1// SPDX-License-Identifier: GPL-2.0 OR MIT
2/**************************************************************************
3 *
4 * Copyright 2009-2023 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 "vmwgfx_bo.h"
29#include "vmwgfx_drv.h"
30#include "vmwgfx_resource_priv.h"
31#include "vmwgfx_so.h"
32#include "vmwgfx_binding.h"
33#include "vmw_surface_cache.h"
34#include "device_include/svga3d_surfacedefs.h"
35
36#include <drm/ttm/ttm_placement.h>
37
38#define SVGA3D_FLAGS_64(upper32, lower32) (((uint64_t)upper32 << 32) | lower32)
39#define SVGA3D_FLAGS_UPPER_32(svga3d_flags) (svga3d_flags >> 32)
40#define SVGA3D_FLAGS_LOWER_32(svga3d_flags) \
41	(svga3d_flags & ((uint64_t)U32_MAX))
42
43/**
44 * struct vmw_user_surface - User-space visible surface resource
45 *
46 * @prime:          The TTM prime object.
47 * @srf:            The surface metadata.
48 * @master:         Master of the creating client. Used for security check.
49 */
50struct vmw_user_surface {
51	struct ttm_prime_object prime;
52	struct vmw_surface srf;
53	struct drm_master *master;
54};
55
56/**
57 * struct vmw_surface_offset - Backing store mip level offset info
58 *
59 * @face:           Surface face.
60 * @mip:            Mip level.
61 * @bo_offset:      Offset into backing store of this mip level.
62 *
63 */
64struct vmw_surface_offset {
65	uint32_t face;
66	uint32_t mip;
67	uint32_t bo_offset;
68};
69
70/**
71 * struct vmw_surface_dirty - Surface dirty-tracker
72 * @cache: Cached layout information of the surface.
73 * @num_subres: Number of subresources.
74 * @boxes: Array of SVGA3dBoxes indicating dirty regions. One per subresource.
75 */
76struct vmw_surface_dirty {
77	struct vmw_surface_cache cache;
78	u32 num_subres;
79	SVGA3dBox boxes[] __counted_by(num_subres);
80};
81
82static void vmw_user_surface_free(struct vmw_resource *res);
83static struct vmw_resource *
84vmw_user_surface_base_to_res(struct ttm_base_object *base);
85static int vmw_legacy_srf_bind(struct vmw_resource *res,
86			       struct ttm_validate_buffer *val_buf);
87static int vmw_legacy_srf_unbind(struct vmw_resource *res,
88				 bool readback,
89				 struct ttm_validate_buffer *val_buf);
90static int vmw_legacy_srf_create(struct vmw_resource *res);
91static int vmw_legacy_srf_destroy(struct vmw_resource *res);
92static int vmw_gb_surface_create(struct vmw_resource *res);
93static int vmw_gb_surface_bind(struct vmw_resource *res,
94			       struct ttm_validate_buffer *val_buf);
95static int vmw_gb_surface_unbind(struct vmw_resource *res,
96				 bool readback,
97				 struct ttm_validate_buffer *val_buf);
98static int vmw_gb_surface_destroy(struct vmw_resource *res);
99static int
100vmw_gb_surface_define_internal(struct drm_device *dev,
101			       struct drm_vmw_gb_surface_create_ext_req *req,
102			       struct drm_vmw_gb_surface_create_rep *rep,
103			       struct drm_file *file_priv);
104static int
105vmw_gb_surface_reference_internal(struct drm_device *dev,
106				  struct drm_vmw_surface_arg *req,
107				  struct drm_vmw_gb_surface_ref_ext_rep *rep,
108				  struct drm_file *file_priv);
109
110static void vmw_surface_dirty_free(struct vmw_resource *res);
111static int vmw_surface_dirty_alloc(struct vmw_resource *res);
112static int vmw_surface_dirty_sync(struct vmw_resource *res);
113static void vmw_surface_dirty_range_add(struct vmw_resource *res, size_t start,
114					size_t end);
115static int vmw_surface_clean(struct vmw_resource *res);
116
117static const struct vmw_user_resource_conv user_surface_conv = {
118	.object_type = VMW_RES_SURFACE,
119	.base_obj_to_res = vmw_user_surface_base_to_res,
120	.res_free = vmw_user_surface_free
121};
122
123const struct vmw_user_resource_conv *user_surface_converter =
124	&user_surface_conv;
125
126static const struct vmw_res_func vmw_legacy_surface_func = {
127	.res_type = vmw_res_surface,
128	.needs_guest_memory = false,
129	.may_evict = true,
130	.prio = 1,
131	.dirty_prio = 1,
132	.type_name = "legacy surfaces",
133	.domain = VMW_BO_DOMAIN_GMR,
134	.busy_domain = VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM,
135	.create = &vmw_legacy_srf_create,
136	.destroy = &vmw_legacy_srf_destroy,
137	.bind = &vmw_legacy_srf_bind,
138	.unbind = &vmw_legacy_srf_unbind
139};
140
141static const struct vmw_res_func vmw_gb_surface_func = {
142	.res_type = vmw_res_surface,
143	.needs_guest_memory = true,
144	.may_evict = true,
145	.prio = 1,
146	.dirty_prio = 2,
147	.type_name = "guest backed surfaces",
148	.domain = VMW_BO_DOMAIN_MOB,
149	.busy_domain = VMW_BO_DOMAIN_MOB,
150	.create = vmw_gb_surface_create,
151	.destroy = vmw_gb_surface_destroy,
152	.bind = vmw_gb_surface_bind,
153	.unbind = vmw_gb_surface_unbind,
154	.dirty_alloc = vmw_surface_dirty_alloc,
155	.dirty_free = vmw_surface_dirty_free,
156	.dirty_sync = vmw_surface_dirty_sync,
157	.dirty_range_add = vmw_surface_dirty_range_add,
158	.clean = vmw_surface_clean,
159};
160
161/*
162 * struct vmw_surface_dma - SVGA3D DMA command
163 */
164struct vmw_surface_dma {
165	SVGA3dCmdHeader header;
166	SVGA3dCmdSurfaceDMA body;
167	SVGA3dCopyBox cb;
168	SVGA3dCmdSurfaceDMASuffix suffix;
169};
170
171/*
172 * struct vmw_surface_define - SVGA3D Surface Define command
173 */
174struct vmw_surface_define {
175	SVGA3dCmdHeader header;
176	SVGA3dCmdDefineSurface body;
177};
178
179/*
180 * struct vmw_surface_destroy - SVGA3D Surface Destroy command
181 */
182struct vmw_surface_destroy {
183	SVGA3dCmdHeader header;
184	SVGA3dCmdDestroySurface body;
185};
186
187
188/**
189 * vmw_surface_dma_size - Compute fifo size for a dma command.
190 *
191 * @srf: Pointer to a struct vmw_surface
192 *
193 * Computes the required size for a surface dma command for backup or
194 * restoration of the surface represented by @srf.
195 */
196static inline uint32_t vmw_surface_dma_size(const struct vmw_surface *srf)
197{
198	return srf->metadata.num_sizes * sizeof(struct vmw_surface_dma);
199}
200
201
202/**
203 * vmw_surface_define_size - Compute fifo size for a surface define command.
204 *
205 * @srf: Pointer to a struct vmw_surface
206 *
207 * Computes the required size for a surface define command for the definition
208 * of the surface represented by @srf.
209 */
210static inline uint32_t vmw_surface_define_size(const struct vmw_surface *srf)
211{
212	return sizeof(struct vmw_surface_define) + srf->metadata.num_sizes *
213		sizeof(SVGA3dSize);
214}
215
216
217/**
218 * vmw_surface_destroy_size - Compute fifo size for a surface destroy command.
219 *
220 * Computes the required size for a surface destroy command for the destruction
221 * of a hw surface.
222 */
223static inline uint32_t vmw_surface_destroy_size(void)
224{
225	return sizeof(struct vmw_surface_destroy);
226}
227
228/**
229 * vmw_surface_destroy_encode - Encode a surface_destroy command.
230 *
231 * @id: The surface id
232 * @cmd_space: Pointer to memory area in which the commands should be encoded.
233 */
234static void vmw_surface_destroy_encode(uint32_t id,
235				       void *cmd_space)
236{
237	struct vmw_surface_destroy *cmd = (struct vmw_surface_destroy *)
238		cmd_space;
239
240	cmd->header.id = SVGA_3D_CMD_SURFACE_DESTROY;
241	cmd->header.size = sizeof(cmd->body);
242	cmd->body.sid = id;
243}
244
245/**
246 * vmw_surface_define_encode - Encode a surface_define command.
247 *
248 * @srf: Pointer to a struct vmw_surface object.
249 * @cmd_space: Pointer to memory area in which the commands should be encoded.
250 */
251static void vmw_surface_define_encode(const struct vmw_surface *srf,
252				      void *cmd_space)
253{
254	struct vmw_surface_define *cmd = (struct vmw_surface_define *)
255		cmd_space;
256	struct drm_vmw_size *src_size;
257	SVGA3dSize *cmd_size;
258	uint32_t cmd_len;
259	int i;
260
261	cmd_len = sizeof(cmd->body) + srf->metadata.num_sizes *
262		sizeof(SVGA3dSize);
263
264	cmd->header.id = SVGA_3D_CMD_SURFACE_DEFINE;
265	cmd->header.size = cmd_len;
266	cmd->body.sid = srf->res.id;
267	/*
268	 * Downcast of surfaceFlags, was upcasted when received from user-space,
269	 * since driver internally stores as 64 bit.
270	 * For legacy surface define only 32 bit flag is supported.
271	 */
272	cmd->body.surfaceFlags = (SVGA3dSurface1Flags)srf->metadata.flags;
273	cmd->body.format = srf->metadata.format;
274	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i)
275		cmd->body.face[i].numMipLevels = srf->metadata.mip_levels[i];
276
277	cmd += 1;
278	cmd_size = (SVGA3dSize *) cmd;
279	src_size = srf->metadata.sizes;
280
281	for (i = 0; i < srf->metadata.num_sizes; ++i, cmd_size++, src_size++) {
282		cmd_size->width = src_size->width;
283		cmd_size->height = src_size->height;
284		cmd_size->depth = src_size->depth;
285	}
286}
287
288/**
289 * vmw_surface_dma_encode - Encode a surface_dma command.
