1 /*
2  * VFIO PCI interrupt handling
3  *
4  * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
5  *     Author: Alex Williamson <alex.williamson@redhat.com>
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  *
11  * Derived from original vfio:
12  * Copyright 2010 Cisco Systems, Inc.  All rights reserved.
13  * Author: Tom Lyon, pugs@cisco.com
14  */
15 
16 #include <linux/device.h>
17 #include <linux/interrupt.h>
18 #include <linux/eventfd.h>
19 #include <linux/pci.h>
20 #include <linux/file.h>
21 #include <linux/poll.h>
22 #include <linux/vfio.h>
23 #include <linux/wait.h>
24 #include <linux/workqueue.h>
25 #include <linux/slab.h>
26 
27 #include "vfio_pci_private.h"
28 
29 /*
30  * IRQfd - generic
31  */
32 struct virqfd {
33 	struct vfio_pci_device	*vdev;
34 	struct eventfd_ctx	*eventfd;
35 	int			(*handler)(struct vfio_pci_device *, void *);
36 	void			(*thread)(struct vfio_pci_device *, void *);
37 	void			*data;
38 	struct work_struct	inject;
39 	wait_queue_t		wait;
40 	poll_table		pt;
41 	struct work_struct	shutdown;
42 	struct virqfd		**pvirqfd;
43 };
44 
45 static struct workqueue_struct *vfio_irqfd_cleanup_wq;
46 
47 int __init vfio_pci_virqfd_init(void)
48 {
49 	vfio_irqfd_cleanup_wq =
50 		create_singlethread_workqueue("vfio-irqfd-cleanup");
51 	if (!vfio_irqfd_cleanup_wq)
52 		return -ENOMEM;
53 
54 	return 0;
55 }
56 
57 void vfio_pci_virqfd_exit(void)
58 {
59 	destroy_workqueue(vfio_irqfd_cleanup_wq);
60 }
61 
62 static void virqfd_deactivate(struct virqfd *virqfd)
63 {
64 	queue_work(vfio_irqfd_cleanup_wq, &virqfd->shutdown);
65 }
66 
67 static int virqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key)
68 {
69 	struct virqfd *virqfd = container_of(wait, struct virqfd, wait);
70 	unsigned long flags = (unsigned long)key;
71 
72 	if (flags & POLLIN) {
73 		/* An event has been signaled, call function */
74 		if ((!virqfd->handler ||
75 		     virqfd->handler(virqfd->vdev, virqfd->data)) &&
76 		    virqfd->thread)
77 			schedule_work(&virqfd->inject);
78 	}
79 
80 	if (flags & POLLHUP) {
81 		unsigned long flags;
82 		spin_lock_irqsave(&virqfd->vdev->irqlock, flags);
83 
84 		/*
85 		 * The eventfd is closing, if the virqfd has not yet been
86 		 * queued for release, as determined by testing whether the
87 		 * vdev pointer to it is still valid, queue it now.  As
88 		 * with kvm irqfds, we know we won't race against the virqfd
89 		 * going away because we hold wqh->lock to get here.
90 		 */
91 		if (*(virqfd->pvirqfd) == virqfd) {
92 			*(virqfd->pvirqfd) = NULL;
93 			virqfd_deactivate(virqfd);
94 		}
95 
96 		spin_unlock_irqrestore(&virqfd->vdev->irqlock, flags);
97 	}
98 
99 	return 0;
100 }
101 
102 static void virqfd_ptable_queue_proc(struct file *file,
103 				     wait_queue_head_t *wqh, poll_table *pt)
104 {
105 	struct virqfd *virqfd = container_of(pt, struct virqfd, pt);
106 	add_wait_queue(wqh, &virqfd->wait);
107 }
108 
109 static void virqfd_shutdown(struct work_struct *work)
110 {
111 	struct virqfd *virqfd = container_of(work, struct virqfd, shutdown);
112 	u64 cnt;
113 
114 	eventfd_ctx_remove_wait_queue(virqfd->eventfd, &virqfd->wait, &cnt);
115 	flush_work(&virqfd->inject);
116 	eventfd_ctx_put(virqfd->eventfd);
