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 * @base:           The TTM base object handling user-space visibility.
48 * @srf:            The surface metadata.
49 * @master:         Master of the creating client. Used for security check.
50 */
51struct vmw_user_surface {
52	struct ttm_prime_object prime;
53	struct vmw_surface srf;
54	struct drm_master *master;
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 */
65struct vmw_surface_offset {
66	uint32_t face;
67	uint32_t mip;
68	uint32_t bo_offset;
69};
70
71/**
72 * struct vmw_surface_dirty - Surface dirty-tracker
73 * @cache: Cached layout information of the surface.
74 * @num_subres: Number of subresources.
75 * @boxes: Array of SVGA3dBoxes indicating dirty regions. One per subresource.
76 */
77struct vmw_surface_dirty {
78	struct vmw_surface_cache cache;
79	u32 num_subres;
80	SVGA3dBox boxes[];
81};
82
83static void vmw_user_surface_free(struct vmw_resource *res);
84static struct vmw_resource *
85vmw_user_surface_base_to_res(struct ttm_base_object *base);
86static int vmw_legacy_srf_bind(struct vmw_resource *res,
87			       struct ttm_validate_buffer *val_buf);
88static int vmw_legacy_srf_unbind(struct vmw_resource *res,
89				 bool readback,
90				 struct ttm_validate_buffer *val_buf);
91static int vmw_legacy_srf_create(struct vmw_resource *res);
92static int vmw_legacy_srf_destroy(struct vmw_resource *res);
93static int vmw_gb_surface_create(struct vmw_resource *res);
94static int vmw_gb_surface_bind(struct vmw_resource *res,
95			       struct ttm_validate_buffer *val_buf);
96static int vmw_gb_surface_unbind(struct vmw_resource *res,
97				 bool readback,
98				 struct ttm_validate_buffer *val_buf);
99static int vmw_gb_surface_destroy(struct vmw_resource *res);
100static int
101vmw_gb_surface_define_internal(struct drm_device *dev,
102			       struct drm_vmw_gb_surface_create_ext_req *req,
103			       struct drm_vmw_gb_surface_create_rep *rep,
104			       struct drm_file *file_priv);
105static int
106vmw_gb_surface_reference_internal(struct drm_device *dev,
107				  struct drm_vmw_surface_arg *req,
108				  struct drm_vmw_gb_surface_ref_ext_rep *rep,
109				  struct drm_file *file_priv);
110
111static void vmw_surface_dirty_free(struct vmw_resource *res);
112static int vmw_surface_dirty_alloc(struct vmw_resource *res);
113static int vmw_surface_dirty_sync(struct vmw_resource *res);
114static void vmw_surface_dirty_range_add(struct vmw_resource *res, size_t start,
115					size_t end);
116static int vmw_surface_clean(struct vmw_resource *res);
117
118static const struct vmw_user_resource_conv user_surface_conv = {
119	.object_type = VMW_RES_SURFACE,
120	.base_obj_to_res = vmw_user_surface_base_to_res,
121	.res_free = vmw_user_surface_free
122};
123
124const struct vmw_user_resource_conv *user_surface_converter =
125	&user_surface_conv;
126
127static const struct vmw_res_func vmw_legacy_surface_func = {
128	.res_type = vmw_res_surface,
129	.needs_guest_memory = false,
130	.may_evict = true,
131	.prio = 1,
132	.dirty_prio = 1,
133	.type_name = "legacy surfaces",
134	.domain = VMW_BO_DOMAIN_GMR,
135	.busy_domain = VMW_BO_DOMAIN_GMR | VMW_BO_DOMAIN_VRAM,
136	.create = &vmw_legacy_srf_create,
137	.destroy = &vmw_legacy_srf_destroy,
138	.bind = &vmw_legacy_srf_bind,
139	.unbind = &vmw_legacy_srf_unbind
140};
141
142static const struct vmw_res_func vmw_gb_surface_func = {
143	.res_type = vmw_res_surface,
144	.needs_guest_memory = true,
145	.may_evict = true,
146	.prio = 1,
147	.dirty_prio = 2,
148	.type_name = "guest backed surfaces",
149	.domain = VMW_BO_DOMAIN_MOB,
150	.busy_domain = VMW_BO_DOMAIN_MOB,
151	.create = vmw_gb_surface_create,
152	.destroy = vmw_gb_surface_destroy,
153	.bind = vmw_gb_surface_bind,
154	.unbind = vmw_gb_surface_unbind,
155	.dirty_alloc = vmw_surface_dirty_alloc,
156	.dirty_free = vmw_surface_dirty_free,
157	.dirty_sync = vmw_surface_dirty_sync,
158	.dirty_range_add = vmw_surface_dirty_range_add,
159	.clean = vmw_surface_clean,
160};
161
162/*
163 * struct vmw_surface_dma - SVGA3D DMA command
164 */
165struct vmw_surface_dma {
166	SVGA3dCmdHeader header;
167	SVGA3dCmdSurfaceDMA body;
168	SVGA3dCopyBox cb;
169	SVGA3dCmdSurfaceDMASuffix suffix;
170};
171
172/*
173 * struct vmw_surface_define - SVGA3D Surface Define command
174 */
175struct vmw_surface_define {
176	SVGA3dCmdHeader header;
177	SVGA3dCmdDefineSurface body;
178};
179
180/*
181 * struct vmw_surface_destroy - SVGA3D Surface Destroy command
182 */
183struct vmw_surface_destroy {
184	SVGA3dCmdHeader header;
185	SVGA3dCmdDestroySurface body;
186};
187
188
189/**
190 * vmw_surface_dma_size - Compute fifo size for a dma command.
191 *
192 * @srf: Pointer to a struct vmw_surface
193 *
194 * Computes the required size for a surface dma command for backup or
195 * restoration of the surface represented by @srf.
196 */
197static inline uint32_t vmw_surface_dma_size(const struct vmw_surface *srf)
198{
199	return srf->metadata.num_sizes * sizeof(struct vmw_surface_dma);
200}
201
202
203/**
204 * vmw_surface_define_size - Compute fifo size for a surface define command.
205 *
206 * @srf: Pointer to a struct vmw_surface
207 *
208 * Computes the required size for a surface define command for the definition
209 * of the surface represented by @srf.
210 */
211static inline uint32_t vmw_surface_define_size(const struct vmw_surface *srf)
212{
213	return sizeof(struct vmw_surface_define) + srf->metadata.num_sizes *
214		sizeof(SVGA3dSize);
215}
216
217
218/**
219 * vmw_surface_destroy_size - Compute fifo size for a surface destroy command.
220 *
221 * Computes the required size for a surface destroy command for the destruction
222 * of a hw surface.
223 */
224static inline uint32_t vmw_surface_destroy_size(void)
225{
226	return sizeof(struct vmw_surface_destroy);
227}
228
229/**
230 * vmw_surface_destroy_encode - Encode a surface_destroy command.
231 *
232 * @id: The surface id
233 * @cmd_space: Pointer to memory area in which the commands should be encoded.
234 */
235static void vmw_surface_destroy_encode(uint32_t id,
236				       void *cmd_space)
237{
238	struct vmw_surface_destroy *cmd = (struct vmw_surface_destroy *)
239		cmd_space;
240
241	cmd->header.id = SVGA_3D_CMD_SURFACE_DESTROY;
242	cmd->header.size = sizeof(cmd->body);
243	cmd->body.sid = id;
244}
245
246/**
247 * vmw_surface_define_encode - Encode a surface_define command.
248 *
249 * @srf: Pointer to a struct vmw_surface object.
250 * @cmd_space: Pointer to memory area in which the commands should be encoded.
251 */
252static void vmw_surface_define_encode(const struct vmw_surface *srf,
253				      void *cmd_space)
254{
255	struct vmw_surface_define *cmd = (struct vmw_surface_define *)
256		cmd_space;
257	struct drm_vmw_size *src_size;
258	SVGA3dSize *cmd_size;
259	uint32_t cmd_len;
260	int i;
261
262	cmd_len = sizeof(cmd->body) + srf->metadata.num_sizes *
263		sizeof(SVGA3dSize);
264
265	cmd->header.id = SVGA_3D_CMD_SURFACE_DEFINE;
266	cmd->header.size = cmd_len;
267	cmd->body.sid = srf->res.id;
268	/*
269	 * Downcast of surfaceFlags, was upcasted when received from user-space,
270	 * since driver internally stores as 64 bit.
271	 * For legacy surface define only 32 bit flag is supported.
272	 */
273	cmd->body.surfaceFlags = (SVGA3dSurface1Flags)srf->metadata.flags;
274	cmd->body.format = srf->metadata.format;
275	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i)
276		cmd->body.face[i].numMipLevels = srf->metadata.mip_levels[i];
277
278	cmd += 1;
279	cmd_size = (SVGA3dSize *) cmd;
280	src_size = srf->metadata.sizes;
281
282	for (i = 0; i < srf->metadata.num_sizes; ++i, cmd_size++, src_size++) {
283		cmd_size->width = src_size->width;
284		cmd_size->height = src_size->height;
285		cmd_size->depth = src_size->depth;
286	}
287}
288
289/**
290 * vmw_surface_dma_encode - Encode a surface_dma command.
291 *
292 * @srf: Pointer to a struct vmw_surface object.
293 * @cmd_space: Pointer to memory area in which the commands should be encoded.
294 * @ptr: Pointer to an SVGAGuestPtr indicating where the surface contents
295 * should be placed or read from.
296 * @to_surface: Boolean whether to DMA to the surface or from the surface.
