xref: /linux/drivers/gpu/drm/vmwgfx/vmwgfx_surface.c (revision 7318234c8d7c0f209f993ee46a7ea148efdb28b9)
1 // SPDX-License-Identifier: GPL-2.0 OR MIT
2 /**************************************************************************
3  *
4  * Copyright 2009-2015 VMware, Inc., Palo Alto, CA., USA
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the
8  * "Software"), to deal in the Software without restriction, including
9  * without limitation the rights to use, copy, modify, merge, publish,
10  * distribute, sub license, and/or sell copies of the Software, and to
11  * permit persons to whom the Software is furnished to do so, subject to
12  * the following conditions:
13  *
14  * The above copyright notice and this permission notice (including the
15  * next paragraph) shall be included in all copies or substantial portions
16  * of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
22  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24  * USE OR OTHER DEALINGS IN THE SOFTWARE.
25  *
26  **************************************************************************/
27 
28 #include <drm/ttm/ttm_placement.h>
29 
30 #include "vmwgfx_drv.h"
31 #include "vmwgfx_resource_priv.h"
32 #include "vmwgfx_so.h"
33 #include "vmwgfx_binding.h"
34 #include "device_include/svga3d_surfacedefs.h"
35 
36 #define SVGA3D_FLAGS_64(upper32, lower32) (((uint64_t)upper32 << 32) | lower32)
37 #define SVGA3D_FLAGS_UPPER_32(svga3d_flags) (svga3d_flags >> 32)
38 #define SVGA3D_FLAGS_LOWER_32(svga3d_flags) \
39 	(svga3d_flags & ((uint64_t)U32_MAX))
40 
41 /**
42  * struct vmw_user_surface - User-space visible surface resource
43  *
44  * @base:           The TTM base object handling user-space visibility.
45  * @srf:            The surface metadata.
46  * @size:           TTM accounting size for the surface.
47  * @master: master of the creating client. Used for security check.
48  */
49 struct vmw_user_surface {
50 	struct ttm_prime_object prime;
51 	struct vmw_surface srf;
52 	uint32_t size;
53 	struct drm_master *master;
54 	struct ttm_base_object *backup_base;
55 };
56 
57 /**
58  * struct vmw_surface_offset - Backing store mip level offset info
59  *
60  * @face:           Surface face.
61  * @mip:            Mip level.
62  * @bo_offset:      Offset into backing store of this mip level.
63  *
64  */
65 struct vmw_surface_offset {
66 	uint32_t face;
67 	uint32_t mip;
68 	uint32_t bo_offset;
69 };
70 
71 /**
72  * vmw_surface_dirty - Surface dirty-tracker
73  * @cache: Cached layout information of the surface.
74  * @size: Accounting size for the struct vmw_surface_dirty.
75  * @num_subres: Number of subresources.
76  * @boxes: Array of SVGA3dBoxes indicating dirty regions. One per subresource.
77  */
78 struct vmw_surface_dirty {
79 	struct svga3dsurface_cache cache;
80 	size_t size;
81 	u32 num_subres;
82 	SVGA3dBox boxes[];
83 };
84 
85 static void vmw_user_surface_free(struct vmw_resource *res);
86 static struct vmw_resource *
87 vmw_user_surface_base_to_res(struct ttm_base_object *base);
88 static int vmw_legacy_srf_bind(struct vmw_resource *res,
89 			       struct ttm_validate_buffer *val_buf);
90 static int vmw_legacy_srf_unbind(struct vmw_resource *res,
91 				 bool readback,
92 				 struct ttm_validate_buffer *val_buf);
93 static int vmw_legacy_srf_create(struct vmw_resource *res);
94 static int vmw_legacy_srf_destroy(struct vmw_resource *res);
95 static int vmw_gb_surface_create(struct vmw_resource *res);
96 static int vmw_gb_surface_bind(struct vmw_resource *res,
97 			       struct ttm_validate_buffer *val_buf);
98 static int vmw_gb_surface_unbind(struct vmw_resource *res,
99 				 bool readback,
100 				 struct ttm_validate_buffer *val_buf);
101 static int vmw_gb_surface_destroy(struct vmw_resource *res);
102 static int
103 vmw_gb_surface_define_internal(struct drm_device *dev,
104 			       struct drm_vmw_gb_surface_create_ext_req *req,
105 			       struct drm_vmw_gb_surface_create_rep *rep,
106 			       struct drm_file *file_priv);
107 static int
108 vmw_gb_surface_reference_internal(struct drm_device *dev,
109 				  struct drm_vmw_surface_arg *req,
110 				  struct drm_vmw_gb_surface_ref_ext_rep *rep,
111 				  struct drm_file *file_priv);
112 
113 static void vmw_surface_dirty_free(struct vmw_resource *res);
114 static int vmw_surface_dirty_alloc(struct vmw_resource *res);
115 static int vmw_surface_dirty_sync(struct vmw_resource *res);
116 static void vmw_surface_dirty_range_add(struct vmw_resource *res, size_t start,
117 					size_t end);
118 static int vmw_surface_clean(struct vmw_resource *res);
119 
120 static const struct vmw_user_resource_conv user_surface_conv = {
121 	.object_type = VMW_RES_SURFACE,
122 	.base_obj_to_res = vmw_user_surface_base_to_res,
123 	.res_free = vmw_user_surface_free
124 };
125 
126 const struct vmw_user_resource_conv *user_surface_converter =
127 	&user_surface_conv;
128 
129 
130 static uint64_t vmw_user_surface_size;
131 
132 static const struct vmw_res_func vmw_legacy_surface_func = {
133 	.res_type = vmw_res_surface,
134 	.needs_backup = false,
135 	.may_evict = true,
136 	.prio = 1,
137 	.dirty_prio = 1,
138 	.type_name = "legacy surfaces",
139 	.backup_placement = &vmw_srf_placement,
140 	.create = &vmw_legacy_srf_create,
141 	.destroy = &vmw_legacy_srf_destroy,
142 	.bind = &vmw_legacy_srf_bind,
143 	.unbind = &vmw_legacy_srf_unbind
144 };
145 
146 static const struct vmw_res_func vmw_gb_surface_func = {
147 	.res_type = vmw_res_surface,
148 	.needs_backup = true,
149 	.may_evict = true,
150 	.prio = 1,
151 	.dirty_prio = 2,
152 	.type_name = "guest backed surfaces",
153 	.backup_placement = &vmw_mob_placement,
154 	.create = vmw_gb_surface_create,
155 	.destroy = vmw_gb_surface_destroy,
156 	.bind = vmw_gb_surface_bind,
157 	.unbind = vmw_gb_surface_unbind,
158 	.dirty_alloc = vmw_surface_dirty_alloc,
159 	.dirty_free = vmw_surface_dirty_free,
160 	.dirty_sync = vmw_surface_dirty_sync,
161 	.dirty_range_add = vmw_surface_dirty_range_add,
162 	.clean = vmw_surface_clean,
163 };
164 
165 /**
166  * struct vmw_surface_dma - SVGA3D DMA command
167  */
168 struct vmw_surface_dma {
169 	SVGA3dCmdHeader header;
170 	SVGA3dCmdSurfaceDMA body;
171 	SVGA3dCopyBox cb;
172 	SVGA3dCmdSurfaceDMASuffix suffix;
173 };
174 
175 /**
176  * struct vmw_surface_define - SVGA3D Surface Define command
177  */
178 struct vmw_surface_define {
179 	SVGA3dCmdHeader header;
180 	SVGA3dCmdDefineSurface body;
181 };
182 
183 /**
184  * struct vmw_surface_destroy - SVGA3D Surface Destroy command
185  */
186 struct vmw_surface_destroy {
187 	SVGA3dCmdHeader header;
188 	SVGA3dCmdDestroySurface body;
189 };
190 
191 
192 /**
193  * vmw_surface_dma_size - Compute fifo size for a dma command.
194  *
195  * @srf: Pointer to a struct vmw_surface
196  *
197  * Computes the required size for a surface dma command for backup or
198  * restoration of the surface represented by @srf.
199  */
200 static inline uint32_t vmw_surface_dma_size(const struct vmw_surface *srf)
201 {
202 	return srf->metadata.num_sizes * sizeof(struct vmw_surface_dma);
203 }
204 
205 
206 /**
207  * vmw_surface_define_size - Compute fifo size for a surface define command.
208  *
209  * @srf: Pointer to a struct vmw_surface
210  *
211  * Computes the required size for a surface define command for the definition
212  * of the surface represented by @srf.
213  */
214 static inline uint32_t vmw_surface_define_size(const struct vmw_surface *srf)
215 {
216 	return sizeof(struct vmw_surface_define) + srf->metadata.num_sizes *
217 		sizeof(SVGA3dSize);
218 }
219 
220 
221 /**
222  * vmw_surface_destroy_size - Compute fifo size for a surface destroy command.
223  *
224  * Computes the required size for a surface destroy command for the destruction
225  * of a hw surface.
226  */
227 static inline uint32_t vmw_surface_destroy_size(void)
228 {
229 	return sizeof(struct vmw_surface_destroy);
230 }
231 
232 /**
233  * vmw_surface_destroy_encode - Encode a surface_destroy command.
234  *
235  * @id: The surface id
236  * @cmd_space: Pointer to memory area in which the commands should be encoded.
237  */
238 static void vmw_surface_destroy_encode(uint32_t id,
239 				       void *cmd_space)
240 {
241 	struct vmw_surface_destroy *cmd = (struct vmw_surface_destroy *)
242 		cmd_space;
243 
244 	cmd->header.id = SVGA_3D_CMD_SURFACE_DESTROY;
245 	cmd->header.size = sizeof(cmd->body);
246 	cmd->body.sid = id;
247 }
248 
249 /**
250  * vmw_surface_define_encode - Encode a surface_define command.
251  *
252  * @srf: Pointer to a struct vmw_surface object.
253  * @cmd_space: Pointer to memory area in which the commands should be encoded.
254  */
255 static void vmw_surface_define_encode(const struct vmw_surface *srf,
256 				      void *cmd_space)
257 {
258 	struct vmw_surface_define *cmd = (struct vmw_surface_define *)
259 		cmd_space;
260 	struct drm_vmw_size *src_size;
261 	SVGA3dSize *cmd_size;
262 	uint32_t cmd_len;
263 	int i;
264 
265 	cmd_len = sizeof(cmd->body) + srf->metadata.num_sizes *
266 		sizeof(SVGA3dSize);
267 
268 	cmd->header.id = SVGA_3D_CMD_SURFACE_DEFINE;
269 	cmd->header.size = cmd_len;
270 	cmd->body.sid = srf->res.id;
271 	/*
272 	 * Downcast of surfaceFlags, was upcasted when received from user-space,
273 	 * since driver internally stores as 64 bit.
274 	 * For legacy surface define only 32 bit flag is supported.
275 	 */
276 	cmd->body.surfaceFlags = (SVGA3dSurface1Flags)srf->metadata.flags;
277 	cmd->body.format = srf->metadata.format;
278 	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i)
279 		cmd->body.face[i].numMipLevels = srf->metadata.mip_levels[i];
280 
281 	cmd += 1;
282 	cmd_size = (SVGA3dSize *) cmd;
283 	src_size = srf->metadata.sizes;
284 
285 	for (i = 0; i < srf->metadata.num_sizes; ++i, cmd_size++, src_size++) {
286 		cmd_size->width = src_size->width;
287 		cmd_size->height = src_size->height;
288 		cmd_size->depth = src_size->depth;
289 	}
290 }
291 
292 /**
293  * vmw_surface_dma_encode - Encode a surface_dma command.
