Merge branch 'modesetting-gem' of ssh://git.freedesktop.org/git/mesa/drm into modesetting-gem

linux-core/Makefile.kernel

@@ -15,7 +15,7 @@ drm-objs    := drm_auth.o drm_bufs.o drm_context.o drm_dma.o drm_drawable.o \
 		drm_hashtab.o drm_mm.o drm_compat.o \
 	        drm_fence.o drm_ttm.o drm_bo.o drm_bo_move.o \
 		drm_crtc.o drm_edid.o drm_modes.o drm_crtc_helper.o \
-		drm_regman.o drm_vm_nopage_compat.o drm_gem.o
+		drm_regman.o drm_vm_nopage_compat.o drm_gem.o drm_uncached.o
 tdfx-objs   := tdfx_drv.o
 r128-objs   := r128_drv.o r128_cce.o r128_state.o r128_irq.o
 mga-objs    := mga_drv.o mga_dma.o mga_state.o mga_warp.o mga_irq.o

linux-core/ObjectID.h

@@ -78,6 +78,10 @@
 #define ENCODER_OBJECT_ID_DP_DP501                0x1D
 #define ENCODER_OBJECT_ID_INTERNAL_UNIPHY         0x1E
 #define ENCODER_OBJECT_ID_INTERNAL_KLDSCP_LVTMA   0x1F
+#define ENCODER_OBJECT_ID_INTERNAL_UNIPHY1        0x20
+#define ENCODER_OBJECT_ID_INTERNAL_UNIPHY2        0x21
+
+#define ENCODER_OBJECT_ID_GENERAL_EXTERNAL_DVO    0xFF
 
 /****************************************************/
 /* Connector Object ID Definition                   */
@@ -118,6 +122,8 @@
 #define GRAPH_OBJECT_ENUM_ID2                     0x02
 #define GRAPH_OBJECT_ENUM_ID3                     0x03
 #define GRAPH_OBJECT_ENUM_ID4                     0x04
+#define GRAPH_OBJECT_ENUM_ID5                     0x05
+#define GRAPH_OBJECT_ENUM_ID6                     0x06
 
 /****************************************************/
 /* Graphics Object ID Bit definition                */
@@ -173,7 +179,7 @@
 #define ENCODER_SI178_ENUM_ID1                   0x2117 
 #define ENCODER_MVPU_FPGA_ENUM_ID1               0x2118
 #define ENCODER_INTERNAL_DDI_ENUM_ID1            0x2119
-#define ENCODER_VT1625_ENUM_ID1               0x211A
+#define ENCODER_VT1625_ENUM_ID1                  0x211A
 #define ENCODER_HDMI_SI1932_ENUM_ID1             0x211B
 #define ENCODER_ENCODER_DP_AN9801_ENUM_ID1       0x211C
 #define ENCODER_DP_DP501_ENUM_ID1                0x211D
@@ -323,6 +329,26 @@
                                                  GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
                                                  ENCODER_OBJECT_ID_INTERNAL_KLDSCP_LVTMA << OBJECT_ID_SHIFT)  
 
+#define ENCODER_INTERNAL_UNIPHY1_ENUM_ID1         ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
+                                                 ENCODER_OBJECT_ID_INTERNAL_UNIPHY1 << OBJECT_ID_SHIFT)
+
+#define ENCODER_INTERNAL_UNIPHY1_ENUM_ID2         ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID2 << ENUM_ID_SHIFT |\
+                                                 ENCODER_OBJECT_ID_INTERNAL_UNIPHY1 << OBJECT_ID_SHIFT)
+
+#define ENCODER_INTERNAL_UNIPHY2_ENUM_ID1         ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
+                                                 ENCODER_OBJECT_ID_INTERNAL_UNIPHY2 << OBJECT_ID_SHIFT)
+
+#define ENCODER_INTERNAL_UNIPHY2_ENUM_ID2         ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID2 << ENUM_ID_SHIFT |\
+                                                 ENCODER_OBJECT_ID_INTERNAL_UNIPHY2 << OBJECT_ID_SHIFT)
+
+#define ENCODER_GENERAL_EXTERNAL_DVO_ENUM_ID1    ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                  GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
+                                                  ENCODER_OBJECT_ID_GENERAL_EXTERNAL_DVO << OBJECT_ID_SHIFT)
+
 /****************************************************/
 /* Connector Object ID definition - Shared with BIOS */
 /****************************************************/
@@ -453,6 +479,14 @@
                                                  GRAPH_OBJECT_ENUM_ID2 << ENUM_ID_SHIFT |\
                                                  CONNECTOR_OBJECT_ID_DISPLAYPORT << OBJECT_ID_SHIFT)
 
