xref: /openbmc/linux/drivers/gpu/drm/i915/gvt/kvmgt.c (revision fb960bd2)
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 	for (;;) {
236 		mutex_lock(&vgpu->vdev.cache_lock);
237 		node = rb_first(&vgpu->vdev.cache);
238 		if (!node) {
239 			mutex_unlock(&vgpu->vdev.cache_lock);
240 			break;
241 		}
242 		dma = rb_entry(node, struct gvt_dma, node);
243 		gvt_dma_unmap_iova(vgpu, dma->iova);
244 		gfn = dma->gfn;
245 		__gvt_cache_remove_entry(vgpu, dma);
246 		mutex_unlock(&vgpu->vdev.cache_lock);
247 		vfio_unpin_pages(dev, &gfn, 1);
248 	}
249 }
250 
251 static void kvmgt_protect_table_init(struct kvmgt_guest_info *info)
252 {
253 	hash_init(info->ptable);
254 }
255 
256 static void kvmgt_protect_table_destroy(struct kvmgt_guest_info *info)
257 {
258 	struct kvmgt_pgfn *p;
259 	struct hlist_node *tmp;
260 	int i;
261 
262 	hash_for_each_safe(info->ptable, i, tmp, p, hnode) {
263 		hash_del(&p->hnode);
264 		kfree(p);
265 	}
266 }
267 
268 static struct kvmgt_pgfn *
269 __kvmgt_protect_table_find(struct kvmgt_guest_info *info, gfn_t gfn)
270 {
271 	struct kvmgt_pgfn *p, *res = NULL;
272 
273 	hash_for_each_possible(info->ptable, p, hnode, gfn) {
274 		if (gfn == p->gfn) {
275 			res = p;
276 			break;
277 		}
278 	}
279 
280 	return res;
281 }
282 
283 static bool kvmgt_gfn_is_write_protected(struct kvmgt_guest_info *info,
284 				gfn_t gfn)
285 {
286 	struct kvmgt_pgfn *p;
287 
288 	p = __kvmgt_protect_table_find(info, gfn);
289 	return !!p;
290 }
291 
292 static void kvmgt_protect_table_add(struct kvmgt_guest_info *info, gfn_t gfn)
293 {
294 	struct kvmgt_pgfn *p;
295 
296 	if (kvmgt_gfn_is_write_protected(info, gfn))
297 		return;
298 
299 	p = kzalloc(sizeof(struct kvmgt_pgfn), GFP_ATOMIC);
300 	if (WARN(!p, "gfn: 0x%llx\n", gfn))
301 		return;
302 
303 	p->gfn = gfn;
304 	hash_add(info->ptable, &p->hnode, gfn);
305 }
306 
307 static void kvmgt_protect_table_del(struct kvmgt_guest_info *info,
308 				gfn_t gfn)
309 {
310 	struct kvmgt_pgfn *p;
311 
312 	p = __kvmgt_protect_table_find(info, gfn);
313 	if (p) {
314 		hash_del(&p->hnode);
315 		kfree(p);
316 	}
317 }
318 
319 static int intel_vgpu_create(struct kobject *kobj, struct mdev_device *mdev)
320 {
321 	struct intel_vgpu *vgpu = NULL;
322 	struct intel_vgpu_type *type;
323 	struct device *pdev;
324 	void *gvt;
325 	int ret;
326 
327 	pdev = mdev_parent_dev(mdev);
328 	gvt = kdev_to_i915(pdev)->gvt;
329 
330 	type = intel_gvt_ops->gvt_find_vgpu_type(gvt, kobject_name(kobj));
331 	if (!type) {
332 		gvt_vgpu_err("failed to find type %s to create\n",
333 						kobject_name(kobj));
334 		ret = -EINVAL;
335 		goto out;
336 	}
337 
338 	vgpu = intel_gvt_ops->vgpu_create(gvt, type);
339 	if (IS_ERR_OR_NULL(vgpu)) {
340 		ret = vgpu == NULL ? -EFAULT : PTR_ERR(vgpu);
341 		gvt_vgpu_err("failed to create intel vgpu: %d\n", ret);
342 		goto out;
343 	}
344 
345 	INIT_WORK(&vgpu->vdev.release_work, intel_vgpu_release_work);
346 
347 	vgpu->vdev.mdev = mdev;
348 	mdev_set_drvdata(mdev, vgpu);
349 
350 	gvt_dbg_core("intel_vgpu_create succeeded for mdev: %s\n",
351 		     dev_name(mdev_dev(mdev)));
352 	ret = 0;
353 
354 out:
355 	return ret;
356 }
357 
358 static int intel_vgpu_remove(struct mdev_device *mdev)
359 {
360 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
361 
362 	if (handle_valid(vgpu->handle))
363 		return -EBUSY;
364 
365 	intel_gvt_ops->vgpu_destroy(vgpu);
366 	return 0;
367 }
368 
369 static int intel_vgpu_iommu_notifier(struct notifier_block *nb,
370 				     unsigned long action, void *data)
371 {
372 	struct intel_vgpu *vgpu = container_of(nb,
373 					struct intel_vgpu,
374 					vdev.iommu_notifier);
375 
376 	if (action == VFIO_IOMMU_NOTIFY_DMA_UNMAP) {
377 		struct vfio_iommu_type1_dma_unmap *unmap = data;
378 		unsigned long gfn, end_gfn;
379 
380 		gfn = unmap->iova >> PAGE_SHIFT;
381 		end_gfn = gfn + unmap->size / PAGE_SIZE;
382 
383 		while (gfn < end_gfn)
384 			gvt_cache_remove(vgpu, gfn++);
385 	}
386 
387 	return NOTIFY_OK;
388 }
389 
390 static int intel_vgpu_group_notifier(struct notifier_block *nb,
391 				     unsigned long action, void *data)