117 
118 	kfree(virqfd);
119 }
120 
121 static void virqfd_inject(struct work_struct *work)
122 {
123 	struct virqfd *virqfd = container_of(work, struct virqfd, inject);
124 	if (virqfd->thread)
125 		virqfd->thread(virqfd->vdev, virqfd->data);
126 }
127 
128 static int virqfd_enable(struct vfio_pci_device *vdev,
129 			 int (*handler)(struct vfio_pci_device *, void *),
130 			 void (*thread)(struct vfio_pci_device *, void *),
131 			 void *data, struct virqfd **pvirqfd, int fd)
132 {
133 	struct fd irqfd;
134 	struct eventfd_ctx *ctx;
135 	struct virqfd *virqfd;
136 	int ret = 0;
137 	unsigned int events;
138 
139 	virqfd = kzalloc(sizeof(*virqfd), GFP_KERNEL);
140 	if (!virqfd)
141 		return -ENOMEM;
142 
143 	virqfd->pvirqfd = pvirqfd;
144 	virqfd->vdev = vdev;
145 	virqfd->handler = handler;
146 	virqfd->thread = thread;
147 	virqfd->data = data;
148 
149 	INIT_WORK(&virqfd->shutdown, virqfd_shutdown);
150 	INIT_WORK(&virqfd->inject, virqfd_inject);
151 
152 	irqfd = fdget(fd);
153 	if (!irqfd.file) {
154 		ret = -EBADF;
155 		goto err_fd;
156 	}
157 
158 	ctx = eventfd_ctx_fileget(irqfd.file);
159 	if (IS_ERR(ctx)) {
160 		ret = PTR_ERR(ctx);
161 		goto err_ctx;
162 	}
163 
164 	virqfd->eventfd = ctx;
165 
166 	/*
167 	 * virqfds can be released by closing the eventfd or directly
168 	 * through ioctl.  These are both done through a workqueue, so
169 	 * we update the pointer to the virqfd under lock to avoid
170 	 * pushing multiple jobs to release the same virqfd.
171 	 */
172 	spin_lock_irq(&vdev->irqlock);
173 
174 	if (*pvirqfd) {
175 		spin_unlock_irq(&vdev->irqlock);
176 		ret = -EBUSY;
177 		goto err_busy;
178 	}
179 	*pvirqfd = virqfd;
180 
181 	spin_unlock_irq(&vdev->irqlock);
182 
183 	/*
184 	 * Install our own custom wake-up handling so we are notified via
185 	 * a callback whenever someone signals the underlying eventfd.
186 	 */
187 	init_waitqueue_func_entry(&virqfd->wait, virqfd_wakeup);
188 	init_poll_funcptr(&virqfd->pt, virqfd_ptable_queue_proc);
189 
190 	events = irqfd.file->f_op->poll(irqfd.file, &virqfd->pt);
191 
192 	/*
193 	 * Check if there was an event already pending on the eventfd
194 	 * before we registered and trigger it as if we didn't miss it.
195 	 */
196 	if (events & POLLIN) {
197 		if ((!handler || handler(vdev, data)) && thread)
198 			schedule_work(&virqfd->inject);
199 	}
200 
201 	/*
202 	 * Do not drop the file until the irqfd is fully initialized,
203 	 * otherwise we might race against the POLLHUP.
204 	 */
205 	fdput(irqfd);
206 
207 	return 0;
208 err_busy:
209 	eventfd_ctx_put(ctx);
210 err_ctx:
211 	fdput(irqfd);
212 err_fd:
213 	kfree(virqfd);
214 
215 	return ret;
216 }
217 
218 static void virqfd_disable(struct vfio_pci_device *vdev,
219 			   struct virqfd **pvirqfd)
220 {
221 	unsigned long flags;
222 
223 	spin_lock_irqsave(&vdev->irqlock, flags);
224 
225 	if (*pvirqfd) {
226 		virqfd_deactivate(*pvirqfd);
227 		*pvirqfd = NULL;
228 	}
229 
230 	spin_unlock_irqrestore(&vdev->irqlock, flags);
231 
232 	/*
233 	 * Block until we know all outstanding shutdown jobs have completed.
234 	 * Even if we don't queue the job, flush the wq to be sure it's
235 	 * been released.