290 *
291 * @srf: Pointer to a struct vmw_surface object.
292 * @cmd_space: Pointer to memory area in which the commands should be encoded.
293 * @ptr: Pointer to an SVGAGuestPtr indicating where the surface contents
294 * should be placed or read from.
295 * @to_surface: Boolean whether to DMA to the surface or from the surface.
296 */
297static void vmw_surface_dma_encode(struct vmw_surface *srf,
298				   void *cmd_space,
299				   const SVGAGuestPtr *ptr,
300				   bool to_surface)
301{
302	uint32_t i;
303	struct vmw_surface_dma *cmd = (struct vmw_surface_dma *)cmd_space;
304	const struct SVGA3dSurfaceDesc *desc =
305		vmw_surface_get_desc(srf->metadata.format);
306
307	for (i = 0; i < srf->metadata.num_sizes; ++i) {
308		SVGA3dCmdHeader *header = &cmd->header;
309		SVGA3dCmdSurfaceDMA *body = &cmd->body;
310		SVGA3dCopyBox *cb = &cmd->cb;
311		SVGA3dCmdSurfaceDMASuffix *suffix = &cmd->suffix;
312		const struct vmw_surface_offset *cur_offset = &srf->offsets[i];
313		const struct drm_vmw_size *cur_size = &srf->metadata.sizes[i];
314
315		header->id = SVGA_3D_CMD_SURFACE_DMA;
316		header->size = sizeof(*body) + sizeof(*cb) + sizeof(*suffix);
317
318		body->guest.ptr = *ptr;
319		body->guest.ptr.offset += cur_offset->bo_offset;
320		body->guest.pitch = vmw_surface_calculate_pitch(desc, cur_size);
321		body->host.sid = srf->res.id;
322		body->host.face = cur_offset->face;
323		body->host.mipmap = cur_offset->mip;
324		body->transfer = ((to_surface) ?  SVGA3D_WRITE_HOST_VRAM :
325				  SVGA3D_READ_HOST_VRAM);
326		cb->x = 0;
327		cb->y = 0;
328		cb->z = 0;
329		cb->srcx = 0;
330		cb->srcy = 0;
331		cb->srcz = 0;
332		cb->w = cur_size->width;
333		cb->h = cur_size->height;
334		cb->d = cur_size->depth;
335
336		suffix->suffixSize = sizeof(*suffix);
337		suffix->maximumOffset =
338			vmw_surface_get_image_buffer_size(desc, cur_size,
339							    body->guest.pitch);
340		suffix->flags.discard = 0;
341		suffix->flags.unsynchronized = 0;
342		suffix->flags.reserved = 0;
343		++cmd;
344	}
345};
346
347
348/**
349 * vmw_hw_surface_destroy - destroy a Device surface
350 *
351 * @res:        Pointer to a struct vmw_resource embedded in a struct
352 *              vmw_surface.
353 *
354 * Destroys a the device surface associated with a struct vmw_surface if
355 * any, and adjusts resource count accordingly.
356 */
357static void vmw_hw_surface_destroy(struct vmw_resource *res)
358{
359
360	struct vmw_private *dev_priv = res->dev_priv;
361	void *cmd;
362
363	if (res->func->destroy == vmw_gb_surface_destroy) {
364		(void) vmw_gb_surface_destroy(res);
365		return;
366	}
367
368	if (res->id != -1) {
369
370		cmd = VMW_CMD_RESERVE(dev_priv, vmw_surface_destroy_size());
371		if (unlikely(!cmd))
372			return;
373
374		vmw_surface_destroy_encode(res->id, cmd);
375		vmw_cmd_commit(dev_priv, vmw_surface_destroy_size());
376
377		/*
378		 * used_memory_size_atomic, or separate lock
379		 * to avoid taking dev_priv::cmdbuf_mutex in
380		 * the destroy path.
381		 */
382
383		mutex_lock(&dev_priv->cmdbuf_mutex);
384		dev_priv->used_memory_size -= res->guest_memory_size;
385		mutex_unlock(&dev_priv->cmdbuf_mutex);
386	}
387}
388
389/**
390 * vmw_legacy_srf_create - Create a device surface as part of the
391 * resource validation process.
392 *
393 * @res: Pointer to a struct vmw_surface.
394 *
395 * If the surface doesn't have a hw id.
396 *
397 * Returns -EBUSY if there wasn't sufficient device resources to
398 * complete the validation. Retry after freeing up resources.
399 *
400 * May return other errors if the kernel is out of guest resources.
401 */
402static int vmw_legacy_srf_create(struct vmw_resource *res)
403{
404	struct vmw_private *dev_priv = res->dev_priv;
405	struct vmw_surface *srf;
406	uint32_t submit_size;
407	uint8_t *cmd;
408	int ret;
409
410	if (likely(res->id != -1))
411		return 0;
412
413	srf = vmw_res_to_srf(res);
414	if (unlikely(dev_priv->used_memory_size + res->guest_memory_size >=
415		     dev_priv->memory_size))
416		return -EBUSY;
417
418	/*
419	 * Alloc id for the resource.
420	 */
421
422	ret = vmw_resource_alloc_id(res);
423	if (unlikely(ret != 0)) {
424		DRM_ERROR("Failed to allocate a surface id.\n");
425		goto out_no_id;
426	}
427
428	if (unlikely(res->id >= SVGA3D_HB_MAX_SURFACE_IDS)) {
429		ret = -EBUSY;
430		goto out_no_fifo;
431	}
432
433	/*
434	 * Encode surface define- commands.
435	 */
436
437	submit_size = vmw_surface_define_size(srf);
438	cmd = VMW_CMD_RESERVE(dev_priv, submit_size);
439	if (unlikely(!cmd)) {
440		ret = -ENOMEM;
441		goto out_no_fifo;
442	}
443
444	vmw_surface_define_encode(srf, cmd);
445	vmw_cmd_commit(dev_priv, submit_size);
446	vmw_fifo_resource_inc(dev_priv);
447
448	/*
449	 * Surface memory usage accounting.
450	 */
451
452	dev_priv->used_memory_size += res->guest_memory_size;
453	return 0;
454
455out_no_fifo:
456	vmw_resource_release_id(res);
457out_no_id:
458	return ret;
459}
460
461/**
462 * vmw_legacy_srf_dma - Copy backup data to or from a legacy surface.
463 *
464 * @res:            Pointer to a struct vmw_res embedded in a struct
465 *                  vmw_surface.
466 * @val_buf:        Pointer to a struct ttm_validate_buffer containing
467 *                  information about the backup buffer.
468 * @bind:           Boolean wether to DMA to the surface.
469 *
470 * Transfer backup data to or from a legacy surface as part of the
471 * validation process.
472 * May return other errors if the kernel is out of guest resources.
473 * The backup buffer will be fenced or idle upon successful completion,
474 * and if the surface needs persistent backup storage, the backup buffer
475 * will also be returned reserved iff @bind is true.
476 */
477static int vmw_legacy_srf_dma(struct vmw_resource *res,
478			      struct ttm_validate_buffer *val_buf,
479			      bool bind)
480{
481	SVGAGuestPtr ptr;
482	struct vmw_fence_obj *fence;
483	uint32_t submit_size;
484	struct vmw_surface *srf = vmw_res_to_srf(res);
485	uint8_t *cmd;
486	struct vmw_private *dev_priv = res->dev_priv;
487
488	BUG_ON(!val_buf->bo);
489	submit_size = vmw_surface_dma_size(srf);
490	cmd = VMW_CMD_RESERVE(dev_priv, submit_size);
491	if (unlikely(!cmd))
492		return -ENOMEM;
493
494	vmw_bo_get_guest_ptr(val_buf->bo, &ptr);
495	vmw_surface_dma_encode(srf, cmd, &ptr, bind);
496
497	vmw_cmd_commit(dev_priv, submit_size);
498
499	/*
500	 * Create a fence object and fence the backup buffer.
501	 */
502
503	(void) vmw_execbuf_fence_commands(NULL, dev_priv,
504					  &fence, NULL);
505
506	vmw_bo_fence_single(val_buf->bo, fence);
507
508	if (likely(fence != NULL))
509		vmw_fence_obj_unreference(&fence);
510
511	return 0;
512}
513
514/**
515 * vmw_legacy_srf_bind - Perform a legacy surface bind as part of the
516 *                       surface validation process.
517 *
518 * @res:            Pointer to a struct vmw_res embedded in a struct
519 *                  vmw_surface.
520 * @val_buf:        Pointer to a struct ttm_validate_buffer containing
521 *                  information about the backup buffer.
522 *
523 * This function will copy backup data to the surface if the
524 * backup buffer is dirty.
525 */
526static int vmw_legacy_srf_bind(struct vmw_resource *res,
527			       struct ttm_validate_buffer *val_buf)
528{
529	if (!res->guest_memory_dirty)
530		return 0;
531
532	return vmw_legacy_srf_dma(res, val_buf, true);
533}
534
535
536/**
537 * vmw_legacy_srf_unbind - Perform a legacy surface unbind as part of the
538 *                         surface eviction process.
539 *
540 * @res:            Pointer to a struct vmw_res embedded in a struct
541 *                  vmw_surface.
542 * @readback:       Readback - only true if dirty
543 * @val_buf:        Pointer to a struct ttm_validate_buffer containing
544 *                  information about the backup buffer.
545 *
546 * This function will copy backup data from the surface.
547 */
548static int vmw_legacy_srf_unbind(struct vmw_resource *res,
549				 bool readback,
550				 struct ttm_validate_buffer *val_buf)
551{
552	if (unlikely(readback))
553		return vmw_legacy_srf_dma(res, val_buf, false);
554	return 0;
555}
556
557/**
558 * vmw_legacy_srf_destroy - Destroy a device surface as part of a
559 *                          resource eviction process.
560 *
561 * @res:            Pointer to a struct vmw_res embedded in a struct
562 *                  vmw_surface.
563 */
564static int vmw_legacy_srf_destroy(struct vmw_resource *res)
565{
566	struct vmw_private *dev_priv = res->dev_priv;
567	uint32_t submit_size;
568	uint8_t *cmd;
569
570	BUG_ON(res->id == -1);
571
572	/*
573	 * Encode the dma- and surface destroy commands.
