xref: /linux/drivers/gpu/drm/i915/gvt/kvmgt.c (revision 062e720cd209d8091c4f3d118d93973f02209aca)
1 /*
2  * KVMGT - the implementation of Intel mediated pass-through framework for KVM
3  *
4  * Copyright(c) 2011-2016 Intel Corporation. All rights reserved.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice (including the next
14  * paragraph) shall be included in all copies or substantial portions of the
15  * Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23  * SOFTWARE.
24  *
25  * Authors:
26  *    Kevin Tian <kevin.tian@intel.com>
27  *    Jike Song <jike.song@intel.com>
28  *    Xiaoguang Chen <xiaoguang.chen@intel.com>
29  *    Eddie Dong <eddie.dong@intel.com>
30  *
31  * Contributors:
32  *    Niu Bing <bing.niu@intel.com>
33  *    Zhi Wang <zhi.a.wang@intel.com>
34  */
35 
36 #include <linux/init.h>
37 #include <linux/mm.h>
38 #include <linux/kthread.h>
39 #include <linux/sched/mm.h>
40 #include <linux/types.h>
41 #include <linux/list.h>
42 #include <linux/rbtree.h>
43 #include <linux/spinlock.h>
44 #include <linux/eventfd.h>
45 #include <linux/mdev.h>
46 #include <linux/debugfs.h>
47 
48 #include <linux/nospec.h>
49 
50 #include <drm/drm_edid.h>
51 
52 #include "i915_drv.h"
53 #include "intel_gvt.h"
54 #include "gvt.h"
55 
56 MODULE_IMPORT_NS(DMA_BUF);
57 MODULE_IMPORT_NS(I915_GVT);
58 
59 /* helper macros copied from vfio-pci */
60 #define VFIO_PCI_OFFSET_SHIFT   40
61 #define VFIO_PCI_OFFSET_TO_INDEX(off)   (off >> VFIO_PCI_OFFSET_SHIFT)
62 #define VFIO_PCI_INDEX_TO_OFFSET(index) ((u64)(index) << VFIO_PCI_OFFSET_SHIFT)
63 #define VFIO_PCI_OFFSET_MASK    (((u64)(1) << VFIO_PCI_OFFSET_SHIFT) - 1)
64 
65 #define EDID_BLOB_OFFSET (PAGE_SIZE/2)
66 
67 #define OPREGION_SIGNATURE "IntelGraphicsMem"
68 
69 struct vfio_region;
70 struct intel_vgpu_regops {
71 	size_t (*rw)(struct intel_vgpu *vgpu, char *buf,
72 			size_t count, loff_t *ppos, bool iswrite);
73 	void (*release)(struct intel_vgpu *vgpu,
74 			struct vfio_region *region);
75 };
76 
77 struct vfio_region {
78 	u32				type;
79 	u32				subtype;
80 	size_t				size;
81 	u32				flags;
82 	const struct intel_vgpu_regops	*ops;
83 	void				*data;
84 };
85 
86 struct vfio_edid_region {
87 	struct vfio_region_gfx_edid vfio_edid_regs;
88 	void *edid_blob;
89 };
90 
91 struct kvmgt_pgfn {
92 	gfn_t gfn;
93 	struct hlist_node hnode;
94 };
95 
96 struct gvt_dma {
97 	struct intel_vgpu *vgpu;
98 	struct rb_node gfn_node;
99 	struct rb_node dma_addr_node;
100 	gfn_t gfn;
101 	dma_addr_t dma_addr;
102 	unsigned long size;
103 	struct kref ref;
104 };
105 
106 #define vfio_dev_to_vgpu(vfio_dev) \
107 	container_of((vfio_dev), struct intel_vgpu, vfio_device)
108 
109 static void kvmgt_page_track_write(struct kvm_vcpu *vcpu, gpa_t gpa,
110 		const u8 *val, int len,
111 		struct kvm_page_track_notifier_node *node);
112 static void kvmgt_page_track_flush_slot(struct kvm *kvm,
113 		struct kvm_memory_slot *slot,
114 		struct kvm_page_track_notifier_node *node);
115 
116 static ssize_t available_instances_show(struct mdev_type *mtype,
117 					struct mdev_type_attribute *attr,
118 					char *buf)
119 {
120 	struct intel_vgpu_type *type =
121 		container_of(mtype, struct intel_vgpu_type, type);
122 
123 	return sprintf(buf, "%u\n", type->avail_instance);
124 }
125 
126 static ssize_t device_api_show(struct mdev_type *mtype,
127 			       struct mdev_type_attribute *attr, char *buf)
128 {
129 	return sprintf(buf, "%s\n", VFIO_DEVICE_API_PCI_STRING);
130 }
131 
132 static ssize_t description_show(struct mdev_type *mtype,
133 				struct mdev_type_attribute *attr, char *buf)
134 {
135 	struct intel_vgpu_type *type =
136 		container_of(mtype, struct intel_vgpu_type, type);
137 
138 	return sprintf(buf, "low_gm_size: %dMB\nhigh_gm_size: %dMB\n"
139 		       "fence: %d\nresolution: %s\n"
140 		       "weight: %d\n",
141 		       BYTES_TO_MB(type->conf->low_mm),
142 		       BYTES_TO_MB(type->conf->high_mm),
143 		       type->conf->fence, vgpu_edid_str(type->conf->edid),
144 		       type->conf->weight);
145 }
146 
147 static ssize_t name_show(struct mdev_type *mtype,
148 			 struct mdev_type_attribute *attr, char *buf)
149 {
150 	return sprintf(buf, "%s\n", mtype->sysfs_name);
151 }
152 
153 static MDEV_TYPE_ATTR_RO(available_instances);
154 static MDEV_TYPE_ATTR_RO(device_api);
155 static MDEV_TYPE_ATTR_RO(description);
156 static MDEV_TYPE_ATTR_RO(name);
157 
158 static const struct attribute *gvt_type_attrs[] = {
159 	&mdev_type_attr_available_instances.attr,
160 	&mdev_type_attr_device_api.attr,
161 	&mdev_type_attr_description.attr,
162 	&mdev_type_attr_name.attr,
163 	NULL,
164 };
165 
166 static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
167 		unsigned long size)
168 {
169 	vfio_unpin_pages(&vgpu->vfio_device, gfn << PAGE_SHIFT,
170 			 DIV_ROUND_UP(size, PAGE_SIZE));
171 }
172 
173 /* Pin a normal or compound guest page for dma. */
174 static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
175 		unsigned long size, struct page **page)
176 {
177 	int total_pages = DIV_ROUND_UP(size, PAGE_SIZE);
178 	struct page *base_page = NULL;
179 	int npage;
180 	int ret;
181 
182 	/*
183 	 * We pin the pages one-by-one to avoid allocating a big arrary
184 	 * on stack to hold pfns.
185 	 */
186 	for (npage = 0; npage < total_pages; npage++) {
187 		dma_addr_t cur_iova = (gfn + npage) << PAGE_SHIFT;
188 		struct page *cur_page;
189 
190 		ret = vfio_pin_pages(&vgpu->vfio_device, cur_iova, 1,
191 				     IOMMU_READ | IOMMU_WRITE, &cur_page);
192 		if (ret != 1) {
193 			gvt_vgpu_err("vfio_pin_pages failed for iova %pad, ret %d\n",
194 				     &cur_iova, ret);
195 			goto err;
196 		}
197 
198 		if (npage == 0)
199 			base_page = cur_page;
200 		else if (base_page + npage != cur_page) {
201 			gvt_vgpu_err("The pages are not continuous\n");
202 			ret = -EINVAL;
203 			npage++;
204 			goto err;
205 		}
206 	}
207 
208 	*page = base_page;
209 	return 0;
210 err:
211 	gvt_unpin_guest_page(vgpu, gfn, npage * PAGE_SIZE);
212 	return ret;
213 }
214 
215 static int gvt_dma_map_page(struct intel_vgpu *vgpu, unsigned long gfn,
216 		dma_addr_t *dma_addr, unsigned long size)
217 {
218 	struct device *dev = vgpu->gvt->gt->i915->drm.dev;
219 	struct page *page = NULL;
220 	int ret;
221 
222 	ret = gvt_pin_guest_page(vgpu, gfn, size, &page);
223 	if (ret)
224 		return ret;
225 
226 	/* Setup DMA mapping. */
227 	*dma_addr = dma_map_page(dev, page, 0, size, DMA_BIDIRECTIONAL);
228 	if (dma_mapping_error(dev, *dma_addr)) {
229 		gvt_vgpu_err("DMA mapping failed for pfn 0x%lx, ret %d\n",
230 			     page_to_pfn(page), ret);
231 		gvt_unpin_guest_page(vgpu, gfn, size);
232 		return -ENOMEM;
233 	}
234 
235 	return 0;
236 }
237 
238 static void gvt_dma_unmap_page(struct intel_vgpu *vgpu, unsigned long gfn,
239 		dma_addr_t dma_addr, unsigned long size)
240 {
241 	struct device *dev = vgpu->gvt->gt->i915->drm.dev;
242 
243 	dma_unmap_page(dev, dma_addr, size, DMA_BIDIRECTIONAL);
244 	gvt_unpin_guest_page(vgpu, gfn, size);
245 }
246 
247 static struct gvt_dma *__gvt_cache_find_dma_addr(struct intel_vgpu *vgpu,
248 		dma_addr_t dma_addr)
249 {
250 	struct rb_node *node = vgpu->dma_addr_cache.rb_node;
251 	struct gvt_dma *itr;
252 
253 	while (node) {
254 		itr = rb_entry(node, struct gvt_dma, dma_addr_node);
255 
256 		if (dma_addr < itr->dma_addr)
257 			node = node->rb_left;
258 		else if (dma_addr > itr->dma_addr)
259 			node = node->rb_right;
260 		else
261 			return itr;
262 	}
263 	return NULL;
264 }
265 
266 static struct gvt_dma *__gvt_cache_find_gfn(struct intel_vgpu *vgpu, gfn_t gfn)
267 {
268 	struct rb_node *node = vgpu->gfn_cache.rb_node;
269 	struct gvt_dma *itr;
270 
271 	while (node) {
272 		itr = rb_entry(node, struct gvt_dma, gfn_node);
273 
274 		if (gfn < itr->gfn)
275 			node = node->rb_left;
276 		else if (gfn > itr->gfn)
277 			node = node->rb_right;
278 		else
279 			return itr;
280 	}
281 	return NULL;
282 }
283 
284 static int __gvt_cache_add(struct intel_vgpu *vgpu, gfn_t gfn,
285 		dma_addr_t dma_addr, unsigned long size)
286 {
287 	struct gvt_dma *new, *itr;
288 	struct rb_node **link, *parent = NULL;
289 
290 	new = kzalloc(sizeof(struct gvt_dma), GFP_KERNEL);
291 	if (!new)
292 		return -ENOMEM;
293 
294 	new->vgpu = vgpu;
295 	new->gfn = gfn;
296 	new->dma_addr = dma_addr;
297 	new->size = size;
298 	kref_init(&new->ref);
299 
300 	/* gfn_cache maps gfn to struct gvt_dma. */
301 	link = &vgpu->gfn_cache.rb_node;
