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>
28 #include <linux/string.h>
29 #include <linux/bitops.h>
30 #include <drm/i915_drm.h>
34 * DOC: buffer object tiling
36 * i915_gem_set_tiling_ioctl() and i915_gem_get_tiling_ioctl() is the userspace
37 * interface to declare fence register requirements.
39 * In principle GEM doesn't care at all about the internal data layout of an
40 * object, and hence it also doesn't care about tiling or swizzling. There's two
43 * - For X and Y tiling the hardware provides detilers for CPU access, so called
44 * fences. Since there's only a limited amount of them the kernel must manage
45 * these, and therefore userspace must tell the kernel the object tiling if it
46 * wants to use fences for detiling.
47 * - On gen3 and gen4 platforms have a swizzling pattern for tiled objects which
48 * depends upon the physical page frame number. When swapping such objects the
49 * page frame number might change and the kernel must be able to fix this up
50 * and hence now the tiling. Note that on a subset of platforms with
51 * asymmetric memory channel population the swizzling pattern changes in an
52 * unknown way, and for those the kernel simply forbids swapping completely.
54 * Since neither of this applies for new tiling layouts on modern platforms like
55 * W, Ys and Yf tiling GEM only allows object tiling to be set to X or Y tiled.
56 * Anything else can be handled in userspace entirely without the kernel's
61 * i915_gem_fence_size - required global GTT size for a fence
64 * @tiling: tiling mode
65 * @stride: tiling stride
67 * Return the required global GTT size for a fence (view of a tiled object),
68 * taking into account potential fence register mapping.
70 u32
i915_gem_fence_size(struct drm_i915_private
*i915
,
71 u32 size
, unsigned int tiling
, unsigned int stride
)
77 if (tiling
== I915_TILING_NONE
)
82 if (INTEL_GEN(i915
) >= 4) {
83 stride
*= i915_gem_tile_height(tiling
);
84 GEM_BUG_ON(!IS_ALIGNED(stride
, I965_FENCE_PAGE
));
85 return roundup(size
, stride
);
88 /* Previous chips need a power-of-two fence region when tiling */
90 ggtt_size
= 1024*1024;
94 while (ggtt_size
< size
)
101 * i915_gem_fence_alignment - required global GTT alignment for a fence
104 * @tiling: tiling mode
105 * @stride: tiling stride
107 * Return the required global GTT alignment for a fence (a view of a tiled
108 * object), taking into account potential fence register mapping.
110 u32
i915_gem_fence_alignment(struct drm_i915_private
*i915
, u32 size
,
111 unsigned int tiling
, unsigned int stride
)
116 * Minimum alignment is 4k (GTT page size), but might be greater
117 * if a fence register is needed for the object.
119 if (tiling
== I915_TILING_NONE
)
120 return I915_GTT_MIN_ALIGNMENT
;
122 if (INTEL_GEN(i915
) >= 4)
123 return I965_FENCE_PAGE
;
126 * Previous chips need to be aligned to the size of the smallest
127 * fence register that can contain the object.
129 return i915_gem_fence_size(i915
, size
, tiling
, stride
);
132 /* Check pitch constriants for all chips & tiling formats */
134 i915_tiling_ok(struct drm_i915_gem_object
*obj
,
135 unsigned int tiling
, unsigned int stride
)
137 struct drm_i915_private
*i915
= to_i915(obj
->base
.dev
);
138 unsigned int tile_width
;
140 /* Linear is always fine */
141 if (tiling
== I915_TILING_NONE
)
144 if (tiling
> I915_TILING_LAST
)
147 /* check maximum stride & object size */
148 /* i965+ stores the end address of the gtt mapping in the fence
149 * reg, so dont bother to check the size */
150 if (INTEL_GEN(i915
) >= 7) {
151 if (stride
/ 128 > GEN7_FENCE_MAX_PITCH_VAL
)
153 } else if (INTEL_GEN(i915
) >= 4) {
154 if (stride
/ 128 > I965_FENCE_MAX_PITCH_VAL
)
160 if (!is_power_of_2(stride
))
164 if (IS_GEN(i915
, 2) ||
165 (tiling
== I915_TILING_Y
&& HAS_128_BYTE_Y_TILING(i915
)))
170 if (!stride
