xref: /linux/drivers/vfio/vfio_main.c (revision e80a48bade619ec5a92230b3d4ae84bfc2746822)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VFIO core
4  *
5  * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
6  *     Author: Alex Williamson <alex.williamson@redhat.com>
7  *
8  * Derived from original vfio:
9  * Copyright 2010 Cisco Systems, Inc.  All rights reserved.
10  * Author: Tom Lyon, pugs@cisco.com
11  */
12 
13 #include <linux/cdev.h>
14 #include <linux/compat.h>
15 #include <linux/device.h>
16 #include <linux/fs.h>
17 #include <linux/idr.h>
18 #include <linux/iommu.h>
19 #include <linux/list.h>
20 #include <linux/miscdevice.h>
21 #include <linux/module.h>
22 #include <linux/mutex.h>
23 #include <linux/pci.h>
24 #include <linux/rwsem.h>
25 #include <linux/sched.h>
26 #include <linux/slab.h>
27 #include <linux/stat.h>
28 #include <linux/string.h>
29 #include <linux/uaccess.h>
30 #include <linux/vfio.h>
31 #include <linux/wait.h>
32 #include <linux/sched/signal.h>
33 #include <linux/pm_runtime.h>
34 #include <linux/interval_tree.h>
35 #include <linux/iova_bitmap.h>
36 #include <linux/iommufd.h>
37 #include "vfio.h"
38 
39 #define DRIVER_VERSION	"0.3"
40 #define DRIVER_AUTHOR	"Alex Williamson <alex.williamson@redhat.com>"
41 #define DRIVER_DESC	"VFIO - User Level meta-driver"
42 
43 static struct vfio {
44 	struct class			*device_class;
45 	struct ida			device_ida;
46 } vfio;
47 
48 static DEFINE_XARRAY(vfio_device_set_xa);
49 
50 int vfio_assign_device_set(struct vfio_device *device, void *set_id)
51 {
52 	unsigned long idx = (unsigned long)set_id;
53 	struct vfio_device_set *new_dev_set;
54 	struct vfio_device_set *dev_set;
55 
56 	if (WARN_ON(!set_id))
57 		return -EINVAL;
58 
59 	/*
60 	 * Atomically acquire a singleton object in the xarray for this set_id
61 	 */
62 	xa_lock(&vfio_device_set_xa);
63 	dev_set = xa_load(&vfio_device_set_xa, idx);
64 	if (dev_set)
65 		goto found_get_ref;
66 	xa_unlock(&vfio_device_set_xa);
67 
68 	new_dev_set = kzalloc(sizeof(*new_dev_set), GFP_KERNEL);
69 	if (!new_dev_set)
70 		return -ENOMEM;
71 	mutex_init(&new_dev_set->lock);
72 	INIT_LIST_HEAD(&new_dev_set->device_list);
73 	new_dev_set->set_id = set_id;
74 
75 	xa_lock(&vfio_device_set_xa);
76 	dev_set = __xa_cmpxchg(&vfio_device_set_xa, idx, NULL, new_dev_set,
77 			       GFP_KERNEL);
78 	if (!dev_set) {
79 		dev_set = new_dev_set;
80 		goto found_get_ref;
81 	}
82 
83 	kfree(new_dev_set);
84 	if (xa_is_err(dev_set)) {
85 		xa_unlock(&vfio_device_set_xa);
86 		return xa_err(dev_set);
87 	}
88 
89 found_get_ref:
90 	dev_set->device_count++;
91 	xa_unlock(&vfio_device_set_xa);
92 	mutex_lock(&dev_set->lock);
93 	device->dev_set = dev_set;
94 	list_add_tail(&device->dev_set_list, &dev_set->device_list);
95 	mutex_unlock(&dev_set->lock);
96 	return 0;
97 }
98 EXPORT_SYMBOL_GPL(vfio_assign_device_set);
99 
100 static void vfio_release_device_set(struct vfio_device *device)
101 {
102 	struct vfio_device_set *dev_set = device->dev_set;
103 
104 	if (!dev_set)
105 		return;
106 
107 	mutex_lock(&dev_set->lock);
108 	list_del(&device->dev_set_list);
109 	mutex_unlock(&dev_set->lock);
110 
111 	xa_lock(&vfio_device_set_xa);
112 	if (!--dev_set->device_count) {
113 		__xa_erase(&vfio_device_set_xa,
114 			   (unsigned long)dev_set->set_id);
115 		mutex_destroy(&dev_set->lock);
116 		kfree(dev_set);
117 	}
118 	xa_unlock(&vfio_device_set_xa);
119 }
120 
121 unsigned int vfio_device_set_open_count(struct vfio_device_set *dev_set)
122 {
123 	struct vfio_device *cur;
124 	unsigned int open_count = 0;
125 
126 	lockdep_assert_held(&dev_set->lock);
127 
128 	list_for_each_entry(cur, &dev_set->device_list, dev_set_list)
129 		open_count += cur->open_count;
130 	return open_count;
131 }
132 EXPORT_SYMBOL_GPL(vfio_device_set_open_count);
133 
134 /*
135  * Device objects - create, release, get, put, search
136  */
137 /* Device reference always implies a group reference */
138 void vfio_device_put_registration(struct vfio_device *device)
139 {
140 	if (refcount_dec_and_test(&device->refcount))
141 		complete(&device->comp);
142 }
143 
144 bool vfio_device_try_get_registration(struct vfio_device *device)
145 {
146 	return refcount_inc_not_zero(&device->refcount);
147 }
148 
149 /*
150  * VFIO driver API
151  */
152 /* Release helper called by vfio_put_device() */
153 static void vfio_device_release(struct device *dev)
154 {
155 	struct vfio_device *device =
156 			container_of(dev, struct vfio_device, device);
157 
158 	vfio_release_device_set(device);
159 	ida_free(&vfio.device_ida, device->index);
160 
161 	if (device->ops->release)
162 		device->ops->release(device);
163 
164 	kvfree(device);
165 }
166 
167 static int vfio_init_device(struct vfio_device *device, struct device *dev,
168 			    const struct vfio_device_ops *ops);
169 
170 /*
171  * Allocate and initialize vfio_device so it can be registered to vfio
172  * core.
173  *
174  * Drivers should use the wrapper vfio_alloc_device() for allocation.
175  * @size is the size of the structure to be allocated, including any
176  * private data used by the driver.
177  *
178  * Driver may provide an @init callback to cover device private data.
179  *
180  * Use vfio_put_device() to release the structure after success return.