574	 */
575
576	submit_size = vmw_surface_destroy_size();
577	cmd = VMW_CMD_RESERVE(dev_priv, submit_size);
578	if (unlikely(!cmd))
579		return -ENOMEM;
580
581	vmw_surface_destroy_encode(res->id, cmd);
582	vmw_cmd_commit(dev_priv, submit_size);
583
584	/*
585	 * Surface memory usage accounting.
586	 */
587
588	dev_priv->used_memory_size -= res->guest_memory_size;
589
590	/*
591	 * Release the surface ID.
592	 */
593
594	vmw_resource_release_id(res);
595	vmw_fifo_resource_dec(dev_priv);
596
597	return 0;
598}
599
600
601/**
602 * vmw_surface_init - initialize a struct vmw_surface
603 *
604 * @dev_priv:       Pointer to a device private struct.
605 * @srf:            Pointer to the struct vmw_surface to initialize.
606 * @res_free:       Pointer to a resource destructor used to free
607 *                  the object.
608 */
609static int vmw_surface_init(struct vmw_private *dev_priv,
610			    struct vmw_surface *srf,
611			    void (*res_free) (struct vmw_resource *res))
612{
613	int ret;
614	struct vmw_resource *res = &srf->res;
615
616	BUG_ON(!res_free);
617	ret = vmw_resource_init(dev_priv, res, true, res_free,
618				(dev_priv->has_mob) ? &vmw_gb_surface_func :
619				&vmw_legacy_surface_func);
620
621	if (unlikely(ret != 0)) {
622		res_free(res);
623		return ret;
624	}
625
626	/*
627	 * The surface won't be visible to hardware until a
628	 * surface validate.
629	 */
630
631	INIT_LIST_HEAD(&srf->view_list);
632	res->hw_destroy = vmw_hw_surface_destroy;
633	return ret;
634}
635
636/**
637 * vmw_user_surface_base_to_res - TTM base object to resource converter for
638 *                                user visible surfaces
639 *
640 * @base:           Pointer to a TTM base object
641 *
642 * Returns the struct vmw_resource embedded in a struct vmw_surface
643 * for the user-visible object identified by the TTM base object @base.
644 */
645static struct vmw_resource *
646vmw_user_surface_base_to_res(struct ttm_base_object *base)
647{
648	return &(container_of(base, struct vmw_user_surface,
649			      prime.base)->srf.res);
650}
651
652/**
653 * vmw_user_surface_free - User visible surface resource destructor
654 *
655 * @res:            A struct vmw_resource embedded in a struct vmw_surface.
656 */
657static void vmw_user_surface_free(struct vmw_resource *res)
658{
659	struct vmw_surface *srf = vmw_res_to_srf(res);
660	struct vmw_user_surface *user_srf =
661	    container_of(srf, struct vmw_user_surface, srf);
662
663	WARN_ON_ONCE(res->dirty);
664	if (user_srf->master)
665		drm_master_put(&user_srf->master);
666	kfree(srf->offsets);
667	kfree(srf->metadata.sizes);
668	kfree(srf->snooper.image);
669	ttm_prime_object_kfree(user_srf, prime);
670}
671
672/**
673 * vmw_user_surface_base_release - User visible surface TTM base object destructor
674 *
675 * @p_base:         Pointer to a pointer to a TTM base object
676 *                  embedded in a struct vmw_user_surface.
677 *
678 * Drops the base object's reference on its resource, and the
679 * pointer pointed to by *p_base is set to NULL.
680 */
681static void vmw_user_surface_base_release(struct ttm_base_object **p_base)
682{
683	struct ttm_base_object *base = *p_base;
684	struct vmw_user_surface *user_srf =
685	    container_of(base, struct vmw_user_surface, prime.base);
686	struct vmw_resource *res = &user_srf->srf.res;
687
688	*p_base = NULL;
689	vmw_resource_unreference(&res);
690}
691
692/**
693 * vmw_surface_destroy_ioctl - Ioctl function implementing
694 *                                  the user surface destroy functionality.
695 *
696 * @dev:            Pointer to a struct drm_device.
697 * @data:           Pointer to data copied from / to user-space.
698 * @file_priv:      Pointer to a drm file private structure.
699 */
700int vmw_surface_destroy_ioctl(struct drm_device *dev, void *data,
701			      struct drm_file *file_priv)
702{
703	struct drm_vmw_surface_arg *arg = (struct drm_vmw_surface_arg *)data;
704	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
705
706	return ttm_ref_object_base_unref(tfile, arg->sid);
707}
708
709/**
710 * vmw_surface_define_ioctl - Ioctl function implementing
711 *                                  the user surface define functionality.
712 *
713 * @dev:            Pointer to a struct drm_device.
714 * @data:           Pointer to data copied from / to user-space.
715 * @file_priv:      Pointer to a drm file private structure.
716 */
717int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
718			     struct drm_file *file_priv)
719{
720	struct vmw_private *dev_priv = vmw_priv(dev);
721	struct vmw_user_surface *user_srf;
722	struct vmw_surface *srf;
723	struct vmw_surface_metadata *metadata;
724	struct vmw_resource *res;
725	struct vmw_resource *tmp;
726	union drm_vmw_surface_create_arg *arg =
727	    (union drm_vmw_surface_create_arg *)data;
728	struct drm_vmw_surface_create_req *req = &arg->req;
729	struct drm_vmw_surface_arg *rep = &arg->rep;
730	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
731	int ret;
732	int i, j;
733	uint32_t cur_bo_offset;
734	struct drm_vmw_size *cur_size;
735	struct vmw_surface_offset *cur_offset;
736	uint32_t num_sizes;
737	const SVGA3dSurfaceDesc *desc;
738
739	num_sizes = 0;
740	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) {
741		if (req->mip_levels[i] > DRM_VMW_MAX_MIP_LEVELS)
742			return -EINVAL;
743		num_sizes += req->mip_levels[i];
744	}
745
746	if (num_sizes > DRM_VMW_MAX_SURFACE_FACES * DRM_VMW_MAX_MIP_LEVELS ||
747	    num_sizes == 0)
748		return -EINVAL;
749
750	desc = vmw_surface_get_desc(req->format);
751	if (unlikely(desc->blockDesc == SVGA3DBLOCKDESC_NONE)) {
752		VMW_DEBUG_USER("Invalid format %d for surface creation.\n",
753			       req->format);
754		return -EINVAL;
755	}
756
757	user_srf = kzalloc(sizeof(*user_srf), GFP_KERNEL);
758	if (unlikely(!user_srf)) {
759		ret = -ENOMEM;
760		goto out_unlock;
761	}
762
763	srf = &user_srf->srf;
764	metadata = &srf->metadata;
765	res = &srf->res;
766
767	/* Driver internally stores as 64-bit flags */
768	metadata->flags = (SVGA3dSurfaceAllFlags)req->flags;
769	metadata->format = req->format;
770	metadata->scanout = req->scanout;
771
772	memcpy(metadata->mip_levels, req->mip_levels,
773	       sizeof(metadata->mip_levels));
774	metadata->num_sizes = num_sizes;
775	metadata->sizes =
776		memdup_array_user((struct drm_vmw_size __user *)(unsigned long)
777			    req->size_addr,
778			    metadata->num_sizes, sizeof(*metadata->sizes));
779	if (IS_ERR(metadata->sizes)) {
780		ret = PTR_ERR(metadata->sizes);
781		goto out_no_sizes;
782	}
783	srf->offsets = kmalloc_array(metadata->num_sizes, sizeof(*srf->offsets),
784				     GFP_KERNEL);
785	if (unlikely(!srf->offsets)) {
786		ret = -ENOMEM;
787		goto out_no_offsets;
788	}
789
790	metadata->base_size = *srf->metadata.sizes;
791	metadata->autogen_filter = SVGA3D_TEX_FILTER_NONE;
792	metadata->multisample_count = 0;
793	metadata->multisample_pattern = SVGA3D_MS_PATTERN_NONE;
794	metadata->quality_level = SVGA3D_MS_QUALITY_NONE;
795
796	cur_bo_offset = 0;
797	cur_offset = srf->offsets;
798	cur_size = metadata->sizes;
799
800	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) {
801		for (j = 0; j < metadata->mip_levels[i]; ++j) {
802			uint32_t stride = vmw_surface_calculate_pitch(
803						  desc, cur_size);
804
805			cur_offset->face = i;
806			cur_offset->mip = j;
807			cur_offset->bo_offset = cur_bo_offset;
808			cur_bo_offset += vmw_surface_get_image_buffer_size
809				(desc, cur_size, stride);
810			++cur_offset;
811			++cur_size;
812		}
813	}
814	res->guest_memory_size = cur_bo_offset;
815	if (metadata->scanout &&
816	    metadata->num_sizes == 1 &&
817	    metadata->sizes[0].width == VMW_CURSOR_SNOOP_WIDTH &&
818	    metadata->sizes[0].height == VMW_CURSOR_SNOOP_HEIGHT &&
819	    metadata->format == VMW_CURSOR_SNOOP_FORMAT) {
820		const struct SVGA3dSurfaceDesc *desc =
821			vmw_surface_get_desc(VMW_CURSOR_SNOOP_FORMAT);
822		const u32 cursor_size_bytes = VMW_CURSOR_SNOOP_WIDTH *
823					      VMW_CURSOR_SNOOP_HEIGHT *
824					      desc->pitchBytesPerBlock;
825		srf->snooper.image = kzalloc(cursor_size_bytes, GFP_KERNEL);
826		if (!srf->snooper.image) {
827			DRM_ERROR("Failed to allocate cursor_image\n");
828			ret = -ENOMEM;
829			goto out_no_copy;
830		}
831	} else {
832		srf->snooper.image = NULL;
833	}
834
835	if (drm_is_primary_client(file_priv))
836		user_srf->master = drm_file_get_master(file_priv);
837
838	/**
839	 * From this point, the generic resource management functions
840	 * destroy the object on failure.
841	 */
842
843	ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free);
844	if (unlikely(ret != 0))
845		goto out_unlock;
846
847	/*
848	 * A gb-aware client referencing a surface will expect a backup
849	 * buffer to be present.
