xref: /dragonfly/sys/dev/drm/i915/i915_gem_tiling.c (revision ffe53622)
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 
28 #include <linux/bitops.h>
29 #include <drm/drmP.h>
30 #include <drm/i915_drm.h>
31 #include "i915_drv.h"
32 
33 /**
34  * DOC: buffer object tiling
35  *
36  * i915_gem_set_tiling() and i915_gem_get_tiling() is the userspace interface to
37  * declare fence register requirements.
38  *
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
41  * exceptions:
42  *
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.
53  *
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
57  * invovlement.
58  */
59 
60 /* Check pitch constriants for all chips & tiling formats */
61 static bool
62 i915_tiling_ok(struct drm_device *dev, int stride, int size, int tiling_mode)
63 {
64 	int tile_width;
65 
66 	/* Linear is always fine */
67 	if (tiling_mode == I915_TILING_NONE)
68 		return true;
69 
70 	if (IS_GEN2(dev) ||
71 	    (tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev)))
72 		tile_width = 128;
73 	else
74 		tile_width = 512;
75 
76 	/* check maximum stride & object size */
77 	/* i965+ stores the end address of the gtt mapping in the fence
78 	 * reg, so dont bother to check the size */
79 	if (INTEL_INFO(dev)->gen >= 7) {
80 		if (stride / 128 > GEN7_FENCE_MAX_PITCH_VAL)
81 			return false;
82 	} else if (INTEL_INFO(dev)->gen >= 4) {
83 		if (stride / 128 > I965_FENCE_MAX_PITCH_VAL)
84 			return false;
85 	} else {
86 		if (stride > 8192)
87 			return false;
88 
89 		if (IS_GEN3(dev)) {
90 			if (size > I830_FENCE_MAX_SIZE_VAL << 20)
91 				return false;
92 		} else {
93 			if (size > I830_FENCE_MAX_SIZE_VAL << 19)
94 				return false;
95 		}
96 	}
97 
98 	if (stride < tile_width)
99 		return false;
100 
101 	/* 965+ just needs multiples of tile width */
102 	if (INTEL_INFO(dev)->gen >= 4) {
103 		if (stride & (tile_width - 1))
104 			return false;
105 		return true;
106 	}
107 
108 	/* Pre-965 needs power of two tile widths */
109 	if (stride & (stride - 1))
110 		return false;
111 
112 	return true;
113 }
114 
115 /* Is the current GTT allocation valid for the change in tiling? */
116 static bool
117 i915_gem_object_fence_ok(struct drm_i915_gem_object *obj, int tiling_mode)
118 {
119 	u32 size;
120 
121 	if (tiling_mode == I915_TILING_NONE)
122 		return true;
123 
124 	if (INTEL_INFO(obj->base.dev)->gen >= 4)
125 		return true;
126 
127 	if (INTEL_INFO(obj->base.dev)->gen == 3) {
128 		if (i915_gem_obj_ggtt_offset(obj) & ~I915_FENCE_START_MASK)
129 			return false;
130 	} else {
131 		if (i915_gem_obj_ggtt_offset(obj) & ~I830_FENCE_START_MASK)
132 			return false;
133 	}
134 
135 	size = i915_gem_get_gtt_size(obj->base.dev, obj->base.size, tiling_mode);
136 	if (i915_gem_obj_ggtt_size(obj) != size)
137 		return false;
138 
139 	if (i915_gem_obj_ggtt_offset(obj) & (size - 1))
140 		return false;
141 
142 	return true;
143 }
144 
145 /**
146  * i915_gem_set_tiling - IOCTL handler to set tiling mode
147  * @dev: DRM device
148  * @data: data pointer for the ioctl
149  * @file: DRM file for the ioctl call
150  *
151  * Sets the tiling mode of an object, returning the required swizzling of
152  * bit 6 of addresses in the object.
153  *
154  * Called by the user via ioctl.
155  *
156  * Returns:
157  * Zero on success, negative errno on failure.
158  */
159 int
160 i915_gem_set_tiling(struct drm_device *dev, void *data,
161 		   struct drm_file *file)
162 {
163 	struct drm_i915_gem_set_tiling *args = data;
164 	struct drm_i915_private *dev_priv = dev->dev_private;
165 	struct drm_i915_gem_object *obj;
166 	int ret = 0;
167 
168 	obj = to_intel_bo(drm_gem_object_lookup(file, args->handle));
169 	if (&obj->base == NULL)
170 		return -ENOENT;
171 
172 	if (!i915_tiling_ok(dev,
173 			    args->stride, obj->base.size, args->tiling_mode)) {
174 		drm_gem_object_unreference_unlocked(&obj->base);
175 		return -EINVAL;
176 	}
177 
178 	intel_runtime_pm_get(dev_priv);
179 
180 	mutex_lock(&dev->struct_mutex);
181 	if (obj->pin_display || obj->framebuffer_references) {
182 		ret = -EBUSY;
183 		goto err;
184 	}
185 
186 	if (args->tiling_mode == I915_TILING_NONE) {
187 		args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
188 		args->stride = 0;
189 	} else {
190 		if (args->tiling_mode == I915_TILING_X)
191 			args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
192 		else
193 			args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y;
194 
195 		/* Hide bit 17 swizzling from the user.  This prevents old Mesa
196 		 * from aborting the application on sw fallbacks to bit 17,
197 		 * and we use the pread/pwrite bit17 paths to swizzle for it.
