xref: /openbmc/linux/drivers/gpu/drm/i915/gvt/kvmgt.c (revision 232b0b08)
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 	atomic_set(&vgpu->vdev.released, 0);
548 	return ret;
549 
550 undo_group:
551 	vfio_unregister_notifier(mdev_dev(mdev), VFIO_GROUP_NOTIFY,
552 					&vgpu->vdev.group_notifier);
553 
554 undo_iommu:
555 	vfio_unregister_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY,
556 					&vgpu->vdev.iommu_notifier);
557 out:
558 	return ret;
559 }
560 
561 static void __intel_vgpu_release(struct intel_vgpu *vgpu)
562 {
563 	struct kvmgt_guest_info *info;
564 	int ret;
565 
566 	if (!handle_valid(vgpu->handle))
567 		return;
568 
569 	if (atomic_cmpxchg(&vgpu->vdev.released, 0, 1))
570 		return;
571 
572 	ret = vfio_unregister_notifier(mdev_dev(vgpu->vdev.mdev), VFIO_IOMMU_NOTIFY,
573 					&vgpu->vdev.iommu_notifier);
574 	WARN(ret, "vfio_unregister_notifier for iommu failed: %d\n", ret);
575 
576 	ret = vfio_unregister_notifier(mdev_dev(vgpu->vdev.mdev), VFIO_GROUP_NOTIFY,
577 					&vgpu->vdev.group_notifier);
578 	WARN(ret, "vfio_unregister_notifier for group failed: %d\n", ret);
579 
580 	info = (struct kvmgt_guest_info *)vgpu->handle;
581 	kvmgt_guest_exit(info);
582 
583 	vgpu->vdev.kvm = NULL;
584 	vgpu->handle = 0;
585 }
586 
587 static void intel_vgpu_release(struct mdev_device *mdev)
588 {
589 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
590 
591 	__intel_vgpu_release(vgpu);
592 }
593 
594 static void intel_vgpu_release_work(struct work_struct *work)
595 {
596 	struct intel_vgpu *vgpu = container_of(work, struct intel_vgpu,
597 					vdev.release_work);
598 
599 	__intel_vgpu_release(vgpu);
600 }
601 
602 static uint64_t intel_vgpu_get_bar0_addr(struct intel_vgpu *vgpu)
603 {
604 	u32 start_lo, start_hi;
605 	u32 mem_type;
606 	int pos = PCI_BASE_ADDRESS_0;
607 
608 	start_lo = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + pos)) &
609 			PCI_BASE_ADDRESS_MEM_MASK;
610 	mem_type = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + pos)) &
611 			PCI_BASE_ADDRESS_MEM_TYPE_MASK;
612 
613 	switch (mem_type) {
614 	case PCI_BASE_ADDRESS_MEM_TYPE_64:
615 		start_hi = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space
616 						+ pos + 4));
617 		break;
618 	case PCI_BASE_ADDRESS_MEM_TYPE_32:
619 	case PCI_BASE_ADDRESS_MEM_TYPE_1M:
620 		/* 1M mem BAR treated as 32-bit BAR */
621 	default:
622 		/* mem unknown type treated as 32-bit BAR */
623 		start_hi = 0;
624 		break;
625 	}
626 
627 	return ((u64)start_hi << 32) | start_lo;
628 }
629 
630 static ssize_t intel_vgpu_rw(struct mdev_device *mdev, char *buf,
631 			size_t count, loff_t *ppos, bool is_write)
632 {
633 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
634 	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
635 	uint64_t pos = *ppos & VFIO_PCI_OFFSET_MASK;
636 	int ret = -EINVAL;
637 
638 
639 	if (index >= VFIO_PCI_NUM_REGIONS) {
640 		gvt_vgpu_err("invalid index: %u\n", index);
641 		return -EINVAL;
642 	}
643 
644 	switch (index) {
645 	case VFIO_PCI_CONFIG_REGION_INDEX:
646 		if (is_write)
647 			ret = intel_gvt_ops->emulate_cfg_write(vgpu, pos,
648 						buf, count);
649 		else
650 			ret = intel_gvt_ops->emulate_cfg_read(vgpu, pos,
651 						buf, count);
652 		break;
653 	case VFIO_PCI_BAR0_REGION_INDEX:
654 	case VFIO_PCI_BAR1_REGION_INDEX:
655 		if (is_write) {
656 			uint64_t bar0_start = intel_vgpu_get_bar0_addr(vgpu);
657 
658 			ret = intel_gvt_ops->emulate_mmio_write(vgpu,
659 						bar0_start + pos, buf, count);
660 		} else {
661 			uint64_t bar0_start = intel_vgpu_get_bar0_addr(vgpu);