302 	while (*link) {
303 		parent = *link;
304 		itr = rb_entry(parent, struct gvt_dma, gfn_node);
305 
306 		if (gfn < itr->gfn)
307 			link = &parent->rb_left;
308 		else
309 			link = &parent->rb_right;
310 	}
311 	rb_link_node(&new->gfn_node, parent, link);
312 	rb_insert_color(&new->gfn_node, &vgpu->gfn_cache);
313 
314 	/* dma_addr_cache maps dma addr to struct gvt_dma. */
315 	parent = NULL;
316 	link = &vgpu->dma_addr_cache.rb_node;
317 	while (*link) {
318 		parent = *link;
319 		itr = rb_entry(parent, struct gvt_dma, dma_addr_node);
320 
321 		if (dma_addr < itr->dma_addr)
322 			link = &parent->rb_left;
323 		else
324 			link = &parent->rb_right;
325 	}
326 	rb_link_node(&new->dma_addr_node, parent, link);
327 	rb_insert_color(&new->dma_addr_node, &vgpu->dma_addr_cache);
328 
329 	vgpu->nr_cache_entries++;
330 	return 0;
331 }
332 
333 static void __gvt_cache_remove_entry(struct intel_vgpu *vgpu,
334 				struct gvt_dma *entry)
335 {
336 	rb_erase(&entry->gfn_node, &vgpu->gfn_cache);
337 	rb_erase(&entry->dma_addr_node, &vgpu->dma_addr_cache);
338 	kfree(entry);
339 	vgpu->nr_cache_entries--;
340 }
341 
342 static void gvt_cache_destroy(struct intel_vgpu *vgpu)
343 {
344 	struct gvt_dma *dma;
345 	struct rb_node *node = NULL;
346 
347 	for (;;) {
348 		mutex_lock(&vgpu->cache_lock);
349 		node = rb_first(&vgpu->gfn_cache);
350 		if (!node) {
351 			mutex_unlock(&vgpu->cache_lock);
352 			break;
353 		}
354 		dma = rb_entry(node, struct gvt_dma, gfn_node);
355 		gvt_dma_unmap_page(vgpu, dma->gfn, dma->dma_addr, dma->size);
356 		__gvt_cache_remove_entry(vgpu, dma);
357 		mutex_unlock(&vgpu->cache_lock);
358 	}
359 }
360 
361 static void gvt_cache_init(struct intel_vgpu *vgpu)
362 {
363 	vgpu->gfn_cache = RB_ROOT;
364 	vgpu->dma_addr_cache = RB_ROOT;
365 	vgpu->nr_cache_entries = 0;
366 	mutex_init(&vgpu->cache_lock);
367 }
368 
369 static void kvmgt_protect_table_init(struct intel_vgpu *info)
370 {
371 	hash_init(info->ptable);
372 }
373 
374 static void kvmgt_protect_table_destroy(struct intel_vgpu *info)
375 {
376 	struct kvmgt_pgfn *p;
377 	struct hlist_node *tmp;
378 	int i;
379 
380 	hash_for_each_safe(info->ptable, i, tmp, p, hnode) {
381 		hash_del(&p->hnode);
382 		kfree(p);
383 	}
384 }
385 
386 static struct kvmgt_pgfn *
387 __kvmgt_protect_table_find(struct intel_vgpu *info, gfn_t gfn)
388 {
389 	struct kvmgt_pgfn *p, *res = NULL;
390 
391 	hash_for_each_possible(info->ptable, p, hnode, gfn) {
392 		if (gfn == p->gfn) {
393 			res = p;
394 			break;
395 		}
396 	}
397 
398 	return res;
399 }
400 
401 static bool kvmgt_gfn_is_write_protected(struct intel_vgpu *info, gfn_t gfn)
402 {
403 	struct kvmgt_pgfn *p;
404 
405 	p = __kvmgt_protect_table_find(info, gfn);
406 	return !!p;
407 }
408 
409 static void kvmgt_protect_table_add(struct intel_vgpu *info, gfn_t gfn)
410 {
411 	struct kvmgt_pgfn *p;
412 
413 	if (kvmgt_gfn_is_write_protected(info, gfn))
414 		return;
415 
416 	p = kzalloc(sizeof(struct kvmgt_pgfn), GFP_ATOMIC);
417 	if (WARN(!p, "gfn: 0x%llx\n", gfn))
418 		return;
419 
420 	p->gfn = gfn;
421 	hash_add(info->ptable, &p->hnode, gfn);
422 }
423 
424 static void kvmgt_protect_table_del(struct intel_vgpu *info, gfn_t gfn)
425 {
426 	struct kvmgt_pgfn *p;
427 
428 	p = __kvmgt_protect_table_find(info, gfn);
429 	if (p) {
430 		hash_del(&p->hnode);
431 		kfree(p);
432 	}
433 }
434 
435 static size_t intel_vgpu_reg_rw_opregion(struct intel_vgpu *vgpu, char *buf,
436 		size_t count, loff_t *ppos, bool iswrite)
437 {
438 	unsigned int i = VFIO_PCI_OFFSET_TO_INDEX(*ppos) -
439 			VFIO_PCI_NUM_REGIONS;
440 	void *base = vgpu->region[i].data;
441 	loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK;
442 
443 
444 	if (pos >= vgpu->region[i].size || iswrite) {
445 		gvt_vgpu_err("invalid op or offset for Intel vgpu OpRegion\n");
446 		return -EINVAL;
447 	}
448 	count = min(count, (size_t)(vgpu->region[i].size - pos));
449 	memcpy(buf, base + pos, count);
450 
451 	return count;
452 }
453 
454 static void intel_vgpu_reg_release_opregion(struct intel_vgpu *vgpu,
455 		struct vfio_region *region)
456 {
457 }
458 
459 static const struct intel_vgpu_regops intel_vgpu_regops_opregion = {
460 	.rw = intel_vgpu_reg_rw_opregion,
461 	.release = intel_vgpu_reg_release_opregion,
462 };
463 
464 static int handle_edid_regs(struct intel_vgpu *vgpu,
465 			struct vfio_edid_region *region, char *buf,
466 			size_t count, u16 offset, bool is_write)
467 {
468 	struct vfio_region_gfx_edid *regs = &region->vfio_edid_regs;
469 	unsigned int data;
470 
471 	if (offset + count > sizeof(*regs))
472 		return -EINVAL;
473 
474 	if (count != 4)
475 		return -EINVAL;
476 
477 	if (is_write) {
478 		data = *((unsigned int *)buf);
479 		switch (offset) {
480 		case offsetof(struct vfio_region_gfx_edid, link_state):
481 			if (data == VFIO_DEVICE_GFX_LINK_STATE_UP) {
482 				if (!drm_edid_block_valid(
483 					(u8 *)region->edid_blob,
484 					0,
485 					true,
486 					NULL)) {
487 					gvt_vgpu_err("invalid EDID blob\n");
488 					return -EINVAL;
489 				}
490 				intel_vgpu_emulate_hotplug(vgpu, true);
491 			} else if (data == VFIO_DEVICE_GFX_LINK_STATE_DOWN)
492 				intel_vgpu_emulate_hotplug(vgpu, false);
493 			else {
494 				gvt_vgpu_err("invalid EDID link state %d\n",
495 					regs->link_state);
496 				return -EINVAL;
497 			}
498 			regs->link_state = data;
499 			break;
500 		case offsetof(struct vfio_region_gfx_edid, edid_size):
501 			if (data > regs->edid_max_size) {
502 				gvt_vgpu_err("EDID size is bigger than %d!\n",
503 					regs->edid_max_size);
504 				return -EINVAL;
505 			}
506 			regs->edid_size = data;
507 			break;
508 		default:
509 			/* read-only regs */
510 			gvt_vgpu_err("write read-only EDID region at offset %d\n",
511 				offset);
512 			return -EPERM;
513 		}
514 	} else {
515 		memcpy(buf, (char *)regs + offset, count);
516 	}
517 
518 	return count;
519 }
520 
521 static int handle_edid_blob(struct vfio_edid_region *region, char *buf,
522 			size_t count, u16 offset, bool is_write)
523 {
524 	if (offset + count > region->vfio_edid_regs.edid_size)
525 		return -EINVAL;
526 
527 	if (is_write)
528 		memcpy(region->edid_blob + offset, buf, count);
529 	else
530 		memcpy(buf, region->edid_blob + offset, count);
531 
532 	return count;
533 }
534 
535 static size_t intel_vgpu_reg_rw_edid(struct intel_vgpu *vgpu, char *buf,
536 		size_t count, loff_t *ppos, bool iswrite)
537 {
538 	int ret;
539 	unsigned int i = VFIO_PCI_OFFSET_TO_INDEX(*ppos) -
540 			VFIO_PCI_NUM_REGIONS;
541 	struct vfio_edid_region *region = vgpu->region[i].data;
542 	loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK;
543 
544 	if (pos < region->vfio_edid_regs.edid_offset) {
545 		ret = handle_edid_regs(vgpu, region, buf, count, pos, iswrite);
546 	} else {
547 		pos -= EDID_BLOB_OFFSET;
548 		ret = handle_edid_blob(region, buf, count, pos, iswrite);
549 	}
550 
551 	if (ret < 0)
552 		gvt_vgpu_err("failed to access EDID region\n");
553 
554 	return ret;
555 }
556 
557 static void intel_vgpu_reg_release_edid(struct intel_vgpu *vgpu,
558 					struct vfio_region *region)
559 {
560 	kfree(region->data);
561 }
562 
563 static const struct intel_vgpu_regops intel_vgpu_regops_edid = {
564 	.rw = intel_vgpu_reg_rw_edid,
565 	.release = intel_vgpu_reg_release_edid,
566 };
567 
568 static int intel_vgpu_register_reg(struct intel_vgpu *vgpu,
569 		unsigned int type, unsigned int subtype,
570 		const struct intel_vgpu_regops *ops,
571 		size_t size, u32 flags, void *data)
572 {
573 	struct vfio_region *region;
574 
575 	region = krealloc(vgpu->region,
576 			(vgpu->num_regions + 1) * sizeof(*region),
577 			GFP_KERNEL);
578 	if (!region)
579 		return -ENOMEM;
580 
581 	vgpu->region = region;
582 	vgpu->region[vgpu->num_regions].type = type;
583 	vgpu->region[vgpu->num_regions].subtype = subtype;
584 	vgpu->region[vgpu->num_regions].ops = ops;
585 	vgpu->region[vgpu->num_regions].size = size;
586 	vgpu->region[vgpu->num_regions].flags = flags;
587 	vgpu->region[vgpu->num_regions].data = data;
588 	vgpu->num_regions++;
589 	return 0;
590 }
591 
592 int intel_gvt_set_opregion(struct intel_vgpu *vgpu)
593 {
594 	void *base;
595 	int ret;
596 
597 	/* Each vgpu has its own opregion, although VFIO would create another
598 	 * one later. This one is used to expose opregion to VFIO. And the
599 	 * other one created by VFIO later, is used by guest actually.