|| !IS_ALIGNED(stride
, tile_width
))
176 static bool i915_vma_fence_prepare(struct i915_vma
*vma
,
177 int tiling_mode
, unsigned int stride
)
179 struct drm_i915_private
*i915
= vma
->vm
->i915
;
182 if (!i915_vma_is_map_and_fenceable(vma
))
185 size
= i915_gem_fence_size(i915
, vma
->size
, tiling_mode
, stride
);
186 if (vma
->node
.size
< size
)
189 alignment
= i915_gem_fence_alignment(i915
, vma
->size
, tiling_mode
, stride
);
190 if (!IS_ALIGNED(vma
->node
.start
, alignment
))
196 /* Make the current GTT allocation valid for the change in tiling. */
198 i915_gem_object_fence_prepare(struct drm_i915_gem_object
*obj
,
199 int tiling_mode
, unsigned int stride
)
201 struct i915_vma
*vma
;
204 if (tiling_mode
== I915_TILING_NONE
)
207 for_each_ggtt_vma(vma
, obj
) {
208 if (i915_vma_fence_prepare(vma
, tiling_mode
, stride
))
211 ret
= i915_vma_unbind(vma
);
220 i915_gem_object_set_tiling(struct drm_i915_gem_object
*obj
,
221 unsigned int tiling
, unsigned int stride
)
223 struct drm_i915_private
*i915
= to_i915(obj
->base
.dev
);
224 struct i915_vma
*vma
;
227 /* Make sure we don't cross-contaminate obj->tiling_and_stride */
228 BUILD_BUG_ON(I915_TILING_LAST
& STRIDE_MASK
);
230 GEM_BUG_ON(!i915_tiling_ok(obj
, tiling
, stride
));
231 GEM_BUG_ON(!stride
^ (tiling
== I915_TILING_NONE
));
232 lockdep_assert_held(&i915
->drm
.struct_mutex
);
234 if ((tiling
| stride
) == obj
->tiling_and_stride
)
237 if (i915_gem_object_is_framebuffer(obj
))
240 /* We need to rebind the object if its current allocation
241 * no longer meets the alignment restrictions for its new
242 * tiling mode. Otherwise we can just leave it alone, but
243 * need to ensure that any fence register is updated before
244 * the next fenced (either through the GTT or by the BLT unit
245 * on older GPUs) access.
247 * After updating the tiling parameters, we then flag whether
248 * we need to update an associated fence register. Note this
249 * has to also include the unfenced register the GPU uses
250 * whilst executing a fenced command for an untiled object.
253 err
= i915_gem_object_fence_prepare(obj
, tiling
, stride
);
257 i915_gem_object_lock(obj
);
258 if (i915_gem_object_is_framebuffer(obj
)) {
259 i915_gem_object_unlock(obj
);
263 /* If the memory has unknown (i.e. varying) swizzling, we pin the
264 * pages to prevent them being swapped out and causing corruption
265 * due to the change in swizzling.
267 mutex_lock(&obj
->mm
.lock
);
268 if (i915_gem_object_has_pages(obj
) &&
269 obj
->mm
.madv
== I915_MADV_WILLNEED
&&
270 i915
->quirks
& QUIRK_PIN_SWIZZLED_PAGES
) {
271 if (tiling
== I915_TILING_NONE
) {
272 GEM_BUG_ON(!obj
->mm
.quirked
);
273 __i915_gem_object_unpin_pages(obj
);
274 obj
->mm
.quirked
= false;
276 if (!i915_gem_object_is_tiled(obj
)) {
277 GEM_BUG_ON(obj
->mm
.quirked
);
278 __i915_gem_object_pin_pages(obj
);
279 obj
->mm
.quirked
= true;
282 mutex_unlock(&obj
->mm
.lock
);
284 for_each_ggtt_vma(vma
, obj
) {
286 i915_gem_fence_size(i915
, vma
->size
, tiling
, stride
);
287 vma
->fence_alignment
=
288 i915_gem_fence_alignment(i915
,
289 vma
->size
, tiling
, stride
);
292 vma
->fence
->dirty
= true;
295 obj
->tiling_and_stride
= tiling
| stride
;
296 i915_gem_object_unlock(obj
);
298 /* Force the fence to be reacquired for GTT access */
299 i915_gem_release_mmap(obj
);
301 /* Try to preallocate memory required to save swizzling on put-pages */
302 if (i915_gem_object_needs_bit17_swizzle(obj
)) {
304 obj
->bit_17
= kcalloc(BITS_TO_LONGS(obj
->base
.size
>> PAGE_SHIFT
),
305 sizeof(long), GFP_KERNEL
);
316 * i915_gem_set_tiling_ioctl - IOCTL handler to set tiling mode
318 * @data: data pointer for the ioctl
319 * @file: DRM file for the ioctl call
321 * Sets the tiling mode of an object, returning the required swizzling of
322 * bit 6 of addresses in the object.
324 * Called by the user via ioctl.