297 */
298static void vmw_surface_dma_encode(struct vmw_surface *srf,
299				   void *cmd_space,
300				   const SVGAGuestPtr *ptr,
301				   bool to_surface)
302{
303	uint32_t i;
304	struct vmw_surface_dma *cmd = (struct vmw_surface_dma *)cmd_space;
305	const struct SVGA3dSurfaceDesc *desc =
306		vmw_surface_get_desc(srf->metadata.format);
307
308	for (i = 0; i < srf->metadata.num_sizes; ++i) {
309		SVGA3dCmdHeader *header = &cmd->header;
310		SVGA3dCmdSurfaceDMA *body = &cmd->body;
311		SVGA3dCopyBox *cb = &cmd->cb;
312		SVGA3dCmdSurfaceDMASuffix *suffix = &cmd->suffix;
313		const struct vmw_surface_offset *cur_offset = &srf->offsets[i];
314		const struct drm_vmw_size *cur_size = &srf->metadata.sizes[i];
315
316		header->id = SVGA_3D_CMD_SURFACE_DMA;
317		header->size = sizeof(*body) + sizeof(*cb) + sizeof(*suffix);
318
319		body->guest.ptr = *ptr;
320		body->guest.ptr.offset += cur_offset->bo_offset;
321		body->guest.pitch = vmw_surface_calculate_pitch(desc, cur_size);
322		body->host.sid = srf->res.id;
323		body->host.face = cur_offset->face;
324		body->host.mipmap = cur_offset->mip;
325		body->transfer = ((to_surface) ?  SVGA3D_WRITE_HOST_VRAM :
326				  SVGA3D_READ_HOST_VRAM);
327		cb->x = 0;
328		cb->y = 0;
329		cb->z = 0;
330		cb->srcx = 0;
331		cb->srcy = 0;
332		cb->srcz = 0;
333		cb->w = cur_size->width;
334		cb->h = cur_size->height;
335		cb->d = cur_size->depth;
336
337		suffix->suffixSize = sizeof(*suffix);
338		suffix->maximumOffset =
339			vmw_surface_get_image_buffer_size(desc, cur_size,
340							    body->guest.pitch);
341		suffix->flags.discard = 0;
342		suffix->flags.unsynchronized = 0;
343		suffix->flags.reserved = 0;
344		++cmd;
345	}
346};
347
348
349/**
350 * vmw_hw_surface_destroy - destroy a Device surface
351 *
352 * @res:        Pointer to a struct vmw_resource embedded in a struct
353 *              vmw_surface.
354 *
355 * Destroys a the device surface associated with a struct vmw_surface if
356 * any, and adjusts resource count accordingly.
357 */
358static void vmw_hw_surface_destroy(struct vmw_resource *res)
359{
360
361	struct vmw_private *dev_priv = res->dev_priv;
362	void *cmd;
363
364	if (res->func->destroy == vmw_gb_surface_destroy) {
365		(void) vmw_gb_surface_destroy(res);
366		return;
367	}
368
369	if (res->id != -1) {
370
371		cmd = VMW_CMD_RESERVE(dev_priv, vmw_surface_destroy_size());
372		if (unlikely(!cmd))
373			return;
374
375		vmw_surface_destroy_encode(res->id, cmd);
376		vmw_cmd_commit(dev_priv, vmw_surface_destroy_size());
377
378		/*
379		 * used_memory_size_atomic, or separate lock
380		 * to avoid taking dev_priv::cmdbuf_mutex in
381		 * the destroy path.
382		 */
383
384		mutex_lock(&dev_priv->cmdbuf_mutex);
385		dev_priv->used_memory_size -= res->guest_memory_size;
386		mutex_unlock(&dev_priv->cmdbuf_mutex);
387	}
388}
389
390/**
391 * vmw_legacy_srf_create - Create a device surface as part of the
392 * resource validation process.
393 *
394 * @res: Pointer to a struct vmw_surface.
395 *
396 * If the surface doesn't have a hw id.
397 *
398 * Returns -EBUSY if there wasn't sufficient device resources to
399 * complete the validation. Retry after freeing up resources.
400 *
401 * May return other errors if the kernel is out of guest resources.
402 */
403static int vmw_legacy_srf_create(struct vmw_resource *res)
404{
405	struct vmw_private *dev_priv = res->dev_priv;
406	struct vmw_surface *srf;
407	uint32_t submit_size;
408	uint8_t *cmd;
409	int ret;
410
411	if (likely(res->id != -1))
412		return 0;
413
414	srf = vmw_res_to_srf(res);
415	if (unlikely(dev_priv->used_memory_size + res->guest_memory_size >=
416		     dev_priv->memory_size))
417		return -EBUSY;
418
419	/*
420	 * Alloc id for the resource.
421	 */
422
423	ret = vmw_resource_alloc_id(res);
424	if (unlikely(ret != 0)) {
425		DRM_ERROR("Failed to allocate a surface id.\n");
426		goto out_no_id;
427	}
428
429	if (unlikely(res->id >= SVGA3D_HB_MAX_SURFACE_IDS)) {
430		ret = -EBUSY;
431		goto out_no_fifo;
432	}
433
434	/*
435	 * Encode surface define- commands.
436	 */
437
438	submit_size = vmw_surface_define_size(srf);
439	cmd = VMW_CMD_RESERVE(dev_priv, submit_size);
440	if (unlikely(!cmd)) {
441		ret = -ENOMEM;
442		goto out_no_fifo;
443	}
444
445	vmw_surface_define_encode(srf, cmd);
446	vmw_cmd_commit(dev_priv, submit_size);
447	vmw_fifo_resource_inc(dev_priv);
448
449	/*
450	 * Surface memory usage accounting.
451	 */
452
453	dev_priv->used_memory_size += res->guest_memory_size;
454	return 0;
455
456out_no_fifo:
457	vmw_resource_release_id(res);
458out_no_id:
459	return ret;
460}
461
462/**
463 * vmw_legacy_srf_dma - Copy backup data to or from a legacy surface.
464 *
465 * @res:            Pointer to a struct vmw_res embedded in a struct
466 *                  vmw_surface.
467 * @val_buf:        Pointer to a struct ttm_validate_buffer containing
468 *                  information about the backup buffer.
469 * @bind:           Boolean wether to DMA to the surface.
470 *
471 * Transfer backup data to or from a legacy surface as part of the
472 * validation process.
473 * May return other errors if the kernel is out of guest resources.
474 * The backup buffer will be fenced or idle upon successful completion,
475 * and if the surface needs persistent backup storage, the backup buffer
476 * will also be returned reserved iff @bind is true.
477 */
478static int vmw_legacy_srf_dma(struct vmw_resource *res,
479			      struct ttm_validate_buffer *val_buf,
480			      bool bind)
481{
482	SVGAGuestPtr ptr;
483	struct vmw_fence_obj *fence;
484	uint32_t submit_size;
485	struct vmw_surface *srf = vmw_res_to_srf(res);
486	uint8_t *cmd;
487	struct vmw_private *dev_priv = res->dev_priv;
488
489	BUG_ON(!val_buf->bo);
490	submit_size = vmw_surface_dma_size(srf);
491	cmd = VMW_CMD_RESERVE(dev_priv, submit_size);
492	if (unlikely(!cmd))
493		return -ENOMEM;
494
495	vmw_bo_get_guest_ptr(val_buf->bo, &ptr);
496	vmw_surface_dma_encode(srf, cmd, &ptr, bind);
497
498	vmw_cmd_commit(dev_priv, submit_size);
499
500	/*
501	 * Create a fence object and fence the backup buffer.
502	 */
503
504	(void) vmw_execbuf_fence_commands(NULL, dev_priv,
505					  &fence, NULL);
506
507	vmw_bo_fence_single(val_buf->bo, fence);
508
509	if (likely(fence != NULL))
510		vmw_fence_obj_unreference(&fence);
511
512	return 0;
513}
514
515/**
516 * vmw_legacy_srf_bind - Perform a legacy surface bind as part of the
517 *                       surface validation process.
518 *
519 * @res:            Pointer to a struct vmw_res embedded in a struct
520 *                  vmw_surface.
521 * @val_buf:        Pointer to a struct ttm_validate_buffer containing
522 *                  information about the backup buffer.
523 *
524 * This function will copy backup data to the surface if the
525 * backup buffer is dirty.
526 */
527static int vmw_legacy_srf_bind(struct vmw_resource *res,
528			       struct ttm_validate_buffer *val_buf)
529{
530	if (!res->guest_memory_dirty)
531		return 0;
532
533	return vmw_legacy_srf_dma(res, val_buf, true);
534}
535
536
537/**
538 * vmw_legacy_srf_unbind - Perform a legacy surface unbind as part of the
539 *                         surface eviction process.
540 *
541 * @res:            Pointer to a struct vmw_res embedded in a struct
542 *                  vmw_surface.
543 * @readback:       Readback - only true if dirty
544 * @val_buf:        Pointer to a struct ttm_validate_buffer containing
545 *                  information about the backup buffer.
546 *
547 * This function will copy backup data from the surface.
548 */
549static int vmw_legacy_srf_unbind(struct vmw_resource *res,
550				 bool readback,
551				 struct ttm_validate_buffer *val_buf)
552{
553	if (unlikely(readback))
554		return vmw_legacy_srf_dma(res, val_buf, false);
555	return 0;
556}
557
558/**
559 * vmw_legacy_srf_destroy - Destroy a device surface as part of a
560 *                          resource eviction process.
561 *
562 * @res:            Pointer to a struct vmw_res embedded in a struct
563 *                  vmw_surface.
564 */
565static int vmw_legacy_srf_destroy(struct vmw_resource *res)
566{
567	struct vmw_private *dev_priv = res->dev_priv;
568	uint32_t submit_size;
569	uint8_t *cmd;
570
571	BUG_ON(res->id == -1);
572
573	/*
574	 * Encode the dma- and surface destroy commands.
575	 */
576
577	submit_size = vmw_surface_destroy_size();
578	cmd = VMW_CMD_RESERVE(dev_priv, submit_size);
579	if (unlikely(!cmd))
580		return -ENOMEM;
581
582	vmw_surface_destroy_encode(res->id, cmd);
583	vmw_cmd_commit(dev_priv, submit_size);
584
585	/*
586	 * Surface memory usage accounting.
587	 */
588
589	dev_priv->used_memory_size -= res->guest_memory_size;
590
591	/*
592	 * Release the surface ID.
593	 */
594
595	vmw_resource_release_id(res);
596	vmw_fifo_resource_dec(dev_priv);
597
598	return 0;
599}
600
601
602/**
603 * vmw_surface_init - initialize a struct vmw_surface
604 *
605 * @dev_priv:       Pointer to a device private struct.
606 * @srf:            Pointer to the struct vmw_surface to initialize.
607 * @res_free:       Pointer to a resource destructor used to free
608 *                  the object.