600 	 */
601 	base = vgpu_opregion(vgpu)->va;
602 	if (!base)
603 		return -ENOMEM;
604 
605 	if (memcmp(base, OPREGION_SIGNATURE, 16)) {
606 		memunmap(base);
607 		return -EINVAL;
608 	}
609 
610 	ret = intel_vgpu_register_reg(vgpu,
611 			PCI_VENDOR_ID_INTEL | VFIO_REGION_TYPE_PCI_VENDOR_TYPE,
612 			VFIO_REGION_SUBTYPE_INTEL_IGD_OPREGION,
613 			&intel_vgpu_regops_opregion, OPREGION_SIZE,
614 			VFIO_REGION_INFO_FLAG_READ, base);
615 
616 	return ret;
617 }
618 
619 int intel_gvt_set_edid(struct intel_vgpu *vgpu, int port_num)
620 {
621 	struct intel_vgpu_port *port = intel_vgpu_port(vgpu, port_num);
622 	struct vfio_edid_region *base;
623 	int ret;
624 
625 	base = kzalloc(sizeof(*base), GFP_KERNEL);
626 	if (!base)
627 		return -ENOMEM;
628 
629 	/* TODO: Add multi-port and EDID extension block support */
630 	base->vfio_edid_regs.edid_offset = EDID_BLOB_OFFSET;
631 	base->vfio_edid_regs.edid_max_size = EDID_SIZE;
632 	base->vfio_edid_regs.edid_size = EDID_SIZE;
633 	base->vfio_edid_regs.max_xres = vgpu_edid_xres(port->id);
634 	base->vfio_edid_regs.max_yres = vgpu_edid_yres(port->id);
635 	base->edid_blob = port->edid->edid_block;
636 
637 	ret = intel_vgpu_register_reg(vgpu,
638 			VFIO_REGION_TYPE_GFX,
639 			VFIO_REGION_SUBTYPE_GFX_EDID,
640 			&intel_vgpu_regops_edid, EDID_SIZE,
641 			VFIO_REGION_INFO_FLAG_READ |
642 			VFIO_REGION_INFO_FLAG_WRITE |
643 			VFIO_REGION_INFO_FLAG_CAPS, base);
644 
645 	return ret;
646 }
647 
648 static void intel_vgpu_dma_unmap(struct vfio_device *vfio_dev, u64 iova,
649 				 u64 length)
650 {
651 	struct intel_vgpu *vgpu = vfio_dev_to_vgpu(vfio_dev);
652 	struct gvt_dma *entry;
653 	u64 iov_pfn = iova >> PAGE_SHIFT;
654 	u64 end_iov_pfn = iov_pfn + length / PAGE_SIZE;
655 
656 	mutex_lock(&vgpu->cache_lock);
657 	for (; iov_pfn < end_iov_pfn; iov_pfn++) {
658 		entry = __gvt_cache_find_gfn(vgpu, iov_pfn);
659 		if (!entry)
660 			continue;
661 
662 		gvt_dma_unmap_page(vgpu, entry->gfn, entry->dma_addr,
663 				   entry->size);
664 		__gvt_cache_remove_entry(vgpu, entry);
665 	}
666 	mutex_unlock(&vgpu->cache_lock);
667 }
668 
669 static bool __kvmgt_vgpu_exist(struct intel_vgpu *vgpu)
670 {
671 	struct intel_vgpu *itr;
672 	int id;
673 	bool ret = false;
674 
675 	mutex_lock(&vgpu->gvt->lock);
676 	for_each_active_vgpu(vgpu->gvt, itr, id) {
677 		if (!itr->attached)
678 			continue;
679 
680 		if (vgpu->vfio_device.kvm == itr->vfio_device.kvm) {
681 			ret = true;
682 			goto out;
683 		}
684 	}
685 out:
686 	mutex_unlock(&vgpu->gvt->lock);
687 	return ret;
688 }
689 
690 static int intel_vgpu_open_device(struct vfio_device *vfio_dev)
691 {
692 	struct intel_vgpu *vgpu = vfio_dev_to_vgpu(vfio_dev);
693 
694 	if (vgpu->attached)
695 		return -EEXIST;
696 
697 	if (!vgpu->vfio_device.kvm ||
698 	    vgpu->vfio_device.kvm->mm != current->mm) {
699 		gvt_vgpu_err("KVM is required to use Intel vGPU\n");
700 		return -ESRCH;
701 	}
702 
703 	kvm_get_kvm(vgpu->vfio_device.kvm);
704 
705 	if (__kvmgt_vgpu_exist(vgpu))
706 		return -EEXIST;
707 
708 	vgpu->attached = true;
709 
710 	kvmgt_protect_table_init(vgpu);
711 	gvt_cache_init(vgpu);
712 
713 	vgpu->track_node.track_write = kvmgt_page_track_write;
714 	vgpu->track_node.track_flush_slot = kvmgt_page_track_flush_slot;
715 	kvm_page_track_register_notifier(vgpu->vfio_device.kvm,
716 					 &vgpu->track_node);
717 
718 	debugfs_create_ulong(KVMGT_DEBUGFS_FILENAME, 0444, vgpu->debugfs,
719 			     &vgpu->nr_cache_entries);
720 
721 	intel_gvt_activate_vgpu(vgpu);
722 
723 	atomic_set(&vgpu->released, 0);
724 	return 0;
725 }
726 
727 static void intel_vgpu_release_msi_eventfd_ctx(struct intel_vgpu *vgpu)
728 {
729 	struct eventfd_ctx *trigger;
730 
731 	trigger = vgpu->msi_trigger;
732 	if (trigger) {
733 		eventfd_ctx_put(trigger);
734 		vgpu->msi_trigger = NULL;
735 	}
736 }
737 
738 static void intel_vgpu_close_device(struct vfio_device *vfio_dev)
739 {
740 	struct intel_vgpu *vgpu = vfio_dev_to_vgpu(vfio_dev);
741 
742 	if (!vgpu->attached)
743 		return;
744 
745 	if (atomic_cmpxchg(&vgpu->released, 0, 1))
746 		return;
747 
748 	intel_gvt_release_vgpu(vgpu);
749 
750 	debugfs_remove(debugfs_lookup(KVMGT_DEBUGFS_FILENAME, vgpu->debugfs));
751 
752 	kvm_page_track_unregister_notifier(vgpu->vfio_device.kvm,
753 					   &vgpu->track_node);
754 	kvmgt_protect_table_destroy(vgpu);
755 	gvt_cache_destroy(vgpu);
756 
757 	intel_vgpu_release_msi_eventfd_ctx(vgpu);
758 
759 	vgpu->attached = false;
760 
761 	if (vgpu->vfio_device.kvm)
762 		kvm_put_kvm(vgpu->vfio_device.kvm);
763 }
764 
765 static u64 intel_vgpu_get_bar_addr(struct intel_vgpu *vgpu, int bar)
766 {
767 	u32 start_lo, start_hi;
768 	u32 mem_type;
769 
770 	start_lo = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + bar)) &
771 			PCI_BASE_ADDRESS_MEM_MASK;
772 	mem_type = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + bar)) &
773 			PCI_BASE_ADDRESS_MEM_TYPE_MASK;
774 
775 	switch (mem_type) {
776 	case PCI_BASE_ADDRESS_MEM_TYPE_64:
777 		start_hi = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space
778 						+ bar + 4));
779 		break;
780 	case PCI_BASE_ADDRESS_MEM_TYPE_32:
781 	case PCI_BASE_ADDRESS_MEM_TYPE_1M:
782 		/* 1M mem BAR treated as 32-bit BAR */
783 	default:
784 		/* mem unknown type treated as 32-bit BAR */
785 		start_hi = 0;
786 		break;
787 	}
788 
789 	return ((u64)start_hi << 32) | start_lo;
790 }
791 
792 static int intel_vgpu_bar_rw(struct intel_vgpu *vgpu, int bar, u64 off,
793 			     void *buf, unsigned int count, bool is_write)
794 {
795 	u64 bar_start = intel_vgpu_get_bar_addr(vgpu, bar);
796 	int ret;
797 
798 	if (is_write)
799 		ret = intel_vgpu_emulate_mmio_write(vgpu,
800 					bar_start + off, buf, count);
801 	else
802 		ret = intel_vgpu_emulate_mmio_read(vgpu,
803 					bar_start + off, buf, count);
804 	return ret;
805 }
806 
807 static inline bool intel_vgpu_in_aperture(struct intel_vgpu *vgpu, u64 off)
808 {
809 	return off >= vgpu_aperture_offset(vgpu) &&
810 	       off < vgpu_aperture_offset(vgpu) + vgpu_aperture_sz(vgpu);
811 }
812 
813 static int intel_vgpu_aperture_rw(struct intel_vgpu *vgpu, u64 off,
814 		void *buf, unsigned long count, bool is_write)
815 {
816 	void __iomem *aperture_va;
817 
818 	if (!intel_vgpu_in_aperture(vgpu, off) ||
819 	    !intel_vgpu_in_aperture(vgpu, off + count)) {
820 		gvt_vgpu_err("Invalid aperture offset %llu\n", off);
821 		return -EINVAL;
822 	}
823 
824 	aperture_va = io_mapping_map_wc(&vgpu->gvt->gt->ggtt->iomap,
825 					ALIGN_DOWN(off, PAGE_SIZE),
826 					count + offset_in_page(off));
827 	if (!aperture_va)
828 		return -EIO;
829 
830 	if (is_write)
831 		memcpy_toio(aperture_va + offset_in_page(off), buf, count);
832 	else
833 		memcpy_fromio(buf, aperture_va + offset_in_page(off), count);
834 
835 	io_mapping_unmap(aperture_va);
836 
837 	return 0;
838 }
839 
840 static ssize_t intel_vgpu_rw(struct intel_vgpu *vgpu, char *buf,
841 			size_t count, loff_t *ppos, bool is_write)
842 {
843 	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
844 	u64 pos = *ppos & VFIO_PCI_OFFSET_MASK;
845 	int ret = -EINVAL;
846 
847 
848 	if (index >= VFIO_PCI_NUM_REGIONS + vgpu->num_regions) {
849 		gvt_vgpu_err("invalid index: %u\n", index);
850 		return -EINVAL;
851 	}
852 
853 	switch (index) {
854 	case VFIO_PCI_CONFIG_REGION_INDEX:
