xref: /openbmc/linux/drivers/gpu/drm/i915/gvt/kvmgt.c (revision 1c2dd16a)
1 /*
2  * KVMGT - the implementation of Intel mediated pass-through framework for KVM
3  *
4  * Copyright(c) 2014-2016 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  */
30 
31 #include <linux/init.h>
32 #include <linux/device.h>
33 #include <linux/mm.h>
34 #include <linux/mmu_context.h>
35 #include <linux/types.h>
36 #include <linux/list.h>
37 #include <linux/rbtree.h>
38 #include <linux/spinlock.h>
39 #include <linux/eventfd.h>
40 #include <linux/uuid.h>
41 #include <linux/kvm_host.h>
42 #include <linux/vfio.h>
43 #include <linux/mdev.h>
44 
45 #include "i915_drv.h"
46 #include "gvt.h"
47 
48 static const struct intel_gvt_ops *intel_gvt_ops;
49 
50 /* helper macros copied from vfio-pci */
51 #define VFIO_PCI_OFFSET_SHIFT   40
52 #define VFIO_PCI_OFFSET_TO_INDEX(off)   (off >> VFIO_PCI_OFFSET_SHIFT)
53 #define VFIO_PCI_INDEX_TO_OFFSET(index) ((u64)(index) << VFIO_PCI_OFFSET_SHIFT)
54 #define VFIO_PCI_OFFSET_MASK    (((u64)(1) << VFIO_PCI_OFFSET_SHIFT) - 1)
55 
56 struct vfio_region {
57 	u32				type;
58 	u32				subtype;
59 	size_t				size;
60 	u32				flags;
61 };
62 
63 struct kvmgt_pgfn {
64 	gfn_t gfn;
65 	struct hlist_node hnode;
66 };
67 
68 struct kvmgt_guest_info {
69 	struct kvm *kvm;
70 	struct intel_vgpu *vgpu;
71 	struct kvm_page_track_notifier_node track_node;
72 #define NR_BKT (1 << 18)
73 	struct hlist_head ptable[NR_BKT];
74 #undef NR_BKT
75 };
76 
77 struct gvt_dma {
78 	struct rb_node node;
79 	gfn_t gfn;
80 	unsigned long iova;
81 };
82 
83 static inline bool handle_valid(unsigned long handle)
84 {
85 	return !!(handle & ~0xff);
86 }
87 
88 static int kvmgt_guest_init(struct mdev_device *mdev);
89 static void intel_vgpu_release_work(struct work_struct *work);
90 static bool kvmgt_guest_exit(struct kvmgt_guest_info *info);
91 
92 static int gvt_dma_map_iova(struct intel_vgpu *vgpu, kvm_pfn_t pfn,
93 		unsigned long *iova)
94 {
95 	struct page *page;
96 	struct device *dev = &vgpu->gvt->dev_priv->drm.pdev->dev;
97 	dma_addr_t daddr;
98 
99 	if (unlikely(!pfn_valid(pfn)))
100 		return -EFAULT;
101 
102 	page = pfn_to_page(pfn);
103 	daddr = dma_map_page(dev, page, 0, PAGE_SIZE,
104 			PCI_DMA_BIDIRECTIONAL);
105 	if (dma_mapping_error(dev, daddr))
106 		return -ENOMEM;
107 
108 	*iova = (unsigned long)(daddr >> PAGE_SHIFT);
109 	return 0;
110 }
111 
112 static void gvt_dma_unmap_iova(struct intel_vgpu *vgpu, unsigned long iova)
113 {
114 	struct device *dev = &vgpu->gvt->dev_priv->drm.pdev->dev;
115 	dma_addr_t daddr;
116 
117 	daddr = (dma_addr_t)(iova << PAGE_SHIFT);
118 	dma_unmap_page(dev, daddr, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
119 }
120 
121 static struct gvt_dma *__gvt_cache_find(struct intel_vgpu *vgpu, gfn_t gfn)
122 {
123 	struct rb_node *node = vgpu->vdev.cache.rb_node;
124 	struct gvt_dma *ret = NULL;
125 
126 	while (node) {
127 		struct gvt_dma *itr = rb_entry(node, struct gvt_dma, node);
128 
129 		if (gfn < itr->gfn)
130 			node = node->rb_left;
131 		else if (gfn > itr->gfn)
132 			node = node->rb_right;
133 		else {
134 			ret = itr;
135 			goto out;
136 		}
137 	}
138 
139 out:
140 	return ret;
141 }
142 
143 static unsigned long gvt_cache_find(struct intel_vgpu *vgpu, gfn_t gfn)
144 {
145 	struct gvt_dma *entry;
146 	unsigned long iova;
147 
148 	mutex_lock(&vgpu->vdev.cache_lock);
149 
150 	entry = __gvt_cache_find(vgpu, gfn);
151 	iova = (entry == NULL) ? INTEL_GVT_INVALID_ADDR : entry->iova;
152 
153 	mutex_unlock(&vgpu->vdev.cache_lock);
154 	return iova;
155 }
156 
157 static void gvt_cache_add(struct intel_vgpu *vgpu, gfn_t gfn,
158 		unsigned long iova)
159 {
160 	struct gvt_dma *new, *itr;
161 	struct rb_node **link = &vgpu->vdev.cache.rb_node, *parent = NULL;
162 
163 	new = kzalloc(sizeof(struct gvt_dma), GFP_KERNEL);
164 	if (!new)
165 		return;
166 
167 	new->gfn = gfn;
168 	new->iova = iova;
169 
170 	mutex_lock(&vgpu->vdev.cache_lock);
171 	while (*link) {
172 		parent = *link;
173 		itr = rb_entry(parent, struct gvt_dma, node);
174 
175 		if (gfn == itr->gfn)
176 			goto out;
177 		else if (gfn < itr->gfn)
178 			link = &parent->rb_left;
179 		else
180 			link = &parent->rb_right;
181 	}
182 
183 	rb_link_node(&new->node, parent, link);
184 	rb_insert_color(&new->node, &vgpu->vdev.cache);
185 	mutex_unlock(&vgpu->vdev.cache_lock);
186 	return;
187 
188 out:
189 	mutex_unlock(&vgpu->vdev.cache_lock);
190 	kfree(new);
191 }
192 
193 static void __gvt_cache_remove_entry(struct intel_vgpu *vgpu,
194 				struct gvt_dma *entry)
195 {
196 	rb_erase(&entry->node, &vgpu->vdev.cache);
197 	kfree(entry);
198 }
199 
200 static void gvt_cache_remove(struct intel_vgpu *vgpu, gfn_t gfn)
201 {
202 	struct device *dev = mdev_dev(vgpu->vdev.mdev);
203 	struct gvt_dma *this;
204 	unsigned long g1;
205 	int rc;
206 
207 	mutex_lock(&vgpu->vdev.cache_lock);
208 	this  = __gvt_cache_find(vgpu, gfn);
