xref: /openbmc/linux/arch/um/drivers/virt-pci.c (revision 806b5228)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2020 Intel Corporation
4  * Author: Johannes Berg <johannes@sipsolutions.net>
5  */
6 #include <linux/module.h>
7 #include <linux/pci.h>
8 #include <linux/virtio.h>
9 #include <linux/virtio_config.h>
10 #include <linux/logic_iomem.h>
11 #include <linux/irqdomain.h>
12 #include <linux/virtio_pcidev.h>
13 #include <linux/virtio-uml.h>
14 #include <linux/delay.h>
15 #include <linux/msi.h>
16 #include <asm/unaligned.h>
17 #include <irq_kern.h>
18 
19 #define MAX_DEVICES 8
20 #define MAX_MSI_VECTORS 32
21 #define CFG_SPACE_SIZE 4096
22 
23 /* for MSI-X we have a 32-bit payload */
24 #define MAX_IRQ_MSG_SIZE (sizeof(struct virtio_pcidev_msg) + sizeof(u32))
25 #define NUM_IRQ_MSGS	10
26 
27 #define HANDLE_NO_FREE(ptr) ((void *)((unsigned long)(ptr) | 1))
28 #define HANDLE_IS_NO_FREE(ptr) ((unsigned long)(ptr) & 1)
29 
30 struct um_pci_device {
31 	struct virtio_device *vdev;
32 
33 	/* for now just standard BARs */
34 	u8 resptr[PCI_STD_NUM_BARS];
35 
36 	struct virtqueue *cmd_vq, *irq_vq;
37 
38 #define UM_PCI_STAT_WAITING	0
39 	unsigned long status;
40 
41 	int irq;
42 };
43 
44 struct um_pci_device_reg {
45 	struct um_pci_device *dev;
46 	void __iomem *iomem;
47 };
48 
49 static struct pci_host_bridge *bridge;
50 static DEFINE_MUTEX(um_pci_mtx);
51 static struct um_pci_device_reg um_pci_devices[MAX_DEVICES];
52 static struct fwnode_handle *um_pci_fwnode;
53 static struct irq_domain *um_pci_inner_domain;
54 static struct irq_domain *um_pci_msi_domain;
55 static unsigned long um_pci_msi_used[BITS_TO_LONGS(MAX_MSI_VECTORS)];
56 
57 #define UM_VIRT_PCI_MAXDELAY 40000
58 
59 struct um_pci_message_buffer {
60 	struct virtio_pcidev_msg hdr;
61 	u8 data[8];
62 };
63 
64 static struct um_pci_message_buffer __percpu *um_pci_msg_bufs;
65 
66 static int um_pci_send_cmd(struct um_pci_device *dev,
67 			   struct virtio_pcidev_msg *cmd,
68 			   unsigned int cmd_size,
69 			   const void *extra, unsigned int extra_size,
70 			   void *out, unsigned int out_size)
71 {
72 	struct scatterlist out_sg, extra_sg, in_sg;
73 	struct scatterlist *sgs_list[] = {
74 		[0] = &out_sg,
75 		[1] = extra ? &extra_sg : &in_sg,
76 		[2] = extra ? &in_sg : NULL,
77 	};
78 	struct um_pci_message_buffer *buf;
79 	int delay_count = 0;
80 	int ret, len;
81 	bool posted;
82 
83 	if (WARN_ON(cmd_size < sizeof(*cmd) || cmd_size > sizeof(*buf)))
84 		return -EINVAL;
85 
86 	switch (cmd->op) {
87 	case VIRTIO_PCIDEV_OP_CFG_WRITE:
88 	case VIRTIO_PCIDEV_OP_MMIO_WRITE:
89 	case VIRTIO_PCIDEV_OP_MMIO_MEMSET:
90 		/* in PCI, writes are posted, so don't wait */
91 		posted = !out;
92 		WARN_ON(!posted);
93 		break;
94 	default:
95 		posted = false;
96 		break;
97 	}
98 
99 	buf = get_cpu_var(um_pci_msg_bufs);
100 	memcpy(buf, cmd, cmd_size);
101 
102 	if (posted) {
103 		u8 *ncmd = kmalloc(cmd_size + extra_size, GFP_ATOMIC);
104 
105 		if (ncmd) {
106 			memcpy(ncmd, cmd, cmd_size);