662 
663 			ret = intel_gvt_ops->emulate_mmio_read(vgpu,
664 						bar0_start + pos, buf, count);
665 		}
666 		break;
667 	case VFIO_PCI_BAR2_REGION_INDEX:
668 	case VFIO_PCI_BAR3_REGION_INDEX:
669 	case VFIO_PCI_BAR4_REGION_INDEX:
670 	case VFIO_PCI_BAR5_REGION_INDEX:
671 	case VFIO_PCI_VGA_REGION_INDEX:
672 	case VFIO_PCI_ROM_REGION_INDEX:
673 	default:
674 		gvt_vgpu_err("unsupported region: %u\n", index);
675 	}
676 
677 	return ret == 0 ? count : ret;
678 }
679 
680 static ssize_t intel_vgpu_read(struct mdev_device *mdev, char __user *buf,
681 			size_t count, loff_t *ppos)
682 {
683 	unsigned int done = 0;
684 	int ret;
685 
686 	while (count) {
687 		size_t filled;
688 
689 		if (count >= 4 && !(*ppos % 4)) {
690 			u32 val;
691 
692 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
693 					ppos, false);
694 			if (ret <= 0)
695 				goto read_err;
696 
697 			if (copy_to_user(buf, &val, sizeof(val)))
698 				goto read_err;
699 
700 			filled = 4;
701 		} else if (count >= 2 && !(*ppos % 2)) {
702 			u16 val;
703 
704 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
705 					ppos, false);
706 			if (ret <= 0)
707 				goto read_err;
708 
709 			if (copy_to_user(buf, &val, sizeof(val)))
710 				goto read_err;
711 
712 			filled = 2;
713 		} else {
714 			u8 val;
715 
716 			ret = intel_vgpu_rw(mdev, &val, sizeof(val), ppos,
717 					false);
718 			if (ret <= 0)
719 				goto read_err;
720 
721 			if (copy_to_user(buf, &val, sizeof(val)))
722 				goto read_err;
723 
724 			filled = 1;
725 		}
726 
727 		count -= filled;
728 		done += filled;
729 		*ppos += filled;
730 		buf += filled;
731 	}
732 
733 	return done;
734 
735 read_err:
736 	return -EFAULT;
737 }
738 
739 static ssize_t intel_vgpu_write(struct mdev_device *mdev,
740 				const char __user *buf,
741 				size_t count, loff_t *ppos)
742 {
743 	unsigned int done = 0;
744 	int ret;
745 
746 	while (count) {
747 		size_t filled;
748 
749 		if (count >= 4 && !(*ppos % 4)) {
750 			u32 val;
751 
752 			if (copy_from_user(&val, buf, sizeof(val)))
753 				goto write_err;
754 
755 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
756 					ppos, true);
757 			if (ret <= 0)
758 				goto write_err;
759 
760 			filled = 4;
761 		} else if (count >= 2 && !(*ppos % 2)) {
762 			u16 val;
763 
764 			if (copy_from_user(&val, buf, sizeof(val)))
765 				goto write_err;
766 
767 			ret = intel_vgpu_rw(mdev, (char *)&val,
768 					sizeof(val), ppos, true);
769 			if (ret <= 0)
770 				goto write_err;
771 
772 			filled = 2;
773 		} else {
774 			u8 val;
775 
776 			if (copy_from_user(&val, buf, sizeof(val)))
777 				goto write_err;
778 
779 			ret = intel_vgpu_rw(mdev, &val, sizeof(val),
780 					ppos, true);
781 			if (ret <= 0)
782 				goto write_err;
783 
784 			filled = 1;
785 		}
786 
787 		count -= filled;
788 		done += filled;
789 		*ppos += filled;
790 		buf += filled;
791 	}
792 
793 	return done;
794 write_err:
795 	return -EFAULT;
796 }
797 
798 static int intel_vgpu_mmap(struct mdev_device *mdev, struct vm_area_struct *vma)
799 {
800 	unsigned int index;
801 	u64 virtaddr;
802 	unsigned long req_size, pgoff = 0;
803 	pgprot_t pg_prot;
804 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
805 
806 	index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
807 	if (index >= VFIO_PCI_ROM_REGION_INDEX)
808 		return -EINVAL;
809 
810 	if (vma->vm_end < vma->vm_start)
811 		return -EINVAL;
812 	if ((vma->vm_flags & VM_SHARED) == 0)
813 		return -EINVAL;
814 	if (index != VFIO_PCI_BAR2_REGION_INDEX)
815 		return -EINVAL;
816 
817 	pg_prot = vma->vm_page_prot;