+#define CONNECTOR_DISPLAYPORT_ENUM_ID3         ( GRAPH_OBJECT_TYPE_CONNECTOR << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID3 << ENUM_ID_SHIFT |\
+                                                 CONNECTOR_OBJECT_ID_DISPLAYPORT << OBJECT_ID_SHIFT)
+
+#define CONNECTOR_DISPLAYPORT_ENUM_ID4         ( GRAPH_OBJECT_TYPE_CONNECTOR << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID4 << ENUM_ID_SHIFT |\
+                                                 CONNECTOR_OBJECT_ID_DISPLAYPORT << OBJECT_ID_SHIFT)
+
 /****************************************************/
 /* Router Object ID definition - Shared with BIOS   */
 /****************************************************/

linux-core/drm_bo.c

@@ -1850,6 +1850,7 @@ int drm_bo_driver_finish(struct drm_device *dev)
 		__free_page(bm->dummy_read_page);
 	}
 
+	drm_uncached_fini();
 out:
 	mutex_unlock(&dev->struct_mutex);
 	return ret;
@@ -1869,6 +1870,8 @@ int drm_bo_driver_init(struct drm_device *dev)
 	struct drm_buffer_manager *bm = &dev->bm;
 	int ret = -EINVAL;
 
+	drm_uncached_init();
+
 	bm->dummy_read_page = NULL;
 	mutex_lock(&dev->struct_mutex);
 	if (!driver)

linux-core/drm_crtc.c

@@ -1423,7 +1423,7 @@ int drm_mode_setcrtc(struct drm_device *dev,
 	set.mode = mode;
 	set.connectors = connector_set;
 	set.num_connectors = crtc_req->count_connectors;
-	set.fb =fb;
+	set.fb = fb;
 	ret = crtc->funcs->set_config(&set);
 
 out:

linux-core/drm_crtc_helper.c

@@ -683,6 +683,8 @@ int drm_crtc_helper_set_config(struct drm_mode_set *set)
 		if (set->crtc->fb != set->fb)
 			set->crtc->fb = set->fb;
 		crtc_funcs->mode_set_base(set->crtc, set->x, set->y);
+		set->crtc->x = set->x;
+		set->crtc->y = set->y;
 	}
 
 	kfree(save_encoders);
@@ -802,3 +804,30 @@ int drm_helper_resume_force_mode(struct drm_device *dev)
 	return 0;
 }
 EXPORT_SYMBOL(drm_helper_resume_force_mode);
+
+void drm_helper_set_connector_dpms(struct drm_connector *connector,
+				  int dpms_mode)
+{
+	int i = 0;
+	struct drm_encoder *encoder;
+	struct drm_encoder_helper_funcs *encoder_funcs;
+	struct drm_mode_object *obj;
+
+	for (i = 0; i < DRM_CONNECTOR_MAX_ENCODER; i++) {
+		if (connector->encoder_ids[i] == 0)
+			break;
+		
+		obj = drm_mode_object_find(connector->dev,
+					   connector->encoder_ids[i],
+					   DRM_MODE_OBJECT_ENCODER);
+		if (!obj)
+			continue;
+		
+		encoder = obj_to_encoder(obj);
+		encoder_funcs = encoder->helper_private;
+		if (encoder_funcs->dpms)
+			encoder_funcs->dpms(encoder, dpms_mode);
+			
+	}
+}
+EXPORT_SYMBOL(drm_helper_set_connector_dpms);