392 {
393 	struct intel_vgpu *vgpu = container_of(nb,
394 					struct intel_vgpu,
395 					vdev.group_notifier);
396 
397 	/* the only action we care about */
398 	if (action == VFIO_GROUP_NOTIFY_SET_KVM) {
399 		vgpu->vdev.kvm = data;
400 
401 		if (!data)
402 			schedule_work(&vgpu->vdev.release_work);
403 	}
404 
405 	return NOTIFY_OK;
406 }
407 
408 static int intel_vgpu_open(struct mdev_device *mdev)
409 {
410 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
411 	unsigned long events;
412 	int ret;
413 
414 	vgpu->vdev.iommu_notifier.notifier_call = intel_vgpu_iommu_notifier;
415 	vgpu->vdev.group_notifier.notifier_call = intel_vgpu_group_notifier;
416 
417 	events = VFIO_IOMMU_NOTIFY_DMA_UNMAP;
418 	ret = vfio_register_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY, &events,
419 				&vgpu->vdev.iommu_notifier);
420 	if (ret != 0) {
421 		gvt_vgpu_err("vfio_register_notifier for iommu failed: %d\n",
422 			ret);
423 		goto out;
424 	}
425 
426 	events = VFIO_GROUP_NOTIFY_SET_KVM;
427 	ret = vfio_register_notifier(mdev_dev(mdev), VFIO_GROUP_NOTIFY, &events,
428 				&vgpu->vdev.group_notifier);
429 	if (ret != 0) {
430 		gvt_vgpu_err("vfio_register_notifier for group failed: %d\n",
431 			ret);
432 		goto undo_iommu;
433 	}
434 
435 	ret = kvmgt_guest_init(mdev);
436 	if (ret)
437 		goto undo_group;
438 
439 	intel_gvt_ops->vgpu_activate(vgpu);
440 
441 	atomic_set(&vgpu->vdev.released, 0);
442 	return ret;
443 
444 undo_group:
445 	vfio_unregister_notifier(mdev_dev(mdev), VFIO_GROUP_NOTIFY,
446 					&vgpu->vdev.group_notifier);
447 
448 undo_iommu:
449 	vfio_unregister_notifier(mdev_dev(mdev), VFIO_IOMMU_NOTIFY,
450 					&vgpu->vdev.iommu_notifier);
451 out:
452 	return ret;
453 }
454 
455 static void __intel_vgpu_release(struct intel_vgpu *vgpu)
456 {
457 	struct kvmgt_guest_info *info;
458 	int ret;
459 
460 	if (!handle_valid(vgpu->handle))
461 		return;
462 
463 	if (atomic_cmpxchg(&vgpu->vdev.released, 0, 1))
464 		return;
465 
466 	intel_gvt_ops->vgpu_deactivate(vgpu);
467 
468 	ret = vfio_unregister_notifier(mdev_dev(vgpu->vdev.mdev), VFIO_IOMMU_NOTIFY,
469 					&vgpu->vdev.iommu_notifier);
470 	WARN(ret, "vfio_unregister_notifier for iommu failed: %d\n", ret);
471 
472 	ret = vfio_unregister_notifier(mdev_dev(vgpu->vdev.mdev), VFIO_GROUP_NOTIFY,
473 					&vgpu->vdev.group_notifier);
474 	WARN(ret, "vfio_unregister_notifier for group failed: %d\n", ret);
475 
476 	info = (struct kvmgt_guest_info *)vgpu->handle;
477 	kvmgt_guest_exit(info);
478 
479 	vgpu->vdev.kvm = NULL;
480 	vgpu->handle = 0;
481 }
482 
483 static void intel_vgpu_release(struct mdev_device *mdev)
484 {
485 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
486 
487 	__intel_vgpu_release(vgpu);
488 }
489 
490 static void intel_vgpu_release_work(struct work_struct *work)
491 {
492 	struct intel_vgpu *vgpu = container_of(work, struct intel_vgpu,
493 					vdev.release_work);
494 
495 	__intel_vgpu_release(vgpu);
496 }
497 
498 static uint64_t intel_vgpu_get_bar_addr(struct intel_vgpu *vgpu, int bar)
499 {
500 	u32 start_lo, start_hi;
501 	u32 mem_type;
502 
503 	start_lo = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + bar)) &
504 			PCI_BASE_ADDRESS_MEM_MASK;
505 	mem_type = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space + bar)) &
506 			PCI_BASE_ADDRESS_MEM_TYPE_MASK;
507 
508 	switch (mem_type) {
509 	case PCI_BASE_ADDRESS_MEM_TYPE_64:
510 		start_hi = (*(u32 *)(vgpu->cfg_space.virtual_cfg_space
511 						+ bar + 4));
512 		break;
513 	case PCI_BASE_ADDRESS_MEM_TYPE_32:
514 	case PCI_BASE_ADDRESS_MEM_TYPE_1M:
515 		/* 1M mem BAR treated as 32-bit BAR */
516 	default:
517 		/* mem unknown type treated as 32-bit BAR */
518 		start_hi = 0;
519 		break;
520 	}
521 
522 	return ((u64)start_hi << 32) | start_lo;
523 }
524 
525 static int intel_vgpu_bar_rw(struct intel_vgpu *vgpu, int bar, uint64_t off,
526 			     void *buf, unsigned int count, bool is_write)
527 {
528 	uint64_t bar_start = intel_vgpu_get_bar_addr(vgpu, bar);
529 	int ret;
530 
531 	if (is_write)
532 		ret = intel_gvt_ops->emulate_mmio_write(vgpu,