818 	virtaddr = vma->vm_start;
819 	req_size = vma->vm_end - vma->vm_start;
820 	pgoff = vgpu_aperture_pa_base(vgpu) >> PAGE_SHIFT;
821 
822 	return remap_pfn_range(vma, virtaddr, pgoff, req_size, pg_prot);
823 }
824 
825 static int intel_vgpu_get_irq_count(struct intel_vgpu *vgpu, int type)
826 {
827 	if (type == VFIO_PCI_INTX_IRQ_INDEX || type == VFIO_PCI_MSI_IRQ_INDEX)
828 		return 1;
829 
830 	return 0;
831 }
832 
833 static int intel_vgpu_set_intx_mask(struct intel_vgpu *vgpu,
834 			unsigned int index, unsigned int start,
835 			unsigned int count, uint32_t flags,
836 			void *data)
837 {
838 	return 0;
839 }
840 
841 static int intel_vgpu_set_intx_unmask(struct intel_vgpu *vgpu,
842 			unsigned int index, unsigned int start,
843 			unsigned int count, uint32_t flags, void *data)
844 {
845 	return 0;
846 }
847 
848 static int intel_vgpu_set_intx_trigger(struct intel_vgpu *vgpu,
849 		unsigned int index, unsigned int start, unsigned int count,
850 		uint32_t flags, void *data)
851 {
852 	return 0;
853 }
854 
855 static int intel_vgpu_set_msi_trigger(struct intel_vgpu *vgpu,
856 		unsigned int index, unsigned int start, unsigned int count,
857 		uint32_t flags, void *data)
858 {
859 	struct eventfd_ctx *trigger;
860 
861 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
862 		int fd = *(int *)data;
863 
864 		trigger = eventfd_ctx_fdget(fd);
865 		if (IS_ERR(trigger)) {
866 			gvt_vgpu_err("eventfd_ctx_fdget failed\n");
867 			return PTR_ERR(trigger);
868 		}
869 		vgpu->vdev.msi_trigger = trigger;
870 	}
871 
872 	return 0;
873 }
874 
875 static int intel_vgpu_set_irqs(struct intel_vgpu *vgpu, uint32_t flags,
876 		unsigned int index, unsigned int start, unsigned int count,
877 		void *data)
878 {
879 	int (*func)(struct intel_vgpu *vgpu, unsigned int index,
880 			unsigned int start, unsigned int count, uint32_t flags,
881 			void *data) = NULL;
882 
883 	switch (index) {
884 	case VFIO_PCI_INTX_IRQ_INDEX:
885 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
886 		case VFIO_IRQ_SET_ACTION_MASK:
887 			func = intel_vgpu_set_intx_mask;
888 			break;
889 		case VFIO_IRQ_SET_ACTION_UNMASK:
890 			func = intel_vgpu_set_intx_unmask;
891 			break;
892 		case VFIO_IRQ_SET_ACTION_TRIGGER:
893 			func = intel_vgpu_set_intx_trigger;
894 			break;
895 		}
896 		break;
897 	case VFIO_PCI_MSI_IRQ_INDEX:
898 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
899 		case VFIO_IRQ_SET_ACTION_MASK:
900 		case VFIO_IRQ_SET_ACTION_UNMASK:
901 			/* XXX Need masking support exported */
902 			break;
903 		case VFIO_IRQ_SET_ACTION_TRIGGER:
904 			func = intel_vgpu_set_msi_trigger;
905 			break;
906 		}
907 		break;
908 	}
909 
910 	if (!func)
911 		return -ENOTTY;
912 
913 	return func(vgpu, index, start, count, flags, data);
914 }
915 
916 static long intel_vgpu_ioctl(struct mdev_device *mdev, unsigned int cmd,
917 			     unsigned long arg)
918 {
919 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
920 	unsigned long minsz;
921 
922 	gvt_dbg_core("vgpu%d ioctl, cmd: %d\n", vgpu->id, cmd);
923 
924 	if (cmd == VFIO_DEVICE_GET_INFO) {
925 		struct vfio_device_info info;
926 
927 		minsz = offsetofend(struct vfio_device_info, num_irqs);
928 
929 		if (copy_from_user(&info, (void __user *)arg, minsz))
930 			return -EFAULT;
931 
932 		if (info.argsz < minsz)
933 			return -EINVAL;
934 
935 		info.flags = VFIO_DEVICE_FLAGS_PCI;
936 		info.flags |= VFIO_DEVICE_FLAGS_RESET;
937 		info.num_regions = VFIO_PCI_NUM_REGIONS;
938 		info.num_irqs = VFIO_PCI_NUM_IRQS;
939 
940 		return copy_to_user((void __user *)arg, &info, minsz) ?
