1 files changed, 278 insertions, 100 deletions

diff --git a/drivers/gpu/drm/i915/i915_gem_tiling.c b/drivers/gpu/drm/i915/i915_gem_tiling.c
index 7fb4191ef93..cb150e8b433 100644
--- a/drivers/gpu/drm/i915/i915_gem_tiling.c
+++ b/drivers/gpu/drm/i915/i915_gem_tiling.c
@@ -25,9 +25,10 @@
  *
  */
 
-#include "drmP.h"
-#include "drm.h"
-#include "i915_drm.h"
+#include <linux/string.h>
+#include <linux/bitops.h>
+#include <drm/drmP.h>
+#include <drm/i915_drm.h>
 #include "i915_drv.h"
 
 /** @file i915_gem_tiling.c
@@ -86,26 +87,53 @@
 void
 i915_gem_detect_bit_6_swizzle(struct drm_device *dev)
 {
-	drm_i915_private_t *dev_priv = dev->dev_private;
+	struct drm_i915_private *dev_priv = dev->dev_private;
 	uint32_t swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN;
 	uint32_t swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN;
 
-	if (!IS_I9XX(dev)) {
+	if (IS_VALLEYVIEW(dev)) {
+		swizzle_x = I915_BIT_6_SWIZZLE_NONE;
+		swizzle_y = I915_BIT_6_SWIZZLE_NONE;
+	} else if (INTEL_INFO(dev)->gen >= 6) {
+		uint32_t dimm_c0, dimm_c1;
+		dimm_c0 = I915_READ(MAD_DIMM_C0);
+		dimm_c1 = I915_READ(MAD_DIMM_C1);
+		dimm_c0 &= MAD_DIMM_A_SIZE_MASK | MAD_DIMM_B_SIZE_MASK;
+		dimm_c1 &= MAD_DIMM_A_SIZE_MASK | MAD_DIMM_B_SIZE_MASK;
+		/* Enable swizzling when the channels are populated with
+		 * identically sized dimms. We don't need to check the 3rd
+		 * channel because no cpu with gpu attached ships in that
+		 * configuration. Also, swizzling only makes sense for 2
+		 * channels anyway. */
+		if (dimm_c0 == dimm_c1) {
+			swizzle_x = I915_BIT_6_SWIZZLE_9_10;
+			swizzle_y = I915_BIT_6_SWIZZLE_9;
+		} else {
+			swizzle_x = I915_BIT_6_SWIZZLE_NONE;
+			swizzle_y = I915_BIT_6_SWIZZLE_NONE;
+		}
+	} else if (IS_GEN5(dev)) {
+		/* On Ironlake whatever DRAM config, GPU always do
+		 * same swizzling setup.
+		 */
+		swizzle_x = I915_BIT_6_SWIZZLE_9_10;
+		swizzle_y = I915_BIT_6_SWIZZLE_9;
+	} else if (IS_GEN2(dev)) {
 		/* As far as we know, the 865 doesn't have these bit 6
 		 * swizzling issues.
 		 */
 		swizzle_x = I915_BIT_6_SWIZZLE_NONE;
 		swizzle_y = I915_BIT_6_SWIZZLE_NONE;
-	} else if ((!IS_I965G(dev) && !IS_G33(dev)) || IS_I965GM(dev) ||
-		   IS_GM45(dev)) {
+	} else if (IS_MOBILE(dev) || (IS_GEN3(dev) && !IS_G33(dev))) {
 		uint32_t dcc;
 