linux-core/drm_crtc_helper.h

@@ -93,4 +93,6 @@ static inline void drm_connector_helper_add(struct drm_connector *connector, con
 }
 
 extern int drm_helper_resume_force_mode(struct drm_device *dev);
+extern void drm_helper_set_connector_dpms(struct drm_connector *connector,
+					  int dpms_mode);
 #endif

linux-core/drm_objects.h

@@ -661,6 +661,9 @@ struct drm_bo_lock {
 #define _DRM_FLAG_MEMTYPE_CMA       0x00000010	/* Can't map aperture */
 #define _DRM_FLAG_MEMTYPE_CSELECT   0x00000020	/* Select caching */
 
+#define _DRM_BM_ALLOCATOR_CACHED 0x0
+#define _DRM_BM_ALLOCATOR_UNCACHED 0x1
+
 struct drm_buffer_manager {
 	struct drm_bo_lock bm_lock;
 	struct mutex evict_mutex;
@@ -679,6 +682,7 @@ struct drm_buffer_manager {
 	unsigned long cur_pages;
 	atomic_t count;
 	struct page *dummy_read_page;
+	int allocator_type;
 };
 
 struct drm_bo_driver {
@@ -894,6 +898,15 @@ extern int drm_mem_reg_ioremap(struct drm_device *dev, struct drm_bo_mem_reg * m
 			       void **virtual);
 extern void drm_mem_reg_iounmap(struct drm_device *dev, struct drm_bo_mem_reg * mem,
 				void *virtual);
+
+/*
+ * drm_uncached.c
+ */
+extern int drm_uncached_init(void);
+extern void drm_uncached_fini(void);
+extern struct page *drm_get_uncached_page(void);
+extern void drm_put_uncached_page(struct page *page);
+
 #ifdef CONFIG_DEBUG_MUTEXES
 #define DRM_ASSERT_LOCKED(_mutex)					\
 	BUG_ON(!mutex_is_locked(_mutex) ||				\

linux-core/drm_ttm.c

@@ -120,14 +120,18 @@ static void drm_ttm_free_page_directory(struct drm_ttm *ttm)
 	ttm->pages = NULL;
 }
 
-static struct page *drm_ttm_alloc_page(void)
+static struct page *drm_ttm_alloc_page(struct drm_ttm *ttm)
 {
 	struct page *page;
 
 	if (drm_alloc_memctl(PAGE_SIZE))
 		return NULL;
 
-	page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA32);
+	if (ttm->dev->bm.allocator_type == _DRM_BM_ALLOCATOR_UNCACHED)
+		page = drm_get_uncached_page();
+	else
+		page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA32);
+
 	if (!page) {
 		drm_free_memctl(PAGE_SIZE);
 		return NULL;
@@ -149,6 +153,9 @@ static int drm_ttm_set_caching(struct drm_ttm *ttm, int noncached)
 	struct page **cur_page;
 	int do_tlbflush = 0;
 
+	if (ttm->dev->bm.allocator_type == _DRM_BM_ALLOCATOR_UNCACHED)
+		return 0;
+
 	if ((ttm->page_flags & DRM_TTM_PAGE_UNCACHED) == noncached)
 		return 0;
 
@@ -215,14 +222,18 @@ static void drm_ttm_free_alloced_pages(struct drm_ttm *ttm)
 	for (i = 0; i < ttm->num_pages; ++i) {
 		cur_page = ttm->pages + i;
 		if (*cur_page) {
+			if (ttm->dev->bm.allocator_type == _DRM_BM_ALLOCATOR_UNCACHED)
+				drm_put_uncached_page(*cur_page);
+			else {
 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15))
-			ClearPageReserved(*cur_page);
+				ClearPageReserved(*cur_page);
 #endif
-			if (page_count(*cur_page) != 1)
+				if (page_count(*cur_page) != 1)
-				DRM_ERROR("Erroneous page count. Leaking pages.\n");
+					DRM_ERROR("Erroneous page count. Leaking pages.\n");
-			if (page_mapped(*cur_page))
+				if (page_mapped(*cur_page))
-				DRM_ERROR("Erroneous map count. Leaking page mappings.\n");
+					DRM_ERROR("Erroneous map count. Leaking page mappings.\n");
-			__free_page(*cur_page);
+				__free_page(*cur_page);
+			}
 			drm_free_memctl(PAGE_SIZE);
 			--bm->cur_pages;
 		}
@@ -268,7 +279,7 @@ struct page *drm_ttm_get_page(struct drm_ttm *ttm, int index)
 	struct drm_buffer_manager *bm = &ttm->dev->bm;
 