941 			-EFAULT : 0;
942 
943 	} else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
944 		struct vfio_region_info info;
945 		struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
946 		int i, ret;
947 		struct vfio_region_info_cap_sparse_mmap *sparse = NULL;
948 		size_t size;
949 		int nr_areas = 1;
950 		int cap_type_id;
951 
952 		minsz = offsetofend(struct vfio_region_info, offset);
953 
954 		if (copy_from_user(&info, (void __user *)arg, minsz))
955 			return -EFAULT;
956 
957 		if (info.argsz < minsz)
958 			return -EINVAL;
959 
960 		switch (info.index) {
961 		case VFIO_PCI_CONFIG_REGION_INDEX:
962 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
963 			info.size = INTEL_GVT_MAX_CFG_SPACE_SZ;
964 			info.flags = VFIO_REGION_INFO_FLAG_READ |
965 				     VFIO_REGION_INFO_FLAG_WRITE;
966 			break;
967 		case VFIO_PCI_BAR0_REGION_INDEX:
968 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
969 			info.size = vgpu->cfg_space.bar[info.index].size;
970 			if (!info.size) {
971 				info.flags = 0;
972 				break;
973 			}
974 
975 			info.flags = VFIO_REGION_INFO_FLAG_READ |
976 				     VFIO_REGION_INFO_FLAG_WRITE;
977 			break;
978 		case VFIO_PCI_BAR1_REGION_INDEX:
979 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
980 			info.size = 0;
981 			info.flags = 0;
982 			break;
983 		case VFIO_PCI_BAR2_REGION_INDEX:
984 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
985 			info.flags = VFIO_REGION_INFO_FLAG_CAPS |
986 					VFIO_REGION_INFO_FLAG_MMAP |
987 					VFIO_REGION_INFO_FLAG_READ |
988 					VFIO_REGION_INFO_FLAG_WRITE;
989 			info.size = gvt_aperture_sz(vgpu->gvt);
990 
991 			size = sizeof(*sparse) +
992 					(nr_areas * sizeof(*sparse->areas));
993 			sparse = kzalloc(size, GFP_KERNEL);
994 			if (!sparse)
995 				return -ENOMEM;
996 
997 			sparse->nr_areas = nr_areas;
998 			cap_type_id = VFIO_REGION_INFO_CAP_SPARSE_MMAP;
999 			sparse->areas[0].offset =
1000 					PAGE_ALIGN(vgpu_aperture_offset(vgpu));
1001 			sparse->areas[0].size = vgpu_aperture_sz(vgpu);
1002 			break;
1003 
1004 		case VFIO_PCI_BAR3_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
1005 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
1006 			info.size = 0;
1007 
1008 			info.flags = 0;
1009 			gvt_dbg_core("get region info bar:%d\n", info.index);
1010 			break;
1011 
1012 		case VFIO_PCI_ROM_REGION_INDEX:
1013 		case VFIO_PCI_VGA_REGION_INDEX:
1014 			gvt_dbg_core("get region info index:%d\n", info.index);
1015 			break;
1016 		default:
1017 			{
1018 				struct vfio_region_info_cap_type cap_type;
1019 
1020 				if (info.index >= VFIO_PCI_NUM_REGIONS +
1021 						vgpu->vdev.num_regions)
1022 					return -EINVAL;
1023 
1024 				i = info.index - VFIO_PCI_NUM_REGIONS;
1025 
1026 				info.offset =
1027 					VFIO_PCI_INDEX_TO_OFFSET(info.index);
1028 				info.size = vgpu->vdev.region[i].size;
1029 				info.flags = vgpu->vdev.region[i].flags;
1030 
1031 				cap_type.type = vgpu->vdev.region[i].type;
1032 				cap_type.subtype = vgpu->vdev.region[i].subtype;
1033 
1034 				ret = vfio_info_add_capability(&caps,
1035 						VFIO_REGION_INFO_CAP_TYPE,
1036 						&cap_type);
1037 				if (ret)
