
-		/*
-		 * Fence creation failed.
-		 * Fall back to synchronous operation and idle the engine.
-		 */
-
-		(void) i915_emit_mi_flush(dev, MI_READ_FLUSH);
-		(void) i915_quiescent(dev);
-
-		if (!(fence_flags & DRM_FENCE_FLAG_NO_USER)) {
-
-			/*
-			 * Communicate to user-space that
-			 * fence creation has failed and that
-			 * the engine is idle.
-			 */
-
-			fence_arg->handle = ~0;
-			fence_arg->error = ret;
-		}
-
-		drm_putback_buffer_objects(dev);
-		if (fence_p)
-		    *fence_p = NULL;
-		return;
-	}
-
-	if (!(fence_flags & DRM_FENCE_FLAG_NO_USER)) {
-
-		ret = drm_fence_add_user_object(file_priv, fence,
-						fence_flags &
-						DRM_FENCE_FLAG_SHAREABLE);
-		if (!ret)
-			drm_fence_fill_arg(fence, fence_arg);
-		else {
-			/*
-			 * Fence user object creation failed.
-			 * We must idle the engine here as well, as user-
-			 * space expects a fence object to wait on. Since we
-			 * have a fence object we wait for it to signal
-			 * to indicate engine "sufficiently" idle.
-			 */
-
-			(void) drm_fence_object_wait(fence, 0, 1,
-						     fence->type);
-			drm_fence_usage_deref_unlocked(&fence);
-			fence_arg->handle = ~0;
-			fence_arg->error = ret;
-		}
-	}
-
-	if (fence_p)
-		*fence_p = fence;
-	else if (fence)
-		drm_fence_usage_deref_unlocked(&fence);
-}
-
-
-static int i915_execbuffer(struct drm_device *dev, void *data,
-			   struct drm_file *file_priv)
-{
-	drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;
-	drm_i915_sarea_t *sarea_priv = (drm_i915_sarea_t *)
-		dev_priv->sarea_priv;
-	struct drm_i915_execbuffer *exec_buf = data;
-	struct drm_i915_batchbuffer *batch = &exec_buf->batch;
-	struct drm_fence_arg *fence_arg = &exec_buf->fence_arg;
-	int num_buffers;
-	int ret;
-
-	if (!dev_priv->allow_batchbuffer) {
-		DRM_ERROR("Batchbuffer ioctl disabled\n");
-		return -EINVAL;
-	}
-
-
-	if (batch->num_cliprects && DRM_VERIFYAREA_READ(batch->cliprects,
-							batch->num_cliprects *
-							sizeof(struct drm_clip_rect)))
-		return -EFAULT;
-
-	if (exec_buf->num_buffers > dev_priv->max_validate_buffers)
-		return -EINVAL;
-
-	ret = drm_bo_read_lock(&dev->bm.bm_lock);
-	if (ret)
-		return ret;
-
-	/*
-	 * The cmdbuf_mutex makes sure the validate-submit-fence
-	 * operation is atomic.
-	 */
-
-	ret = mutex_lock_interruptible(&dev_priv->cmdbuf_mutex);
-	if (ret) {
-		drm_bo_read_unlock(&dev->bm.bm_lock);
-		return -EAGAIN;
-	}
-
-	num_buffers = exec_buf->num_buffers;
-
-	if (!dev_priv->val_bufs) {
-		dev_priv->val_bufs =
-			vmalloc(sizeof(struct drm_i915_validate_buffer)*
-				dev_priv->max_validate_buffers);
-	}
-	if (!dev_priv->val_bufs) {
-		drm_bo_read_unlock(&dev->bm.bm_lock);
-		mutex_unlock(&dev_priv->cmdbuf_mutex);
-		return -ENOMEM;
-	}
-
-	/* validate buffer list + fixup relocations */
-	ret = i915_validate_buffer_list(file_priv, 0, exec_buf->ops_list,
-					dev_priv->val_bufs, &num_buffers);
-	if (ret)
-		goto out_err0;
-
-	/* make sure all previous memory operations have passed */
-	DRM_MEMORYBARRIER();
-	drm_agp_chipset_flush(dev);
-
-	/* submit buffer */
-	batch->start = dev_priv->val_bufs[num_buffers-1].buffer->offset;
-
-	DRM_DEBUG("i915 exec batchbuffer, start %x used %d cliprects %d\n",
-		  batch->start, batch->used, batch->num_cliprects);
-
-	ret = i915_dispatch_batchbuffer(dev, batch);
-	if (ret)
-		goto out_err0;
-
-	if (sarea_priv)
-		sarea_priv->last_dispatch = READ_BREADCRUMB(dev_priv);
-
-	i915_fence_or_sync(file_priv, fence_arg->flags, fence_arg, NULL);
-
-out_err0:
-
-	/* handle errors */
-	ret = i915_handle_copyback(dev, dev_priv->val_bufs, num_buffers, ret);
-	mutex_lock(&dev->struct_mutex);
-	i915_dereference_buffers_locked(dev_priv->val_bufs, num_buffers);
-	mutex_unlock(&dev->struct_mutex);
-
-	mutex_unlock(&dev_priv->cmdbuf_mutex);
-	drm_bo_read_unlock(&dev->bm.bm_lock);
-	return ret;
-}
-#endif
-
 static int i915_do_cleanup_pageflip(struct drm_device * dev)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;
diff --git a/shared-core/i915_drm.h b/shared-core/i915_drm.h
index b93df670..de463614 100644
--- a/shared-core/i915_drm.h
+++ b/shared-core/i915_drm.h
@@ -358,8 +358,22 @@ typedef struct drm_i915_hws_addr {
  * 2 - buffer handle
  * 3 - reserved (for optimisations later).
  */
+/*
+ * type 1 relocation has 4-uint32_t stride.
+ * Hangs off the first item in the op list.
+ * Performed after all valiations are done.
+ * Try to group relocs into the same relocatee together for
+ * performance reasons.
+ * 0 - offset into buffer
+ * 1 - delta to add in
+ * 2 - buffer index in op list.
+ * 3 - relocatee index in op list.
+ */
 #define I915_RELOC_TYPE_0 0
 #define I915_RELOC0_STRIDE 4
+#define I915_RELOC_TYPE_1 1
+#define I915_RELOC1_STRIDE 4
+
 