181  */
182 struct vfio_device *_vfio_alloc_device(size_t size, struct device *dev,
183 				       const struct vfio_device_ops *ops)
184 {
185 	struct vfio_device *device;
186 	int ret;
187 
188 	if (WARN_ON(size < sizeof(struct vfio_device)))
189 		return ERR_PTR(-EINVAL);
190 
191 	device = kvzalloc(size, GFP_KERNEL);
192 	if (!device)
193 		return ERR_PTR(-ENOMEM);
194 
195 	ret = vfio_init_device(device, dev, ops);
196 	if (ret)
197 		goto out_free;
198 	return device;
199 
200 out_free:
201 	kvfree(device);
202 	return ERR_PTR(ret);
203 }
204 EXPORT_SYMBOL_GPL(_vfio_alloc_device);
205 
206 /*
207  * Initialize a vfio_device so it can be registered to vfio core.
208  */
209 static int vfio_init_device(struct vfio_device *device, struct device *dev,
210 			    const struct vfio_device_ops *ops)
211 {
212 	int ret;
213 
214 	ret = ida_alloc_max(&vfio.device_ida, MINORMASK, GFP_KERNEL);
215 	if (ret < 0) {
216 		dev_dbg(dev, "Error to alloc index\n");
217 		return ret;
218 	}
219 
220 	device->index = ret;
221 	init_completion(&device->comp);
222 	device->dev = dev;
223 	device->ops = ops;
224 
225 	if (ops->init) {
226 		ret = ops->init(device);
227 		if (ret)
228 			goto out_uninit;
229 	}
230 
231 	device_initialize(&device->device);
232 	device->device.release = vfio_device_release;
233 	device->device.class = vfio.device_class;
234 	device->device.parent = device->dev;
235 	return 0;
236 
237 out_uninit:
238 	vfio_release_device_set(device);
239 	ida_free(&vfio.device_ida, device->index);
240 	return ret;
241 }
242 
243 static int __vfio_register_dev(struct vfio_device *device,
244 			       enum vfio_group_type type)
245 {
246 	int ret;
247 
248 	if (WARN_ON(device->ops->bind_iommufd &&
249 		    (!device->ops->unbind_iommufd ||
250 		     !device->ops->attach_ioas)))
251 		return -EINVAL;
252 
253 	/*
254 	 * If the driver doesn't specify a set then the device is added to a
255 	 * singleton set just for itself.
256 	 */
257 	if (!device->dev_set)
258 		vfio_assign_device_set(device, device);
259 
260 	ret = dev_set_name(&device->device, "vfio%d", device->index);
261 	if (ret)
262 		return ret;
263 
264 	ret = vfio_device_set_group(device, type);
265 	if (ret)
266 		return ret;
267 
268 	ret = device_add(&device->device);
269 	if (ret)
270 		goto err_out;
271 
272 	/* Refcounting can't start until the driver calls register */
273 	refcount_set(&device->refcount, 1);
274 
275 	vfio_device_group_register(device);
276 
277 	return 0;
278 err_out:
279 	vfio_device_remove_group(device);
280 	return ret;
281 }
282 
283 int vfio_register_group_dev(struct vfio_device *device)
284 {
285 	return __vfio_register_dev(device, VFIO_IOMMU);
286 }
287 EXPORT_SYMBOL_GPL(vfio_register_group_dev);
288 
289 /*
290  * Register a virtual device without IOMMU backing.  The user of this
291  * device must not be able to directly trigger unmediated DMA.
292  */
293 int vfio_register_emulated_iommu_dev(struct vfio_device *device)
294 {
295 	return __vfio_register_dev(device, VFIO_EMULATED_IOMMU);
296 }
297 EXPORT_SYMBOL_GPL(vfio_register_emulated_iommu_dev);
298 
299 /*
300  * Decrement the device reference count and wait for the device to be
301  * removed.  Open file descriptors for the device... */
302 void vfio_unregister_group_dev(struct vfio_device *device)
303 {
304 	unsigned int i = 0;
305 	bool interrupted = false;
306 	long rc;
307 
308 	vfio_device_put_registration(device);
309 	rc = try_wait_for_completion(&device->comp);
310 	while (rc <= 0) {
311 		if (device->ops->request)
312 			device->ops->request(device, i++);
313 
314 		if (interrupted) {
315 			rc = wait_for_completion_timeout(&device->comp,
316 							 HZ * 10);
317 		} else {
318 			rc = wait_for_completion_interruptible_timeout(
319 				&device->comp, HZ * 10);
320 			if (rc < 0) {
321 				interrupted = true;
322 				dev_warn(device->dev,
323 					 "Device is currently in use, task"
324 					 " \"%s\" (%d) "
325 					 "blocked until device is released",
326 					 current->comm, task_pid_nr(current));
327 			}
328 		}
329 	}
330 
331 	vfio_device_group_unregister(device);
332 
333 	/* Balances device_add in register path */
334 	device_del(&device->device);
335 
336 	/* Balances vfio_device_set_group in register path */
337 	vfio_device_remove_group(device);
338 }
339 EXPORT_SYMBOL_GPL(vfio_unregister_group_dev);
340 
341 /* true if the vfio_device has open_device() called but not close_device() */
342 static bool vfio_assert_device_open(struct vfio_device *device)
343 {
344 	return !WARN_ON_ONCE(!READ_ONCE(device->open_count));
345 }
346 
347 static int vfio_device_first_open(struct vfio_device *device,
348 				  struct iommufd_ctx *iommufd, struct kvm *kvm)
349 {
350 	int ret;
351 
352 	lockdep_assert_held(&device->dev_set->lock);
353 
354 	if (!try_module_get(device->dev->driver->owner))
355 		return -ENODEV;
356 
357 	if (iommufd)
358 		ret = vfio_iommufd_bind(device, iommufd);
359 	else
360 		ret = vfio_device_group_use_iommu(device);
361 	if (ret)
362 		goto err_module_put;
363 
364 	device->kvm = kvm;
365 	if (device->ops->open_device) {
366 		ret = device->ops->open_device(device);
367 		if (ret)
368 			goto err_unuse_iommu;
369 	}
370 	return 0;
371 
372 err_unuse_iommu:
373 	device->kvm = NULL;
374 	if (iommufd)
375 		vfio_iommufd_unbind(device);
376 	else
377 		vfio_device_group_unuse_iommu(device);
378 err_module_put:
379 	module_put(device->dev->driver->owner);
380 	return ret;
381 }
382 
383 static void vfio_device_last_close(struct vfio_device *device,
384 				   struct iommufd_ctx *iommufd)
385 {
386 	lockdep_assert_held(&device->dev_set->lock);
387 
388 	if (device->ops->close_device)
389 		device->ops->close_device(device);
390 	device->kvm = NULL;
391 	if (iommufd)
392 		vfio_iommufd_unbind(device);
393 	else
394 		vfio_device_group_unuse_iommu(device);
395 	module_put(device->dev->driver->owner);
396 }
397 
398 int vfio_device_open(struct vfio_device *device,
399 		     struct iommufd_ctx *iommufd, struct kvm *kvm)
400 {
401 	int ret = 0;
402 
403 	mutex_lock(&device->dev_set->lock);
404 	device->open_count++;
405 	if (device->open_count == 1) {
406 		ret = vfio_device_first_open(device, iommufd, kvm);
407 		if (ret)
408 			device->open_count--;
409 	}
410 	mutex_unlock(&device->dev_set->lock);
411 
412 	return ret;
413 }
414 
415 void vfio_device_close(struct vfio_device *device,
416 		       struct iommufd_ctx *iommufd)
417 {
418 	mutex_lock(&device->dev_set->lock);
419 	vfio_assert_device_open(device);
420 	if (device->open_count == 1)
421 		vfio_device_last_close(device, iommufd);
422 	device->open_count--;
423 	mutex_unlock(&device->dev_set->lock);
424 }
425 
426 /*
427  * Wrapper around pm_runtime_resume_and_get().