1038 					return ret;
1039 			}
1040 		}
1041 
1042 		if ((info.flags & VFIO_REGION_INFO_FLAG_CAPS) && sparse) {
1043 			switch (cap_type_id) {
1044 			case VFIO_REGION_INFO_CAP_SPARSE_MMAP:
1045 				ret = vfio_info_add_capability(&caps,
1046 					VFIO_REGION_INFO_CAP_SPARSE_MMAP,
1047 					sparse);
1048 				kfree(sparse);
1049 				if (ret)
1050 					return ret;
1051 				break;
1052 			default:
1053 				return -EINVAL;
1054 			}
1055 		}
1056 
1057 		if (caps.size) {
1058 			if (info.argsz < sizeof(info) + caps.size) {
1059 				info.argsz = sizeof(info) + caps.size;
1060 				info.cap_offset = 0;
1061 			} else {
1062 				vfio_info_cap_shift(&caps, sizeof(info));
1063 				if (copy_to_user((void __user *)arg +
1064 						  sizeof(info), caps.buf,
1065 						  caps.size)) {
1066 					kfree(caps.buf);
1067 					return -EFAULT;
1068 				}
1069 				info.cap_offset = sizeof(info);
1070 			}
1071 
1072 			kfree(caps.buf);
1073 		}
1074 
1075 		return copy_to_user((void __user *)arg, &info, minsz) ?
1076 			-EFAULT : 0;
1077 	} else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
1078 		struct vfio_irq_info info;
1079 
1080 		minsz = offsetofend(struct vfio_irq_info, count);
1081 
1082 		if (copy_from_user(&info, (void __user *)arg, minsz))
1083 			return -EFAULT;
1084 
1085 		if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
1086 			return -EINVAL;
1087 
1088 		switch (info.index) {
1089 		case VFIO_PCI_INTX_IRQ_INDEX:
1090 		case VFIO_PCI_MSI_IRQ_INDEX:
1091 			break;
1092 		default:
1093 			return -EINVAL;
1094 		}
1095 
1096 		info.flags = VFIO_IRQ_INFO_EVENTFD;
1097 
1098 		info.count = intel_vgpu_get_irq_count(vgpu, info.index);
1099 
1100 		if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
1101 			info.flags |= (VFIO_IRQ_INFO_MASKABLE |
1102 				       VFIO_IRQ_INFO_AUTOMASKED);
1103 		else
1104 			info.flags |= VFIO_IRQ_INFO_NORESIZE;
1105 
1106 		return copy_to_user((void __user *)arg, &info, minsz) ?
1107 			-EFAULT : 0;
1108 	} else if (cmd == VFIO_DEVICE_SET_IRQS) {
1109 		struct vfio_irq_set hdr;
1110 		u8 *data = NULL;
1111 		int ret = 0;
1112 		size_t data_size = 0;
1113 
1114 		minsz = offsetofend(struct vfio_irq_set, count);
1115 
1116 		if (copy_from_user(&hdr, (void __user *)arg, minsz))
1117 			return -EFAULT;
1118 
1119 		if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) {
1120 			int max = intel_vgpu_get_irq_count(vgpu, hdr.index);
1121 
1122 			ret = vfio_set_irqs_validate_and_prepare(&hdr, max,
1123 						VFIO_PCI_NUM_IRQS, &data_size);
1124 			if (ret) {
1125 				gvt_vgpu_err("intel:vfio_set_irqs_validate_and_prepare failed\n");
1126 				return -EINVAL;
1127 			}
1128 			if (data_size) {
1129 				data = memdup_user((void __user *)(arg + minsz),
1130 						   data_size);
1131 				if (IS_ERR(data))
1132 					return PTR_ERR(data);
1133 			}
1134 		}
1135 
1136 		ret = intel_vgpu_set_irqs(vgpu, hdr.flags, hdr.index,
1137 					hdr.start, hdr.count, data);
1138 		kfree(data);
1139 
1140 		return ret;
1141 	} else if (cmd == VFIO_DEVICE_RESET) {
1142 		intel_gvt_ops->vgpu_reset(vgpu);
1143 		return 0;
1144 	}
1145 
1146 	return 0;
1147 }
1148 
1149 static const struct mdev_parent_ops intel_vgpu_ops = {