533 					bar_start + off, buf, count);
534 	else
535 		ret = intel_gvt_ops->emulate_mmio_read(vgpu,
536 					bar_start + off, buf, count);
537 	return ret;
538 }
539 
540 static ssize_t intel_vgpu_rw(struct mdev_device *mdev, char *buf,
541 			size_t count, loff_t *ppos, bool is_write)
542 {
543 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
544 	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
545 	uint64_t pos = *ppos & VFIO_PCI_OFFSET_MASK;
546 	int ret = -EINVAL;
547 
548 
549 	if (index >= VFIO_PCI_NUM_REGIONS) {
550 		gvt_vgpu_err("invalid index: %u\n", index);
551 		return -EINVAL;
552 	}
553 
554 	switch (index) {
555 	case VFIO_PCI_CONFIG_REGION_INDEX:
556 		if (is_write)
557 			ret = intel_gvt_ops->emulate_cfg_write(vgpu, pos,
558 						buf, count);
559 		else
560 			ret = intel_gvt_ops->emulate_cfg_read(vgpu, pos,
561 						buf, count);
562 		break;
563 	case VFIO_PCI_BAR0_REGION_INDEX:
564 		ret = intel_vgpu_bar_rw(vgpu, PCI_BASE_ADDRESS_0, pos,
565 					buf, count, is_write);
566 		break;
567 	case VFIO_PCI_BAR2_REGION_INDEX:
568 		ret = intel_vgpu_bar_rw(vgpu, PCI_BASE_ADDRESS_2, pos,
569 					buf, count, is_write);
570 		break;
571 	case VFIO_PCI_BAR1_REGION_INDEX:
572 	case VFIO_PCI_BAR3_REGION_INDEX:
573 	case VFIO_PCI_BAR4_REGION_INDEX:
574 	case VFIO_PCI_BAR5_REGION_INDEX:
575 	case VFIO_PCI_VGA_REGION_INDEX:
576 	case VFIO_PCI_ROM_REGION_INDEX:
577 	default:
578 		gvt_vgpu_err("unsupported region: %u\n", index);
579 	}
580 
581 	return ret == 0 ? count : ret;
582 }
583 
584 static ssize_t intel_vgpu_read(struct mdev_device *mdev, char __user *buf,
585 			size_t count, loff_t *ppos)
586 {
587 	unsigned int done = 0;
588 	int ret;
589 
590 	while (count) {
591 		size_t filled;
592 
593 		if (count >= 4 && !(*ppos % 4)) {
594 			u32 val;
595 
596 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
597 					ppos, false);
598 			if (ret <= 0)
599 				goto read_err;
600 
601 			if (copy_to_user(buf, &val, sizeof(val)))
602 				goto read_err;
603 
604 			filled = 4;
605 		} else if (count >= 2 && !(*ppos % 2)) {
606 			u16 val;
607 
608 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
609 					ppos, false);
610 			if (ret <= 0)
611 				goto read_err;
612 
613 			if (copy_to_user(buf, &val, sizeof(val)))
614 				goto read_err;
615 
616 			filled = 2;
617 		} else {
618 			u8 val;
619 
620 			ret = intel_vgpu_rw(mdev, &val, sizeof(val), ppos,
621 					false);
622 			if (ret <= 0)
623 				goto read_err;
624 
625 			if (copy_to_user(buf, &val, sizeof(val)))
626 				goto read_err;
627 
628 			filled = 1;
629 		}
630 
631 		count -= filled;
632 		done += filled;
633 		*ppos += filled;
634 		buf += filled;
635 	}
636 
637 	return done;
638 
639 read_err:
640 	return -EFAULT;
641 }
642 
643 static ssize_t intel_vgpu_write(struct mdev_device *mdev,
644 				const char __user *buf,
645 				size_t count, loff_t *ppos)
646 {
647 	unsigned int done = 0;
648 	int ret;
649 
650 	while (count) {
651 		size_t filled;
652 
653 		if (count >= 4 && !(*ppos % 4)) {
654 			u32 val;
655 
656 			if (copy_from_user(&val, buf, sizeof(val)))
657 				goto write_err;
658 
659 			ret = intel_vgpu_rw(mdev, (char *)&val, sizeof(val),
660 					ppos, true);
661 			if (ret <= 0)
662 				goto write_err;
663 
664 			filled = 4;
665 		} else if (count >= 2 && !(*ppos % 2)) {
666 			u16 val;
667 
668 			if (copy_from_user(&val, buf, sizeof(val)))
669 				goto write_err;
670 
671 			ret = intel_vgpu_rw(mdev, (char *)&val,
672 					sizeof(val), ppos, true);
673 			if (ret <= 0)
674 				goto write_err;
675 
676 			filled = 2;
677 		} else {
678 			u8 val;
679 
680 			if (copy_from_user(&val, buf, sizeof(val)))
681 				goto write_err;
682 
683 			ret = intel_vgpu_rw(mdev, &val, sizeof(val),
684 					ppos, true);
685 			if (ret <= 0)
686 				goto write_err;
687 
688 			filled = 1;
689 		}
690 
691 		count -= filled;
692 		done += filled;
693 		*ppos += filled;
694 		buf += filled;
695 	}
696 
697 	return done;
698 write_err:
699 	return -EFAULT;
700 }
701 
702 static int intel_vgpu_mmap(struct mdev_device *mdev, struct vm_area_struct *vma)
703 {
