]>
Commit | Line | Data |
---|---|---|
673a394b EA |
1 | /* |
2 | * Copyright © 2008 Intel Corporation | |
3 | * | |
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: | |
10 | * | |
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 | |
13 | * Software. | |
14 | * | |
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 | |
21 | * IN THE SOFTWARE. | |
22 | * | |
23 | * Authors: | |
24 | * Eric Anholt <eric@anholt.net> | |
25 | * | |
26 | */ | |
27 | ||
760285e7 DH |
28 | #include <linux/string.h> |
29 | #include <linux/bitops.h> | |
30 | #include <drm/drmP.h> | |
31 | #include <drm/i915_drm.h> | |
673a394b EA |
32 | #include "i915_drv.h" |
33 | ||
3271dca4 DV |
34 | /** |
35 | * DOC: buffer object tiling | |
673a394b | 36 | * |
3271dca4 DV |
37 | * i915_gem_set_tiling() and i915_gem_get_tiling() is the userspace interface to |
38 | * declare fence register requirements. | |
673a394b | 39 | * |
3271dca4 DV |
40 | * In principle GEM doesn't care at all about the internal data layout of an |
41 | * object, and hence it also doesn't care about tiling or swizzling. There's two | |
42 | * exceptions: | |
673a394b | 43 | * |
3271dca4 DV |
44 | * - For X and Y tiling the hardware provides detilers for CPU access, so called |
45 | * fences. Since there's only a limited amount of them the kernel must manage | |
46 | * these, and therefore userspace must tell the kernel the object tiling if it | |
47 | * wants to use fences for detiling. | |
48 | * - On gen3 and gen4 platforms have a swizzling pattern for tiled objects which | |
49 | * depends upon the physical page frame number. When swapping such objects the | |
50 | * page frame number might change and the kernel must be able to fix this up | |
51 | * and hence now the tiling. Note that on a subset of platforms with | |
52 | * asymmetric memory channel population the swizzling pattern changes in an | |
53 | * unknown way, and for those the kernel simply forbids swapping completely. | |
673a394b | 54 | * |
3271dca4 DV |
55 | * Since neither of this applies for new tiling layouts on modern platforms like |
56 | * W, Ys and Yf tiling GEM only allows object tiling to be set to X or Y tiled. | |
57 | * Anything else can be handled in userspace entirely without the kernel's | |
58 | * invovlement. | |
673a394b EA |
59 | */ |
60 | ||
91d4e0aa CW |
61 | /** |
62 | * i915_gem_fence_size - required global GTT size for a fence | |
63 | * @i915: i915 device | |
64 | * @size: object size | |
65 | * @tiling: tiling mode | |
66 | * @stride: tiling stride | |
67 | * | |
68 | * Return the required global GTT size for a fence (view of a tiled object), | |
69 | * taking into account potential fence register mapping. | |
70 | */ | |
71 | u32 i915_gem_fence_size(struct drm_i915_private *i915, | |
72 | u32 size, unsigned int tiling, unsigned int stride) | |
73 | { | |
74 | u32 ggtt_size; | |
75 | ||
76 | GEM_BUG_ON(!size); | |
77 | ||
78 | if (tiling == I915_TILING_NONE) | |
79 | return size; | |
80 | ||
81 | GEM_BUG_ON(!stride); | |
82 | ||
83 | if (INTEL_GEN(i915) >= 4) { | |
84 | stride *= i915_gem_tile_height(tiling); | |
85 | GEM_BUG_ON(stride & 4095); | |
86 | return roundup(size, stride); | |
87 | } | |
88 | ||
89 | /* Previous chips need a power-of-two fence region when tiling */ | |
90 | if (IS_GEN3(i915)) | |
91 | ggtt_size = 1024*1024; | |
92 | else | |
93 | ggtt_size = 512*1024; | |
94 | ||
95 | while (ggtt_size < size) | |
96 | ggtt_size <<= 1; | |
97 | ||