428  * Return error code on failure or 0 on success.
429  */
430 static inline int vfio_device_pm_runtime_get(struct vfio_device *device)
431 {
432 	struct device *dev = device->dev;
433 
434 	if (dev->driver && dev->driver->pm) {
435 		int ret;
436 
437 		ret = pm_runtime_resume_and_get(dev);
438 		if (ret) {
439 			dev_info_ratelimited(dev,
440 				"vfio: runtime resume failed %d\n", ret);
441 			return -EIO;
442 		}
443 	}
444 
445 	return 0;
446 }
447 
448 /*
449  * Wrapper around pm_runtime_put().
450  */
451 static inline void vfio_device_pm_runtime_put(struct vfio_device *device)
452 {
453 	struct device *dev = device->dev;
454 
455 	if (dev->driver && dev->driver->pm)
456 		pm_runtime_put(dev);
457 }
458 
459 /*
460  * VFIO Device fd
461  */
462 static int vfio_device_fops_release(struct inode *inode, struct file *filep)
463 {
464 	struct vfio_device *device = filep->private_data;
465 
466 	vfio_device_group_close(device);
467 
468 	vfio_device_put_registration(device);
469 
470 	return 0;
471 }
472 
473 /*
474  * vfio_mig_get_next_state - Compute the next step in the FSM
475  * @cur_fsm - The current state the device is in
476  * @new_fsm - The target state to reach
477  * @next_fsm - Pointer to the next step to get to new_fsm
478  *
479  * Return 0 upon success, otherwise -errno
480  * Upon success the next step in the state progression between cur_fsm and
481  * new_fsm will be set in next_fsm.
482  *
483  * This breaks down requests for combination transitions into smaller steps and
484  * returns the next step to get to new_fsm. The function may need to be called
485  * multiple times before reaching new_fsm.
486  *
487  */
488 int vfio_mig_get_next_state(struct vfio_device *device,
489 			    enum vfio_device_mig_state cur_fsm,
490 			    enum vfio_device_mig_state new_fsm,
491 			    enum vfio_device_mig_state *next_fsm)
492 {
493 	enum { VFIO_DEVICE_NUM_STATES = VFIO_DEVICE_STATE_PRE_COPY_P2P + 1 };
494 	/*
495 	 * The coding in this table requires the driver to implement the
496 	 * following FSM arcs:
497 	 *         RESUMING -> STOP
498 	 *         STOP -> RESUMING
499 	 *         STOP -> STOP_COPY
500 	 *         STOP_COPY -> STOP
501 	 *
502 	 * If P2P is supported then the driver must also implement these FSM
503 	 * arcs:
504 	 *         RUNNING -> RUNNING_P2P
505 	 *         RUNNING_P2P -> RUNNING
506 	 *         RUNNING_P2P -> STOP
507 	 *         STOP -> RUNNING_P2P
508 	 *
509 	 * If precopy is supported then the driver must support these additional
510 	 * FSM arcs:
511 	 *         RUNNING -> PRE_COPY
512 	 *         PRE_COPY -> RUNNING
513 	 *         PRE_COPY -> STOP_COPY
514 	 * However, if precopy and P2P are supported together then the driver
515 	 * must support these additional arcs beyond the P2P arcs above:
516 	 *         PRE_COPY -> RUNNING
517 	 *         PRE_COPY -> PRE_COPY_P2P
518 	 *         PRE_COPY_P2P -> PRE_COPY
519 	 *         PRE_COPY_P2P -> RUNNING_P2P
520 	 *         PRE_COPY_P2P -> STOP_COPY
521 	 *         RUNNING -> PRE_COPY
522 	 *         RUNNING_P2P -> PRE_COPY_P2P
523 	 *
524 	 * Without P2P and precopy the driver must implement:
525 	 *         RUNNING -> STOP
526 	 *         STOP -> RUNNING
527 	 *
528 	 * The coding will step through multiple states for some combination
529 	 * transitions; if all optional features are supported, this means the
530 	 * following ones:
531 	 *         PRE_COPY -> PRE_COPY_P2P -> STOP_COPY
532 	 *         PRE_COPY -> RUNNING -> RUNNING_P2P
533 	 *         PRE_COPY -> RUNNING -> RUNNING_P2P -> STOP
534 	 *         PRE_COPY -> RUNNING -> RUNNING_P2P -> STOP -> RESUMING
535 	 *         PRE_COPY_P2P -> RUNNING_P2P -> RUNNING
536 	 *         PRE_COPY_P2P -> RUNNING_P2P -> STOP
537 	 *         PRE_COPY_P2P -> RUNNING_P2P -> STOP -> RESUMING
538 	 *         RESUMING -> STOP -> RUNNING_P2P
539 	 *         RESUMING -> STOP -> RUNNING_P2P -> PRE_COPY_P2P
540 	 *         RESUMING -> STOP -> RUNNING_P2P -> RUNNING
541 	 *         RESUMING -> STOP -> RUNNING_P2P -> RUNNING -> PRE_COPY
542 	 *         RESUMING -> STOP -> STOP_COPY
543 	 *         RUNNING -> RUNNING_P2P -> PRE_COPY_P2P
544 	 *         RUNNING -> RUNNING_P2P -> STOP
545 	 *         RUNNING -> RUNNING_P2P -> STOP -> RESUMING
546 	 *         RUNNING -> RUNNING_P2P -> STOP -> STOP_COPY
547 	 *         RUNNING_P2P -> RUNNING -> PRE_COPY
548 	 *         RUNNING_P2P -> STOP -> RESUMING
549 	 *         RUNNING_P2P -> STOP -> STOP_COPY
550 	 *         STOP -> RUNNING_P2P -> PRE_COPY_P2P
551 	 *         STOP -> RUNNING_P2P -> RUNNING
552 	 *         STOP -> RUNNING_P2P -> RUNNING -> PRE_COPY
553 	 *         STOP_COPY -> STOP -> RESUMING
554 	 *         STOP_COPY -> STOP -> RUNNING_P2P
555 	 *         STOP_COPY -> STOP -> RUNNING_P2P -> RUNNING
556 	 *
557 	 *  The following transitions are blocked:
558 	 *         STOP_COPY -> PRE_COPY
559 	 *         STOP_COPY -> PRE_COPY_P2P
560 	 */
561 	static const u8 vfio_from_fsm_table[VFIO_DEVICE_NUM_STATES][VFIO_DEVICE_NUM_STATES] = {
562 		[VFIO_DEVICE_STATE_STOP] = {
563 			[VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_STOP,
564 			[VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_RUNNING_P2P,
565 			[VFIO_DEVICE_STATE_PRE_COPY] = VFIO_DEVICE_STATE_RUNNING_P2P,
566 			[VFIO_DEVICE_STATE_PRE_COPY_P2P] = VFIO_DEVICE_STATE_RUNNING_P2P,
567 			[VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_STOP_COPY,
568 			[VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_RESUMING,
569 			[VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_RUNNING_P2P,