 	while(NULL == (p = ttm->pages[index])) {
-		p = drm_ttm_alloc_page();
+		p = drm_ttm_alloc_page(ttm);
 		if (!p)
 			return NULL;
 

linux-core/drm_uncached.c

@@ -0,0 +1,138 @@
+/*
+ * Copyright (c) Red Hat Inc.
+
+ * 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 above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Dave Airlie <airlied@redhat.com>
+ */
+
+/* simple list based uncached page allocator 
+ * - Add chunks of 1MB to the allocator at a time.
+ * - Use page->lru to keep a free list
+ * - doesn't track currently in use pages
+ * 
+ *  TODO: Add shrinker support
+ */
+
+#include "drmP.h"
+#include <asm/agp.h>
+
+static struct list_head uncached_free_list;
+
+static struct mutex uncached_mutex;
+static int uncached_inited;
+static int total_uncached_pages;
+
+/* add 1MB at a time */
+#define NUM_PAGES_TO_ADD 256
+
+static void drm_uncached_page_put(struct page *page)
+{
+	unmap_page_from_agp(page);
+	put_page(page);
+	__free_page(page);
+}
+
+int drm_uncached_add_pages_locked(int num_pages)
+{
+	struct page *page;
+	int i;
+
+	DRM_DEBUG("adding uncached memory %ld\n", num_pages * PAGE_SIZE);
+	for (i = 0; i < num_pages; i++) {
+
+		page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA32);
+		if (!page) {
+			DRM_ERROR("unable to get page %d\n", i);
+			return i;
+		}
+
+		get_page(page);
+#ifdef CONFIG_X86
+		set_memory_wc((unsigned long)page_address(page), 1);
+#else
+		map_page_into_agp(page);
+#endif
+
+		list_add(&page->lru, &uncached_free_list);
+		total_uncached_pages++;
+	}
+	return i;
+}
+
+struct page *drm_get_uncached_page(void)
+{
+	struct page *page = NULL;
+	int ret;
+
+	mutex_lock(&uncached_mutex);
+	if (list_empty(&uncached_free_list)) {
+		ret = drm_uncached_add_pages_locked(NUM_PAGES_TO_ADD);
+		if (ret == 0)
+			return NULL;
+	}
+
+	page = list_first_entry(&uncached_free_list, struct page, lru);
+	list_del(&page->lru);
+
+	mutex_unlock(&uncached_mutex);
+	return page;
+}
+
+void drm_put_uncached_page(struct page *page)
+{
+	mutex_lock(&uncached_mutex);
+	list_add(&page->lru, &uncached_free_list);
+	mutex_unlock(&uncached_mutex);
+}
+
+void drm_uncached_release_all_pages(void)
+{
+	struct page *page, *tmp;
+
+	list_for_each_entry_safe(page, tmp, &uncached_free_list, lru) {
+		list_del(&page->lru);
+		drm_uncached_page_put(page);
+	}
+}
+
+int drm_uncached_init(void)
+{
+
+	if (uncached_inited)
+		return 0;
+
+	INIT_LIST_HEAD(&uncached_free_list);
+
+	mutex_init(&uncached_mutex);
+	uncached_inited = 1;
+	return 0;
+
+}
+
+void drm_uncached_fini(void)
+{
+	if (!uncached_inited)
+		return;
+	
+	uncached_inited = 0;
+	drm_uncached_release_all_pages();
+}
+

linux-core/radeon_connectors.c

@@ -77,6 +77,22 @@ static struct drm_display_mode *radeon_fp_native_mode(struct drm_encoder *encode
 	return mode;
 }
 