98 | return ggtt_size; | |
99 | } | |
100 | ||
101 | /** | |
102 | * i915_gem_fence_alignment - required global GTT alignment for a fence | |
103 | * @i915: i915 device | |
104 | * @size: object size | |
105 | * @tiling: tiling mode | |
106 | * @stride: tiling stride | |
107 | * | |
108 | * Return the required global GTT alignment for a fence (a view of a tiled | |
109 | * object), taking into account potential fence register mapping. | |
110 | */ | |
111 | u32 i915_gem_fence_alignment(struct drm_i915_private *i915, u32 size, | |
112 | unsigned int tiling, unsigned int stride) | |
113 | { | |
114 | GEM_BUG_ON(!size); | |
115 | ||
116 | /* | |
117 | * Minimum alignment is 4k (GTT page size), but might be greater | |
118 | * if a fence register is needed for the object. | |
119 | */ | |
120 | if (INTEL_GEN(i915) >= 4 || tiling == I915_TILING_NONE) | |
121 | return 4096; | |
122 | ||
123 | /* | |
124 | * Previous chips need to be aligned to the size of the smallest | |
125 | * fence register that can contain the object. | |
126 | */ | |
127 | return i915_gem_fence_size(i915, size, tiling, stride); | |
128 | } | |
129 | ||
0f973f27 | 130 | /* Check pitch constriants for all chips & tiling formats */ |
a00b10c3 | 131 | static bool |
118bb9fb TU |
132 | i915_tiling_ok(struct drm_i915_private *dev_priv, |
133 | int stride, int size, int tiling_mode) | |
0f973f27 | 134 | { |
0ee537ab | 135 | int tile_width; |
0f973f27 JB |
136 | |
137 | /* Linear is always fine */ | |
138 | if (tiling_mode == I915_TILING_NONE) | |
139 | return true; | |
140 | ||
deeb1519 CW |
141 | if (tiling_mode > I915_TILING_LAST) |
142 | return false; | |
143 | ||
5db94019 | 144 | if (IS_GEN2(dev_priv) || |
50a0bc90 | 145 | (tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev_priv))) |
0f973f27 JB |
146 | tile_width = 128; |
147 | else | |
148 | tile_width = 512; | |
149 | ||
8d7773a3 | 150 | /* check maximum stride & object size */ |
3a062478 VS |
151 | /* i965+ stores the end address of the gtt mapping in the fence |
152 | * reg, so dont bother to check the size */ | |
118bb9fb | 153 | if (INTEL_GEN(dev_priv) >= 7) { |
3a062478 VS |
154 | if (stride / 128 > GEN7_FENCE_MAX_PITCH_VAL) |
155 | return false; | |
118bb9fb | 156 | } else if (INTEL_GEN(dev_priv) >= 4) { |
8d7773a3 DV |
157 | if (stride / 128 > I965_FENCE_MAX_PITCH_VAL) |
158 | return false; | |
a6c45cf0 | 159 | } else { |
c36a2a6d | 160 | if (stride > 8192) |
8d7773a3 | 161 | return false; |
e76a16de | 162 | |
5db94019 | 163 | if (IS_GEN3(dev_priv)) { |
c36a2a6d DV |
164 | if (size > I830_FENCE_MAX_SIZE_VAL << 20) |
165 | return false; | |
166 | } else { | |
167 | if (size > I830_FENCE_MAX_SIZE_VAL << 19) | |
168 | return false; | |
169 | } | |
8d7773a3 DV |
170 | } |
171 | ||
fe48d8de VS |
172 | if (stride < tile_width) |
173 | return false; | |
174 | ||
0f973f27 | 175 | /* 965+ just needs multiples of tile width */ |
118bb9fb | 176 | if (INTEL_GEN(dev_priv) >= 4) { |
0f973f27 JB |
177 | if (stride & (tile_width - 1)) |
178 | return false; | |
179 | return true; | |
180 | } | |
181 | ||
182 | /* Pre-965 needs power of two tile widths */ | |
0f973f27 JB |
183 | if (stride & (stride - 1)) |
184 | return false; | |
185 | ||
0f973f27 JB |
186 | return true; |
187 | } | |
188 | ||
5b30694b CW |
189 | static bool i915_vma_fence_prepare(struct i915_vma *vma, |
190 | int tiling_mode, unsigned int stride) | |
49ef5294 | 191 | { |
944397f0 CW |
192 | struct drm_i915_private *i915 = vma->vm->i915; |