570 			[VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
571 		},
572 		[VFIO_DEVICE_STATE_RUNNING] = {
573 			[VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_RUNNING_P2P,
574 			[VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_RUNNING,
575 			[VFIO_DEVICE_STATE_PRE_COPY] = VFIO_DEVICE_STATE_PRE_COPY,
576 			[VFIO_DEVICE_STATE_PRE_COPY_P2P] = VFIO_DEVICE_STATE_RUNNING_P2P,
577 			[VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_RUNNING_P2P,
578 			[VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_RUNNING_P2P,
579 			[VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_RUNNING_P2P,
580 			[VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
581 		},
582 		[VFIO_DEVICE_STATE_PRE_COPY] = {
583 			[VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_RUNNING,
584 			[VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_RUNNING,
585 			[VFIO_DEVICE_STATE_PRE_COPY] = VFIO_DEVICE_STATE_PRE_COPY,
586 			[VFIO_DEVICE_STATE_PRE_COPY_P2P] = VFIO_DEVICE_STATE_PRE_COPY_P2P,
587 			[VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_PRE_COPY_P2P,
588 			[VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_RUNNING,
589 			[VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_RUNNING,
590 			[VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
591 		},
592 		[VFIO_DEVICE_STATE_PRE_COPY_P2P] = {
593 			[VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_RUNNING_P2P,
594 			[VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_RUNNING_P2P,
595 			[VFIO_DEVICE_STATE_PRE_COPY] = VFIO_DEVICE_STATE_PRE_COPY,
596 			[VFIO_DEVICE_STATE_PRE_COPY_P2P] = VFIO_DEVICE_STATE_PRE_COPY_P2P,
597 			[VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_STOP_COPY,
598 			[VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_RUNNING_P2P,
599 			[VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_RUNNING_P2P,
600 			[VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
601 		},
602 		[VFIO_DEVICE_STATE_STOP_COPY] = {
603 			[VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_STOP,
604 			[VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_STOP,
605 			[VFIO_DEVICE_STATE_PRE_COPY] = VFIO_DEVICE_STATE_ERROR,
606 			[VFIO_DEVICE_STATE_PRE_COPY_P2P] = VFIO_DEVICE_STATE_ERROR,
607 			[VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_STOP_COPY,
608 			[VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_STOP,
609 			[VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_STOP,
610 			[VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
611 		},
612 		[VFIO_DEVICE_STATE_RESUMING] = {
613 			[VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_STOP,
614 			[VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_STOP,
615 			[VFIO_DEVICE_STATE_PRE_COPY] = VFIO_DEVICE_STATE_STOP,
616 			[VFIO_DEVICE_STATE_PRE_COPY_P2P] = VFIO_DEVICE_STATE_STOP,
617 			[VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_STOP,
618 			[VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_RESUMING,
619 			[VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_STOP,
620 			[VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
621 		},
622 		[VFIO_DEVICE_STATE_RUNNING_P2P] = {
623 			[VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_STOP,
624 			[VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_RUNNING,
625 			[VFIO_DEVICE_STATE_PRE_COPY] = VFIO_DEVICE_STATE_RUNNING,
626 			[VFIO_DEVICE_STATE_PRE_COPY_P2P] = VFIO_DEVICE_STATE_PRE_COPY_P2P,
627 			[VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_STOP,
628 			[VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_STOP,
629 			[VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_RUNNING_P2P,
630 			[VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
631 		},
632 		[VFIO_DEVICE_STATE_ERROR] = {
633 			[VFIO_DEVICE_STATE_STOP] = VFIO_DEVICE_STATE_ERROR,
634 			[VFIO_DEVICE_STATE_RUNNING] = VFIO_DEVICE_STATE_ERROR,
635 			[VFIO_DEVICE_STATE_PRE_COPY] = VFIO_DEVICE_STATE_ERROR,
636 			[VFIO_DEVICE_STATE_PRE_COPY_P2P] = VFIO_DEVICE_STATE_ERROR,
637 			[VFIO_DEVICE_STATE_STOP_COPY] = VFIO_DEVICE_STATE_ERROR,
638 			[VFIO_DEVICE_STATE_RESUMING] = VFIO_DEVICE_STATE_ERROR,
639 			[VFIO_DEVICE_STATE_RUNNING_P2P] = VFIO_DEVICE_STATE_ERROR,
640 			[VFIO_DEVICE_STATE_ERROR] = VFIO_DEVICE_STATE_ERROR,
641 		},
642 	};
643 
644 	static const unsigned int state_flags_table[VFIO_DEVICE_NUM_STATES] = {
645 		[VFIO_DEVICE_STATE_STOP] = VFIO_MIGRATION_STOP_COPY,
646 		[VFIO_DEVICE_STATE_RUNNING] = VFIO_MIGRATION_STOP_COPY,
647 		[VFIO_DEVICE_STATE_PRE_COPY] =
648 			VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_PRE_COPY,
649 		[VFIO_DEVICE_STATE_PRE_COPY_P2P] = VFIO_MIGRATION_STOP_COPY |
650 						   VFIO_MIGRATION_P2P |
651 						   VFIO_MIGRATION_PRE_COPY,
652 		[VFIO_DEVICE_STATE_STOP_COPY] = VFIO_MIGRATION_STOP_COPY,
653 		[VFIO_DEVICE_STATE_RESUMING] = VFIO_MIGRATION_STOP_COPY,
654 		[VFIO_DEVICE_STATE_RUNNING_P2P] =
655 			VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P,
656 		[VFIO_DEVICE_STATE_ERROR] = ~0U,
657 	};
658 
659 	if (WARN_ON(cur_fsm >= ARRAY_SIZE(vfio_from_fsm_table) ||
660 		    (state_flags_table[cur_fsm] & device->migration_flags) !=
661 			state_flags_table[cur_fsm]))
662 		return -EINVAL;
663 
664 	if (new_fsm >= ARRAY_SIZE(vfio_from_fsm_table) ||
665 	   (state_flags_table[new_fsm] & device->migration_flags) !=
666 			state_flags_table[new_fsm])
667 		return -EINVAL;
668 
669 	/*
670 	 * Arcs touching optional and unsupported states are skipped over. The
671 	 * driver will instead see an arc from the original state to the next
672 	 * logical state, as per the above comment.