198 		 * If there was a user that was relying on the swizzle
199 		 * information for drm_intel_bo_map()ed reads/writes this would
200 		 * break it, but we don't have any of those.
201 		 */
202 		if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_17)
203 			args->swizzle_mode = I915_BIT_6_SWIZZLE_9;
204 		if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_10_17)
205 			args->swizzle_mode = I915_BIT_6_SWIZZLE_9_10;
206 
207 		/* If we can't handle the swizzling, make it untiled. */
208 		if (args->swizzle_mode == I915_BIT_6_SWIZZLE_UNKNOWN) {
209 			args->tiling_mode = I915_TILING_NONE;
210 			args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
211 			args->stride = 0;
212 		}
213 	}
214 
215 	if (args->tiling_mode != obj->tiling_mode ||
216 	    args->stride != obj->stride) {
217 		/* We need to rebind the object if its current allocation
218 		 * no longer meets the alignment restrictions for its new
219 		 * tiling mode. Otherwise we can just leave it alone, but
220 		 * need to ensure that any fence register is updated before
221 		 * the next fenced (either through the GTT or by the BLT unit
222 		 * on older GPUs) access.
223 		 *
224 		 * After updating the tiling parameters, we then flag whether
225 		 * we need to update an associated fence register. Note this
226 		 * has to also include the unfenced register the GPU uses
227 		 * whilst executing a fenced command for an untiled object.
228 		 */
229 		if (obj->map_and_fenceable &&
230 		    !i915_gem_object_fence_ok(obj, args->tiling_mode))
231 			ret = i915_gem_object_ggtt_unbind(obj);
232 
233 		if (ret == 0) {
234 			if (obj->pages &&
235 			    obj->madv == I915_MADV_WILLNEED &&
236 			    dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES) {
237 				if (args->tiling_mode == I915_TILING_NONE)
238 					i915_gem_object_unpin_pages(obj);
239 				if (obj->tiling_mode == I915_TILING_NONE)
240 					i915_gem_object_pin_pages(obj);
241 			}
242 
243 			obj->fence_dirty =
244 				obj->last_fenced_req ||
245 				obj->fence_reg != I915_FENCE_REG_NONE;
246 
247 			obj->tiling_mode = args->tiling_mode;
248 			obj->stride = args->stride;
249 
250 			/* Force the fence to be reacquired for GTT access */
251 			i915_gem_release_mmap(obj);
252 		}
253 	}
254 	/* we have to maintain this existing ABI... */
255 	args->stride = obj->stride;
256 	args->tiling_mode = obj->tiling_mode;
257 
258 	/* Try to preallocate memory required to save swizzling on put-pages */
259 	if (i915_gem_object_needs_bit17_swizzle(obj)) {
260 		if (obj->bit_17 == NULL) {
261 			obj->bit_17 = kcalloc(BITS_TO_LONGS(obj->base.size >> PAGE_SHIFT),
262 					      sizeof(long), GFP_KERNEL);
263 		}
264 	} else {
265 		kfree(obj->bit_17);
266 		obj->bit_17 = NULL;
267 	}
268 
269 err:
270 	drm_gem_object_unreference(&obj->base);
271 	mutex_unlock(&dev->struct_mutex);
272 
273 	intel_runtime_pm_put(dev_priv);
274 
275 	return ret;
276 }
277 
278 /**
279  * i915_gem_get_tiling - IOCTL handler to get tiling mode
280  * @dev: DRM device
281  * @data: data pointer for the ioctl
282  * @file: DRM file for the ioctl call
283  *
284  * Returns the current tiling mode and required bit 6 swizzling for the object.
285  *
286  * Called by the user via ioctl.
287  *
288  * Returns:
289  * Zero on success, negative errno on failure.
290  */
291 int
292 i915_gem_get_tiling(struct drm_device *dev, void *data,
293 		   struct drm_file *file)
294 {
295 	struct drm_i915_gem_get_tiling *args = data;
296 	struct drm_i915_private *dev_priv = dev->dev_private;
297 	struct drm_i915_gem_object *obj;
298 
299 	obj = to_intel_bo(drm_gem_object_lookup(file, args->handle));
300 	if (&obj->base == NULL)
301 		return -ENOENT;
302 
303 	mutex_lock(&dev->struct_mutex);
304 
305 	args->tiling_mode = obj->tiling_mode;
306 	switch (obj->tiling_mode) {
307 	case I915_TILING_X:
308 		args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
309 		break;
310 	case I915_TILING_Y:
311 		args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y;
312 		break;
313 	case I915_TILING_NONE:
314 		args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
315 		break;
316 	default:
317 		DRM_ERROR("unknown tiling mode\n");
318 	}
319 
320 	/* Hide bit 17 from the user -- see comment in i915_gem_set_tiling */
321 	if (dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES)
322 		args->phys_swizzle_mode = I915_BIT_6_SWIZZLE_UNKNOWN;
323 	else
324 		args->phys_swizzle_mode = args->swizzle_mode;
325 	if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_17)
326 		args->swizzle_mode = I915_BIT_6_SWIZZLE_9;
327 	if (args->swizzle_mode == I915_BIT_6_SWIZZLE_9_10_17)
328 		args->swizzle_mode = I915_BIT_6_SWIZZLE_9_10;
329 
330 	drm_gem_object_unreference(&obj->base);
331 	mutex_unlock(&dev->struct_mutex);
332 
333 	return 0;
334 }
335