294  *
295  * @srf: Pointer to a struct vmw_surface object.
296  * @cmd_space: Pointer to memory area in which the commands should be encoded.
297  * @ptr: Pointer to an SVGAGuestPtr indicating where the surface contents
298  * should be placed or read from.
299  * @to_surface: Boolean whether to DMA to the surface or from the surface.
300  */
301 static void vmw_surface_dma_encode(struct vmw_surface *srf,
302 				   void *cmd_space,
303 				   const SVGAGuestPtr *ptr,
304 				   bool to_surface)
305 {
306 	uint32_t i;
307 	struct vmw_surface_dma *cmd = (struct vmw_surface_dma *)cmd_space;
308 	const struct svga3d_surface_desc *desc =
309 		svga3dsurface_get_desc(srf->metadata.format);
310 
311 	for (i = 0; i < srf->metadata.num_sizes; ++i) {
312 		SVGA3dCmdHeader *header = &cmd->header;
313 		SVGA3dCmdSurfaceDMA *body = &cmd->body;
314 		SVGA3dCopyBox *cb = &cmd->cb;
315 		SVGA3dCmdSurfaceDMASuffix *suffix = &cmd->suffix;
316 		const struct vmw_surface_offset *cur_offset = &srf->offsets[i];
317 		const struct drm_vmw_size *cur_size = &srf->metadata.sizes[i];
318 
319 		header->id = SVGA_3D_CMD_SURFACE_DMA;
320 		header->size = sizeof(*body) + sizeof(*cb) + sizeof(*suffix);
321 
322 		body->guest.ptr = *ptr;
323 		body->guest.ptr.offset += cur_offset->bo_offset;
324 		body->guest.pitch = svga3dsurface_calculate_pitch(desc,
325 								  cur_size);
326 		body->host.sid = srf->res.id;
327 		body->host.face = cur_offset->face;
328 		body->host.mipmap = cur_offset->mip;
329 		body->transfer = ((to_surface) ?  SVGA3D_WRITE_HOST_VRAM :
330 				  SVGA3D_READ_HOST_VRAM);
331 		cb->x = 0;
332 		cb->y = 0;
333 		cb->z = 0;
334 		cb->srcx = 0;
335 		cb->srcy = 0;
336 		cb->srcz = 0;
337 		cb->w = cur_size->width;
338 		cb->h = cur_size->height;
339 		cb->d = cur_size->depth;
340 
341 		suffix->suffixSize = sizeof(*suffix);
342 		suffix->maximumOffset =
343 			svga3dsurface_get_image_buffer_size(desc, cur_size,
344 							    body->guest.pitch);
345 		suffix->flags.discard = 0;
346 		suffix->flags.unsynchronized = 0;
347 		suffix->flags.reserved = 0;
348 		++cmd;
349 	}
350 };
351 
352 
353 /**
354  * vmw_hw_surface_destroy - destroy a Device surface
355  *
356  * @res:        Pointer to a struct vmw_resource embedded in a struct
357  *              vmw_surface.
358  *
359  * Destroys a the device surface associated with a struct vmw_surface if
360  * any, and adjusts accounting and resource count accordingly.
361  */
362 static void vmw_hw_surface_destroy(struct vmw_resource *res)
363 {
364 
365 	struct vmw_private *dev_priv = res->dev_priv;
366 	void *cmd;
367 
368 	if (res->func->destroy == vmw_gb_surface_destroy) {
369 		(void) vmw_gb_surface_destroy(res);
370 		return;
371 	}
372 
373 	if (res->id != -1) {
374 
375 		cmd = VMW_FIFO_RESERVE(dev_priv, vmw_surface_destroy_size());
376 		if (unlikely(!cmd))
377 			return;
378 
379 		vmw_surface_destroy_encode(res->id, cmd);
380 		vmw_fifo_commit(dev_priv, vmw_surface_destroy_size());
381 
382 		/*
383 		 * used_memory_size_atomic, or separate lock
384 		 * to avoid taking dev_priv::cmdbuf_mutex in
385 		 * the destroy path.
386 		 */
387 
388 		mutex_lock(&dev_priv->cmdbuf_mutex);
389 		dev_priv->used_memory_size -= res->backup_size;
390 		mutex_unlock(&dev_priv->cmdbuf_mutex);
391 	}
392 }
393 
394 /**
395  * vmw_legacy_srf_create - Create a device surface as part of the
396  * resource validation process.
397  *
398  * @res: Pointer to a struct vmw_surface.
399  *
400  * If the surface doesn't have a hw id.
401  *
402  * Returns -EBUSY if there wasn't sufficient device resources to
403  * complete the validation. Retry after freeing up resources.
404  *
405  * May return other errors if the kernel is out of guest resources.
406  */
407 static int vmw_legacy_srf_create(struct vmw_resource *res)
408 {
409 	struct vmw_private *dev_priv = res->dev_priv;
410 	struct vmw_surface *srf;
411 	uint32_t submit_size;
412 	uint8_t *cmd;
413 	int ret;
414 
415 	if (likely(res->id != -1))
416 		return 0;
417 
418 	srf = vmw_res_to_srf(res);
419 	if (unlikely(dev_priv->used_memory_size + res->backup_size >=
420 		     dev_priv->memory_size))
421 		return -EBUSY;
422 
423 	/*
424 	 * Alloc id for the resource.
425 	 */
426 
427 	ret = vmw_resource_alloc_id(res);
428 	if (unlikely(ret != 0)) {
429 		DRM_ERROR("Failed to allocate a surface id.\n");
430 		goto out_no_id;
431 	}
432 
433 	if (unlikely(res->id >= SVGA3D_MAX_SURFACE_IDS)) {
434 		ret = -EBUSY;
435 		goto out_no_fifo;
436 	}
437 
438 	/*
439 	 * Encode surface define- commands.
440 	 */
441 
442 	submit_size = vmw_surface_define_size(srf);
443 	cmd = VMW_FIFO_RESERVE(dev_priv, submit_size);
444 	if (unlikely(!cmd)) {
445 		ret = -ENOMEM;
446 		goto out_no_fifo;
447 	}
448 
449 	vmw_surface_define_encode(srf, cmd);
450 	vmw_fifo_commit(dev_priv, submit_size);
451 	vmw_fifo_resource_inc(dev_priv);
452 
453 	/*
454 	 * Surface memory usage accounting.
455 	 */
456 
457 	dev_priv->used_memory_size += res->backup_size;
458 	return 0;
459 
460 out_no_fifo:
461 	vmw_resource_release_id(res);
462 out_no_id:
463 	return ret;
464 }
465 
466 /**
467  * vmw_legacy_srf_dma - Copy backup data to or from a legacy surface.
468  *
469  * @res:            Pointer to a struct vmw_res embedded in a struct
470  *                  vmw_surface.
471  * @val_buf:        Pointer to a struct ttm_validate_buffer containing
472  *                  information about the backup buffer.
473  * @bind:           Boolean wether to DMA to the surface.
474  *
475  * Transfer backup data to or from a legacy surface as part of the
476  * validation process.
477  * May return other errors if the kernel is out of guest resources.
478  * The backup buffer will be fenced or idle upon successful completion,
479  * and if the surface needs persistent backup storage, the backup buffer
480  * will also be returned reserved iff @bind is true.
481  */
482 static int vmw_legacy_srf_dma(struct vmw_resource *res,
483 			      struct ttm_validate_buffer *val_buf,
484 			      bool bind)
485 {
486 	SVGAGuestPtr ptr;
487 	struct vmw_fence_obj *fence;
488 	uint32_t submit_size;
489 	struct vmw_surface *srf = vmw_res_to_srf(res);
490 	uint8_t *cmd;
491 	struct vmw_private *dev_priv = res->dev_priv;
492 
493 	BUG_ON(!val_buf->bo);
494 	submit_size = vmw_surface_dma_size(srf);
495 	cmd = VMW_FIFO_RESERVE(dev_priv, submit_size);
496 	if (unlikely(!cmd))
497 		return -ENOMEM;
498 
499 	vmw_bo_get_guest_ptr(val_buf->bo, &ptr);
500 	vmw_surface_dma_encode(srf, cmd, &ptr, bind);
501 
502 	vmw_fifo_commit(dev_priv, submit_size);
503 
504 	/*
505 	 * Create a fence object and fence the backup buffer.
506 	 */
507 
508 	(void) vmw_execbuf_fence_commands(NULL, dev_priv,
509 					  &fence, NULL);
510 
511 	vmw_bo_fence_single(val_buf->bo, fence);
512 
513 	if (likely(fence != NULL))
514 		vmw_fence_obj_unreference(&fence);
515 
516 	return 0;
517 }
518 
519 /**
520  * vmw_legacy_srf_bind - Perform a legacy surface bind as part of the
521  *                       surface validation process.
522  *
523  * @res:            Pointer to a struct vmw_res embedded in a struct
524  *                  vmw_surface.
525  * @val_buf:        Pointer to a struct ttm_validate_buffer containing
526  *                  information about the backup buffer.
527  *
528  * This function will copy backup data to the surface if the
529  * backup buffer is dirty.
530  */
531 static int vmw_legacy_srf_bind(struct vmw_resource *res,
532 			       struct ttm_validate_buffer *val_buf)
533 {
534 	if (!res->backup_dirty)
535 		return 0;
536 
537 	return vmw_legacy_srf_dma(res, val_buf, true);
538 }
539 
540 
541 /**
542  * vmw_legacy_srf_unbind - Perform a legacy surface unbind as part of the
543  *                         surface eviction process.
544  *
545  * @res:            Pointer to a struct vmw_res embedded in a struct
546  *                  vmw_surface.
547  * @val_buf:        Pointer to a struct ttm_validate_buffer containing
548  *                  information about the backup buffer.
549  *
550  * This function will copy backup data from the surface.
551  */
552 static int vmw_legacy_srf_unbind(struct vmw_resource *res,
553 				 bool readback,
554 				 struct ttm_validate_buffer *val_buf)
555 {
556 	if (unlikely(readback))
557 		return vmw_legacy_srf_dma(res, val_buf, false);
558 	return 0;
559 }
560 
561 /**
562  * vmw_legacy_srf_destroy - Destroy a device surface as part of a
563  *                          resource eviction process.
564  *
565  * @res:            Pointer to a struct vmw_res embedded in a struct
566  *                  vmw_surface.
567  */
568 static int vmw_legacy_srf_destroy(struct vmw_resource *res)
569 {
570 	struct vmw_private *dev_priv = res->dev_priv;
571 	uint32_t submit_size;
572 	uint8_t *cmd;
573 
574 	BUG_ON(res->id == -1);
575 
576 	/*
577 	 * Encode the dma- and surface destroy commands.
578 	 */
579 
580 	submit_size = vmw_surface_destroy_size();
581 	cmd = VMW_FIFO_RESERVE(dev_priv, submit_size);
582 	if (unlikely(!cmd))
583 		return -ENOMEM;
584 
585 	vmw_surface_destroy_encode(res->id, cmd);
586 	vmw_fifo_commit(dev_priv, submit_size);
587 
588 	/*
589 	 * Surface memory usage accounting.
