xref: /openbmc/linux/drivers/gpu/drm/i915/gvt/kvmgt.c (revision a36954f5)
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 		       "weight: %d\n",
300 		       BYTES_TO_MB(type->low_gm_size),
301 		       BYTES_TO_MB(type->high_gm_size),
302 		       type->fence, vgpu_edid_str(type->resolution),
303 		       type->weight);
304 }
305 
306 static MDEV_TYPE_ATTR_RO(available_instances);
307 static MDEV_TYPE_ATTR_RO(device_api);
308 static MDEV_TYPE_ATTR_RO(description);
309 
310 static struct attribute *type_attrs[] = {
311 	&mdev_type_attr_available_instances.attr,
312 	&mdev_type_attr_device_api.attr,
313 	&mdev_type_attr_description.attr,
314 	NULL,
315 };
316 
317 static struct attribute_group *intel_vgpu_type_groups[] = {
318 	[0 ... NR_MAX_INTEL_VGPU_TYPES - 1] = NULL,
319 };
320 
321 static bool intel_gvt_init_vgpu_type_groups(struct intel_gvt *gvt)
322 {
323 	int i, j;
324 	struct intel_vgpu_type *type;
325 	struct attribute_group *group;
326 
327 	for (i = 0; i < gvt->num_types; i++) {
328 		type = &gvt->types[i];
329 
330 		group = kzalloc(sizeof(struct attribute_group), GFP_KERNEL);
331 		if (WARN_ON(!group))
332 			goto unwind;
333 
334 		group->name = type->name;
335 		group->attrs = type_attrs;
336 		intel_vgpu_type_groups[i] = group;
337 	}
338 
339 	return true;
340 
341 unwind:
342 	for (j = 0; j < i; j++) {
343 		group = intel_vgpu_type_groups[j];
344 		kfree(group);
345 	}
346 
347 	return false;
348 }
349 
350 static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt)
351 {
352 	int i;
353 	struct attribute_group *group;
354 
355 	for (i = 0; i < gvt->num_types; i++) {
356 		group = intel_vgpu_type_groups[i];
357 		kfree(group);
358 	}
359 }
360 
361 static void kvmgt_protect_table_init(struct kvmgt_guest_info *info)
362 {
363 	hash_init(info->ptable);
364 }
365 
366 static void kvmgt_protect_table_destroy(struct kvmgt_guest_info *info)
367 {
368 	struct kvmgt_pgfn *p;
369 	struct hlist_node *tmp;
370 	int i;
371 
372 	hash_for_each_safe(info->ptable, i, tmp, p, hnode) {
373 		hash_del(&p->hnode);
374 		kfree(p);
375 	}
376 }
377 
378 static struct kvmgt_pgfn *
379 __kvmgt_protect_table_find(struct kvmgt_guest_info *info, gfn_t gfn)
380 {
381 	struct kvmgt_pgfn *p, *res = NULL;
382 
383 	hash_for_each_possible(info->ptable, p, hnode, gfn) {
384 		if (gfn == p->gfn) {
385 			res = p;
386 			break;
387 		}
388 	}
389 
390 	return res;
391 }
392 
393 static bool kvmgt_gfn_is_write_protected(struct kvmgt_guest_info *info,
394 				gfn_t gfn)
395 {
396 	struct kvmgt_pgfn *p;
397 
398 	p = __kvmgt_protect_table_find(info, gfn);
399 	return !!p;
400 }
401 
402 static void kvmgt_protect_table_add(struct kvmgt_guest_info *info, gfn_t gfn)
403 {
404 	struct kvmgt_pgfn *p;
405 
406 	if (kvmgt_gfn_is_write_protected(info, gfn))
407 		return;
408 
409 	p = kzalloc(sizeof(struct kvmgt_pgfn), GFP_ATOMIC);
410 	if (WARN(!p, "gfn: 0x%llx\n", gfn))
411 		return;
412 
413 	p->gfn = gfn;
414 	hash_add(info->ptable, &p->hnode, gfn);
415 }
416 
417 static void kvmgt_protect_table_del(struct kvmgt_guest_info *info,
418 				gfn_t gfn)
419 {
420 	struct kvmgt_pgfn *p;
421 
422 	p = __kvmgt_protect_table_find(info, gfn);
423 	if (p) {
424 		hash_del(&p->hnode);
425 		kfree(p);
426 	}
427 }
428 
429 static int intel_vgpu_create(struct kobject *kobj, struct mdev_device *mdev)
430 {
431 	struct intel_vgpu *vgpu = NULL;
432 	struct intel_vgpu_type *type;
433 	struct device *pdev;
434 	void *gvt;
435 	int ret;
436 
437 	pdev = mdev_parent_dev(mdev);
438 	gvt = kdev_to_i915(pdev)->gvt;
439 
440 	type = intel_gvt_find_vgpu_type(gvt, kobject_name(kobj));
441 	if (!type) {
442 		gvt_vgpu_err("failed to find type %s to create\n",
443 						kobject_name(kobj));
444 		ret = -EINVAL;
445 		goto out;
446 	}
447 
448 	vgpu = intel_gvt_ops->vgpu_create(gvt, type);
449 	if (IS_ERR_OR_NULL(vgpu)) {
450 		ret = vgpu == NULL ? -EFAULT : PTR_ERR(vgpu);
451 		gvt_vgpu_err("failed to create intel vgpu: %d\n", ret);
452 		goto out;
453 	}
454 
455 	INIT_WORK(&vgpu->vdev.release_work, intel_vgpu_release_work);
456 
457 	vgpu->vdev.mdev = mdev;
458 	mdev_set_drvdata(mdev, vgpu);
459 
460 	gvt_dbg_core("intel_vgpu_create succeeded for mdev: %s\n",
461 		     dev_name(mdev_dev(mdev)));
462 	ret = 0;
463 
464 out:
465 	return ret;
466 }
467 
468 static int intel_vgpu_remove(struct mdev_device *mdev)