673 	 */
674 	*next_fsm = vfio_from_fsm_table[cur_fsm][new_fsm];
675 	while ((state_flags_table[*next_fsm] & device->migration_flags) !=
676 			state_flags_table[*next_fsm])
677 		*next_fsm = vfio_from_fsm_table[*next_fsm][new_fsm];
678 
679 	return (*next_fsm != VFIO_DEVICE_STATE_ERROR) ? 0 : -EINVAL;
680 }
681 EXPORT_SYMBOL_GPL(vfio_mig_get_next_state);
682 
683 /*
684  * Convert the drivers's struct file into a FD number and return it to userspace
685  */
686 static int vfio_ioct_mig_return_fd(struct file *filp, void __user *arg,
687 				   struct vfio_device_feature_mig_state *mig)
688 {
689 	int ret;
690 	int fd;
691 
692 	fd = get_unused_fd_flags(O_CLOEXEC);
693 	if (fd < 0) {
694 		ret = fd;
695 		goto out_fput;
696 	}
697 
698 	mig->data_fd = fd;
699 	if (copy_to_user(arg, mig, sizeof(*mig))) {
700 		ret = -EFAULT;
701 		goto out_put_unused;
702 	}
703 	fd_install(fd, filp);
704 	return 0;
705 
706 out_put_unused:
707 	put_unused_fd(fd);
708 out_fput:
709 	fput(filp);
710 	return ret;
711 }
712 
713 static int
714 vfio_ioctl_device_feature_mig_device_state(struct vfio_device *device,
715 					   u32 flags, void __user *arg,
716 					   size_t argsz)
717 {
718 	size_t minsz =
719 		offsetofend(struct vfio_device_feature_mig_state, data_fd);
720 	struct vfio_device_feature_mig_state mig;
721 	struct file *filp = NULL;
722 	int ret;
723 
724 	if (!device->mig_ops)
725 		return -ENOTTY;
726 
727 	ret = vfio_check_feature(flags, argsz,
728 				 VFIO_DEVICE_FEATURE_SET |
729 				 VFIO_DEVICE_FEATURE_GET,
730 				 sizeof(mig));
731 	if (ret != 1)
732 		return ret;
733 
734 	if (copy_from_user(&mig, arg, minsz))
735 		return -EFAULT;
736 
737 	if (flags & VFIO_DEVICE_FEATURE_GET) {
738 		enum vfio_device_mig_state curr_state;
739 
740 		ret = device->mig_ops->migration_get_state(device,
741 							   &curr_state);
742 		if (ret)
743 			return ret;
744 		mig.device_state = curr_state;
745 		goto out_copy;
746 	}
747 
748 	/* Handle the VFIO_DEVICE_FEATURE_SET */
749 	filp = device->mig_ops->migration_set_state(device, mig.device_state);
750 	if (IS_ERR(filp) || !filp)
751 		goto out_copy;
752 
753 	return vfio_ioct_mig_return_fd(filp, arg, &mig);
754 out_copy:
755 	mig.data_fd = -1;
756 	if (copy_to_user(arg, &mig, sizeof(mig)))
757 		return -EFAULT;
758 	if (IS_ERR(filp))
759 		return PTR_ERR(filp);
760 	return 0;
761 }
762 
763 static int
764 vfio_ioctl_device_feature_migration_data_size(struct vfio_device *device,
765 					      u32 flags, void __user *arg,
766 					      size_t argsz)
767 {
768 	struct vfio_device_feature_mig_data_size data_size = {};
769 	unsigned long stop_copy_length;
770 	int ret;
771 
772 	if (!device->mig_ops)
773 		return -ENOTTY;
774 
775 	ret = vfio_check_feature(flags, argsz, VFIO_DEVICE_FEATURE_GET,
776 				 sizeof(data_size));
777 	if (ret != 1)
778 		return ret;
779 
780 	ret = device->mig_ops->migration_get_data_size(device, &stop_copy_length);
781 	if (ret)
782 		return ret;
783 
784 	data_size.stop_copy_length = stop_copy_length;
785 	if (copy_to_user(arg, &data_size, sizeof(data_size)))
786 		return -EFAULT;
787 
788 	return 0;
789 }
790 
791 static int vfio_ioctl_device_feature_migration(struct vfio_device *device,
792 					       u32 flags, void __user *arg,
793 					       size_t argsz)
794 {
795 	struct vfio_device_feature_migration mig = {
796 		.flags = device->migration_flags,
797 	};
798 	int ret;
799 
800 	if (!device->mig_ops)
801 		return -ENOTTY;
802 
803 	ret = vfio_check_feature(flags, argsz, VFIO_DEVICE_FEATURE_GET,
804 				 sizeof(mig));
805 	if (ret != 1)
806 		return ret;
807 	if (copy_to_user(arg, &mig, sizeof(mig)))
808 		return -EFAULT;
809 	return 0;
810 }
811 
812 /* Ranges should fit into a single kernel page */
813 #define LOG_MAX_RANGES \
814 	(PAGE_SIZE / sizeof(struct vfio_device_feature_dma_logging_range))
815 
816 static int
817 vfio_ioctl_device_feature_logging_start(struct vfio_device *device,
818 					u32 flags, void __user *arg,
819 					size_t argsz)
820 {
821 	size_t minsz =
822 		offsetofend(struct vfio_device_feature_dma_logging_control,
823 			    ranges);
824 	struct vfio_device_feature_dma_logging_range __user *ranges;
825 	struct vfio_device_feature_dma_logging_control control;
826 	struct vfio_device_feature_dma_logging_range range;
827 	struct rb_root_cached root = RB_ROOT_CACHED;
828 	struct interval_tree_node *nodes;
829 	u64 iova_end;
830 	u32 nnodes;
831 	int i, ret;
832 
833 	if (!device->log_ops)
834 		return -ENOTTY;
835 
836 	ret = vfio_check_feature(flags, argsz,
837 				 VFIO_DEVICE_FEATURE_SET,
838 				 sizeof(control));
839 	if (ret != 1)
840 		return ret;
841 
842 	if (copy_from_user(&control, arg, minsz))
843 		return -EFAULT;
844 
845 	nnodes = control.num_ranges;
846 	if (!nnodes)
847 		return -EINVAL;
848 
849 	if (nnodes > LOG_MAX_RANGES)
850 		return -E2BIG;
851 
852 	ranges = u64_to_user_ptr(control.ranges);
853 	nodes = kmalloc_array(nnodes, sizeof(struct interval_tree_node),
854 			      GFP_KERNEL);
855 	if (!nodes)
856 		return -ENOMEM;
857 
858 	for (i = 0; i < nnodes; i++) {
859 		if (copy_from_user(&range, &ranges[i], sizeof(range))) {
860 			ret = -EFAULT;
861 			goto end;
862 		}
863 		if (!IS_ALIGNED(range.iova, control.page_size) ||
864 		    !IS_ALIGNED(range.length, control.page_size)) {