704 	unsigned int index;
705 	u64 virtaddr;
706 	unsigned long req_size, pgoff = 0;
707 	pgprot_t pg_prot;
708 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
709 
710 	index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
711 	if (index >= VFIO_PCI_ROM_REGION_INDEX)
712 		return -EINVAL;
713 
714 	if (vma->vm_end < vma->vm_start)
715 		return -EINVAL;
716 	if ((vma->vm_flags & VM_SHARED) == 0)
717 		return -EINVAL;
718 	if (index != VFIO_PCI_BAR2_REGION_INDEX)
719 		return -EINVAL;
720 
721 	pg_prot = vma->vm_page_prot;
722 	virtaddr = vma->vm_start;
723 	req_size = vma->vm_end - vma->vm_start;
724 	pgoff = vgpu_aperture_pa_base(vgpu) >> PAGE_SHIFT;
725 
726 	return remap_pfn_range(vma, virtaddr, pgoff, req_size, pg_prot);
727 }
728 
729 static int intel_vgpu_get_irq_count(struct intel_vgpu *vgpu, int type)
730 {
731 	if (type == VFIO_PCI_INTX_IRQ_INDEX || type == VFIO_PCI_MSI_IRQ_INDEX)
732 		return 1;
733 
734 	return 0;
735 }
736 
737 static int intel_vgpu_set_intx_mask(struct intel_vgpu *vgpu,
738 			unsigned int index, unsigned int start,
739 			unsigned int count, uint32_t flags,
740 			void *data)
741 {
742 	return 0;
743 }
744 
745 static int intel_vgpu_set_intx_unmask(struct intel_vgpu *vgpu,
746 			unsigned int index, unsigned int start,
747 			unsigned int count, uint32_t flags, void *data)
748 {
749 	return 0;
750 }
751 
752 static int intel_vgpu_set_intx_trigger(struct intel_vgpu *vgpu,
753 		unsigned int index, unsigned int start, unsigned int count,
754 		uint32_t flags, void *data)
755 {
756 	return 0;
757 }
758 
759 static int intel_vgpu_set_msi_trigger(struct intel_vgpu *vgpu,
760 		unsigned int index, unsigned int start, unsigned int count,
761 		uint32_t flags, void *data)
762 {
763 	struct eventfd_ctx *trigger;
764 
765 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
766 		int fd = *(int *)data;
767 
768 		trigger = eventfd_ctx_fdget(fd);
769 		if (IS_ERR(trigger)) {
770 			gvt_vgpu_err("eventfd_ctx_fdget failed\n");
771 			return PTR_ERR(trigger);
772 		}
773 		vgpu->vdev.msi_trigger = trigger;
774 	}
775 
776 	return 0;
777 }
778 
779 static int intel_vgpu_set_irqs(struct intel_vgpu *vgpu, uint32_t flags,
780 		unsigned int index, unsigned int start, unsigned int count,
781 		void *data)
782 {
783 	int (*func)(struct intel_vgpu *vgpu, unsigned int index,
784 			unsigned int start, unsigned int count, uint32_t flags,
785 			void *data) = NULL;
786 
787 	switch (index) {
788 	case VFIO_PCI_INTX_IRQ_INDEX:
789 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
790 		case VFIO_IRQ_SET_ACTION_MASK:
791 			func = intel_vgpu_set_intx_mask;
792 			break;
793 		case VFIO_IRQ_SET_ACTION_UNMASK:
794 			func = intel_vgpu_set_intx_unmask;
795 			break;
796 		case VFIO_IRQ_SET_ACTION_TRIGGER:
797 			func = intel_vgpu_set_intx_trigger;
798 			break;
799 		}
800 		break;
801 	case VFIO_PCI_MSI_IRQ_INDEX:
802 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
803 		case VFIO_IRQ_SET_ACTION_MASK:
804 		case VFIO_IRQ_SET_ACTION_UNMASK:
805 			/* XXX Need masking support exported */
806 			break;
807 		case VFIO_IRQ_SET_ACTION_TRIGGER:
808 			func = intel_vgpu_set_msi_trigger;
809 			break;
810 		}
811 		break;
812 	}
813 
814 	if (!func)
815 		return -ENOTTY;
816 
817 	return func(vgpu, index, start, count, flags, data);
818 }
819 
820 static long intel_vgpu_ioctl(struct mdev_device *mdev, unsigned int cmd,
821 			     unsigned long arg)
822 {
823 	struct intel_vgpu *vgpu = mdev_get_drvdata(mdev);
824 	unsigned long minsz;
825 
826 	gvt_dbg_core("vgpu%d ioctl, cmd: %d\n", vgpu->id, cmd);
827 
828 	if (cmd == VFIO_DEVICE_GET_INFO) {
829 		struct vfio_device_info info;
830 
831 		minsz = offsetofend(struct vfio_device_info, num_irqs);
832 
833 		if (copy_from_user(&info, (void __user *)arg, minsz))
834 			return -EFAULT;
835 
836 		if (info.argsz < minsz)
837 			return -EINVAL;
838 
839 		info.flags = VFIO_DEVICE_FLAGS_PCI;
840 		info.flags |= VFIO_DEVICE_FLAGS_RESET;
841 		info.num_regions = VFIO_PCI_NUM_REGIONS;
842 		info.num_irqs = VFIO_PCI_NUM_IRQS;
843 
844 		return copy_to_user((void __user *)arg, &info, minsz) ?