590 	 */
591 
592 	dev_priv->used_memory_size -= res->backup_size;
593 
594 	/*
595 	 * Release the surface ID.
596 	 */
597 
598 	vmw_resource_release_id(res);
599 	vmw_fifo_resource_dec(dev_priv);
600 
601 	return 0;
602 }
603 
604 
605 /**
606  * vmw_surface_init - initialize a struct vmw_surface
607  *
608  * @dev_priv:       Pointer to a device private struct.
609  * @srf:            Pointer to the struct vmw_surface to initialize.
610  * @res_free:       Pointer to a resource destructor used to free
611  *                  the object.
612  */
613 static int vmw_surface_init(struct vmw_private *dev_priv,
614 			    struct vmw_surface *srf,
615 			    void (*res_free) (struct vmw_resource *res))
616 {
617 	int ret;
618 	struct vmw_resource *res = &srf->res;
619 
620 	BUG_ON(!res_free);
621 	ret = vmw_resource_init(dev_priv, res, true, res_free,
622 				(dev_priv->has_mob) ? &vmw_gb_surface_func :
623 				&vmw_legacy_surface_func);
624 
625 	if (unlikely(ret != 0)) {
626 		res_free(res);
627 		return ret;
628 	}
629 
630 	/*
631 	 * The surface won't be visible to hardware until a
632 	 * surface validate.
633 	 */
634 
635 	INIT_LIST_HEAD(&srf->view_list);
636 	res->hw_destroy = vmw_hw_surface_destroy;
637 	return ret;
638 }
639 
640 /**
641  * vmw_user_surface_base_to_res - TTM base object to resource converter for
642  *                                user visible surfaces
643  *
644  * @base:           Pointer to a TTM base object
645  *
646  * Returns the struct vmw_resource embedded in a struct vmw_surface
647  * for the user-visible object identified by the TTM base object @base.
648  */
649 static struct vmw_resource *
650 vmw_user_surface_base_to_res(struct ttm_base_object *base)
651 {
652 	return &(container_of(base, struct vmw_user_surface,
653 			      prime.base)->srf.res);
654 }
655 
656 /**
657  * vmw_user_surface_free - User visible surface resource destructor
658  *
659  * @res:            A struct vmw_resource embedded in a struct vmw_surface.
660  */
661 static void vmw_user_surface_free(struct vmw_resource *res)
662 {
663 	struct vmw_surface *srf = vmw_res_to_srf(res);
664 	struct vmw_user_surface *user_srf =
665 	    container_of(srf, struct vmw_user_surface, srf);
666 	struct vmw_private *dev_priv = srf->res.dev_priv;
667 	uint32_t size = user_srf->size;
668 
669 	WARN_ON_ONCE(res->dirty);
670 	if (user_srf->master)
671 		drm_master_put(&user_srf->master);
672 	kfree(srf->offsets);
673 	kfree(srf->metadata.sizes);
674 	kfree(srf->snooper.image);
675 	ttm_prime_object_kfree(user_srf, prime);
676 	ttm_mem_global_free(vmw_mem_glob(dev_priv), size);
677 }
678 
679 /**
680  * vmw_user_surface_free - User visible surface TTM base object destructor
681  *
682  * @p_base:         Pointer to a pointer to a TTM base object
683  *                  embedded in a struct vmw_user_surface.
684  *
685  * Drops the base object's reference on its resource, and the
686  * pointer pointed to by *p_base is set to NULL.
687  */
688 static void vmw_user_surface_base_release(struct ttm_base_object **p_base)
689 {
690 	struct ttm_base_object *base = *p_base;
691 	struct vmw_user_surface *user_srf =
692 	    container_of(base, struct vmw_user_surface, prime.base);
693 	struct vmw_resource *res = &user_srf->srf.res;
694 
695 	*p_base = NULL;
696 	if (user_srf->backup_base)
697 		ttm_base_object_unref(&user_srf->backup_base);
698 	vmw_resource_unreference(&res);
699 }
700 
701 /**
702  * vmw_user_surface_destroy_ioctl - Ioctl function implementing
703  *                                  the user surface destroy functionality.
704  *
705  * @dev:            Pointer to a struct drm_device.
706  * @data:           Pointer to data copied from / to user-space.
707  * @file_priv:      Pointer to a drm file private structure.
708  */
709 int vmw_surface_destroy_ioctl(struct drm_device *dev, void *data,
710 			      struct drm_file *file_priv)
711 {
712 	struct drm_vmw_surface_arg *arg = (struct drm_vmw_surface_arg *)data;
713 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
714 
715 	return ttm_ref_object_base_unref(tfile, arg->sid, TTM_REF_USAGE);
716 }
717 
718 /**
719  * vmw_user_surface_define_ioctl - Ioctl function implementing
720  *                                  the user surface define functionality.
721  *
722  * @dev:            Pointer to a struct drm_device.
723  * @data:           Pointer to data copied from / to user-space.
724  * @file_priv:      Pointer to a drm file private structure.
725  */
726 int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
727 			     struct drm_file *file_priv)
728 {
729 	struct vmw_private *dev_priv = vmw_priv(dev);
730 	struct vmw_user_surface *user_srf;
731 	struct vmw_surface *srf;
732 	struct vmw_surface_metadata *metadata;
733 	struct vmw_resource *res;
734 	struct vmw_resource *tmp;
735 	union drm_vmw_surface_create_arg *arg =
736 	    (union drm_vmw_surface_create_arg *)data;
737 	struct drm_vmw_surface_create_req *req = &arg->req;
738 	struct drm_vmw_surface_arg *rep = &arg->rep;
739 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
740 	struct ttm_operation_ctx ctx = {
741 		.interruptible = true,
742 		.no_wait_gpu = false
743 	};
744 	int ret;
745 	int i, j;
746 	uint32_t cur_bo_offset;
747 	struct drm_vmw_size *cur_size;
748 	struct vmw_surface_offset *cur_offset;
749 	uint32_t num_sizes;
750 	uint32_t size;
751 	const struct svga3d_surface_desc *desc;
752 
753 	if (unlikely(vmw_user_surface_size == 0))
754 		vmw_user_surface_size = ttm_round_pot(sizeof(*user_srf)) +
755 			VMW_IDA_ACC_SIZE + TTM_OBJ_EXTRA_SIZE;
756 
757 	num_sizes = 0;
758 	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) {
759 		if (req->mip_levels[i] > DRM_VMW_MAX_MIP_LEVELS)
760 			return -EINVAL;
761 		num_sizes += req->mip_levels[i];
762 	}
763 
764 	if (num_sizes > DRM_VMW_MAX_SURFACE_FACES * DRM_VMW_MAX_MIP_LEVELS ||
765 	    num_sizes == 0)
766 		return -EINVAL;
767 
768 	size = vmw_user_surface_size +
769 		ttm_round_pot(num_sizes * sizeof(struct drm_vmw_size)) +
770 		ttm_round_pot(num_sizes * sizeof(struct vmw_surface_offset));
771 
772 	desc = svga3dsurface_get_desc(req->format);
773 	if (unlikely(desc->block_desc == SVGA3DBLOCKDESC_NONE)) {
774 		VMW_DEBUG_USER("Invalid format %d for surface creation.\n",
775 			       req->format);
776 		return -EINVAL;
777 	}
778 
779 	ret = ttm_read_lock(&dev_priv->reservation_sem, true);
780 	if (unlikely(ret != 0))
781 		return ret;
782 
783 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
784 				   size, &ctx);
785 	if (unlikely(ret != 0)) {
786 		if (ret != -ERESTARTSYS)
787 			DRM_ERROR("Out of graphics memory for surface.\n");
788 		goto out_unlock;
789 	}
790 
791 	user_srf = kzalloc(sizeof(*user_srf), GFP_KERNEL);
792 	if (unlikely(!user_srf)) {
793 		ret = -ENOMEM;
794 		goto out_no_user_srf;
795 	}
796 
797 	srf = &user_srf->srf;
798 	metadata = &srf->metadata;
799 	res = &srf->res;
800 
801 	/* Driver internally stores as 64-bit flags */
802 	metadata->flags = (SVGA3dSurfaceAllFlags)req->flags;
803 	metadata->format = req->format;
804 	metadata->scanout = req->scanout;
805 
806 	memcpy(metadata->mip_levels, req->mip_levels,
807 	       sizeof(metadata->mip_levels));
808 	metadata->num_sizes = num_sizes;
809 	user_srf->size = size;
810 	metadata->sizes =
811 		memdup_user((struct drm_vmw_size __user *)(unsigned long)
812 			    req->size_addr,
813 			    sizeof(*metadata->sizes) * metadata->num_sizes);
814 	if (IS_ERR(metadata->sizes)) {
815 		ret = PTR_ERR(metadata->sizes);
816 		goto out_no_sizes;
817 	}
818 	srf->offsets = kmalloc_array(metadata->num_sizes, sizeof(*srf->offsets),
819 				     GFP_KERNEL);
820 	if (unlikely(!srf->offsets)) {
821 		ret = -ENOMEM;
822 		goto out_no_offsets;
823 	}
824 
825 	metadata->base_size = *srf->metadata.sizes;
826 	metadata->autogen_filter = SVGA3D_TEX_FILTER_NONE;
827 	metadata->multisample_count = 0;
828 	metadata->multisample_pattern = SVGA3D_MS_PATTERN_NONE;
829 	metadata->quality_level = SVGA3D_MS_QUALITY_NONE;
830 
831 	cur_bo_offset = 0;
832 	cur_offset = srf->offsets;
833 	cur_size = metadata->sizes;
834 
835 	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) {
836 		for (j = 0; j < metadata->mip_levels[i]; ++j) {
837 			uint32_t stride = svga3dsurface_calculate_pitch
838 				(desc, cur_size);
839 
840 			cur_offset->face = i;
841 			cur_offset->mip = j;
842 			cur_offset->bo_offset = cur_bo_offset;
843 			cur_bo_offset += svga3dsurface_get_image_buffer_size
844 				(desc, cur_size, stride);
845 			++cur_offset;
846 			++cur_size;
847 		}
848 	}
849 	res->backup_size = cur_bo_offset;
850 	if (metadata->scanout &&
851 	    metadata->num_sizes == 1 &&
852 	    metadata->sizes[0].width == 64 &&
853 	    metadata->sizes[0].height == 64 &&
854 	    metadata->format == SVGA3D_A8R8G8B8) {
855 
856 		srf->snooper.image = kzalloc(64 * 64 * 4, GFP_KERNEL);
857 		if (!srf->snooper.image) {
858 			DRM_ERROR("Failed to allocate cursor_image\n");
859 			ret = -ENOMEM;
860 			goto out_no_copy;
861 		}
862 	} else {
863 		srf->snooper.image = NULL;
864 	}
865 
866 	user_srf->prime.base.shareable = false;
867 	user_srf->prime.base.tfile = NULL;
868 	if (drm_is_primary_client(file_priv))
869 		user_srf->master = drm_master_get(file_priv->master);
870 
871 	/**
872 	 * From this point, the generic resource management functions
873 	 * destroy the object on failure.
874 	 */
875 
876 	ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free);
877 	if (unlikely(ret != 0))
878 		goto out_unlock;
879 
880 	/*
881 	 * A gb-aware client referencing a shared surface will
882 	 * expect a backup buffer to be present.