469 {
470 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
471 
472 	if (handle_valid(vgpu->handle))
473 		return -EBUSY;
474 
475 	intel_gvt_ops->vgpu_destroy(vgpu);
476 	return 0;
477 }
478 
479 static int intel_vgpu_iommu_notifier(struct notifier_block *nb,
480 				     unsigned long action, void *data)
481 {
482 	struct intel_vgpu *vgpu = container_of(nb,
483 					struct intel_vgpu,
484 					vdev.iommu_notifier);
485 
486 	if (action == VFIO_IOMMU_NOTIFY_DMA_UNMAP) {
487 		struct vfio_iommu_type1_dma_unmap *unmap = data;
488 		unsigned long gfn, end_gfn;
489 
490 		gfn = unmap->iova >> PAGE_SHIFT;
491 		end_gfn = gfn + unmap->size / PAGE_SIZE;
492 
493 		while (gfn < end_gfn)
494 			gvt_cache_remove(vgpu, gfn++);
495 	}
496 
497 	return NOTIFY_OK;
498 }
499 
500 static int intel_vgpu_group_notifier(struct notifier_block *nb,
501 				     unsigned long action, void *data)
502 {
503 	struct intel_vgpu *vgpu = container_of(nb,
504 					struct intel_vgpu,
505 					vdev.group_notifier);
506 
507 	/* the only action we care about */
508 	if (action == VFIO_GROUP_NOTIFY_SET_KVM) {
509 		vgpu->vdev.kvm = data;
510 
511 		if (!data)
512 			schedule_work(&vgpu->vdev.release_work);
513 	}
514 
515 	return NOTIFY_OK;
516 }
517 
518 static int intel_vgpu_open(struct mdev_device *mdev)
519 {
520 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
521 	unsigned long events;
522 	int ret;
523 
524 	vgpu->vdev.iommu_notifier.notifier_call = intel_vgpu_iommu_notifier;
525 	vgpu->vdev.group_notifier.notifier_call = intel_vgpu_group_notifier;
526 
527 	events = VFIO_IOMMU_NOTIFY_DMA_UNMAP;
528 	ret = vfio_register_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY, &events,
529 				&vgpu->vdev.iommu_notifier);
530 	if (ret != 0) {
531 		gvt_vgpu_err("vfio_register_notifier for iommu failed: %d\n",
532 			ret);
533 		goto out;
534 	}
535 
536 	events = VFIO_GROUP_NOTIFY_SET_KVM;
537 	ret = vfio_register_notifier(mdev_dev(mdev), VFIO_GROUP_NOTIFY, &events,
538 				&vgpu->vdev.group_notifier);
539 	if (ret != 0) {
540 		gvt_vgpu_err("vfio_register_notifier for group failed: %d\n",
541 			ret);
542 		goto undo_iommu;
543 	}
544 
545 	ret = kvmgt_guest_init(mdev);
546 	if (ret)
547 		goto undo_group;
548 
549 	intel_gvt_ops->vgpu_activate(vgpu);
550 
551 	atomic_set(&vgpu->vdev.released, 0);
552 	return ret;
553 
554 undo_group:
555 	vfio_unregister_notifier(mdev_dev(mdev), VFIO_GROUP_NOTIFY,
556 					&vgpu->vdev.group_notifier);
557 
558 undo_iommu:
559 	vfio_unregister_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY,
560 					&vgpu->vdev.iommu_notifier);
561 out:
562 	return ret;
563 }
564 
565 static void __intel_vgpu_release(struct intel_vgpu *vgpu)
566 {
567 	struct kvmgt_guest_info *info;
568 	int ret;
569 
570 	if (!handle_valid(vgpu->handle))
571 		return;
572 
573 	if (atomic_cmpxchg(&vgpu->vdev.released, 0, 1))
574 		return;
575 
576 	intel_gvt_ops->vgpu_deactivate(vgpu);
577 
578 	ret = vfio_unregister_notifier(mdev_dev(vgpu->vdev.mdev), VFIO_IOMMU_NOTIFY,
579 					&vgpu->vdev.iommu_notifier);
580 	WARN(ret, "vfio_unregister_notifier for iommu failed: %d\n", ret);
581 
582 	ret = vfio_unregister_notifier(mdev_dev(vgpu->vdev.mdev), VFIO_GROUP_NOTIFY,
583 					&vgpu->vdev.group_notifier);
584 	WARN(ret, "vfio_unregister_notifier for group failed: %d\n", ret);
585 
586 	info = (struct kvmgt_guest_info *)vgpu->handle;
587 	kvmgt_guest_exit(info);
588 
589 	vgpu->vdev.kvm = NULL;
590 	vgpu->handle = 0;
591 }
592 
593 static void intel_vgpu_release(struct mdev_device *mdev)
594 {
595 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
596 
597 	__intel_vgpu_release(vgpu);
598 }
599 
600 static void intel_vgpu_release_work(struct work_struct *work)
601 {
602 	struct intel_vgpu *vgpu = container_of(work, struct intel_vgpu,
603 					vdev.release_work);
604 
605 	__intel_vgpu_release(vgpu);
606 }
607 
608 static uint64_t intel_vgpu_get_bar0_addr(struct intel_vgpu *vgpu)
609 {
610 	u32 start_lo, start_hi;
611 	u32 mem_type;
612 	int pos = PCI_BASE_ADDRESS_0;
613 
614 	start_lo = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + pos)) &
615 			PCI_BASE_ADDRESS_MEM_MASK;
616 	mem_type = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + pos)) &
617 			PCI_BASE_ADDRESS_MEM_TYPE_MASK;
618 
619 	switch (mem_type) {
