nouveau: allocate drm-use vram buffers from end of vram.

This avoids seeing garbage from engine setup etc before X gets around
to pointing the CRTCs at a new scanout buffer.  Not actually a noticable
problem before G80 as PRAMIN is forced to the end of VRAM by the hardware
already.

5 files changed, 42 insertions, 26 deletions

diff --git a/shared-core/nouveau_drm.h b/shared-core/nouveau_drm.h
index cf762052..bbb51bc4 100644
--- a/shared-core/nouveau_drm.h
+++ b/shared-core/nouveau_drm.h
@@ -88,9 +88,11 @@ struct drm_nouveau_gpuobj_free {
 #define NOUVEAU_MEM_INSTANCE		0x00000200 /* internal */
 #define NOUVEAU_MEM_NOTIFIER            0x00000400 /* internal */
 #define NOUVEAU_MEM_NOVM		0x00000800 /* internal */
+#define NOUVEAU_MEM_USER		0x00001000 /* internal */
 #define NOUVEAU_MEM_INTERNAL (NOUVEAU_MEM_INSTANCE | \
 			      NOUVEAU_MEM_NOTIFIER | \
-			      NOUVEAU_MEM_NOVM)
+			      NOUVEAU_MEM_NOVM | \
+			      NOUVEAU_MEM_USER)
 
 struct drm_nouveau_mem_alloc {
 	int flags;
diff --git a/shared-core/nouveau_drv.h b/shared-core/nouveau_drv.h
index a51e552c..1aaf1d74 100644
--- a/shared-core/nouveau_drv.h
+++ b/shared-core/nouveau_drv.h
@@ -350,7 +350,7 @@ extern int  nouveau_mem_init_heap(struct mem_block **, uint64_t start,
 				 uint64_t size);
 extern struct mem_block *nouveau_mem_alloc_block(struct mem_block *,
 						 uint64_t size, int align2,
-						 struct drm_file *);
+						 struct drm_file *, int tail);
 extern void nouveau_mem_takedown(struct mem_block **heap);
 extern void nouveau_mem_free_block(struct mem_block *);
 extern uint64_t nouveau_mem_fb_amount(struct drm_device *);
diff --git a/shared-core/nouveau_mem.c b/shared-core/nouveau_mem.c
index 2cf8807d..5511e5b6 100644
--- a/shared-core/nouveau_mem.c
+++ b/shared-core/nouveau_mem.c
@@ -35,8 +35,9 @@
 #include "drm_sarea.h"
 #include "nouveau_drv.h"
 
-static struct mem_block *split_block(struct mem_block *p, uint64_t start, uint64_t size,
-		struct drm_file *file_priv)
+static struct mem_block *
+split_block(struct mem_block *p, uint64_t start, uint64_t size,
+	    struct drm_file *file_priv)
 {
 	/* Maybe cut off the start of an existing block */
 	if (start > p->start) {
@@ -77,10 +78,9 @@ out:
 	return p;
 }
 
-struct mem_block *nouveau_mem_alloc_block(struct mem_block *heap,
-					  uint64_t size,
-					  int align2,
-					  struct drm_file *file_priv)
+struct mem_block *
+nouveau_mem_alloc_block(struct mem_block *heap, uint64_t size,
+			int align2, struct drm_file *file_priv, int tail)
 {
 	struct mem_block *p;
 	uint64_t mask = (1 << align2) - 1;
@@ -88,10 +88,22 @@ struct mem_block *nouveau_mem_alloc_block(struct mem_block *heap,
 	if (!heap)
 		return NULL;
 
-	list_for_each(p, heap) {
-		uint64_t start = (p->start + mask) & ~mask;
-		if (p->file_priv == 0 && start + size <= p->start + p->size)
-			return split_block(p, start, size, file_priv);
+	if (tail) {
+		list_for_each_prev(p, heap) {
+			uint64_t start = ((p->start + p->size) - size) & ~mask;
+
+			if (p->file_priv == 0 && start >= p->start &&
+			    start + size <= p->start + p->size)
+				return split_block(p, start, size, file_priv);
+		}
+	} else {
+		list_for_each(p, heap) {
+			uint64_t start = (p->start + mask) & ~mask;
+
+			if (p->file_priv == 0 &&
+			    start + size <= p->start + p->size)
+				return split_block(p, start, size, file_priv);
+		}
 	}
 
 	return NULL;
@@ -563,13 +575,13 @@ int nouveau_mem_init(struct drm_device *dev)
 	return 0;
 }
 
