2 * Copyright © 2008 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Eric Anholt <eric@anholt.net>
33 /** @file i915_gem_tiling.c
35 * Support for managing tiling state of buffer objects.
37 * The idea behind tiling is to increase cache hit rates by rearranging
38 * pixel data so that a group of pixel accesses are in the same cacheline.
39 * Performance improvement from doing this on the back/depth buffer are on
42 * Intel architectures make this somewhat more complicated, though, by
43 * adjustments made to addressing of data when the memory is in interleaved
44 * mode (matched pairs of DIMMS) to improve memory bandwidth.
45 * For interleaved memory, the CPU sends every sequential 64 bytes
46 * to an alternate memory channel so it can get the bandwidth from both.
48 * The GPU also rearranges its accesses for increased bandwidth to interleaved
49 * memory, and it matches what the CPU does for non-tiled. However, when tiled
50 * it does it a little differently, since one walks addresses not just in the
51 * X direction but also Y. So, along with alternating channels when bit
52 * 6 of the address flips, it also alternates when other bits flip -- Bits 9
53 * (every 512 bytes, an X tile scanline) and 10 (every two X tile scanlines)
54 * are common to both the 915 and 965-class hardware.
56 * The CPU also sometimes XORs in higher bits as well, to improve
57 * bandwidth doing strided access like we do so frequently in graphics. This
58 * is called "Channel XOR Randomization" in the MCH documentation. The result
59 * is that the CPU is XORing in either bit 11 or bit 17 to bit 6 of its address
62 * All of this bit 6 XORing has an effect on our memory management,
63 * as we need to make sure that the 3d driver can correctly address object
66 * If we don't have interleaved memory, all tiling is safe and no swizzling is
69 * When bit 17 is XORed in, we simply refuse to tile at all. Bit
70 * 17 is not just a page offset, so as we page an objet out and back in,
71 * individual pages in it will have different bit 17 addresses, resulting in
72 * each 64 bytes being swapped with its neighbor!
74 * Otherwise, if interleaved, we have to tell the 3d driver what the address
75 * swizzling it needs to do is, since it's writing with the CPU to the pages
76 * (bit 6 and potentially bit 11 XORed in), and the GPU is reading from the
77 * pages (bit 6, 9, and 10 XORed in), resulting in a cumulative bit swizzling
78 * required by the CPU of XORing in bit 6, 9, 10, and potentially 11, in order
79 * to match what the GPU expects.
83 * Detects bit 6 swizzling of address lookup between IGD access and CPU
84 * access through main memory.
87 i915_gem_detect_bit_6_swizzle(struct drm_device *dev)
89 drm_i915_private_t *dev_priv = dev->dev_private;
90 uint32_t swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN;
91 uint32_t swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN;
94 /* As far as we know, the 865 doesn't have these bit 6
97 swizzle_x = I915_BIT_6_SWIZZLE_NONE;
98 swizzle_y = I915_BIT_6_SWIZZLE_NONE;
99 } else if ((!IS_I965G(dev) && !IS_G33(dev)) || IS_I965GM(dev) ||
103 /* On 915-945 and GM965, channel interleave by the CPU is
104 * determined by DCC. The CPU will alternate based on bit 6
105 * in interleaved mode, and the GPU will then also alternate
106 * on bit 6, 9, and 10 for X, but the CPU may also optionally
107 * alternate based on bit 17 (XOR not disabled and XOR
110 dcc = I915_READ(DCC);
111 switch (dcc & DCC_ADDRESSING_MODE_MASK) {
112 case DCC_ADDRESSING_MODE_SINGLE_CHANNEL:
113 case DCC_ADDRESSING_MODE_DUAL_CHANNEL_ASYMMETRIC:
114 swizzle_x = I915_BIT_6_SWIZZLE_NONE;
115 swizzle_y = I915_BIT_6_SWIZZLE_NONE;
117 case DCC_ADDRESSING_MODE_DUAL_CHANNEL_INTERLEAVED:
118 if (IS_I915G(dev) || IS_I915GM(dev) ||
119 dcc & DCC_CHANNEL_XOR_DISABLE) {
120 swizzle_x = I915_BIT_6_SWIZZLE_9_10;
121 swizzle_y = I915_BIT_6_SWIZZLE_9;
122 } else if ((IS_I965GM(dev) || IS_GM45(dev)) &&
123 (dcc & DCC_CHANNEL_XOR_BIT_17) == 0) {
124 /* GM965/GM45 does either bit 11 or bit 17
127 swizzle_x = I915_BIT_6_SWIZZLE_9_10_11;
128 swizzle_y = I915_BIT_6_SWIZZLE_9_11;
130 /* Bit 17 or perhaps other swizzling */
131 swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN;
132 swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN;
136 if (dcc == 0xffffffff) {
137 DRM_ERROR("Couldn't read from MCHBAR. "
138 "Disabling tiling.\n");
139 swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN;
140 swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN;
143 /* The 965, G33, and newer, have a very flexible memory
144 * configuration. It will enable dual-channel mode
145 * (interleaving) on as much memory as it can, and the GPU
146 * will additionally sometimes enable different bit 6
147 * swizzling for tiled objects from the CPU.