620 	case PCI_BASE_ADDRESS_MEM_TYPE_64:
621 		start_hi = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space
622 						+ pos + 4));
623 		break;
624 	case PCI_BASE_ADDRESS_MEM_TYPE_32:
625 	case PCI_BASE_ADDRESS_MEM_TYPE_1M:
626 		/* 1M mem BAR treated as 32-bit BAR */
627 	default:
628 		/* mem unknown type treated as 32-bit BAR */
629 		start_hi = 0;
630 		break;
631 	}
632 
633 	return ((u64)start_hi << 32) | start_lo;
634 }
635 
636 static ssize_t intel_vgpu_rw(struct mdev_device *mdev, char *buf,
637 			size_t count, loff_t *ppos, bool is_write)
638 {
639 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
640 	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
641 	uint64_t pos = *ppos & VFIO_PCI_OFFSET_MASK;
642 	int ret = -EINVAL;
643 
644 
645 	if (index >= VFIO_PCI_NUM_REGIONS) {
646 		gvt_vgpu_err("invalid index: %u\n", index);
647 		return -EINVAL;
648 	}
649 
650 	switch (index) {
651 	case VFIO_PCI_CONFIG_REGION_INDEX:
652 		if (is_write)
653 			ret = intel_gvt_ops->emulate_cfg_write(vgpu, pos,
654 						buf, count);
655 		else
656 			ret = intel_gvt_ops->emulate_cfg_read(vgpu, pos,
657 						buf, count);
658 		break;
659 	case VFIO_PCI_BAR0_REGION_INDEX:
660 	case VFIO_PCI_BAR1_REGION_INDEX:
661 		if (is_write) {
662 			uint64_t bar0_start = intel_vgpu_get_bar0_addr(vgpu);
663 
664 			ret = intel_gvt_ops->emulate_mmio_write(vgpu,
665 						bar0_start + pos, buf, count);
666 		} else {
667 			uint64_t bar0_start = intel_vgpu_get_bar0_addr(vgpu);
668 
669 			ret = intel_gvt_ops->emulate_mmio_read(vgpu,
670 						bar0_start + pos, buf, count);
671 		}
672 		break;
673 	case VFIO_PCI_BAR2_REGION_INDEX:
674 	case VFIO_PCI_BAR3_REGION_INDEX:
675 	case VFIO_PCI_BAR4_REGION_INDEX:
676 	case VFIO_PCI_BAR5_REGION_INDEX:
677 	case VFIO_PCI_VGA_REGION_INDEX:
678 	case VFIO_PCI_ROM_REGION_INDEX:
679 	default:
680 		gvt_vgpu_err("unsupported region: %u\n", index);
681 	}
682 
683 	return ret == 0 ? count : ret;
684 }
685 
686 static ssize_t intel_vgpu_read(struct mdev_device *mdev, char __user *buf,
687 			size_t count, loff_t *ppos)
688 {
689 	unsigned int done = 0;
690 	int ret;
691 
692 	while (count) {
693 		size_t filled;
694 
695 		if (count >= 4 && !(*ppos % 4)) {
696 			u32 val;
697 
698 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
699 					ppos, false);
700 			if (ret <= 0)
701 				goto read_err;
702 
703 			if (copy_to_user(buf, &val, sizeof(val)))
704 				goto read_err;
705 
706 			filled = 4;
707 		} else if (count >= 2 && !(*ppos % 2)) {
708 			u16 val;
709 
710 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
711 					ppos, false);
712 			if (ret <= 0)
713 				goto read_err;
714 
715 			if (copy_to_user(buf, &val, sizeof(val)))
716 				goto read_err;
717 
718 			filled = 2;
719 		} else {
720 			u8 val;
721 
722 			ret = intel_vgpu_rw(mdev, &val, sizeof(val), ppos,
723 					false);
724 			if (ret <= 0)
725 				goto read_err;
726 
727 			if (copy_to_user(buf, &val, sizeof(val)))
728 				goto read_err;
729 
730 			filled = 1;
731 		}
732 
733 		count -= filled;
734 		done += filled;
735 		*ppos += filled;
736 		buf += filled;
737 	}
738 
739 	return done;
740 
741 read_err:
742 	return -EFAULT;
743 }
744 
745 static ssize_t intel_vgpu_write(struct mdev_device *mdev,
746 				const char __user *buf,
747 				size_t count, loff_t *ppos)
748 {
749 	unsigned int done = 0;
750 	int ret;
751 
752 	while (count) {
753 		size_t filled;
754 
755 		if (count >= 4 && !(*ppos % 4)) {
756 			u32 val;
757 
758 			if (copy_from_user(&val, buf, sizeof(val)))
759 				goto write_err;
760 
761 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
762 					ppos, true);
763 			if (ret <= 0)
764 				goto write_err;
765 
766 			filled = 4;
767 		} else if (count >= 2 && !(*ppos % 2)) {
768 			u16 val;
769 
770 			if (copy_from_user(&val, buf, sizeof(val)))
771 				goto write_err;
772 
773 			ret = intel_vgpu_rw(mdev, (char *)&val,
774 					sizeof(val), ppos, true);
775 			if (ret <= 0)
776 				goto write_err;
777 
778 			filled = 2;
779 		} else {
780 			u8 val;
781 
782 			if (copy_from_user(&val, buf, sizeof(val)))
783 				goto write_err;
784 
785 			ret = intel_vgpu_rw(mdev, &val, sizeof(val),
786 					ppos, true);
787 			if (ret <= 0)
788 				goto write_err;
789 
790 			filled = 1;
791 		}
792 
793 		count -= filled;
794 		done += filled;