865 			ret = -EINVAL;
866 			goto end;
867 		}
868 
869 		if (check_add_overflow(range.iova, range.length, &iova_end) ||
870 		    iova_end > ULONG_MAX) {
871 			ret = -EOVERFLOW;
872 			goto end;
873 		}
874 
875 		nodes[i].start = range.iova;
876 		nodes[i].last = range.iova + range.length - 1;
877 		if (interval_tree_iter_first(&root, nodes[i].start,
878 					     nodes[i].last)) {
879 			/* Range overlapping */
880 			ret = -EINVAL;
881 			goto end;
882 		}
883 		interval_tree_insert(nodes + i, &root);
884 	}
885 
886 	ret = device->log_ops->log_start(device, &root, nnodes,
887 					 &control.page_size);
888 	if (ret)
889 		goto end;
890 
891 	if (copy_to_user(arg, &control, sizeof(control))) {
892 		ret = -EFAULT;
893 		device->log_ops->log_stop(device);
894 	}
895 
896 end:
897 	kfree(nodes);
898 	return ret;
899 }
900 
901 static int
902 vfio_ioctl_device_feature_logging_stop(struct vfio_device *device,
903 				       u32 flags, void __user *arg,
904 				       size_t argsz)
905 {
906 	int ret;
907 
908 	if (!device->log_ops)
909 		return -ENOTTY;
910 
911 	ret = vfio_check_feature(flags, argsz,
912 				 VFIO_DEVICE_FEATURE_SET, 0);
913 	if (ret != 1)
914 		return ret;
915 
916 	return device->log_ops->log_stop(device);
917 }
918 
919 static int vfio_device_log_read_and_clear(struct iova_bitmap *iter,
920 					  unsigned long iova, size_t length,
921 					  void *opaque)
922 {
923 	struct vfio_device *device = opaque;
924 
925 	return device->log_ops->log_read_and_clear(device, iova, length, iter);
926 }
927 
928 static int
929 vfio_ioctl_device_feature_logging_report(struct vfio_device *device,
930 					 u32 flags, void __user *arg,
931 					 size_t argsz)
932 {
933 	size_t minsz =
934 		offsetofend(struct vfio_device_feature_dma_logging_report,
935 			    bitmap);
936 	struct vfio_device_feature_dma_logging_report report;
937 	struct iova_bitmap *iter;
938 	u64 iova_end;
939 	int ret;
940 
941 	if (!device->log_ops)
942 		return -ENOTTY;
943 
944 	ret = vfio_check_feature(flags, argsz,
945 				 VFIO_DEVICE_FEATURE_GET,
946 				 sizeof(report));
947 	if (ret != 1)
948 		return ret;
949 
950 	if (copy_from_user(&report, arg, minsz))
951 		return -EFAULT;
952 
953 	if (report.page_size < SZ_4K || !is_power_of_2(report.page_size))
954 		return -EINVAL;
955 
956 	if (check_add_overflow(report.iova, report.length, &iova_end) ||
957 	    iova_end > ULONG_MAX)
958 		return -EOVERFLOW;
959 
960 	iter = iova_bitmap_alloc(report.iova, report.length,
961 				 report.page_size,
962 				 u64_to_user_ptr(report.bitmap));
963 	if (IS_ERR(iter))
964 		return PTR_ERR(iter);
965 
966 	ret = iova_bitmap_for_each(iter, device,
967 				   vfio_device_log_read_and_clear);
968 
969 	iova_bitmap_free(iter);
970 	return ret;
971 }
972 
973 static int vfio_ioctl_device_feature(struct vfio_device *device,
974 				     struct vfio_device_feature __user *arg)
975 {
976 	size_t minsz = offsetofend(struct vfio_device_feature, flags);
977 	struct vfio_device_feature feature;
978 
979 	if (copy_from_user(&feature, arg, minsz))
980 		return -EFAULT;
981 
982 	if (feature.argsz < minsz)
983 		return -EINVAL;
984 
985 	/* Check unknown flags */
986 	if (feature.flags &
987 	    ~(VFIO_DEVICE_FEATURE_MASK | VFIO_DEVICE_FEATURE_SET |
988 	      VFIO_DEVICE_FEATURE_GET | VFIO_DEVICE_FEATURE_PROBE))
989 		return -EINVAL;
990 
991 	/* GET & SET are mutually exclusive except with PROBE */
992 	if (!(feature.flags & VFIO_DEVICE_FEATURE_PROBE) &&
993 	    (feature.flags & VFIO_DEVICE_FEATURE_SET) &&
994 	    (feature.flags & VFIO_DEVICE_FEATURE_GET))
995 		return -EINVAL;
996 
997 	switch (feature.flags & VFIO_DEVICE_FEATURE_MASK) {
998 	case VFIO_DEVICE_FEATURE_MIGRATION:
999 		return vfio_ioctl_device_feature_migration(
1000 			device, feature.flags, arg->data,
1001 			feature.argsz - minsz);
1002 	case VFIO_DEVICE_FEATURE_MIG_DEVICE_STATE:
1003 		return vfio_ioctl_device_feature_mig_device_state(
1004 			device, feature.flags, arg->data,
1005 			feature.argsz - minsz);
1006 	case VFIO_DEVICE_FEATURE_DMA_LOGGING_START:
1007 		return vfio_ioctl_device_feature_logging_start(
1008 			device, feature.flags, arg->data,
1009 			feature.argsz - minsz);
1010 	case VFIO_DEVICE_FEATURE_DMA_LOGGING_STOP:
1011 		return vfio_ioctl_device_feature_logging_stop(
1012 			device, feature.flags, arg->data,
1013 			feature.argsz - minsz);
1014 	case VFIO_DEVICE_FEATURE_DMA_LOGGING_REPORT:
1015 		return vfio_ioctl_device_feature_logging_report(
1016 			device, feature.flags, arg->data,
1017 			feature.argsz - minsz);
1018 	case VFIO_DEVICE_FEATURE_MIG_DATA_SIZE:
1019 		return vfio_ioctl_device_feature_migration_data_size(
1020 			device, feature.flags, arg->data,
1021 			feature.argsz - minsz);
1022 	default:
1023 		if (unlikely(!device->ops->device_feature))
1024 			return -EINVAL;
1025 		return device->ops->device_feature(device, feature.flags,
1026 						   arg->data,
1027 						   feature.argsz - minsz);
1028 	}
1029 }
1030 
1031 static long vfio_device_fops_unl_ioctl(struct file *filep,
1032 				       unsigned int cmd, unsigned long arg)
1033 {
1034 	struct vfio_device *device = filep->private_data;
1035 	int ret;
1036 
1037 	ret = vfio_device_pm_runtime_get(device);
1038 	if (ret)
1039 		return ret;