+int radeon_connector_set_property(struct drm_connector *connector, struct drm_property *property,
+				  uint64_t val)
+{
+	struct drm_device *dev = connector->dev;
+		
+	if (property == dev->mode_config.dpms_property) {
+		if (val > 3)
+			return -EINVAL;
+		
+		drm_helper_set_connector_dpms(connector, val);
+
+	}
+	return 0;
+}
+
+
 static int radeon_lvds_get_modes(struct drm_connector *connector)
 {
 	struct radeon_connector *radeon_connector = to_radeon_connector(connector);
@@ -146,6 +162,7 @@ struct drm_connector_funcs radeon_lvds_connector_funcs = {
 	.detect = radeon_lvds_detect,
 	.fill_modes = drm_helper_probe_single_connector_modes,
 	.destroy = radeon_connector_destroy,
+	.set_property = radeon_connector_set_property,
 };
 
 static int radeon_vga_get_modes(struct drm_connector *connector)
@@ -197,6 +214,7 @@ struct drm_connector_funcs radeon_vga_connector_funcs = {
 	.detect = radeon_vga_detect,
 	.fill_modes = drm_helper_probe_single_connector_modes,
 	.destroy = radeon_connector_destroy,
+	.set_property = radeon_connector_set_property,
 };
 
 
@@ -289,6 +307,7 @@ struct drm_connector_helper_funcs radeon_dvi_connector_helper_funcs = {
 struct drm_connector_funcs radeon_dvi_connector_funcs = {
 	.detect = radeon_dvi_detect,
 	.fill_modes = drm_helper_probe_single_connector_modes,
+	.set_property = radeon_connector_set_property,
 	.destroy = radeon_connector_destroy,
 };
 

linux-core/radeon_cursor.c

@@ -204,6 +204,10 @@ int radeon_crtc_cursor_move(struct drm_crtc *crtc,
 
 	radeon_lock_cursor(crtc, true);
 	if (radeon_is_avivo(dev_priv)) {
+		/* avivo cursor are offset into the total surface */
+		x += crtc->x;
+		y += crtc->y;
+		DRM_DEBUG("x %d y %d c->x %d c->y %d\n", x, y, crtc->x, crtc->y);
 		RADEON_WRITE(AVIVO_D1CUR_POSITION + radeon_crtc->crtc_offset,
 			     ((xorigin ? 0: x) << 16) |
 			     (yorigin ? 0 : y));

linux-core/radeon_fb.c

@@ -707,6 +707,7 @@ int radeonfb_create(struct drm_device *dev, uint32_t fb_width, uint32_t fb_heigh
 		   uint32_t surface_width, uint32_t surface_height,
 		   struct radeon_framebuffer **radeon_fb_p)
 {
+	struct drm_radeon_private *dev_priv = dev->dev_private;
 	struct fb_info *info;
 	struct radeonfb_par *par;
 	struct drm_framebuffer *fb;
@@ -743,6 +744,8 @@ int radeonfb_create(struct drm_device *dev, uint32_t fb_width, uint32_t fb_heigh
 		goto out_unref;
 	}
 
+	dev_priv->mm.vram_visible -= aligned_size;
+
 	mutex_lock(&dev->struct_mutex);
 	fb = radeon_framebuffer_create(dev, &mode_cmd, fbo);
 	if (!fb) {
@@ -1136,6 +1139,7 @@ EXPORT_SYMBOL(radeonfb_probe);
 
 int radeonfb_remove(struct drm_device *dev, struct drm_framebuffer *fb)
 {
+	struct drm_radeon_private *dev_priv = dev->dev_private;
 	struct fb_info *info;
 	struct radeon_framebuffer *radeon_fb = to_radeon_framebuffer(fb);
 