795 		*ppos += filled;
796 		buf += filled;
797 	}
798 
799 	return done;
800 write_err:
801 	return -EFAULT;
802 }
803 
804 static int intel_vgpu_mmap(struct mdev_device *mdev, struct vm_area_struct *vma)
805 {
806 	unsigned int index;
807 	u64 virtaddr;
808 	unsigned long req_size, pgoff = 0;
809 	pgprot_t pg_prot;
810 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
811 
812 	index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
813 	if (index >= VFIO_PCI_ROM_REGION_INDEX)
814 		return -EINVAL;
815 
816 	if (vma->vm_end < vma->vm_start)
817 		return -EINVAL;
818 	if ((vma->vm_flags & VM_SHARED) == 0)
819 		return -EINVAL;
820 	if (index != VFIO_PCI_BAR2_REGION_INDEX)
821 		return -EINVAL;
822 
823 	pg_prot = vma->vm_page_prot;
824 	virtaddr = vma->vm_start;
825 	req_size = vma->vm_end - vma->vm_start;
826 	pgoff = vgpu_aperture_pa_base(vgpu) >> PAGE_SHIFT;
827 
828 	return remap_pfn_range(vma, virtaddr, pgoff, req_size, pg_prot);
829 }
830 
831 static int intel_vgpu_get_irq_count(struct intel_vgpu *vgpu, int type)
832 {
833 	if (type == VFIO_PCI_INTX_IRQ_INDEX || type == VFIO_PCI_MSI_IRQ_INDEX)
834 		return 1;
835 
836 	return 0;
837 }
838 
839 static int intel_vgpu_set_intx_mask(struct intel_vgpu *vgpu,
840 			unsigned int index, unsigned int start,
841 			unsigned int count, uint32_t flags,
842 			void *data)
843 {
844 	return 0;
845 }
846 
847 static int intel_vgpu_set_intx_unmask(struct intel_vgpu *vgpu,
848 			unsigned int index, unsigned int start,
849 			unsigned int count, uint32_t flags, void *data)
850 {
851 	return 0;
852 }
853 
854 static int intel_vgpu_set_intx_trigger(struct intel_vgpu *vgpu,
855 		unsigned int index, unsigned int start, unsigned int count,
856 		uint32_t flags, void *data)
857 {
858 	return 0;
859 }
860 
861 static int intel_vgpu_set_msi_trigger(struct intel_vgpu *vgpu,
862 		unsigned int index, unsigned int start, unsigned int count,
863 		uint32_t flags, void *data)
864 {
865 	struct eventfd_ctx *trigger;
866 
867 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
868 		int fd = *(int *)data;
869 
870 		trigger = eventfd_ctx_fdget(fd);
871 		if (IS_ERR(trigger)) {
872 			gvt_vgpu_err("eventfd_ctx_fdget failed\n");
873 			return PTR_ERR(trigger);
874 		}
875 		vgpu->vdev.msi_trigger = trigger;
876 	}
877 
878 	return 0;
879 }
880 
881 static int intel_vgpu_set_irqs(struct intel_vgpu *vgpu, uint32_t flags,
882 		unsigned int index, unsigned int start, unsigned int count,
883 		void *data)
884 {
885 	int (*func)(struct intel_vgpu *vgpu, unsigned int index,
886 			unsigned int start, unsigned int count, uint32_t flags,
887 			void *data) = NULL;
888 
889 	switch (index) {
890 	case VFIO_PCI_INTX_IRQ_INDEX:
891 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
892 		case VFIO_IRQ_SET_ACTION_MASK:
893 			func = intel_vgpu_set_intx_mask;
894 			break;
895 		case VFIO_IRQ_SET_ACTION_UNMASK:
896 			func = intel_vgpu_set_intx_unmask;
897 			break;
898 		case VFIO_IRQ_SET_ACTION_TRIGGER:
899 			func = intel_vgpu_set_intx_trigger;
900 			break;
901 		}
902 		break;
903 	case VFIO_PCI_MSI_IRQ_INDEX:
904 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
905 		case VFIO_IRQ_SET_ACTION_MASK:
906 		case VFIO_IRQ_SET_ACTION_UNMASK:
907 			/* XXX Need masking support exported */
908 			break;
909 		case VFIO_IRQ_SET_ACTION_TRIGGER:
910 			func = intel_vgpu_set_msi_trigger;
911 			break;
912 		}
913 		break;
914 	}
915 
916 	if (!func)
917 		return -ENOTTY;
918 
919 	return func(vgpu, index, start, count, flags, data);
920 }
921 
922 static long intel_vgpu_ioctl(struct mdev_device *mdev, unsigned int cmd,
923 			     unsigned long arg)
924 {
925 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
926 	unsigned long minsz;
927 
928 	gvt_dbg_core("vgpu%d ioctl, cmd: %d\n", vgpu->id, cmd);
929 
930 	if (cmd == VFIO_DEVICE_GET_INFO) {
931 		struct vfio_device_info info;
932 
933 		minsz = offsetofend(struct vfio_device_info, num_irqs);
934 
935 		if (copy_from_user(&info, (void __user *)arg, minsz))
936 			return -EFAULT;
937 
938 		if (info.argsz < minsz)
939 			return -EINVAL;
940 
941 		info.flags = VFIO_DEVICE_FLAGS_PCI;
942 		info.flags |= VFIO_DEVICE_FLAGS_RESET;
943 		info.num_regions = VFIO_PCI_NUM_REGIONS;
944 		info.num_irqs = VFIO_PCI_NUM_IRQS;
945 
946 		return copy_to_user((void __user *)arg, &info, minsz) ?