609 */
610static int vmw_surface_init(struct vmw_private *dev_priv,
611			    struct vmw_surface *srf,
612			    void (*res_free) (struct vmw_resource *res))
613{
614	int ret;
615	struct vmw_resource *res = &srf->res;
616
617	BUG_ON(!res_free);
618	ret = vmw_resource_init(dev_priv, res, true, res_free,
619				(dev_priv->has_mob) ? &vmw_gb_surface_func :
620				&vmw_legacy_surface_func);
621
622	if (unlikely(ret != 0)) {
623		res_free(res);
624		return ret;
625	}
626
627	/*
628	 * The surface won't be visible to hardware until a
629	 * surface validate.
630	 */
631
632	INIT_LIST_HEAD(&srf->view_list);
633	res->hw_destroy = vmw_hw_surface_destroy;
634	return ret;
635}
636
637/**
638 * vmw_user_surface_base_to_res - TTM base object to resource converter for
639 *                                user visible surfaces
640 *
641 * @base:           Pointer to a TTM base object
642 *
643 * Returns the struct vmw_resource embedded in a struct vmw_surface
644 * for the user-visible object identified by the TTM base object @base.
645 */
646static struct vmw_resource *
647vmw_user_surface_base_to_res(struct ttm_base_object *base)
648{
649	return &(container_of(base, struct vmw_user_surface,
650			      prime.base)->srf.res);
651}
652
653/**
654 * vmw_user_surface_free - User visible surface resource destructor
655 *
656 * @res:            A struct vmw_resource embedded in a struct vmw_surface.
657 */
658static void vmw_user_surface_free(struct vmw_resource *res)
659{
660	struct vmw_surface *srf = vmw_res_to_srf(res);
661	struct vmw_user_surface *user_srf =
662	    container_of(srf, struct vmw_user_surface, srf);
663
664	WARN_ON_ONCE(res->dirty);
665	if (user_srf->master)
666		drm_master_put(&user_srf->master);
667	kfree(srf->offsets);
668	kfree(srf->metadata.sizes);
669	kfree(srf->snooper.image);
670	ttm_prime_object_kfree(user_srf, prime);
671}
672
673/**
674 * vmw_user_surface_base_release - User visible surface TTM base object destructor
675 *
676 * @p_base:         Pointer to a pointer to a TTM base object
677 *                  embedded in a struct vmw_user_surface.
678 *
679 * Drops the base object's reference on its resource, and the
680 * pointer pointed to by *p_base is set to NULL.
681 */
682static void vmw_user_surface_base_release(struct ttm_base_object **p_base)
683{
684	struct ttm_base_object *base = *p_base;
685	struct vmw_user_surface *user_srf =
686	    container_of(base, struct vmw_user_surface, prime.base);
687	struct vmw_resource *res = &user_srf->srf.res;
688
689	*p_base = NULL;
690	vmw_resource_unreference(&res);
691}
692
693/**
694 * vmw_surface_destroy_ioctl - Ioctl function implementing
695 *                                  the user surface destroy functionality.
696 *
697 * @dev:            Pointer to a struct drm_device.
698 * @data:           Pointer to data copied from / to user-space.
699 * @file_priv:      Pointer to a drm file private structure.
700 */
701int vmw_surface_destroy_ioctl(struct drm_device *dev, void *data,
702			      struct drm_file *file_priv)
703{
704	struct drm_vmw_surface_arg *arg = (struct drm_vmw_surface_arg *)data;
705	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
706
707	return ttm_ref_object_base_unref(tfile, arg->sid);
708}
709
710/**
711 * vmw_surface_define_ioctl - Ioctl function implementing
712 *                                  the user surface define functionality.
713 *
714 * @dev:            Pointer to a struct drm_device.
715 * @data:           Pointer to data copied from / to user-space.
716 * @file_priv:      Pointer to a drm file private structure.
717 */
718int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
719			     struct drm_file *file_priv)
720{
721	struct vmw_private *dev_priv = vmw_priv(dev);
722	struct vmw_user_surface *user_srf;
723	struct vmw_surface *srf;
724	struct vmw_surface_metadata *metadata;
725	struct vmw_resource *res;
726	struct vmw_resource *tmp;
727	union drm_vmw_surface_create_arg *arg =
728	    (union drm_vmw_surface_create_arg *)data;
729	struct drm_vmw_surface_create_req *req = &arg->req;
730	struct drm_vmw_surface_arg *rep = &arg->rep;
731	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
732	int ret;
733	int i, j;
734	uint32_t cur_bo_offset;
735	struct drm_vmw_size *cur_size;
736	struct vmw_surface_offset *cur_offset;
737	uint32_t num_sizes;
738	const SVGA3dSurfaceDesc *desc;
739
740	num_sizes = 0;
741	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) {
742		if (req->mip_levels[i] > DRM_VMW_MAX_MIP_LEVELS)
743			return -EINVAL;
744		num_sizes += req->mip_levels[i];
745	}
746
747	if (num_sizes > DRM_VMW_MAX_SURFACE_FACES * DRM_VMW_MAX_MIP_LEVELS ||
748	    num_sizes == 0)
749		return -EINVAL;
750
751	desc = vmw_surface_get_desc(req->format);
752	if (unlikely(desc->blockDesc == SVGA3DBLOCKDESC_NONE)) {
753		VMW_DEBUG_USER("Invalid format %d for surface creation.\n",
754			       req->format);
755		return -EINVAL;
756	}
757
758	user_srf = kzalloc(sizeof(*user_srf), GFP_KERNEL);
759	if (unlikely(!user_srf)) {
760		ret = -ENOMEM;
761		goto out_unlock;
762	}
763
764	srf = &user_srf->srf;
765	metadata = &srf->metadata;
766	res = &srf->res;
767
768	/* Driver internally stores as 64-bit flags */
769	metadata->flags = (SVGA3dSurfaceAllFlags)req->flags;
770	metadata->format = req->format;
771	metadata->scanout = req->scanout;
772
773	memcpy(metadata->mip_levels, req->mip_levels,
774	       sizeof(metadata->mip_levels));
775	metadata->num_sizes = num_sizes;
776	metadata->sizes =
777		memdup_array_user((struct drm_vmw_size __user *)(unsigned long)
778			    req->size_addr,
779			    metadata->num_sizes, sizeof(*metadata->sizes));
780	if (IS_ERR(metadata->sizes)) {
781		ret = PTR_ERR(metadata->sizes);
782		goto out_no_sizes;
783	}
784	srf->offsets = kmalloc_array(metadata->num_sizes, sizeof(*srf->offsets),
785				     GFP_KERNEL);
786	if (unlikely(!srf->offsets)) {
787		ret = -ENOMEM;
788		goto out_no_offsets;
789	}
790
791	metadata->base_size = *srf->metadata.sizes;
792	metadata->autogen_filter = SVGA3D_TEX_FILTER_NONE;
793	metadata->multisample_count = 0;
794	metadata->multisample_pattern = SVGA3D_MS_PATTERN_NONE;
795	metadata->quality_level = SVGA3D_MS_QUALITY_NONE;
796
797	cur_bo_offset = 0;
798	cur_offset = srf->offsets;
799	cur_size = metadata->sizes;
800
801	for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) {
802		for (j = 0; j < metadata->mip_levels[i]; ++j) {
803			uint32_t stride = vmw_surface_calculate_pitch(
804						  desc, cur_size);
805
806			cur_offset->face = i;
807			cur_offset->mip = j;
808			cur_offset->bo_offset = cur_bo_offset;
809			cur_bo_offset += vmw_surface_get_image_buffer_size
810				(desc, cur_size, stride);
811			++cur_offset;
812			++cur_size;
813		}
814	}
815	res->guest_memory_size = cur_bo_offset;
816	if (metadata->scanout &&
817	    metadata->num_sizes == 1 &&
818	    metadata->sizes[0].width == VMW_CURSOR_SNOOP_WIDTH &&
819	    metadata->sizes[0].height == VMW_CURSOR_SNOOP_HEIGHT &&
820	    metadata->format == VMW_CURSOR_SNOOP_FORMAT) {
821		const struct SVGA3dSurfaceDesc *desc =
822			vmw_surface_get_desc(VMW_CURSOR_SNOOP_FORMAT);
823		const u32 cursor_size_bytes = VMW_CURSOR_SNOOP_WIDTH *
824					      VMW_CURSOR_SNOOP_HEIGHT *
825					      desc->pitchBytesPerBlock;
826		srf->snooper.image = kzalloc(cursor_size_bytes, GFP_KERNEL);
827		if (!srf->snooper.image) {
828			DRM_ERROR("Failed to allocate cursor_image\n");
829			ret = -ENOMEM;
830			goto out_no_copy;
831		}
832	} else {
833		srf->snooper.image = NULL;
834	}
835
836	user_srf->prime.base.shareable = false;
837	user_srf->prime.base.tfile = NULL;
838	if (drm_is_primary_client(file_priv))
839		user_srf->master = drm_file_get_master(file_priv);
840
841	/**
842	 * From this point, the generic resource management functions
843	 * destroy the object on failure.
844	 */
845
846	ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free);
847	if (unlikely(ret != 0))
848		goto out_unlock;
849
850	/*
851	 * A gb-aware client referencing a shared surface will
852	 * expect a backup buffer to be present.