855 		if (is_write)
856 			ret = intel_vgpu_emulate_cfg_write(vgpu, pos,
857 						buf, count);
858 		else
859 			ret = intel_vgpu_emulate_cfg_read(vgpu, pos,
860 						buf, count);
861 		break;
862 	case VFIO_PCI_BAR0_REGION_INDEX:
863 		ret = intel_vgpu_bar_rw(vgpu, PCI_BASE_ADDRESS_0, pos,
864 					buf, count, is_write);
865 		break;
866 	case VFIO_PCI_BAR2_REGION_INDEX:
867 		ret = intel_vgpu_aperture_rw(vgpu, pos, buf, count, is_write);
868 		break;
869 	case VFIO_PCI_BAR1_REGION_INDEX:
870 	case VFIO_PCI_BAR3_REGION_INDEX:
871 	case VFIO_PCI_BAR4_REGION_INDEX:
872 	case VFIO_PCI_BAR5_REGION_INDEX:
873 	case VFIO_PCI_VGA_REGION_INDEX:
874 	case VFIO_PCI_ROM_REGION_INDEX:
875 		break;
876 	default:
877 		if (index >= VFIO_PCI_NUM_REGIONS + vgpu->num_regions)
878 			return -EINVAL;
879 
880 		index -= VFIO_PCI_NUM_REGIONS;
881 		return vgpu->region[index].ops->rw(vgpu, buf, count,
882 				ppos, is_write);
883 	}
884 
885 	return ret == 0 ? count : ret;
886 }
887 
888 static bool gtt_entry(struct intel_vgpu *vgpu, loff_t *ppos)
889 {
890 	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
891 	struct intel_gvt *gvt = vgpu->gvt;
892 	int offset;
893 
894 	/* Only allow MMIO GGTT entry access */
895 	if (index != PCI_BASE_ADDRESS_0)
896 		return false;
897 
898 	offset = (u64)(*ppos & VFIO_PCI_OFFSET_MASK) -
899 		intel_vgpu_get_bar_gpa(vgpu, PCI_BASE_ADDRESS_0);
900 
901 	return (offset >= gvt->device_info.gtt_start_offset &&
902 		offset < gvt->device_info.gtt_start_offset + gvt_ggtt_sz(gvt)) ?
903 			true : false;
904 }
905 
906 static ssize_t intel_vgpu_read(struct vfio_device *vfio_dev, char __user *buf,
907 			size_t count, loff_t *ppos)
908 {
909 	struct intel_vgpu *vgpu = vfio_dev_to_vgpu(vfio_dev);
910 	unsigned int done = 0;
911 	int ret;
912 
913 	while (count) {
914 		size_t filled;
915 
916 		/* Only support GGTT entry 8 bytes read */
917 		if (count >= 8 && !(*ppos % 8) &&
918 			gtt_entry(vgpu, ppos)) {
919 			u64 val;
920 
921 			ret = intel_vgpu_rw(vgpu, (char *)&val, sizeof(val),
922 					ppos, false);
923 			if (ret <= 0)
924 				goto read_err;
925 
926 			if (copy_to_user(buf, &val, sizeof(val)))
927 				goto read_err;
928 
929 			filled = 8;
930 		} else if (count >= 4 && !(*ppos % 4)) {
931 			u32 val;
932 
933 			ret = intel_vgpu_rw(vgpu, (char *)&val, sizeof(val),
934 					ppos, false);
935 			if (ret <= 0)
936 				goto read_err;
937 
938 			if (copy_to_user(buf, &val, sizeof(val)))
939 				goto read_err;
940 
941 			filled = 4;
942 		} else if (count >= 2 && !(*ppos % 2)) {
943 			u16 val;
944 
945 			ret = intel_vgpu_rw(vgpu, (char *)&val, sizeof(val),
946 					ppos, false);
947 			if (ret <= 0)
948 				goto read_err;
949 
950 			if (copy_to_user(buf, &val, sizeof(val)))
951 				goto read_err;
952 
953 			filled = 2;
954 		} else {
955 			u8 val;
956 
957 			ret = intel_vgpu_rw(vgpu, &val, sizeof(val), ppos,
958 					false);
959 			if (ret <= 0)
960 				goto read_err;
961 
962 			if (copy_to_user(buf, &val, sizeof(val)))
963 				goto read_err;
964 
965 			filled = 1;
966 		}
967 
968 		count -= filled;
969 		done += filled;
970 		*ppos += filled;
971 		buf += filled;
972 	}
973 
974 	return done;
975 
976 read_err:
977 	return -EFAULT;
978 }
979 
980 static ssize_t intel_vgpu_write(struct vfio_device *vfio_dev,
981 				const char __user *buf,
982 				size_t count, loff_t *ppos)
983 {
984 	struct intel_vgpu *vgpu = vfio_dev_to_vgpu(vfio_dev);
985 	unsigned int done = 0;
986 	int ret;
987 
988 	while (count) {
989 		size_t filled;
990 
991 		/* Only support GGTT entry 8 bytes write */
992 		if (count >= 8 && !(*ppos % 8) &&
993 			gtt_entry(vgpu, ppos)) {
994 			u64 val;
995 
996 			if (copy_from_user(&val, buf, sizeof(val)))
997 				goto write_err;
998 
999 			ret = intel_vgpu_rw(vgpu, (char *)&val, sizeof(val),
1000 					ppos, true);
1001 			if (ret <= 0)
1002 				goto write_err;
1003 
1004 			filled = 8;
1005 		} else if (count >= 4 && !(*ppos % 4)) {
1006 			u32 val;
1007 
1008 			if (copy_from_user(&val, buf, sizeof(val)))
1009 				goto write_err;
1010 
1011 			ret = intel_vgpu_rw(vgpu, (char *)&val, sizeof(val),
1012 					ppos, true);
1013 			if (ret <= 0)
1014 				goto write_err;
1015 
1016 			filled = 4;
1017 		} else if (count >= 2 && !(*ppos % 2)) {
1018 			u16 val;
1019 
1020 			if (copy_from_user(&val, buf, sizeof(val)))
1021 				goto write_err;
1022 
1023 			ret = intel_vgpu_rw(vgpu, (char *)&val,
1024 					sizeof(val), ppos, true);
1025 			if (ret <= 0)
1026 				goto write_err;
1027 
1028 			filled = 2;
1029 		} else {
1030 			u8 val;
1031 
1032 			if (copy_from_user(&val, buf, sizeof(val)))
1033 				goto write_err;
1034 
1035 			ret = intel_vgpu_rw(vgpu, &val, sizeof(val),
1036 					ppos, true);
1037 			if (ret <= 0)
1038 				goto write_err;
1039 
1040 			filled = 1;
1041 		}
1042 
1043 		count -= filled;
1044 		done += filled;
1045 		*ppos += filled;
1046 		buf += filled;
1047 	}
1048 
1049 	return done;
1050 write_err:
1051 	return -EFAULT;
1052 }
1053 
1054 static int intel_vgpu_mmap(struct vfio_device *vfio_dev,
1055 		struct vm_area_struct *vma)
1056 {
1057 	struct intel_vgpu *vgpu = vfio_dev_to_vgpu(vfio_dev);
1058 	unsigned int index;
1059 	u64 virtaddr;
1060 	unsigned long req_size, pgoff, req_start;
1061 	pgprot_t pg_prot;
1062 
1063 	index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
1064 	if (index >= VFIO_PCI_ROM_REGION_INDEX)
1065 		return -EINVAL;
1066 
1067 	if (vma->vm_end < vma->vm_start)
1068 		return -EINVAL;
1069 	if ((vma->vm_flags & VM_SHARED) == 0)
1070 		return -EINVAL;
1071 	if (index != VFIO_PCI_BAR2_REGION_INDEX)
1072 		return -EINVAL;
1073 
1074 	pg_prot = vma->vm_page_prot;
1075 	virtaddr = vma->vm_start;
1076 	req_size = vma->vm_end - vma->vm_start;
1077 	pgoff = vma->vm_pgoff &
1078 		((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1);
1079 	req_start = pgoff << PAGE_SHIFT;
1080 
1081 	if (!intel_vgpu_in_aperture(vgpu, req_start))
1082 		return -EINVAL;
1083 	if (req_start + req_size >
1084 	    vgpu_aperture_offset(vgpu) + vgpu_aperture_sz(vgpu))
1085 		return -EINVAL;
1086 
1087 	pgoff = (gvt_aperture_pa_base(vgpu->gvt) >> PAGE_SHIFT) + pgoff;
1088 
1089 	return remap_pfn_range(vma, virtaddr, pgoff, req_size, pg_prot);
1090 }
1091 
1092 static int intel_vgpu_get_irq_count(struct intel_vgpu *vgpu, int type)
1093 {
1094 	if (type == VFIO_PCI_INTX_IRQ_INDEX || type == VFIO_PCI_MSI_IRQ_INDEX)
1095 		return 1;
1096 
1097 	return 0;
1098 }
1099 
1100 static int intel_vgpu_set_intx_mask(struct intel_vgpu *vgpu,
1101 			unsigned int index, unsigned int start,
1102 			unsigned int count, u32 flags,
1103 			void *data)
1104 {
1105 	return 0;
1106 }
1107 
1108 static int intel_vgpu_set_intx_unmask(struct intel_vgpu *vgpu,
1109 			unsigned int index, unsigned int start,
1110 			unsigned int count, u32 flags, void *data)
1111 {
1112 	return 0;
1113 }
1114 
1115 static int intel_vgpu_set_intx_trigger(struct intel_vgpu *vgpu,
1116 		unsigned int index, unsigned int start, unsigned int count,
1117 		u32 flags, void *data)
1118 {
1119 	return 0;
1120 }
1121 
1122 static int intel_vgpu_set_msi_trigger(struct intel_vgpu *vgpu,
1123 		unsigned int index, unsigned int start, unsigned int count,
1124 		u32 flags, void *data)
1125 {
1126 	struct eventfd_ctx *trigger;
1127 