850	 */
851	if (dev_priv->has_mob) {
852		struct vmw_bo_params params = {
853			.domain = VMW_BO_DOMAIN_SYS,
854			.busy_domain = VMW_BO_DOMAIN_SYS,
855			.bo_type = ttm_bo_type_device,
856			.size = res->guest_memory_size,
857			.pin = false
858		};
859
860		ret = vmw_gem_object_create(dev_priv,
861					    &params,
862					    &res->guest_memory_bo);
863		if (unlikely(ret != 0)) {
864			vmw_resource_unreference(&res);
865			goto out_unlock;
866		}
867	}
868
869	tmp = vmw_resource_reference(&srf->res);
870	ret = ttm_prime_object_init(tfile, res->guest_memory_size,
871				    &user_srf->prime,
872				    VMW_RES_SURFACE,
873				    &vmw_user_surface_base_release);
874
875	if (unlikely(ret != 0)) {
876		vmw_resource_unreference(&tmp);
877		vmw_resource_unreference(&res);
878		goto out_unlock;
879	}
880
881	rep->sid = user_srf->prime.base.handle;
882	vmw_resource_unreference(&res);
883
884	return 0;
885out_no_copy:
886	kfree(srf->offsets);
887out_no_offsets:
888	kfree(metadata->sizes);
889out_no_sizes:
890	ttm_prime_object_kfree(user_srf, prime);
891out_unlock:
892	return ret;
893}
894
895
896static int
897vmw_surface_handle_reference(struct vmw_private *dev_priv,
898			     struct drm_file *file_priv,
899			     uint32_t u_handle,
900			     enum drm_vmw_handle_type handle_type,
901			     struct ttm_base_object **base_p)
902{
903	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
904	struct vmw_user_surface *user_srf;
905	uint32_t handle;
906	struct ttm_base_object *base;
907	int ret;
908
909	if (handle_type == DRM_VMW_HANDLE_PRIME) {
910		ret = ttm_prime_fd_to_handle(tfile, u_handle, &handle);
911		if (unlikely(ret != 0))
912			return ret;
913	} else {
914		handle = u_handle;
915	}
916
917	ret = -EINVAL;
918	base = ttm_base_object_lookup_for_ref(dev_priv->tdev, handle);
919	if (unlikely(!base)) {
920		VMW_DEBUG_USER("Could not find surface to reference.\n");
921		goto out_no_lookup;
922	}
923
924	if (unlikely(ttm_base_object_type(base) != VMW_RES_SURFACE)) {
925		VMW_DEBUG_USER("Referenced object is not a surface.\n");
926		goto out_bad_resource;
927	}
928	if (handle_type != DRM_VMW_HANDLE_PRIME) {
929		bool require_exist = false;
930
931		user_srf = container_of(base, struct vmw_user_surface,
932					prime.base);
933
934		/* Error out if we are unauthenticated primary */
935		if (drm_is_primary_client(file_priv) &&
936		    !file_priv->authenticated) {
937			ret = -EACCES;
938			goto out_bad_resource;
939		}
940
941		/*
942		 * Make sure the surface creator has the same
943		 * authenticating master, or is already registered with us.
944		 */
945		if (drm_is_primary_client(file_priv) &&
946		    user_srf->master != file_priv->master)
947			require_exist = true;
948
949		if (unlikely(drm_is_render_client(file_priv)))
950			require_exist = true;
951
952		ret = ttm_ref_object_add(tfile, base, NULL, require_exist);
953		if (unlikely(ret != 0)) {
954			DRM_ERROR("Could not add a reference to a surface.\n");
955			goto out_bad_resource;
956		}
957	}
958
959	*base_p = base;
960	return 0;
961
962out_bad_resource:
963	ttm_base_object_unref(&base);
964out_no_lookup:
965	if (handle_type == DRM_VMW_HANDLE_PRIME)
966		(void) ttm_ref_object_base_unref(tfile, handle);
967
968	return ret;
969}
970
971/**
972 * vmw_surface_reference_ioctl - Ioctl function implementing
973 *                                  the user surface reference functionality.
974 *
975 * @dev:            Pointer to a struct drm_device.
976 * @data:           Pointer to data copied from / to user-space.
977 * @file_priv:      Pointer to a drm file private structure.
978 */
979int vmw_surface_reference_ioctl(struct drm_device *dev, void *data,
980				struct drm_file *file_priv)
981{
982	struct vmw_private *dev_priv = vmw_priv(dev);
983	union drm_vmw_surface_reference_arg *arg =
984	    (union drm_vmw_surface_reference_arg *)data;
985	struct drm_vmw_surface_arg *req = &arg->req;
986	struct drm_vmw_surface_create_req *rep = &arg->rep;
987	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
988	struct vmw_surface *srf;
989	struct vmw_user_surface *user_srf;
990	struct drm_vmw_size __user *user_sizes;
991	struct ttm_base_object *base;
992	int ret;
993
994	ret = vmw_surface_handle_reference(dev_priv, file_priv, req->sid,
995					   req->handle_type, &base);
996	if (unlikely(ret != 0))
997		return ret;
998
999	user_srf = container_of(base, struct vmw_user_surface, prime.base);
1000	srf = &user_srf->srf;
1001
1002	/* Downcast of flags when sending back to user space */
1003	rep->flags = (uint32_t)srf->metadata.flags;
1004	rep->format = srf->metadata.format;
1005	memcpy(rep->mip_levels, srf->metadata.mip_levels,
1006	       sizeof(srf->metadata.mip_levels));
1007	user_sizes = (struct drm_vmw_size __user *)(unsigned long)
1008	    rep->size_addr;
1009
1010	if (user_sizes)
1011		ret = copy_to_user(user_sizes, &srf->metadata.base_size,
1012				   sizeof(srf->metadata.base_size));
1013	if (unlikely(ret != 0)) {
1014		VMW_DEBUG_USER("copy_to_user failed %p %u\n", user_sizes,
1015			       srf->metadata.num_sizes);
1016		ttm_ref_object_base_unref(tfile, base->handle);
1017		ret = -EFAULT;
1018	}
1019
1020	ttm_base_object_unref(&base);
1021
1022	return ret;
1023}
1024
1025/**
1026 * vmw_gb_surface_create - Encode a surface_define command.
1027 *
1028 * @res:        Pointer to a struct vmw_resource embedded in a struct
1029 *              vmw_surface.
1030 */
1031static int vmw_gb_surface_create(struct vmw_resource *res)
1032{
1033	struct vmw_private *dev_priv = res->dev_priv;
1034	struct vmw_surface *srf = vmw_res_to_srf(res);
1035	struct vmw_surface_metadata *metadata = &srf->metadata;
1036	uint32_t cmd_len, cmd_id, submit_len;
1037	int ret;
1038	struct {
1039		SVGA3dCmdHeader header;
1040		SVGA3dCmdDefineGBSurface body;
1041	} *cmd;
1042	struct {
1043		SVGA3dCmdHeader header;
1044		SVGA3dCmdDefineGBSurface_v2 body;
1045	} *cmd2;
1046	struct {
1047		SVGA3dCmdHeader header;
1048		SVGA3dCmdDefineGBSurface_v3 body;
1049	} *cmd3;
1050	struct {
1051		SVGA3dCmdHeader header;
1052		SVGA3dCmdDefineGBSurface_v4 body;
1053	} *cmd4;
1054
1055	if (likely(res->id != -1))
1056		return 0;
1057
1058	vmw_fifo_resource_inc(dev_priv);
1059	ret = vmw_resource_alloc_id(res);
1060	if (unlikely(ret != 0)) {
1061		DRM_ERROR("Failed to allocate a surface id.\n");
1062		goto out_no_id;
1063	}
1064
1065	if (unlikely(res->id >= VMWGFX_NUM_GB_SURFACE)) {
1066		ret = -EBUSY;
1067		goto out_no_fifo;
1068	}
1069
1070	if (has_sm5_context(dev_priv) && metadata->array_size > 0) {
1071		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE_V4;
1072		cmd_len = sizeof(cmd4->body);
1073		submit_len = sizeof(*cmd4);
1074	} else if (has_sm4_1_context(dev_priv) && metadata->array_size > 0) {
1075		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE_V3;
1076		cmd_len = sizeof(cmd3->body);
1077		submit_len = sizeof(*cmd3);
1078	} else if (metadata->array_size > 0) {
1079		/* VMW_SM_4 support verified at creation time. */
1080		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE_V2;
1081		cmd_len = sizeof(cmd2->body);
1082		submit_len = sizeof(*cmd2);
1083	} else {
1084		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE;
1085		cmd_len = sizeof(cmd->body);
1086		submit_len = sizeof(*cmd);
1087	}
1088
1089	cmd = VMW_CMD_RESERVE(dev_priv, submit_len);
1090	cmd2 = (typeof(cmd2))cmd;
1091	cmd3 = (typeof(cmd3))cmd;
1092	cmd4 = (typeof(cmd4))cmd;
1093	if (unlikely(!cmd)) {
1094		ret = -ENOMEM;
1095		goto out_no_fifo;
1096	}
1097
1098	if (has_sm5_context(dev_priv) && metadata->array_size > 0) {
1099		cmd4->header.id = cmd_id;
1100		cmd4->header.size = cmd_len;
1101		cmd4->body.sid = srf->res.id;
1102		cmd4->body.surfaceFlags = metadata->flags;
1103		cmd4->body.format = metadata->format;
1104		cmd4->body.numMipLevels = metadata->mip_levels[0];
1105		cmd4->body.multisampleCount = metadata->multisample_count;
1106		cmd4->body.multisamplePattern = metadata->multisample_pattern;
1107		cmd4->body.qualityLevel = metadata->quality_level;
1108		cmd4->body.autogenFilter = metadata->autogen_filter;