-struct mem_block* nouveau_mem_alloc(struct drm_device *dev, int alignment,
-				    uint64_t size, int flags,
-				    struct drm_file *file_priv)
+struct mem_block *
+nouveau_mem_alloc(struct drm_device *dev, int alignment, uint64_t size,
+		  int flags, struct drm_file *file_priv)
 {
-	struct mem_block *block;
-	int type;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct mem_block *block;
+	int type, tail = !(flags & NOUVEAU_MEM_USER);
 
 	/*
 	 * Make things easier on ourselves: all allocations are page-aligned.
@@ -600,14 +612,14 @@ struct mem_block* nouveau_mem_alloc(struct drm_device *dev, int alignment,
 #define NOUVEAU_MEM_ALLOC_AGP {\
 		type=NOUVEAU_MEM_AGP;\
                 block = nouveau_mem_alloc_block(dev_priv->agp_heap, size,\
-                                                alignment, file_priv); \
+                                                alignment, file_priv, tail); \
                 if (block) goto alloc_ok;\
 	        }
 
 #define NOUVEAU_MEM_ALLOC_PCI {\
                 type = NOUVEAU_MEM_PCI;\
                 block = nouveau_mem_alloc_block(dev_priv->pci_heap, size, \
-						alignment, file_priv); \
+						alignment, file_priv, tail); \
                 if ( block ) goto alloc_ok;\
 	        }
 
@@ -616,11 +628,11 @@ struct mem_block* nouveau_mem_alloc(struct drm_device *dev, int alignment,
                 if (!(flags&NOUVEAU_MEM_MAPPED)) {\
                         block = nouveau_mem_alloc_block(dev_priv->fb_nomap_heap,\
                                                         size, alignment, \
-							file_priv); \
+							file_priv, tail); \
                         if (block) goto alloc_ok;\
                 }\
                 block = nouveau_mem_alloc_block(dev_priv->fb_heap, size,\
-                                                alignment, file_priv);\
+                                                alignment, file_priv, tail);\
                 if (block) goto alloc_ok;\
 	        }
 
@@ -738,7 +750,9 @@ out_free:
  * Ioctls
  */
 
-int nouveau_ioctl_mem_alloc(struct drm_device *dev, void *data, struct drm_file *file_priv)
+int
+nouveau_ioctl_mem_alloc(struct drm_device *dev, void *data,
+			struct drm_file *file_priv)
 {
 	struct drm_nouveau_mem_alloc *alloc = data;
 	struct mem_block *block;
@@ -748,8 +762,8 @@ int nouveau_ioctl_mem_alloc(struct drm_device *dev, void *data, struct drm_file
 	if (alloc->flags & NOUVEAU_MEM_INTERNAL)
 		return -EINVAL;
 
-	block=nouveau_mem_alloc(dev, alloc->alignment, alloc->size,
-				alloc->flags, file_priv);
+	block = nouveau_mem_alloc(dev, alloc->alignment, alloc->size,
+				  alloc->flags | NOUVEAU_MEM_USER, file_priv);
 	if (!block)
 		return -ENOMEM;
 	alloc->map_handle=block->map_handle;
diff --git a/shared-core/nouveau_notifier.c b/shared-core/nouveau_notifier.c
index 82c8ab7d..edece4da 100644
--- a/shared-core/nouveau_notifier.c
+++ b/shared-core/nouveau_notifier.c
@@ -94,7 +94,7 @@ nouveau_notifier_alloc(struct nouveau_channel *chan, uint32_t handle,
 	}
 
 	mem = nouveau_mem_alloc_block(chan->notifier_heap, count*32, 0,
-				      (struct drm_file *)-2);
+				      (struct drm_file *)-2, 0);
 	if (!mem) {
 		DRM_ERROR("Channel %d notifier block full\n", chan->id);
 		return -ENOMEM;
diff --git a/shared-core/nouveau_object.c b/shared-core/nouveau_object.c
index 09f9027a..5664bfc8 100644
--- a/shared-core/nouveau_object.c
+++ b/shared-core/nouveau_object.c
@@ -248,7 +248,7 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 	/* Allocate a chunk of the PRAMIN aperture */
 	gpuobj->im_pramin = nouveau_mem_alloc_block(pramin, size,
 						    drm_order(align),
-						    (struct drm_file *)-2);
+						    (struct drm_file *)-2, 0);
 	if (!gpuobj->im_pramin) {
 		nouveau_gpuobj_del(dev, &gpuobj);
 		return -ENOMEM;