1128 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
1129 		int fd = *(int *)data;
1130 
1131 		trigger = eventfd_ctx_fdget(fd);
1132 		if (IS_ERR(trigger)) {
1133 			gvt_vgpu_err("eventfd_ctx_fdget failed\n");
1134 			return PTR_ERR(trigger);
1135 		}
1136 		vgpu->msi_trigger = trigger;
1137 	} else if ((flags & VFIO_IRQ_SET_DATA_NONE) && !count)
1138 		intel_vgpu_release_msi_eventfd_ctx(vgpu);
1139 
1140 	return 0;
1141 }
1142 
1143 static int intel_vgpu_set_irqs(struct intel_vgpu *vgpu, u32 flags,
1144 		unsigned int index, unsigned int start, unsigned int count,
1145 		void *data)
1146 {
1147 	int (*func)(struct intel_vgpu *vgpu, unsigned int index,
1148 			unsigned int start, unsigned int count, u32 flags,
1149 			void *data) = NULL;
1150 
1151 	switch (index) {
1152 	case VFIO_PCI_INTX_IRQ_INDEX:
1153 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
1154 		case VFIO_IRQ_SET_ACTION_MASK:
1155 			func = intel_vgpu_set_intx_mask;
1156 			break;
1157 		case VFIO_IRQ_SET_ACTION_UNMASK:
1158 			func = intel_vgpu_set_intx_unmask;
1159 			break;
1160 		case VFIO_IRQ_SET_ACTION_TRIGGER:
1161 			func = intel_vgpu_set_intx_trigger;
1162 			break;
1163 		}
1164 		break;
1165 	case VFIO_PCI_MSI_IRQ_INDEX:
1166 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
1167 		case VFIO_IRQ_SET_ACTION_MASK:
1168 		case VFIO_IRQ_SET_ACTION_UNMASK:
1169 			/* XXX Need masking support exported */
1170 			break;
1171 		case VFIO_IRQ_SET_ACTION_TRIGGER:
1172 			func = intel_vgpu_set_msi_trigger;
1173 			break;
1174 		}
1175 		break;
1176 	}
1177 
1178 	if (!func)
1179 		return -ENOTTY;
1180 
1181 	return func(vgpu, index, start, count, flags, data);
1182 }
1183 
1184 static long intel_vgpu_ioctl(struct vfio_device *vfio_dev, unsigned int cmd,
1185 			     unsigned long arg)
1186 {
1187 	struct intel_vgpu *vgpu = vfio_dev_to_vgpu(vfio_dev);
1188 	unsigned long minsz;
1189 
1190 	gvt_dbg_core("vgpu%d ioctl, cmd: %d\n", vgpu->id, cmd);
1191 
1192 	if (cmd == VFIO_DEVICE_GET_INFO) {
1193 		struct vfio_device_info info;
1194 
1195 		minsz = offsetofend(struct vfio_device_info, num_irqs);
1196 
1197 		if (copy_from_user(&info, (void __user *)arg, minsz))
1198 			return -EFAULT;
1199 
1200 		if (info.argsz < minsz)
1201 			return -EINVAL;
1202 
1203 		info.flags = VFIO_DEVICE_FLAGS_PCI;
1204 		info.flags |= VFIO_DEVICE_FLAGS_RESET;
1205 		info.num_regions = VFIO_PCI_NUM_REGIONS +
1206 				vgpu->num_regions;
1207 		info.num_irqs = VFIO_PCI_NUM_IRQS;
1208 
1209 		return copy_to_user((void __user *)arg, &info, minsz) ?
1210 			-EFAULT : 0;
1211 
1212 	} else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
1213 		struct vfio_region_info info;
1214 		struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
1215 		unsigned int i;
1216 		int ret;
1217 		struct vfio_region_info_cap_sparse_mmap *sparse = NULL;
1218 		int nr_areas = 1;
1219 		int cap_type_id;
1220 
1221 		minsz = offsetofend(struct vfio_region_info, offset);
1222 
1223 		if (copy_from_user(&info, (void __user *)arg, minsz))
1224 			return -EFAULT;
1225 
1226 		if (info.argsz < minsz)
1227 			return -EINVAL;
1228 
1229 		switch (info.index) {
1230 		case VFIO_PCI_CONFIG_REGION_INDEX:
1231 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
1232 			info.size = vgpu->gvt->device_info.cfg_space_size;
1233 			info.flags = VFIO_REGION_INFO_FLAG_READ |
1234 				     VFIO_REGION_INFO_FLAG_WRITE;
1235 			break;
1236 		case VFIO_PCI_BAR0_REGION_INDEX:
1237 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
1238 			info.size = vgpu->cfg_space.bar[info.index].size;
1239 			if (!info.size) {
1240 				info.flags = 0;
1241 				break;
1242 			}
1243 
1244 			info.flags = VFIO_REGION_INFO_FLAG_READ |
1245 				     VFIO_REGION_INFO_FLAG_WRITE;
1246 			break;
1247 		case VFIO_PCI_BAR1_REGION_INDEX:
1248 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
1249 			info.size = 0;
1250 			info.flags = 0;
1251 			break;
1252 		case VFIO_PCI_BAR2_REGION_INDEX:
1253 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
1254 			info.flags = VFIO_REGION_INFO_FLAG_CAPS |
1255 					VFIO_REGION_INFO_FLAG_MMAP |
1256 					VFIO_REGION_INFO_FLAG_READ |
1257 					VFIO_REGION_INFO_FLAG_WRITE;
1258 			info.size = gvt_aperture_sz(vgpu->gvt);
1259 
1260 			sparse = kzalloc(struct_size(sparse, areas, nr_areas),
1261 					 GFP_KERNEL);
1262 			if (!sparse)
1263 				return -ENOMEM;
1264 
1265 			sparse->header.id = VFIO_REGION_INFO_CAP_SPARSE_MMAP;
1266 			sparse->header.version = 1;
1267 			sparse->nr_areas = nr_areas;
1268 			cap_type_id = VFIO_REGION_INFO_CAP_SPARSE_MMAP;
1269 			sparse->areas[0].offset =
1270 					PAGE_ALIGN(vgpu_aperture_offset(vgpu));
1271 			sparse->areas[0].size = vgpu_aperture_sz(vgpu);
1272 			break;
1273 
1274 		case VFIO_PCI_BAR3_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
1275 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
1276 			info.size = 0;
1277 			info.flags = 0;
1278 
1279 			gvt_dbg_core("get region info bar:%d\n", info.index);
1280 			break;
1281 
1282 		case VFIO_PCI_ROM_REGION_INDEX:
1283 		case VFIO_PCI_VGA_REGION_INDEX:
1284 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
1285 			info.size = 0;
1286 			info.flags = 0;
1287 
1288 			gvt_dbg_core("get region info index:%d\n", info.index);
1289 			break;
1290 		default:
1291 			{
1292 				struct vfio_region_info_cap_type cap_type = {
1293 					.header.id = VFIO_REGION_INFO_CAP_TYPE,
1294 					.header.version = 1 };
1295 
1296 				if (info.index >= VFIO_PCI_NUM_REGIONS +
1297 						vgpu->num_regions)
1298 					return -EINVAL;
1299 				info.index =
1300 					array_index_nospec(info.index,
1301 							VFIO_PCI_NUM_REGIONS +
1302 							vgpu->num_regions);
1303 
1304 				i = info.index - VFIO_PCI_NUM_REGIONS;
1305 
1306 				info.offset =
1307 					VFIO_PCI_INDEX_TO_OFFSET(info.index);
1308 				info.size = vgpu->region[i].size;
1309 				info.flags = vgpu->region[i].flags;
1310 
1311 				cap_type.type = vgpu->region[i].type;
1312 				cap_type.subtype = vgpu->region[i].subtype;
1313 
1314 				ret = vfio_info_add_capability(&caps,
1315 							&cap_type.header,
1316 							sizeof(cap_type));
1317 				if (ret)
1318 					return ret;
1319 			}
1320 		}
1321 
1322 		if ((info.flags & VFIO_REGION_INFO_FLAG_CAPS) && sparse) {
1323 			switch (cap_type_id) {
1324 			case VFIO_REGION_INFO_CAP_SPARSE_MMAP:
1325 				ret = vfio_info_add_capability(&caps,
1326 					&sparse->header,
1327 					struct_size(sparse, areas,
1328 						    sparse->nr_areas));
1329 				if (ret) {
1330 					kfree(sparse);
1331 					return ret;
1332 				}
1333 				break;
1334 			default:
1335 				kfree(sparse);
1336 				return -EINVAL;
1337 			}
1338 		}
1339 
1340 		if (caps.size) {
1341 			info.flags |= VFIO_REGION_INFO_FLAG_CAPS;
1342 			if (info.argsz < sizeof(info) + caps.size) {
1343 				info.argsz = sizeof(info) + caps.size;
1344 				info.cap_offset = 0;
1345 			} else {
1346 				vfio_info_cap_shift(&caps, sizeof(info));
1347 				if (copy_to_user((void __user *)arg +
1348 						  sizeof(info), caps.buf,
1349 						  caps.size)) {
1350 					kfree(caps.buf);
1351 					kfree(sparse);
1352 					return -EFAULT;
1353 				}
1354 				info.cap_offset = sizeof(info);
1355 			}
1356 
1357 			kfree(caps.buf);
1358 		}
1359 
1360 		kfree(sparse);
1361 		return copy_to_user((void __user *)arg, &info, minsz) ?