107 			if (extra)
108 				memcpy(ncmd + cmd_size, extra, extra_size);
109 			cmd = (void *)ncmd;
110 			cmd_size += extra_size;
111 			extra = NULL;
112 			extra_size = 0;
113 		} else {
114 			/* try without allocating memory */
115 			posted = false;
116 			cmd = (void *)buf;
117 		}
118 	} else {
119 		cmd = (void *)buf;
120 	}
121 
122 	sg_init_one(&out_sg, cmd, cmd_size);
123 	if (extra)
124 		sg_init_one(&extra_sg, extra, extra_size);
125 	if (out)
126 		sg_init_one(&in_sg, out, out_size);
127 
128 	/* add to internal virtio queue */
129 	ret = virtqueue_add_sgs(dev->cmd_vq, sgs_list,
130 				extra ? 2 : 1,
131 				out ? 1 : 0,
132 				posted ? cmd : HANDLE_NO_FREE(cmd),
133 				GFP_ATOMIC);
134 	if (ret)
135 		goto out;
136 
137 	if (posted) {
138 		virtqueue_kick(dev->cmd_vq);
139 		ret = 0;
140 		goto out;
141 	}
142 
143 	/* kick and poll for getting a response on the queue */
144 	set_bit(UM_PCI_STAT_WAITING, &dev->status);
145 	virtqueue_kick(dev->cmd_vq);
146 
147 	while (1) {
148 		void *completed = virtqueue_get_buf(dev->cmd_vq, &len);
149 
150 		if (completed == HANDLE_NO_FREE(cmd))
151 			break;
152 
153 		if (completed && !HANDLE_IS_NO_FREE(completed))
154 			kfree(completed);
155 
156 		if (WARN_ONCE(virtqueue_is_broken(dev->cmd_vq) ||
157 			      ++delay_count > UM_VIRT_PCI_MAXDELAY,
158 			      "um virt-pci delay: %d", delay_count)) {
159 			ret = -EIO;
160 			break;
161 		}
162 		udelay(1);
163 	}
164 	clear_bit(UM_PCI_STAT_WAITING, &dev->status);
165 
166 out:
167 	put_cpu_var(um_pci_msg_bufs);
168 	return ret;
169 }
170 
171 static unsigned long um_pci_cfgspace_read(void *priv, unsigned int offset,
172 					  int size)
173 {
174 	struct um_pci_device_reg *reg = priv;
175 	struct um_pci_device *dev = reg->dev;
176 	struct virtio_pcidev_msg hdr = {
177 		.op = VIRTIO_PCIDEV_OP_CFG_READ,
178 		.size = size,
179 		.addr = offset,
180 	};
181 	/* buf->data is maximum size - we may only use parts of it */
182 	struct um_pci_message_buffer *buf;
183 	u8 *data;
184 	unsigned long ret = ULONG_MAX;
185 
186 	if (!dev)
187 		return ULONG_MAX;
188 
189 	buf = get_cpu_var(um_pci_msg_bufs);
190 	data = buf->data;
191 
192 	memset(buf->data, 0xff, sizeof(buf->data));
193 
194 	switch (size) {
195 	case 1:
196 	case 2:
197 	case 4:
198 #ifdef CONFIG_64BIT
199 	case 8:
200 #endif
201 		break;
202 	default:
203 		WARN(1, "invalid config space read size %d\n", size);
204 		goto out;
205 	}
206 
207 	if (um_pci_send_cmd(dev, &hdr, sizeof(hdr), NULL, 0, data, 8))
208 		goto out;
209 
210 	switch (size) {
211 	case 1:
212 		ret = data[0];
213 		break;
214 	case 2:
215 		ret = le16_to_cpup((void *)data);
216 		break;
217 	case 4:
218 		ret = le32_to_cpup((void *)data);
219 		break;
220 #ifdef CONFIG_64BIT
221 	case 8:
222 		ret = le64_to_cpup((void *)data);
223 		break;
224 #endif
225 	default:
226 		break;
227 	}
228 
229 out:
230 	put_cpu_var(um_pci_msg_bufs);
231 	return ret;
232 }
233 
234 static void um_pci_cfgspace_write(void *priv, unsigned int offset, int size,