883 	 */
884 	if (dev_priv->has_mob && req->shareable) {
885 		uint32_t backup_handle;
886 
887 		ret = vmw_user_bo_alloc(dev_priv, tfile,
888 					res->backup_size,
889 					true,
890 					&backup_handle,
891 					&res->backup,
892 					&user_srf->backup_base);
893 		if (unlikely(ret != 0)) {
894 			vmw_resource_unreference(&res);
895 			goto out_unlock;
896 		}
897 	}
898 
899 	tmp = vmw_resource_reference(&srf->res);
900 	ret = ttm_prime_object_init(tfile, res->backup_size, &user_srf->prime,
901 				    req->shareable, VMW_RES_SURFACE,
902 				    &vmw_user_surface_base_release, NULL);
903 
904 	if (unlikely(ret != 0)) {
905 		vmw_resource_unreference(&tmp);
906 		vmw_resource_unreference(&res);
907 		goto out_unlock;
908 	}
909 
910 	rep->sid = user_srf->prime.base.handle;
911 	vmw_resource_unreference(&res);
912 
913 	ttm_read_unlock(&dev_priv->reservation_sem);
914 	return 0;
915 out_no_copy:
916 	kfree(srf->offsets);
917 out_no_offsets:
918 	kfree(metadata->sizes);
919 out_no_sizes:
920 	ttm_prime_object_kfree(user_srf, prime);
921 out_no_user_srf:
922 	ttm_mem_global_free(vmw_mem_glob(dev_priv), size);
923 out_unlock:
924 	ttm_read_unlock(&dev_priv->reservation_sem);
925 	return ret;
926 }
927 
928 
929 static int
930 vmw_surface_handle_reference(struct vmw_private *dev_priv,
931 			     struct drm_file *file_priv,
932 			     uint32_t u_handle,
933 			     enum drm_vmw_handle_type handle_type,
934 			     struct ttm_base_object **base_p)
935 {
936 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
937 	struct vmw_user_surface *user_srf;
938 	uint32_t handle;
939 	struct ttm_base_object *base;
940 	int ret;
941 
942 	if (handle_type == DRM_VMW_HANDLE_PRIME) {
943 		ret = ttm_prime_fd_to_handle(tfile, u_handle, &handle);
944 		if (unlikely(ret != 0))
945 			return ret;
946 	} else {
947 		handle = u_handle;
948 	}
949 
950 	ret = -EINVAL;
951 	base = ttm_base_object_lookup_for_ref(dev_priv->tdev, handle);
952 	if (unlikely(!base)) {
953 		VMW_DEBUG_USER("Could not find surface to reference.\n");
954 		goto out_no_lookup;
955 	}
956 
957 	if (unlikely(ttm_base_object_type(base) != VMW_RES_SURFACE)) {
958 		VMW_DEBUG_USER("Referenced object is not a surface.\n");
959 		goto out_bad_resource;
960 	}
961 
962 	if (handle_type != DRM_VMW_HANDLE_PRIME) {
963 		bool require_exist = false;
964 
965 		user_srf = container_of(base, struct vmw_user_surface,
966 					prime.base);
967 
968 		/* Error out if we are unauthenticated primary */
969 		if (drm_is_primary_client(file_priv) &&
970 		    !file_priv->authenticated) {
971 			ret = -EACCES;
972 			goto out_bad_resource;
973 		}
974 
975 		/*
976 		 * Make sure the surface creator has the same
977 		 * authenticating master, or is already registered with us.
978 		 */
979 		if (drm_is_primary_client(file_priv) &&
980 		    user_srf->master != file_priv->master)
981 			require_exist = true;
982 
983 		if (unlikely(drm_is_render_client(file_priv)))
984 			require_exist = true;
985 
986 		ret = ttm_ref_object_add(tfile, base, TTM_REF_USAGE, NULL,
987 					 require_exist);
988 		if (unlikely(ret != 0)) {
989 			DRM_ERROR("Could not add a reference to a surface.\n");
990 			goto out_bad_resource;
991 		}
992 	}
993 
994 	*base_p = base;
995 	return 0;
996 
997 out_bad_resource:
998 	ttm_base_object_unref(&base);
999 out_no_lookup:
1000 	if (handle_type == DRM_VMW_HANDLE_PRIME)
1001 		(void) ttm_ref_object_base_unref(tfile, handle, TTM_REF_USAGE);
1002 
1003 	return ret;
1004 }
1005 
1006 /**
1007  * vmw_user_surface_define_ioctl - Ioctl function implementing
1008  *                                  the user surface reference functionality.
1009  *
1010  * @dev:            Pointer to a struct drm_device.
1011  * @data:           Pointer to data copied from / to user-space.
1012  * @file_priv:      Pointer to a drm file private structure.
1013  */
1014 int vmw_surface_reference_ioctl(struct drm_device *dev, void *data,
1015 				struct drm_file *file_priv)
1016 {
1017 	struct vmw_private *dev_priv = vmw_priv(dev);
1018 	union drm_vmw_surface_reference_arg *arg =
1019 	    (union drm_vmw_surface_reference_arg *)data;
1020 	struct drm_vmw_surface_arg *req = &arg->req;
1021 	struct drm_vmw_surface_create_req *rep = &arg->rep;
1022 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
1023 	struct vmw_surface *srf;
1024 	struct vmw_user_surface *user_srf;
1025 	struct drm_vmw_size __user *user_sizes;
1026 	struct ttm_base_object *base;
1027 	int ret;
1028 
1029 	ret = vmw_surface_handle_reference(dev_priv, file_priv, req->sid,
1030 					   req->handle_type, &base);
1031 	if (unlikely(ret != 0))
1032 		return ret;
1033 
1034 	user_srf = container_of(base, struct vmw_user_surface, prime.base);
1035 	srf = &user_srf->srf;
1036 
1037 	/* Downcast of flags when sending back to user space */
1038 	rep->flags = (uint32_t)srf->metadata.flags;
1039 	rep->format = srf->metadata.format;
1040 	memcpy(rep->mip_levels, srf->metadata.mip_levels,
1041 	       sizeof(srf->metadata.mip_levels));
1042 	user_sizes = (struct drm_vmw_size __user *)(unsigned long)
1043 	    rep->size_addr;
1044 
1045 	if (user_sizes)
1046 		ret = copy_to_user(user_sizes, &srf->metadata.base_size,
1047 				   sizeof(srf->metadata.base_size));
1048 	if (unlikely(ret != 0)) {
1049 		VMW_DEBUG_USER("copy_to_user failed %p %u\n", user_sizes,
1050 			       srf->metadata.num_sizes);
1051 		ttm_ref_object_base_unref(tfile, base->handle, TTM_REF_USAGE);
1052 		ret = -EFAULT;
1053 	}
1054 
1055 	ttm_base_object_unref(&base);
1056 
1057 	return ret;
1058 }
1059 
1060 /**
1061  * vmw_surface_define_encode - Encode a surface_define command.
1062  *
1063  * @srf: Pointer to a struct vmw_surface object.
1064  * @cmd_space: Pointer to memory area in which the commands should be encoded.
1065  */
1066 static int vmw_gb_surface_create(struct vmw_resource *res)
1067 {
1068 	struct vmw_private *dev_priv = res->dev_priv;
1069 	struct vmw_surface *srf = vmw_res_to_srf(res);
1070 	struct vmw_surface_metadata *metadata = &srf->metadata;
1071 	uint32_t cmd_len, cmd_id, submit_len;
1072 	int ret;
1073 	struct {
1074 		SVGA3dCmdHeader header;
1075 		SVGA3dCmdDefineGBSurface body;
1076 	} *cmd;
1077 	struct {
1078 		SVGA3dCmdHeader header;
1079 		SVGA3dCmdDefineGBSurface_v2 body;
1080 	} *cmd2;
1081 	struct {
1082 		SVGA3dCmdHeader header;
1083 		SVGA3dCmdDefineGBSurface_v3 body;
1084 	} *cmd3;
1085 	struct {
1086 		SVGA3dCmdHeader header;
1087 		SVGA3dCmdDefineGBSurface_v4 body;
1088 	} *cmd4;
1089 
1090 	if (likely(res->id != -1))
1091 		return 0;
1092 
1093 	vmw_fifo_resource_inc(dev_priv);
1094 	ret = vmw_resource_alloc_id(res);
1095 	if (unlikely(ret != 0)) {
1096 		DRM_ERROR("Failed to allocate a surface id.\n");
1097 		goto out_no_id;
1098 	}
1099 
1100 	if (unlikely(res->id >= VMWGFX_NUM_GB_SURFACE)) {
1101 		ret = -EBUSY;
1102 		goto out_no_fifo;
1103 	}
1104 
1105 	if (has_sm5_context(dev_priv) && metadata->array_size > 0) {
1106 		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE_V4;
1107 		cmd_len = sizeof(cmd4->body);
1108 		submit_len = sizeof(*cmd4);
1109 	} else if (has_sm4_1_context(dev_priv) && metadata->array_size > 0) {
1110 		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE_V3;
1111 		cmd_len = sizeof(cmd3->body);
1112 		submit_len = sizeof(*cmd3);
1113 	} else if (metadata->array_size > 0) {
1114 		/* VMW_SM_4 support verified at creation time. */
1115 		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE_V2;
1116 		cmd_len = sizeof(cmd2->body);
1117 		submit_len = sizeof(*cmd2);
1118 	} else {
1119 		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE;
1120 		cmd_len = sizeof(cmd->body);
1121 		submit_len = sizeof(*cmd);
1122 	}
1123 
1124 	cmd = VMW_FIFO_RESERVE(dev_priv, submit_len);
1125 	cmd2 = (typeof(cmd2))cmd;
1126 	cmd3 = (typeof(cmd3))cmd;
1127 	cmd4 = (typeof(cmd4))cmd;
1128 	if (unlikely(!cmd)) {
1129 		ret = -ENOMEM;
1130 		goto out_no_fifo;
1131 	}
1132 
1133 	if (has_sm5_context(dev_priv) && metadata->array_size > 0) {
1134 		cmd4->header.id = cmd_id;
1135 		cmd4->header.size = cmd_len;
1136 		cmd4->body.sid = srf->res.id;
1137 		cmd4->body.surfaceFlags = metadata->flags;
1138 		cmd4->body.format = metadata->format;
1139 		cmd4->body.numMipLevels = metadata->mip_levels[0];
1140 		cmd4->body.multisampleCount = metadata->multisample_count;
1141 		cmd4->body.multisamplePattern = metadata->multisample_pattern;
1142 		cmd4->body.qualityLevel = metadata->quality_level;
1143 		cmd4->body.autogenFilter = metadata->autogen_filter;
1144 		cmd4->body.size.width = metadata->base_size.width;
1145 		cmd4->body.size.height = metadata->base_size.height;
1146 		cmd4->body.size.depth = metadata->base_size.depth;
1147 		cmd4->body.arraySize = metadata->array_size;