-		/* On 915-945 and GM965, channel interleave by the CPU is
-		 * determined by DCC.  The CPU will alternate based on bit 6
-		 * in interleaved mode, and the GPU will then also alternate
-		 * on bit 6, 9, and 10 for X, but the CPU may also optionally
-		 * alternate based on bit 17 (XOR not disabled and XOR
-		 * bit == 17).
+		/* On 9xx chipsets, channel interleave by the CPU is
+		 * determined by DCC.  For single-channel, neither the CPU
+		 * nor the GPU do swizzling.  For dual channel interleaved,
+		 * the GPU's interleave is bit 9 and 10 for X tiled, and bit
+		 * 9 for Y tiled.  The CPU's interleave is independent, and
+		 * can be based on either bit 11 (haven't seen this yet) or
+		 * bit 17 (common).
 		 */
 		dcc = I915_READ(DCC);
 		switch (dcc & DCC_ADDRESSING_MODE_MASK) {
@@ -115,21 +143,20 @@ i915_gem_detect_bit_6_swizzle(struct drm_device *dev)
 			swizzle_y = I915_BIT_6_SWIZZLE_NONE;
 			break;
 		case DCC_ADDRESSING_MODE_DUAL_CHANNEL_INTERLEAVED:
-			if (IS_I915G(dev) || IS_I915GM(dev) ||
-			    dcc & DCC_CHANNEL_XOR_DISABLE) {
+			if (dcc & DCC_CHANNEL_XOR_DISABLE) {
+				/* This is the base swizzling by the GPU for
+				 * tiled buffers.
+				 */
 				swizzle_x = I915_BIT_6_SWIZZLE_9_10;
 				swizzle_y = I915_BIT_6_SWIZZLE_9;
-			} else if ((IS_I965GM(dev) || IS_GM45(dev)) &&
-				   (dcc & DCC_CHANNEL_XOR_BIT_17) == 0) {
-				/* GM965/GM45 does either bit 11 or bit 17
-				 * swizzling.
-				 */
+			} else if ((dcc & DCC_CHANNEL_XOR_BIT_17) == 0) {
+				/* Bit 11 swizzling by the CPU in addition. */
 				swizzle_x = I915_BIT_6_SWIZZLE_9_10_11;
 				swizzle_y = I915_BIT_6_SWIZZLE_9_11;
 			} else {
-				/* Bit 17 or perhaps other swizzling */
-				swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN;
-				swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN;
+				/* Bit 17 swizzling by the CPU in addition. */
+				swizzle_x = I915_BIT_6_SWIZZLE_9_10_17;
+				swizzle_y = I915_BIT_6_SWIZZLE_9_17;
 			}
 			break;
 		}
@@ -173,35 +200,6 @@ i915_gem_detect_bit_6_swizzle(struct drm_device *dev)
 	dev_priv->mm.bit_6_swizzle_y = swizzle_y;
 }
 
-
-/**
- * Returns the size of the fence for a tiled object of the given size.
- */
-static int
-i915_get_fence_size(struct drm_device *dev, int size)
-{
-	int i;
-	int start;
-
-	if (IS_I965G(dev)) {
-		/* The 965 can have fences at any page boundary. */
-		return ALIGN(size, 4096);
-	} else {
-		/* Align the size to a power of two greater than the smallest
-		 * fence size.
-		 */
-		if (IS_I9XX(dev))
-			start = 1024 * 1024;
-		else
-			start = 512 * 1024;
-
-		for (i = start; i < size; i <<= 1)
-			;
-
-		return i;
-	}
-}
-
 /* Check pitch constriants for all chips & tiling formats */
 static bool
 i915_tiling_ok(struct drm_device *dev, int stride, int size, int tiling_mode)
@@ -212,29 +210,76 @@ i915_tiling_ok(struct drm_device *dev, int stride, int size, int tiling_mode)
 	if (tiling_mode == I915_TILING_NONE)
 		return true;
 
-	if (tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev))
+	if (IS_GEN2(dev) ||
+	    (tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev)))
 		tile_width = 128;
 	else
 		tile_width = 512;
 