209 	if (!this) {
210 		mutex_unlock(&vgpu->vdev.cache_lock);
211 		return;
212 	}
213 
214 	g1 = gfn;
215 	gvt_dma_unmap_iova(vgpu, this->iova);
216 	rc = vfio_unpin_pages(dev, &g1, 1);
217 	WARN_ON(rc != 1);
218 	__gvt_cache_remove_entry(vgpu, this);
219 	mutex_unlock(&vgpu->vdev.cache_lock);
220 }
221 
222 static void gvt_cache_init(struct intel_vgpu *vgpu)
223 {
224 	vgpu->vdev.cache = RB_ROOT;
225 	mutex_init(&vgpu->vdev.cache_lock);
226 }
227 
228 static void gvt_cache_destroy(struct intel_vgpu *vgpu)
229 {
230 	struct gvt_dma *dma;
231 	struct rb_node *node = NULL;
232 	struct device *dev = mdev_dev(vgpu->vdev.mdev);
233 	unsigned long gfn;
234 
235 	mutex_lock(&vgpu->vdev.cache_lock);
236 	while ((node = rb_first(&vgpu->vdev.cache))) {
237 		dma = rb_entry(node, struct gvt_dma, node);
238 		gvt_dma_unmap_iova(vgpu, dma->iova);
239 		gfn = dma->gfn;
240 
241 		vfio_unpin_pages(dev, &gfn, 1);
242 		__gvt_cache_remove_entry(vgpu, dma);
243 	}
244 	mutex_unlock(&vgpu->vdev.cache_lock);
245 }
246 
247 static struct intel_vgpu_type *intel_gvt_find_vgpu_type(struct intel_gvt *gvt,
248 		const char *name)
249 {
250 	int i;
251 	struct intel_vgpu_type *t;
252 	const char *driver_name = dev_driver_string(
253 			&gvt->dev_priv->drm.pdev->dev);
254 
255 	for (i = 0; i < gvt->num_types; i++) {
256 		t = &gvt->types[i];
257 		if (!strncmp(t->name, name + strlen(driver_name) + 1,
258 			sizeof(t->name)))
259 			return t;
260 	}
261 
262 	return NULL;
263 }
264 
265 static ssize_t available_instances_show(struct kobject *kobj,
266 					struct device *dev, char *buf)
267 {
268 	struct intel_vgpu_type *type;
269 	unsigned int num = 0;
270 	void *gvt = kdev_to_i915(dev)->gvt;
271 
272 	type = intel_gvt_find_vgpu_type(gvt, kobject_name(kobj));
273 	if (!type)
274 		num = 0;
275 	else
276 		num = type->avail_instance;
277 
278 	return sprintf(buf, "%u\n", num);
279 }
280 
281 static ssize_t device_api_show(struct kobject *kobj, struct device *dev,
282 		char *buf)
283 {
284 	return sprintf(buf, "%s\n", VFIO_DEVICE_API_PCI_STRING);
285 }
286 
287 static ssize_t description_show(struct kobject *kobj, struct device *dev,
288 		char *buf)
289 {
290 	struct intel_vgpu_type *type;
291 	void *gvt = kdev_to_i915(dev)->gvt;
292 
293 	type = intel_gvt_find_vgpu_type(gvt, kobject_name(kobj));
294 	if (!type)
295 		return 0;
296 
297 	return sprintf(buf, "low_gm_size: %dMB\nhigh_gm_size: %dMB\n"
298 		       "fence: %d\nresolution: %s\n",
299 		       BYTES_TO_MB(type->low_gm_size),
300 		       BYTES_TO_MB(type->high_gm_size),
301 		       type->fence, vgpu_edid_str(type->resolution));
302 }
303 
304 static MDEV_TYPE_ATTR_RO(available_instances);
305 static MDEV_TYPE_ATTR_RO(device_api);
306 static MDEV_TYPE_ATTR_RO(description);
307 
308 static struct attribute *type_attrs[] = {
309 	&mdev_type_attr_available_instances.attr,
310 	&mdev_type_attr_device_api.attr,
311 	&mdev_type_attr_description.attr,
312 	NULL,
313 };
314 
315 static struct attribute_group *intel_vgpu_type_groups[] = {
316 	[0 ... NR_MAX_INTEL_VGPU_TYPES - 1] = NULL,
317 };
318 
319 static bool intel_gvt_init_vgpu_type_groups(struct intel_gvt *gvt)
320 {
321 	int i, j;
322 	struct intel_vgpu_type *type;
323 	struct attribute_group *group;
324 
325 	for (i = 0; i < gvt->num_types; i++) {
326 		type = &gvt->types[i];
327 
328 		group = kzalloc(sizeof(struct attribute_group), GFP_KERNEL);
329 		if (WARN_ON(!group))
330 			goto unwind;
331 
332 		group->name = type->name;
333 		group->attrs = type_attrs;
334 		intel_vgpu_type_groups[i] = group;
335 	}
336 
337 	return true;
338 
339 unwind:
340 	for (j = 0; j < i; j++) {
341 		group = intel_vgpu_type_groups[j];
342 		kfree(group);
343 	}
344 
345 	return false;
346 }
347 
348 static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt)
349 {
350 	int i;
351 	struct attribute_group *group;
352 
353 	for (i = 0; i < gvt->num_types; i++) {
354 		group = intel_vgpu_type_groups[i];
355 		kfree(group);
356 	}
357 }
358 
359 static void kvmgt_protect_table_init(struct kvmgt_guest_info *info)
360 {
361 	hash_init(info->ptable);
362 }
363 
364 static void kvmgt_protect_table_destroy(struct kvmgt_guest_info *info)
365 {
366 	struct kvmgt_pgfn *p;
367 	struct hlist_node *tmp;
368 	int i;
369 
370 	hash_for_each_safe(info->ptable, i, tmp, p, hnode) {
371 		hash_del(&p->hnode);
372 		kfree(p);
373 	}
374 }
375 
376 static struct kvmgt_pgfn *
377 __kvmgt_protect_table_find(struct kvmgt_guest_info *info, gfn_t gfn)
378 {
379 	struct kvmgt_pgfn *p, *res = NULL;
380 
381 	hash_for_each_possible(info->ptable, p, hnode, gfn) {
382 		if (gfn == p->gfn) {
383 			res = p;
384 			break;
385 		}
386 	}
387 
388 	return res;
389 }
390 
391 static bool kvmgt_gfn_is_write_protected(struct kvmgt_guest_info *info,
392 				gfn_t gfn)
393 {
394 	struct kvmgt_pgfn *p;
395 
396 	p = __kvmgt_protect_table_find(info, gfn);
397 	return !!p;