1109		cmd4->body.size.width = metadata->base_size.width;
1110		cmd4->body.size.height = metadata->base_size.height;
1111		cmd4->body.size.depth = metadata->base_size.depth;
1112		cmd4->body.arraySize = metadata->array_size;
1113		cmd4->body.bufferByteStride = metadata->buffer_byte_stride;
1114	} else if (has_sm4_1_context(dev_priv) && metadata->array_size > 0) {
1115		cmd3->header.id = cmd_id;
1116		cmd3->header.size = cmd_len;
1117		cmd3->body.sid = srf->res.id;
1118		cmd3->body.surfaceFlags = metadata->flags;
1119		cmd3->body.format = metadata->format;
1120		cmd3->body.numMipLevels = metadata->mip_levels[0];
1121		cmd3->body.multisampleCount = metadata->multisample_count;
1122		cmd3->body.multisamplePattern = metadata->multisample_pattern;
1123		cmd3->body.qualityLevel = metadata->quality_level;
1124		cmd3->body.autogenFilter = metadata->autogen_filter;
1125		cmd3->body.size.width = metadata->base_size.width;
1126		cmd3->body.size.height = metadata->base_size.height;
1127		cmd3->body.size.depth = metadata->base_size.depth;
1128		cmd3->body.arraySize = metadata->array_size;
1129	} else if (metadata->array_size > 0) {
1130		cmd2->header.id = cmd_id;
1131		cmd2->header.size = cmd_len;
1132		cmd2->body.sid = srf->res.id;
1133		cmd2->body.surfaceFlags = metadata->flags;
1134		cmd2->body.format = metadata->format;
1135		cmd2->body.numMipLevels = metadata->mip_levels[0];
1136		cmd2->body.multisampleCount = metadata->multisample_count;
1137		cmd2->body.autogenFilter = metadata->autogen_filter;
1138		cmd2->body.size.width = metadata->base_size.width;
1139		cmd2->body.size.height = metadata->base_size.height;
1140		cmd2->body.size.depth = metadata->base_size.depth;
1141		cmd2->body.arraySize = metadata->array_size;
1142	} else {
1143		cmd->header.id = cmd_id;
1144		cmd->header.size = cmd_len;
1145		cmd->body.sid = srf->res.id;
1146		cmd->body.surfaceFlags = metadata->flags;
1147		cmd->body.format = metadata->format;
1148		cmd->body.numMipLevels = metadata->mip_levels[0];
1149		cmd->body.multisampleCount = metadata->multisample_count;
1150		cmd->body.autogenFilter = metadata->autogen_filter;
1151		cmd->body.size.width = metadata->base_size.width;
1152		cmd->body.size.height = metadata->base_size.height;
1153		cmd->body.size.depth = metadata->base_size.depth;
1154	}
1155
1156	vmw_cmd_commit(dev_priv, submit_len);
1157
1158	return 0;
1159
1160out_no_fifo:
1161	vmw_resource_release_id(res);
1162out_no_id:
1163	vmw_fifo_resource_dec(dev_priv);
1164	return ret;
1165}
1166
1167
1168static int vmw_gb_surface_bind(struct vmw_resource *res,
1169			       struct ttm_validate_buffer *val_buf)
1170{
1171	struct vmw_private *dev_priv = res->dev_priv;
1172	struct {
1173		SVGA3dCmdHeader header;
1174		SVGA3dCmdBindGBSurface body;
1175	} *cmd1;
1176	struct {
1177		SVGA3dCmdHeader header;
1178		SVGA3dCmdUpdateGBSurface body;
1179	} *cmd2;
1180	uint32_t submit_size;
1181	struct ttm_buffer_object *bo = val_buf->bo;
1182
1183	BUG_ON(bo->resource->mem_type != VMW_PL_MOB);
1184
1185	submit_size = sizeof(*cmd1) + (res->guest_memory_dirty ? sizeof(*cmd2) : 0);
1186
1187	cmd1 = VMW_CMD_RESERVE(dev_priv, submit_size);
1188	if (unlikely(!cmd1))
1189		return -ENOMEM;
1190
1191	cmd1->header.id = SVGA_3D_CMD_BIND_GB_SURFACE;
1192	cmd1->header.size = sizeof(cmd1->body);
1193	cmd1->body.sid = res->id;
1194	cmd1->body.mobid = bo->resource->start;
1195	if (res->guest_memory_dirty) {
1196		cmd2 = (void *) &cmd1[1];
1197		cmd2->header.id = SVGA_3D_CMD_UPDATE_GB_SURFACE;
1198		cmd2->header.size = sizeof(cmd2->body);
1199		cmd2->body.sid = res->id;
1200	}
1201	vmw_cmd_commit(dev_priv, submit_size);
1202
1203	if (res->guest_memory_bo->dirty && res->guest_memory_dirty) {
1204		/* We've just made a full upload. Cear dirty regions. */
1205		vmw_bo_dirty_clear_res(res);
1206	}
1207
1208	res->guest_memory_dirty = false;
1209
1210	return 0;
1211}
1212
1213static int vmw_gb_surface_unbind(struct vmw_resource *res,
1214				 bool readback,
1215				 struct ttm_validate_buffer *val_buf)
1216{
1217	struct vmw_private *dev_priv = res->dev_priv;
1218	struct ttm_buffer_object *bo = val_buf->bo;
1219	struct vmw_fence_obj *fence;
1220
1221	struct {
1222		SVGA3dCmdHeader header;
1223		SVGA3dCmdReadbackGBSurface body;
1224	} *cmd1;
1225	struct {
1226		SVGA3dCmdHeader header;
1227		SVGA3dCmdInvalidateGBSurface body;
1228	} *cmd2;
1229	struct {
1230		SVGA3dCmdHeader header;
1231		SVGA3dCmdBindGBSurface body;
1232	} *cmd3;
1233	uint32_t submit_size;
1234	uint8_t *cmd;
1235
1236
1237	BUG_ON(bo->resource->mem_type != VMW_PL_MOB);
1238
1239	submit_size = sizeof(*cmd3) + (readback ? sizeof(*cmd1) : sizeof(*cmd2));
1240	cmd = VMW_CMD_RESERVE(dev_priv, submit_size);
1241	if (unlikely(!cmd))
1242		return -ENOMEM;
1243
1244	if (readback) {
1245		cmd1 = (void *) cmd;
1246		cmd1->header.id = SVGA_3D_CMD_READBACK_GB_SURFACE;
1247		cmd1->header.size = sizeof(cmd1->body);
1248		cmd1->body.sid = res->id;
1249		cmd3 = (void *) &cmd1[1];
1250	} else {
1251		cmd2 = (void *) cmd;
1252		cmd2->header.id = SVGA_3D_CMD_INVALIDATE_GB_SURFACE;
1253		cmd2->header.size = sizeof(cmd2->body);
1254		cmd2->body.sid = res->id;
1255		cmd3 = (void *) &cmd2[1];
1256	}
1257
1258	cmd3->header.id = SVGA_3D_CMD_BIND_GB_SURFACE;
1259	cmd3->header.size = sizeof(cmd3->body);
1260	cmd3->body.sid = res->id;
1261	cmd3->body.mobid = SVGA3D_INVALID_ID;
1262
1263	vmw_cmd_commit(dev_priv, submit_size);
1264
1265	/*
1266	 * Create a fence object and fence the backup buffer.
1267	 */
1268
1269	(void) vmw_execbuf_fence_commands(NULL, dev_priv,
1270					  &fence, NULL);
1271
1272	vmw_bo_fence_single(val_buf->bo, fence);
1273
1274	if (likely(fence != NULL))
1275		vmw_fence_obj_unreference(&fence);
1276
1277	return 0;
1278}
1279
1280static int vmw_gb_surface_destroy(struct vmw_resource *res)
1281{
1282	struct vmw_private *dev_priv = res->dev_priv;
1283	struct vmw_surface *srf = vmw_res_to_srf(res);
1284	struct {
1285		SVGA3dCmdHeader header;
1286		SVGA3dCmdDestroyGBSurface body;
1287	} *cmd;
1288
1289	if (likely(res->id == -1))
1290		return 0;
1291
1292	mutex_lock(&dev_priv->binding_mutex);
1293	vmw_view_surface_list_destroy(dev_priv, &srf->view_list);
1294	vmw_binding_res_list_scrub(&res->binding_head);
1295
1296	cmd = VMW_CMD_RESERVE(dev_priv, sizeof(*cmd));
1297	if (unlikely(!cmd)) {
1298		mutex_unlock(&dev_priv->binding_mutex);
1299		return -ENOMEM;
1300	}
1301
1302	cmd->header.id = SVGA_3D_CMD_DESTROY_GB_SURFACE;
1303	cmd->header.size = sizeof(cmd->body);
1304	cmd->body.sid = res->id;
1305	vmw_cmd_commit(dev_priv, sizeof(*cmd));
1306	mutex_unlock(&dev_priv->binding_mutex);
1307	vmw_resource_release_id(res);
1308	vmw_fifo_resource_dec(dev_priv);
1309
1310	return 0;
1311}
1312
1313/**
1314 * vmw_gb_surface_define_ioctl - Ioctl function implementing
1315 * the user surface define functionality.
1316 *
1317 * @dev: Pointer to a struct drm_device.
1318 * @data: Pointer to data copied from / to user-space.
1319 * @file_priv: Pointer to a drm file private structure.
1320 */
1321int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
1322				struct drm_file *file_priv)
1323{
1324	union drm_vmw_gb_surface_create_arg *arg =
1325	    (union drm_vmw_gb_surface_create_arg *)data;
1326	struct drm_vmw_gb_surface_create_rep *rep = &arg->rep;
1327	struct drm_vmw_gb_surface_create_ext_req req_ext;
1328
1329	req_ext.base = arg->req;
1330	req_ext.version = drm_vmw_gb_surface_v1;
1331	req_ext.svga3d_flags_upper_32_bits = 0;
1332	req_ext.multisample_pattern = SVGA3D_MS_PATTERN_NONE;
1333	req_ext.quality_level = SVGA3D_MS_QUALITY_NONE;
1334	req_ext.buffer_byte_stride = 0;
1335	req_ext.must_be_zero = 0;
1336
1337	return vmw_gb_surface_define_internal(dev, &req_ext, rep, file_priv);
1338}
1339
1340/**
1341 * vmw_gb_surface_reference_ioctl - Ioctl function implementing
1342 * the user surface reference functionality.
1343 *
1344 * @dev: Pointer to a struct drm_device.
1345 * @data: Pointer to data copied from / to user-space.
1346 * @file_priv: Pointer to a drm file private structure.