236 	 */
237 	flush_workqueue(vfio_irqfd_cleanup_wq);
238 }
239 
240 /*
241  * INTx
242  */
243 static void vfio_send_intx_eventfd(struct vfio_pci_device *vdev, void *unused)
244 {
245 	if (likely(is_intx(vdev) && !vdev->virq_disabled))
246 		eventfd_signal(vdev->ctx[0].trigger, 1);
247 }
248 
249 void vfio_pci_intx_mask(struct vfio_pci_device *vdev)
250 {
251 	struct pci_dev *pdev = vdev->pdev;
252 	unsigned long flags;
253 
254 	spin_lock_irqsave(&vdev->irqlock, flags);
255 
256 	/*
257 	 * Masking can come from interrupt, ioctl, or config space
258 	 * via INTx disable.  The latter means this can get called
259 	 * even when not using intx delivery.  In this case, just
260 	 * try to have the physical bit follow the virtual bit.
261 	 */
262 	if (unlikely(!is_intx(vdev))) {
263 		if (vdev->pci_2_3)
264 			pci_intx(pdev, 0);
265 	} else if (!vdev->ctx[0].masked) {
266 		/*
267 		 * Can't use check_and_mask here because we always want to
268 		 * mask, not just when something is pending.
269 		 */
270 		if (vdev->pci_2_3)
271 			pci_intx(pdev, 0);
272 		else
273 			disable_irq_nosync(pdev->irq);
274 
275 		vdev->ctx[0].masked = true;
276 	}
277 
278 	spin_unlock_irqrestore(&vdev->irqlock, flags);
279 }
280 
281 /*
282  * If this is triggered by an eventfd, we can't call eventfd_signal
283  * or else we'll deadlock on the eventfd wait queue.  Return >0 when
284  * a signal is necessary, which can then be handled via a work queue
285  * or directly depending on the caller.
286  */
287 static int vfio_pci_intx_unmask_handler(struct vfio_pci_device *vdev,
288 					void *unused)
289 {
290 	struct pci_dev *pdev = vdev->pdev;
291 	unsigned long flags;
292 	int ret = 0;
293 
294 	spin_lock_irqsave(&vdev->irqlock, flags);
295 
296 	/*
297 	 * Unmasking comes from ioctl or config, so again, have the
298 	 * physical bit follow the virtual even when not using INTx.
299 	 */
300 	if (unlikely(!is_intx(vdev))) {
301 		if (vdev->pci_2_3)
302 			pci_intx(pdev, 1);
303 	} else if (vdev->ctx[0].masked && !vdev->virq_disabled) {
304 		/*
305 		 * A pending interrupt here would immediately trigger,
306 		 * but we can avoid that overhead by just re-sending
307 		 * the interrupt to the user.
308 		 */
309 		if (vdev->pci_2_3) {
310 			if (!pci_check_and_unmask_intx(pdev))
311 				ret = 1;
312 		} else
313 			enable_irq(pdev->irq);
314 
315 		vdev->ctx[0].masked = (ret > 0);
316 	}
317 
318 	spin_unlock_irqrestore(&vdev->irqlock, flags);
319 
320 	return ret;
321 }
322 
323 void vfio_pci_intx_unmask(struct vfio_pci_device *vdev)
324 {
325 	if (vfio_pci_intx_unmask_handler(vdev, NULL) > 0)
326 		vfio_send_intx_eventfd(vdev, NULL);
327 }
328 
329 static irqreturn_t vfio_intx_handler(int irq, void *dev_id)
330 {
331 	struct vfio_pci_device *vdev = dev_id;
332 	unsigned long flags;
333 	int ret = IRQ_NONE;
334 
335 	spin_lock_irqsave(&vdev->irqlock, flags);
336 
337 	if (!vdev->pci_2_3) {
338 		disable_irq_nosync(vdev->pdev->irq);
339 		vdev->ctx[0].masked = true;
340 		ret = IRQ_HANDLED;
341 	} else if (!vdev->ctx[0].masked &&  /* may be shared */
342 		   pci_check_and_mask_intx(vdev->pdev)) {
343 		vdev->ctx[0].masked = true;
344 		ret = IRQ_HANDLED;
345 	}
346 
347 	spin_unlock_irqrestore(&vdev->irqlock, flags);
348 
349 	if (ret == IRQ_HANDLED)
350 		vfio_send_intx_eventfd(vdev, NULL);
351 
352 	return ret;
353 }
354 
355 static int vfio_intx_enable(struct vfio_pci_device *vdev)
356 {
357 	if (!is_irq_none(vdev))
358 		return -EINVAL;
359 
360 	if (!vdev->pdev->irq)
361 		return -ENODEV;
362 
363 	vdev->ctx = kzalloc(sizeof(struct vfio_pci_irq_ctx), GFP_KERNEL);
364 	if (!vdev->ctx)
365 		return -ENOMEM;
366 
367 	vdev->num_ctx = 1;
368 
369 	/*
370 	 * If the virtual interrupt is masked, restore it.  Devices
371 	 * supporting DisINTx can be masked at the hardware level
372 	 * here, non-PCI-2.3 devices will have to wait until the
373 	 * interrupt is enabled.