947 			-EFAULT : 0;
948 
949 	} else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
950 		struct vfio_region_info info;
951 		struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
952 		int i, ret;
953 		struct vfio_region_info_cap_sparse_mmap *sparse = NULL;
954 		size_t size;
955 		int nr_areas = 1;
956 		int cap_type_id;
957 
958 		minsz = offsetofend(struct vfio_region_info, offset);
959 
960 		if (copy_from_user(&info, (void __user *)arg, minsz))
961 			return -EFAULT;
962 
963 		if (info.argsz < minsz)
964 			return -EINVAL;
965 
966 		switch (info.index) {
967 		case VFIO_PCI_CONFIG_REGION_INDEX:
968 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
969 			info.size = INTEL_GVT_MAX_CFG_SPACE_SZ;
970 			info.flags = VFIO_REGION_INFO_FLAG_READ |
971 				     VFIO_REGION_INFO_FLAG_WRITE;
972 			break;
973 		case VFIO_PCI_BAR0_REGION_INDEX:
974 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
975 			info.size = vgpu->cfg_space.bar[info.index].size;
976 			if (!info.size) {
977 				info.flags = 0;
978 				break;
979 			}
980 
981 			info.flags = VFIO_REGION_INFO_FLAG_READ |
982 				     VFIO_REGION_INFO_FLAG_WRITE;
983 			break;
984 		case VFIO_PCI_BAR1_REGION_INDEX:
985 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
986 			info.size = 0;
987 			info.flags = 0;
988 			break;
989 		case VFIO_PCI_BAR2_REGION_INDEX:
990 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
991 			info.flags = VFIO_REGION_INFO_FLAG_CAPS |
992 					VFIO_REGION_INFO_FLAG_MMAP |
993 					VFIO_REGION_INFO_FLAG_READ |
994 					VFIO_REGION_INFO_FLAG_WRITE;
995 			info.size = gvt_aperture_sz(vgpu->gvt);
996 
997 			size = sizeof(*sparse) +
998 					(nr_areas * sizeof(*sparse->areas));
999 			sparse = kzalloc(size, GFP_KERNEL);
1000 			if (!sparse)
1001 				return -ENOMEM;
1002 
1003 			sparse->nr_areas = nr_areas;
1004 			cap_type_id = VFIO_REGION_INFO_CAP_SPARSE_MMAP;
1005 			sparse->areas[0].offset =
1006 					PAGE_ALIGN(vgpu_aperture_offset(vgpu));
1007 			sparse->areas[0].size = vgpu_aperture_sz(vgpu);
1008 			break;
1009 
1010 		case VFIO_PCI_BAR3_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
1011 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
1012 			info.size = 0;
1013 
1014 			info.flags = 0;
1015 			gvt_dbg_core("get region info bar:%d\n", info.index);
1016 			break;
1017 
1018 		case VFIO_PCI_ROM_REGION_INDEX:
1019 		case VFIO_PCI_VGA_REGION_INDEX:
1020 			gvt_dbg_core("get region info index:%d\n", info.index);
1021 			break;
1022 		default:
1023 			{
1024 				struct vfio_region_info_cap_type cap_type;
1025 
1026 				if (info.index >= VFIO_PCI_NUM_REGIONS +
1027 						vgpu->vdev.num_regions)
1028 					return -EINVAL;
1029 
1030 				i = info.index - VFIO_PCI_NUM_REGIONS;
1031 
1032 				info.offset =
1033 					VFIO_PCI_INDEX_TO_OFFSET(info.index);
1034 				info.size = vgpu->vdev.region[i].size;
1035 				info.flags = vgpu->vdev.region[i].flags;
1036 
1037 				cap_type.type = vgpu->vdev.region[i].type;
1038 				cap_type.subtype = vgpu->vdev.region[i].subtype;
1039 
1040 				ret = vfio_info_add_capability(&caps,
1041 						VFIO_REGION_INFO_CAP_TYPE,
1042 						&cap_type);
1043 				if (ret)
1044 					return ret;
1045 			}
1046 		}
1047 
1048 		if ((info.flags & VFIO_REGION_INFO_FLAG_CAPS) && sparse) {
1049 			switch (cap_type_id) {
1050 			case VFIO_REGION_INFO_CAP_SPARSE_MMAP:
1051 				ret = vfio_info_add_capability(&caps,
1052 					VFIO_REGION_INFO_CAP_SPARSE_MMAP,
1053 					sparse);
1054 				kfree(sparse);
1055 				if (ret)
1056 					return ret;
1057 				break;
1058 			default:
1059 				return -EINVAL;
1060 			}
1061 		}
1062 
1063 		if (caps.size) {
1064 			if (info.argsz < sizeof(info) + caps.size) {
1065 				info.argsz = sizeof(info) + caps.size;
1066 				info.cap_offset = 0;
1067 			} else {
1068 				vfio_info_cap_shift(&caps, sizeof(info));
1069 				if (copy_to_user((void __user *)arg +
1070 						  sizeof(info), caps.buf,
1071 						  caps.size)) {
1072 					kfree(caps.buf);
1073 					return -EFAULT;
1074 				}
1075 				info.cap_offset = sizeof(info);
1076 			}
1077 
1078 			kfree(caps.buf);
1079 		}
1080 
1081 		return copy_to_user((void __user *)arg, &info, minsz) ?