1150 	.supported_type_groups	= intel_vgpu_type_groups,
1151 	.create			= intel_vgpu_create,
1152 	.remove			= intel_vgpu_remove,
1153 
1154 	.open			= intel_vgpu_open,
1155 	.release		= intel_vgpu_release,
1156 
1157 	.read			= intel_vgpu_read,
1158 	.write			= intel_vgpu_write,
1159 	.mmap			= intel_vgpu_mmap,
1160 	.ioctl			= intel_vgpu_ioctl,
1161 };
1162 
1163 static int kvmgt_host_init(struct device *dev, void *gvt, const void *ops)
1164 {
1165 	if (!intel_gvt_init_vgpu_type_groups(gvt))
1166 		return -EFAULT;
1167 
1168 	intel_gvt_ops = ops;
1169 
1170 	return mdev_register_device(dev, &intel_vgpu_ops);
1171 }
1172 
1173 static void kvmgt_host_exit(struct device *dev, void *gvt)
1174 {
1175 	intel_gvt_cleanup_vgpu_type_groups(gvt);
1176 	mdev_unregister_device(dev);
1177 }
1178 
1179 static int kvmgt_write_protect_add(unsigned long handle, u64 gfn)
1180 {
1181 	struct kvmgt_guest_info *info;
1182 	struct kvm *kvm;
1183 	struct kvm_memory_slot *slot;
1184 	int idx;
1185 
1186 	if (!handle_valid(handle))
1187 		return -ESRCH;
1188 
1189 	info = (struct kvmgt_guest_info *)handle;
1190 	kvm = info->kvm;
1191 
1192 	idx = srcu_read_lock(&kvm->srcu);
1193 	slot = gfn_to_memslot(kvm, gfn);
1194 	if (!slot) {
1195 		srcu_read_unlock(&kvm->srcu, idx);
1196 		return -EINVAL;
1197 	}
1198 
1199 	spin_lock(&kvm->mmu_lock);
1200 
1201 	if (kvmgt_gfn_is_write_protected(info, gfn))
1202 		goto out;
1203 
1204 	kvm_slot_page_track_add_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1205 	kvmgt_protect_table_add(info, gfn);
1206 
1207 out:
1208 	spin_unlock(&kvm->mmu_lock);
1209 	srcu_read_unlock(&kvm->srcu, idx);
1210 	return 0;
1211 }
1212 
1213 static int kvmgt_write_protect_remove(unsigned long handle, u64 gfn)
1214 {
1215 	struct kvmgt_guest_info *info;
1216 	struct kvm *kvm;
1217 	struct kvm_memory_slot *slot;
1218 	int idx;
1219 
1220 	if (!handle_valid(handle))
1221 		return 0;
1222 
1223 	info = (struct kvmgt_guest_info *)handle;
1224 	kvm = info->kvm;
1225 
1226 	idx = srcu_read_lock(&kvm->srcu);
1227 	slot = gfn_to_memslot(kvm, gfn);
1228 	if (!slot) {
1229 		srcu_read_unlock(&kvm->srcu, idx);
1230 		return -EINVAL;
1231 	}
1232 
1233 	spin_lock(&kvm->mmu_lock);
1234 
1235 	if (!kvmgt_gfn_is_write_protected(info, gfn))
1236 		goto out;
1237 
1238 	kvm_slot_page_track_remove_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1239 	kvmgt_protect_table_del(info, gfn);
1240 
1241 out:
1242 	spin_unlock(&kvm->mmu_lock);
1243 	srcu_read_unlock(&kvm->srcu, idx);
1244 	return 0;
1245 }
1246 
1247 static void kvmgt_page_track_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1248 		const u8 *val, int len,
1249 		struct kvm_page_track_notifier_node *node)
1250 {
1251 	struct kvmgt_guest_info *info = container_of(node,
1252 					struct kvmgt_guest_info, track_node);
1253 
1254 	if (kvmgt_gfn_is_write_protected(info, gpa_to_gfn(gpa)))
1255 		intel_gvt_ops->emulate_mmio_write(info->vgpu, gpa,
1256 					(void *)val, len);
1257 }
1258 
1259 static void kvmgt_page_track_flush_slot(struct kvm *kvm,
1260 		struct kvm_memory_slot *slot,
1261 		struct kvm_page_track_notifier_node *node)
1262 {
1263 	int i;