+	/* check maximum stride & object size */
+	/* i965+ stores the end address of the gtt mapping in the fence
+	 * reg, so dont bother to check the size */
+	if (INTEL_INFO(dev)->gen >= 7) {
+		if (stride / 128 > GEN7_FENCE_MAX_PITCH_VAL)
+			return false;
+	} else if (INTEL_INFO(dev)->gen >= 4) {
+		if (stride / 128 > I965_FENCE_MAX_PITCH_VAL)
+			return false;
+	} else {
+		if (stride > 8192)
+			return false;
+
+		if (IS_GEN3(dev)) {
+			if (size > I830_FENCE_MAX_SIZE_VAL << 20)
+				return false;
+		} else {
+			if (size > I830_FENCE_MAX_SIZE_VAL << 19)
+				return false;
+		}
+	}
+
+	if (stride < tile_width)
+		return false;
+
 	/* 965+ just needs multiples of tile width */
-	if (IS_I965G(dev)) {
+	if (INTEL_INFO(dev)->gen >= 4) {
 		if (stride & (tile_width - 1))
 			return false;
 		return true;
 	}
 
 	/* Pre-965 needs power of two tile widths */
-	if (stride < tile_width)
+	if (stride & (stride - 1))
 		return false;
 
-	if (stride & (stride - 1))
+	return true;
+}
+
+/* Is the current GTT allocation valid for the change in tiling? */
+static bool
+i915_gem_object_fence_ok(struct drm_i915_gem_object *obj, int tiling_mode)
+{
+	u32 size;
+
+	if (tiling_mode == I915_TILING_NONE)
+		return true;
+
+	if (INTEL_INFO(obj->base.dev)->gen >= 4)
+		return true;
+
+	if (INTEL_INFO(obj->base.dev)->gen == 3) {
+		if (i915_gem_obj_ggtt_offset(obj) & ~I915_FENCE_START_MASK)
+			return false;
+	} else {
+		if (i915_gem_obj_ggtt_offset(obj) & ~I830_FENCE_START_MASK)
+			return false;
+	}
+
+	size = i915_gem_get_gtt_size(obj->base.dev, obj->base.size, tiling_mode);
+	if (i915_gem_obj_ggtt_size(obj) != size)
 		return false;
 
-	/* We don't handle the aperture area covered by the fence being bigger
-	 * than the object size.
-	 */
-	if (i915_get_fence_size(dev, size) != size)
+	if (i915_gem_obj_ggtt_offset(obj) & (size - 1))
 		return false;
 
 	return true;
@@ -246,63 +291,120 @@ i915_tiling_ok(struct drm_device *dev, int stride, int size, int tiling_mode)
  */
 int
 i915_gem_set_tiling(struct drm_device *dev, void *data,
-		   struct drm_file *file_priv)
+		   struct drm_file *file)
 {
 	struct drm_i915_gem_set_tiling *args = data;
-	drm_i915_private_t *dev_priv = dev->dev_private;
-	struct drm_gem_object *obj;
-	struct drm_i915_gem_object *obj_priv;
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct drm_i915_gem_object *obj;
+	int ret = 0;
 
-	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
-	if (obj == NULL)
-		return -EINVAL;
-	obj_priv = obj->driver_private;
+	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
+	if (&obj->base == NULL)
+		return -ENOENT;
 
-	if (!i915_tiling_ok(dev, args->stride, obj->size, args->tiling_mode)) {
-		drm_gem_object_unreference(obj);
+	if (!i915_tiling_ok(dev,
+			    args->stride, obj->base.size, args->tiling_mode)) {
+		drm_gem_object_unreference_unlocked(&obj->base);
 		return -EINVAL;
 	}
 