845 			-EFAULT : 0;
846 
847 	} else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
848 		struct vfio_region_info info;
849 		struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
850 		int i, ret;
851 		struct vfio_region_info_cap_sparse_mmap *sparse = NULL;
852 		size_t size;
853 		int nr_areas = 1;
854 		int cap_type_id;
855 
856 		minsz = offsetofend(struct vfio_region_info, offset);
857 
858 		if (copy_from_user(&info, (void __user *)arg, minsz))
859 			return -EFAULT;
860 
861 		if (info.argsz < minsz)
862 			return -EINVAL;
863 
864 		switch (info.index) {
865 		case VFIO_PCI_CONFIG_REGION_INDEX:
866 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
867 			info.size = vgpu->gvt->device_info.cfg_space_size;
868 			info.flags = VFIO_REGION_INFO_FLAG_READ |
869 				     VFIO_REGION_INFO_FLAG_WRITE;
870 			break;
871 		case VFIO_PCI_BAR0_REGION_INDEX:
872 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
873 			info.size = vgpu->cfg_space.bar[info.index].size;
874 			if (!info.size) {
875 				info.flags = 0;
876 				break;
877 			}
878 
879 			info.flags = VFIO_REGION_INFO_FLAG_READ |
880 				     VFIO_REGION_INFO_FLAG_WRITE;
881 			break;
882 		case VFIO_PCI_BAR1_REGION_INDEX:
883 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
884 			info.size = 0;
885 			info.flags = 0;
886 			break;
887 		case VFIO_PCI_BAR2_REGION_INDEX:
888 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
889 			info.flags = VFIO_REGION_INFO_FLAG_CAPS |
890 					VFIO_REGION_INFO_FLAG_MMAP |
891 					VFIO_REGION_INFO_FLAG_READ |
892 					VFIO_REGION_INFO_FLAG_WRITE;
893 			info.size = gvt_aperture_sz(vgpu->gvt);
894 
895 			size = sizeof(*sparse) +
896 					(nr_areas * sizeof(*sparse->areas));
897 			sparse = kzalloc(size, GFP_KERNEL);
898 			if (!sparse)
899 				return -ENOMEM;
900 
901 			sparse->nr_areas = nr_areas;
902 			cap_type_id = VFIO_REGION_INFO_CAP_SPARSE_MMAP;
903 			sparse->areas[0].offset =
904 					PAGE_ALIGN(vgpu_aperture_offset(vgpu));
905 			sparse->areas[0].size = vgpu_aperture_sz(vgpu);
906 			break;
907 
908 		case VFIO_PCI_BAR3_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
909 			info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
910 			info.size = 0;
911 
912 			info.flags = 0;
913 			gvt_dbg_core("get region info bar:%d\n", info.index);
914 			break;
915 
916 		case VFIO_PCI_ROM_REGION_INDEX:
917 		case VFIO_PCI_VGA_REGION_INDEX:
918 			gvt_dbg_core("get region info index:%d\n", info.index);
919 			break;
920 		default:
921 			{
922 				struct vfio_region_info_cap_type cap_type;
923 
924 				if (info.index >= VFIO_PCI_NUM_REGIONS +
925 						vgpu->vdev.num_regions)
926 					return -EINVAL;
927 
928 				i = info.index - VFIO_PCI_NUM_REGIONS;
929 
930 				info.offset =
931 					VFIO_PCI_INDEX_TO_OFFSET(info.index);
932 				info.size = vgpu->vdev.region[i].size;
933 				info.flags = vgpu->vdev.region[i].flags;
934 
935 				cap_type.type = vgpu->vdev.region[i].type;
936 				cap_type.subtype = vgpu->vdev.region[i].subtype;
937 
938 				ret = vfio_info_add_capability(&caps,
939 						VFIO_REGION_INFO_CAP_TYPE,
940 						&cap_type);
941 				if (ret)
942 					return ret;
943 			}
944 		}
945 
946 		if ((info.flags & VFIO_REGION_INFO_FLAG_CAPS) && sparse) {
947 			switch (cap_type_id) {
948 			case VFIO_REGION_INFO_CAP_SPARSE_MMAP:
949 				ret = vfio_info_add_capability(&caps,
950 					VFIO_REGION_INFO_CAP_SPARSE_MMAP,
951 					sparse);
952 				kfree(sparse);
953 				if (ret)
954 					return ret;
955 				break;
956 			default:
957 				return -EINVAL;
958 			}
959 		}
960 
961 		if (caps.size) {
962 			if (info.argsz < sizeof(info) + caps.size) {
963 				info.argsz = sizeof(info) + caps.size;
964 				info.cap_offset = 0;
965 			} else {
966 				vfio_info_cap_shift(&caps, sizeof(info));
967 				if (copy_to_user((void __user *)arg +
968 						  sizeof(info), caps.buf,
969 						  caps.size)) {
970 					kfree(caps.buf);
971 					return -EFAULT;
972 				}
973 				info.cap_offset = sizeof(info);
974 			}
975 
976 			kfree(caps.buf);
977 		}
978 
979 		return copy_to_user((void __user *)arg, &info, minsz) ?
980 			-EFAULT : 0;
981 	} else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
982 		struct vfio_irq_info info;
983 
984 		minsz = offsetofend(struct vfio_irq_info, count);
985 
986 		if (copy_from_user(&info, (void __user *)arg, minsz))
987 			return -EFAULT;
988 
989 		if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
990 			return -EINVAL;
991 
992 		switch (info.index) {
993 		case VFIO_PCI_INTX_IRQ_INDEX:
994 		case VFIO_PCI_MSI_IRQ_INDEX:
995 			break;
996 		default:
997 			return -EINVAL;
998 		}
999 
1000 		info.flags = VFIO_IRQ_INFO_EVENTFD;
1001 
1002 		info.count = intel_vgpu_get_irq_count(vgpu, info.index);
1003 
1004 		if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
1005 			info.flags |= (VFIO_IRQ_INFO_MASKABLE |
1006 				       VFIO_IRQ_INFO_AUTOMASKED);
1007 		else
1008 			info.flags |= VFIO_IRQ_INFO_NORESIZE;
1009 
1010 		return copy_to_user((void __user *)arg, &info, minsz) ?