1040 
1041 	switch (cmd) {
1042 	case VFIO_DEVICE_FEATURE:
1043 		ret = vfio_ioctl_device_feature(device, (void __user *)arg);
1044 		break;
1045 
1046 	default:
1047 		if (unlikely(!device->ops->ioctl))
1048 			ret = -EINVAL;
1049 		else
1050 			ret = device->ops->ioctl(device, cmd, arg);
1051 		break;
1052 	}
1053 
1054 	vfio_device_pm_runtime_put(device);
1055 	return ret;
1056 }
1057 
1058 static ssize_t vfio_device_fops_read(struct file *filep, char __user *buf,
1059 				     size_t count, loff_t *ppos)
1060 {
1061 	struct vfio_device *device = filep->private_data;
1062 
1063 	if (unlikely(!device->ops->read))
1064 		return -EINVAL;
1065 
1066 	return device->ops->read(device, buf, count, ppos);
1067 }
1068 
1069 static ssize_t vfio_device_fops_write(struct file *filep,
1070 				      const char __user *buf,
1071 				      size_t count, loff_t *ppos)
1072 {
1073 	struct vfio_device *device = filep->private_data;
1074 
1075 	if (unlikely(!device->ops->write))
1076 		return -EINVAL;
1077 
1078 	return device->ops->write(device, buf, count, ppos);
1079 }
1080 
1081 static int vfio_device_fops_mmap(struct file *filep, struct vm_area_struct *vma)
1082 {
1083 	struct vfio_device *device = filep->private_data;
1084 
1085 	if (unlikely(!device->ops->mmap))
1086 		return -EINVAL;
1087 
1088 	return device->ops->mmap(device, vma);
1089 }
1090 
1091 const struct file_operations vfio_device_fops = {
1092 	.owner		= THIS_MODULE,
1093 	.release	= vfio_device_fops_release,
1094 	.read		= vfio_device_fops_read,
1095 	.write		= vfio_device_fops_write,
1096 	.unlocked_ioctl	= vfio_device_fops_unl_ioctl,
1097 	.compat_ioctl	= compat_ptr_ioctl,
1098 	.mmap		= vfio_device_fops_mmap,
1099 };
1100 
1101 /*
1102  * Sub-module support
1103  */
1104 /*
1105  * Helper for managing a buffer of info chain capabilities, allocate or
1106  * reallocate a buffer with additional @size, filling in @id and @version
1107  * of the capability.  A pointer to the new capability is returned.
1108  *
1109  * NB. The chain is based at the head of the buffer, so new entries are
1110  * added to the tail, vfio_info_cap_shift() should be called to fixup the
1111  * next offsets prior to copying to the user buffer.
1112  */
1113 struct vfio_info_cap_header *vfio_info_cap_add(struct vfio_info_cap *caps,
1114 					       size_t size, u16 id, u16 version)
1115 {
1116 	void *buf;
1117 	struct vfio_info_cap_header *header, *tmp;
1118 
1119 	buf = krealloc(caps->buf, caps->size + size, GFP_KERNEL);
1120 	if (!buf) {
1121 		kfree(caps->buf);
1122 		caps->buf = NULL;
1123 		caps->size = 0;
1124 		return ERR_PTR(-ENOMEM);
1125 	}
1126 
1127 	caps->buf = buf;
1128 	header = buf + caps->size;
1129 
1130 	/* Eventually copied to user buffer, zero */
1131 	memset(header, 0, size);
1132 
1133 	header->id = id;
1134 	header->version = version;
1135 
1136 	/* Add to the end of the capability chain */
1137 	for (tmp = buf; tmp->next; tmp = buf + tmp->next)
1138 		; /* nothing */
1139 
1140 	tmp->next = caps->size;
1141 	caps->size += size;
1142 
1143 	return header;
1144 }
1145 EXPORT_SYMBOL_GPL(vfio_info_cap_add);
1146 
1147 void vfio_info_cap_shift(struct vfio_info_cap *caps, size_t offset)
1148 {
1149 	struct vfio_info_cap_header *tmp;
1150 	void *buf = (void *)caps->buf;
1151 
1152 	for (tmp = buf; tmp->next; tmp = buf + tmp->next - offset)
1153 		tmp->next += offset;
1154 }
1155 EXPORT_SYMBOL(vfio_info_cap_shift);
1156 
1157 int vfio_info_add_capability(struct vfio_info_cap *caps,
1158 			     struct vfio_info_cap_header *cap, size_t size)
1159 {
1160 	struct vfio_info_cap_header *header;
1161 
1162 	header = vfio_info_cap_add(caps, size, cap->id, cap->version);
1163 	if (IS_ERR(header))
1164 		return PTR_ERR(header);
1165 
1166 	memcpy(header + 1, cap + 1, size - sizeof(*header));
1167 
1168 	return 0;
1169 }
1170 EXPORT_SYMBOL(vfio_info_add_capability);
1171 
1172 int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr, int num_irqs,
1173 				       int max_irq_type, size_t *data_size)
1174 {
1175 	unsigned long minsz;
1176 	size_t size;
1177 
1178 	minsz = offsetofend(struct vfio_irq_set, count);
1179 
1180 	if ((hdr->argsz < minsz) || (hdr->index >= max_irq_type) ||
1181 	    (hdr->count >= (U32_MAX - hdr->start)) ||
1182 	    (hdr->flags & ~(VFIO_IRQ_SET_DATA_TYPE_MASK |
1183 				VFIO_IRQ_SET_ACTION_TYPE_MASK)))
1184 		return -EINVAL;
1185 
1186 	if (data_size)
1187 		*data_size = 0;
1188 
1189 	if (hdr->start >= num_irqs || hdr->start + hdr->count > num_irqs)
1190 		return -EINVAL;
1191 
1192 	switch (hdr->flags & VFIO_IRQ_SET_DATA_TYPE_MASK) {
1193 	case VFIO_IRQ_SET_DATA_NONE:
1194 		size = 0;
1195 		break;
1196 	case VFIO_IRQ_SET_DATA_BOOL:
1197 		size = sizeof(uint8_t);
1198 		break;
1199 	case VFIO_IRQ_SET_DATA_EVENTFD:
1200 		size = sizeof(int32_t);
1201 		break;
1202 	default:
1203 		return -EINVAL;
1204 	}
1205 
1206 	if (size) {
1207 		if (hdr->argsz - minsz < hdr->count * size)
1208 			return -EINVAL;
1209 
1210 		if (!data_size)
1211 			return -EINVAL;
1212 
1213 		*data_size = hdr->count * size;
1214 	}
1215 
1216 	return 0;
1217 }
1218 EXPORT_SYMBOL(vfio_set_irqs_validate_and_prepare);
1219 
1220 /*
1221  * Pin contiguous user pages and return their associated host pages for local
1222  * domain only.