374 	 */
375 	vdev->ctx[0].masked = vdev->virq_disabled;
376 	if (vdev->pci_2_3)
377 		pci_intx(vdev->pdev, !vdev->ctx[0].masked);
378 
379 	vdev->irq_type = VFIO_PCI_INTX_IRQ_INDEX;
380 
381 	return 0;
382 }
383 
384 static int vfio_intx_set_signal(struct vfio_pci_device *vdev, int fd)
385 {
386 	struct pci_dev *pdev = vdev->pdev;
387 	unsigned long irqflags = IRQF_SHARED;
388 	struct eventfd_ctx *trigger;
389 	unsigned long flags;
390 	int ret;
391 
392 	if (vdev->ctx[0].trigger) {
393 		free_irq(pdev->irq, vdev);
394 		kfree(vdev->ctx[0].name);
395 		eventfd_ctx_put(vdev->ctx[0].trigger);
396 		vdev->ctx[0].trigger = NULL;
397 	}
398 
399 	if (fd < 0) /* Disable only */
400 		return 0;
401 
402 	vdev->ctx[0].name = kasprintf(GFP_KERNEL, "vfio-intx(%s)",
403 				      pci_name(pdev));
404 	if (!vdev->ctx[0].name)
405 		return -ENOMEM;
406 
407 	trigger = eventfd_ctx_fdget(fd);
408 	if (IS_ERR(trigger)) {
409 		kfree(vdev->ctx[0].name);
410 		return PTR_ERR(trigger);
411 	}
412 
413 	vdev->ctx[0].trigger = trigger;
414 
415 	if (!vdev->pci_2_3)
416 		irqflags = 0;
417 
418 	ret = request_irq(pdev->irq, vfio_intx_handler,
419 			  irqflags, vdev->ctx[0].name, vdev);
420 	if (ret) {
421 		vdev->ctx[0].trigger = NULL;
422 		kfree(vdev->ctx[0].name);
423 		eventfd_ctx_put(trigger);
424 		return ret;
425 	}
426 
427 	/*
428 	 * INTx disable will stick across the new irq setup,
429 	 * disable_irq won't.
430 	 */
431 	spin_lock_irqsave(&vdev->irqlock, flags);
432 	if (!vdev->pci_2_3 && vdev->ctx[0].masked)
433 		disable_irq_nosync(pdev->irq);
434 	spin_unlock_irqrestore(&vdev->irqlock, flags);
435 
436 	return 0;
437 }
438 
439 static void vfio_intx_disable(struct vfio_pci_device *vdev)
440 {
441 	vfio_intx_set_signal(vdev, -1);
442 	virqfd_disable(vdev, &vdev->ctx[0].unmask);
443 	virqfd_disable(vdev, &vdev->ctx[0].mask);
444 	vdev->irq_type = VFIO_PCI_NUM_IRQS;
445 	vdev->num_ctx = 0;
446 	kfree(vdev->ctx);
447 }
448 
449 /*
450  * MSI/MSI-X
451  */
452 static irqreturn_t vfio_msihandler(int irq, void *arg)
453 {
454 	struct eventfd_ctx *trigger = arg;
455 
456 	eventfd_signal(trigger, 1);
457 	return IRQ_HANDLED;
458 }
459 
460 static int vfio_msi_enable(struct vfio_pci_device *vdev, int nvec, bool msix)
461 {
462 	struct pci_dev *pdev = vdev->pdev;
463 	int ret;
464 
465 	if (!is_irq_none(vdev))
466 		return -EINVAL;
467 
468 	vdev->ctx = kzalloc(nvec * sizeof(struct vfio_pci_irq_ctx), GFP_KERNEL);
469 	if (!vdev->ctx)
470 		return -ENOMEM;
471 
472 	if (msix) {
473 		int i;
474 
475 		vdev->msix = kzalloc(nvec * sizeof(struct msix_entry),
476 				     GFP_KERNEL);
477 		if (!vdev->msix) {
478 			kfree(vdev->ctx);
479 			return -ENOMEM;
480 		}
481 
482 		for (i = 0; i < nvec; i++)
483 			vdev->msix[i].entry = i;
484 
485 		ret = pci_enable_msix_range(pdev, vdev->msix, 1, nvec);
486 		if (ret < nvec) {
487 			if (ret > 0)
488 				pci_disable_msix(pdev);
489 			kfree(vdev->msix);
490 			kfree(vdev->ctx);
491 			return ret;
492 		}
493 	} else {
494 		ret = pci_enable_msi_range(pdev, 1, nvec);
495 		if (ret < nvec) {
496 			if (ret > 0)
497 				pci_disable_msi(pdev);
498 			kfree(vdev->ctx);
499 			return ret;
500 		}
501 	}
502 
503 	vdev->num_ctx = nvec;
504 	vdev->irq_type = msix ? VFIO_PCI_MSIX_IRQ_INDEX :
505 				VFIO_PCI_MSI_IRQ_INDEX;
506 
507 	if (!msix) {
508 		/*
509 		 * Compute the virtual hardware field for max msi vectors -
510 		 * it is the log base 2 of the number of vectors.