1011 			-EFAULT : 0;
1012 	} else if (cmd == VFIO_DEVICE_SET_IRQS) {
1013 		struct vfio_irq_set hdr;
1014 		u8 *data = NULL;
1015 		int ret = 0;
1016 		size_t data_size = 0;
1017 
1018 		minsz = offsetofend(struct vfio_irq_set, count);
1019 
1020 		if (copy_from_user(&hdr, (void __user *)arg, minsz))
1021 			return -EFAULT;
1022 
1023 		if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) {
1024 			int max = intel_vgpu_get_irq_count(vgpu, hdr.index);
1025 
1026 			ret = vfio_set_irqs_validate_and_prepare(&hdr, max,
1027 						VFIO_PCI_NUM_IRQS, &data_size);
1028 			if (ret) {
1029 				gvt_vgpu_err("intel:vfio_set_irqs_validate_and_prepare failed\n");
1030 				return -EINVAL;
1031 			}
1032 			if (data_size) {
1033 				data = memdup_user((void __user *)(arg + minsz),
1034 						   data_size);
1035 				if (IS_ERR(data))
1036 					return PTR_ERR(data);
1037 			}
1038 		}
1039 
1040 		ret = intel_vgpu_set_irqs(vgpu, hdr.flags, hdr.index,
1041 					hdr.start, hdr.count, data);
1042 		kfree(data);
1043 
1044 		return ret;
1045 	} else if (cmd == VFIO_DEVICE_RESET) {
1046 		intel_gvt_ops->vgpu_reset(vgpu);
1047 		return 0;
1048 	}
1049 
1050 	return 0;
1051 }
1052 
1053 static ssize_t
1054 vgpu_id_show(struct device *dev, struct device_attribute *attr,
1055 	     char *buf)
1056 {
1057 	struct mdev_device *mdev = mdev_from_dev(dev);
1058 
1059 	if (mdev) {
1060 		struct intel_vgpu *vgpu = (struct intel_vgpu *)
1061 			mdev_get_drvdata(mdev);
1062 		return sprintf(buf, "%d\n", vgpu->id);
1063 	}
1064 	return sprintf(buf, "\n");
1065 }
1066 
1067 static ssize_t
1068 hw_id_show(struct device *dev, struct device_attribute *attr,
1069 	   char *buf)
1070 {
1071 	struct mdev_device *mdev = mdev_from_dev(dev);
1072 
1073 	if (mdev) {
1074 		struct intel_vgpu *vgpu = (struct intel_vgpu *)
1075 			mdev_get_drvdata(mdev);
1076 		return sprintf(buf, "%u\n",
1077 			       vgpu->submission.shadow_ctx->hw_id);
1078 	}
1079 	return sprintf(buf, "\n");
1080 }
1081 
1082 static DEVICE_ATTR_RO(vgpu_id);
1083 static DEVICE_ATTR_RO(hw_id);
1084 
1085 static struct attribute *intel_vgpu_attrs[] = {
1086 	&dev_attr_vgpu_id.attr,
1087 	&dev_attr_hw_id.attr,
1088 	NULL
1089 };
1090 
1091 static const struct attribute_group intel_vgpu_group = {
1092 	.name = "intel_vgpu",
1093 	.attrs = intel_vgpu_attrs,
1094 };
1095 
1096 static const struct attribute_group *intel_vgpu_groups[] = {
1097 	&intel_vgpu_group,
1098 	NULL,
1099 };
1100 
1101 static struct mdev_parent_ops intel_vgpu_ops = {
1102 	.mdev_attr_groups       = intel_vgpu_groups,
1103 	.create			= intel_vgpu_create,
1104 	.remove			= intel_vgpu_remove,
1105 
1106 	.open			= intel_vgpu_open,
1107 	.release		= intel_vgpu_release,
1108 
1109 	.read			= intel_vgpu_read,
1110 	.write			= intel_vgpu_write,
1111 	.mmap			= intel_vgpu_mmap,
1112 	.ioctl			= intel_vgpu_ioctl,
1113 };
1114 
1115 static int kvmgt_host_init(struct device *dev, void *gvt, const void *ops)
1116 {
1117 	struct attribute **kvm_type_attrs;
1118 	struct attribute_group **kvm_vgpu_type_groups;
1119 
1120 	intel_gvt_ops = ops;
1121 	if (!intel_gvt_ops->get_gvt_attrs(&kvm_type_attrs,
1122 			&kvm_vgpu_type_groups))
1123 		return -EFAULT;
1124 	intel_vgpu_ops.supported_type_groups = kvm_vgpu_type_groups;
1125 
1126 	return mdev_register_device(dev, &intel_vgpu_ops);
1127 }
1128 
1129 static void kvmgt_host_exit(struct device *dev, void *gvt)
1130 {
1131 	mdev_unregister_device(dev);
1132 }
1133 
1134 static int kvmgt_write_protect_add(unsigned long handle, u64 gfn)
1135 {
1136 	struct kvmgt_guest_info *info;
1137 	struct kvm *kvm;
1138 	struct kvm_memory_slot *slot;
1139 	int idx;
1140 
1141 	if (!handle_valid(handle))
1142 		return -ESRCH;
1143 
1144 	info = (struct kvmgt_guest_info *)handle;
1145 	kvm = info->kvm;
1146 
1147 	idx = srcu_read_lock(&kvm->srcu);
1148 	slot = gfn_to_memslot(kvm, gfn);
1149 	if (!slot) {
1150 		srcu_read_unlock(&kvm->srcu, idx);
1151 		return -EINVAL;
1152 	}
1153 
1154 	spin_lock(&kvm->mmu_lock);
1155 