1082 			-EFAULT : 0;
1083 	} else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
1084 		struct vfio_irq_info info;
1085 
1086 		minsz = offsetofend(struct vfio_irq_info, count);
1087 
1088 		if (copy_from_user(&info, (void __user *)arg, minsz))
1089 			return -EFAULT;
1090 
1091 		if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
1092 			return -EINVAL;
1093 
1094 		switch (info.index) {
1095 		case VFIO_PCI_INTX_IRQ_INDEX:
1096 		case VFIO_PCI_MSI_IRQ_INDEX:
1097 			break;
1098 		default:
1099 			return -EINVAL;
1100 		}
1101 
1102 		info.flags = VFIO_IRQ_INFO_EVENTFD;
1103 
1104 		info.count = intel_vgpu_get_irq_count(vgpu, info.index);
1105 
1106 		if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
1107 			info.flags |= (VFIO_IRQ_INFO_MASKABLE |
1108 				       VFIO_IRQ_INFO_AUTOMASKED);
1109 		else
1110 			info.flags |= VFIO_IRQ_INFO_NORESIZE;
1111 
1112 		return copy_to_user((void __user *)arg, &info, minsz) ?
1113 			-EFAULT : 0;
1114 	} else if (cmd == VFIO_DEVICE_SET_IRQS) {
1115 		struct vfio_irq_set hdr;
1116 		u8 *data = NULL;
1117 		int ret = 0;
1118 		size_t data_size = 0;
1119 
1120 		minsz = offsetofend(struct vfio_irq_set, count);
1121 
1122 		if (copy_from_user(&hdr, (void __user *)arg, minsz))
1123 			return -EFAULT;
1124 
1125 		if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) {
1126 			int max = intel_vgpu_get_irq_count(vgpu, hdr.index);
1127 
1128 			ret = vfio_set_irqs_validate_and_prepare(&hdr, max,
1129 						VFIO_PCI_NUM_IRQS, &data_size);
1130 			if (ret) {
1131 				gvt_vgpu_err("intel:vfio_set_irqs_validate_and_prepare failed\n");
1132 				return -EINVAL;
1133 			}
1134 			if (data_size) {
1135 				data = memdup_user((void __user *)(arg + minsz),
1136 						   data_size);
1137 				if (IS_ERR(data))
1138 					return PTR_ERR(data);
1139 			}
1140 		}
1141 
1142 		ret = intel_vgpu_set_irqs(vgpu, hdr.flags, hdr.index,
1143 					hdr.start, hdr.count, data);
1144 		kfree(data);
1145 
1146 		return ret;
1147 	} else if (cmd == VFIO_DEVICE_RESET) {
1148 		intel_gvt_ops->vgpu_reset(vgpu);
1149 		return 0;
1150 	}
1151 
1152 	return 0;
1153 }
1154 
1155 static ssize_t
1156 vgpu_id_show(struct device *dev, struct device_attribute *attr,
1157 	     char *buf)
1158 {
1159 	struct mdev_device *mdev = mdev_from_dev(dev);
1160 
1161 	if (mdev) {
1162 		struct intel_vgpu *vgpu = (struct intel_vgpu *)
1163 			mdev_get_drvdata(mdev);
1164 		return sprintf(buf, "%d\n", vgpu->id);
1165 	}
1166 	return sprintf(buf, "\n");
1167 }
1168 
1169 static DEVICE_ATTR_RO(vgpu_id);
1170 
1171 static struct attribute *intel_vgpu_attrs[] = {
1172 	&dev_attr_vgpu_id.attr,
1173 	NULL
1174 };
1175 
1176 static const struct attribute_group intel_vgpu_group = {
1177 	.name = "intel_vgpu",
1178 	.attrs = intel_vgpu_attrs,
1179 };
1180 
1181 static const struct attribute_group *intel_vgpu_groups[] = {
1182 	&intel_vgpu_group,
1183 	NULL,
1184 };
1185 
1186 static const struct mdev_parent_ops intel_vgpu_ops = {
1187 	.supported_type_groups	= intel_vgpu_type_groups,
1188 	.mdev_attr_groups       = intel_vgpu_groups,
1189 	.create			= intel_vgpu_create,
1190 	.remove			= intel_vgpu_remove,
1191 
1192 	.open			= intel_vgpu_open,
1193 	.release		= intel_vgpu_release,
1194 
1195 	.read			= intel_vgpu_read,
1196 	.write			= intel_vgpu_write,
1197 	.mmap			= intel_vgpu_mmap,
1198 	.ioctl			= intel_vgpu_ioctl,
1199 };
1200 
1201 static int kvmgt_host_init(struct device *dev, void *gvt, const void *ops)
1202 {
1203 	if (!intel_gvt_init_vgpu_type_groups(gvt))
1204 		return -EFAULT;
1205 
1206 	intel_gvt_ops = ops;
1207 
1208 	return mdev_register_device(dev, &intel_vgpu_ops);
1209 }
1210 
1211 static void kvmgt_host_exit(struct device *dev, void *gvt)
1212 {
1213 	intel_gvt_cleanup_vgpu_type_groups(gvt);
1214 	mdev_unregister_device(dev);
1215 }
1216 
1217 static int kvmgt_write_protect_add(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 -ESRCH;