1148 		cmd4->body.bufferByteStride = metadata->buffer_byte_stride;
1149 	} else if (has_sm4_1_context(dev_priv) && metadata->array_size > 0) {
1150 		cmd3->header.id = cmd_id;
1151 		cmd3->header.size = cmd_len;
1152 		cmd3->body.sid = srf->res.id;
1153 		cmd3->body.surfaceFlags = metadata->flags;
1154 		cmd3->body.format = metadata->format;
1155 		cmd3->body.numMipLevels = metadata->mip_levels[0];
1156 		cmd3->body.multisampleCount = metadata->multisample_count;
1157 		cmd3->body.multisamplePattern = metadata->multisample_pattern;
1158 		cmd3->body.qualityLevel = metadata->quality_level;
1159 		cmd3->body.autogenFilter = metadata->autogen_filter;
1160 		cmd3->body.size.width = metadata->base_size.width;
1161 		cmd3->body.size.height = metadata->base_size.height;
1162 		cmd3->body.size.depth = metadata->base_size.depth;
1163 		cmd3->body.arraySize = metadata->array_size;
1164 	} else if (metadata->array_size > 0) {
1165 		cmd2->header.id = cmd_id;
1166 		cmd2->header.size = cmd_len;
1167 		cmd2->body.sid = srf->res.id;
1168 		cmd2->body.surfaceFlags = metadata->flags;
1169 		cmd2->body.format = metadata->format;
1170 		cmd2->body.numMipLevels = metadata->mip_levels[0];
1171 		cmd2->body.multisampleCount = metadata->multisample_count;
1172 		cmd2->body.autogenFilter = metadata->autogen_filter;
1173 		cmd2->body.size.width = metadata->base_size.width;
1174 		cmd2->body.size.height = metadata->base_size.height;
1175 		cmd2->body.size.depth = metadata->base_size.depth;
1176 		cmd2->body.arraySize = metadata->array_size;
1177 	} else {
1178 		cmd->header.id = cmd_id;
1179 		cmd->header.size = cmd_len;
1180 		cmd->body.sid = srf->res.id;
1181 		cmd->body.surfaceFlags = metadata->flags;
1182 		cmd->body.format = metadata->format;
1183 		cmd->body.numMipLevels = metadata->mip_levels[0];
1184 		cmd->body.multisampleCount = metadata->multisample_count;
1185 		cmd->body.autogenFilter = metadata->autogen_filter;
1186 		cmd->body.size.width = metadata->base_size.width;
1187 		cmd->body.size.height = metadata->base_size.height;
1188 		cmd->body.size.depth = metadata->base_size.depth;
1189 	}
1190 
1191 	vmw_fifo_commit(dev_priv, submit_len);
1192 
1193 	return 0;
1194 
1195 out_no_fifo:
1196 	vmw_resource_release_id(res);
1197 out_no_id:
1198 	vmw_fifo_resource_dec(dev_priv);
1199 	return ret;
1200 }
1201 
1202 
1203 static int vmw_gb_surface_bind(struct vmw_resource *res,
1204 			       struct ttm_validate_buffer *val_buf)
1205 {
1206 	struct vmw_private *dev_priv = res->dev_priv;
1207 	struct {
1208 		SVGA3dCmdHeader header;
1209 		SVGA3dCmdBindGBSurface body;
1210 	} *cmd1;
1211 	struct {
1212 		SVGA3dCmdHeader header;
1213 		SVGA3dCmdUpdateGBSurface body;
1214 	} *cmd2;
1215 	uint32_t submit_size;
1216 	struct ttm_buffer_object *bo = val_buf->bo;
1217 
1218 	BUG_ON(bo->mem.mem_type != VMW_PL_MOB);
1219 
1220 	submit_size = sizeof(*cmd1) + (res->backup_dirty ? sizeof(*cmd2) : 0);
1221 
1222 	cmd1 = VMW_FIFO_RESERVE(dev_priv, submit_size);
1223 	if (unlikely(!cmd1))
1224 		return -ENOMEM;
1225 
1226 	cmd1->header.id = SVGA_3D_CMD_BIND_GB_SURFACE;
1227 	cmd1->header.size = sizeof(cmd1->body);
1228 	cmd1->body.sid = res->id;
1229 	cmd1->body.mobid = bo->mem.start;
1230 	if (res->backup_dirty) {
1231 		cmd2 = (void *) &cmd1[1];
1232 		cmd2->header.id = SVGA_3D_CMD_UPDATE_GB_SURFACE;
1233 		cmd2->header.size = sizeof(cmd2->body);
1234 		cmd2->body.sid = res->id;
1235 	}
1236 	vmw_fifo_commit(dev_priv, submit_size);
1237 
1238 	if (res->backup->dirty && res->backup_dirty) {
1239 		/* We've just made a full upload. Cear dirty regions. */
1240 		vmw_bo_dirty_clear_res(res);
1241 	}
1242 
1243 	res->backup_dirty = false;
1244 
1245 	return 0;
1246 }
1247 
1248 static int vmw_gb_surface_unbind(struct vmw_resource *res,
1249 				 bool readback,
1250 				 struct ttm_validate_buffer *val_buf)
1251 {
1252 	struct vmw_private *dev_priv = res->dev_priv;
1253 	struct ttm_buffer_object *bo = val_buf->bo;
1254 	struct vmw_fence_obj *fence;
1255 
1256 	struct {
1257 		SVGA3dCmdHeader header;
1258 		SVGA3dCmdReadbackGBSurface body;
1259 	} *cmd1;
1260 	struct {
1261 		SVGA3dCmdHeader header;
1262 		SVGA3dCmdInvalidateGBSurface body;
1263 	} *cmd2;
1264 	struct {
1265 		SVGA3dCmdHeader header;
1266 		SVGA3dCmdBindGBSurface body;
1267 	} *cmd3;
1268 	uint32_t submit_size;
1269 	uint8_t *cmd;
1270 
1271 
1272 	BUG_ON(bo->mem.mem_type != VMW_PL_MOB);
1273 
1274 	submit_size = sizeof(*cmd3) + (readback ? sizeof(*cmd1) : sizeof(*cmd2));
1275 	cmd = VMW_FIFO_RESERVE(dev_priv, submit_size);
1276 	if (unlikely(!cmd))
1277 		return -ENOMEM;
1278 
1279 	if (readback) {
1280 		cmd1 = (void *) cmd;
1281 		cmd1->header.id = SVGA_3D_CMD_READBACK_GB_SURFACE;
1282 		cmd1->header.size = sizeof(cmd1->body);
1283 		cmd1->body.sid = res->id;
1284 		cmd3 = (void *) &cmd1[1];
1285 	} else {
1286 		cmd2 = (void *) cmd;
1287 		cmd2->header.id = SVGA_3D_CMD_INVALIDATE_GB_SURFACE;
1288 		cmd2->header.size = sizeof(cmd2->body);
1289 		cmd2->body.sid = res->id;
1290 		cmd3 = (void *) &cmd2[1];
1291 	}
1292 
1293 	cmd3->header.id = SVGA_3D_CMD_BIND_GB_SURFACE;
1294 	cmd3->header.size = sizeof(cmd3->body);
1295 	cmd3->body.sid = res->id;
1296 	cmd3->body.mobid = SVGA3D_INVALID_ID;
1297 
1298 	vmw_fifo_commit(dev_priv, submit_size);
1299 
1300 	/*
1301 	 * Create a fence object and fence the backup buffer.
1302 	 */
1303 
1304 	(void) vmw_execbuf_fence_commands(NULL, dev_priv,
1305 					  &fence, NULL);
1306 
1307 	vmw_bo_fence_single(val_buf->bo, fence);
1308 
1309 	if (likely(fence != NULL))
1310 		vmw_fence_obj_unreference(&fence);
1311 
1312 	return 0;
1313 }
1314 
1315 static int vmw_gb_surface_destroy(struct vmw_resource *res)
1316 {
1317 	struct vmw_private *dev_priv = res->dev_priv;
1318 	struct vmw_surface *srf = vmw_res_to_srf(res);
1319 	struct {
1320 		SVGA3dCmdHeader header;
1321 		SVGA3dCmdDestroyGBSurface body;
1322 	} *cmd;
1323 
1324 	if (likely(res->id == -1))
1325 		return 0;
1326 
1327 	mutex_lock(&dev_priv->binding_mutex);
1328 	vmw_view_surface_list_destroy(dev_priv, &srf->view_list);
1329 	vmw_binding_res_list_scrub(&res->binding_head);
1330 
1331 	cmd = VMW_FIFO_RESERVE(dev_priv, sizeof(*cmd));
1332 	if (unlikely(!cmd)) {
1333 		mutex_unlock(&dev_priv->binding_mutex);
1334 		return -ENOMEM;
1335 	}
1336 
1337 	cmd->header.id = SVGA_3D_CMD_DESTROY_GB_SURFACE;
1338 	cmd->header.size = sizeof(cmd->body);
1339 	cmd->body.sid = res->id;
1340 	vmw_fifo_commit(dev_priv, sizeof(*cmd));
1341 	mutex_unlock(&dev_priv->binding_mutex);
1342 	vmw_resource_release_id(res);
1343 	vmw_fifo_resource_dec(dev_priv);
1344 
1345 	return 0;
1346 }
1347 
1348 /**
1349  * vmw_gb_surface_define_ioctl - Ioctl function implementing
1350  * the user surface define functionality.
1351  *
1352  * @dev: Pointer to a struct drm_device.
1353  * @data: Pointer to data copied from / to user-space.
1354  * @file_priv: Pointer to a drm file private structure.
1355  */
1356 int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
1357 				struct drm_file *file_priv)
1358 {
1359 	union drm_vmw_gb_surface_create_arg *arg =
1360 	    (union drm_vmw_gb_surface_create_arg *)data;
1361 	struct drm_vmw_gb_surface_create_rep *rep = &arg->rep;
1362 	struct drm_vmw_gb_surface_create_ext_req req_ext;
1363 
1364 	req_ext.base = arg->req;
1365 	req_ext.version = drm_vmw_gb_surface_v1;
1366 	req_ext.svga3d_flags_upper_32_bits = 0;
1367 	req_ext.multisample_pattern = SVGA3D_MS_PATTERN_NONE;
1368 	req_ext.quality_level = SVGA3D_MS_QUALITY_NONE;
1369 	req_ext.buffer_byte_stride = 0;
1370 	req_ext.must_be_zero = 0;
1371 
1372 	return vmw_gb_surface_define_internal(dev, &req_ext, rep, file_priv);
1373 }
1374 
1375 /**
1376  * vmw_gb_surface_reference_ioctl - Ioctl function implementing
1377  * the user surface reference functionality.
1378  *
1379  * @dev: Pointer to a struct drm_device.
1380  * @data: Pointer to data copied from / to user-space.
1381  * @file_priv: Pointer to a drm file private structure.
1382  */
1383 int vmw_gb_surface_reference_ioctl(struct drm_device *dev, void *data,
1384 				   struct drm_file *file_priv)
1385 {
1386 	union drm_vmw_gb_surface_reference_arg *arg =
1387 	    (union drm_vmw_gb_surface_reference_arg *)data;
1388 	struct drm_vmw_surface_arg *req = &arg->req;
1389 	struct drm_vmw_gb_surface_ref_rep *rep = &arg->rep;
1390 	struct drm_vmw_gb_surface_ref_ext_rep rep_ext;
1391 	int ret;
1392 
1393 	ret = vmw_gb_surface_reference_internal(dev, req, &rep_ext, file_priv);
1394 
1395 	if (unlikely(ret != 0))
1396 		return ret;
1397 
1398 	rep->creq = rep_ext.creq.base;
1399 	rep->crep = rep_ext.crep;
1400 
1401 	return ret;
1402 }
1403 
1404 /**
1405  * vmw_gb_surface_define_ext_ioctl - Ioctl function implementing
1406  * the user surface define functionality.