1362 			-EFAULT : 0;
1363 	} else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
1364 		struct vfio_irq_info info;
1365 
1366 		minsz = offsetofend(struct vfio_irq_info, count);
1367 
1368 		if (copy_from_user(&info, (void __user *)arg, minsz))
1369 			return -EFAULT;
1370 
1371 		if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
1372 			return -EINVAL;
1373 
1374 		switch (info.index) {
1375 		case VFIO_PCI_INTX_IRQ_INDEX:
1376 		case VFIO_PCI_MSI_IRQ_INDEX:
1377 			break;
1378 		default:
1379 			return -EINVAL;
1380 		}
1381 
1382 		info.flags = VFIO_IRQ_INFO_EVENTFD;
1383 
1384 		info.count = intel_vgpu_get_irq_count(vgpu, info.index);
1385 
1386 		if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
1387 			info.flags |= (VFIO_IRQ_INFO_MASKABLE |
1388 				       VFIO_IRQ_INFO_AUTOMASKED);
1389 		else
1390 			info.flags |= VFIO_IRQ_INFO_NORESIZE;
1391 
1392 		return copy_to_user((void __user *)arg, &info, minsz) ?
1393 			-EFAULT : 0;
1394 	} else if (cmd == VFIO_DEVICE_SET_IRQS) {
1395 		struct vfio_irq_set hdr;
1396 		u8 *data = NULL;
1397 		int ret = 0;
1398 		size_t data_size = 0;
1399 
1400 		minsz = offsetofend(struct vfio_irq_set, count);
1401 
1402 		if (copy_from_user(&hdr, (void __user *)arg, minsz))
1403 			return -EFAULT;
1404 
1405 		if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) {
1406 			int max = intel_vgpu_get_irq_count(vgpu, hdr.index);
1407 
1408 			ret = vfio_set_irqs_validate_and_prepare(&hdr, max,
1409 						VFIO_PCI_NUM_IRQS, &data_size);
1410 			if (ret) {
1411 				gvt_vgpu_err("intel:vfio_set_irqs_validate_and_prepare failed\n");
1412 				return -EINVAL;
1413 			}
1414 			if (data_size) {
1415 				data = memdup_user((void __user *)(arg + minsz),
1416 						   data_size);
1417 				if (IS_ERR(data))
1418 					return PTR_ERR(data);
1419 			}
1420 		}
1421 
1422 		ret = intel_vgpu_set_irqs(vgpu, hdr.flags, hdr.index,
1423 					hdr.start, hdr.count, data);
1424 		kfree(data);
1425 
1426 		return ret;
1427 	} else if (cmd == VFIO_DEVICE_RESET) {
1428 		intel_gvt_reset_vgpu(vgpu);
1429 		return 0;
1430 	} else if (cmd == VFIO_DEVICE_QUERY_GFX_PLANE) {
1431 		struct vfio_device_gfx_plane_info dmabuf;
1432 		int ret = 0;
1433 
1434 		minsz = offsetofend(struct vfio_device_gfx_plane_info,
1435 				    dmabuf_id);
1436 		if (copy_from_user(&dmabuf, (void __user *)arg, minsz))
1437 			return -EFAULT;
1438 		if (dmabuf.argsz < minsz)
1439 			return -EINVAL;
1440 
1441 		ret = intel_vgpu_query_plane(vgpu, &dmabuf);
1442 		if (ret != 0)
1443 			return ret;
1444 
1445 		return copy_to_user((void __user *)arg, &dmabuf, minsz) ?
1446 								-EFAULT : 0;
1447 	} else if (cmd == VFIO_DEVICE_GET_GFX_DMABUF) {
1448 		__u32 dmabuf_id;
1449 
1450 		if (get_user(dmabuf_id, (__u32 __user *)arg))
1451 			return -EFAULT;
1452 		return intel_vgpu_get_dmabuf(vgpu, dmabuf_id);
1453 	}
1454 
1455 	return -ENOTTY;
1456 }
1457 
1458 static ssize_t
1459 vgpu_id_show(struct device *dev, struct device_attribute *attr,
1460 	     char *buf)
1461 {
1462 	struct intel_vgpu *vgpu = dev_get_drvdata(dev);
1463 
1464 	return sprintf(buf, "%d\n", vgpu->id);
1465 }
1466 
1467 static DEVICE_ATTR_RO(vgpu_id);
1468 
1469 static struct attribute *intel_vgpu_attrs[] = {
1470 	&dev_attr_vgpu_id.attr,
1471 	NULL
1472 };
1473 
1474 static const struct attribute_group intel_vgpu_group = {
1475 	.name = "intel_vgpu",
1476 	.attrs = intel_vgpu_attrs,
1477 };
1478 
1479 static const struct attribute_group *intel_vgpu_groups[] = {
1480 	&intel_vgpu_group,
1481 	NULL,
1482 };
1483 
1484 static int intel_vgpu_init_dev(struct vfio_device *vfio_dev)
1485 {
1486 	struct mdev_device *mdev = to_mdev_device(vfio_dev->dev);
1487 	struct intel_vgpu *vgpu = vfio_dev_to_vgpu(vfio_dev);
1488 	struct intel_vgpu_type *type =
1489 		container_of(mdev->type, struct intel_vgpu_type, type);
1490 
1491 	vgpu->gvt = kdev_to_i915(mdev->type->parent->dev)->gvt;
1492 	return intel_gvt_create_vgpu(vgpu, type->conf);
1493 }
1494 
1495 static void intel_vgpu_release_dev(struct vfio_device *vfio_dev)
1496 {
1497 	struct intel_vgpu *vgpu = vfio_dev_to_vgpu(vfio_dev);
1498 
1499 	intel_gvt_destroy_vgpu(vgpu);
1500 	vfio_free_device(vfio_dev);
1501 }
1502 
1503 static const struct vfio_device_ops intel_vgpu_dev_ops = {
1504 	.init		= intel_vgpu_init_dev,
1505 	.release	= intel_vgpu_release_dev,
1506 	.open_device	= intel_vgpu_open_device,
1507 	.close_device	= intel_vgpu_close_device,
1508 	.read		= intel_vgpu_read,
1509 	.write		= intel_vgpu_write,
1510 	.mmap		= intel_vgpu_mmap,
1511 	.ioctl		= intel_vgpu_ioctl,
1512 	.dma_unmap	= intel_vgpu_dma_unmap,
1513 };
1514 
1515 static int intel_vgpu_probe(struct mdev_device *mdev)
1516 {
1517 	struct intel_vgpu *vgpu;
1518 	int ret;
1519 
1520 	vgpu = vfio_alloc_device(intel_vgpu, vfio_device, &mdev->dev,
1521 				 &intel_vgpu_dev_ops);
1522 	if (IS_ERR(vgpu)) {
1523 		gvt_err("failed to create intel vgpu: %ld\n", PTR_ERR(vgpu));
1524 		return PTR_ERR(vgpu);
1525 	}
1526 
1527 	dev_set_drvdata(&mdev->dev, vgpu);
1528 	ret = vfio_register_emulated_iommu_dev(&vgpu->vfio_device);
1529 	if (ret)
1530 		goto out_put_vdev;
1531 
1532 	gvt_dbg_core("intel_vgpu_create succeeded for mdev: %s\n",
1533 		     dev_name(mdev_dev(mdev)));
1534 	return 0;
1535 
1536 out_put_vdev:
1537 	vfio_put_device(&vgpu->vfio_device);
1538 	return ret;
1539 }
1540 
1541 static void intel_vgpu_remove(struct mdev_device *mdev)
1542 {
1543 	struct intel_vgpu *vgpu = dev_get_drvdata(&mdev->dev);
1544 
1545 	if (WARN_ON_ONCE(vgpu->attached))
1546 		return;
1547 
1548 	vfio_unregister_group_dev(&vgpu->vfio_device);
1549 	vfio_put_device(&vgpu->vfio_device);
1550 }
1551 
1552 static struct mdev_driver intel_vgpu_mdev_driver = {
1553 	.driver = {
1554 		.name		= "intel_vgpu_mdev",
1555 		.owner		= THIS_MODULE,
1556 		.dev_groups	= intel_vgpu_groups,
1557 	},
1558 	.probe		= intel_vgpu_probe,
1559 	.remove		= intel_vgpu_remove,
1560 	.types_attrs	= gvt_type_attrs,
1561 };
1562 
1563 int intel_gvt_page_track_add(struct intel_vgpu *info, u64 gfn)
1564 {
1565 	struct kvm *kvm = info->vfio_device.kvm;
1566 	struct kvm_memory_slot *slot;
1567 	int idx;
1568 
1569 	if (!info->attached)
1570 		return -ESRCH;
1571 
1572 	idx = srcu_read_lock(&kvm->srcu);
1573 	slot = gfn_to_memslot(kvm, gfn);
1574 	if (!slot) {
1575 		srcu_read_unlock(&kvm->srcu, idx);
1576 		return -EINVAL;
1577 	}
1578 
1579 	write_lock(&kvm->mmu_lock);
1580 
1581 	if (kvmgt_gfn_is_write_protected(info, gfn))
1582 		goto out;
1583 
1584 	kvm_slot_page_track_add_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1585 	kvmgt_protect_table_add(info, gfn);
1586 
1587 out:
1588 	write_unlock(&kvm->mmu_lock);
1589 	srcu_read_unlock(&kvm->srcu, idx);
1590 	return 0;
1591 }
1592 
1593 int intel_gvt_page_track_remove(struct intel_vgpu *info, u64 gfn)
1594 {
1595 	struct kvm *kvm = info->vfio_device.kvm;
1596 	struct kvm_memory_slot *slot;
1597 	int idx;
1598 
1599 	if (!info->attached)
1600 		return 0;
1601 
1602 	idx = srcu_read_lock(&kvm->srcu);