853	 */
854	if (dev_priv->has_mob && req->shareable) {
855		struct vmw_bo_params params = {
856			.domain = VMW_BO_DOMAIN_SYS,
857			.busy_domain = VMW_BO_DOMAIN_SYS,
858			.bo_type = ttm_bo_type_device,
859			.size = res->guest_memory_size,
860			.pin = false
861		};
862
863		ret = vmw_gem_object_create(dev_priv,
864					    &params,
865					    &res->guest_memory_bo);
866		if (unlikely(ret != 0)) {
867			vmw_resource_unreference(&res);
868			goto out_unlock;
869		}
870	}
871
872	tmp = vmw_resource_reference(&srf->res);
873	ret = ttm_prime_object_init(tfile, res->guest_memory_size, &user_srf->prime,
874				    req->shareable, VMW_RES_SURFACE,
875				    &vmw_user_surface_base_release);
876
877	if (unlikely(ret != 0)) {
878		vmw_resource_unreference(&tmp);
879		vmw_resource_unreference(&res);
880		goto out_unlock;
881	}
882
883	rep->sid = user_srf->prime.base.handle;
884	vmw_resource_unreference(&res);
885
886	return 0;
887out_no_copy:
888	kfree(srf->offsets);
889out_no_offsets:
890	kfree(metadata->sizes);
891out_no_sizes:
892	ttm_prime_object_kfree(user_srf, prime);
893out_unlock:
894	return ret;
895}
896
897
898static int
899vmw_surface_handle_reference(struct vmw_private *dev_priv,
900			     struct drm_file *file_priv,
901			     uint32_t u_handle,
902			     enum drm_vmw_handle_type handle_type,
903			     struct ttm_base_object **base_p)
904{
905	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
906	struct vmw_user_surface *user_srf;
907	uint32_t handle;
908	struct ttm_base_object *base;
909	int ret;
910
911	if (handle_type == DRM_VMW_HANDLE_PRIME) {
912		ret = ttm_prime_fd_to_handle(tfile, u_handle, &handle);
913		if (unlikely(ret != 0))
914			return ret;
915	} else {
916		handle = u_handle;
917	}
918
919	ret = -EINVAL;
920	base = ttm_base_object_lookup_for_ref(dev_priv->tdev, handle);
921	if (unlikely(!base)) {
922		VMW_DEBUG_USER("Could not find surface to reference.\n");
923		goto out_no_lookup;
924	}
925
926	if (unlikely(ttm_base_object_type(base) != VMW_RES_SURFACE)) {
927		VMW_DEBUG_USER("Referenced object is not a surface.\n");
928		goto out_bad_resource;
929	}
930	if (handle_type != DRM_VMW_HANDLE_PRIME) {
931		bool require_exist = false;
932
933		user_srf = container_of(base, struct vmw_user_surface,
934					prime.base);
935
936		/* Error out if we are unauthenticated primary */
937		if (drm_is_primary_client(file_priv) &&
938		    !file_priv->authenticated) {
939			ret = -EACCES;
940			goto out_bad_resource;
941		}
942
943		/*
944		 * Make sure the surface creator has the same
945		 * authenticating master, or is already registered with us.
946		 */
947		if (drm_is_primary_client(file_priv) &&
948		    user_srf->master != file_priv->master)
949			require_exist = true;
950
951		if (unlikely(drm_is_render_client(file_priv)))
952			require_exist = true;
953
954		ret = ttm_ref_object_add(tfile, base, NULL, require_exist);
955		if (unlikely(ret != 0)) {
956			DRM_ERROR("Could not add a reference to a surface.\n");
957			goto out_bad_resource;
958		}
959	}
960
961	*base_p = base;
962	return 0;
963
964out_bad_resource:
965	ttm_base_object_unref(&base);
966out_no_lookup:
967	if (handle_type == DRM_VMW_HANDLE_PRIME)
968		(void) ttm_ref_object_base_unref(tfile, handle);
969
970	return ret;
971}
972
973/**
974 * vmw_surface_reference_ioctl - Ioctl function implementing
975 *                                  the user surface reference functionality.
976 *
977 * @dev:            Pointer to a struct drm_device.
978 * @data:           Pointer to data copied from / to user-space.
979 * @file_priv:      Pointer to a drm file private structure.
980 */
981int vmw_surface_reference_ioctl(struct drm_device *dev, void *data,
982				struct drm_file *file_priv)
983{
984	struct vmw_private *dev_priv = vmw_priv(dev);
985	union drm_vmw_surface_reference_arg *arg =
986	    (union drm_vmw_surface_reference_arg *)data;
987	struct drm_vmw_surface_arg *req = &arg->req;
988	struct drm_vmw_surface_create_req *rep = &arg->rep;
989	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
990	struct vmw_surface *srf;
991	struct vmw_user_surface *user_srf;
992	struct drm_vmw_size __user *user_sizes;
993	struct ttm_base_object *base;
994	int ret;
995
996	ret = vmw_surface_handle_reference(dev_priv, file_priv, req->sid,
997					   req->handle_type, &base);
998	if (unlikely(ret != 0))
999		return ret;
1000
1001	user_srf = container_of(base, struct vmw_user_surface, prime.base);
1002	srf = &user_srf->srf;
1003
1004	/* Downcast of flags when sending back to user space */
1005	rep->flags = (uint32_t)srf->metadata.flags;
1006	rep->format = srf->metadata.format;
1007	memcpy(rep->mip_levels, srf->metadata.mip_levels,
1008	       sizeof(srf->metadata.mip_levels));
1009	user_sizes = (struct drm_vmw_size __user *)(unsigned long)
1010	    rep->size_addr;
1011
1012	if (user_sizes)
1013		ret = copy_to_user(user_sizes, &srf->metadata.base_size,
1014				   sizeof(srf->metadata.base_size));
1015	if (unlikely(ret != 0)) {
1016		VMW_DEBUG_USER("copy_to_user failed %p %u\n", user_sizes,
1017			       srf->metadata.num_sizes);
1018		ttm_ref_object_base_unref(tfile, base->handle);
1019		ret = -EFAULT;
1020	}
1021
1022	ttm_base_object_unref(&base);
1023
1024	return ret;
1025}
1026
1027/**
1028 * vmw_gb_surface_create - Encode a surface_define command.
1029 *
1030 * @res:        Pointer to a struct vmw_resource embedded in a struct
1031 *              vmw_surface.
1032 */
1033static int vmw_gb_surface_create(struct vmw_resource *res)
1034{
1035	struct vmw_private *dev_priv = res->dev_priv;
1036	struct vmw_surface *srf = vmw_res_to_srf(res);
1037	struct vmw_surface_metadata *metadata = &srf->metadata;
1038	uint32_t cmd_len, cmd_id, submit_len;
1039	int ret;
1040	struct {
1041		SVGA3dCmdHeader header;
1042		SVGA3dCmdDefineGBSurface body;
1043	} *cmd;
1044	struct {
1045		SVGA3dCmdHeader header;
1046		SVGA3dCmdDefineGBSurface_v2 body;
1047	} *cmd2;
1048	struct {
1049		SVGA3dCmdHeader header;
1050		SVGA3dCmdDefineGBSurface_v3 body;
1051	} *cmd3;
1052	struct {
1053		SVGA3dCmdHeader header;
1054		SVGA3dCmdDefineGBSurface_v4 body;
1055	} *cmd4;
1056
1057	if (likely(res->id != -1))
1058		return 0;
1059
1060	vmw_fifo_resource_inc(dev_priv);
1061	ret = vmw_resource_alloc_id(res);
1062	if (unlikely(ret != 0)) {
1063		DRM_ERROR("Failed to allocate a surface id.\n");
1064		goto out_no_id;
1065	}
1066
1067	if (unlikely(res->id >= VMWGFX_NUM_GB_SURFACE)) {
1068		ret = -EBUSY;
1069		goto out_no_fifo;
1070	}
1071
1072	if (has_sm5_context(dev_priv) && metadata->array_size > 0) {
1073		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE_V4;
1074		cmd_len = sizeof(cmd4->body);
1075		submit_len = sizeof(*cmd4);
1076	} else if (has_sm4_1_context(dev_priv) && metadata->array_size > 0) {
1077		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE_V3;
1078		cmd_len = sizeof(cmd3->body);
1079		submit_len = sizeof(*cmd3);
1080	} else if (metadata->array_size > 0) {
1081		/* VMW_SM_4 support verified at creation time. */
1082		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE_V2;
1083		cmd_len = sizeof(cmd2->body);
1084		submit_len = sizeof(*cmd2);
1085	} else {
1086		cmd_id = SVGA_3D_CMD_DEFINE_GB_SURFACE;
1087		cmd_len = sizeof(cmd->body);
1088		submit_len = sizeof(*cmd);
1089	}
1090
1091	cmd = VMW_CMD_RESERVE(dev_priv, submit_len);
1092	cmd2 = (typeof(cmd2))cmd;
1093	cmd3 = (typeof(cmd3))cmd;
1094	cmd4 = (typeof(cmd4))cmd;
1095	if (unlikely(!cmd)) {
1096		ret = -ENOMEM;
1097		goto out_no_fifo;
1098	}
1099
1100	if (has_sm5_context(dev_priv) && metadata->array_size > 0) {
1101		cmd4->header.id = cmd_id;
1102		cmd4->header.size = cmd_len;
1103		cmd4->body.sid = srf->res.id;
1104		cmd4->body.surfaceFlags = metadata->flags;
1105		cmd4->body.format = metadata->format;
1106		cmd4->body.numMipLevels = metadata->mip_levels[0];
1107		cmd4->body.multisampleCount = metadata->multisample_count;
1108		cmd4->body.multisamplePattern = metadata->multisample_pattern;
1109		cmd4->body.qualityLevel = metadata->quality_level;
1110		cmd4->body.autogenFilter = metadata->autogen_filter;
1111		cmd4->body.size.width = metadata->base_size.width;
1112		cmd4->body.size.height = metadata->base_size.height;
1113		cmd4->body.size.depth = metadata->base_size.depth;
1114		cmd4->body.arraySize = metadata->array_size;
1115		cmd4->body.bufferByteStride = metadata->buffer_byte_stride;
1116	} else if (has_sm4_1_context(dev_priv) && metadata->array_size > 0) {