1407  *
1408  * @dev: Pointer to a struct drm_device.
1409  * @data: Pointer to data copied from / to user-space.
1410  * @file_priv: Pointer to a drm file private structure.
1411  */
1412 int vmw_gb_surface_define_ext_ioctl(struct drm_device *dev, void *data,
1413 				struct drm_file *file_priv)
1414 {
1415 	union drm_vmw_gb_surface_create_ext_arg *arg =
1416 	    (union drm_vmw_gb_surface_create_ext_arg *)data;
1417 	struct drm_vmw_gb_surface_create_ext_req *req = &arg->req;
1418 	struct drm_vmw_gb_surface_create_rep *rep = &arg->rep;
1419 
1420 	return vmw_gb_surface_define_internal(dev, req, rep, file_priv);
1421 }
1422 
1423 /**
1424  * vmw_gb_surface_reference_ext_ioctl - Ioctl function implementing
1425  * the user surface reference functionality.
1426  *
1427  * @dev: Pointer to a struct drm_device.
1428  * @data: Pointer to data copied from / to user-space.
1429  * @file_priv: Pointer to a drm file private structure.
1430  */
1431 int vmw_gb_surface_reference_ext_ioctl(struct drm_device *dev, void *data,
1432 				   struct drm_file *file_priv)
1433 {
1434 	union drm_vmw_gb_surface_reference_ext_arg *arg =
1435 	    (union drm_vmw_gb_surface_reference_ext_arg *)data;
1436 	struct drm_vmw_surface_arg *req = &arg->req;
1437 	struct drm_vmw_gb_surface_ref_ext_rep *rep = &arg->rep;
1438 
1439 	return vmw_gb_surface_reference_internal(dev, req, rep, file_priv);
1440 }
1441 
1442 /**
1443  * vmw_gb_surface_define_internal - Ioctl function implementing
1444  * the user surface define functionality.
1445  *
1446  * @dev: Pointer to a struct drm_device.
1447  * @req: Request argument from user-space.
1448  * @rep: Response argument to user-space.
1449  * @file_priv: Pointer to a drm file private structure.
1450  */
1451 static int
1452 vmw_gb_surface_define_internal(struct drm_device *dev,
1453 			       struct drm_vmw_gb_surface_create_ext_req *req,
1454 			       struct drm_vmw_gb_surface_create_rep *rep,
1455 			       struct drm_file *file_priv)
1456 {
1457 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
1458 	struct vmw_private *dev_priv = vmw_priv(dev);
1459 	struct vmw_user_surface *user_srf;
1460 	struct vmw_surface_metadata metadata = {0};
1461 	struct vmw_surface *srf;
1462 	struct vmw_resource *res;
1463 	struct vmw_resource *tmp;
1464 	int ret = 0;
1465 	uint32_t size;
1466 	uint32_t backup_handle = 0;
1467 	SVGA3dSurfaceAllFlags svga3d_flags_64 =
1468 		SVGA3D_FLAGS_64(req->svga3d_flags_upper_32_bits,
1469 				req->base.svga3d_flags);
1470 
1471 	/* array_size must be null for non-GL3 host. */
1472 	if (req->base.array_size > 0 && !has_sm4_context(dev_priv)) {
1473 		VMW_DEBUG_USER("SM4 surface not supported.\n");
1474 		return -EINVAL;
1475 	}
1476 
1477 	if (!has_sm4_1_context(dev_priv)) {
1478 		if (req->svga3d_flags_upper_32_bits != 0)
1479 			ret = -EINVAL;
1480 
1481 		if (req->base.multisample_count != 0)
1482 			ret = -EINVAL;
1483 
1484 		if (req->multisample_pattern != SVGA3D_MS_PATTERN_NONE)
1485 			ret = -EINVAL;
1486 
1487 		if (req->quality_level != SVGA3D_MS_QUALITY_NONE)
1488 			ret = -EINVAL;
1489 
1490 		if (ret) {
1491 			VMW_DEBUG_USER("SM4.1 surface not supported.\n");
1492 			return ret;
1493 		}
1494 	}
1495 
1496 	if (req->buffer_byte_stride > 0 && !has_sm5_context(dev_priv)) {
1497 		VMW_DEBUG_USER("SM5 surface not supported.\n");
1498 		return -EINVAL;
1499 	}
1500 
1501 	if ((svga3d_flags_64 & SVGA3D_SURFACE_MULTISAMPLE) &&
1502 	    req->base.multisample_count == 0) {
1503 		VMW_DEBUG_USER("Invalid sample count.\n");
1504 		return -EINVAL;
1505 	}
1506 
1507 	if (req->base.mip_levels > DRM_VMW_MAX_MIP_LEVELS) {
1508 		VMW_DEBUG_USER("Invalid mip level.\n");
1509 		return -EINVAL;
1510 	}
1511 
1512 	if (unlikely(vmw_user_surface_size == 0))
1513 		vmw_user_surface_size = ttm_round_pot(sizeof(*user_srf)) +
1514 			VMW_IDA_ACC_SIZE + TTM_OBJ_EXTRA_SIZE;
1515 
1516 	size = vmw_user_surface_size;
1517 
1518 	metadata.flags = svga3d_flags_64;
1519 	metadata.format = req->base.format;
1520 	metadata.mip_levels[0] = req->base.mip_levels;
1521 	metadata.multisample_count = req->base.multisample_count;
1522 	metadata.multisample_pattern = req->multisample_pattern;
1523 	metadata.quality_level = req->quality_level;
1524 	metadata.array_size = req->base.array_size;
1525 	metadata.buffer_byte_stride = req->buffer_byte_stride;
1526 	metadata.num_sizes = 1;
1527 	metadata.base_size = req->base.base_size;
1528 	metadata.scanout = req->base.drm_surface_flags &
1529 		drm_vmw_surface_flag_scanout;
1530 
1531 	/* Define a surface based on the parameters. */
1532 	ret = vmw_gb_surface_define(dev_priv, size, &metadata, &srf);
1533 	if (ret != 0) {
1534 		VMW_DEBUG_USER("Failed to define surface.\n");
1535 		return ret;
1536 	}
1537 
1538 	user_srf = container_of(srf, struct vmw_user_surface, srf);
1539 	if (drm_is_primary_client(file_priv))
1540 		user_srf->master = drm_master_get(file_priv->master);
1541 
1542 	ret = ttm_read_lock(&dev_priv->reservation_sem, true);
1543 	if (unlikely(ret != 0))
1544 		return ret;
1545 
1546 	res = &user_srf->srf.res;
1547 
1548 	if (req->base.buffer_handle != SVGA3D_INVALID_ID) {
1549 		ret = vmw_user_bo_lookup(tfile, req->base.buffer_handle,
1550 					 &res->backup,
1551 					 &user_srf->backup_base);
1552 		if (ret == 0) {
1553 			if (res->backup->base.num_pages * PAGE_SIZE <
1554 			    res->backup_size) {
1555 				VMW_DEBUG_USER("Surface backup buffer too small.\n");
1556 				vmw_bo_unreference(&res->backup);
1557 				ret = -EINVAL;
1558 				goto out_unlock;
1559 			} else {
1560 				backup_handle = req->base.buffer_handle;
1561 			}
1562 		}
1563 	} else if (req->base.drm_surface_flags &
1564 		   (drm_vmw_surface_flag_create_buffer |
1565 		    drm_vmw_surface_flag_coherent))
1566 		ret = vmw_user_bo_alloc(dev_priv, tfile,
1567 					res->backup_size,
1568 					req->base.drm_surface_flags &
1569 					drm_vmw_surface_flag_shareable,
1570 					&backup_handle,
1571 					&res->backup,
1572 					&user_srf->backup_base);
1573 
1574 	if (unlikely(ret != 0)) {
1575 		vmw_resource_unreference(&res);
1576 		goto out_unlock;
1577 	}
1578 
1579 	if (req->base.drm_surface_flags & drm_vmw_surface_flag_coherent) {
1580 		struct vmw_buffer_object *backup = res->backup;
1581 
1582 		ttm_bo_reserve(&backup->base, false, false, NULL);
1583 		if (!res->func->dirty_alloc)
1584 			ret = -EINVAL;
1585 		if (!ret)
1586 			ret = vmw_bo_dirty_add(backup);
1587 		if (!ret) {
1588 			res->coherent = true;
1589 			ret = res->func->dirty_alloc(res);
1590 		}
1591 		ttm_bo_unreserve(&backup->base);
1592 		if (ret) {
1593 			vmw_resource_unreference(&res);
1594 			goto out_unlock;
1595 		}
1596 
1597 	}
1598 
1599 	tmp = vmw_resource_reference(res);
1600 	ret = ttm_prime_object_init(tfile, res->backup_size, &user_srf->prime,
1601 				    req->base.drm_surface_flags &
1602 				    drm_vmw_surface_flag_shareable,
1603 				    VMW_RES_SURFACE,
1604 				    &vmw_user_surface_base_release, NULL);
1605 
1606 	if (unlikely(ret != 0)) {
1607 		vmw_resource_unreference(&tmp);
1608 		vmw_resource_unreference(&res);
1609 		goto out_unlock;
1610 	}
1611 
1612 	rep->handle      = user_srf->prime.base.handle;
1613 	rep->backup_size = res->backup_size;
1614 	if (res->backup) {
1615 		rep->buffer_map_handle =
1616 			drm_vma_node_offset_addr(&res->backup->base.base.vma_node);
1617 		rep->buffer_size = res->backup->base.num_pages * PAGE_SIZE;
1618 		rep->buffer_handle = backup_handle;
1619 	} else {
1620 		rep->buffer_map_handle = 0;
1621 		rep->buffer_size = 0;
1622 		rep->buffer_handle = SVGA3D_INVALID_ID;
1623 	}
1624 
1625 	vmw_resource_unreference(&res);
1626 
1627 out_unlock:
1628 	ttm_read_unlock(&dev_priv->reservation_sem);
1629 	return ret;
1630 }
1631 
1632 /**
1633  * vmw_gb_surface_reference_internal - Ioctl function implementing
1634  * the user surface reference functionality.
1635  *
1636  * @dev: Pointer to a struct drm_device.
1637  * @req: Pointer to user-space request surface arg.
1638  * @rep: Pointer to response to user-space.
1639  * @file_priv: Pointer to a drm file private structure.