511 		 */
512 		vdev->msi_qmax = fls(nvec * 2 - 1) - 1;
513 	}
514 
515 	return 0;
516 }
517 
518 static int vfio_msi_set_vector_signal(struct vfio_pci_device *vdev,
519 				      int vector, int fd, bool msix)
520 {
521 	struct pci_dev *pdev = vdev->pdev;
522 	int irq = msix ? vdev->msix[vector].vector : pdev->irq + vector;
523 	char *name = msix ? "vfio-msix" : "vfio-msi";
524 	struct eventfd_ctx *trigger;
525 	int ret;
526 
527 	if (vector >= vdev->num_ctx)
528 		return -EINVAL;
529 
530 	if (vdev->ctx[vector].trigger) {
531 		free_irq(irq, vdev->ctx[vector].trigger);
532 		kfree(vdev->ctx[vector].name);
533 		eventfd_ctx_put(vdev->ctx[vector].trigger);
534 		vdev->ctx[vector].trigger = NULL;
535 	}
536 
537 	if (fd < 0)
538 		return 0;
539 
540 	vdev->ctx[vector].name = kasprintf(GFP_KERNEL, "%s[%d](%s)",
541 					   name, vector, pci_name(pdev));
542 	if (!vdev->ctx[vector].name)
543 		return -ENOMEM;
544 
545 	trigger = eventfd_ctx_fdget(fd);
546 	if (IS_ERR(trigger)) {
547 		kfree(vdev->ctx[vector].name);
548 		return PTR_ERR(trigger);
549 	}
550 
551 	ret = request_irq(irq, vfio_msihandler, 0,
552 			  vdev->ctx[vector].name, trigger);
553 	if (ret) {
554 		kfree(vdev->ctx[vector].name);
555 		eventfd_ctx_put(trigger);
556 		return ret;
557 	}
558 
559 	vdev->ctx[vector].trigger = trigger;
560 
561 	return 0;
562 }
563 
564 static int vfio_msi_set_block(struct vfio_pci_device *vdev, unsigned start,
565 			      unsigned count, int32_t *fds, bool msix)
566 {
567 	int i, j, ret = 0;
568 
569 	if (start + count > vdev->num_ctx)
570 		return -EINVAL;
571 
572 	for (i = 0, j = start; i < count && !ret; i++, j++) {
573 		int fd = fds ? fds[i] : -1;
574 		ret = vfio_msi_set_vector_signal(vdev, j, fd, msix);
575 	}
576 
577 	if (ret) {
578 		for (--j; j >= start; j--)
579 			vfio_msi_set_vector_signal(vdev, j, -1, msix);
580 	}
581 
582 	return ret;
583 }
584 
585 static void vfio_msi_disable(struct vfio_pci_device *vdev, bool msix)
586 {
587 	struct pci_dev *pdev = vdev->pdev;
588 	int i;
589 
590 	vfio_msi_set_block(vdev, 0, vdev->num_ctx, NULL, msix);
591 
592 	for (i = 0; i < vdev->num_ctx; i++) {
593 		virqfd_disable(vdev, &vdev->ctx[i].unmask);
594 		virqfd_disable(vdev, &vdev->ctx[i].mask);
595 	}
596 
597 	if (msix) {
598 		pci_disable_msix(vdev->pdev);
599 		kfree(vdev->msix);
600 	} else
601 		pci_disable_msi(pdev);
602 
603 	vdev->irq_type = VFIO_PCI_NUM_IRQS;
604 	vdev->num_ctx = 0;
605 	kfree(vdev->ctx);
606 }
607 
608 /*
609  * IOCTL support
610  */
611 static int vfio_pci_set_intx_unmask(struct vfio_pci_device *vdev,
612 				    unsigned index, unsigned start,
613 				    unsigned count, uint32_t flags, void *data)
614 {
615 	if (!is_intx(vdev) || start != 0 || count != 1)
616 		return -EINVAL;
617 
618 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
619 		vfio_pci_intx_unmask(vdev);
620 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
621 		uint8_t unmask = *(uint8_t *)data;
622 		if (unmask)
623 			vfio_pci_intx_unmask(vdev);
624 	} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
625 		int32_t fd = *(int32_t *)data;