 struct drm_i915_op_arg {
 	uint64_t next;
diff --git a/shared-core/i915_drv.h b/shared-core/i915_drv.h
index 3bc39e2a..59f515c1 100644
--- a/shared-core/i915_drv.h
+++ b/shared-core/i915_drv.h
@@ -37,7 +37,7 @@
 
 #define DRIVER_NAME		"i915"
 #define DRIVER_DESC		"Intel Graphics"
-#define DRIVER_DATE		"20070209"
+#define DRIVER_DATE		"20080312"
 
 #if defined(__linux__)
 #define I915_HAVE_FENCE
@@ -61,7 +61,7 @@
  */
 #define DRIVER_MAJOR		1
 #if defined(I915_HAVE_FENCE) && defined(I915_HAVE_BUFFER)
-#define DRIVER_MINOR		12
+#define DRIVER_MINOR		13
 #else
 #define DRIVER_MINOR		6
 #endif
@@ -265,6 +265,9 @@ extern void i915_emit_breadcrumb(struct drm_device *dev);
 extern void i915_dispatch_flip(struct drm_device * dev, int pipes, int sync);
 extern int i915_emit_mi_flush(struct drm_device *dev, uint32_t flush);
 extern int i915_driver_firstopen(struct drm_device *dev);
+extern int i915_dispatch_batchbuffer(struct drm_device * dev,
+				     drm_i915_batchbuffer_t * batch);
+extern int i915_quiescent(struct drm_device *dev);
 
 /* i915_irq.c */
 extern int i915_irq_emit(struct drm_device *dev, void *data,
@@ -321,6 +324,10 @@ extern uint64_t i915_evict_flags(struct drm_buffer_object *bo);
 extern int i915_move(struct drm_buffer_object *bo, int evict,
 		int no_wait, struct drm_bo_mem_reg *new_mem);
 void i915_flush_ttm(struct drm_ttm *ttm);
+/* i915_execbuf.c */
+int i915_execbuffer(struct drm_device *dev, void *data,
+				   struct drm_file *file_priv);
+
 #endif
 
 #ifdef __linux__
@@ -416,6 +423,7 @@ extern int i915_wait_ring(struct drm_device * dev, int n, const char *caller);
 #define GFX_OP_USER_INTERRUPT		((0<<29)|(2<<23))
 #define GFX_OP_BREAKPOINT_INTERRUPT	((0<<29)|(1<<23))
 #define CMD_REPORT_HEAD			(7<<23)
+#define CMD_STORE_DWORD_IMM             ((0x20<<23) | (0x1 << 22) | 0x1)
 #define CMD_STORE_DWORD_IDX		((0x21<<23) | 0x1)
 #define CMD_OP_BATCH_BUFFER  ((0x0<<29)|(0x30<<23)|0x1)