235 				  unsigned long val)
236 {
237 	struct um_pci_device_reg *reg = priv;
238 	struct um_pci_device *dev = reg->dev;
239 	struct {
240 		struct virtio_pcidev_msg hdr;
241 		/* maximum size - we may only use parts of it */
242 		u8 data[8];
243 	} msg = {
244 		.hdr = {
245 			.op = VIRTIO_PCIDEV_OP_CFG_WRITE,
246 			.size = size,
247 			.addr = offset,
248 		},
249 	};
250 
251 	if (!dev)
252 		return;
253 
254 	switch (size) {
255 	case 1:
256 		msg.data[0] = (u8)val;
257 		break;
258 	case 2:
259 		put_unaligned_le16(val, (void *)msg.data);
260 		break;
261 	case 4:
262 		put_unaligned_le32(val, (void *)msg.data);
263 		break;
264 #ifdef CONFIG_64BIT
265 	case 8:
266 		put_unaligned_le64(val, (void *)msg.data);
267 		break;
268 #endif
269 	default:
270 		WARN(1, "invalid config space write size %d\n", size);
271 		return;
272 	}
273 
274 	WARN_ON(um_pci_send_cmd(dev, &msg.hdr, sizeof(msg), NULL, 0, NULL, 0));
275 }
276 
277 static const struct logic_iomem_ops um_pci_device_cfgspace_ops = {
278 	.read = um_pci_cfgspace_read,
279 	.write = um_pci_cfgspace_write,
280 };
281 
282 static void um_pci_bar_copy_from(void *priv, void *buffer,
283 				 unsigned int offset, int size)
284 {
285 	u8 *resptr = priv;
286 	struct um_pci_device *dev = container_of(resptr - *resptr,
287 						 struct um_pci_device,
288 						 resptr[0]);
289 	struct virtio_pcidev_msg hdr = {
290 		.op = VIRTIO_PCIDEV_OP_MMIO_READ,
291 		.bar = *resptr,
292 		.size = size,
293 		.addr = offset,
294 	};
295 
296 	memset(buffer, 0xff, size);
297 
298 	um_pci_send_cmd(dev, &hdr, sizeof(hdr), NULL, 0, buffer, size);
299 }
300 
301 static unsigned long um_pci_bar_read(void *priv, unsigned int offset,
302 				     int size)
303 {
304 	/* buf->data is maximum size - we may only use parts of it */
305 	struct um_pci_message_buffer *buf;
306 	u8 *data;
307 	unsigned long ret = ULONG_MAX;
308 
309 	buf = get_cpu_var(um_pci_msg_bufs);
310 	data = buf->data;
311 
312 	switch (size) {
313 	case 1:
314 	case 2:
315 	case 4:
316 #ifdef CONFIG_64BIT
317 	case 8:
318 #endif
319 		break;
320 	default:
321 		WARN(1, "invalid config space read size %d\n", size);
322 		goto out;
323 	}
324 
325 	um_pci_bar_copy_from(priv, data, offset, size);
326 
327 	switch (size) {
328 	case 1:
329 		ret = data[0];
330 		break;
331 	case 2:
332 		ret = le16_to_cpup((void *)data);
333 		break;
334 	case 4:
335 		ret = le32_to_cpup((void *)data);
336 		break;
337 #ifdef CONFIG_64BIT
338 	case 8:
339 		ret = le64_to_cpup((void *)data);
340 		break;
341 #endif
342 	default:
343 		break;
344 	}
345 
346 out:
347 	put_cpu_var(um_pci_msg_bufs);
348 	return ret;
349 }
350 
351 static void um_pci_bar_copy_to(void *priv, unsigned int offset,
352 			       const void *buffer, int size)
353 {
354 	u8 *resptr = priv;
355 	struct um_pci_device *dev = container_of(resptr - *resptr,
356 						 struct um_pci_device,
357 						 resptr[0]);
358 	struct virtio_pcidev_msg hdr = {
359 		.op = VIRTIO_PCIDEV_OP_MMIO_WRITE,
360 		.bar = *resptr,