1117		cmd3->header.id = cmd_id;
1118		cmd3->header.size = cmd_len;
1119		cmd3->body.sid = srf->res.id;
1120		cmd3->body.surfaceFlags = metadata->flags;
1121		cmd3->body.format = metadata->format;
1122		cmd3->body.numMipLevels = metadata->mip_levels[0];
1123		cmd3->body.multisampleCount = metadata->multisample_count;
1124		cmd3->body.multisamplePattern = metadata->multisample_pattern;
1125		cmd3->body.qualityLevel = metadata->quality_level;
1126		cmd3->body.autogenFilter = metadata->autogen_filter;
1127		cmd3->body.size.width = metadata->base_size.width;
1128		cmd3->body.size.height = metadata->base_size.height;
1129		cmd3->body.size.depth = metadata->base_size.depth;
1130		cmd3->body.arraySize = metadata->array_size;
1131	} else if (metadata->array_size > 0) {
1132		cmd2->header.id = cmd_id;
1133		cmd2->header.size = cmd_len;
1134		cmd2->body.sid = srf->res.id;
1135		cmd2->body.surfaceFlags = metadata->flags;
1136		cmd2->body.format = metadata->format;
1137		cmd2->body.numMipLevels = metadata->mip_levels[0];
1138		cmd2->body.multisampleCount = metadata->multisample_count;
1139		cmd2->body.autogenFilter = metadata->autogen_filter;
1140		cmd2->body.size.width = metadata->base_size.width;
1141		cmd2->body.size.height = metadata->base_size.height;
1142		cmd2->body.size.depth = metadata->base_size.depth;
1143		cmd2->body.arraySize = metadata->array_size;
1144	} else {
1145		cmd->header.id = cmd_id;
1146		cmd->header.size = cmd_len;
1147		cmd->body.sid = srf->res.id;
1148		cmd->body.surfaceFlags = metadata->flags;
1149		cmd->body.format = metadata->format;
1150		cmd->body.numMipLevels = metadata->mip_levels[0];
1151		cmd->body.multisampleCount = metadata->multisample_count;
1152		cmd->body.autogenFilter = metadata->autogen_filter;
1153		cmd->body.size.width = metadata->base_size.width;
1154		cmd->body.size.height = metadata->base_size.height;
1155		cmd->body.size.depth = metadata->base_size.depth;
1156	}
1157
1158	vmw_cmd_commit(dev_priv, submit_len);
1159
1160	return 0;
1161
1162out_no_fifo:
1163	vmw_resource_release_id(res);
1164out_no_id:
1165	vmw_fifo_resource_dec(dev_priv);
1166	return ret;
1167}
1168
1169
1170static int vmw_gb_surface_bind(struct vmw_resource *res,
1171			       struct ttm_validate_buffer *val_buf)
1172{
1173	struct vmw_private *dev_priv = res->dev_priv;
1174	struct {
1175		SVGA3dCmdHeader header;
1176		SVGA3dCmdBindGBSurface body;
1177	} *cmd1;
1178	struct {
1179		SVGA3dCmdHeader header;
1180		SVGA3dCmdUpdateGBSurface body;
1181	} *cmd2;
1182	uint32_t submit_size;
1183	struct ttm_buffer_object *bo = val_buf->bo;
1184
1185	BUG_ON(bo->resource->mem_type != VMW_PL_MOB);
1186
1187	submit_size = sizeof(*cmd1) + (res->guest_memory_dirty ? sizeof(*cmd2) : 0);
1188
1189	cmd1 = VMW_CMD_RESERVE(dev_priv, submit_size);
1190	if (unlikely(!cmd1))
1191		return -ENOMEM;
1192
1193	cmd1->header.id = SVGA_3D_CMD_BIND_GB_SURFACE;
1194	cmd1->header.size = sizeof(cmd1->body);
1195	cmd1->body.sid = res->id;
1196	cmd1->body.mobid = bo->resource->start;
1197	if (res->guest_memory_dirty) {
1198		cmd2 = (void *) &cmd1[1];
1199		cmd2->header.id = SVGA_3D_CMD_UPDATE_GB_SURFACE;
1200		cmd2->header.size = sizeof(cmd2->body);
1201		cmd2->body.sid = res->id;
1202	}
1203	vmw_cmd_commit(dev_priv, submit_size);
1204
1205	if (res->guest_memory_bo->dirty && res->guest_memory_dirty) {
1206		/* We've just made a full upload. Cear dirty regions. */
1207		vmw_bo_dirty_clear_res(res);
1208	}
1209
1210	res->guest_memory_dirty = false;
1211
1212	return 0;
1213}
1214
1215static int vmw_gb_surface_unbind(struct vmw_resource *res,
1216				 bool readback,
1217				 struct ttm_validate_buffer *val_buf)
1218{
1219	struct vmw_private *dev_priv = res->dev_priv;
1220	struct ttm_buffer_object *bo = val_buf->bo;
1221	struct vmw_fence_obj *fence;
1222
1223	struct {
1224		SVGA3dCmdHeader header;
1225		SVGA3dCmdReadbackGBSurface body;
1226	} *cmd1;
1227	struct {
1228		SVGA3dCmdHeader header;
1229		SVGA3dCmdInvalidateGBSurface body;
1230	} *cmd2;
1231	struct {
1232		SVGA3dCmdHeader header;
1233		SVGA3dCmdBindGBSurface body;
1234	} *cmd3;
1235	uint32_t submit_size;
1236	uint8_t *cmd;
1237
1238
1239	BUG_ON(bo->resource->mem_type != VMW_PL_MOB);
1240
1241	submit_size = sizeof(*cmd3) + (readback ? sizeof(*cmd1) : sizeof(*cmd2));
1242	cmd = VMW_CMD_RESERVE(dev_priv, submit_size);
1243	if (unlikely(!cmd))
1244		return -ENOMEM;
1245
1246	if (readback) {
1247		cmd1 = (void *) cmd;
1248		cmd1->header.id = SVGA_3D_CMD_READBACK_GB_SURFACE;
1249		cmd1->header.size = sizeof(cmd1->body);
1250		cmd1->body.sid = res->id;
1251		cmd3 = (void *) &cmd1[1];
1252	} else {
1253		cmd2 = (void *) cmd;
1254		cmd2->header.id = SVGA_3D_CMD_INVALIDATE_GB_SURFACE;
1255		cmd2->header.size = sizeof(cmd2->body);
1256		cmd2->body.sid = res->id;
1257		cmd3 = (void *) &cmd2[1];
1258	}
1259
1260	cmd3->header.id = SVGA_3D_CMD_BIND_GB_SURFACE;
1261	cmd3->header.size = sizeof(cmd3->body);
1262	cmd3->body.sid = res->id;
1263	cmd3->body.mobid = SVGA3D_INVALID_ID;
1264
1265	vmw_cmd_commit(dev_priv, submit_size);
1266
1267	/*
1268	 * Create a fence object and fence the backup buffer.
1269	 */
1270
1271	(void) vmw_execbuf_fence_commands(NULL, dev_priv,
1272					  &fence, NULL);
1273
1274	vmw_bo_fence_single(val_buf->bo, fence);
1275
1276	if (likely(fence != NULL))
1277		vmw_fence_obj_unreference(&fence);
1278
1279	return 0;
1280}
1281
1282static int vmw_gb_surface_destroy(struct vmw_resource *res)
1283{
1284	struct vmw_private *dev_priv = res->dev_priv;
1285	struct vmw_surface *srf = vmw_res_to_srf(res);
1286	struct {
1287		SVGA3dCmdHeader header;
1288		SVGA3dCmdDestroyGBSurface body;
1289	} *cmd;
1290
1291	if (likely(res->id == -1))
1292		return 0;
1293
1294	mutex_lock(&dev_priv->binding_mutex);
1295	vmw_view_surface_list_destroy(dev_priv, &srf->view_list);
1296	vmw_binding_res_list_scrub(&res->binding_head);
1297
1298	cmd = VMW_CMD_RESERVE(dev_priv, sizeof(*cmd));
1299	if (unlikely(!cmd)) {
1300		mutex_unlock(&dev_priv->binding_mutex);
1301		return -ENOMEM;
1302	}
1303
1304	cmd->header.id = SVGA_3D_CMD_DESTROY_GB_SURFACE;
1305	cmd->header.size = sizeof(cmd->body);
1306	cmd->body.sid = res->id;
1307	vmw_cmd_commit(dev_priv, sizeof(*cmd));
1308	mutex_unlock(&dev_priv->binding_mutex);
1309	vmw_resource_release_id(res);
1310	vmw_fifo_resource_dec(dev_priv);
1311
1312	return 0;
1313}
1314
1315/**
1316 * vmw_gb_surface_define_ioctl - Ioctl function implementing
1317 * the user surface define functionality.
1318 *
1319 * @dev: Pointer to a struct drm_device.
1320 * @data: Pointer to data copied from / to user-space.
1321 * @file_priv: Pointer to a drm file private structure.
1322 */
1323int vmw_gb_surface_define_ioctl(struct drm_device *dev, void *data,
1324				struct drm_file *file_priv)
1325{
1326	union drm_vmw_gb_surface_create_arg *arg =
1327	    (union drm_vmw_gb_surface_create_arg *)data;
1328	struct drm_vmw_gb_surface_create_rep *rep = &arg->rep;
1329	struct drm_vmw_gb_surface_create_ext_req req_ext;
1330
1331	req_ext.base = arg->req;
1332	req_ext.version = drm_vmw_gb_surface_v1;
1333	req_ext.svga3d_flags_upper_32_bits = 0;
1334	req_ext.multisample_pattern = SVGA3D_MS_PATTERN_NONE;
1335	req_ext.quality_level = SVGA3D_MS_QUALITY_NONE;
1336	req_ext.buffer_byte_stride = 0;
1337	req_ext.must_be_zero = 0;
1338
1339	return vmw_gb_surface_define_internal(dev, &req_ext, rep, file_priv);
1340}
1341
1342/**
1343 * vmw_gb_surface_reference_ioctl - Ioctl function implementing
1344 * the user surface reference functionality.
1345 *
1346 * @dev: Pointer to a struct drm_device.
1347 * @data: Pointer to data copied from / to user-space.
1348 * @file_priv: Pointer to a drm file private structure.
1349 */
1350int vmw_gb_surface_reference_ioctl(struct drm_device *dev, void *data,
1351				   struct drm_file *file_priv)
1352{
1353	union drm_vmw_gb_surface_reference_arg *arg =
1354	    (union drm_vmw_gb_surface_reference_arg *)data;
1355	struct drm_vmw_surface_arg *req = &arg->req;
1356	struct drm_vmw_gb_surface_ref_rep *rep = &arg->rep;
1357	struct drm_vmw_gb_surface_ref_ext_rep rep_ext;
1358	int ret;
1359
1360	ret = vmw_gb_surface_reference_internal(dev, req, &rep_ext, file_priv);
1361
1362	if (unlikely(ret != 0))
1363		return ret;
1364
1365	rep->creq = rep_ext.creq.base;
1366	rep->crep = rep_ext.crep;
1367
1368	return ret;
1369}
1370
1371/**
1372 * vmw_gb_surface_define_ext_ioctl - Ioctl function implementing
1373 * the user surface define functionality.