398 }
399 
400 static void kvmgt_protect_table_add(struct kvmgt_guest_info *info, gfn_t gfn)
401 {
402 	struct kvmgt_pgfn *p;
403 
404 	if (kvmgt_gfn_is_write_protected(info, gfn))
405 		return;
406 
407 	p = kzalloc(sizeof(struct kvmgt_pgfn), GFP_ATOMIC);
408 	if (WARN(!p, "gfn: 0x%llx\n", gfn))
409 		return;
410 
411 	p->gfn = gfn;
412 	hash_add(info->ptable, &p->hnode, gfn);
413 }
414 
415 static void kvmgt_protect_table_del(struct kvmgt_guest_info *info,
416 				gfn_t gfn)
417 {
418 	struct kvmgt_pgfn *p;
419 
420 	p = __kvmgt_protect_table_find(info, gfn);
421 	if (p) {
422 		hash_del(&p->hnode);
423 		kfree(p);
424 	}
425 }
426 
427 static int intel_vgpu_create(struct kobject *kobj, struct mdev_device *mdev)
428 {
429 	struct intel_vgpu *vgpu = NULL;
430 	struct intel_vgpu_type *type;
431 	struct device *pdev;
432 	void *gvt;
433 	int ret;
434 
435 	pdev = mdev_parent_dev(mdev);
436 	gvt = kdev_to_i915(pdev)->gvt;
437 
438 	type = intel_gvt_find_vgpu_type(gvt, kobject_name(kobj));
439 	if (!type) {
440 		gvt_vgpu_err("failed to find type %s to create\n",
441 						kobject_name(kobj));
442 		ret = -EINVAL;
443 		goto out;
444 	}
445 
446 	vgpu = intel_gvt_ops->vgpu_create(gvt, type);
447 	if (IS_ERR_OR_NULL(vgpu)) {
448 		ret = vgpu == NULL ? -EFAULT : PTR_ERR(vgpu);
449 		gvt_vgpu_err("failed to create intel vgpu: %d\n", ret);
450 		goto out;
451 	}
452 
453 	INIT_WORK(&vgpu->vdev.release_work, intel_vgpu_release_work);
454 
455 	vgpu->vdev.mdev = mdev;
456 	mdev_set_drvdata(mdev, vgpu);
457 
458 	gvt_dbg_core("intel_vgpu_create succeeded for mdev: %s\n",
459 		     dev_name(mdev_dev(mdev)));
460 	ret = 0;
461 
462 out:
463 	return ret;
464 }
465 
466 static int intel_vgpu_remove(struct mdev_device *mdev)
467 {
468 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
469 
470 	if (handle_valid(vgpu->handle))
471 		return -EBUSY;
472 
473 	intel_gvt_ops->vgpu_destroy(vgpu);
474 	return 0;
475 }
476 
477 static int intel_vgpu_iommu_notifier(struct notifier_block *nb,
478 				     unsigned long action, void *data)
479 {
480 	struct intel_vgpu *vgpu = container_of(nb,
481 					struct intel_vgpu,
482 					vdev.iommu_notifier);
483 
484 	if (action == VFIO_IOMMU_NOTIFY_DMA_UNMAP) {
485 		struct vfio_iommu_type1_dma_unmap *unmap = data;
486 		unsigned long gfn, end_gfn;
487 
488 		gfn = unmap->iova >> PAGE_SHIFT;
489 		end_gfn = gfn + unmap->size / PAGE_SIZE;
490 
491 		while (gfn < end_gfn)
492 			gvt_cache_remove(vgpu, gfn++);
493 	}
494 
495 	return NOTIFY_OK;
496 }
497 
498 static int intel_vgpu_group_notifier(struct notifier_block *nb,
499 				     unsigned long action, void *data)
500 {
501 	struct intel_vgpu *vgpu = container_of(nb,
502 					struct intel_vgpu,
503 					vdev.group_notifier);
504 
505 	/* the only action we care about */
506 	if (action == VFIO_GROUP_NOTIFY_SET_KVM) {
507 		vgpu->vdev.kvm = data;
508 
509 		if (!data)
510 			schedule_work(&vgpu->vdev.release_work);
511 	}
512 
513 	return NOTIFY_OK;
514 }
515 
516 static int intel_vgpu_open(struct mdev_device *mdev)
517 {
518 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
519 	unsigned long events;
520 	int ret;
521 
522 	vgpu->vdev.iommu_notifier.notifier_call = intel_vgpu_iommu_notifier;
523 	vgpu->vdev.group_notifier.notifier_call = intel_vgpu_group_notifier;
524 
525 	events = VFIO_IOMMU_NOTIFY_DMA_UNMAP;
526 	ret = vfio_register_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY, &events,
527 				&vgpu->vdev.iommu_notifier);
528 	if (ret != 0) {
529 		gvt_vgpu_err("vfio_register_notifier for iommu failed: %d\n",
530 			ret);
531 		goto out;
532 	}
533 
534 	events = VFIO_GROUP_NOTIFY_SET_KVM;
535 	ret = vfio_register_notifier(mdev_dev(mdev), VFIO_GROUP_NOTIFY, &events,
536 				&vgpu->vdev.group_notifier);
537 	if (ret != 0) {
538 		gvt_vgpu_err("vfio_register_notifier for group failed: %d\n",
539 			ret);
540 		goto undo_iommu;
541 	}
542 
543 	ret = kvmgt_guest_init(mdev);
544 	if (ret)
545 		goto undo_group;
546 
547 	intel_gvt_ops->vgpu_activate(vgpu);
548 
549 	atomic_set(&vgpu->vdev.released, 0);
550 	return ret;
551 
552 undo_group:
553 	vfio_unregister_notifier(mdev_dev(mdev), VFIO_GROUP_NOTIFY,
554 					&vgpu->vdev.group_notifier);
555 
556 undo_iommu:
557 	vfio_unregister_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY,
558 					&vgpu->vdev.iommu_notifier);
559 out:
560 	return ret;
561 }
562 
563 static void __intel_vgpu_release(struct intel_vgpu *vgpu)
564 {
565 	struct kvmgt_guest_info *info;
566 	int ret;
567 
568 	if (!handle_valid(vgpu->handle))
569 		return;
570 
571 	if (atomic_cmpxchg(&vgpu->vdev.released, 0, 1))
572 		return;
573 
574 	intel_gvt_ops->vgpu_deactivate(vgpu);
575 
576 	ret = vfio_unregister_notifier(mdev_dev(vgpu->vdev.mdev), VFIO_IOMMU_NOTIFY,
577 					&vgpu->vdev.iommu_notifier);
578 	WARN(ret, "vfio_unregister_notifier for iommu failed: %d\n", ret);
579 