361 		.size = size,
362 		.addr = offset,
363 	};
364 
365 	um_pci_send_cmd(dev, &hdr, sizeof(hdr), buffer, size, NULL, 0);
366 }
367 
368 static void um_pci_bar_write(void *priv, unsigned int offset, int size,
369 			     unsigned long val)
370 {
371 	/* maximum size - we may only use parts of it */
372 	u8 data[8];
373 
374 	switch (size) {
375 	case 1:
376 		data[0] = (u8)val;
377 		break;
378 	case 2:
379 		put_unaligned_le16(val, (void *)data);
380 		break;
381 	case 4:
382 		put_unaligned_le32(val, (void *)data);
383 		break;
384 #ifdef CONFIG_64BIT
385 	case 8:
386 		put_unaligned_le64(val, (void *)data);
387 		break;
388 #endif
389 	default:
390 		WARN(1, "invalid config space write size %d\n", size);
391 		return;
392 	}
393 
394 	um_pci_bar_copy_to(priv, offset, data, size);
395 }
396 
397 static void um_pci_bar_set(void *priv, unsigned int offset, u8 value, int size)
398 {
399 	u8 *resptr = priv;
400 	struct um_pci_device *dev = container_of(resptr - *resptr,
401 						 struct um_pci_device,
402 						 resptr[0]);
403 	struct {
404 		struct virtio_pcidev_msg hdr;
405 		u8 data;
406 	} msg = {
407 		.hdr = {
408 			.op = VIRTIO_PCIDEV_OP_CFG_WRITE,
409 			.bar = *resptr,
410 			.size = size,
411 			.addr = offset,
412 		},
413 		.data = value,
414 	};
415 
416 	um_pci_send_cmd(dev, &msg.hdr, sizeof(msg), NULL, 0, NULL, 0);
417 }
418 
419 static const struct logic_iomem_ops um_pci_device_bar_ops = {
420 	.read = um_pci_bar_read,
421 	.write = um_pci_bar_write,
422 	.set = um_pci_bar_set,
423 	.copy_from = um_pci_bar_copy_from,
424 	.copy_to = um_pci_bar_copy_to,
425 };
426 
427 static void __iomem *um_pci_map_bus(struct pci_bus *bus, unsigned int devfn,
428 				    int where)
429 {
430 	struct um_pci_device_reg *dev;
431 	unsigned int busn = bus->number;
432 
433 	if (busn > 0)
434 		return NULL;
435 
436 	/* not allowing functions for now ... */
437 	if (devfn % 8)
438 		return NULL;
439 
440 	if (devfn / 8 >= ARRAY_SIZE(um_pci_devices))
441 		return NULL;
442 
443 	dev = &um_pci_devices[devfn / 8];
444 	if (!dev)
445 		return NULL;
446 
447 	return (void __iomem *)((unsigned long)dev->iomem + where);
448 }
449 
450 static struct pci_ops um_pci_ops = {
451 	.map_bus = um_pci_map_bus,
452 	.read = pci_generic_config_read,
453 	.write = pci_generic_config_write,
454 };
455 
456 static void um_pci_rescan(void)
457 {
458 	pci_lock_rescan_remove();
459 	pci_rescan_bus(bridge->bus);
460 	pci_unlock_rescan_remove();
461 }
462 
463 static void um_pci_irq_vq_addbuf(struct virtqueue *vq, void *buf, bool kick)
464 {
465 	struct scatterlist sg[1];
466 
467 	sg_init_one(sg, buf, MAX_IRQ_MSG_SIZE);
468 	if (virtqueue_add_inbuf(vq, sg, 1, buf, GFP_ATOMIC))
469 		kfree(buf);
470 	else if (kick)
471 		virtqueue_kick(vq);
472 }
473 
474 static void um_pci_handle_irq_message(struct virtqueue *vq,
475 				      struct virtio_pcidev_msg *msg)
476 {
477 	struct virtio_device *vdev = vq->vdev;
478 	struct um_pci_device *dev = vdev->priv;
479 