1156 	if (kvmgt_gfn_is_write_protected(info, gfn))
1157 		goto out;
1158 
1159 	kvm_slot_page_track_add_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1160 	kvmgt_protect_table_add(info, gfn);
1161 
1162 out:
1163 	spin_unlock(&kvm->mmu_lock);
1164 	srcu_read_unlock(&kvm->srcu, idx);
1165 	return 0;
1166 }
1167 
1168 static int kvmgt_write_protect_remove(unsigned long handle, u64 gfn)
1169 {
1170 	struct kvmgt_guest_info *info;
1171 	struct kvm *kvm;
1172 	struct kvm_memory_slot *slot;
1173 	int idx;
1174 
1175 	if (!handle_valid(handle))
1176 		return 0;
1177 
1178 	info = (struct kvmgt_guest_info *)handle;
1179 	kvm = info->kvm;
1180 
1181 	idx = srcu_read_lock(&kvm->srcu);
1182 	slot = gfn_to_memslot(kvm, gfn);
1183 	if (!slot) {
1184 		srcu_read_unlock(&kvm->srcu, idx);
1185 		return -EINVAL;
1186 	}
1187 
1188 	spin_lock(&kvm->mmu_lock);
1189 
1190 	if (!kvmgt_gfn_is_write_protected(info, gfn))
1191 		goto out;
1192 
1193 	kvm_slot_page_track_remove_page(kvm, slot, gfn, KVM_PAGE_TRACK_WRITE);
1194 	kvmgt_protect_table_del(info, gfn);
1195 
1196 out:
1197 	spin_unlock(&kvm->mmu_lock);
1198 	srcu_read_unlock(&kvm->srcu, idx);
1199 	return 0;
1200 }
1201 
1202 static void kvmgt_page_track_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1203 		const u8 *val, int len,
1204 		struct kvm_page_track_notifier_node *node)
1205 {
1206 	struct kvmgt_guest_info *info = container_of(node,
1207 					struct kvmgt_guest_info, track_node);
1208 
1209 	if (kvmgt_gfn_is_write_protected(info, gpa_to_gfn(gpa)))
1210 		intel_gvt_ops->emulate_mmio_write(info->vgpu, gpa,
1211 					(void *)val, len);
1212 }
1213 
1214 static void kvmgt_page_track_flush_slot(struct kvm *kvm,
1215 		struct kvm_memory_slot *slot,
1216 		struct kvm_page_track_notifier_node *node)
1217 {
1218 	int i;
1219 	gfn_t gfn;
1220 	struct kvmgt_guest_info *info = container_of(node,
1221 					struct kvmgt_guest_info, track_node);
1222 
1223 	spin_lock(&kvm->mmu_lock);
1224 	for (i = 0; i < slot->npages; i++) {
1225 		gfn = slot->base_gfn + i;
1226 		if (kvmgt_gfn_is_write_protected(info, gfn)) {
1227 			kvm_slot_page_track_remove_page(kvm, slot, gfn,
1228 						KVM_PAGE_TRACK_WRITE);
1229 			kvmgt_protect_table_del(info, gfn);
1230 		}
1231 	}
1232 	spin_unlock(&kvm->mmu_lock);
1233 }
1234 
1235 static bool __kvmgt_vgpu_exist(struct intel_vgpu *vgpu, struct kvm *kvm)
1236 {
1237 	struct intel_vgpu *itr;
1238 	struct kvmgt_guest_info *info;
1239 	int id;
1240 	bool ret = false;
1241 
1242 	mutex_lock(&vgpu->gvt->lock);
1243 	for_each_active_vgpu(vgpu->gvt, itr, id) {
1244 		if (!handle_valid(itr->handle))
1245 			continue;
1246 
1247 		info = (struct kvmgt_guest_info *)itr->handle;
1248 		if (kvm && kvm == info->kvm) {
1249 			ret = true;
1250 			goto out;
1251 		}
1252 	}
1253 out:
1254 	mutex_unlock(&vgpu->gvt->lock);
1255 	return ret;
1256 }
1257 
1258 static int kvmgt_guest_init(struct mdev_device *mdev)
1259 {
1260 	struct kvmgt_guest_info *info;
1261 	struct intel_vgpu *vgpu;
1262 	struct kvm *kvm;
1263 
1264 	vgpu = mdev_get_drvdata(mdev);
1265 	if (handle_valid(vgpu->handle))
1266 		return -EEXIST;
1267 
1268 	kvm = vgpu->vdev.kvm;
1269 	if (!kvm || kvm->mm != current->mm) {
1270 		gvt_vgpu_err("KVM is required to use Intel vGPU\n");
1271 		return -ESRCH;
1272 	}
1273 
1274 	if (__kvmgt_vgpu_exist(vgpu, kvm))
1275 		return -EEXIST;
1276 
1277 	info = vzalloc(sizeof(struct kvmgt_guest_info));
1278 	if (!info)
1279 		return -ENOMEM;
1280 
1281 	vgpu->handle = (unsigned long)info;
1282 	info->vgpu = vgpu;
1283 	info->kvm = kvm;
1284 	kvm_get_kvm(info->kvm);
1285 
1286 	kvmgt_protect_table_init(info);
1287 	gvt_cache_init(vgpu);
1288 
1289 	info->track_node.track_write = kvmgt_page_track_write;
1290 	info->track_node.track_flush_slot = kvmgt_page_track_flush_slot;
1291 	kvm_page_track_register_notifier(kvm, &info->track_node);
1292 
1293 	return 0;
1294 }
1295 
1296 static bool kvmgt_guest_exit(struct kvmgt_guest_info *info)
1297 {
1298 	kvm_page_track_unregister_notifier(info->kvm, &info->track_node);