193 | u32 size, alignment; | |
49ef5294 CW |
194 | |
195 | if (!i915_vma_is_map_and_fenceable(vma)) | |
196 | return true; | |
197 | ||
91d4e0aa | 198 | size = i915_gem_fence_size(i915, vma->size, tiling_mode, stride); |
49ef5294 CW |
199 | if (vma->node.size < size) |
200 | return false; | |
201 | ||
91d4e0aa | 202 | alignment = i915_gem_fence_alignment(i915, vma->size, tiling_mode, stride); |
944397f0 | 203 | if (vma->node.start & (alignment - 1)) |
49ef5294 CW |
204 | return false; |
205 | ||
206 | return true; | |
207 | } | |
208 | ||
f23eda8c CW |
209 | /* Make the current GTT allocation valid for the change in tiling. */ |
210 | static int | |
5b30694b CW |
211 | i915_gem_object_fence_prepare(struct drm_i915_gem_object *obj, |
212 | int tiling_mode, unsigned int stride) | |
52dc7d32 | 213 | { |
f23eda8c | 214 | struct i915_vma *vma; |
49ef5294 | 215 | int ret; |
52dc7d32 CW |
216 | |
217 | if (tiling_mode == I915_TILING_NONE) | |
f23eda8c | 218 | return 0; |
52dc7d32 | 219 | |
49ef5294 | 220 | list_for_each_entry(vma, &obj->vma_list, obj_link) { |
944397f0 CW |
221 | if (!i915_vma_is_ggtt(vma)) |
222 | break; | |
223 | ||
5b30694b | 224 | if (i915_vma_fence_prepare(vma, tiling_mode, stride)) |
49ef5294 | 225 | continue; |
a6c45cf0 | 226 | |
49ef5294 CW |
227 | ret = i915_vma_unbind(vma); |
228 | if (ret) | |
229 | return ret; | |
df153158 CW |
230 | } |
231 | ||
f23eda8c | 232 | return 0; |
52dc7d32 CW |
233 | } |
234 | ||
673a394b | 235 | /** |
3271dca4 DV |
236 | * i915_gem_set_tiling - IOCTL handler to set tiling mode |
237 | * @dev: DRM device | |
238 | * @data: data pointer for the ioctl | |
239 | * @file: DRM file for the ioctl call | |
240 | * | |
673a394b EA |
241 | * Sets the tiling mode of an object, returning the required swizzling of |
242 | * bit 6 of addresses in the object. | |
3271dca4 DV |
243 | * |
244 | * Called by the user via ioctl. | |
245 | * | |
246 | * Returns: | |
247 | * Zero on success, negative errno on failure. | |
673a394b EA |
248 | */ |
249 | int | |
250 | i915_gem_set_tiling(struct drm_device *dev, void *data, | |
05394f39 | 251 | struct drm_file *file) |
673a394b EA |
252 | { |
253 | struct drm_i915_gem_set_tiling *args = data; | |
fac5e23e | 254 | struct drm_i915_private *dev_priv = to_i915(dev); |
05394f39 | 255 | struct drm_i915_gem_object *obj; |
f23eda8c | 256 | int err = 0; |
673a394b | 257 | |
3e510a8e CW |
258 | /* Make sure we don't cross-contaminate obj->tiling_and_stride */ |
259 | BUILD_BUG_ON(I915_TILING_LAST & STRIDE_MASK); | |
260 | ||
03ac0642 CW |
261 | obj = i915_gem_object_lookup(file, args->handle); |
262 | if (!obj) | |
bf79cb91 | 263 | return -ENOENT; |
673a394b | 264 | |
118bb9fb | 265 | if (!i915_tiling_ok(dev_priv, |
05394f39 | 266 | args->stride, obj->base.size, args->tiling_mode)) { |
f0cd5182 | 267 | i915_gem_object_put(obj); |
0f973f27 | 268 | return -EINVAL; |
72daad40 | 269 | } |
0f973f27 | 270 | |
6c31a614 | 271 | mutex_lock(&dev->struct_mutex); |
1f30a614 | 272 | if (obj->pin_display || obj->framebuffer_references) { |
f23eda8c | 273 | err = -EBUSY; |
6c31a614 | 274 | goto err; |
31770bd4 DV |
275 | } |
276 | ||
673a394b | 277 | if (args->tiling_mode == I915_TILING_NONE) { |
673a394b | 278 | args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE; |
52dc7d32 | 279 | args->stride = 0; |
673a394b EA |
280 | } else { |
281 | if (args->tiling_mode == I915_TILING_X) | |
282 | args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x; | |
