Menu
▾
▴
[PATCH] drm/i915: Allow tiling of objects with bit 17 swizzling by the CPU.
|
From: Eric A. <er...@an...> - 2009-04-04 00:03:46
|
Save the bit 17 state of the pages when freeing the page list, and
reswizzle them if necessary when rebinding the pages. This exposes two new
swizzle enums to userland, which would have to be respected for correct
operation of software fallbacks on these objects. Given that users
complain that 3D "acceleration" is slower than software rendering without
tiling, it seems that reenabling things even if some operations aren't correct
with old userland is worth the cost.
Userland won't be able to respect the two new tiling enums with
drm_intel_bo_map or pread/pwrite. However, GTT maps are available and should
prove sufficient.
Signed-off-by: Eric Anholt <er...@an...>
---
I've tested this, and it does take swapping of tiled objects on an
A17-affected machine from failing to working. I haven't written the
Mesa patch to move to GTT mapping yet, so this would regress Mesa and I
won't submit it for inclusion until I do that.
Or, if there's objection to breaking backwards compat in this way, I
could write pread/pwrite compatibility paths that do the swizzling as
well and just tell the user that it's got the swizzling without A17,
which would cover Mesa SW fallbacks. But either way we'll want to
change Mesa's behavior on at least A17 machines.
drivers/gpu/drm/i915/i915_drv.h | 5 ++
drivers/gpu/drm/i915/i915_gem.c | 8 +++
drivers/gpu/drm/i915/i915_gem_tiling.c | 85 +++++++++++++++++++++++++++++++-
include/drm/i915_drm.h | 3 +
4 files changed, 99 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index efcd610..f968fe8 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -446,6 +446,9 @@ struct drm_i915_gem_object {
uint32_t tiling_mode;
uint32_t stride;
+ /** Record of address bit 17 of each page at last unbind. */
+ char *bit_17;
+
/** AGP mapping type (AGP_USER_MEMORY or AGP_USER_CACHED_MEMORY */
uint32_t agp_type;
@@ -640,6 +643,8 @@ void i915_gem_object_put_pages(struct drm_gem_object *obj);
/* i915_gem_tiling.c */
void i915_gem_detect_bit_6_swizzle(struct drm_device *dev);
+void i915_gem_object_do_bit_17_swizzle(struct drm_gem_object *obj);
+void i915_gem_object_save_bit_17_swizzle(struct drm_gem_object *obj);
/* i915_gem_debug.c */
void i915_gem_dump_object(struct drm_gem_object *obj, int len,
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 33ab07b..14db0e4 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1295,6 +1295,9 @@ i915_gem_object_put_pages(struct drm_gem_object *obj)
if (--obj_priv->pages_refcount != 0)
return;
+ if (obj_priv->tiling_mode != I915_TILING_NONE)
+ i915_gem_object_save_bit_17_swizzle(obj);
+
for (i = 0; i < page_count; i++)
if (obj_priv->pages[i] != NULL) {
if (obj_priv->dirty)
@@ -1920,6 +1923,10 @@ i915_gem_object_get_pages(struct drm_gem_object *obj)
}
obj_priv->pages[i] = page;
}
+
+ if (obj_priv->tiling_mode != I915_TILING_NONE)
+ i915_gem_object_do_bit_17_swizzle(obj);
+
return 0;
}
@@ -3591,6 +3598,7 @@ void i915_gem_free_object(struct drm_gem_object *obj)
i915_gem_free_mmap_offset(obj);
drm_free(obj_priv->page_cpu_valid, 1, DRM_MEM_DRIVER);
+ kfree(obj_priv->bit_17);
drm_free(obj->driver_private, 1, DRM_MEM_DRIVER);
}
diff --git a/drivers/gpu/drm/i915/i915_gem_tiling.c b/drivers/gpu/drm/i915/i915_gem_tiling.c
index 6be3f92..5544102 100644
--- a/drivers/gpu/drm/i915/i915_gem_tiling.c
+++ b/drivers/gpu/drm/i915/i915_gem_tiling.c
@@ -25,6 +25,7 @@
*
*/
+#include "linux/string.h"
#include "drmP.h"
#include "drm.h"
#include "i915_drm.h"
@@ -127,8 +128,8 @@ i915_gem_detect_bit_6_swizzle(struct drm_device *dev)
swizzle_y = I915_BIT_6_SWIZZLE_9_11;
} else {
/* Bit 17 swizzling by the CPU in addition. */
- swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN;
- swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN;
+ swizzle_x = I915_BIT_6_SWIZZLE_9_10_17;
+ swizzle_y = I915_BIT_6_SWIZZLE_9_17;
}
break;
}
@@ -359,3 +360,83 @@ i915_gem_get_tiling(struct drm_device *dev, void *data,
return 0;
}
+
+/**
+ * Swap every 64 bytes of this page around, to account for it having a new
+ * bit 17 of its physical address and therefore being interpreted differently
+ * by the GPU.