1264 	gfn_t gfn;
1265 	struct kvmgt_guest_info *info = container_of(node,
1266 					struct kvmgt_guest_info, track_node);
1267 
1268 	spin_lock(&kvm->mmu_lock);
1269 	for (i = 0; i < slot->npages; i++) {
1270 		gfn = slot->base_gfn + i;
1271 		if (kvmgt_gfn_is_write_protected(info, gfn)) {
1272 			kvm_slot_page_track_remove_page(kvm, slot, gfn,
1273 						KVM_PAGE_TRACK_WRITE);
1274 			kvmgt_protect_table_del(info, gfn);
1275 		}
1276 	}
1277 	spin_unlock(&kvm->mmu_lock);
1278 }
1279 
1280 static bool __kvmgt_vgpu_exist(struct intel_vgpu *vgpu, struct kvm *kvm)
1281 {
1282 	struct intel_vgpu *itr;
1283 	struct kvmgt_guest_info *info;
1284 	int id;
1285 	bool ret = false;
1286 
1287 	mutex_lock(&vgpu->gvt->lock);
1288 	for_each_active_vgpu(vgpu->gvt, itr, id) {
1289 		if (!handle_valid(itr->handle))
1290 			continue;
1291 
1292 		info = (struct kvmgt_guest_info *)itr->handle;
1293 		if (kvm && kvm == info->kvm) {
1294 			ret = true;
1295 			goto out;
1296 		}
1297 	}
1298 out:
1299 	mutex_unlock(&vgpu->gvt->lock);
1300 	return ret;
1301 }
1302 
1303 static int kvmgt_guest_init(struct mdev_device *mdev)
1304 {
1305 	struct kvmgt_guest_info *info;
1306 	struct intel_vgpu *vgpu;
1307 	struct kvm *kvm;
1308 
1309 	vgpu = mdev_get_drvdata(mdev);
1310 	if (handle_valid(vgpu->handle))
1311 		return -EEXIST;
1312 
1313 	kvm = vgpu->vdev.kvm;
1314 	if (!kvm || kvm->mm != current->mm) {
1315 		gvt_vgpu_err("KVM is required to use Intel vGPU\n");
1316 		return -ESRCH;
1317 	}
1318 
1319 	if (__kvmgt_vgpu_exist(vgpu, kvm))
1320 		return -EEXIST;
1321 
1322 	info = vzalloc(sizeof(struct kvmgt_guest_info));
1323 	if (!info)
1324 		return -ENOMEM;
1325 
1326 	vgpu->handle = (unsigned long)info;
1327 	info->vgpu = vgpu;
1328 	info->kvm = kvm;
1329 
1330 	kvmgt_protect_table_init(info);
1331 	gvt_cache_init(vgpu);
1332 
1333 	info->track_node.track_write = kvmgt_page_track_write;
1334 	info->track_node.track_flush_slot = kvmgt_page_track_flush_slot;
1335 	kvm_page_track_register_notifier(kvm, &info->track_node);
1336 
1337 	return 0;
1338 }
1339 
1340 static bool kvmgt_guest_exit(struct kvmgt_guest_info *info)
1341 {
1342 	struct intel_vgpu *vgpu = info->vgpu;
1343 
1344 	if (!info) {
1345 		gvt_vgpu_err("kvmgt_guest_info invalid\n");
1346 		return false;
1347 	}
1348 
1349 	kvm_page_track_unregister_notifier(info->kvm, &info->track_node);
1350 	kvmgt_protect_table_destroy(info);
1351 	gvt_cache_destroy(info->vgpu);
1352 	vfree(info);
1353 
1354 	return true;
1355 }
1356 
1357 static int kvmgt_attach_vgpu(void *vgpu, unsigned long *handle)
1358 {
1359 	/* nothing to do here */
1360 	return 0;
1361 }
1362 
1363 static void kvmgt_detach_vgpu(unsigned long handle)
1364 {
1365 	/* nothing to do here */
1366 }
1367 
1368 static int kvmgt_inject_msi(unsigned long handle, u32 addr, u16 data)
1369 {
1370 	struct kvmgt_guest_info *info;
1371 	struct intel_vgpu *vgpu;
1372 
1373 	if (!handle_valid(handle))
1374 		return -ESRCH;
1375 
1376 	info = (struct kvmgt_guest_info *)handle;
1377 	vgpu = info->vgpu;
1378 
1379 	if (eventfd_signal(vgpu->vdev.msi_trigger, 1) == 1)
1380 		return 0;
1381 
1382 	return -EFAULT;
1383 }
1384 