480 	/* we should properly chain interrupts, but on ARCH=um we don't care */
481 
482 	switch (msg->op) {
483 	case VIRTIO_PCIDEV_OP_INT:
484 		generic_handle_irq(dev->irq);
485 		break;
486 	case VIRTIO_PCIDEV_OP_MSI:
487 		/* our MSI message is just the interrupt number */
488 		if (msg->size == sizeof(u32))
489 			generic_handle_irq(le32_to_cpup((void *)msg->data));
490 		else
491 			generic_handle_irq(le16_to_cpup((void *)msg->data));
492 		break;
493 	case VIRTIO_PCIDEV_OP_PME:
494 		/* nothing to do - we already woke up due to the message */
495 		break;
496 	default:
497 		dev_err(&vdev->dev, "unexpected virt-pci message %d\n", msg->op);
498 		break;
499 	}
500 }
501 
502 static void um_pci_cmd_vq_cb(struct virtqueue *vq)
503 {
504 	struct virtio_device *vdev = vq->vdev;
505 	struct um_pci_device *dev = vdev->priv;
506 	void *cmd;
507 	int len;
508 
509 	if (test_bit(UM_PCI_STAT_WAITING, &dev->status))
510 		return;
511 
512 	while ((cmd = virtqueue_get_buf(vq, &len))) {
513 		if (WARN_ON(HANDLE_IS_NO_FREE(cmd)))
514 			continue;
515 		kfree(cmd);
516 	}
517 }
518 
519 static void um_pci_irq_vq_cb(struct virtqueue *vq)
520 {
521 	struct virtio_pcidev_msg *msg;
522 	int len;
523 
524 	while ((msg = virtqueue_get_buf(vq, &len))) {
525 		if (len >= sizeof(*msg))
526 			um_pci_handle_irq_message(vq, msg);
527 
528 		/* recycle the message buffer */
529 		um_pci_irq_vq_addbuf(vq, msg, true);
530 	}
531 }
532 
533 static int um_pci_init_vqs(struct um_pci_device *dev)
534 {
535 	struct virtqueue *vqs[2];
536 	static const char *const names[2] = { "cmd", "irq" };
537 	vq_callback_t *cbs[2] = { um_pci_cmd_vq_cb, um_pci_irq_vq_cb };
538 	int err, i;
539 
540 	err = virtio_find_vqs(dev->vdev, 2, vqs, cbs, names, NULL);
541 	if (err)
542 		return err;
543 
544 	dev->cmd_vq = vqs[0];
545 	dev->irq_vq = vqs[1];
546 
547 	for (i = 0; i < NUM_IRQ_MSGS; i++) {
548 		void *msg = kzalloc(MAX_IRQ_MSG_SIZE, GFP_KERNEL);
549 
550 		if (msg)
551 			um_pci_irq_vq_addbuf(dev->irq_vq, msg, false);
552 	}
553 
554 	virtqueue_kick(dev->irq_vq);
555 
556 	return 0;
557 }
558 
559 static int um_pci_virtio_probe(struct virtio_device *vdev)
560 {
561 	struct um_pci_device *dev;
562 	int i, free = -1;
563 	int err = -ENOSPC;
564 
565 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
566 	if (!dev)
567 		return -ENOMEM;
568 
569 	dev->vdev = vdev;
570 	vdev->priv = dev;
571 
572 	mutex_lock(&um_pci_mtx);
573 	for (i = 0; i < MAX_DEVICES; i++) {
574 		if (um_pci_devices[i].dev)
575 			continue;
576 		free = i;
577 		break;
578 	}
579 
580 	if (free < 0)
581 		goto error;
582 
583 	err = um_pci_init_vqs(dev);
584 	if (err)
585 		goto error;
586 
587 	dev->irq = irq_alloc_desc(numa_node_id());
588 	if (dev->irq < 0) {
589 		err = dev->irq;
590 		goto error;
591 	}
592 	um_pci_devices[free].dev = dev;
593 	vdev->priv = dev;
594 
595 	mutex_unlock(&um_pci_mtx);
596 
597 	device_set_wakeup_enable(&vdev->dev, true);
598 
599 	/*
600 	 * In order to do suspend-resume properly, don't allow VQs
601 	 * to be suspended.