283 | else | |
284 | args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y; | |
280b713b EA |
285 | |
286 | /* Hide bit 17 swizzling from the user. This prevents old Mesa | |
287 | * from aborting the application on sw fallbacks to bit 17, | |
288 | * and we use the pread/pwrite bit17 paths to swizzle for it. | |
289 | * If there was a user that was relying on the swizzle | |
290 | * information for drm_intel_bo_map()ed reads/writes this would | |
291 | * break it, but we don't have any of those. | |
292 | */ | |
293 | if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_17) | |
294 | args->swizzle_mode = I915_BIT_6_SWIZZLE_9; | |
295 | if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_10_17) | |
296 | args->swizzle_mode = I915_BIT_6_SWIZZLE_9_10; | |
297 | ||
673a394b EA |
298 | /* If we can't handle the swizzling, make it untiled. */ |
299 | if (args->swizzle_mode == I915_BIT_6_SWIZZLE_UNKNOWN) { | |
300 | args->tiling_mode = I915_TILING_NONE; | |
301 | args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE; | |
52dc7d32 | 302 | args->stride = 0; |
673a394b EA |
303 | } |
304 | } | |
0f973f27 | 305 | |
3e510a8e CW |
306 | if (args->tiling_mode != i915_gem_object_get_tiling(obj) || |
307 | args->stride != i915_gem_object_get_stride(obj)) { | |
52dc7d32 CW |
308 | /* We need to rebind the object if its current allocation |
309 | * no longer meets the alignment restrictions for its new | |
310 | * tiling mode. Otherwise we can just leave it alone, but | |
1869b620 CW |
311 | * need to ensure that any fence register is updated before |
312 | * the next fenced (either through the GTT or by the BLT unit | |
313 | * on older GPUs) access. | |
5d82e3e6 CW |
314 | * |
315 | * After updating the tiling parameters, we then flag whether | |
316 | * we need to update an associated fence register. Note this | |
317 | * has to also include the unfenced register the GPU uses | |
318 | * whilst executing a fenced command for an untiled object. | |
0f973f27 | 319 | */ |
467cffba | 320 | |
5b30694b CW |
321 | err = i915_gem_object_fence_prepare(obj, |
322 | args->tiling_mode, | |
323 | args->stride); | |
f23eda8c | 324 | if (!err) { |
49ef5294 CW |
325 | struct i915_vma *vma; |
326 | ||
1233e2db | 327 | mutex_lock(&obj->mm.lock); |
a4f5ea64 CW |
328 | if (obj->mm.pages && |
329 | obj->mm.madv == I915_MADV_WILLNEED && | |
656bfa3a | 330 | dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES) { |
bc0629a7 CW |
331 | if (args->tiling_mode == I915_TILING_NONE) { |
332 | GEM_BUG_ON(!obj->mm.quirked); | |
a4f5ea64 | 333 | __i915_gem_object_unpin_pages(obj); |
bc0629a7 CW |
334 | obj->mm.quirked = false; |
335 | } | |
336 | if (!i915_gem_object_is_tiled(obj)) { | |
2c3a3f44 | 337 | GEM_BUG_ON(!obj->mm.quirked); |
a4f5ea64 | 338 | __i915_gem_object_pin_pages(obj); |
bc0629a7 CW |
339 | obj->mm.quirked = true; |
340 | } | |
656bfa3a | 341 | } |
1233e2db | 342 | mutex_unlock(&obj->mm.lock); |
656bfa3a | 343 | |
49ef5294 | 344 | list_for_each_entry(vma, &obj->vma_list, obj_link) { |
944397f0 CW |
345 | if (!i915_vma_is_ggtt(vma)) |
346 | break; | |
347 | ||
91d4e0aa CW |
348 | vma->fence_size = i915_gem_fence_size(dev_priv, vma->size, |
349 | args->tiling_mode, | |
350 | args->stride); | |
351 | vma->fence_alignment = i915_gem_fence_alignment(dev_priv, vma->size, | |
352 | args->tiling_mode, | |
353 | args->stride); | |
944397f0 CW |
354 | |
355 | if (vma->fence) | |