1299 	kvm_put_kvm(info->kvm);
1300 	kvmgt_protect_table_destroy(info);
1301 	gvt_cache_destroy(info->vgpu);
1302 	vfree(info);
1303 
1304 	return true;
1305 }
1306 
1307 static int kvmgt_attach_vgpu(void *vgpu, unsigned long *handle)
1308 {
1309 	/* nothing to do here */
1310 	return 0;
1311 }
1312 
1313 static void kvmgt_detach_vgpu(unsigned long handle)
1314 {
1315 	/* nothing to do here */
1316 }
1317 
1318 static int kvmgt_inject_msi(unsigned long handle, u32 addr, u16 data)
1319 {
1320 	struct kvmgt_guest_info *info;
1321 	struct intel_vgpu *vgpu;
1322 
1323 	if (!handle_valid(handle))
1324 		return -ESRCH;
1325 
1326 	info = (struct kvmgt_guest_info *)handle;
1327 	vgpu = info->vgpu;
1328 
1329 	if (eventfd_signal(vgpu->vdev.msi_trigger, 1) == 1)
1330 		return 0;
1331 
1332 	return -EFAULT;
1333 }
1334 
1335 static unsigned long kvmgt_gfn_to_pfn(unsigned long handle, unsigned long gfn)
1336 {
1337 	unsigned long iova, pfn;
1338 	struct kvmgt_guest_info *info;
1339 	struct device *dev;
1340 	struct intel_vgpu *vgpu;
1341 	int rc;
1342 
1343 	if (!handle_valid(handle))
1344 		return INTEL_GVT_INVALID_ADDR;
1345 
1346 	info = (struct kvmgt_guest_info *)handle;
1347 	vgpu = info->vgpu;
1348 	iova = gvt_cache_find(info->vgpu, gfn);
1349 	if (iova != INTEL_GVT_INVALID_ADDR)
1350 		return iova;
1351 
1352 	pfn = INTEL_GVT_INVALID_ADDR;
1353 	dev = mdev_dev(info->vgpu->vdev.mdev);
1354 	rc = vfio_pin_pages(dev, &gfn, 1, IOMMU_READ | IOMMU_WRITE, &pfn);
1355 	if (rc != 1) {
1356 		gvt_vgpu_err("vfio_pin_pages failed for gfn 0x%lx: %d\n",
1357 			gfn, rc);
1358 		return INTEL_GVT_INVALID_ADDR;
1359 	}
1360 	/* transfer to host iova for GFX to use DMA */
1361 	rc = gvt_dma_map_iova(info->vgpu, pfn, &iova);
1362 	if (rc) {
1363 		gvt_vgpu_err("gvt_dma_map_iova failed for gfn: 0x%lx\n", gfn);
1364 		vfio_unpin_pages(dev, &gfn, 1);
1365 		return INTEL_GVT_INVALID_ADDR;
1366 	}
1367 
1368 	gvt_cache_add(info->vgpu, gfn, iova);
1369 	return iova;
1370 }
1371 
1372 static int kvmgt_rw_gpa(unsigned long handle, unsigned long gpa,
1373 			void *buf, unsigned long len, bool write)
1374 {
1375 	struct kvmgt_guest_info *info;
1376 	struct kvm *kvm;
1377 	int idx, ret;
1378 	bool kthread = current->mm == NULL;
1379 
1380 	if (!handle_valid(handle))
1381 		return -ESRCH;
1382 
1383 	info = (struct kvmgt_guest_info *)handle;
1384 	kvm = info->kvm;
1385 
1386 	if (kthread)
1387 		use_mm(kvm->mm);
1388 
1389 	idx = srcu_read_lock(&kvm->srcu);
1390 	ret = write ? kvm_write_guest(kvm, gpa, buf, len) :
1391 		      kvm_read_guest(kvm, gpa, buf, len);
1392 	srcu_read_unlock(&kvm->srcu, idx);
1393 
1394 	if (kthread)
1395 		unuse_mm(kvm->mm);
1396 
1397 	return ret;
1398 }
1399 
1400 static int kvmgt_read_gpa(unsigned long handle, unsigned long gpa,
1401 			void *buf, unsigned long len)
1402 {
1403 	return kvmgt_rw_gpa(handle, gpa, buf, len, false);
1404 }
1405 
1406 static int kvmgt_write_gpa(unsigned long handle, unsigned long gpa,
1407 			void *buf, unsigned long len)
1408 {
1409 	return kvmgt_rw_gpa(handle, gpa, buf, len, true);
1410 }
1411 
1412 static unsigned long kvmgt_virt_to_pfn(void *addr)
1413 {
1414 	return PFN_DOWN(__pa(addr));
1415 }
1416 
1417 struct intel_gvt_mpt kvmgt_mpt = {
1418 	.host_init = kvmgt_host_init,
1419 	.host_exit = kvmgt_host_exit,
1420 	.attach_vgpu = kvmgt_attach_vgpu,
1421 	.detach_vgpu = kvmgt_detach_vgpu,
1422 	.inject_msi = kvmgt_inject_msi,
1423 	.from_virt_to_mfn = kvmgt_virt_to_pfn,
1424 	.set_wp_page = kvmgt_write_protect_add,
1425 	.unset_wp_page = kvmgt_write_protect_remove,
1426 	.read_gpa = kvmgt_read_gpa,
1427 	.write_gpa = kvmgt_write_gpa,
1428 	.gfn_to_mfn = kvmgt_gfn_to_pfn,
1429 };
1430 EXPORT_SYMBOL_GPL(kvmgt_mpt);
1431 
1432 static int __init kvmgt_init(void)
1433 {
1434 	return 0;
1435 }
1436 
1437 static void __exit kvmgt_exit(void)
1438 {
1439 }
1440 
1441 module_init(kvmgt_init);
1442 module_exit(kvmgt_exit);
1443 
1444 MODULE_LICENSE("GPL and additional rights");
1445 MODULE_AUTHOR("Intel Corporation");
1446