602 	 */
603 	virtio_uml_set_no_vq_suspend(vdev, true);
604 
605 	um_pci_rescan();
606 	return 0;
607 error:
608 	mutex_unlock(&um_pci_mtx);
609 	kfree(dev);
610 	return err;
611 }
612 
613 static void um_pci_virtio_remove(struct virtio_device *vdev)
614 {
615 	struct um_pci_device *dev = vdev->priv;
616 	int i;
617 
618         /* Stop all virtqueues */
619         virtio_reset_device(vdev);
620         vdev->config->del_vqs(vdev);
621 
622 	device_set_wakeup_enable(&vdev->dev, false);
623 
624 	mutex_lock(&um_pci_mtx);
625 	for (i = 0; i < MAX_DEVICES; i++) {
626 		if (um_pci_devices[i].dev != dev)
627 			continue;
628 		um_pci_devices[i].dev = NULL;
629 		irq_free_desc(dev->irq);
630 	}
631 	mutex_unlock(&um_pci_mtx);
632 
633 	um_pci_rescan();
634 
635 	kfree(dev);
636 }
637 
638 static struct virtio_device_id id_table[] = {
639 	{ CONFIG_UML_PCI_OVER_VIRTIO_DEVICE_ID, VIRTIO_DEV_ANY_ID },
640 	{ 0 },
641 };
642 MODULE_DEVICE_TABLE(virtio, id_table);
643 
644 static struct virtio_driver um_pci_virtio_driver = {
645 	.driver.name = "virtio-pci",
646 	.driver.owner = THIS_MODULE,
647 	.id_table = id_table,
648 	.probe = um_pci_virtio_probe,
649 	.remove = um_pci_virtio_remove,
650 };
651 
652 static struct resource virt_cfgspace_resource = {
653 	.name = "PCI config space",
654 	.start = 0xf0000000 - MAX_DEVICES * CFG_SPACE_SIZE,
655 	.end = 0xf0000000 - 1,
656 	.flags = IORESOURCE_MEM,
657 };
658 
659 static long um_pci_map_cfgspace(unsigned long offset, size_t size,
660 				const struct logic_iomem_ops **ops,
661 				void **priv)
662 {
663 	if (WARN_ON(size > CFG_SPACE_SIZE || offset % CFG_SPACE_SIZE))
664 		return -EINVAL;
665 
666 	if (offset / CFG_SPACE_SIZE < MAX_DEVICES) {
667 		*ops = &um_pci_device_cfgspace_ops;
668 		*priv = &um_pci_devices[offset / CFG_SPACE_SIZE];
669 		return 0;
670 	}
671 
672 	WARN(1, "cannot map offset 0x%lx/0x%zx\n", offset, size);
673 	return -ENOENT;
674 }
675 
676 static const struct logic_iomem_region_ops um_pci_cfgspace_ops = {
677 	.map = um_pci_map_cfgspace,
678 };
679 
680 static struct resource virt_iomem_resource = {
681 	.name = "PCI iomem",
682 	.start = 0xf0000000,
683 	.end = 0xffffffff,
684 	.flags = IORESOURCE_MEM,
685 };
686 
687 struct um_pci_map_iomem_data {
688 	unsigned long offset;
689 	size_t size;
690 	const struct logic_iomem_ops **ops;
691 	void **priv;
692 	long ret;
693 };
694 
695 static int um_pci_map_iomem_walk(struct pci_dev *pdev, void *_data)
696 {
697 	struct um_pci_map_iomem_data *data = _data;
698 	struct um_pci_device_reg *reg = &um_pci_devices[pdev->devfn / 8];
699 	struct um_pci_device *dev;
700 	int i;
701 
702 	if (!reg->dev)
703 		return 0;
704 
705 	for (i = 0; i < ARRAY_SIZE(dev->resptr); i++) {
706 		struct resource *r = &pdev->resource[i];
707 
708 		if ((r->flags & IORESOURCE_TYPE_BITS) != IORESOURCE_MEM)
709 			continue;
710 
711 		/*
712 		 * must be the whole or part of the resource,
713 		 * not allowed to only overlap
714 		 */
715 		if (data->offset < r->start || data->offset > r->end)
716 			continue;
717 		if (data->offset + data->size - 1 > r->end)
718 			continue;
719 
720 		dev = reg->dev;
721 		*data->ops = &um_pci_device_bar_ops;
722 		dev->resptr[i] = i;
723 		*data->priv = &dev->resptr[i];
724 		data->ret = data->offset - r->start;
725 
726 		/* no need to continue */
727 		return 1;
728 	}
729 
730 	return 0;
731 }
732 
733 static long um_pci_map_iomem(unsigned long offset, size_t size,
734 			     const struct logic_iomem_ops **ops,
735 			     void **priv)
736 {
737 	struct um_pci_map_iomem_data data = {
738 		/* we want the full address here */
739 		.offset = offset + virt_iomem_resource.start,
740 		.size = size,
741 		.ops = ops,
742 		.priv = priv,
743 		.ret = -ENOENT,
744 	};
745 
746 	pci_walk_bus(bridge->bus, um_pci_map_iomem_walk, &data);
747 	return data.ret;
748 }
749 
750 static const struct logic_iomem_region_ops um_pci_iomem_ops = {
751 	.map = um_pci_map_iomem,
752 };
753 
754 static void um_pci_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
755 {
756 	/*
757 	 * This is a very low address and not actually valid 'physical' memory
758 	 * in UML, so we can simply map MSI(-X) vectors to there, it cannot be
759 	 * legitimately written to by the device in any other way.