356 | vma->fence->dirty = true; | |
49ef5294 | 357 | } |
3e510a8e CW |
358 | obj->tiling_and_stride = |
359 | args->stride | args->tiling_mode; | |
1869b620 CW |
360 | |
361 | /* Force the fence to be reacquired for GTT access */ | |
362 | i915_gem_release_mmap(obj); | |
467cffba | 363 | } |
0f973f27 | 364 | } |
467cffba | 365 | /* we have to maintain this existing ABI... */ |
3e510a8e CW |
366 | args->stride = i915_gem_object_get_stride(obj); |
367 | args->tiling_mode = i915_gem_object_get_tiling(obj); | |
e9b73c67 CW |
368 | |
369 | /* Try to preallocate memory required to save swizzling on put-pages */ | |
370 | if (i915_gem_object_needs_bit17_swizzle(obj)) { | |
371 | if (obj->bit_17 == NULL) { | |
a1e22653 | 372 | obj->bit_17 = kcalloc(BITS_TO_LONGS(obj->base.size >> PAGE_SHIFT), |
e9b73c67 CW |
373 | sizeof(long), GFP_KERNEL); |
374 | } | |
375 | } else { | |
376 | kfree(obj->bit_17); | |
377 | obj->bit_17 = NULL; | |
378 | } | |
379 | ||
6c31a614 | 380 | err: |
f8c417cd | 381 | i915_gem_object_put(obj); |
d6873102 | 382 | mutex_unlock(&dev->struct_mutex); |
673a394b | 383 | |
f23eda8c | 384 | return err; |
673a394b EA |
385 | } |
386 | ||
387 | /** | |
3271dca4 DV |
388 | * i915_gem_get_tiling - IOCTL handler to get tiling mode |
389 | * @dev: DRM device | |
390 | * @data: data pointer for the ioctl | |
391 | * @file: DRM file for the ioctl call | |
392 | * | |
673a394b | 393 | * Returns the current tiling mode and required bit 6 swizzling for the object. |
3271dca4 DV |
394 | * |
395 | * Called by the user via ioctl. | |
396 | * | |
397 | * Returns: | |
398 | * Zero on success, negative errno on failure. | |
673a394b EA |
399 | */ |
400 | int | |
401 | i915_gem_get_tiling(struct drm_device *dev, void *data, | |
05394f39 | 402 | struct drm_file *file) |
673a394b EA |
403 | { |
404 | struct drm_i915_gem_get_tiling *args = data; | |
fac5e23e | 405 | struct drm_i915_private *dev_priv = to_i915(dev); |
05394f39 | 406 | struct drm_i915_gem_object *obj; |
fbbd37b3 CW |
407 | int err = -ENOENT; |
408 | ||
409 | rcu_read_lock(); | |
410 | obj = i915_gem_object_lookup_rcu(file, args->handle); | |
411 | if (obj) { | |
412 | args->tiling_mode = | |
413 | READ_ONCE(obj->tiling_and_stride) & TILING_MASK; | |
414 | err = 0; | |
415 | } | |
416 | rcu_read_unlock(); | |
417 | if (unlikely(err)) | |
418 | return err; | |
673a394b | 419 | |
9ad36761 | 420 | switch (args->tiling_mode) { |
673a394b EA |
421 | case I915_TILING_X: |
422 | args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x; | |
423 | break; | |
424 | case I915_TILING_Y: | |
425 | args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y; | |
426 | break; | |
fbbd37b3 | 427 | default: |
673a394b EA |
428 | case I915_TILING_NONE: |
429 | args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE; | |
430 | break; | |
673a394b EA |
431 | } |
432 | ||
280b713b | 433 | /* Hide bit 17 from the user -- see comment in i915_gem_set_tiling */ |
5eb3e5a5 CW |
434 | if (dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES) |
435 | args->phys_swizzle_mode = I915_BIT_6_SWIZZLE_UNKNOWN; | |
436 | else | |
437 | args->phys_swizzle_mode = args->swizzle_mode; | |
280b713b EA |
438 | if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_17) |
439 | args->swizzle_mode = I915_BIT_6_SWIZZLE_9; | |
440 | if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_10_17) | |
441 | args->swizzle_mode = I915_BIT_6_SWIZZLE_9_10; | |
442 | ||
673a394b EA |
443 | return 0; |
444 | } |