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_add_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1243 	kvmgt_protect_table_add(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 int kvmgt_write_protect_remove(unsigned long handle, u64 gfn)
1252 {
1253 	struct kvmgt_guest_info *info;
1254 	struct kvm *kvm;
1255 	struct kvm_memory_slot *slot;
1256 	int idx;
1257 
1258 	if (!handle_valid(handle))
1259 		return 0;
1260 
1261 	info = (struct kvmgt_guest_info *)handle;
1262 	kvm = info->kvm;
1263 
1264 	idx = srcu_read_lock(&kvm->srcu);
1265 	slot = gfn_to_memslot(kvm, gfn);
1266 	if (!slot) {
1267 		srcu_read_unlock(&kvm->srcu, idx);
1268 		return -EINVAL;
1269 	}
1270 
1271 	spin_lock(&kvm->mmu_lock);
1272 
1273 	if (!kvmgt_gfn_is_write_protected(info, gfn))
1274 		goto out;
1275 
1276 	kvm_slot_page_track_remove_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1277 	kvmgt_protect_table_del(info, gfn);
1278 
1279 out:
1280 	spin_unlock(&kvm->mmu_lock);
1281 	srcu_read_unlock(&kvm->srcu, idx);
1282 	return 0;
1283 }
1284 
1285 static void kvmgt_page_track_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1286 		const u8 *val, int len,
1287 		struct kvm_page_track_notifier_node *node)
1288 {
1289 	struct kvmgt_guest_info *info = container_of(node,
1290 					struct kvmgt_guest_info, track_node);
1291 
1292 	if (kvmgt_gfn_is_write_protected(info, gpa_to_gfn(gpa)))
1293 		intel_gvt_ops->emulate_mmio_write(info->vgpu, gpa,
1294 					(void *)val, len);
1295 }
1296 
1297 static void kvmgt_page_track_flush_slot(struct kvm *kvm,
1298 		struct kvm_memory_slot *slot,
1299 		struct kvm_page_track_notifier_node *node)
1300 {
1301 	int i;
1302 	gfn_t gfn;
1303 	struct kvmgt_guest_info *info = container_of(node,
1304 					struct kvmgt_guest_info, track_node);
1305 
1306 	spin_lock(&kvm->mmu_lock);
1307 	for (i = 0; i < slot->npages; i++) {
1308 		gfn = slot->base_gfn + i;
1309 		if (kvmgt_gfn_is_write_protected(info, gfn)) {
1310 			kvm_slot_page_track_remove_page(kvm, slot, gfn,
1311 						KVM_PAGE_TRACK_WRITE);
1312 			kvmgt_protect_table_del(info, gfn);
1313 		}
1314 	}
1315 	spin_unlock(&kvm->mmu_lock);
1316 }
1317 
1318 static bool __kvmgt_vgpu_exist(struct intel_vgpu *vgpu, struct kvm *kvm)
1319 {
1320 	struct intel_vgpu *itr;
1321 	struct kvmgt_guest_info *info;
1322 	int id;
1323 	bool ret = false;
1324 
1325 	mutex_lock(&vgpu->gvt->lock);
1326 	for_each_active_vgpu(vgpu->gvt, itr, id) {
1327 		if (!handle_valid(itr->handle))
1328 			continue;
1329 
1330 		info = (struct kvmgt_guest_info *)itr->handle;
1331 		if (kvm && kvm == info->kvm) {
1332 			ret = true;
1333 			goto out;
1334 		}
1335 	}
1336 out:
1337 	mutex_unlock(&vgpu->gvt->lock);
1338 	return ret;
1339 }
1340 
1341 static int kvmgt_guest_init(struct mdev_device *mdev)
1342 {
1343 	struct kvmgt_guest_info *info;
1344 	struct intel_vgpu *vgpu;
1345 	struct kvm *kvm;
1346 
1347 	vgpu = mdev_get_drvdata(mdev);
1348 	if (handle_valid(vgpu->handle))
1349 		return -EEXIST;
1350 
1351 	kvm = vgpu->vdev.kvm;
1352 	if (!kvm || kvm->mm != current->mm) {
1353 		gvt_vgpu_err("KVM is required to use Intel vGPU\n");
1354 		return -ESRCH;
1355 	}
1356 
1357 	if (__kvmgt_vgpu_exist(vgpu, kvm))
1358 		return -EEXIST;
1359 
1360 	info = vzalloc(sizeof(struct kvmgt_guest_info));
1361 	if (!info)
1362 		return -ENOMEM;
1363 
1364 	vgpu->handle = (unsigned long)info;
1365 	info->vgpu = vgpu;
1366 	info->kvm = kvm;
1367 	kvm_get_kvm(info->kvm);
1368 
1369 	kvmgt_protect_table_init(info);
1370 	gvt_cache_init(vgpu);
1371 
1372 	info->track_node.track_write = kvmgt_page_track_write;
1373 	info->track_node.track_flush_slot = kvmgt_page_track_flush_slot;
1374 	kvm_page_track_register_notifier(kvm, &info->track_node);
1375 
1376 	return 0;
1377 }
1378 
1379 static bool kvmgt_guest_exit(struct kvmgt_guest_info *info)
1380 {
1381 	kvm_page_track_unregister_notifier(info->kvm, &info->track_node);
1382 	kvm_put_kvm(info->kvm);
1383 	kvmgt_protect_table_destroy(info);