1223  * @device [in]  : device
1224  * @iova [in]    : starting IOVA of user pages to be pinned.
1225  * @npage [in]   : count of pages to be pinned.  This count should not
1226  *		   be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
1227  * @prot [in]    : protection flags
1228  * @pages[out]   : array of host pages
1229  * Return error or number of pages pinned.
1230  *
1231  * A driver may only call this function if the vfio_device was created
1232  * by vfio_register_emulated_iommu_dev() due to vfio_device_container_pin_pages().
1233  */
1234 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
1235 		   int npage, int prot, struct page **pages)
1236 {
1237 	/* group->container cannot change while a vfio device is open */
1238 	if (!pages || !npage || WARN_ON(!vfio_assert_device_open(device)))
1239 		return -EINVAL;
1240 	if (vfio_device_has_container(device))
1241 		return vfio_device_container_pin_pages(device, iova,
1242 						       npage, prot, pages);
1243 	if (device->iommufd_access) {
1244 		int ret;
1245 
1246 		if (iova > ULONG_MAX)
1247 			return -EINVAL;
1248 		/*
1249 		 * VFIO ignores the sub page offset, npages is from the start of
1250 		 * a PAGE_SIZE chunk of IOVA. The caller is expected to recover
1251 		 * the sub page offset by doing:
1252 		 *     pages[0] + (iova % PAGE_SIZE)
1253 		 */
1254 		ret = iommufd_access_pin_pages(
1255 			device->iommufd_access, ALIGN_DOWN(iova, PAGE_SIZE),
1256 			npage * PAGE_SIZE, pages,
1257 			(prot & IOMMU_WRITE) ? IOMMUFD_ACCESS_RW_WRITE : 0);
1258 		if (ret)
1259 			return ret;
1260 		return npage;
1261 	}
1262 	return -EINVAL;
1263 }
1264 EXPORT_SYMBOL(vfio_pin_pages);
1265 
1266 /*
1267  * Unpin contiguous host pages for local domain only.
1268  * @device [in]  : device
1269  * @iova [in]    : starting address of user pages to be unpinned.
1270  * @npage [in]   : count of pages to be unpinned.  This count should not
1271  *                 be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
1272  */
1273 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage)
1274 {
1275 	if (WARN_ON(!vfio_assert_device_open(device)))
1276 		return;
1277 
1278 	if (vfio_device_has_container(device)) {
1279 		vfio_device_container_unpin_pages(device, iova, npage);
1280 		return;
1281 	}
1282 	if (device->iommufd_access) {
1283 		if (WARN_ON(iova > ULONG_MAX))
1284 			return;
1285 		iommufd_access_unpin_pages(device->iommufd_access,
1286 					   ALIGN_DOWN(iova, PAGE_SIZE),
1287 					   npage * PAGE_SIZE);
1288 		return;
1289 	}
1290 }
1291 EXPORT_SYMBOL(vfio_unpin_pages);
1292 
1293 /*
1294  * This interface allows the CPUs to perform some sort of virtual DMA on
1295  * behalf of the device.
1296  *
1297  * CPUs read/write from/into a range of IOVAs pointing to user space memory
1298  * into/from a kernel buffer.
1299  *
1300  * As the read/write of user space memory is conducted via the CPUs and is
1301  * not a real device DMA, it is not necessary to pin the user space memory.
1302  *
1303  * @device [in]		: VFIO device
1304  * @iova [in]		: base IOVA of a user space buffer
1305  * @data [in]		: pointer to kernel buffer
1306  * @len [in]		: kernel buffer length
1307  * @write		: indicate read or write
1308  * Return error code on failure or 0 on success.
1309  */
1310 int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova, void *data,
1311 		size_t len, bool write)
1312 {
1313 	if (!data || len <= 0 || !vfio_assert_device_open(device))
1314 		return -EINVAL;
1315 
1316 	if (vfio_device_has_container(device))
1317 		return vfio_device_container_dma_rw(device, iova,
1318 						    data, len, write);
1319 
1320 	if (device->iommufd_access) {
1321 		unsigned int flags = 0;
1322 
1323 		if (iova > ULONG_MAX)
1324 			return -EINVAL;
1325 
1326 		/* VFIO historically tries to auto-detect a kthread */
1327 		if (!current->mm)
1328 			flags |= IOMMUFD_ACCESS_RW_KTHREAD;
1329 		if (write)
1330 			flags |= IOMMUFD_ACCESS_RW_WRITE;
1331 		return iommufd_access_rw(device->iommufd_access, iova, data,
1332 					 len, flags);
1333 	}
1334 	return -EINVAL;
1335 }
1336 EXPORT_SYMBOL(vfio_dma_rw);
1337 
1338 /*
1339  * Module/class support
1340  */
1341 static int __init vfio_init(void)
1342 {
1343 	int ret;
1344 
1345 	ida_init(&vfio.device_ida);
1346 
1347 	ret = vfio_group_init();
1348 	if (ret)
1349 		return ret;
1350 
1351 	ret = vfio_virqfd_init();
1352 	if (ret)
1353 		goto err_virqfd;
1354 
1355 	/* /sys/class/vfio-dev/vfioX */
1356 	vfio.device_class = class_create(THIS_MODULE, "vfio-dev");
1357 	if (IS_ERR(vfio.device_class)) {
1358 		ret = PTR_ERR(vfio.device_class);
1359 		goto err_dev_class;
1360 	}
1361 
1362 	pr_info(DRIVER_DESC " version: " DRIVER_VERSION "\n");
1363 	return 0;
1364 
1365 err_dev_class:
1366 	vfio_virqfd_exit();
1367 err_virqfd:
1368 	vfio_group_cleanup();
1369 	return ret;
1370 }
1371 
1372 static void __exit vfio_cleanup(void)
1373 {
1374 	ida_destroy(&vfio.device_ida);
1375 	class_destroy(vfio.device_class);
1376 	vfio.device_class = NULL;
1377 	vfio_virqfd_exit();
1378 	vfio_group_cleanup();
1379 	xa_destroy(&vfio_device_set_xa);
1380 }
1381 
1382 module_init(vfio_init);
1383 module_exit(vfio_cleanup);
1384 
1385 MODULE_VERSION(DRIVER_VERSION);
1386 MODULE_LICENSE("GPL v2");
1387 MODULE_AUTHOR(DRIVER_AUTHOR);
1388 MODULE_DESCRIPTION(DRIVER_DESC);
1389 MODULE_SOFTDEP("post: vfio_iommu_type1 vfio_iommu_spapr_tce");
1390