760 	 * We use the (virtual) IRQ number here as the message to simplify the
761 	 * code that receives the message, where for now we simply trust the
762 	 * device to send the correct message.
763 	 */
764 	msg->address_hi = 0;
765 	msg->address_lo = 0xa0000;
766 	msg->data = data->irq;
767 }
768 
769 static struct irq_chip um_pci_msi_bottom_irq_chip = {
770 	.name = "UM virtio MSI",
771 	.irq_compose_msi_msg = um_pci_compose_msi_msg,
772 };
773 
774 static int um_pci_inner_domain_alloc(struct irq_domain *domain,
775 				     unsigned int virq, unsigned int nr_irqs,
776 				     void *args)
777 {
778 	unsigned long bit;
779 
780 	WARN_ON(nr_irqs != 1);
781 
782 	mutex_lock(&um_pci_mtx);
783 	bit = find_first_zero_bit(um_pci_msi_used, MAX_MSI_VECTORS);
784 	if (bit >= MAX_MSI_VECTORS) {
785 		mutex_unlock(&um_pci_mtx);
786 		return -ENOSPC;
787 	}
788 
789 	set_bit(bit, um_pci_msi_used);
790 	mutex_unlock(&um_pci_mtx);
791 
792 	irq_domain_set_info(domain, virq, bit, &um_pci_msi_bottom_irq_chip,
793 			    domain->host_data, handle_simple_irq,
794 			    NULL, NULL);
795 
796 	return 0;
797 }
798 
799 static void um_pci_inner_domain_free(struct irq_domain *domain,
800 				     unsigned int virq, unsigned int nr_irqs)
801 {
802 	struct irq_data *d = irq_domain_get_irq_data(domain, virq);
803 
804 	mutex_lock(&um_pci_mtx);
805 
806 	if (!test_bit(d->hwirq, um_pci_msi_used))
807 		pr_err("trying to free unused MSI#%lu\n", d->hwirq);
808 	else
809 		__clear_bit(d->hwirq, um_pci_msi_used);
810 
811 	mutex_unlock(&um_pci_mtx);
812 }
813 
814 static const struct irq_domain_ops um_pci_inner_domain_ops = {
815 	.alloc = um_pci_inner_domain_alloc,
816 	.free = um_pci_inner_domain_free,
817 };
818 
819 static struct irq_chip um_pci_msi_irq_chip = {
820 	.name = "UM virtio PCIe MSI",
821 	.irq_mask = pci_msi_mask_irq,
822 	.irq_unmask = pci_msi_unmask_irq,
823 };
824 
825 static struct msi_domain_info um_pci_msi_domain_info = {
826 	.flags	= MSI_FLAG_USE_DEF_DOM_OPS |
827 		  MSI_FLAG_USE_DEF_CHIP_OPS |
828 		  MSI_FLAG_PCI_MSIX,
829 	.chip	= &um_pci_msi_irq_chip,
830 };
831 
832 static struct resource busn_resource = {
833 	.name	= "PCI busn",
834 	.start	= 0,
835 	.end	= 0,
836 	.flags	= IORESOURCE_BUS,
837 };
838 
839 static int um_pci_map_irq(const struct pci_dev *pdev, u8 slot, u8 pin)
840 {
841 	struct um_pci_device_reg *reg = &um_pci_devices[pdev->devfn / 8];
842 
843 	if (WARN_ON(!reg->dev))
844 		return -EINVAL;
845 
846 	/* Yes, we map all pins to the same IRQ ... doesn't matter for now. */
847 	return reg->dev->irq;
848 }
849 
850 void *pci_root_bus_fwnode(struct pci_bus *bus)
851 {
852 	return um_pci_fwnode;