626 		if (fd >= 0)
627 			return virqfd_enable(vdev, vfio_pci_intx_unmask_handler,
628 					     vfio_send_intx_eventfd, NULL,
629 					     &vdev->ctx[0].unmask, fd);
630 
631 		virqfd_disable(vdev, &vdev->ctx[0].unmask);
632 	}
633 
634 	return 0;
635 }
636 
637 static int vfio_pci_set_intx_mask(struct vfio_pci_device *vdev,
638 				  unsigned index, unsigned start,
639 				  unsigned count, uint32_t flags, void *data)
640 {
641 	if (!is_intx(vdev) || start != 0 || count != 1)
642 		return -EINVAL;
643 
644 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
645 		vfio_pci_intx_mask(vdev);
646 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
647 		uint8_t mask = *(uint8_t *)data;
648 		if (mask)
649 			vfio_pci_intx_mask(vdev);
650 	} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
651 		return -ENOTTY; /* XXX implement me */
652 	}
653 
654 	return 0;
655 }
656 
657 static int vfio_pci_set_intx_trigger(struct vfio_pci_device *vdev,
658 				     unsigned index, unsigned start,
659 				     unsigned count, uint32_t flags, void *data)
660 {
661 	if (is_intx(vdev) && !count && (flags & VFIO_IRQ_SET_DATA_NONE)) {
662 		vfio_intx_disable(vdev);
663 		return 0;
664 	}
665 
666 	if (!(is_intx(vdev) || is_irq_none(vdev)) || start != 0 || count != 1)
667 		return -EINVAL;
668 
669 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
670 		int32_t fd = *(int32_t *)data;
671 		int ret;
672 
673 		if (is_intx(vdev))
674 			return vfio_intx_set_signal(vdev, fd);
675 
676 		ret = vfio_intx_enable(vdev);
677 		if (ret)
678 			return ret;
679 
680 		ret = vfio_intx_set_signal(vdev, fd);
681 		if (ret)
682 			vfio_intx_disable(vdev);
683 
684 		return ret;
685 	}
686 
687 	if (!is_intx(vdev))
688 		return -EINVAL;
689 
690 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
691 		vfio_send_intx_eventfd(vdev, NULL);
692 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
693 		uint8_t trigger = *(uint8_t *)data;
694 		if (trigger)
695 			vfio_send_intx_eventfd(vdev, NULL);
696 	}
697 	return 0;
698 }
699 
700 static int vfio_pci_set_msi_trigger(struct vfio_pci_device *vdev,
701 				    unsigned index, unsigned start,
702 				    unsigned count, uint32_t flags, void *data)
703 {
704 	int i;
705 	bool msix = (index == VFIO_PCI_MSIX_IRQ_INDEX) ? true : false;
706 
707 	if (irq_is(vdev, index) && !count && (flags & VFIO_IRQ_SET_DATA_NONE)) {
708 		vfio_msi_disable(vdev, msix);
709 		return 0;
710 	}
711 
712 	if (!(irq_is(vdev, index) || is_irq_none(vdev)))
713 		return -EINVAL;
714 
715 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
716 		int32_t *fds = data;
717 		int ret;
718 
719 		if (vdev->irq_type == index)
720 			return vfio_msi_set_block(vdev, start, count,
721 						  fds, msix);
722 
723 		ret = vfio_msi_enable(vdev, start + count, msix);
724 		if (ret)
725 			return ret;
726 
727 		ret = vfio_msi_set_block(vdev, start, count, fds, msix);
728 		if (ret)
729 			vfio_msi_disable(vdev, msix);
730 
731 		return ret;
732 	}
733 
734 	if (!irq_is(vdev, index) || start + count > vdev->num_ctx)
735 		return -EINVAL;
736 