580 	ret = vfio_unregister_notifier(mdev_dev(vgpu->vdev.mdev), VFIO_GROUP_NOTIFY,
581 					&vgpu->vdev.group_notifier);
582 	WARN(ret, "vfio_unregister_notifier for group failed: %d\n", ret);
583 
584 	info = (struct kvmgt_guest_info *)vgpu->handle;
585 	kvmgt_guest_exit(info);
586 
587 	vgpu->vdev.kvm = NULL;
588 	vgpu->handle = 0;
589 }
590 
591 static void intel_vgpu_release(struct mdev_device *mdev)
592 {
593 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
594 
595 	__intel_vgpu_release(vgpu);
596 }
597 
598 static void intel_vgpu_release_work(struct work_struct *work)
599 {
600 	struct intel_vgpu *vgpu = container_of(work, struct intel_vgpu,
601 					vdev.release_work);
602 
603 	__intel_vgpu_release(vgpu);
604 }
605 
606 static uint64_t intel_vgpu_get_bar0_addr(struct intel_vgpu *vgpu)
607 {
608 	u32 start_lo, start_hi;
609 	u32 mem_type;
610 	int pos = PCI_BASE_ADDRESS_0;
611 
612 	start_lo = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + pos)) &
613 			PCI_BASE_ADDRESS_MEM_MASK;
614 	mem_type = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + pos)) &
615 			PCI_BASE_ADDRESS_MEM_TYPE_MASK;
616 
617 	switch (mem_type) {
618 	case PCI_BASE_ADDRESS_MEM_TYPE_64:
619 		start_hi = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space
620 						+ pos + 4));
621 		break;
622 	case PCI_BASE_ADDRESS_MEM_TYPE_32:
623 	case PCI_BASE_ADDRESS_MEM_TYPE_1M:
624 		/* 1M mem BAR treated as 32-bit BAR */
625 	default:
626 		/* mem unknown type treated as 32-bit BAR */
627 		start_hi = 0;
628 		break;
629 	}
630 
631 	return ((u64)start_hi << 32) | start_lo;
632 }
633 
634 static ssize_t intel_vgpu_rw(struct mdev_device *mdev, char *buf,
635 			size_t count, loff_t *ppos, bool is_write)
636 {
637 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
638 	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
639 	uint64_t pos = *ppos & VFIO_PCI_OFFSET_MASK;
640 	int ret = -EINVAL;
641 
642 
643 	if (index >= VFIO_PCI_NUM_REGIONS) {
644 		gvt_vgpu_err("invalid index: %u\n", index);
645 		return -EINVAL;
646 	}
647 
648 	switch (index) {
649 	case VFIO_PCI_CONFIG_REGION_INDEX:
650 		if (is_write)
651 			ret = intel_gvt_ops->emulate_cfg_write(vgpu, pos,
652 						buf, count);
653 		else
654 			ret = intel_gvt_ops->emulate_cfg_read(vgpu, pos,
655 						buf, count);
656 		break;
657 	case VFIO_PCI_BAR0_REGION_INDEX:
658 	case VFIO_PCI_BAR1_REGION_INDEX:
659 		if (is_write) {
660 			uint64_t bar0_start = intel_vgpu_get_bar0_addr(vgpu);
661 
662 			ret = intel_gvt_ops->emulate_mmio_write(vgpu,
663 						bar0_start + pos, buf, count);
664 		} else {
665 			uint64_t bar0_start = intel_vgpu_get_bar0_addr(vgpu);
666 
667 			ret = intel_gvt_ops->emulate_mmio_read(vgpu,
668 						bar0_start + pos, buf, count);
669 		}
670 		break;
671 	case VFIO_PCI_BAR2_REGION_INDEX:
672 	case VFIO_PCI_BAR3_REGION_INDEX:
673 	case VFIO_PCI_BAR4_REGION_INDEX:
674 	case VFIO_PCI_BAR5_REGION_INDEX:
675 	case VFIO_PCI_VGA_REGION_INDEX:
676 	case VFIO_PCI_ROM_REGION_INDEX:
677 	default:
678 		gvt_vgpu_err("unsupported region: %u\n", index);
679 	}
680 
681 	return ret == 0 ? count : ret;
682 }
683 
684 static ssize_t intel_vgpu_read(struct mdev_device *mdev, char __user *buf,
685 			size_t count, loff_t *ppos)
686 {
687 	unsigned int done = 0;
688 	int ret;
689 
690 	while (count) {
691 		size_t filled;
692 
693 		if (count >= 4 && !(*ppos % 4)) {
694 			u32 val;
695 
696 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
697 					ppos, false);
698 			if (ret <= 0)
699 				goto read_err;
700 
701 			if (copy_to_user(buf, &val, sizeof(val)))
702 				goto read_err;
703 
704 			filled = 4;
705 		} else if (count >= 2 && !(*ppos % 2)) {
706 			u16 val;
707 
708 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
709 					ppos, false);
710 			if (ret <= 0)
711 				goto read_err;
712 
713 			if (copy_to_user(buf, &val, sizeof(val)))
714 				goto read_err;
715 
716 			filled = 2;
717 		} else {
718 			u8 val;
719 
720 			ret = intel_vgpu_rw(mdev, &val, sizeof(val), ppos,
721 					false);
722 			if (ret <= 0)
723 				goto read_err;
724 
725 			if (copy_to_user(buf, &val, sizeof(val)))
726 				goto read_err;
727 
728 			filled = 1;
729 		}
730 
731 		count -= filled;
732 		done += filled;
733 		*ppos += filled;
734 		buf += filled;
735 	}
736 
737 	return done;
738 
739 read_err:
740 	return -EFAULT;
741 }
742 
743 static ssize_t intel_vgpu_write(struct mdev_device *mdev,
744 				const char __user *buf,
745 				size_t count, loff_t *ppos)
746 {
747 	unsigned int done = 0;
748 	int ret;
749 
750 	while (count) {
751 		size_t filled;
752 
753 		if (count >= 4 && !(*ppos % 4)) {
754 			u32 val;
755 
756 			if (copy_from_user(&val, buf, sizeof(val)))
757 				goto write_err;
758 
759 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
760 					ppos, true);
761 			if (ret <= 0)
762 				goto write_err;
763 
764 			filled = 4;
765 		} else if (count >= 2 && !(*ppos % 2)) {