1347 */
1348int vmw_gb_surface_reference_ioctl(struct drm_device *dev, void *data,
1349				   struct drm_file *file_priv)
1350{
1351	union drm_vmw_gb_surface_reference_arg *arg =
1352	    (union drm_vmw_gb_surface_reference_arg *)data;
1353	struct drm_vmw_surface_arg *req = &arg->req;
1354	struct drm_vmw_gb_surface_ref_rep *rep = &arg->rep;
1355	struct drm_vmw_gb_surface_ref_ext_rep rep_ext;
1356	int ret;
1357
1358	ret = vmw_gb_surface_reference_internal(dev, req, &rep_ext, file_priv);
1359
1360	if (unlikely(ret != 0))
1361		return ret;
1362
1363	rep->creq = rep_ext.creq.base;
1364	rep->crep = rep_ext.crep;
1365
1366	return ret;
1367}
1368
1369/**
1370 * vmw_gb_surface_define_ext_ioctl - Ioctl function implementing
1371 * the user surface define functionality.
1372 *
1373 * @dev: Pointer to a struct drm_device.
1374 * @data: Pointer to data copied from / to user-space.
1375 * @file_priv: Pointer to a drm file private structure.
1376 */
1377int vmw_gb_surface_define_ext_ioctl(struct drm_device *dev, void *data,
1378				struct drm_file *file_priv)
1379{
1380	union drm_vmw_gb_surface_create_ext_arg *arg =
1381	    (union drm_vmw_gb_surface_create_ext_arg *)data;
1382	struct drm_vmw_gb_surface_create_ext_req *req = &arg->req;
1383	struct drm_vmw_gb_surface_create_rep *rep = &arg->rep;
1384
1385	return vmw_gb_surface_define_internal(dev, req, rep, file_priv);
1386}
1387
1388/**
1389 * vmw_gb_surface_reference_ext_ioctl - Ioctl function implementing
1390 * the user surface reference functionality.
1391 *
1392 * @dev: Pointer to a struct drm_device.
1393 * @data: Pointer to data copied from / to user-space.
1394 * @file_priv: Pointer to a drm file private structure.
1395 */
1396int vmw_gb_surface_reference_ext_ioctl(struct drm_device *dev, void *data,
1397				   struct drm_file *file_priv)
1398{
1399	union drm_vmw_gb_surface_reference_ext_arg *arg =
1400	    (union drm_vmw_gb_surface_reference_ext_arg *)data;
1401	struct drm_vmw_surface_arg *req = &arg->req;
1402	struct drm_vmw_gb_surface_ref_ext_rep *rep = &arg->rep;
1403
1404	return vmw_gb_surface_reference_internal(dev, req, rep, file_priv);
1405}
1406
1407/**
1408 * vmw_gb_surface_define_internal - Ioctl function implementing
1409 * the user surface define functionality.
1410 *
1411 * @dev: Pointer to a struct drm_device.
1412 * @req: Request argument from user-space.
1413 * @rep: Response argument to user-space.
1414 * @file_priv: Pointer to a drm file private structure.
1415 */
1416static int
1417vmw_gb_surface_define_internal(struct drm_device *dev,
1418			       struct drm_vmw_gb_surface_create_ext_req *req,
1419			       struct drm_vmw_gb_surface_create_rep *rep,
1420			       struct drm_file *file_priv)
1421{
1422	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
1423	struct vmw_private *dev_priv = vmw_priv(dev);
1424	struct vmw_user_surface *user_srf;
1425	struct vmw_surface_metadata metadata = {0};
1426	struct vmw_surface *srf;
1427	struct vmw_resource *res;
1428	struct vmw_resource *tmp;
1429	int ret = 0;
1430	uint32_t backup_handle = 0;
1431	SVGA3dSurfaceAllFlags svga3d_flags_64 =
1432		SVGA3D_FLAGS_64(req->svga3d_flags_upper_32_bits,
1433				req->base.svga3d_flags);
1434
1435	/* array_size must be null for non-GL3 host. */
1436	if (req->base.array_size > 0 && !has_sm4_context(dev_priv)) {
1437		VMW_DEBUG_USER("SM4 surface not supported.\n");
1438		return -EINVAL;
1439	}
1440
1441	if (!has_sm4_1_context(dev_priv)) {
1442		if (req->svga3d_flags_upper_32_bits != 0)
1443			ret = -EINVAL;
1444
1445		if (req->base.multisample_count != 0)
1446			ret = -EINVAL;
1447
1448		if (req->multisample_pattern != SVGA3D_MS_PATTERN_NONE)
1449			ret = -EINVAL;
1450
1451		if (req->quality_level != SVGA3D_MS_QUALITY_NONE)
1452			ret = -EINVAL;
1453
1454		if (ret) {
1455			VMW_DEBUG_USER("SM4.1 surface not supported.\n");
1456			return ret;
1457		}
1458	}
1459
1460	if (req->buffer_byte_stride > 0 && !has_sm5_context(dev_priv)) {
1461		VMW_DEBUG_USER("SM5 surface not supported.\n");
1462		return -EINVAL;
1463	}
1464
1465	if ((svga3d_flags_64 & SVGA3D_SURFACE_MULTISAMPLE) &&
1466	    req->base.multisample_count == 0) {
1467		VMW_DEBUG_USER("Invalid sample count.\n");
1468		return -EINVAL;
1469	}
1470
1471	if (req->base.mip_levels > DRM_VMW_MAX_MIP_LEVELS) {
1472		VMW_DEBUG_USER("Invalid mip level.\n");
1473		return -EINVAL;
1474	}
1475
1476	metadata.flags = svga3d_flags_64;
1477	metadata.format = req->base.format;
1478	metadata.mip_levels[0] = req->base.mip_levels;
1479	metadata.multisample_count = req->base.multisample_count;
1480	metadata.multisample_pattern = req->multisample_pattern;
1481	metadata.quality_level = req->quality_level;
1482	metadata.array_size = req->base.array_size;
1483	metadata.buffer_byte_stride = req->buffer_byte_stride;
1484	metadata.num_sizes = 1;
1485	metadata.base_size = req->base.base_size;
1486	metadata.scanout = req->base.drm_surface_flags &
1487		drm_vmw_surface_flag_scanout;
1488
1489	/* Define a surface based on the parameters. */
1490	ret = vmw_gb_surface_define(dev_priv, &metadata, &srf);
1491	if (ret != 0) {
1492		VMW_DEBUG_USER("Failed to define surface.\n");
1493		return ret;
1494	}
1495
1496	user_srf = container_of(srf, struct vmw_user_surface, srf);
1497	if (drm_is_primary_client(file_priv))
1498		user_srf->master = drm_file_get_master(file_priv);
1499
1500	res = &user_srf->srf.res;
1501
1502	if (req->base.buffer_handle != SVGA3D_INVALID_ID) {
1503		ret = vmw_user_bo_lookup(file_priv, req->base.buffer_handle,
1504					 &res->guest_memory_bo);
1505		if (ret == 0) {
1506			if (res->guest_memory_bo->tbo.base.size < res->guest_memory_size) {
1507				VMW_DEBUG_USER("Surface backup buffer too small.\n");
1508				vmw_user_bo_unref(&res->guest_memory_bo);
1509				ret = -EINVAL;
1510				goto out_unlock;
1511			} else {
1512				backup_handle = req->base.buffer_handle;
1513			}
1514		}
1515	} else if (req->base.drm_surface_flags &
1516		   (drm_vmw_surface_flag_create_buffer |
1517		    drm_vmw_surface_flag_coherent)) {
1518		ret = vmw_gem_object_create_with_handle(dev_priv, file_priv,
1519							res->guest_memory_size,
1520							&backup_handle,
1521							&res->guest_memory_bo);
1522	}
1523
1524	if (unlikely(ret != 0)) {
1525		vmw_resource_unreference(&res);
1526		goto out_unlock;
1527	}
1528
1529	if (req->base.drm_surface_flags & drm_vmw_surface_flag_coherent) {
1530		struct vmw_bo *backup = res->guest_memory_bo;
1531
1532		ttm_bo_reserve(&backup->tbo, false, false, NULL);
1533		if (!res->func->dirty_alloc)
1534			ret = -EINVAL;
1535		if (!ret)
1536			ret = vmw_bo_dirty_add(backup);
1537		if (!ret) {
1538			res->coherent = true;
1539			ret = res->func->dirty_alloc(res);
1540		}
1541		ttm_bo_unreserve(&backup->tbo);
1542		if (ret) {
1543			vmw_resource_unreference(&res);
1544			goto out_unlock;
1545		}
1546
1547	}
1548
1549	tmp = vmw_resource_reference(res);
1550	ret = ttm_prime_object_init(tfile, res->guest_memory_size, &user_srf->prime,
1551				    VMW_RES_SURFACE,
1552				    &vmw_user_surface_base_release);
1553
1554	if (unlikely(ret != 0)) {
1555		vmw_resource_unreference(&tmp);
1556		vmw_resource_unreference(&res);
1557		goto out_unlock;
1558	}
1559
1560	rep->handle      = user_srf->prime.base.handle;
1561	rep->backup_size = res->guest_memory_size;
1562	if (res->guest_memory_bo) {
1563		rep->buffer_map_handle =
1564			drm_vma_node_offset_addr(&res->guest_memory_bo->tbo.base.vma_node);
1565		rep->buffer_size = res->guest_memory_bo->tbo.base.size;
1566		rep->buffer_handle = backup_handle;
1567	} else {
1568		rep->buffer_map_handle = 0;
1569		rep->buffer_size = 0;
1570		rep->buffer_handle = SVGA3D_INVALID_ID;
1571	}
1572	vmw_resource_unreference(&res);
1573
1574out_unlock:
1575	return ret;
1576}
1577
1578/**
1579 * vmw_gb_surface_reference_internal - Ioctl function implementing
1580 * the user surface reference functionality.
1581 *
1582 * @dev: Pointer to a struct drm_device.
1583 * @req: Pointer to user-space request surface arg.
1584 * @rep: Pointer to response to user-space.
1585 * @file_priv: Pointer to a drm file private structure.