737 	for (i = start; i < start + count; i++) {
738 		if (!vdev->ctx[i].trigger)
739 			continue;
740 		if (flags & VFIO_IRQ_SET_DATA_NONE) {
741 			eventfd_signal(vdev->ctx[i].trigger, 1);
742 		} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
743 			uint8_t *bools = data;
744 			if (bools[i - start])
745 				eventfd_signal(vdev->ctx[i].trigger, 1);
746 		}
747 	}
748 	return 0;
749 }
750 
751 static int vfio_pci_set_err_trigger(struct vfio_pci_device *vdev,
752 				    unsigned index, unsigned start,
753 				    unsigned count, uint32_t flags, void *data)
754 {
755 	int32_t fd = *(int32_t *)data;
756 
757 	if ((index != VFIO_PCI_ERR_IRQ_INDEX) ||
758 	    !(flags & VFIO_IRQ_SET_DATA_TYPE_MASK))
759 		return -EINVAL;
760 
761 	/* DATA_NONE/DATA_BOOL enables loopback testing */
762 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
763 		if (vdev->err_trigger)
764 			eventfd_signal(vdev->err_trigger, 1);
765 		return 0;
766 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
767 		uint8_t trigger = *(uint8_t *)data;
768 		if (trigger && vdev->err_trigger)
769 			eventfd_signal(vdev->err_trigger, 1);
770 		return 0;
771 	}
772 
773 	/* Handle SET_DATA_EVENTFD */
774 	if (fd == -1) {
775 		if (vdev->err_trigger)
776 			eventfd_ctx_put(vdev->err_trigger);
777 		vdev->err_trigger = NULL;
778 		return 0;
779 	} else if (fd >= 0) {
780 		struct eventfd_ctx *efdctx;
781 		efdctx = eventfd_ctx_fdget(fd);
782 		if (IS_ERR(efdctx))
783 			return PTR_ERR(efdctx);
784 		if (vdev->err_trigger)
785 			eventfd_ctx_put(vdev->err_trigger);
786 		vdev->err_trigger = efdctx;
787 		return 0;
788 	} else
789 		return -EINVAL;
790 }
791 int vfio_pci_set_irqs_ioctl(struct vfio_pci_device *vdev, uint32_t flags,
792 			    unsigned index, unsigned start, unsigned count,
793 			    void *data)
794 {
795 	int (*func)(struct vfio_pci_device *vdev, unsigned index,
796 		    unsigned start, unsigned count, uint32_t flags,
797 		    void *data) = NULL;
798 
799 	switch (index) {
800 	case VFIO_PCI_INTX_IRQ_INDEX:
801 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
802 		case VFIO_IRQ_SET_ACTION_MASK:
803 			func = vfio_pci_set_intx_mask;
804 			break;
805 		case VFIO_IRQ_SET_ACTION_UNMASK:
806 			func = vfio_pci_set_intx_unmask;
807 			break;
808 		case VFIO_IRQ_SET_ACTION_TRIGGER:
809 			func = vfio_pci_set_intx_trigger;
810 			break;
811 		}
812 		break;
813 	case VFIO_PCI_MSI_IRQ_INDEX:
814 	case VFIO_PCI_MSIX_IRQ_INDEX:
815 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
816 		case VFIO_IRQ_SET_ACTION_MASK:
817 		case VFIO_IRQ_SET_ACTION_UNMASK:
818 			/* XXX Need masking support exported */
819 			break;
820 		case VFIO_IRQ_SET_ACTION_TRIGGER:
821 			func = vfio_pci_set_msi_trigger;
822 			break;
823 		}
824 		break;
825 	case VFIO_PCI_ERR_IRQ_INDEX:
826 		switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
827 		case VFIO_IRQ_SET_ACTION_TRIGGER:
828 			if (pci_is_pcie(vdev->pdev))
829 				func = vfio_pci_set_err_trigger;
830 			break;
831 		}
832 	}
833 
834 	if (!func)
835 		return -ENOTTY;
836 
837 	return func(vdev, index, start, count, flags, data);
838 }
839