766 			u16 val;
767 
768 			if (copy_from_user(&val, buf, sizeof(val)))
769 				goto write_err;
770 
771 			ret = intel_vgpu_rw(mdev, (char *)&val,
772 					sizeof(val), ppos, true);
773 			if (ret <= 0)
774 				goto write_err;
775 
776 			filled = 2;
777 		} else {
778 			u8 val;
779 
780 			if (copy_from_user(&val, buf, sizeof(val)))
781 				goto write_err;
782 
783 			ret = intel_vgpu_rw(mdev, &val, sizeof(val),
784 					ppos, true);
785 			if (ret <= 0)
786 				goto write_err;
787 
788 			filled = 1;
789 		}
790 
791 		count -= filled;
792 		done += filled;
793 		*ppos += filled;
794 		buf += filled;
795 	}
796 
797 	return done;
798 write_err:
799 	return -EFAULT;
800 }
801 
802 static int intel_vgpu_mmap(struct mdev_device *mdev, struct vm_area_struct *vma)
803 {
804 	unsigned int index;
805 	u64 virtaddr;
806 	unsigned long req_size, pgoff = 0;
807 	pgprot_t pg_prot;
808 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
809 
810 	index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
811 	if (index >= VFIO_PCI_ROM_REGION_INDEX)
812 		return -EINVAL;
813 
814 	if (vma->vm_end < vma->vm_start)
815 		return -EINVAL;
816 	if ((vma->vm_flags & VM_SHARED) == 0)
817 		return -EINVAL;
818 	if (index != VFIO_PCI_BAR2_REGION_INDEX)
819 		return -EINVAL;
820 
821 	pg_prot = vma->vm_page_prot;
822 	virtaddr = vma->vm_start;
823 	req_size = vma->vm_end - vma->vm_start;
824 	pgoff = vgpu_aperture_pa_base(vgpu) >> PAGE_SHIFT;
825 
826 	return remap_pfn_range(vma, virtaddr, pgoff, req_size, pg_prot);
827 }
828 
829 static int intel_vgpu_get_irq_count(struct intel_vgpu *vgpu, int type)
830 {
831 	if (type == VFIO_PCI_INTX_IRQ_INDEX || type == VFIO_PCI_MSI_IRQ_INDEX)
832 		return 1;
833 
834 	return 0;
835 }
836 
837 static int intel_vgpu_set_intx_mask(struct intel_vgpu *vgpu,
838 			unsigned int index, unsigned int start,
839 			unsigned int count, uint32_t flags,
840 			void *data)
841 {
842 	return 0;
843 }
844 
845 static int intel_vgpu_set_intx_unmask(struct intel_vgpu *vgpu,
846 			unsigned int index, unsigned int start,
847 			unsigned int count, uint32_t flags, void *data)
848 {
849 	return 0;
850 }
851 
852 static int intel_vgpu_set_intx_trigger(struct intel_vgpu *vgpu,
853 		unsigned int index, unsigned int start, unsigned int count,
854 		uint32_t flags, void *data)
855 {
856 	return 0;
857 }
858 
859 static int intel_vgpu_set_msi_trigger(struct intel_vgpu *vgpu,
860 		unsigned int index, unsigned int start, unsigned int count,
861 		uint32_t flags, void *data)
862 {
863 	struct eventfd_ctx *trigger;
864 
865 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
866 		int fd = *(int *)data;
867 
868 		trigger = eventfd_ctx_fdget(fd);
869 		if (IS_ERR(trigger)) {
870 			gvt_vgpu_err("eventfd_ctx_fdget failed\n");
871 			return PTR_ERR(trigger);
872 		}
873 		vgpu->vdev.msi_trigger = trigger;
874 	}
875 
876 	return 0;
877 }
878 
879 static int intel_vgpu_set_irqs(struct intel_vgpu *vgpu, uint32_t flags,
880 		unsigned int index, unsigned int start, unsigned int count,
881 		void *data)
882 {
883 	int (*func)(struct intel_vgpu *vgpu, unsigned int index,
884 			unsigned int start, unsigned int count, uint32_t flags,
885 			void *data) = NULL;
886 
887 	switch (index) {
888 	case VFIO_PCI_INTX_IRQ_INDEX:
889 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
890 		case VFIO_IRQ_SET_ACTION_MASK:
891 			func = intel_vgpu_set_intx_mask;
892 			break;
893 		case VFIO_IRQ_SET_ACTION_UNMASK:
894 			func = intel_vgpu_set_intx_unmask;
895 			break;
896 		case VFIO_IRQ_SET_ACTION_TRIGGER:
897 			func = intel_vgpu_set_intx_trigger;
898 			break;
899 		}
900 		break;
901 	case VFIO_PCI_MSI_IRQ_INDEX:
902 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
903 		case VFIO_IRQ_SET_ACTION_MASK:
904 		case VFIO_IRQ_SET_ACTION_UNMASK:
905 			/* XXX Need masking support exported */
906 			break;
907 		case VFIO_IRQ_SET_ACTION_TRIGGER:
908 			func = intel_vgpu_set_msi_trigger;
909 			break;
910 		}
911 		break;
912 	}
913 
914 	if (!func)
915 		return -ENOTTY;
916 
917 	return func(vgpu, index, start, count, flags, data);
918 }
919 
920 static long intel_vgpu_ioctl(struct mdev_device *mdev, unsigned int cmd,
921 			     unsigned long arg)
922 {
923 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
924 	unsigned long minsz;
925 
926 	gvt_dbg_core("vgpu%d ioctl, cmd: %d\n", vgpu->id, cmd);
927 
928 	if (cmd == VFIO_DEVICE_GET_INFO) {
929 		struct vfio_device_info info;
930 
931 		minsz = offsetofend(struct vfio_device_info, num_irqs);
932 
933 		if (copy_from_user(&info, (void __user *)arg, minsz))
934 			return -EFAULT;
935 
936 		if (info.argsz < minsz)
937 			return -EINVAL;
938 
939 		info.flags = VFIO_DEVICE_FLAGS_PCI;
940 		info.flags |= VFIO_DEVICE_FLAGS_RESET;
941 		info.num_regions = VFIO_PCI_NUM_REGIONS;
942 		info.num_irqs = VFIO_PCI_NUM_IRQS;
943 
944 		return copy_to_user((void __user *)arg, &info, minsz) ?