1374 *
1375 * @dev: Pointer to a struct drm_device.
1376 * @data: Pointer to data copied from / to user-space.
1377 * @file_priv: Pointer to a drm file private structure.
1378 */
1379int vmw_gb_surface_define_ext_ioctl(struct drm_device *dev, void *data,
1380				struct drm_file *file_priv)
1381{
1382	union drm_vmw_gb_surface_create_ext_arg *arg =
1383	    (union drm_vmw_gb_surface_create_ext_arg *)data;
1384	struct drm_vmw_gb_surface_create_ext_req *req = &arg->req;
1385	struct drm_vmw_gb_surface_create_rep *rep = &arg->rep;
1386
1387	return vmw_gb_surface_define_internal(dev, req, rep, file_priv);
1388}
1389
1390/**
1391 * vmw_gb_surface_reference_ext_ioctl - Ioctl function implementing
1392 * the user surface reference functionality.
1393 *
1394 * @dev: Pointer to a struct drm_device.
1395 * @data: Pointer to data copied from / to user-space.
1396 * @file_priv: Pointer to a drm file private structure.
1397 */
1398int vmw_gb_surface_reference_ext_ioctl(struct drm_device *dev, void *data,
1399				   struct drm_file *file_priv)
1400{
1401	union drm_vmw_gb_surface_reference_ext_arg *arg =
1402	    (union drm_vmw_gb_surface_reference_ext_arg *)data;
1403	struct drm_vmw_surface_arg *req = &arg->req;
1404	struct drm_vmw_gb_surface_ref_ext_rep *rep = &arg->rep;
1405
1406	return vmw_gb_surface_reference_internal(dev, req, rep, file_priv);
1407}
1408
1409/**
1410 * vmw_gb_surface_define_internal - Ioctl function implementing
1411 * the user surface define functionality.
1412 *
1413 * @dev: Pointer to a struct drm_device.
1414 * @req: Request argument from user-space.
1415 * @rep: Response argument to user-space.
1416 * @file_priv: Pointer to a drm file private structure.
1417 */
1418static int
1419vmw_gb_surface_define_internal(struct drm_device *dev,
1420			       struct drm_vmw_gb_surface_create_ext_req *req,
1421			       struct drm_vmw_gb_surface_create_rep *rep,
1422			       struct drm_file *file_priv)
1423{
1424	struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
1425	struct vmw_private *dev_priv = vmw_priv(dev);
1426	struct vmw_user_surface *user_srf;
1427	struct vmw_surface_metadata metadata = {0};
1428	struct vmw_surface *srf;
1429	struct vmw_resource *res;
1430	struct vmw_resource *tmp;
1431	int ret = 0;
1432	uint32_t backup_handle = 0;
1433	SVGA3dSurfaceAllFlags svga3d_flags_64 =
1434		SVGA3D_FLAGS_64(req->svga3d_flags_upper_32_bits,
1435				req->base.svga3d_flags);
1436
1437	/* array_size must be null for non-GL3 host. */
1438	if (req->base.array_size > 0 && !has_sm4_context(dev_priv)) {
1439		VMW_DEBUG_USER("SM4 surface not supported.\n");
1440		return -EINVAL;
1441	}
1442
1443	if (!has_sm4_1_context(dev_priv)) {
1444		if (req->svga3d_flags_upper_32_bits != 0)
1445			ret = -EINVAL;
1446
1447		if (req->base.multisample_count != 0)
1448			ret = -EINVAL;
1449
1450		if (req->multisample_pattern != SVGA3D_MS_PATTERN_NONE)
1451			ret = -EINVAL;
1452
1453		if (req->quality_level != SVGA3D_MS_QUALITY_NONE)
1454			ret = -EINVAL;
1455
1456		if (ret) {
1457			VMW_DEBUG_USER("SM4.1 surface not supported.\n");
1458			return ret;
1459		}
1460	}
1461
1462	if (req->buffer_byte_stride > 0 && !has_sm5_context(dev_priv)) {
1463		VMW_DEBUG_USER("SM5 surface not supported.\n");
1464		return -EINVAL;
1465	}
1466
1467	if ((svga3d_flags_64 & SVGA3D_SURFACE_MULTISAMPLE) &&
1468	    req->base.multisample_count == 0) {
1469		VMW_DEBUG_USER("Invalid sample count.\n");
1470		return -EINVAL;
1471	}
1472
1473	if (req->base.mip_levels > DRM_VMW_MAX_MIP_LEVELS) {
1474		VMW_DEBUG_USER("Invalid mip level.\n");
1475		return -EINVAL;
1476	}
1477
1478	metadata.flags = svga3d_flags_64;
1479	metadata.format = req->base.format;
1480	metadata.mip_levels[0] = req->base.mip_levels;
1481	metadata.multisample_count = req->base.multisample_count;
1482	metadata.multisample_pattern = req->multisample_pattern;
1483	metadata.quality_level = req->quality_level;
1484	metadata.array_size = req->base.array_size;
1485	metadata.buffer_byte_stride = req->buffer_byte_stride;
1486	metadata.num_sizes = 1;
1487	metadata.base_size = req->base.base_size;
1488	metadata.scanout = req->base.drm_surface_flags &
1489		drm_vmw_surface_flag_scanout;
1490
1491	/* Define a surface based on the parameters. */
1492	ret = vmw_gb_surface_define(dev_priv, &metadata, &srf);
1493	if (ret != 0) {
1494		VMW_DEBUG_USER("Failed to define surface.\n");
1495		return ret;
1496	}
1497
1498	user_srf = container_of(srf, struct vmw_user_surface, srf);
1499	if (drm_is_primary_client(file_priv))
1500		user_srf->master = drm_file_get_master(file_priv);
1501
1502	res = &user_srf->srf.res;
1503
1504	if (req->base.buffer_handle != SVGA3D_INVALID_ID) {
1505		ret = vmw_user_bo_lookup(file_priv, req->base.buffer_handle,
1506					 &res->guest_memory_bo);
1507		if (ret == 0) {
1508			if (res->guest_memory_bo->tbo.base.size < res->guest_memory_size) {
1509				VMW_DEBUG_USER("Surface backup buffer too small.\n");
1510				vmw_user_bo_unref(&res->guest_memory_bo);
1511				ret = -EINVAL;
1512				goto out_unlock;
1513			} else {
1514				backup_handle = req->base.buffer_handle;
1515			}
1516		}
1517	} else if (req->base.drm_surface_flags &
1518		   (drm_vmw_surface_flag_create_buffer |
1519		    drm_vmw_surface_flag_coherent)) {
1520		ret = vmw_gem_object_create_with_handle(dev_priv, file_priv,
1521							res->guest_memory_size,
1522							&backup_handle,
1523							&res->guest_memory_bo);
1524	}
1525
1526	if (unlikely(ret != 0)) {
1527		vmw_resource_unreference(&res);
1528		goto out_unlock;
1529	}
1530
1531	if (req->base.drm_surface_flags & drm_vmw_surface_flag_coherent) {
1532		struct vmw_bo *backup = res->guest_memory_bo;
1533
1534		ttm_bo_reserve(&backup->tbo, false, false, NULL);
1535		if (!res->func->dirty_alloc)
1536			ret = -EINVAL;
1537		if (!ret)
1538			ret = vmw_bo_dirty_add(backup);
1539		if (!ret) {
1540			res->coherent = true;
1541			ret = res->func->dirty_alloc(res);
1542		}
1543		ttm_bo_unreserve(&backup->tbo);
1544		if (ret) {
1545			vmw_resource_unreference(&res);
1546			goto out_unlock;
1547		}
1548
1549	}
1550
1551	tmp = vmw_resource_reference(res);
1552	ret = ttm_prime_object_init(tfile, res->guest_memory_size, &user_srf->prime,
1553				    req->base.drm_surface_flags &
1554				    drm_vmw_surface_flag_shareable,
1555				    VMW_RES_SURFACE,
1556				    &vmw_user_surface_base_release);
1557
1558	if (unlikely(ret != 0)) {
1559		vmw_resource_unreference(&tmp);
1560		vmw_resource_unreference(&res);
1561		goto out_unlock;
1562	}
1563
1564	rep->handle      = user_srf->prime.base.handle;
1565	rep->backup_size = res->guest_memory_size;
1566	if (res->guest_memory_bo) {
1567		rep->buffer_map_handle =
1568			drm_vma_node_offset_addr(&res->guest_memory_bo->tbo.base.vma_node);
1569		rep->buffer_size = res->guest_memory_bo->tbo.base.size;
1570		rep->buffer_handle = backup_handle;
1571	} else {
1572		rep->buffer_map_handle = 0;
1573		rep->buffer_size = 0;
1574		rep->buffer_handle = SVGA3D_INVALID_ID;
1575	}
1576	vmw_resource_unreference(&res);
1577
1578out_unlock:
1579	return ret;
1580}
1581
1582/**
1583 * vmw_gb_surface_reference_internal - Ioctl function implementing
1584 * the user surface reference functionality.
1585 *
1586 * @dev: Pointer to a struct drm_device.
1587 * @req: Pointer to user-space request surface arg.
1588 * @rep: Pointer to response to user-space.
1589 * @file_priv: Pointer to a drm file private structure.