1640  */
1641 static int
1642 vmw_gb_surface_reference_internal(struct drm_device *dev,
1643 				  struct drm_vmw_surface_arg *req,
1644 				  struct drm_vmw_gb_surface_ref_ext_rep *rep,
1645 				  struct drm_file *file_priv)
1646 {
1647 	struct vmw_private *dev_priv = vmw_priv(dev);
1648 	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
1649 	struct vmw_surface *srf;
1650 	struct vmw_user_surface *user_srf;
1651 	struct vmw_surface_metadata *metadata;
1652 	struct ttm_base_object *base;
1653 	uint32_t backup_handle;
1654 	int ret;
1655 
1656 	ret = vmw_surface_handle_reference(dev_priv, file_priv, req->sid,
1657 					   req->handle_type, &base);
1658 	if (unlikely(ret != 0))
1659 		return ret;
1660 
1661 	user_srf = container_of(base, struct vmw_user_surface, prime.base);
1662 	srf = &user_srf->srf;
1663 	if (!srf->res.backup) {
1664 		DRM_ERROR("Shared GB surface is missing a backup buffer.\n");
1665 		goto out_bad_resource;
1666 	}
1667 	metadata = &srf->metadata;
1668 
1669 	mutex_lock(&dev_priv->cmdbuf_mutex); /* Protect res->backup */
1670 	ret = vmw_user_bo_reference(tfile, srf->res.backup, &backup_handle);
1671 	mutex_unlock(&dev_priv->cmdbuf_mutex);
1672 
1673 	if (unlikely(ret != 0)) {
1674 		DRM_ERROR("Could not add a reference to a GB surface "
1675 			  "backup buffer.\n");
1676 		(void) ttm_ref_object_base_unref(tfile, base->handle,
1677 						 TTM_REF_USAGE);
1678 		goto out_bad_resource;
1679 	}
1680 
1681 	rep->creq.base.svga3d_flags = SVGA3D_FLAGS_LOWER_32(metadata->flags);
1682 	rep->creq.base.format = metadata->format;
1683 	rep->creq.base.mip_levels = metadata->mip_levels[0];
1684 	rep->creq.base.drm_surface_flags = 0;
1685 	rep->creq.base.multisample_count = metadata->multisample_count;
1686 	rep->creq.base.autogen_filter = metadata->autogen_filter;
1687 	rep->creq.base.array_size = metadata->array_size;
1688 	rep->creq.base.buffer_handle = backup_handle;
1689 	rep->creq.base.base_size = metadata->base_size;
1690 	rep->crep.handle = user_srf->prime.base.handle;
1691 	rep->crep.backup_size = srf->res.backup_size;
1692 	rep->crep.buffer_handle = backup_handle;
1693 	rep->crep.buffer_map_handle =
1694 		drm_vma_node_offset_addr(&srf->res.backup->base.base.vma_node);
1695 	rep->crep.buffer_size = srf->res.backup->base.num_pages * PAGE_SIZE;
1696 
1697 	rep->creq.version = drm_vmw_gb_surface_v1;
1698 	rep->creq.svga3d_flags_upper_32_bits =
1699 		SVGA3D_FLAGS_UPPER_32(metadata->flags);
1700 	rep->creq.multisample_pattern = metadata->multisample_pattern;
1701 	rep->creq.quality_level = metadata->quality_level;
1702 	rep->creq.must_be_zero = 0;
1703 
1704 out_bad_resource:
1705 	ttm_base_object_unref(&base);
1706 
1707 	return ret;
1708 }
1709 
1710 /**
1711  * vmw_subres_dirty_add - Add a dirty region to a subresource
1712  * @dirty: The surfaces's dirty tracker.
1713  * @loc_start: The location corresponding to the start of the region.
1714  * @loc_end: The location corresponding to the end of the region.
1715  *
1716  * As we are assuming that @loc_start and @loc_end represent a sequential
1717  * range of backing store memory, if the region spans multiple lines then
1718  * regardless of the x coordinate, the full lines are dirtied.
1719  * Correspondingly if the region spans multiple z slices, then full rather
1720  * than partial z slices are dirtied.
1721  */
1722 static void vmw_subres_dirty_add(struct vmw_surface_dirty *dirty,
1723 				 const struct svga3dsurface_loc *loc_start,
1724 				 const struct svga3dsurface_loc *loc_end)
1725 {
1726 	const struct svga3dsurface_cache *cache = &dirty->cache;
1727 	SVGA3dBox *box = &dirty->boxes[loc_start->sub_resource];
1728 	u32 mip = loc_start->sub_resource % cache->num_mip_levels;
1729 	const struct drm_vmw_size *size = &cache->mip[mip].size;
1730 	u32 box_c2 = box->z + box->d;
1731 
1732 	if (WARN_ON(loc_start->sub_resource >= dirty->num_subres))
1733 		return;
1734 
1735 	if (box->d == 0 || box->z > loc_start->z)
1736 		box->z = loc_start->z;
1737 	if (box_c2 < loc_end->z)
1738 		box->d = loc_end->z - box->z;
1739 
1740 	if (loc_start->z + 1 == loc_end->z) {
1741 		box_c2 = box->y + box->h;
1742 		if (box->h == 0 || box->y > loc_start->y)
1743 			box->y = loc_start->y;
1744 		if (box_c2 < loc_end->y)
1745 			box->h = loc_end->y - box->y;
1746 
1747 		if (loc_start->y + 1 == loc_end->y) {
1748 			box_c2 = box->x + box->w;
1749 			if (box->w == 0 || box->x > loc_start->x)
1750 				box->x = loc_start->x;
1751 			if (box_c2 < loc_end->x)
1752 				box->w = loc_end->x - box->x;
1753 		} else {
1754 			box->x = 0;
1755 			box->w = size->width;
1756 		}
1757 	} else {
1758 		box->y = 0;
1759 		box->h = size->height;
1760 		box->x = 0;
1761 		box->w = size->width;
1762 	}
1763 }
1764 
1765 /**
1766  * vmw_subres_dirty_full - Mark a full subresource as dirty
1767  * @dirty: The surface's dirty tracker.
1768  * @subres: The subresource
1769  */
1770 static void vmw_subres_dirty_full(struct vmw_surface_dirty *dirty, u32 subres)
1771 {
1772 	const struct svga3dsurface_cache *cache = &dirty->cache;
1773 	u32 mip = subres % cache->num_mip_levels;
1774 	const struct drm_vmw_size *size = &cache->mip[mip].size;
1775 	SVGA3dBox *box = &dirty->boxes[subres];
1776 
1777 	box->x = 0;
1778 	box->y = 0;
1779 	box->z = 0;
1780 	box->w = size->width;
1781 	box->h = size->height;
1782 	box->d = size->depth;
1783 }
1784 
1785 /*
1786  * vmw_surface_tex_dirty_add_range - The dirty_add_range callback for texture
1787  * surfaces.
1788  */
1789 static void vmw_surface_tex_dirty_range_add(struct vmw_resource *res,
1790 					    size_t start, size_t end)
1791 {
1792 	struct vmw_surface_dirty *dirty =
1793 		(struct vmw_surface_dirty *) res->dirty;
1794 	size_t backup_end = res->backup_offset + res->backup_size;
1795 	struct svga3dsurface_loc loc1, loc2;
1796 	const struct svga3dsurface_cache *cache;
1797 
1798 	start = max_t(size_t, start, res->backup_offset) - res->backup_offset;
1799 	end = min(end, backup_end) - res->backup_offset;
1800 	cache = &dirty->cache;
1801 	svga3dsurface_get_loc(cache, &loc1, start);
1802 	svga3dsurface_get_loc(cache, &loc2, end - 1);
1803 	svga3dsurface_inc_loc(cache, &loc2);
1804 
1805 	if (loc1.sub_resource + 1 == loc2.sub_resource) {
1806 		/* Dirty range covers a single sub-resource */
1807 		vmw_subres_dirty_add(dirty, &loc1, &loc2);
1808 	} else {
1809 		/* Dirty range covers multiple sub-resources */
1810 		struct svga3dsurface_loc loc_min, loc_max;
1811 		u32 sub_res;
1812 
1813 		svga3dsurface_max_loc(cache, loc1.sub_resource, &loc_max);
1814 		vmw_subres_dirty_add(dirty, &loc1, &loc_max);
1815 		svga3dsurface_min_loc(cache, loc2.sub_resource - 1, &loc_min);
1816 		vmw_subres_dirty_add(dirty, &loc_min, &loc2);
1817 		for (sub_res = loc1.sub_resource + 1;
1818 		     sub_res < loc2.sub_resource - 1; ++sub_res)
1819 			vmw_subres_dirty_full(dirty, sub_res);
1820 	}
1821 }
1822 
1823 /*
1824  * vmw_surface_tex_dirty_add_range - The dirty_add_range callback for buffer
1825  * surfaces.
1826  */
1827 static void vmw_surface_buf_dirty_range_add(struct vmw_resource *res,
1828 					    size_t start, size_t end)
1829 {
1830 	struct vmw_surface_dirty *dirty =
1831 		(struct vmw_surface_dirty *) res->dirty;
1832 	const struct svga3dsurface_cache *cache = &dirty->cache;
1833 	size_t backup_end = res->backup_offset + cache->mip_chain_bytes;
1834 	SVGA3dBox *box = &dirty->boxes[0];
1835 	u32 box_c2;
1836 
1837 	box->h = box->d = 1;
1838 	start = max_t(size_t, start, res->backup_offset) - res->backup_offset;
1839 	end = min(end, backup_end) - res->backup_offset;
1840 	box_c2 = box->x + box->w;
1841 	if (box->w == 0 || box->x > start)
1842 		box->x = start;
1843 	if (box_c2 < end)
1844 		box->w = end - box->x;
1845 }
1846 
1847 /*
1848  * vmw_surface_tex_dirty_add_range - The dirty_add_range callback for surfaces
1849  */
1850 static void vmw_surface_dirty_range_add(struct vmw_resource *res, size_t start,
1851 					size_t end)
1852 {
1853 	struct vmw_surface *srf = vmw_res_to_srf(res);
1854 
1855 	if (WARN_ON(end <= res->backup_offset ||
1856 		    start >= res->backup_offset + res->backup_size))
1857 		return;
1858 
1859 	if (srf->metadata.format == SVGA3D_BUFFER)
1860 		vmw_surface_buf_dirty_range_add(res, start, end);
1861 	else
1862 		vmw_surface_tex_dirty_range_add(res, start, end);
1863 }
1864 
1865 /*
1866  * vmw_surface_dirty_sync - The surface's dirty_sync callback.
1867  */
1868 static int vmw_surface_dirty_sync(struct vmw_resource *res)
1869 {
1870 	struct vmw_private *dev_priv = res->dev_priv;
1871 	bool has_dx = 0;
1872 	u32 i, num_dirty;
1873 	struct vmw_surface_dirty *dirty =
1874 		(struct vmw_surface_dirty *) res->dirty;
1875 	size_t alloc_size;
1876 	const struct svga3dsurface_cache *cache = &dirty->cache;
1877 	struct {
1878 		SVGA3dCmdHeader header;
1879 		SVGA3dCmdDXUpdateSubResource body;
1880 	} *cmd1;
1881 	struct {
1882 		SVGA3dCmdHeader header;
1883 		SVGA3dCmdUpdateGBImage body;
1884 	} *cmd2;
1885 	void *cmd;
1886 
1887 	num_dirty = 0;
1888 	for (i = 0; i < dirty->num_subres; ++i) {
1889 		const SVGA3dBox *box = &dirty->boxes[i];
1890 
1891 		if (box->d)
1892 			num_dirty++;
1893 	}
1894 
1895 	if (!num_dirty)
1896 		goto out;
1897 
1898 	alloc_size = num_dirty * ((has_dx) ? sizeof(*cmd1) : sizeof(*cmd2));
1899 	cmd = VMW_FIFO_RESERVE(dev_priv, alloc_size);
1900 	if (!cmd)
1901 		return -ENOMEM;
1902 
1903 	cmd1 = cmd;
1904 	cmd2 = cmd;
1905 
1906 	for (i = 0; i < dirty->num_subres; ++i) {
1907 		const SVGA3dBox *box = &dirty->boxes[i];
1908 
1909 		if (!box->d)
1910 			continue;
1911 
1912 		/*
1913 		 * DX_UPDATE_SUBRESOURCE is aware of array surfaces.
1914 		 * UPDATE_GB_IMAGE is not.