-	mutex_lock(&dev->struct_mutex);
+	if (i915_gem_obj_is_pinned(obj) || obj->framebuffer_references) {
+		drm_gem_object_unreference_unlocked(&obj->base);
+		return -EBUSY;
+	}
 
 	if (args->tiling_mode == I915_TILING_NONE) {
-		obj_priv->tiling_mode = I915_TILING_NONE;
 		args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
+		args->stride = 0;
 	} else {
 		if (args->tiling_mode == I915_TILING_X)
 			args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
 		else
 			args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y;
+
+		/* Hide bit 17 swizzling from the user.  This prevents old Mesa
+		 * from aborting the application on sw fallbacks to bit 17,
+		 * and we use the pread/pwrite bit17 paths to swizzle for it.
+		 * If there was a user that was relying on the swizzle
+		 * information for drm_intel_bo_map()ed reads/writes this would
+		 * break it, but we don't have any of those.
+		 */
+		if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_17)
+			args->swizzle_mode = I915_BIT_6_SWIZZLE_9;
+		if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_10_17)
+			args->swizzle_mode = I915_BIT_6_SWIZZLE_9_10;
+
 		/* If we can't handle the swizzling, make it untiled. */
 		if (args->swizzle_mode == I915_BIT_6_SWIZZLE_UNKNOWN) {
 			args->tiling_mode = I915_TILING_NONE;
 			args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
+			args->stride = 0;
 		}
 	}
-	if (args->tiling_mode != obj_priv->tiling_mode) {
-		int ret;
 
-		/* Unbind the object, as switching tiling means we're
-		 * switching the cache organization due to fencing, probably.
+	mutex_lock(&dev->struct_mutex);
+	if (args->tiling_mode != obj->tiling_mode ||
+	    args->stride != obj->stride) {
+		/* We need to rebind the object if its current allocation
+		 * no longer meets the alignment restrictions for its new
+		 * tiling mode. Otherwise we can just leave it alone, but
+		 * need to ensure that any fence register is updated before
+		 * the next fenced (either through the GTT or by the BLT unit
+		 * on older GPUs) access.
+		 *
+		 * After updating the tiling parameters, we then flag whether
+		 * we need to update an associated fence register. Note this
+		 * has to also include the unfenced register the GPU uses
+		 * whilst executing a fenced command for an untiled object.
 		 */
-		ret = i915_gem_object_unbind(obj);
-		if (ret != 0) {
-			WARN(ret != -ERESTARTSYS,
-			     "failed to unbind object for tiling switch");
-			args->tiling_mode = obj_priv->tiling_mode;
-			mutex_unlock(&dev->struct_mutex);
-			drm_gem_object_unreference(obj);
-
-			return ret;
+
+		obj->map_and_fenceable =
+			!i915_gem_obj_ggtt_bound(obj) ||
+			(i915_gem_obj_ggtt_offset(obj) +
+			 obj->base.size <= dev_priv->gtt.mappable_end &&
+			 i915_gem_object_fence_ok(obj, args->tiling_mode));
+
+		/* Rebind if we need a change of alignment */
+		if (!obj->map_and_fenceable) {
+			u32 unfenced_align =
+				i915_gem_get_gtt_alignment(dev, obj->base.size,
+							    args->tiling_mode,
+							    false);
+			if (i915_gem_obj_ggtt_offset(obj) & (unfenced_align - 1))
+				ret = i915_gem_object_ggtt_unbind(obj);
 		}
-		obj_priv->tiling_mode = args->tiling_mode;
+
+		if (ret == 0) {
+			obj->fence_dirty =
+				obj->fenced_gpu_access ||
+				obj->fence_reg != I915_FENCE_REG_NONE;
+
+			obj->tiling_mode = args->tiling_mode;
+			obj->stride = args->stride;
+
+			/* Force the fence to be reacquired for GTT access */
+			i915_gem_release_mmap(obj);
+		}
+	}
+	/* we have to maintain this existing ABI... */
+	args->stride = obj->stride;
+	args->tiling_mode = obj->tiling_mode;
+
+	/* Try to preallocate memory required to save swizzling on put-pages */
+	if (i915_gem_object_needs_bit17_swizzle(obj)) {
+		if (obj->bit_17 == NULL) {
+			obj->bit_17 = kcalloc(BITS_TO_LONGS(obj->base.size >> PAGE_SHIFT),
+					      sizeof(long), GFP_KERNEL);
+		}
+	} else {
+		kfree(obj->bit_17);
+		obj->bit_17 = NULL;
 	}
-	obj_priv->stride = args->stride;
 