945 			-EFAULT : 0;
946 
947 	} else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
948 		struct vfio_region_info info;
949 		struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
950 		int i, ret;
951 		struct vfio_region_info_cap_sparse_mmap *sparse = NULL;
952 		size_t size;
953 		int nr_areas = 1;
954 		int cap_type_id;
955 
956 		minsz = offsetofend(struct vfio_region_info, offset);
957 
958 		if (copy_from_user(&info, (void __user *)arg, minsz))
959 			return -EFAULT;
960 
961 		if (info.argsz < minsz)
962 			return -EINVAL;
963 
964 		switch (info.index) {
965 		case VFIO_PCI_CONFIG_REGION_INDEX:
966 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
967 			info.size = INTEL_GVT_MAX_CFG_SPACE_SZ;
968 			info.flags = VFIO_REGION_INFO_FLAG_READ |
969 				     VFIO_REGION_INFO_FLAG_WRITE;
970 			break;
971 		case VFIO_PCI_BAR0_REGION_INDEX:
972 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
973 			info.size = vgpu->cfg_space.bar[info.index].size;
974 			if (!info.size) {
975 				info.flags = 0;
976 				break;
977 			}
978 
979 			info.flags = VFIO_REGION_INFO_FLAG_READ |
980 				     VFIO_REGION_INFO_FLAG_WRITE;
981 			break;
982 		case VFIO_PCI_BAR1_REGION_INDEX:
983 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
984 			info.size = 0;
985 			info.flags = 0;
986 			break;
987 		case VFIO_PCI_BAR2_REGION_INDEX:
988 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
989 			info.flags = VFIO_REGION_INFO_FLAG_CAPS |
990 					VFIO_REGION_INFO_FLAG_MMAP |
991 					VFIO_REGION_INFO_FLAG_READ |
992 					VFIO_REGION_INFO_FLAG_WRITE;
993 			info.size = gvt_aperture_sz(vgpu->gvt);
994 
995 			size = sizeof(*sparse) +
996 					(nr_areas * sizeof(*sparse->areas));
997 			sparse = kzalloc(size, GFP_KERNEL);
998 			if (!sparse)
999 				return -ENOMEM;
1000 
1001 			sparse->nr_areas = nr_areas;
1002 			cap_type_id = VFIO_REGION_INFO_CAP_SPARSE_MMAP;
1003 			sparse->areas[0].offset =
1004 					PAGE_ALIGN(vgpu_aperture_offset(vgpu));
1005 			sparse->areas[0].size = vgpu_aperture_sz(vgpu);
1006 			break;
1007 
1008 		case VFIO_PCI_BAR3_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
1009 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
1010 			info.size = 0;
1011 
1012 			info.flags = 0;
1013 			gvt_dbg_core("get region info bar:%d\n", info.index);
1014 			break;
1015 
1016 		case VFIO_PCI_ROM_REGION_INDEX:
1017 		case VFIO_PCI_VGA_REGION_INDEX:
1018 			gvt_dbg_core("get region info index:%d\n", info.index);
1019 			break;
1020 		default:
1021 			{
1022 				struct vfio_region_info_cap_type cap_type;
1023 
1024 				if (info.index >= VFIO_PCI_NUM_REGIONS +
1025 						vgpu->vdev.num_regions)
1026 					return -EINVAL;
1027 
1028 				i = info.index - VFIO_PCI_NUM_REGIONS;
1029 
1030 				info.offset =
1031 					VFIO_PCI_INDEX_TO_OFFSET(info.index);
1032 				info.size = vgpu->vdev.region[i].size;
1033 				info.flags = vgpu->vdev.region[i].flags;
1034 
1035 				cap_type.type = vgpu->vdev.region[i].type;
1036 				cap_type.subtype = vgpu->vdev.region[i].subtype;
1037 
1038 				ret = vfio_info_add_capability(&caps,
1039 						VFIO_REGION_INFO_CAP_TYPE,
1040 						&cap_type);
1041 				if (ret)
1042 					return ret;
1043 			}
1044 		}
1045 
1046 		if ((info.flags & VFIO_REGION_INFO_FLAG_CAPS) && sparse) {
1047 			switch (cap_type_id) {
1048 			case VFIO_REGION_INFO_CAP_SPARSE_MMAP:
1049 				ret = vfio_info_add_capability(&caps,
1050 					VFIO_REGION_INFO_CAP_SPARSE_MMAP,
1051 					sparse);
1052 				kfree(sparse);
1053 				if (ret)
1054 					return ret;
1055 				break;
1056 			default:
1057 				return -EINVAL;
1058 			}
1059 		}
1060 
1061 		if (caps.size) {
1062 			if (info.argsz < sizeof(info) + caps.size) {
1063 				info.argsz = sizeof(info) + caps.size;
1064 				info.cap_offset = 0;
1065 			} else {
1066 				vfio_info_cap_shift(&caps, sizeof(info));
1067 				if (copy_to_user((void __user *)arg +
1068 						  sizeof(info), caps.buf,
1069 						  caps.size)) {
1070 					kfree(caps.buf);
1071 					return -EFAULT;
1072 				}
1073 				info.cap_offset = sizeof(info);
1074 			}
1075 
1076 			kfree(caps.buf);
1077 		}
1078 
1079 		return copy_to_user((void __user *)arg, &info, minsz) ?
1080 			-EFAULT : 0;
1081 	} else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
1082 		struct vfio_irq_info info;
1083 
1084 		minsz = offsetofend(struct vfio_irq_info, count);
1085 
1086 		if (copy_from_user(&info, (void __user *)arg, minsz))
1087 			return -EFAULT;
1088 
1089 		if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
1090 			return -EINVAL;
1091 
1092 		switch (info.index) {
1093 		case VFIO_PCI_INTX_IRQ_INDEX:
1094 		case VFIO_PCI_MSI_IRQ_INDEX:
1095 			break;
1096 		default:
1097 			return -EINVAL;
1098 		}
1099 
1100 		info.flags = VFIO_IRQ_INFO_EVENTFD;
1101 
1102 		info.count = intel_vgpu_get_irq_count(vgpu, info.index);
1103 
1104 		if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
1105 			info.flags |= (VFIO_IRQ_INFO_MASKABLE |
1106 				       VFIO_IRQ_INFO_AUTOMASKED);
1107 		else
1108 			info.flags |= VFIO_IRQ_INFO_NORESIZE;
1109 
1110 		return copy_to_user((void __user *)arg, &info, minsz) ?