1586 */
1587static int
1588vmw_gb_surface_reference_internal(struct drm_device *dev,
1589				  struct drm_vmw_surface_arg *req,
1590				  struct drm_vmw_gb_surface_ref_ext_rep *rep,
1591				  struct drm_file *file_priv)
1592{
1593	struct vmw_private *dev_priv = vmw_priv(dev);
1594	struct vmw_surface *srf;
1595	struct vmw_user_surface *user_srf;
1596	struct vmw_surface_metadata *metadata;
1597	struct ttm_base_object *base;
1598	u32 backup_handle;
1599	int ret;
1600
1601	ret = vmw_surface_handle_reference(dev_priv, file_priv, req->sid,
1602					   req->handle_type, &base);
1603	if (unlikely(ret != 0))
1604		return ret;
1605
1606	user_srf = container_of(base, struct vmw_user_surface, prime.base);
1607	srf = &user_srf->srf;
1608	if (!srf->res.guest_memory_bo) {
1609		DRM_ERROR("Shared GB surface is missing a backup buffer.\n");
1610		goto out_bad_resource;
1611	}
1612	metadata = &srf->metadata;
1613
1614	mutex_lock(&dev_priv->cmdbuf_mutex); /* Protect res->backup */
1615	ret = drm_gem_handle_create(file_priv, &srf->res.guest_memory_bo->tbo.base,
1616				    &backup_handle);
1617	mutex_unlock(&dev_priv->cmdbuf_mutex);
1618	if (ret != 0) {
1619		drm_err(dev, "Wasn't able to create a backing handle for surface sid = %u.\n",
1620			req->sid);
1621		goto out_bad_resource;
1622	}
1623
1624	rep->creq.base.svga3d_flags = SVGA3D_FLAGS_LOWER_32(metadata->flags);
1625	rep->creq.base.format = metadata->format;
1626	rep->creq.base.mip_levels = metadata->mip_levels[0];
1627	rep->creq.base.drm_surface_flags = 0;
1628	rep->creq.base.multisample_count = metadata->multisample_count;
1629	rep->creq.base.autogen_filter = metadata->autogen_filter;
1630	rep->creq.base.array_size = metadata->array_size;
1631	rep->creq.base.buffer_handle = backup_handle;
1632	rep->creq.base.base_size = metadata->base_size;
1633	rep->crep.handle = user_srf->prime.base.handle;
1634	rep->crep.backup_size = srf->res.guest_memory_size;
1635	rep->crep.buffer_handle = backup_handle;
1636	rep->crep.buffer_map_handle =
1637		drm_vma_node_offset_addr(&srf->res.guest_memory_bo->tbo.base.vma_node);
1638	rep->crep.buffer_size = srf->res.guest_memory_bo->tbo.base.size;
1639
1640	rep->creq.version = drm_vmw_gb_surface_v1;
1641	rep->creq.svga3d_flags_upper_32_bits =
1642		SVGA3D_FLAGS_UPPER_32(metadata->flags);
1643	rep->creq.multisample_pattern = metadata->multisample_pattern;
1644	rep->creq.quality_level = metadata->quality_level;
1645	rep->creq.must_be_zero = 0;
1646
1647out_bad_resource:
1648	ttm_base_object_unref(&base);
1649
1650	return ret;
1651}
1652
1653/**
1654 * vmw_subres_dirty_add - Add a dirty region to a subresource
1655 * @dirty: The surfaces's dirty tracker.
1656 * @loc_start: The location corresponding to the start of the region.
1657 * @loc_end: The location corresponding to the end of the region.
1658 *
1659 * As we are assuming that @loc_start and @loc_end represent a sequential
1660 * range of backing store memory, if the region spans multiple lines then
1661 * regardless of the x coordinate, the full lines are dirtied.
1662 * Correspondingly if the region spans multiple z slices, then full rather
1663 * than partial z slices are dirtied.
1664 */
1665static void vmw_subres_dirty_add(struct vmw_surface_dirty *dirty,
1666				 const struct vmw_surface_loc *loc_start,
1667				 const struct vmw_surface_loc *loc_end)
1668{
1669	const struct vmw_surface_cache *cache = &dirty->cache;
1670	SVGA3dBox *box = &dirty->boxes[loc_start->sub_resource];
1671	u32 mip = loc_start->sub_resource % cache->num_mip_levels;
1672	const struct drm_vmw_size *size = &cache->mip[mip].size;
1673	u32 box_c2 = box->z + box->d;
1674
1675	if (WARN_ON(loc_start->sub_resource >= dirty->num_subres))
1676		return;
1677
1678	if (box->d == 0 || box->z > loc_start->z)
1679		box->z = loc_start->z;
1680	if (box_c2 < loc_end->z)
1681		box->d = loc_end->z - box->z;
1682
1683	if (loc_start->z + 1 == loc_end->z) {
1684		box_c2 = box->y + box->h;
1685		if (box->h == 0 || box->y > loc_start->y)
1686			box->y = loc_start->y;
1687		if (box_c2 < loc_end->y)
1688			box->h = loc_end->y - box->y;
1689
1690		if (loc_start->y + 1 == loc_end->y) {
1691			box_c2 = box->x + box->w;
1692			if (box->w == 0 || box->x > loc_start->x)
1693				box->x = loc_start->x;
1694			if (box_c2 < loc_end->x)
1695				box->w = loc_end->x - box->x;
1696		} else {
1697			box->x = 0;
1698			box->w = size->width;
1699		}
1700	} else {
1701		box->y = 0;
1702		box->h = size->height;
1703		box->x = 0;
1704		box->w = size->width;
1705	}
1706}
1707
1708/**
1709 * vmw_subres_dirty_full - Mark a full subresource as dirty
1710 * @dirty: The surface's dirty tracker.
1711 * @subres: The subresource
1712 */
1713static void vmw_subres_dirty_full(struct vmw_surface_dirty *dirty, u32 subres)
1714{
1715	const struct vmw_surface_cache *cache = &dirty->cache;
1716	u32 mip = subres % cache->num_mip_levels;
1717	const struct drm_vmw_size *size = &cache->mip[mip].size;
1718	SVGA3dBox *box = &dirty->boxes[subres];
1719
1720	box->x = 0;
1721	box->y = 0;
1722	box->z = 0;
1723	box->w = size->width;
1724	box->h = size->height;
1725	box->d = size->depth;
1726}
1727
1728/*
1729 * vmw_surface_tex_dirty_add_range - The dirty_add_range callback for texture
1730 * surfaces.
1731 */
1732static void vmw_surface_tex_dirty_range_add(struct vmw_resource *res,
1733					    size_t start, size_t end)
1734{
1735	struct vmw_surface_dirty *dirty =
1736		(struct vmw_surface_dirty *) res->dirty;
1737	size_t backup_end = res->guest_memory_offset + res->guest_memory_size;
1738	struct vmw_surface_loc loc1, loc2;
1739	const struct vmw_surface_cache *cache;
1740
1741	start = max_t(size_t, start, res->guest_memory_offset) - res->guest_memory_offset;
1742	end = min(end, backup_end) - res->guest_memory_offset;
1743	cache = &dirty->cache;
1744	vmw_surface_get_loc(cache, &loc1, start);
1745	vmw_surface_get_loc(cache, &loc2, end - 1);
1746	vmw_surface_inc_loc(cache, &loc2);
1747
1748	if (loc1.sheet != loc2.sheet) {
1749		u32 sub_res;
1750
1751		/*
1752		 * Multiple multisample sheets. To do this in an optimized
1753		 * fashion, compute the dirty region for each sheet and the
1754		 * resulting union. Since this is not a common case, just dirty
1755		 * the whole surface.
1756		 */
1757		for (sub_res = 0; sub_res < dirty->num_subres; ++sub_res)
1758			vmw_subres_dirty_full(dirty, sub_res);
1759		return;
1760	}
1761	if (loc1.sub_resource + 1 == loc2.sub_resource) {
1762		/* Dirty range covers a single sub-resource */
1763		vmw_subres_dirty_add(dirty, &loc1, &loc2);
1764	} else {
1765		/* Dirty range covers multiple sub-resources */
1766		struct vmw_surface_loc loc_min, loc_max;
1767		u32 sub_res;
1768
1769		vmw_surface_max_loc(cache, loc1.sub_resource, &loc_max);
1770		vmw_subres_dirty_add(dirty, &loc1, &loc_max);
1771		vmw_surface_min_loc(cache, loc2.sub_resource - 1, &loc_min);
1772		vmw_subres_dirty_add(dirty, &loc_min, &loc2);
1773		for (sub_res = loc1.sub_resource + 1;
1774		     sub_res < loc2.sub_resource - 1; ++sub_res)
1775			vmw_subres_dirty_full(dirty, sub_res);
1776	}
1777}
1778
1779/*
1780 * vmw_surface_tex_dirty_add_range - The dirty_add_range callback for buffer
1781 * surfaces.
1782 */
1783static void vmw_surface_buf_dirty_range_add(struct vmw_resource *res,
1784					    size_t start, size_t end)
1785{
1786	struct vmw_surface_dirty *dirty =
1787		(struct vmw_surface_dirty *) res->dirty;
1788	const struct vmw_surface_cache *cache = &dirty->cache;
1789	size_t backup_end = res->guest_memory_offset + cache->mip_chain_bytes;
1790	SVGA3dBox *box = &dirty->boxes[0];
1791	u32 box_c2;
1792
1793	box->h = box->d = 1;
1794	start = max_t(size_t, start, res->guest_memory_offset) - res->guest_memory_offset;
1795	end = min(end, backup_end) - res->guest_memory_offset;
1796	box_c2 = box->x + box->w;
1797	if (box->w == 0 || box->x > start)
1798		box->x = start;
1799	if (box_c2 < end)
1800		box->w = end - box->x;
1801}
1802
1803/*
1804 * vmw_surface_tex_dirty_add_range - The dirty_add_range callback for surfaces
1805 */
1806static void vmw_surface_dirty_range_add(struct vmw_resource *res, size_t start,
1807					size_t end)
1808{
1809	struct vmw_surface *srf = vmw_res_to_srf(res);
1810
1811	if (WARN_ON(end <= res->guest_memory_offset ||
1812		    start >= res->guest_memory_offset + res->guest_memory_size))
1813		return;
1814
1815	if (srf->metadata.format == SVGA3D_BUFFER)
1816		vmw_surface_buf_dirty_range_add(res, start, end);
1817	else
1818		vmw_surface_tex_dirty_range_add(res, start, end);
1819}
1820
1821/*
1822 * vmw_surface_dirty_sync - The surface's dirty_sync callback.
