drm/i915: add fence register management to execbuf
Adds code to set up fence registers at execbuf time on pre-965 chips as
necessary. Also fixes up a few bugs in the pre-965 tile register support
(get_order != ffs). The number of fences available to the kernel defaults
to the hw limit minus 3 (for legacy X front/back/depth), but a new parameter
allows userspace to override that as needed.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@linux.ie>
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index e1f831f..6a9e3a8 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -52,7 +52,7 @@
static int i915_gem_object_wait_rendering(struct drm_gem_object *obj);
static int i915_gem_object_bind_to_gtt(struct drm_gem_object *obj,
unsigned alignment);
-static int i915_gem_object_get_fence_reg(struct drm_gem_object *obj);
+static int i915_gem_object_get_fence_reg(struct drm_gem_object *obj, bool write);
static void i915_gem_clear_fence_reg(struct drm_gem_object *obj);
static int i915_gem_evict_something(struct drm_device *dev);
static int i915_gem_phys_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
@@ -567,6 +567,7 @@
pgoff_t page_offset;
unsigned long pfn;
int ret = 0;
+ bool write = !!(vmf->flags & FAULT_FLAG_WRITE);
/* We don't use vmf->pgoff since that has the fake offset */
page_offset = ((unsigned long)vmf->virtual_address - vma->vm_start) >>
@@ -586,7 +587,7 @@
/* Need a new fence register? */
if (obj_priv->fence_reg == I915_FENCE_REG_NONE &&
obj_priv->tiling_mode != I915_TILING_NONE) {
- ret = i915_gem_object_get_fence_reg(obj);
+ ret = i915_gem_object_get_fence_reg(obj, write);
if (ret != 0)
return VM_FAULT_SIGBUS;
}
@@ -1214,7 +1215,7 @@
/**
* Unbinds an object from the GTT aperture.
*/
-static int
+int
i915_gem_object_unbind(struct drm_gem_object *obj)
{
struct drm_device *dev = obj->dev;
@@ -1448,21 +1449,26 @@
drm_i915_private_t *dev_priv = dev->dev_private;
struct drm_i915_gem_object *obj_priv = obj->driver_private;
int regnum = obj_priv->fence_reg;
+ int tile_width;
uint32_t val;
uint32_t pitch_val;
if ((obj_priv->gtt_offset & ~I915_FENCE_START_MASK) ||
(obj_priv->gtt_offset & (obj->size - 1))) {
- WARN(1, "%s: object not 1M or size aligned\n", __func__);
+ WARN(1, "%s: object 0x%08x not 1M or size (0x%x) aligned\n",
+ __func__, obj_priv->gtt_offset, obj->size);
return;
}
- if (obj_priv->tiling_mode == I915_TILING_Y && (IS_I945G(dev) ||
- IS_I945GM(dev) ||
- IS_G33(dev)))
- pitch_val = (obj_priv->stride / 128) - 1;
+ if (obj_priv->tiling_mode == I915_TILING_Y &&
+ HAS_128_BYTE_Y_TILING(dev))
+ tile_width = 128;
else
- pitch_val = (obj_priv->stride / 512) - 1;
+ tile_width = 512;
+
+ /* Note: pitch better be a power of two tile widths */
+ pitch_val = obj_priv->stride / tile_width;
+ pitch_val = ffs(pitch_val) - 1;
val = obj_priv->gtt_offset;
if (obj_priv->tiling_mode == I915_TILING_Y)
@@ -1486,7 +1492,8 @@
if ((obj_priv->gtt_offset & ~I915_FENCE_START_MASK) ||
(obj_priv->gtt_offset & (obj->size - 1))) {
- WARN(1, "%s: object not 1M or size aligned\n", __func__);
+ WARN(1, "%s: object 0x%08x not 1M or size aligned\n",
+ __func__, obj_priv->gtt_offset);
return;
}
@@ -1506,6 +1513,7 @@
/**
* i915_gem_object_get_fence_reg - set up a fence reg for an object
* @obj: object to map through a fence reg
+ * @write: object is about to be written
*
* When mapping objects through the GTT, userspace wants to be able to write
* to them without having to worry about swizzling if the object is tiled.
@@ -1517,7 +1525,7 @@
* and tiling format.
*/
static int
-i915_gem_object_get_fence_reg(struct drm_gem_object *obj)
+i915_gem_object_get_fence_reg(struct drm_gem_object *obj, bool write)
{
struct drm_device *dev = obj->dev;
struct drm_i915_private *dev_priv = dev->dev_private;
@@ -1530,12 +1538,18 @@
WARN(1, "allocating a fence for non-tiled object?\n");
break;
case I915_TILING_X:
- WARN(obj_priv->stride & (512 - 1),
- "object is X tiled but has non-512B pitch\n");
+ if (!obj_priv->stride)
+ return -EINVAL;
+ WARN((obj_priv->stride & (512 - 1)),
+ "object 0x%08x is X tiled but has non-512B pitch\n",
+ obj_priv->gtt_offset);
break;
case I915_TILING_Y:
- WARN(obj_priv->stride & (128 - 1),
- "object is Y tiled but has non-128B pitch\n");
+ if (!obj_priv->stride)
+ return -EINVAL;
+ WARN((obj_priv->stride & (128 - 1)),
+ "object 0x%08x is Y tiled but has non-128B pitch\n",
+ obj_priv->gtt_offset);
break;
}
@@ -1637,7 +1651,7 @@
if (dev_priv->mm.suspended)
return -EBUSY;
if (alignment == 0)
- alignment = PAGE_SIZE;
+ alignment = i915_gem_get_gtt_alignment(obj);
if (alignment & (PAGE_SIZE - 1)) {
DRM_ERROR("Invalid object alignment requested %u\n", alignment);
return -EINVAL;
@@ -2658,6 +2672,14 @@
DRM_ERROR("Failure to bind: %d", ret);
return ret;
}
+ /*
+ * Pre-965 chips need a fence register set up in order to
+ * properly handle tiled surfaces.
+ */
+ if (!IS_I965G(dev) &&
+ obj_priv->fence_reg == I915_FENCE_REG_NONE &&
+ obj_priv->tiling_mode != I915_TILING_NONE)
+ i915_gem_object_get_fence_reg(obj, true);
}
obj_priv->pin_count++;
@@ -3297,7 +3319,7 @@
/* Old X drivers will take 0-2 for front, back, depth buffers */
dev_priv->fence_reg_start = 3;
- if (IS_I965G(dev))
+ if (IS_I965G(dev) || IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev))
dev_priv->num_fence_regs = 16;
else
dev_priv->num_fence_regs = 8;