149 * Here's what I found on the G965:
150 * slot fill memory size swizzling
151 * 0A 0B 1A 1B 1-ch 2-ch
153 * 512 0 512 0 16 1008 X
154 * 512 0 0 512 16 1008 X
155 * 0 512 0 512 16 1008 X
156 * 1024 1024 1024 0 2048 1024 O
158 * We could probably detect this based on either the DRB
159 * matching, which was the case for the swizzling required in
160 * the table above, or from the 1-ch value being less than
161 * the minimum size of a rank.
163 if (I915_READ16(C0DRB3) != I915_READ16(C1DRB3)) {
164 swizzle_x = I915_BIT_6_SWIZZLE_NONE;
165 swizzle_y = I915_BIT_6_SWIZZLE_NONE;
167 swizzle_x = I915_BIT_6_SWIZZLE_9_10;
168 swizzle_y = I915_BIT_6_SWIZZLE_9;
172 dev_priv->mm.bit_6_swizzle_x = swizzle_x;
173 dev_priv->mm.bit_6_swizzle_y = swizzle_y;
177 * Sets the tiling mode of an object, returning the required swizzling of
178 * bit 6 of addresses in the object.
181 i915_gem_set_tiling(struct drm_device *dev, void *data,
182 struct drm_file *file_priv)
184 struct drm_i915_gem_set_tiling *args = data;
185 drm_i915_private_t *dev_priv = dev->dev_private;
186 struct drm_gem_object *obj;
187 struct drm_i915_gem_object *obj_priv;
189 obj = drm_gem_object_lookup(dev, file_priv, args->handle);
192 obj_priv = obj->driver_private;
194 mutex_lock(&dev->struct_mutex);
196 if (args->tiling_mode == I915_TILING_NONE) {
197 obj_priv->tiling_mode = I915_TILING_NONE;
198 args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
200 if (args->tiling_mode == I915_TILING_X)
201 args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
203 args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y;
204 /* If we can't handle the swizzling, make it untiled. */
205 if (args->swizzle_mode == I915_BIT_6_SWIZZLE_UNKNOWN) {
206 args->tiling_mode = I915_TILING_NONE;
207 args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
210 obj_priv->tiling_mode = args->tiling_mode;
212 mutex_unlock(&dev->struct_mutex);
214 drm_gem_object_unreference(obj);
220 * Returns the current tiling mode and required bit 6 swizzling for the object.
223 i915_gem_get_tiling(struct drm_device *dev, void *data,
224 struct drm_file *file_priv)
226 struct drm_i915_gem_get_tiling *args = data;
227 drm_i915_private_t *dev_priv = dev->dev_private;
228 struct drm_gem_object *obj;
229 struct drm_i915_gem_object *obj_priv;
231 obj = drm_gem_object_lookup(dev, file_priv, args->handle);
234 obj_priv = obj->driver_private;
236 mutex_lock(&dev->struct_mutex);
238 args->tiling_mode = obj_priv->tiling_mode;
239 switch (obj_priv->tiling_mode) {
241 args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
244 args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y;
246 case I915_TILING_NONE:
247 args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
250 DRM_ERROR("unknown tiling mode\n");
253 mutex_unlock(&dev->struct_mutex);
255 drm_gem_object_unreference(obj);