1823 */
1824static int vmw_surface_dirty_sync(struct vmw_resource *res)
1825{
1826	struct vmw_private *dev_priv = res->dev_priv;
1827	u32 i, num_dirty;
1828	struct vmw_surface_dirty *dirty =
1829		(struct vmw_surface_dirty *) res->dirty;
1830	size_t alloc_size;
1831	const struct vmw_surface_cache *cache = &dirty->cache;
1832	struct {
1833		SVGA3dCmdHeader header;
1834		SVGA3dCmdDXUpdateSubResource body;
1835	} *cmd1;
1836	struct {
1837		SVGA3dCmdHeader header;
1838		SVGA3dCmdUpdateGBImage body;
1839	} *cmd2;
1840	void *cmd;
1841
1842	num_dirty = 0;
1843	for (i = 0; i < dirty->num_subres; ++i) {
1844		const SVGA3dBox *box = &dirty->boxes[i];
1845
1846		if (box->d)
1847			num_dirty++;
1848	}
1849
1850	if (!num_dirty)
1851		goto out;
1852
1853	alloc_size = num_dirty * ((has_sm4_context(dev_priv)) ? sizeof(*cmd1) : sizeof(*cmd2));
1854	cmd = VMW_CMD_RESERVE(dev_priv, alloc_size);
1855	if (!cmd)
1856		return -ENOMEM;
1857
1858	cmd1 = cmd;
1859	cmd2 = cmd;
1860
1861	for (i = 0; i < dirty->num_subres; ++i) {
1862		const SVGA3dBox *box = &dirty->boxes[i];
1863
1864		if (!box->d)
1865			continue;
1866
1867		/*
1868		 * DX_UPDATE_SUBRESOURCE is aware of array surfaces.
1869		 * UPDATE_GB_IMAGE is not.
1870		 */
1871		if (has_sm4_context(dev_priv)) {
1872			cmd1->header.id = SVGA_3D_CMD_DX_UPDATE_SUBRESOURCE;
1873			cmd1->header.size = sizeof(cmd1->body);
1874			cmd1->body.sid = res->id;
1875			cmd1->body.subResource = i;
1876			cmd1->body.box = *box;
1877			cmd1++;
1878		} else {
1879			cmd2->header.id = SVGA_3D_CMD_UPDATE_GB_IMAGE;
1880			cmd2->header.size = sizeof(cmd2->body);
1881			cmd2->body.image.sid = res->id;
1882			cmd2->body.image.face = i / cache->num_mip_levels;
1883			cmd2->body.image.mipmap = i -
1884				(cache->num_mip_levels * cmd2->body.image.face);
1885			cmd2->body.box = *box;
1886			cmd2++;
1887		}
1888
1889	}
1890	vmw_cmd_commit(dev_priv, alloc_size);
1891 out:
1892	memset(&dirty->boxes[0], 0, sizeof(dirty->boxes[0]) *
1893	       dirty->num_subres);
1894
1895	return 0;
1896}
1897
1898/*
1899 * vmw_surface_dirty_alloc - The surface's dirty_alloc callback.
1900 */
1901static int vmw_surface_dirty_alloc(struct vmw_resource *res)
1902{
1903	struct vmw_surface *srf = vmw_res_to_srf(res);
1904	const struct vmw_surface_metadata *metadata = &srf->metadata;
1905	struct vmw_surface_dirty *dirty;
1906	u32 num_layers = 1;
1907	u32 num_mip;
1908	u32 num_subres;
1909	u32 num_samples;
1910	size_t dirty_size;
1911	int ret;
1912
1913	if (metadata->array_size)
1914		num_layers = metadata->array_size;
1915	else if (metadata->flags & SVGA3D_SURFACE_CUBEMAP)
1916		num_layers *= SVGA3D_MAX_SURFACE_FACES;
1917
1918	num_mip = metadata->mip_levels[0];
1919	if (!num_mip)
1920		num_mip = 1;
1921
1922	num_subres = num_layers * num_mip;
1923	dirty_size = struct_size(dirty, boxes, num_subres);
1924
1925	dirty = kvzalloc(dirty_size, GFP_KERNEL);
1926	if (!dirty) {
1927		ret = -ENOMEM;
1928		goto out_no_dirty;
1929	}
1930
1931	num_samples = max_t(u32, 1, metadata->multisample_count);
1932	ret = vmw_surface_setup_cache(&metadata->base_size, metadata->format,
1933				      num_mip, num_layers, num_samples,
1934				      &dirty->cache);
1935	if (ret)
1936		goto out_no_cache;
1937
1938	dirty->num_subres = num_subres;
1939	res->dirty = (struct vmw_resource_dirty *) dirty;
1940
1941	return 0;
1942
1943out_no_cache:
1944	kvfree(dirty);
1945out_no_dirty:
1946	return ret;
1947}
1948
1949/*
1950 * vmw_surface_dirty_free - The surface's dirty_free callback
1951 */
1952static void vmw_surface_dirty_free(struct vmw_resource *res)
1953{
1954	struct vmw_surface_dirty *dirty =
1955		(struct vmw_surface_dirty *) res->dirty;
1956
1957	kvfree(dirty);
1958	res->dirty = NULL;
1959}
1960
1961/*
1962 * vmw_surface_clean - The surface's clean callback
1963 */
1964static int vmw_surface_clean(struct vmw_resource *res)
1965{
1966	struct vmw_private *dev_priv = res->dev_priv;
1967	size_t alloc_size;
1968	struct {
1969		SVGA3dCmdHeader header;
1970		SVGA3dCmdReadbackGBSurface body;
1971	} *cmd;
1972
1973	alloc_size = sizeof(*cmd);
1974	cmd = VMW_CMD_RESERVE(dev_priv, alloc_size);
1975	if (!cmd)
1976		return -ENOMEM;
1977
1978	cmd->header.id = SVGA_3D_CMD_READBACK_GB_SURFACE;
1979	cmd->header.size = sizeof(cmd->body);
1980	cmd->body.sid = res->id;
1981	vmw_cmd_commit(dev_priv, alloc_size);
1982
1983	return 0;
1984}
1985
1986/*
1987 * vmw_gb_surface_define - Define a private GB surface
1988 *
1989 * @dev_priv: Pointer to a device private.
1990 * @metadata: Metadata representing the surface to create.
1991 * @user_srf_out: allocated user_srf. Set to NULL on failure.
1992 *
1993 * GB surfaces allocated by this function will not have a user mode handle, and
1994 * thus will only be visible to vmwgfx.  For optimization reasons the
1995 * surface may later be given a user mode handle by another function to make
1996 * it available to user mode drivers.
1997 */
1998int vmw_gb_surface_define(struct vmw_private *dev_priv,
1999			  const struct vmw_surface_metadata *req,
2000			  struct vmw_surface **srf_out)
2001{
2002	struct vmw_surface_metadata *metadata;
2003	struct vmw_user_surface *user_srf;
2004	struct vmw_surface *srf;
2005	u32 sample_count = 1;
2006	u32 num_layers = 1;
2007	int ret;
2008
2009	*srf_out = NULL;
2010
2011	if (req->scanout) {
2012		if (!vmw_surface_is_screen_target_format(req->format)) {
2013			VMW_DEBUG_USER("Invalid Screen Target surface format.");
2014			return -EINVAL;
2015		}
2016
2017		if (req->base_size.width > dev_priv->texture_max_width ||
2018		    req->base_size.height > dev_priv->texture_max_height) {
2019			VMW_DEBUG_USER("%ux%u\n, exceed max surface size %ux%u",
2020				       req->base_size.width,
2021				       req->base_size.height,
2022				       dev_priv->texture_max_width,
2023				       dev_priv->texture_max_height);
2024			return -EINVAL;
2025		}
2026	} else {
2027		const SVGA3dSurfaceDesc *desc =
2028			vmw_surface_get_desc(req->format);
2029
2030		if (desc->blockDesc == SVGA3DBLOCKDESC_NONE) {
2031			VMW_DEBUG_USER("Invalid surface format.\n");
2032			return -EINVAL;
2033		}
2034	}
2035
2036	if (req->autogen_filter != SVGA3D_TEX_FILTER_NONE)
2037		return -EINVAL;
2038
2039	if (req->num_sizes != 1)
2040		return -EINVAL;
2041
2042	if (req->sizes != NULL)
2043		return -EINVAL;
2044
2045	user_srf = kzalloc(sizeof(*user_srf), GFP_KERNEL);
2046	if (unlikely(!user_srf)) {
2047		ret = -ENOMEM;
2048		goto out_unlock;
2049	}
2050
2051	*srf_out  = &user_srf->srf;
2052
2053	srf = &user_srf->srf;
2054	srf->metadata = *req;
2055	srf->offsets = NULL;
2056
2057	metadata = &srf->metadata;
2058
2059	if (metadata->array_size)
2060		num_layers = req->array_size;
2061	else if (metadata->flags & SVGA3D_SURFACE_CUBEMAP)
2062		num_layers = SVGA3D_MAX_SURFACE_FACES;
2063
2064	if (metadata->flags & SVGA3D_SURFACE_MULTISAMPLE)
2065		sample_count = metadata->multisample_count;
2066
2067	srf->res.guest_memory_size =
2068		vmw_surface_get_serialized_size_extended(
2069				metadata->format,
2070				metadata->base_size,
2071				metadata->mip_levels[0],
2072				num_layers,
2073				sample_count);
2074
2075	if (metadata->flags & SVGA3D_SURFACE_BIND_STREAM_OUTPUT)
2076		srf->res.guest_memory_size += sizeof(SVGA3dDXSOState);
2077
2078	/*
2079	 * Don't set SVGA3D_SURFACE_SCREENTARGET flag for a scanout surface with
2080	 * size greater than STDU max width/height. This is really a workaround
2081	 * to support creation of big framebuffer requested by some user-space
2082	 * for whole topology. That big framebuffer won't really be used for
2083	 * binding with screen target as during prepare_fb a separate surface is
2084	 * created so it's safe to ignore SVGA3D_SURFACE_SCREENTARGET flag.
2085	 */
2086	if (dev_priv->active_display_unit == vmw_du_screen_target &&
2087	    metadata->scanout &&
2088	    metadata->base_size.width <= dev_priv->stdu_max_width &&
2089	    metadata->base_size.height <= dev_priv->stdu_max_height)
2090		metadata->flags |= SVGA3D_SURFACE_SCREENTARGET;
2091
2092	/*
2093	 * From this point, the generic resource management functions
2094	 * destroy the object on failure.
2095	 */
2096	ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free);
2097
2098	return ret;
2099
2100out_unlock:
2101	return ret;
2102}
2103