853 }
854 
855 static int um_pci_init(void)
856 {
857 	int err, i;
858 
859 	WARN_ON(logic_iomem_add_region(&virt_cfgspace_resource,
860 				       &um_pci_cfgspace_ops));
861 	WARN_ON(logic_iomem_add_region(&virt_iomem_resource,
862 				       &um_pci_iomem_ops));
863 
864 	if (WARN(CONFIG_UML_PCI_OVER_VIRTIO_DEVICE_ID < 0,
865 		 "No virtio device ID configured for PCI - no PCI support\n"))
866 		return 0;
867 
868 	um_pci_msg_bufs = alloc_percpu(struct um_pci_message_buffer);
869 	if (!um_pci_msg_bufs)
870 		return -ENOMEM;
871 
872 	bridge = pci_alloc_host_bridge(0);
873 	if (!bridge) {
874 		err = -ENOMEM;
875 		goto free;
876 	}
877 
878 	um_pci_fwnode = irq_domain_alloc_named_fwnode("um-pci");
879 	if (!um_pci_fwnode) {
880 		err = -ENOMEM;
881 		goto free;
882 	}
883 
884 	um_pci_inner_domain = __irq_domain_add(um_pci_fwnode, MAX_MSI_VECTORS,
885 					       MAX_MSI_VECTORS, 0,
886 					       &um_pci_inner_domain_ops, NULL);
887 	if (!um_pci_inner_domain) {
888 		err = -ENOMEM;
889 		goto free;
890 	}
891 
892 	um_pci_msi_domain = pci_msi_create_irq_domain(um_pci_fwnode,
893 						      &um_pci_msi_domain_info,
894 						      um_pci_inner_domain);
895 	if (!um_pci_msi_domain) {
896 		err = -ENOMEM;
897 		goto free;
898 	}
899 
900 	pci_add_resource(&bridge->windows, &virt_iomem_resource);
901 	pci_add_resource(&bridge->windows, &busn_resource);
902 	bridge->ops = &um_pci_ops;
903 	bridge->map_irq = um_pci_map_irq;
904 
905 	for (i = 0; i < MAX_DEVICES; i++) {
906 		resource_size_t start;
907 
908 		start = virt_cfgspace_resource.start + i * CFG_SPACE_SIZE;
909 		um_pci_devices[i].iomem = ioremap(start, CFG_SPACE_SIZE);
910 		if (WARN(!um_pci_devices[i].iomem, "failed to map %d\n", i)) {
911 			err = -ENOMEM;
912 			goto free;
913 		}
914 	}
915 
916 	err = pci_host_probe(bridge);
917 	if (err)
918 		goto free;
919 
920 	err = register_virtio_driver(&um_pci_virtio_driver);
921 	if (err)
922 		goto free;
923 	return 0;
924 free:
925 	if (um_pci_inner_domain)
926 		irq_domain_remove(um_pci_inner_domain);
927 	if (um_pci_fwnode)
928 		irq_domain_free_fwnode(um_pci_fwnode);
929 	if (bridge) {
930 		pci_free_resource_list(&bridge->windows);
931 		pci_free_host_bridge(bridge);
932 	}
933 	free_percpu(um_pci_msg_bufs);
934 	return err;
935 }
936 module_init(um_pci_init);
937 
938 static void um_pci_exit(void)
939 {
940 	unregister_virtio_driver(&um_pci_virtio_driver);
941 	irq_domain_remove(um_pci_msi_domain);
942 	irq_domain_remove(um_pci_inner_domain);
943 	pci_free_resource_list(&bridge->windows);
944 	pci_free_host_bridge(bridge);
945 	free_percpu(um_pci_msg_bufs);
946 }
947 module_exit(um_pci_exit);
948