1603 	slot = gfn_to_memslot(kvm, gfn);
1604 	if (!slot) {
1605 		srcu_read_unlock(&kvm->srcu, idx);
1606 		return -EINVAL;
1607 	}
1608 
1609 	write_lock(&kvm->mmu_lock);
1610 
1611 	if (!kvmgt_gfn_is_write_protected(info, gfn))
1612 		goto out;
1613 
1614 	kvm_slot_page_track_remove_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1615 	kvmgt_protect_table_del(info, gfn);
1616 
1617 out:
1618 	write_unlock(&kvm->mmu_lock);
1619 	srcu_read_unlock(&kvm->srcu, idx);
1620 	return 0;
1621 }
1622 
1623 static void kvmgt_page_track_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1624 		const u8 *val, int len,
1625 		struct kvm_page_track_notifier_node *node)
1626 {
1627 	struct intel_vgpu *info =
1628 		container_of(node, struct intel_vgpu, track_node);
1629 
1630 	if (kvmgt_gfn_is_write_protected(info, gpa_to_gfn(gpa)))
1631 		intel_vgpu_page_track_handler(info, gpa,
1632 						     (void *)val, len);
1633 }
1634 
1635 static void kvmgt_page_track_flush_slot(struct kvm *kvm,
1636 		struct kvm_memory_slot *slot,
1637 		struct kvm_page_track_notifier_node *node)
1638 {
1639 	int i;
1640 	gfn_t gfn;
1641 	struct intel_vgpu *info =
1642 		container_of(node, struct intel_vgpu, track_node);
1643 
1644 	write_lock(&kvm->mmu_lock);
1645 	for (i = 0; i < slot->npages; i++) {
1646 		gfn = slot->base_gfn + i;
1647 		if (kvmgt_gfn_is_write_protected(info, gfn)) {
1648 			kvm_slot_page_track_remove_page(kvm, slot, gfn,
1649 						KVM_PAGE_TRACK_WRITE);
1650 			kvmgt_protect_table_del(info, gfn);
1651 		}
1652 	}
1653 	write_unlock(&kvm->mmu_lock);
1654 }
1655 
1656 void intel_vgpu_detach_regions(struct intel_vgpu *vgpu)
1657 {
1658 	int i;
1659 
1660 	if (!vgpu->region)
1661 		return;
1662 
1663 	for (i = 0; i < vgpu->num_regions; i++)
1664 		if (vgpu->region[i].ops->release)
1665 			vgpu->region[i].ops->release(vgpu,
1666 					&vgpu->region[i]);
1667 	vgpu->num_regions = 0;
1668 	kfree(vgpu->region);
1669 	vgpu->region = NULL;
1670 }
1671 
1672 int intel_gvt_dma_map_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
1673 		unsigned long size, dma_addr_t *dma_addr)
1674 {
1675 	struct gvt_dma *entry;
1676 	int ret;
1677 
1678 	if (!vgpu->attached)
1679 		return -EINVAL;
1680 
1681 	mutex_lock(&vgpu->cache_lock);
1682 
1683 	entry = __gvt_cache_find_gfn(vgpu, gfn);
1684 	if (!entry) {
1685 		ret = gvt_dma_map_page(vgpu, gfn, dma_addr, size);
1686 		if (ret)
1687 			goto err_unlock;
1688 
1689 		ret = __gvt_cache_add(vgpu, gfn, *dma_addr, size);
1690 		if (ret)
1691 			goto err_unmap;
1692 	} else if (entry->size != size) {
1693 		/* the same gfn with different size: unmap and re-map */
1694 		gvt_dma_unmap_page(vgpu, gfn, entry->dma_addr, entry->size);
1695 		__gvt_cache_remove_entry(vgpu, entry);
1696 
1697 		ret = gvt_dma_map_page(vgpu, gfn, dma_addr, size);
1698 		if (ret)
1699 			goto err_unlock;
1700 
1701 		ret = __gvt_cache_add(vgpu, gfn, *dma_addr, size);
1702 		if (ret)
1703 			goto err_unmap;
1704 	} else {
1705 		kref_get(&entry->ref);
1706 		*dma_addr = entry->dma_addr;
1707 	}
1708 
1709 	mutex_unlock(&vgpu->cache_lock);
1710 	return 0;
1711 
1712 err_unmap:
1713 	gvt_dma_unmap_page(vgpu, gfn, *dma_addr, size);
1714 err_unlock:
1715 	mutex_unlock(&vgpu->cache_lock);
1716 	return ret;
1717 }
1718 
1719 int intel_gvt_dma_pin_guest_page(struct intel_vgpu *vgpu, dma_addr_t dma_addr)
1720 {
1721 	struct gvt_dma *entry;
1722 	int ret = 0;
1723 
1724 	if (!vgpu->attached)
1725 		return -ENODEV;
1726 
1727 	mutex_lock(&vgpu->cache_lock);
1728 	entry = __gvt_cache_find_dma_addr(vgpu, dma_addr);
1729 	if (entry)
1730 		kref_get(&entry->ref);
1731 	else
1732 		ret = -ENOMEM;
1733 	mutex_unlock(&vgpu->cache_lock);
1734 
1735 	return ret;
1736 }
1737 
1738 static void __gvt_dma_release(struct kref *ref)
1739 {
1740 	struct gvt_dma *entry = container_of(ref, typeof(*entry), ref);
1741 
1742 	gvt_dma_unmap_page(entry->vgpu, entry->gfn, entry->dma_addr,
1743 			   entry->size);
1744 	__gvt_cache_remove_entry(entry->vgpu, entry);
1745 }
1746 
1747 void intel_gvt_dma_unmap_guest_page(struct intel_vgpu *vgpu,
1748 		dma_addr_t dma_addr)
1749 {
1750 	struct gvt_dma *entry;
1751 
1752 	if (!vgpu->attached)
1753 		return;
1754 
1755 	mutex_lock(&vgpu->cache_lock);
1756 	entry = __gvt_cache_find_dma_addr(vgpu, dma_addr);
1757 	if (entry)
1758 		kref_put(&entry->ref, __gvt_dma_release);
1759 	mutex_unlock(&vgpu->cache_lock);
1760 }
1761 
1762 static void init_device_info(struct intel_gvt *gvt)
1763 {
1764 	struct intel_gvt_device_info *info = &gvt->device_info;
1765 	struct pci_dev *pdev = to_pci_dev(gvt->gt->i915->drm.dev);
1766 
1767 	info->max_support_vgpus = 8;
1768 	info->cfg_space_size = PCI_CFG_SPACE_EXP_SIZE;
1769 	info->mmio_size = 2 * 1024 * 1024;
1770 	info->mmio_bar = 0;
1771 	info->gtt_start_offset = 8 * 1024 * 1024;
1772 	info->gtt_entry_size = 8;
1773 	info->gtt_entry_size_shift = 3;
1774 	info->gmadr_bytes_in_cmd = 8;
1775 	info->max_surface_size = 36 * 1024 * 1024;
1776 	info->msi_cap_offset = pdev->msi_cap;
1777 }
1778 
1779 static void intel_gvt_test_and_emulate_vblank(struct intel_gvt *gvt)
1780 {
1781 	struct intel_vgpu *vgpu;
1782 	int id;
1783 
1784 	mutex_lock(&gvt->lock);
1785 	idr_for_each_entry((&(gvt)->vgpu_idr), (vgpu), (id)) {
1786 		if (test_and_clear_bit(INTEL_GVT_REQUEST_EMULATE_VBLANK + id,
1787 				       (void *)&gvt->service_request)) {
1788 			if (vgpu->active)
1789 				intel_vgpu_emulate_vblank(vgpu);
1790 		}
1791 	}
1792 	mutex_unlock(&gvt->lock);
1793 }
1794 
1795 static int gvt_service_thread(void *data)
1796 {
1797 	struct intel_gvt *gvt = (struct intel_gvt *)data;
1798 	int ret;
1799 
1800 	gvt_dbg_core("service thread start\n");
1801 
1802 	while (!kthread_should_stop()) {
1803 		ret = wait_event_interruptible(gvt->service_thread_wq,
1804 				kthread_should_stop() || gvt->service_request);
1805 
1806 		if (kthread_should_stop())
1807 			break;
1808 
1809 		if (WARN_ONCE(ret, "service thread is waken up by signal.\n"))
1810 			continue;
1811 
1812 		intel_gvt_test_and_emulate_vblank(gvt);
1813 
1814 		if (test_bit(INTEL_GVT_REQUEST_SCHED,
1815 				(void *)&gvt->service_request) ||
1816 			test_bit(INTEL_GVT_REQUEST_EVENT_SCHED,
1817 					(void *)&gvt->service_request)) {
1818 			intel_gvt_schedule(gvt);
1819 		}
1820 	}
1821 
1822 	return 0;
1823 }
1824 
1825 static void clean_service_thread(struct intel_gvt *gvt)
1826 {
1827 	kthread_stop(gvt->service_thread);
1828 }
1829 
1830 static int init_service_thread(struct intel_gvt *gvt)
1831 {
1832 	init_waitqueue_head(&gvt->service_thread_wq);
1833 
1834 	gvt->service_thread = kthread_run(gvt_service_thread,
1835 			gvt, "gvt_service_thread");
1836 	if (IS_ERR(gvt->service_thread)) {
1837 		gvt_err("fail to start service thread.\n");
1838 		return PTR_ERR(gvt->service_thread);
1839 	}
1840 	return 0;
1841 }
1842 
1843 /**
1844  * intel_gvt_clean_device - clean a GVT device
1845  * @i915: i915 private
1846  *
1847  * This function is called at the driver unloading stage, to free the
1848  * resources owned by a GVT device.