1385 static unsigned long kvmgt_gfn_to_pfn(unsigned long handle, unsigned long gfn)
1386 {
1387 	unsigned long iova, pfn;
1388 	struct kvmgt_guest_info *info;
1389 	struct device *dev;
1390 	struct intel_vgpu *vgpu;
1391 	int rc;
1392 
1393 	if (!handle_valid(handle))
1394 		return INTEL_GVT_INVALID_ADDR;
1395 
1396 	info = (struct kvmgt_guest_info *)handle;
1397 	vgpu = info->vgpu;
1398 	iova = gvt_cache_find(info->vgpu, gfn);
1399 	if (iova != INTEL_GVT_INVALID_ADDR)
1400 		return iova;
1401 
1402 	pfn = INTEL_GVT_INVALID_ADDR;
1403 	dev = mdev_dev(info->vgpu->vdev.mdev);
1404 	rc = vfio_pin_pages(dev, &gfn, 1, IOMMU_READ | IOMMU_WRITE, &pfn);
1405 	if (rc != 1) {
1406 		gvt_vgpu_err("vfio_pin_pages failed for gfn 0x%lx: %d\n",
1407 			gfn, rc);
1408 		return INTEL_GVT_INVALID_ADDR;
1409 	}
1410 	/* transfer to host iova for GFX to use DMA */
1411 	rc = gvt_dma_map_iova(info->vgpu, pfn, &iova);
1412 	if (rc) {
1413 		gvt_vgpu_err("gvt_dma_map_iova failed for gfn: 0x%lx\n", gfn);
1414 		vfio_unpin_pages(dev, &gfn, 1);
1415 		return INTEL_GVT_INVALID_ADDR;
1416 	}
1417 
1418 	gvt_cache_add(info->vgpu, gfn, iova);
1419 	return iova;
1420 }
1421 
1422 static int kvmgt_rw_gpa(unsigned long handle, unsigned long gpa,
1423 			void *buf, unsigned long len, bool write)
1424 {
1425 	struct kvmgt_guest_info *info;
1426 	struct kvm *kvm;
1427 	int idx, ret;
1428 	bool kthread = current->mm == NULL;
1429 
1430 	if (!handle_valid(handle))
1431 		return -ESRCH;
1432 
1433 	info = (struct kvmgt_guest_info *)handle;
1434 	kvm = info->kvm;
1435 
1436 	if (kthread)
1437 		use_mm(kvm->mm);
1438 
1439 	idx = srcu_read_lock(&kvm->srcu);
1440 	ret = write ? kvm_write_guest(kvm, gpa, buf, len) :
1441 		      kvm_read_guest(kvm, gpa, buf, len);
1442 	srcu_read_unlock(&kvm->srcu, idx);
1443 
1444 	if (kthread)
1445 		unuse_mm(kvm->mm);
1446 
1447 	return ret;
1448 }
1449 
1450 static int kvmgt_read_gpa(unsigned long handle, unsigned long gpa,
1451 			void *buf, unsigned long len)
1452 {
1453 	return kvmgt_rw_gpa(handle, gpa, buf, len, false);
1454 }
1455 
1456 static int kvmgt_write_gpa(unsigned long handle, unsigned long gpa,
1457 			void *buf, unsigned long len)
1458 {
1459 	return kvmgt_rw_gpa(handle, gpa, buf, len, true);
1460 }
1461 
1462 static unsigned long kvmgt_virt_to_pfn(void *addr)
1463 {
1464 	return PFN_DOWN(__pa(addr));
1465 }
1466 
1467 struct intel_gvt_mpt kvmgt_mpt = {
1468 	.host_init = kvmgt_host_init,
1469 	.host_exit = kvmgt_host_exit,
1470 	.attach_vgpu = kvmgt_attach_vgpu,
1471 	.detach_vgpu = kvmgt_detach_vgpu,
1472 	.inject_msi = kvmgt_inject_msi,
1473 	.from_virt_to_mfn = kvmgt_virt_to_pfn,
1474 	.set_wp_page = kvmgt_write_protect_add,
1475 	.unset_wp_page = kvmgt_write_protect_remove,
1476 	.read_gpa = kvmgt_read_gpa,
1477 	.write_gpa = kvmgt_write_gpa,
1478 	.gfn_to_mfn = kvmgt_gfn_to_pfn,
1479 };
1480 EXPORT_SYMBOL_GPL(kvmgt_mpt);
1481 
1482 static int __init kvmgt_init(void)
1483 {
1484 	return 0;
1485 }
1486 
1487 static void __exit kvmgt_exit(void)
1488 {
1489 }
1490 
1491 module_init(kvmgt_init);
1492 module_exit(kvmgt_exit);
1493 
1494 MODULE_LICENSE("GPL and additional rights");
1495 MODULE_AUTHOR("Intel Corporation");
1496