1111 			-EFAULT : 0;
1112 	} else if (cmd == VFIO_DEVICE_SET_IRQS) {
1113 		struct vfio_irq_set hdr;
1114 		u8 *data = NULL;
1115 		int ret = 0;
1116 		size_t data_size = 0;
1117 
1118 		minsz = offsetofend(struct vfio_irq_set, count);
1119 
1120 		if (copy_from_user(&hdr, (void __user *)arg, minsz))
1121 			return -EFAULT;
1122 
1123 		if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) {
1124 			int max = intel_vgpu_get_irq_count(vgpu, hdr.index);
1125 
1126 			ret = vfio_set_irqs_validate_and_prepare(&hdr, max,
1127 						VFIO_PCI_NUM_IRQS, &data_size);
1128 			if (ret) {
1129 				gvt_vgpu_err("intel:vfio_set_irqs_validate_and_prepare failed\n");
1130 				return -EINVAL;
1131 			}
1132 			if (data_size) {
1133 				data = memdup_user((void __user *)(arg + minsz),
1134 						   data_size);
1135 				if (IS_ERR(data))
1136 					return PTR_ERR(data);
1137 			}
1138 		}
1139 
1140 		ret = intel_vgpu_set_irqs(vgpu, hdr.flags, hdr.index,
1141 					hdr.start, hdr.count, data);
1142 		kfree(data);
1143 
1144 		return ret;
1145 	} else if (cmd == VFIO_DEVICE_RESET) {
1146 		intel_gvt_ops->vgpu_reset(vgpu);
1147 		return 0;
1148 	}
1149 
1150 	return 0;
1151 }
1152 
1153 static const struct mdev_parent_ops intel_vgpu_ops = {
1154 	.supported_type_groups	= intel_vgpu_type_groups,
1155 	.create			= intel_vgpu_create,
1156 	.remove			= intel_vgpu_remove,
1157 
1158 	.open			= intel_vgpu_open,
1159 	.release		= intel_vgpu_release,
1160 
1161 	.read			= intel_vgpu_read,
1162 	.write			= intel_vgpu_write,
1163 	.mmap			= intel_vgpu_mmap,
1164 	.ioctl			= intel_vgpu_ioctl,
1165 };
1166 
1167 static int kvmgt_host_init(struct device *dev, void *gvt, const void *ops)
1168 {
1169 	if (!intel_gvt_init_vgpu_type_groups(gvt))
1170 		return -EFAULT;
1171 
1172 	intel_gvt_ops = ops;
1173 
1174 	return mdev_register_device(dev, &intel_vgpu_ops);
1175 }
1176 
1177 static void kvmgt_host_exit(struct device *dev, void *gvt)
1178 {
1179 	intel_gvt_cleanup_vgpu_type_groups(gvt);
1180 	mdev_unregister_device(dev);
1181 }
1182 
1183 static int kvmgt_write_protect_add(unsigned long handle, u64 gfn)
1184 {
1185 	struct kvmgt_guest_info *info;
1186 	struct kvm *kvm;
1187 	struct kvm_memory_slot *slot;
1188 	int idx;
1189 
1190 	if (!handle_valid(handle))
1191 		return -ESRCH;
1192 
1193 	info = (struct kvmgt_guest_info *)handle;
1194 	kvm = info->kvm;
1195 
1196 	idx = srcu_read_lock(&kvm->srcu);
1197 	slot = gfn_to_memslot(kvm, gfn);
1198 	if (!slot) {
1199 		srcu_read_unlock(&kvm->srcu, idx);
1200 		return -EINVAL;
1201 	}
1202 
1203 	spin_lock(&kvm->mmu_lock);
1204 
1205 	if (kvmgt_gfn_is_write_protected(info, gfn))
1206 		goto out;
1207 
1208 	kvm_slot_page_track_add_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1209 	kvmgt_protect_table_add(info, gfn);
1210 
1211 out:
1212 	spin_unlock(&kvm->mmu_lock);
1213 	srcu_read_unlock(&kvm->srcu, idx);
1214 	return 0;
1215 }
1216 
1217 static int kvmgt_write_protect_remove(unsigned long handle, u64 gfn)
1218 {
1219 	struct kvmgt_guest_info *info;
1220 	struct kvm *kvm;
1221 	struct kvm_memory_slot *slot;
1222 	int idx;
1223 
1224 	if (!handle_valid(handle))
1225 		return 0;
1226 
1227 	info = (struct kvmgt_guest_info *)handle;
1228 	kvm = info->kvm;
1229 
1230 	idx = srcu_read_lock(&kvm->srcu);
1231 	slot = gfn_to_memslot(kvm, gfn);
1232 	if (!slot) {
1233 		srcu_read_unlock(&kvm->srcu, idx);
1234 		return -EINVAL;
1235 	}
1236 
1237 	spin_lock(&kvm->mmu_lock);
1238 
1239 	if (!kvmgt_gfn_is_write_protected(info, gfn))
1240 		goto out;
1241 
1242 	kvm_slot_page_track_remove_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1243 	kvmgt_protect_table_del(info, gfn);
1244 
1245 out:
1246 	spin_unlock(&kvm->mmu_lock);
1247 	srcu_read_unlock(&kvm->srcu, idx);
1248 	return 0;
1249 }
1250 
1251 static void kvmgt_page_track_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1252 		const u8 *val, int len,
1253 		struct kvm_page_track_notifier_node *node)
1254 {
1255 	struct kvmgt_guest_info *info = container_of(node,
1256 					struct kvmgt_guest_info, track_node);
1257 
1258 	if (kvmgt_gfn_is_write_protected(info, gpa_to_gfn(gpa)))
1259 		intel_gvt_ops->emulate_mmio_write(info->vgpu, gpa,
1260 					(void *)val, len);
1261 }
1262 
1263 static void kvmgt_page_track_flush_slot(struct kvm *kvm,
1264 		struct kvm_memory_slot *slot,
1265 		struct kvm_page_track_notifier_node *node)
1266 {
1267 	int i;
1268 	gfn_t gfn;
1269 	struct kvmgt_guest_info *info = container_of(node,
1270 					struct kvmgt_guest_info, track_node);
1271 
1272 	spin_lock(&kvm->mmu_lock);
1273 	for (i = 0; i < slot->npages; i++) {
1274 		gfn = slot->base_gfn + i;
1275 		if (kvmgt_gfn_is_write_protected(info, gfn)) {
1276 			kvm_slot_page_track_remove_page(kvm, slot, gfn,
1277 						KVM_PAGE_TRACK_WRITE);
1278 			kvmgt_protect_table_del(info, gfn);
1279 		}
1280 	}
1281 	spin_unlock(&kvm->mmu_lock);
1282 }
1283 
1284 static bool __kvmgt_vgpu_exist(struct intel_vgpu *vgpu, struct kvm *kvm)
1285 {
1286 	struct intel_vgpu *itr;
1287 	struct kvmgt_guest_info *info;
1288 	int id;
1289 	bool ret = false;
1290 
1291 	mutex_lock(&vgpu->gvt->lock);
1292 	for_each_active_vgpu(vgpu->gvt, itr, id) {
1293 		if (!handle_valid(itr->handle))
1294 			continue;
1295 
1296 		info = (struct kvmgt_guest_info *)itr->handle;
1297 		if (kvm && kvm == info->kvm) {