1915 		 */
1916 		if (has_dx) {
1917 			cmd1->header.id = SVGA_3D_CMD_DX_UPDATE_SUBRESOURCE;
1918 			cmd1->header.size = sizeof(cmd1->body);
1919 			cmd1->body.sid = res->id;
1920 			cmd1->body.subResource = i;
1921 			cmd1->body.box = *box;
1922 			cmd1++;
1923 		} else {
1924 			cmd2->header.id = SVGA_3D_CMD_UPDATE_GB_IMAGE;
1925 			cmd2->header.size = sizeof(cmd2->body);
1926 			cmd2->body.image.sid = res->id;
1927 			cmd2->body.image.face = i / cache->num_mip_levels;
1928 			cmd2->body.image.mipmap = i -
1929 				(cache->num_mip_levels * cmd2->body.image.face);
1930 			cmd2->body.box = *box;
1931 			cmd2++;
1932 		}
1933 
1934 	}
1935 	vmw_fifo_commit(dev_priv, alloc_size);
1936  out:
1937 	memset(&dirty->boxes[0], 0, sizeof(dirty->boxes[0]) *
1938 	       dirty->num_subres);
1939 
1940 	return 0;
1941 }
1942 
1943 /*
1944  * vmw_surface_dirty_alloc - The surface's dirty_alloc callback.
1945  */
1946 static int vmw_surface_dirty_alloc(struct vmw_resource *res)
1947 {
1948 	struct vmw_surface *srf = vmw_res_to_srf(res);
1949 	const struct vmw_surface_metadata *metadata = &srf->metadata;
1950 	struct vmw_surface_dirty *dirty;
1951 	u32 num_layers = 1;
1952 	u32 num_mip;
1953 	u32 num_subres;
1954 	u32 num_samples;
1955 	size_t dirty_size, acc_size;
1956 	static struct ttm_operation_ctx ctx = {
1957 		.interruptible = false,
1958 		.no_wait_gpu = false
1959 	};
1960 	int ret;
1961 
1962 	if (metadata->array_size)
1963 		num_layers = metadata->array_size;
1964 	else if (metadata->flags & SVGA3D_SURFACE_CUBEMAP)
1965 		num_layers *= SVGA3D_MAX_SURFACE_FACES;
1966 
1967 	num_mip = metadata->mip_levels[0];
1968 	if (!num_mip)
1969 		num_mip = 1;
1970 
1971 	num_subres = num_layers * num_mip;
1972 	dirty_size = sizeof(*dirty) + num_subres * sizeof(dirty->boxes[0]);
1973 	acc_size = ttm_round_pot(dirty_size);
1974 	ret = ttm_mem_global_alloc(vmw_mem_glob(res->dev_priv),
1975 				   acc_size, &ctx);
1976 	if (ret) {
1977 		VMW_DEBUG_USER("Out of graphics memory for surface "
1978 			       "dirty tracker.\n");
1979 		return ret;
1980 	}
1981 
1982 	dirty = kvzalloc(dirty_size, GFP_KERNEL);
1983 	if (!dirty) {
1984 		ret = -ENOMEM;
1985 		goto out_no_dirty;
1986 	}
1987 
1988 	num_samples = max_t(u32, 1, metadata->multisample_count);
1989 	ret = svga3dsurface_setup_cache(&metadata->base_size, metadata->format,
1990 					num_mip, num_layers, num_samples,
1991 					&dirty->cache);
1992 	if (ret)
1993 		goto out_no_cache;
1994 
1995 	dirty->num_subres = num_subres;
1996 	dirty->size = acc_size;
1997 	res->dirty = (struct vmw_resource_dirty *) dirty;
1998 
1999 	return 0;
2000 
2001 out_no_cache:
2002 	kvfree(dirty);
2003 out_no_dirty:
2004 	ttm_mem_global_free(vmw_mem_glob(res->dev_priv), acc_size);
2005 	return ret;
2006 }
2007 
2008 /*
2009  * vmw_surface_dirty_free - The surface's dirty_free callback
2010  */
2011 static void vmw_surface_dirty_free(struct vmw_resource *res)
2012 {
2013 	struct vmw_surface_dirty *dirty =
2014 		(struct vmw_surface_dirty *) res->dirty;
2015 	size_t acc_size = dirty->size;
2016 
2017 	kvfree(dirty);
2018 	ttm_mem_global_free(vmw_mem_glob(res->dev_priv), acc_size);
2019 	res->dirty = NULL;
2020 }
2021 
2022 /*
2023  * vmw_surface_clean - The surface's clean callback
2024  */
2025 static int vmw_surface_clean(struct vmw_resource *res)
2026 {
2027 	struct vmw_private *dev_priv = res->dev_priv;
2028 	size_t alloc_size;
2029 	struct {
2030 		SVGA3dCmdHeader header;
2031 		SVGA3dCmdReadbackGBSurface body;
2032 	} *cmd;
2033 
2034 	alloc_size = sizeof(*cmd);
2035 	cmd = VMW_FIFO_RESERVE(dev_priv, alloc_size);
2036 	if (!cmd)
2037 		return -ENOMEM;
2038 
2039 	cmd->header.id = SVGA_3D_CMD_READBACK_GB_SURFACE;
2040 	cmd->header.size = sizeof(cmd->body);
2041 	cmd->body.sid = res->id;
2042 	vmw_fifo_commit(dev_priv, alloc_size);
2043 
2044 	return 0;
2045 }
2046 
2047 /*
2048  * vmw_gb_surface_define - Define a private GB surface
2049  *
2050  * @dev_priv: Pointer to a device private.
2051  * @user_accounting_size:  Used to track user-space memory usage, set
2052  *                         to 0 for kernel mode only memory
2053  * @metadata: Metadata representing the surface to create.
2054  * @user_srf_out: allocated user_srf. Set to NULL on failure.
2055  *
2056  * GB surfaces allocated by this function will not have a user mode handle, and
2057  * thus will only be visible to vmwgfx.  For optimization reasons the
2058  * surface may later be given a user mode handle by another function to make
2059  * it available to user mode drivers.
2060  */
2061 int vmw_gb_surface_define(struct vmw_private *dev_priv,
2062 			  uint32_t user_accounting_size,
2063 			  const struct vmw_surface_metadata *req,
2064 			  struct vmw_surface **srf_out)
2065 {
2066 	struct vmw_surface_metadata *metadata;
2067 	struct vmw_user_surface *user_srf;
2068 	struct vmw_surface *srf;
2069 	struct ttm_operation_ctx ctx = {
2070 		.interruptible = true,
2071 		.no_wait_gpu = false
2072 	};
2073 	u32 sample_count = 1;
2074 	u32 num_layers = 1;
2075 	int ret;
2076 
2077 	*srf_out = NULL;
2078 
2079 	if (req->scanout) {
2080 		if (!svga3dsurface_is_screen_target_format(req->format)) {
2081 			VMW_DEBUG_USER("Invalid Screen Target surface format.");
2082 			return -EINVAL;
2083 		}
2084 
2085 		if (req->base_size.width > dev_priv->texture_max_width ||
2086 		    req->base_size.height > dev_priv->texture_max_height) {
2087 			VMW_DEBUG_USER("%ux%u\n, exceed max surface size %ux%u",
2088 				       req->base_size.width,
2089 				       req->base_size.height,
2090 				       dev_priv->texture_max_width,
2091 				       dev_priv->texture_max_height);
2092 			return -EINVAL;
2093 		}
2094 	} else {
2095 		const struct svga3d_surface_desc *desc =
2096 			svga3dsurface_get_desc(req->format);
2097 
2098 		if (desc->block_desc == SVGA3DBLOCKDESC_NONE) {
2099 			VMW_DEBUG_USER("Invalid surface format.\n");
2100 			return -EINVAL;
2101 		}
2102 	}
2103 
2104 	if (req->autogen_filter != SVGA3D_TEX_FILTER_NONE)
2105 		return -EINVAL;
2106 
2107 	if (req->num_sizes != 1)
2108 		return -EINVAL;
2109 
2110 	if (req->sizes != NULL)
2111 		return -EINVAL;
2112 
2113 	ret = ttm_read_lock(&dev_priv->reservation_sem, true);
2114 	if (unlikely(ret != 0))
2115 		return ret;
2116 
2117 	ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
2118 				   user_accounting_size, &ctx);
2119 	if (ret != 0) {
2120 		if (ret != -ERESTARTSYS)
2121 			DRM_ERROR("Out of graphics memory for surface.\n");
2122 		goto out_unlock;
2123 	}
2124 
2125 	user_srf = kzalloc(sizeof(*user_srf), GFP_KERNEL);
2126 	if (unlikely(!user_srf)) {
2127 		ret = -ENOMEM;
2128 		goto out_no_user_srf;
2129 	}
2130 
2131 	*srf_out  = &user_srf->srf;
2132 	user_srf->size = user_accounting_size;
2133 	user_srf->prime.base.shareable = false;
2134 	user_srf->prime.base.tfile = NULL;
2135 
2136 	srf = &user_srf->srf;
2137 	srf->metadata = *req;
2138 	srf->offsets = NULL;
2139 
2140 	metadata = &srf->metadata;
2141 
2142 	if (metadata->array_size)
2143 		num_layers = req->array_size;
2144 	else if (metadata->flags & SVGA3D_SURFACE_CUBEMAP)
2145 		num_layers = SVGA3D_MAX_SURFACE_FACES;
2146 
2147 	if (metadata->flags & SVGA3D_SURFACE_MULTISAMPLE)
2148 		sample_count = metadata->multisample_count;
2149 
2150 	srf->res.backup_size =
2151 		svga3dsurface_get_serialized_size_extended(metadata->format,
2152 							   metadata->base_size,
2153 							   metadata->mip_levels[0],
2154 							   num_layers,
2155 							   sample_count);
2156 
2157 	if (metadata->flags & SVGA3D_SURFACE_BIND_STREAM_OUTPUT)
2158 		srf->res.backup_size += sizeof(SVGA3dDXSOState);
2159 
2160 	/*
2161 	 * Don't set SVGA3D_SURFACE_SCREENTARGET flag for a scanout surface with
2162 	 * size greater than STDU max width/height. This is really a workaround
2163 	 * to support creation of big framebuffer requested by some user-space
2164 	 * for whole topology. That big framebuffer won't really be used for
2165 	 * binding with screen target as during prepare_fb a separate surface is
2166 	 * created so it's safe to ignore SVGA3D_SURFACE_SCREENTARGET flag.
2167 	 */
2168 	if (dev_priv->active_display_unit == vmw_du_screen_target &&
2169 	    metadata->scanout &&
2170 	    metadata->base_size.width <= dev_priv->stdu_max_width &&
2171 	    metadata->base_size.height <= dev_priv->stdu_max_height)
2172 		metadata->flags |= SVGA3D_SURFACE_SCREENTARGET;
2173 
2174 	/*
2175 	 * From this point, the generic resource management functions
2176 	 * destroy the object on failure.
2177 	 */
2178 	ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free);
2179 
2180 	ttm_read_unlock(&dev_priv->reservation_sem);
2181 	return ret;
2182 
2183 out_no_user_srf:
2184 	ttm_mem_global_free(vmw_mem_glob(dev_priv), user_accounting_size);
2185 
2186 out_unlock:
2187 	ttm_read_unlock(&dev_priv->reservation_sem);
2188 	return ret;
2189 }
2190