1849  *
1850  */
1851 static void intel_gvt_clean_device(struct drm_i915_private *i915)
1852 {
1853 	struct intel_gvt *gvt = fetch_and_zero(&i915->gvt);
1854 
1855 	if (drm_WARN_ON(&i915->drm, !gvt))
1856 		return;
1857 
1858 	mdev_unregister_parent(&gvt->parent);
1859 	intel_gvt_destroy_idle_vgpu(gvt->idle_vgpu);
1860 	intel_gvt_clean_vgpu_types(gvt);
1861 
1862 	intel_gvt_debugfs_clean(gvt);
1863 	clean_service_thread(gvt);
1864 	intel_gvt_clean_cmd_parser(gvt);
1865 	intel_gvt_clean_sched_policy(gvt);
1866 	intel_gvt_clean_workload_scheduler(gvt);
1867 	intel_gvt_clean_gtt(gvt);
1868 	intel_gvt_free_firmware(gvt);
1869 	intel_gvt_clean_mmio_info(gvt);
1870 	idr_destroy(&gvt->vgpu_idr);
1871 
1872 	kfree(i915->gvt);
1873 }
1874 
1875 /**
1876  * intel_gvt_init_device - initialize a GVT device
1877  * @i915: drm i915 private data
1878  *
1879  * This function is called at the initialization stage, to initialize
1880  * necessary GVT components.
1881  *
1882  * Returns:
1883  * Zero on success, negative error code if failed.
1884  *
1885  */
1886 static int intel_gvt_init_device(struct drm_i915_private *i915)
1887 {
1888 	struct intel_gvt *gvt;
1889 	struct intel_vgpu *vgpu;
1890 	int ret;
1891 
1892 	if (drm_WARN_ON(&i915->drm, i915->gvt))
1893 		return -EEXIST;
1894 
1895 	gvt = kzalloc(sizeof(struct intel_gvt), GFP_KERNEL);
1896 	if (!gvt)
1897 		return -ENOMEM;
1898 
1899 	gvt_dbg_core("init gvt device\n");
1900 
1901 	idr_init_base(&gvt->vgpu_idr, 1);
1902 	spin_lock_init(&gvt->scheduler.mmio_context_lock);
1903 	mutex_init(&gvt->lock);
1904 	mutex_init(&gvt->sched_lock);
1905 	gvt->gt = to_gt(i915);
1906 	i915->gvt = gvt;
1907 
1908 	init_device_info(gvt);
1909 
1910 	ret = intel_gvt_setup_mmio_info(gvt);
1911 	if (ret)
1912 		goto out_clean_idr;
1913 
1914 	intel_gvt_init_engine_mmio_context(gvt);
1915 
1916 	ret = intel_gvt_load_firmware(gvt);
1917 	if (ret)
1918 		goto out_clean_mmio_info;
1919 
1920 	ret = intel_gvt_init_irq(gvt);
1921 	if (ret)
1922 		goto out_free_firmware;
1923 
1924 	ret = intel_gvt_init_gtt(gvt);
1925 	if (ret)
1926 		goto out_free_firmware;
1927 
1928 	ret = intel_gvt_init_workload_scheduler(gvt);
1929 	if (ret)
1930 		goto out_clean_gtt;
1931 
1932 	ret = intel_gvt_init_sched_policy(gvt);
1933 	if (ret)
1934 		goto out_clean_workload_scheduler;
1935 
1936 	ret = intel_gvt_init_cmd_parser(gvt);
1937 	if (ret)
1938 		goto out_clean_sched_policy;
1939 
1940 	ret = init_service_thread(gvt);
1941 	if (ret)
1942 		goto out_clean_cmd_parser;
1943 
1944 	ret = intel_gvt_init_vgpu_types(gvt);
1945 	if (ret)
1946 		goto out_clean_thread;
1947 
1948 	vgpu = intel_gvt_create_idle_vgpu(gvt);
1949 	if (IS_ERR(vgpu)) {
1950 		ret = PTR_ERR(vgpu);
1951 		gvt_err("failed to create idle vgpu\n");
1952 		goto out_clean_types;
1953 	}
1954 	gvt->idle_vgpu = vgpu;
1955 
1956 	intel_gvt_debugfs_init(gvt);
1957 
1958 	ret = mdev_register_parent(&gvt->parent, i915->drm.dev,
1959 				   &intel_vgpu_mdev_driver,
1960 				   gvt->mdev_types, gvt->num_types);
1961 	if (ret)
1962 		goto out_destroy_idle_vgpu;
1963 
1964 	gvt_dbg_core("gvt device initialization is done\n");
1965 	return 0;
1966 
1967 out_destroy_idle_vgpu:
1968 	intel_gvt_destroy_idle_vgpu(gvt->idle_vgpu);
1969 	intel_gvt_debugfs_clean(gvt);
1970 out_clean_types:
1971 	intel_gvt_clean_vgpu_types(gvt);
1972 out_clean_thread:
1973 	clean_service_thread(gvt);
1974 out_clean_cmd_parser:
1975 	intel_gvt_clean_cmd_parser(gvt);
1976 out_clean_sched_policy:
1977 	intel_gvt_clean_sched_policy(gvt);
1978 out_clean_workload_scheduler:
1979 	intel_gvt_clean_workload_scheduler(gvt);
1980 out_clean_gtt:
1981 	intel_gvt_clean_gtt(gvt);
1982 out_free_firmware:
1983 	intel_gvt_free_firmware(gvt);
1984 out_clean_mmio_info:
1985 	intel_gvt_clean_mmio_info(gvt);
1986 out_clean_idr:
1987 	idr_destroy(&gvt->vgpu_idr);
1988 	kfree(gvt);
1989 	i915->gvt = NULL;
1990 	return ret;
1991 }
1992 
1993 static void intel_gvt_pm_resume(struct drm_i915_private *i915)
1994 {
1995 	struct intel_gvt *gvt = i915->gvt;
1996 
1997 	intel_gvt_restore_fence(gvt);
1998 	intel_gvt_restore_mmio(gvt);
1999 	intel_gvt_restore_ggtt(gvt);
2000 }
2001 
2002 static const struct intel_vgpu_ops intel_gvt_vgpu_ops = {
2003 	.init_device	= intel_gvt_init_device,
2004 	.clean_device	= intel_gvt_clean_device,
2005 	.pm_resume	= intel_gvt_pm_resume,
2006 };
2007 
2008 static int __init kvmgt_init(void)
2009 {
2010 	int ret;
2011 
2012 	ret = intel_gvt_set_ops(&intel_gvt_vgpu_ops);
2013 	if (ret)
2014 		return ret;
2015 
2016 	ret = mdev_register_driver(&intel_vgpu_mdev_driver);
2017 	if (ret)
2018 		intel_gvt_clear_ops(&intel_gvt_vgpu_ops);
2019 	return ret;
2020 }
2021 
2022 static void __exit kvmgt_exit(void)
2023 {
2024 	mdev_unregister_driver(&intel_vgpu_mdev_driver);
2025 	intel_gvt_clear_ops(&intel_gvt_vgpu_ops);
2026 }
2027 
2028 module_init(kvmgt_init);
2029 module_exit(kvmgt_exit);
2030 
2031 MODULE_LICENSE("GPL and additional rights");
2032 MODULE_AUTHOR("Intel Corporation");
2033