1298 			ret = true;
1299 			goto out;
1300 		}
1301 	}
1302 out:
1303 	mutex_unlock(&vgpu->gvt->lock);
1304 	return ret;
1305 }
1306 
1307 static int kvmgt_guest_init(struct mdev_device *mdev)
1308 {
1309 	struct kvmgt_guest_info *info;
1310 	struct intel_vgpu *vgpu;
1311 	struct kvm *kvm;
1312 
1313 	vgpu = mdev_get_drvdata(mdev);
1314 	if (handle_valid(vgpu->handle))
1315 		return -EEXIST;
1316 
1317 	kvm = vgpu->vdev.kvm;
1318 	if (!kvm || kvm->mm != current->mm) {
1319 		gvt_vgpu_err("KVM is required to use Intel vGPU\n");
1320 		return -ESRCH;
1321 	}
1322 
1323 	if (__kvmgt_vgpu_exist(vgpu, kvm))
1324 		return -EEXIST;
1325 
1326 	info = vzalloc(sizeof(struct kvmgt_guest_info));
1327 	if (!info)
1328 		return -ENOMEM;
1329 
1330 	vgpu->handle = (unsigned long)info;
1331 	info->vgpu = vgpu;
1332 	info->kvm = kvm;
1333 	kvm_get_kvm(info->kvm);
1334 
1335 	kvmgt_protect_table_init(info);
1336 	gvt_cache_init(vgpu);
1337 
1338 	info->track_node.track_write = kvmgt_page_track_write;
1339 	info->track_node.track_flush_slot = kvmgt_page_track_flush_slot;
1340 	kvm_page_track_register_notifier(kvm, &info->track_node);
1341 
1342 	return 0;
1343 }
1344 
1345 static bool kvmgt_guest_exit(struct kvmgt_guest_info *info)
1346 {
1347 	kvm_page_track_unregister_notifier(info->kvm, &info->track_node);
1348 	kvm_put_kvm(info->kvm);
1349 	kvmgt_protect_table_destroy(info);
1350 	gvt_cache_destroy(info->vgpu);
1351 	vfree(info);
1352 
1353 	return true;
1354 }
1355 
1356 static int kvmgt_attach_vgpu(void *vgpu, unsigned long *handle)
1357 {
1358 	/* nothing to do here */
1359 	return 0;
1360 }
1361 
1362 static void kvmgt_detach_vgpu(unsigned long handle)
1363 {
1364 	/* nothing to do here */
1365 }
1366 
1367 static int kvmgt_inject_msi(unsigned long handle, u32 addr, u16 data)
1368 {
1369 	struct kvmgt_guest_info *info;
1370 	struct intel_vgpu *vgpu;
1371 
1372 	if (!handle_valid(handle))
1373 		return -ESRCH;
1374 
1375 	info = (struct kvmgt_guest_info *)handle;
1376 	vgpu = info->vgpu;
1377 
1378 	if (eventfd_signal(vgpu->vdev.msi_trigger, 1) == 1)
1379 		return 0;
1380 
1381 	return -EFAULT;
1382 }
1383 
1384 static unsigned long kvmgt_gfn_to_pfn(unsigned long handle, unsigned long gfn)
1385 {
1386 	unsigned long iova, pfn;
1387 	struct kvmgt_guest_info *info;
1388 	struct device *dev;
1389 	struct intel_vgpu *vgpu;
1390 	int rc;
1391 
1392 	if (!handle_valid(handle))
1393 		return INTEL_GVT_INVALID_ADDR;
1394 
1395 	info = (struct kvmgt_guest_info *)handle;
1396 	vgpu = info->vgpu;
1397 	iova = gvt_cache_find(info->vgpu, gfn);
1398 	if (iova != INTEL_GVT_INVALID_ADDR)
1399 		return iova;
1400 
1401 	pfn = INTEL_GVT_INVALID_ADDR;
1402 	dev = mdev_dev(info->vgpu->vdev.mdev);
1403 	rc = vfio_pin_pages(dev, &gfn, 1, IOMMU_READ | IOMMU_WRITE, &pfn);
1404 	if (rc != 1) {
1405 		gvt_vgpu_err("vfio_pin_pages failed for gfn 0x%lx: %d\n",
1406 			gfn, rc);
1407 		return INTEL_GVT_INVALID_ADDR;
1408 	}
1409 	/* transfer to host iova for GFX to use DMA */
1410 	rc = gvt_dma_map_iova(info->vgpu, pfn, &iova);
1411 	if (rc) {
1412 		gvt_vgpu_err("gvt_dma_map_iova failed for gfn: 0x%lx\n", gfn);
1413 		vfio_unpin_pages(dev, &gfn, 1);
1414 		return INTEL_GVT_INVALID_ADDR;
1415 	}
1416 
1417 	gvt_cache_add(info->vgpu, gfn, iova);
1418 	return iova;
1419 }
1420 
1421 static int kvmgt_rw_gpa(unsigned long handle, unsigned long gpa,
1422 			void *buf, unsigned long len, bool write)
1423 {
1424 	struct kvmgt_guest_info *info;
1425 	struct kvm *kvm;
1426 	int idx, ret;
1427 	bool kthread = current->mm == NULL;
1428 
1429 	if (!handle_valid(handle))
1430 		return -ESRCH;
1431 
1432 	info = (struct kvmgt_guest_info *)handle;
1433 	kvm = info->kvm;
1434 
1435 	if (kthread)
1436 		use_mm(kvm->mm);
1437 
1438 	idx = srcu_read_lock(&kvm->srcu);
1439 	ret = write ? kvm_write_guest(kvm, gpa, buf, len) :
1440 		      kvm_read_guest(kvm, gpa, buf, len);
1441 	srcu_read_unlock(&kvm->srcu, idx);
1442 
1443 	if (kthread)
1444 		unuse_mm(kvm->mm);
1445 
1446 	return ret;
1447 }
1448 
1449 static int kvmgt_read_gpa(unsigned long handle, unsigned long gpa,
1450 			void *buf, unsigned long len)
1451 {
1452 	return kvmgt_rw_gpa(handle, gpa, buf, len, false);
1453 }
1454 
1455 static int kvmgt_write_gpa(unsigned long handle, unsigned long gpa,
1456 			void *buf, unsigned long len)
1457 {
1458 	return kvmgt_rw_gpa(handle, gpa, buf, len, true);
1459 }
1460 
1461 static unsigned long kvmgt_virt_to_pfn(void *addr)
1462 {
1463 	return PFN_DOWN(__pa(addr));
1464 }
1465 
1466 struct intel_gvt_mpt kvmgt_mpt = {
1467 	.host_init = kvmgt_host_init,
1468 	.host_exit = kvmgt_host_exit,
1469 	.attach_vgpu = kvmgt_attach_vgpu,
1470 	.detach_vgpu = kvmgt_detach_vgpu,
1471 	.inject_msi = kvmgt_inject_msi,
1472 	.from_virt_to_mfn = kvmgt_virt_to_pfn,
1473 	.set_wp_page = kvmgt_write_protect_add,
1474 	.unset_wp_page = kvmgt_write_protect_remove,
1475 	.read_gpa = kvmgt_read_gpa,
1476 	.write_gpa = kvmgt_write_gpa,
1477 	.gfn_to_mfn = kvmgt_gfn_to_pfn,
1478 };
1479 EXPORT_SYMBOL_GPL(kvmgt_mpt);
1480 
1481 static int __init kvmgt_init(void)
1482 {
1483 	return 0;
1484 }
1485 
1486 static void __exit kvmgt_exit(void)
1487 {
1488 }
1489 
1490 module_init(kvmgt_init);
1491 module_exit(kvmgt_exit);
1492 
1493 MODULE_LICENSE("GPL and additional rights");
1494 MODULE_AUTHOR("Intel Corporation");
1495