1590 */
1591static int
1592vmw_gb_surface_reference_internal(struct drm_device *dev,
1593				  struct drm_vmw_surface_arg *req,
1594				  struct drm_vmw_gb_surface_ref_ext_rep *rep,
1595				  struct drm_file *file_priv)
1596{
1597	struct vmw_private *dev_priv = vmw_priv(dev);
1598	struct vmw_surface *srf;
1599	struct vmw_user_surface *user_srf;
1600	struct vmw_surface_metadata *metadata;
1601	struct ttm_base_object *base;
1602	u32 backup_handle;
1603	int ret;
1604
1605	ret = vmw_surface_handle_reference(dev_priv, file_priv, req->sid,
1606					   req->handle_type, &base);
1607	if (unlikely(ret != 0))
1608		return ret;
1609
1610	user_srf = container_of(base, struct vmw_user_surface, prime.base);
1611	srf = &user_srf->srf;
1612	if (!srf->res.guest_memory_bo) {
1613		DRM_ERROR("Shared GB surface is missing a backup buffer.\n");
1614		goto out_bad_resource;
1615	}
1616	metadata = &srf->metadata;
1617
1618	mutex_lock(&dev_priv->cmdbuf_mutex); /* Protect res->backup */
1619	ret = drm_gem_handle_create(file_priv, &srf->res.guest_memory_bo->tbo.base,
1620				    &backup_handle);
1621	mutex_unlock(&dev_priv->cmdbuf_mutex);
1622	if (ret != 0) {
1623		drm_err(dev, "Wasn't able to create a backing handle for surface sid = %u.\n",
1624			req->sid);
1625		goto out_bad_resource;
1626	}
1627
1628	rep->creq.base.svga3d_flags = SVGA3D_FLAGS_LOWER_32(metadata->flags);
1629	rep->creq.base.format = metadata->format;
1630	rep->creq.base.mip_levels = metadata->mip_levels[0];
1631	rep->creq.base.drm_surface_flags = 0;
1632	rep->creq.base.multisample_count = metadata->multisample_count;
1633	rep->creq.base.autogen_filter = metadata->autogen_filter;
1634	rep->creq.base.array_size = metadata->array_size;
1635	rep->creq.base.buffer_handle = backup_handle;
1636	rep->creq.base.base_size = metadata->base_size;
1637	rep->crep.handle = user_srf->prime.base.handle;
1638	rep->crep.backup_size = srf->res.guest_memory_size;
1639	rep->crep.buffer_handle = backup_handle;
1640	rep->crep.buffer_map_handle =
1641		drm_vma_node_offset_addr(&srf->res.guest_memory_bo->tbo.base.vma_node);
1642	rep->crep.buffer_size = srf->res.guest_memory_bo->tbo.base.size;
1643
1644	rep->creq.version = drm_vmw_gb_surface_v1;
1645	rep->creq.svga3d_flags_upper_32_bits =
1646		SVGA3D_FLAGS_UPPER_32(metadata->flags);
1647	rep->creq.multisample_pattern = metadata->multisample_pattern;
1648	rep->creq.quality_level = metadata->quality_level;
1649	rep->creq.must_be_zero = 0;
1650
1651out_bad_resource:
1652	ttm_base_object_unref(&base);
1653
1654	return ret;
1655}
1656
1657/**
1658 * vmw_subres_dirty_add - Add a dirty region to a subresource
1659 * @dirty: The surfaces's dirty tracker.
1660 * @loc_start: The location corresponding to the start of the region.
1661 * @loc_end: The location corresponding to the end of the region.
1662 *
1663 * As we are assuming that @loc_start and @loc_end represent a sequential
1664 * range of backing store memory, if the region spans multiple lines then
1665 * regardless of the x coordinate, the full lines are dirtied.
1666 * Correspondingly if the region spans multiple z slices, then full rather
1667 * than partial z slices are dirtied.
1668 */
1669static void vmw_subres_dirty_add(struct vmw_surface_dirty *dirty,
1670				 const struct vmw_surface_loc *loc_start,
1671				 const struct vmw_surface_loc *loc_end)
1672{
1673	const struct vmw_surface_cache *cache = &dirty->cache;
1674	SVGA3dBox *box = &dirty->boxes[loc_start->sub_resource];
1675	u32 mip = loc_start->sub_resource % cache->num_mip_levels;
1676	const struct drm_vmw_size *size = &cache->mip[mip].size;
1677	u32 box_c2 = box->z + box->d;
1678
1679	if (WARN_ON(loc_start->sub_resource >= dirty->num_subres))
1680		return;
1681
1682	if (box->d == 0 || box->z > loc_start->z)
1683		box->z = loc_start->z;
1684	if (box_c2 < loc_end->z)
1685		box->d = loc_end->z - box->z;
1686
1687	if (loc_start->z + 1 == loc_end->z) {
1688		box_c2 = box->y + box->h;
1689		if (box->h == 0 || box->y > loc_start->y)
1690			box->y = loc_start->y;
1691		if (box_c2 < loc_end->y)
1692			box->h = loc_end->y - box->y;
1693
1694		if (loc_start->y + 1 == loc_end->y) {
1695			box_c2 = box->x + box->w;
1696			if (box->w == 0 || box->x > loc_start->x)
1697				box->x = loc_start->x;
1698			if (box_c2 < loc_end->x)
1699				box->w = loc_end->x - box->x;
1700		} else {
1701			box->x = 0;
1702			box->w = size->width;
1703		}
1704	} else {
1705		box->y = 0;
1706		box->h = size->height;
1707		box->x = 0;
1708		box->w = size->width;
1709	}
1710}
1711
1712/**
1713 * vmw_subres_dirty_full - Mark a full subresource as dirty
1714 * @dirty: The surface's dirty tracker.
1715 * @subres: The subresource
1716 */
1717static void vmw_subres_dirty_full(struct vmw_surface_dirty *dirty, u32 subres)
1718{
1719	const struct vmw_surface_cache *cache = &dirty->cache;
1720	u32 mip = subres % cache->num_mip_levels;
1721	const struct drm_vmw_size *size = &cache->mip[mip].size;
1722	SVGA3dBox *box = &dirty->boxes[subres];
1723
1724	box->x = 0;
1725	box->y = 0;
1726	box->z = 0;
1727	box->w = size->width;
1728	box->h = size->height;
1729	box->d = size->depth;
1730}
1731
1732/*
1733 * vmw_surface_tex_dirty_add_range - The dirty_add_range callback for texture
1734 * surfaces.
1735 */
1736static void vmw_surface_tex_dirty_range_add(struct vmw_resource *res,
1737					    size_t start, size_t end)
1738{
1739	struct vmw_surface_dirty *dirty =
1740		(struct vmw_surface_dirty *) res->dirty;
1741	size_t backup_end = res->guest_memory_offset + res->guest_memory_size;
1742	struct vmw_surface_loc loc1, loc2;
1743	const struct vmw_surface_cache *cache;
1744
1745	start = max_t(size_t, start, res->guest_memory_offset) - res->guest_memory_offset;
1746	end = min(end, backup_end) - res->guest_memory_offset;
1747	cache = &dirty->cache;
1748	vmw_surface_get_loc(cache, &loc1, start);
1749	vmw_surface_get_loc(cache, &loc2, end - 1);
1750	vmw_surface_inc_loc(cache, &loc2);
1751
1752	if (loc1.sheet != loc2.sheet) {
1753		u32 sub_res;
1754
1755		/*
1756		 * Multiple multisample sheets. To do this in an optimized
1757		 * fashion, compute the dirty region for each sheet and the
1758		 * resulting union. Since this is not a common case, just dirty
1759		 * the whole surface.
1760		 */
1761		for (sub_res = 0; sub_res < dirty->num_subres; ++sub_res)
1762			vmw_subres_dirty_full(dirty, sub_res);
1763		return;
1764	}
1765	if (loc1.sub_resource + 1 == loc2.sub_resource) {
1766		/* Dirty range covers a single sub-resource */
1767		vmw_subres_dirty_add(dirty, &loc1, &loc2);
1768	} else {
1769		/* Dirty range covers multiple sub-resources */
1770		struct vmw_surface_loc loc_min, loc_max;
1771		u32 sub_res;
1772
1773		vmw_surface_max_loc(cache, loc1.sub_resource, &loc_max);
1774		vmw_subres_dirty_add(dirty, &loc1, &loc_max);
1775		vmw_surface_min_loc(cache, loc2.sub_resource - 1, &loc_min);
1776		vmw_subres_dirty_add(dirty, &loc_min, &loc2);
1777		for (sub_res = loc1.sub_resource + 1;
1778		     sub_res < loc2.sub_resource - 1; ++sub_res)
1779			vmw_subres_dirty_full(dirty, sub_res);
1780	}
1781}
1782
1783/*
1784 * vmw_surface_tex_dirty_add_range - The dirty_add_range callback for buffer
1785 * surfaces.
1786 */
1787static void vmw_surface_buf_dirty_range_add(struct vmw_resource *res,
1788					    size_t start, size_t end)
1789{
1790	struct vmw_surface_dirty *dirty =
1791		(struct vmw_surface_dirty *) res->dirty;
1792	const struct vmw_surface_cache *cache = &dirty->cache;
1793	size_t backup_end = res->guest_memory_offset + cache->mip_chain_bytes;
1794	SVGA3dBox *box = &dirty->boxes[0];
1795	u32 box_c2;
1796
1797	box->h = box->d = 1;
1798	start = max_t(size_t, start, res->guest_memory_offset) - res->guest_memory_offset;
1799	end = min(end, backup_end) - res->guest_memory_offset;
1800	box_c2 = box->x + box->w;
1801	if (box->w == 0 || box->x > start)
1802		box->x = start;
1803	if (box_c2 < end)
1804		box->w = end - box->x;
1805}
1806
1807/*
1808 * vmw_surface_tex_dirty_add_range - The dirty_add_range callback for surfaces
1809 */
1810static void vmw_surface_dirty_range_add(struct vmw_resource *res, size_t start,
1811					size_t end)
1812{
1813	struct vmw_surface *srf = vmw_res_to_srf(res);
1814
1815	if (WARN_ON(end <= res->guest_memory_offset ||
1816		    start >= res->guest_memory_offset + res->guest_memory_size))
1817		return;
1818
1819	if (srf->metadata.format == SVGA3D_BUFFER)
1820		vmw_surface_buf_dirty_range_add(res, start, end);
1821	else
1822		vmw_surface_tex_dirty_range_add(res, start, end);
1823}
1824
1825/*
1826 * vmw_surface_dirty_sync - The surface's dirty_sync callback.