@@ -1147,6 +1151,7 @@ int radeonfb_remove(struct drm_device *dev, struct drm_framebuffer *fb)
 	if (info) {
 		unregister_framebuffer(info);
 		drm_bo_kunmap(&radeon_fb->kmap_obj);
+		dev_priv->mm.vram_visible += radeon_fb->obj->size;
 		mutex_lock(&dev->struct_mutex);
 		drm_gem_object_unreference(radeon_fb->obj);
 		radeon_fb->obj = NULL;

linux-core/radeon_fence.c

@@ -57,14 +57,15 @@ static void radeon_fence_poll(struct drm_device *dev, uint32_t fence_class,
 {
 	struct drm_radeon_private *dev_priv = (struct drm_radeon_private *) dev->dev_private;
 	uint32_t sequence;
-        if (waiting_types & DRM_FENCE_TYPE_EXE) {
 
-                sequence = READ_BREADCRUMB(dev_priv);
+	sequence = RADEON_READ(RADEON_SCRATCH_REG3);
+	/* this used to be READ_BREADCRUMB(dev_priv); but it caused
+	 * a race somewhere in the fencing irq
+	 */
 
-		DRM_DEBUG("polling %d\n", sequence);
+	DRM_DEBUG("polling %d\n", sequence);
-                drm_fence_handler(dev, 0, sequence,
+	drm_fence_handler(dev, 0, sequence,
-                                  DRM_FENCE_TYPE_EXE, 0);
+       			  DRM_FENCE_TYPE_EXE, 0);
-        }
 }
 
 void radeon_fence_handler(struct drm_device * dev)

linux-core/radeon_gem.c

@@ -68,7 +68,7 @@ int radeon_gem_info_ioctl(struct drm_device *dev, void *data,
 	args->vram_visible = dev_priv->mm.vram_visible;
 
 	args->gart_start = dev_priv->mm.gart_start;
-	args->gart_size = dev_priv->mm.gart_size;
+	args->gart_size = dev_priv->mm.gart_useable;
 
 	return 0;
 }
@@ -586,6 +586,9 @@ static int radeon_gart_init(struct drm_device *dev)
 		if (ret)
 			return -EINVAL;
 
+		/* subtract from VRAM value reporting to userspace */
+		dev_priv->mm.vram_visible -= RADEON_PCIGART_TABLE_SIZE;
+
 		dev_priv->mm.pcie_table_backup = kzalloc(RADEON_PCIGART_TABLE_SIZE, GFP_KERNEL);
 		if (!dev_priv->mm.pcie_table_backup)
 			return -EINVAL;
@@ -677,6 +680,8 @@ int radeon_alloc_gart_objects(struct drm_device *dev)
 		  dev_priv->mm.ring.bo, dev_priv->mm.ring.bo->offset, dev_priv->mm.ring.kmap.virtual,
 		  dev_priv->mm.ring_read.bo, dev_priv->mm.ring_read.bo->offset, dev_priv->mm.ring_read.kmap.virtual);
 
+	dev_priv->mm.gart_useable -= RADEON_DEFAULT_RING_SIZE + PAGE_SIZE;
+
 	/* init the indirect buffers */
 	radeon_gem_ib_init(dev);
 	radeon_gem_dma_bufs_init(dev);
@@ -963,6 +968,9 @@ int radeon_gem_mm_init(struct drm_device *dev)
 	/* init TTM underneath */
 	drm_bo_driver_init(dev);
 
+	/* use the uncached allocator */
+	dev->bm.allocator_type = _DRM_BM_ALLOCATOR_UNCACHED;
+
 	/* size the mappable VRAM memory for now */
 	radeon_vram_setup(dev);
 	
@@ -983,6 +991,7 @@ int radeon_gem_mm_init(struct drm_device *dev)
 
 	dev_priv->mm.gart_size = (32 * 1024 * 1024);
 	dev_priv->mm.gart_start = 0;
+	dev_priv->mm.gart_useable = dev_priv->mm.gart_size;
 	ret = radeon_gart_init(dev);
 	if (ret)
 		return -EINVAL;
@@ -1287,6 +1296,7 @@ static int radeon_gem_ib_init(struct drm_device *dev)
 			goto free_all;
 	}
 