327 * Zero on success, negative errno on failure.
330 i915_gem_set_tiling_ioctl(struct drm_device
*dev
, void *data
,
331 struct drm_file
*file
)
333 struct drm_i915_gem_set_tiling
*args
= data
;
334 struct drm_i915_gem_object
*obj
;
337 obj
= i915_gem_object_lookup(file
, args
->handle
);
342 * The tiling mode of proxy objects is handled by its generator, and
343 * not allowed to be changed by userspace.
345 if (i915_gem_object_is_proxy(obj
)) {
350 if (!i915_tiling_ok(obj
, args
->tiling_mode
, args
->stride
)) {
355 if (args
->tiling_mode
== I915_TILING_NONE
) {
356 args
->swizzle_mode
= I915_BIT_6_SWIZZLE_NONE
;
359 if (args
->tiling_mode
== I915_TILING_X
)
360 args
->swizzle_mode
= to_i915(dev
)->mm
.bit_6_swizzle_x
;
362 args
->swizzle_mode
= to_i915(dev
)->mm
.bit_6_swizzle_y
;
364 /* Hide bit 17 swizzling from the user. This prevents old Mesa
365 * from aborting the application on sw fallbacks to bit 17,
366 * and we use the pread/pwrite bit17 paths to swizzle for it.
367 * If there was a user that was relying on the swizzle
368 * information for drm_intel_bo_map()ed reads/writes this would
369 * break it, but we don't have any of those.
371 if (args
->swizzle_mode
== I915_BIT_6_SWIZZLE_9_17
)
372 args
->swizzle_mode
= I915_BIT_6_SWIZZLE_9
;
373 if (args
->swizzle_mode
== I915_BIT_6_SWIZZLE_9_10_17
)
374 args
->swizzle_mode
= I915_BIT_6_SWIZZLE_9_10
;
376 /* If we can't handle the swizzling, make it untiled. */
377 if (args
->swizzle_mode
== I915_BIT_6_SWIZZLE_UNKNOWN
) {
378 args
->tiling_mode
= I915_TILING_NONE
;
379 args
->swizzle_mode
= I915_BIT_6_SWIZZLE_NONE
;
384 err
= mutex_lock_interruptible(&dev
->struct_mutex
);
388 err
= i915_gem_object_set_tiling(obj
, args
->tiling_mode
, args
->stride
);
389 mutex_unlock(&dev
->struct_mutex
);
391 /* We have to maintain this existing ABI... */
392 args
->stride
= i915_gem_object_get_stride(obj
);
393 args
->tiling_mode
= i915_gem_object_get_tiling(obj
);
396 i915_gem_object_put(obj
);
401 * i915_gem_get_tiling_ioctl - IOCTL handler to get tiling mode
403 * @data: data pointer for the ioctl
404 * @file: DRM file for the ioctl call
406 * Returns the current tiling mode and required bit 6 swizzling for the object.
408 * Called by the user via ioctl.
411 * Zero on success, negative errno on failure.
414 i915_gem_get_tiling_ioctl(struct drm_device
*dev
, void *data
,
415 struct drm_file
*file
)
417 struct drm_i915_gem_get_tiling
*args
= data
;
418 struct drm_i915_private
*dev_priv
= to_i915(dev
);
419 struct drm_i915_gem_object
*obj
;
423 obj
= i915_gem_object_lookup_rcu(file
, args
->handle
);
426 READ_ONCE(obj
->tiling_and_stride
) & TILING_MASK
;
433 switch (args
->tiling_mode
) {
435 args
->swizzle_mode
= dev_priv
->mm
.bit_6_swizzle_x
;
438 args
->swizzle_mode
= dev_priv
->mm
.bit_6_swizzle_y
;
441 case I915_TILING_NONE
:
442 args
->swizzle_mode
= I915_BIT_6_SWIZZLE_NONE
;
446 /* Hide bit 17 from the user -- see comment in i915_gem_set_tiling */
447 if (dev_priv
->quirks
& QUIRK_PIN_SWIZZLED_PAGES
)
448 args
->phys_swizzle_mode
= I915_BIT_6_SWIZZLE_UNKNOWN
;
450 args
->phys_swizzle_mode
= args
->swizzle_mode
;
451 if (args
->swizzle_mode
== I915_BIT_6_SWIZZLE_9_17
)
452 args
->swizzle_mode
= I915_BIT_6_SWIZZLE_9
;
453 if (args
->swizzle_mode
== I915_BIT_6_SWIZZLE_9_10_17
)
454 args
->swizzle_mode
= I915_BIT_6_SWIZZLE_9_10
;