1827 */
1828static int vmw_surface_dirty_sync(struct vmw_resource *res)
1829{
1830	struct vmw_private *dev_priv = res->dev_priv;
1831	u32 i, num_dirty;
1832	struct vmw_surface_dirty *dirty =
1833		(struct vmw_surface_dirty *) res->dirty;
1834	size_t alloc_size;
1835	const struct vmw_surface_cache *cache = &dirty->cache;
1836	struct {
1837		SVGA3dCmdHeader header;
1838		SVGA3dCmdDXUpdateSubResource body;
1839	} *cmd1;
1840	struct {
1841		SVGA3dCmdHeader header;
1842		SVGA3dCmdUpdateGBImage body;
1843	} *cmd2;
1844	void *cmd;
1845
1846	num_dirty = 0;
1847	for (i = 0; i < dirty->num_subres; ++i) {
1848		const SVGA3dBox *box = &dirty->boxes[i];
1849
1850		if (box->d)
1851			num_dirty++;
1852	}
1853
1854	if (!num_dirty)
1855		goto out;
1856
1857	alloc_size = num_dirty * ((has_sm4_context(dev_priv)) ? sizeof(*cmd1) : sizeof(*cmd2));
1858	cmd = VMW_CMD_RESERVE(dev_priv, alloc_size);
1859	if (!cmd)
1860		return -ENOMEM;
1861
1862	cmd1 = cmd;
1863	cmd2 = cmd;
1864
1865	for (i = 0; i < dirty->num_subres; ++i) {
1866		const SVGA3dBox *box = &dirty->boxes[i];
1867
1868		if (!box->d)
1869			continue;
1870
1871		/*
1872		 * DX_UPDATE_SUBRESOURCE is aware of array surfaces.
1873		 * UPDATE_GB_IMAGE is not.
1874		 */
1875		if (has_sm4_context(dev_priv)) {
1876			cmd1->header.id = SVGA_3D_CMD_DX_UPDATE_SUBRESOURCE;
1877			cmd1->header.size = sizeof(cmd1->body);
1878			cmd1->body.sid = res->id;
1879			cmd1->body.subResource = i;
1880			cmd1->body.box = *box;
1881			cmd1++;
1882		} else {
1883			cmd2->header.id = SVGA_3D_CMD_UPDATE_GB_IMAGE;
1884			cmd2->header.size = sizeof(cmd2->body);
1885			cmd2->body.image.sid = res->id;
1886			cmd2->body.image.face = i / cache->num_mip_levels;
1887			cmd2->body.image.mipmap = i -
1888				(cache->num_mip_levels * cmd2->body.image.face);
1889			cmd2->body.box = *box;
1890			cmd2++;
1891		}
1892
1893	}
1894	vmw_cmd_commit(dev_priv, alloc_size);
1895 out:
1896	memset(&dirty->boxes[0], 0, sizeof(dirty->boxes[0]) *
1897	       dirty->num_subres);
1898
1899	return 0;
1900}
1901
1902/*
1903 * vmw_surface_dirty_alloc - The surface's dirty_alloc callback.
1904 */
1905static int vmw_surface_dirty_alloc(struct vmw_resource *res)
1906{
1907	struct vmw_surface *srf = vmw_res_to_srf(res);
1908	const struct vmw_surface_metadata *metadata = &srf->metadata;
1909	struct vmw_surface_dirty *dirty;
1910	u32 num_layers = 1;
1911	u32 num_mip;
1912	u32 num_subres;
1913	u32 num_samples;
1914	size_t dirty_size;
1915	int ret;
1916
1917	if (metadata->array_size)
1918		num_layers = metadata->array_size;
1919	else if (metadata->flags & SVGA3D_SURFACE_CUBEMAP)
1920		num_layers *= SVGA3D_MAX_SURFACE_FACES;
1921
1922	num_mip = metadata->mip_levels[0];
1923	if (!num_mip)
1924		num_mip = 1;
1925
1926	num_subres = num_layers * num_mip;
1927	dirty_size = struct_size(dirty, boxes, num_subres);
1928
1929	dirty = kvzalloc(dirty_size, GFP_KERNEL);
1930	if (!dirty) {
1931		ret = -ENOMEM;
1932		goto out_no_dirty;
1933	}
1934
1935	num_samples = max_t(u32, 1, metadata->multisample_count);
1936	ret = vmw_surface_setup_cache(&metadata->base_size, metadata->format,
1937				      num_mip, num_layers, num_samples,
1938				      &dirty->cache);
1939	if (ret)
1940		goto out_no_cache;
1941
1942	dirty->num_subres = num_subres;
1943	res->dirty = (struct vmw_resource_dirty *) dirty;
1944
1945	return 0;
1946
1947out_no_cache:
1948	kvfree(dirty);
1949out_no_dirty:
1950	return ret;
1951}
1952
1953/*
1954 * vmw_surface_dirty_free - The surface's dirty_free callback
1955 */
1956static void vmw_surface_dirty_free(struct vmw_resource *res)
1957{
1958	struct vmw_surface_dirty *dirty =
1959		(struct vmw_surface_dirty *) res->dirty;
1960
1961	kvfree(dirty);
1962	res->dirty = NULL;
1963}
1964
1965/*
1966 * vmw_surface_clean - The surface's clean callback
1967 */
1968static int vmw_surface_clean(struct vmw_resource *res)
1969{
1970	struct vmw_private *dev_priv = res->dev_priv;
1971	size_t alloc_size;
1972	struct {
1973		SVGA3dCmdHeader header;
1974		SVGA3dCmdReadbackGBSurface body;
1975	} *cmd;
1976
1977	alloc_size = sizeof(*cmd);
1978	cmd = VMW_CMD_RESERVE(dev_priv, alloc_size);
1979	if (!cmd)
1980		return -ENOMEM;
1981
1982	cmd->header.id = SVGA_3D_CMD_READBACK_GB_SURFACE;
1983	cmd->header.size = sizeof(cmd->body);
1984	cmd->body.sid = res->id;
1985	vmw_cmd_commit(dev_priv, alloc_size);
1986
1987	return 0;
1988}
1989
1990/*
1991 * vmw_gb_surface_define - Define a private GB surface
1992 *
1993 * @dev_priv: Pointer to a device private.
1994 * @metadata: Metadata representing the surface to create.
1995 * @user_srf_out: allocated user_srf. Set to NULL on failure.
1996 *
1997 * GB surfaces allocated by this function will not have a user mode handle, and
1998 * thus will only be visible to vmwgfx.  For optimization reasons the
1999 * surface may later be given a user mode handle by another function to make
2000 * it available to user mode drivers.
2001 */
2002int vmw_gb_surface_define(struct vmw_private *dev_priv,
2003			  const struct vmw_surface_metadata *req,
2004			  struct vmw_surface **srf_out)
2005{
2006	struct vmw_surface_metadata *metadata;
2007	struct vmw_user_surface *user_srf;
2008	struct vmw_surface *srf;
2009	u32 sample_count = 1;
2010	u32 num_layers = 1;
2011	int ret;
2012
2013	*srf_out = NULL;
2014
2015	if (req->scanout) {
2016		if (!vmw_surface_is_screen_target_format(req->format)) {
2017			VMW_DEBUG_USER("Invalid Screen Target surface format.");
2018			return -EINVAL;
2019		}
2020
2021		if (req->base_size.width > dev_priv->texture_max_width ||
2022		    req->base_size.height > dev_priv->texture_max_height) {
2023			VMW_DEBUG_USER("%ux%u\n, exceed max surface size %ux%u",
2024				       req->base_size.width,
2025				       req->base_size.height,
2026				       dev_priv->texture_max_width,
2027				       dev_priv->texture_max_height);
2028			return -EINVAL;
2029		}
2030	} else {
2031		const SVGA3dSurfaceDesc *desc =
2032			vmw_surface_get_desc(req->format);
2033
2034		if (desc->blockDesc == SVGA3DBLOCKDESC_NONE) {
2035			VMW_DEBUG_USER("Invalid surface format.\n");
2036			return -EINVAL;
2037		}
2038	}
2039
2040	if (req->autogen_filter != SVGA3D_TEX_FILTER_NONE)
2041		return -EINVAL;
2042
2043	if (req->num_sizes != 1)
2044		return -EINVAL;
2045
2046	if (req->sizes != NULL)
2047		return -EINVAL;
2048
2049	user_srf = kzalloc(sizeof(*user_srf), GFP_KERNEL);
2050	if (unlikely(!user_srf)) {
2051		ret = -ENOMEM;
2052		goto out_unlock;
2053	}
2054
2055	*srf_out  = &user_srf->srf;
2056	user_srf->prime.base.shareable = false;
2057	user_srf->prime.base.tfile = NULL;
2058
2059	srf = &user_srf->srf;
2060	srf->metadata = *req;
2061	srf->offsets = NULL;
2062
2063	metadata = &srf->metadata;
2064
2065	if (metadata->array_size)
2066		num_layers = req->array_size;
2067	else if (metadata->flags & SVGA3D_SURFACE_CUBEMAP)
2068		num_layers = SVGA3D_MAX_SURFACE_FACES;
2069
2070	if (metadata->flags & SVGA3D_SURFACE_MULTISAMPLE)
2071		sample_count = metadata->multisample_count;
2072
2073	srf->res.guest_memory_size =
2074		vmw_surface_get_serialized_size_extended(
2075				metadata->format,
2076				metadata->base_size,
2077				metadata->mip_levels[0],
2078				num_layers,
2079				sample_count);
2080
2081	if (metadata->flags & SVGA3D_SURFACE_BIND_STREAM_OUTPUT)
2082		srf->res.guest_memory_size += sizeof(SVGA3dDXSOState);
2083
2084	/*
2085	 * Don't set SVGA3D_SURFACE_SCREENTARGET flag for a scanout surface with
2086	 * size greater than STDU max width/height. This is really a workaround
2087	 * to support creation of big framebuffer requested by some user-space
2088	 * for whole topology. That big framebuffer won't really be used for
2089	 * binding with screen target as during prepare_fb a separate surface is
2090	 * created so it's safe to ignore SVGA3D_SURFACE_SCREENTARGET flag.
2091	 */
2092	if (dev_priv->active_display_unit == vmw_du_screen_target &&
2093	    metadata->scanout &&
2094	    metadata->base_size.width <= dev_priv->stdu_max_width &&
2095	    metadata->base_size.height <= dev_priv->stdu_max_height)
2096		metadata->flags |= SVGA3D_SURFACE_SCREENTARGET;
2097
2098	/*
2099	 * From this point, the generic resource management functions
2100	 * destroy the object on failure.
2101	 */
2102	ret = vmw_surface_init(dev_priv, srf, vmw_user_surface_free);
2103
2104	return ret;
2105
2106out_unlock:
2107	return ret;
2108}
2109