-	drm_gem_object_unreference(obj);
+	drm_gem_object_unreference(&obj->base);
 	mutex_unlock(&dev->struct_mutex);
 
-	return 0;
+	return ret;
 }
 
 /**
@@ -310,22 +412,20 @@ i915_gem_set_tiling(struct drm_device *dev, void *data,
  */
 int
 i915_gem_get_tiling(struct drm_device *dev, void *data,
-		   struct drm_file *file_priv)
+		   struct drm_file *file)
 {
 	struct drm_i915_gem_get_tiling *args = data;
-	drm_i915_private_t *dev_priv = dev->dev_private;
-	struct drm_gem_object *obj;
-	struct drm_i915_gem_object *obj_priv;
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	struct drm_i915_gem_object *obj;
 
-	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
-	if (obj == NULL)
-		return -EINVAL;
-	obj_priv = obj->driver_private;
+	obj = to_intel_bo(drm_gem_object_lookup(dev, file, args->handle));
+	if (&obj->base == NULL)
+		return -ENOENT;
 
 	mutex_lock(&dev->struct_mutex);
 
-	args->tiling_mode = obj_priv->tiling_mode;
-	switch (obj_priv->tiling_mode) {
+	args->tiling_mode = obj->tiling_mode;
+	switch (obj->tiling_mode) {
 	case I915_TILING_X:
 		args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
 		break;
@@ -339,8 +439,86 @@ i915_gem_get_tiling(struct drm_device *dev, void *data,
 		DRM_ERROR("unknown tiling mode\n");
 	}
 
-	drm_gem_object_unreference(obj);
+	/* Hide bit 17 from the user -- see comment in i915_gem_set_tiling */
+	if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_17)
+		args->swizzle_mode = I915_BIT_6_SWIZZLE_9;
+	if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_10_17)
+		args->swizzle_mode = I915_BIT_6_SWIZZLE_9_10;
+
+	drm_gem_object_unreference(&obj->base);
 	mutex_unlock(&dev->struct_mutex);
 
 	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 void
+i915_gem_swizzle_page(struct page *page)
+{
+	char temp[64];
+	char *vaddr;
+	int i;
+
+	vaddr = kmap(page);
+
+	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);
+}
+
+void
+i915_gem_object_do_bit_17_swizzle(struct drm_i915_gem_object *obj)
+{
+	struct sg_page_iter sg_iter;
+	int i;
+
+	if (obj->bit_17 == NULL)
+		return;
+
+	i = 0;
+	for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, 0) {
+		struct page *page = sg_page_iter_page(&sg_iter);
+		char new_bit_17 = page_to_phys(page) >> 17;
+		if ((new_bit_17 & 0x1) !=
+		    (test_bit(i, obj->bit_17) != 0)) {
+			i915_gem_swizzle_page(page);
+			set_page_dirty(page);
+		}
+		i++;
+	}
+}
+
+void
+i915_gem_object_save_bit_17_swizzle(struct drm_i915_gem_object *obj)
+{
+	struct sg_page_iter sg_iter;
+	int page_count = obj->base.size >> PAGE_SHIFT;
+	int i;
+
+	if (obj->bit_17 == NULL) {
+		obj->bit_17 = kcalloc(BITS_TO_LONGS(page_count),
+				      sizeof(long), GFP_KERNEL);
+		if (obj->bit_17 == NULL) {
+			DRM_ERROR("Failed to allocate memory for bit 17 "
+				  "record\n");
+			return;
+		}
+	}
+
+	i = 0;
+	for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, 0) {
+		if (page_to_phys(sg_page_iter_page(&sg_iter)) & (1 << 17))
+			__set_bit(i, obj->bit_17);
+		else
+			__clear_bit(i, obj->bit_17);
+		i++;
+	}
+}