+ */
+static int
+i915_gem_swizzle_page(struct page *page)
+{
+ char *vaddr;
+ int i;
+ char temp[64];
+
+ vaddr = kmap(page);
+ if (vaddr == NULL)
+ return -ENOMEM;
+
+ for (i = 0; i < PAGE_SIZE; i += 128) {
+ memcpy(temp, &vaddr[i], 64);
+ memcpy(&vaddr[i], &vaddr[i + 64], 64);
+ memcpy(&vaddr[i + 64], temp, 64);
+ }
+
+ kunmap(page);
+
+ return 0;
+}
+
+void
+i915_gem_object_do_bit_17_swizzle(struct drm_gem_object *obj)
+{
+ struct drm_device *dev = obj->dev;
+ drm_i915_private_t *dev_priv = dev->dev_private;
+ struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ int page_count = obj->size >> PAGE_SHIFT;
+ int i;
+
+ if (dev_priv->mm.bit_6_swizzle_x != I915_BIT_6_SWIZZLE_9_10_17)
+ return;
+
+ if (obj_priv->bit_17 == NULL)
+ return;
+
+ for (i = 0; i < page_count; i++) {
+ char new_bit_17 = page_to_phys(obj_priv->pages[i]) >> 17;
+ if ((new_bit_17 & 0x1) != (obj_priv->bit_17[i] & 0x1)) {
+ int ret = i915_gem_swizzle_page(obj_priv->pages[i]);
+ if (ret != 0) {
+ DRM_ERROR("Failed to swizzle page\n");
+ return;
+ }
+ }
+ }
+}
+
+void
+i915_gem_object_save_bit_17_swizzle(struct drm_gem_object *obj)
+{
+ struct drm_device *dev = obj->dev;
+ drm_i915_private_t *dev_priv = dev->dev_private;
+ struct drm_i915_gem_object *obj_priv = obj->driver_private;
+ int page_count = obj->size >> PAGE_SHIFT;
+ int i;
+
+ if (dev_priv->mm.bit_6_swizzle_x != I915_BIT_6_SWIZZLE_9_10_17)
+ return;
+
+ if (obj_priv->bit_17 == NULL) {
+ obj_priv->bit_17 = kmalloc(page_count, GFP_KERNEL);
+ if (obj_priv->bit_17 == NULL) {
+ DRM_ERROR("Failed to allocate memory for bit 17 "
+ "record\n");
+ return;
+ }
+ }
+
+ for (i = 0; i < page_count; i++)
+ obj_priv->bit_17[i] =
+ page_to_phys(obj_priv->pages[i]) >> 17;
+}
diff --git a/include/drm/i915_drm.h b/include/drm/i915_drm.h
index 67e3353..95962fa 100644
--- a/include/drm/i915_drm.h
+++ b/include/drm/i915_drm.h
@@ -594,6 +594,9 @@ struct drm_i915_gem_busy {
#define I915_BIT_6_SWIZZLE_9_10_11 4
/* Not seen by userland */
#define I915_BIT_6_SWIZZLE_UNKNOWN 5
+/* Seen by userland. */
+#define I915_BIT_6_SWIZZLE_9_17 6
+#define I915_BIT_6_SWIZZLE_9_10_17 7
struct drm_i915_gem_set_tiling {
/** Handle of the buffer to have its tiling state updated */
--
1.6.2.1
--
Eric Anholt
er...@an... eri...@in...
|