1384 	gvt_cache_destroy(info->vgpu);
1385 	vfree(info);
1386 
1387 	return true;
1388 }
1389 
1390 static int kvmgt_attach_vgpu(void *vgpu, unsigned long *handle)
1391 {
1392 	/* nothing to do here */
1393 	return 0;
1394 }
1395 
1396 static void kvmgt_detach_vgpu(unsigned long handle)
1397 {
1398 	/* nothing to do here */
1399 }
1400 
1401 static int kvmgt_inject_msi(unsigned long handle, u32 addr, u16 data)
1402 {
1403 	struct kvmgt_guest_info *info;
1404 	struct intel_vgpu *vgpu;
1405 
1406 	if (!handle_valid(handle))
1407 		return -ESRCH;
1408 
1409 	info = (struct kvmgt_guest_info *)handle;
1410 	vgpu = info->vgpu;
1411 
1412 	if (eventfd_signal(vgpu->vdev.msi_trigger, 1) == 1)
1413 		return 0;
1414 
1415 	return -EFAULT;
1416 }
1417 
1418 static unsigned long kvmgt_gfn_to_pfn(unsigned long handle, unsigned long gfn)
1419 {
1420 	unsigned long iova, pfn;
1421 	struct kvmgt_guest_info *info;
1422 	struct device *dev;
1423 	struct intel_vgpu *vgpu;
1424 	int rc;
1425 
1426 	if (!handle_valid(handle))
1427 		return INTEL_GVT_INVALID_ADDR;
1428 
1429 	info = (struct kvmgt_guest_info *)handle;
1430 	vgpu = info->vgpu;
1431 	iova = gvt_cache_find(info->vgpu, gfn);
1432 	if (iova != INTEL_GVT_INVALID_ADDR)
1433 		return iova;
1434 
1435 	pfn = INTEL_GVT_INVALID_ADDR;
1436 	dev = mdev_dev(info->vgpu->vdev.mdev);
1437 	rc = vfio_pin_pages(dev, &gfn, 1, IOMMU_READ | IOMMU_WRITE, &pfn);
1438 	if (rc != 1) {
1439 		gvt_vgpu_err("vfio_pin_pages failed for gfn 0x%lx: %d\n",
1440 			gfn, rc);
1441 		return INTEL_GVT_INVALID_ADDR;
1442 	}
1443 	/* transfer to host iova for GFX to use DMA */
1444 	rc = gvt_dma_map_iova(info->vgpu, pfn, &iova);
1445 	if (rc) {
1446 		gvt_vgpu_err("gvt_dma_map_iova failed for gfn: 0x%lx\n", gfn);
1447 		vfio_unpin_pages(dev, &gfn, 1);
1448 		return INTEL_GVT_INVALID_ADDR;
1449 	}
1450 
1451 	gvt_cache_add(info->vgpu, gfn, iova);
1452 	return iova;
1453 }
1454 
1455 static int kvmgt_rw_gpa(unsigned long handle, unsigned long gpa,
1456 			void *buf, unsigned long len, bool write)
1457 {
1458 	struct kvmgt_guest_info *info;
1459 	struct kvm *kvm;
1460 	int idx, ret;
1461 	bool kthread = current->mm == NULL;
1462 
1463 	if (!handle_valid(handle))
1464 		return -ESRCH;
1465 
1466 	info = (struct kvmgt_guest_info *)handle;
1467 	kvm = info->kvm;
1468 
1469 	if (kthread)
1470 		use_mm(kvm->mm);
1471 
1472 	idx = srcu_read_lock(&kvm->srcu);
1473 	ret = write ? kvm_write_guest(kvm, gpa, buf, len) :
1474 		      kvm_read_guest(kvm, gpa, buf, len);
1475 	srcu_read_unlock(&kvm->srcu, idx);
1476 
1477 	if (kthread)
1478 		unuse_mm(kvm->mm);
1479 
1480 	return ret;
1481 }
1482 
1483 static int kvmgt_read_gpa(unsigned long handle, unsigned long gpa,
1484 			void *buf, unsigned long len)
1485 {
1486 	return kvmgt_rw_gpa(handle, gpa, buf, len, false);
1487 }
1488 
1489 static int kvmgt_write_gpa(unsigned long handle, unsigned long gpa,
1490 			void *buf, unsigned long len)
1491 {
1492 	return kvmgt_rw_gpa(handle, gpa, buf, len, true);
1493 }
1494 
1495 static unsigned long kvmgt_virt_to_pfn(void *addr)
1496 {
1497 	return PFN_DOWN(__pa(addr));
1498 }
1499 
1500 struct intel_gvt_mpt kvmgt_mpt = {
1501 	.host_init = kvmgt_host_init,
1502 	.host_exit = kvmgt_host_exit,
1503 	.attach_vgpu = kvmgt_attach_vgpu,
1504 	.detach_vgpu = kvmgt_detach_vgpu,
1505 	.inject_msi = kvmgt_inject_msi,
1506 	.from_virt_to_mfn = kvmgt_virt_to_pfn,
1507 	.set_wp_page = kvmgt_write_protect_add,
1508 	.unset_wp_page = kvmgt_write_protect_remove,
1509 	.read_gpa = kvmgt_read_gpa,
1510 	.write_gpa = kvmgt_write_gpa,
1511 	.gfn_to_mfn = kvmgt_gfn_to_pfn,
1512 };
1513 EXPORT_SYMBOL_GPL(kvmgt_mpt);
1514 
1515 static int __init kvmgt_init(void)
1516 {
1517 	return 0;
1518 }
1519 
1520 static void __exit kvmgt_exit(void)
1521 {
1522 }
1523 
1524 module_init(kvmgt_init);
1525 module_exit(kvmgt_exit);
1526 
1527 MODULE_LICENSE("GPL and additional rights");
1528 MODULE_AUTHOR("Intel Corporation");
1529