+	dev_priv->mm.gart_useable -= RADEON_IB_SIZE * RADEON_NUM_IB;
 	dev_priv->ib_alloc_bitmap = 0;
 
 	dev_priv->cs.ib_get = radeon_gem_ib_get;
@@ -1523,6 +1533,7 @@ static int radeon_gem_dma_bufs_init(struct drm_device *dev)
 		DRM_ERROR("Failed to mmap DMA buffers\n");
 		return -ENOMEM;
 	}
+	dev_priv->mm.gart_useable -= size;
 	DRM_DEBUG("\n");
 	radeon_gem_addbufs(dev);
 

shared-core/radeon_cp.c

@@ -1313,9 +1313,7 @@ static int radeon_do_init_cp(struct drm_device *dev, drm_radeon_init_t *init,
 	dev_priv->ring.rptr_update = /* init->rptr_update */ 4096;
 	dev_priv->ring.rptr_update_l2qw = drm_order( /* init->rptr_update */ 4096 / 8);
 
-	dev_priv->ring.fetch_size = /* init->fetch_size */ 32;
+	dev_priv->ring.fetch_size_l2ow = 2;
-	dev_priv->ring.fetch_size_l2ow = drm_order( /* init->fetch_size */ 32 / 16);
-
 	dev_priv->ring.tail_mask = (dev_priv->ring.size / sizeof(u32)) - 1;
 
 	dev_priv->ring.high_mark = RADEON_RING_HIGH_MARK;
@@ -2513,8 +2511,7 @@ int radeon_modeset_cp_init(struct drm_device *dev)
 	dev_priv->ring.size_l2qw = drm_order(dev_priv->ring.size / 8);
 	dev_priv->ring.rptr_update = 4096;
 	dev_priv->ring.rptr_update_l2qw = drm_order(4096 / 8);
-	dev_priv->ring.fetch_size = 32;
+	dev_priv->ring.fetch_size_l2ow = 2; /* do what tcore does */
-	dev_priv->ring.fetch_size_l2ow = drm_order(32 / 16);
 	dev_priv->ring.tail_mask = (dev_priv->ring.size / sizeof(u32)) - 1;
 	dev_priv->ring.high_mark = RADEON_RING_HIGH_MARK;
 
@@ -2522,6 +2519,9 @@ int radeon_modeset_cp_init(struct drm_device *dev)
 
 	r300_init_reg_flags(dev);
 
+	/* turn off HDP read cache for now */
+	RADEON_WRITE(RADEON_HOST_PATH_CNTL, RADEON_READ(RADEON_HOST_PATH_CNTL) | RADEON_HP_LIN_RD_CACHE_DIS);
+	      
 #if __OS_HAS_AGP
 	if (dev_priv->flags & RADEON_IS_AGP)
 		radeon_modeset_agp_init(dev);
@@ -2841,3 +2841,35 @@ void radeon_gart_flush(struct drm_device *dev)
 	}
 	
 }
+
+void radeon_commit_ring(drm_radeon_private_t *dev_priv)
+{
+	int i;
+	u32 *ring;
+	int tail_aligned;
+
+	/* check if the ring is padded out to 16-dword alignment */
+	
+	tail_aligned = dev_priv->ring.tail & 0xf;
+	if (tail_aligned) {
+		int num_p2 = 16 - tail_aligned;
+
+		ring = dev_priv->ring.start;
+		/* pad with some CP_PACKET2 */
+		for (i = 0; i < num_p2; i++)
+			ring[dev_priv->ring.tail + i] = CP_PACKET2();
+	
+		dev_priv->ring.tail += i;
+
+		dev_priv->ring.space -= num_p2 * sizeof(u32);
+	}
+
+	dev_priv->ring.tail &= dev_priv->ring.tail_mask;
+		
+	DRM_MEMORYBARRIER();
+	GET_RING_HEAD( dev_priv );
+
+	RADEON_WRITE( RADEON_CP_RB_WPTR, dev_priv->ring.tail );
+	/* read from PCI bus to ensure correct posting */
+	RADEON_READ( RADEON_CP_RB_RPTR );
+}

shared-core/radeon_drv.h

@@ -219,7 +219,6 @@ typedef struct drm_radeon_ring_buffer {
 	int rptr_update; /* Double Words */
 	int rptr_update_l2qw; /* log2 Quad Words */
 
-	int fetch_size; /* Double Words */
 	int fetch_size_l2ow; /* log2 Oct Words */
 
 	u32 tail;
@@ -275,6 +274,8 @@ struct radeon_mm_info {
 	uint64_t gart_start;
 	uint64_t gart_size;
 
+	uint64_t gart_useable;
+
 	void *pcie_table_backup;
 	
 	struct radeon_mm_obj pcie_table;
@@ -792,8 +793,10 @@ int radeon_resume(struct drm_device *dev);
 #       define R500_DISPLAY_INT_STATUS          (1 << 0)
 
 #define RADEON_HOST_PATH_CNTL               0x0130
-#       define RADEON_HDP_SOFT_RESET        (1 << 26)
 #       define RADEON_HDP_APER_CNTL         (1 << 23)
+#       define RADEON_HP_LIN_RD_CACHE_DIS   (1 << 24)
+#       define RADEON_HDP_SOFT_RESET        (1 << 26)
+#       define RADEON_HDP_READ_BUFFER_INVALIDATED (1 << 27)
 
 #define RADEON_NB_TOM                       0x15c
 
@@ -1515,15 +1518,16 @@ do {									\
 
 #define RADEON_VERBOSE	0
 
-#define RING_LOCALS	int write, _nr; unsigned int mask; u32 *ring;
+#define RING_LOCALS	int write, _nr, _align_nr; unsigned int mask; u32 *ring;
 
 #define BEGIN_RING( n ) do {						\
 	if ( RADEON_VERBOSE ) {						\
 		DRM_INFO( "BEGIN_RING( %d )\n", (n));			\
 	}								\
-	if ( dev_priv->ring.space <= (n) * sizeof(u32) ) {		\
+	_align_nr = (n + 0xf) & ~0xf;					\
+	if (dev_priv->ring.space <= (_align_nr * sizeof(u32))) {	\
 		COMMIT_RING();						\
-		radeon_wait_ring( dev_priv, (n) * sizeof(u32) );	\
+		radeon_wait_ring(dev_priv, _align_nr * sizeof(u32));	\
 	}								\
 	_nr = n; dev_priv->ring.space -= (n) * sizeof(u32);		\
 	ring = dev_priv->ring.start;					\
@@ -1540,19 +1544,14 @@ do {									\
 		DRM_ERROR(						\
 			"ADVANCE_RING(): mismatch: nr: %x write: %x line: %d\n",	\
 			((dev_priv->ring.tail + _nr) & mask),		\
-			write, __LINE__);						\
+			write, __LINE__);				\
 	} else								\
 		dev_priv->ring.tail = write;				\
 } while (0)
 
 #define COMMIT_RING() do {						\
-	/* Flush writes to ring */					\
+		radeon_commit_ring(dev_priv);				\
-	DRM_MEMORYBARRIER();						\
+	} while(0)
-	GET_RING_HEAD( dev_priv );					\
-	RADEON_WRITE( RADEON_CP_RB_WPTR, dev_priv->ring.tail );		\
-	/* read from PCI bus to ensure correct posting */		\
-	RADEON_READ( RADEON_CP_RB_RPTR );				\
-} while (0)
 
 #define OUT_RING( x ) do {						\
 	if ( RADEON_VERBOSE ) {						\
@@ -1731,6 +1730,8 @@ extern void radeon_gem_proc_cleanup(struct drm_minor *minor);
 #define MARK_CHECK_OFFSET	2
 #define MARK_CHECK_SCISSOR	3
 
+extern void radeon_commit_ring(drm_radeon_private_t *dev_priv);
+
 extern int r300_check_range(unsigned reg, int count);
 extern int r300_get_reg_flags(unsigned reg);
 #endif				/* __RADEON_DRV_H__ */