• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
um_pci_send_cmd(struct um_pci_device * dev,struct virtio_pcidev_msg * cmd,unsigned int cmd_size,const void * extra,unsigned int extra_size,void * out,unsigned int out_size)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 		if (posted)
136 			kfree(cmd);
137 		goto out;
138 	}
139 
140 	if (posted) {
141 		virtqueue_kick(dev->cmd_vq);
142 		ret = 0;
143 		goto out;
144 	}
145 
146 	/* kick and poll for getting a response on the queue */
147 	set_bit(UM_PCI_STAT_WAITING, &dev->status);
148 	virtqueue_kick(dev->cmd_vq);
149 
150 	while (1) {
151 		void *completed = virtqueue_get_buf(dev->cmd_vq, &len);
152 
153 		if (completed == HANDLE_NO_FREE(cmd))
154 			break;
155 
156 		if (completed && !HANDLE_IS_NO_FREE(completed))
157 			kfree(completed);
158 
159 		if (WARN_ONCE(virtqueue_is_broken(dev->cmd_vq) ||
160 			      ++delay_count > UM_VIRT_PCI_MAXDELAY,
161 			      "um virt-pci delay: %d", delay_count)) {
162 			ret = -EIO;
163 			break;
164 		}
165 		udelay(1);
166 	}
167 	clear_bit(UM_PCI_STAT_WAITING, &dev->status);
168 
169 out:
170 	put_cpu_var(um_pci_msg_bufs);
171 	return ret;
172 }
173 
um_pci_cfgspace_read(void * priv,unsigned int offset,int size)174 static unsigned long um_pci_cfgspace_read(void *priv, unsigned int offset,
175 					  int size)
176 {
177 	struct um_pci_device_reg *reg = priv;
178 	struct um_pci_device *dev = reg->dev;
179 	struct virtio_pcidev_msg hdr = {
180 		.op = VIRTIO_PCIDEV_OP_CFG_READ,
181 		.size = size,
182 		.addr = offset,
183 	};
184 	/* buf->data is maximum size - we may only use parts of it */
185 	struct um_pci_message_buffer *buf;
186 	u8 *data;
187 	unsigned long ret = ULONG_MAX;
188 
189 	if (!dev)
190 		return ULONG_MAX;
191 
192 	buf = get_cpu_var(um_pci_msg_bufs);
193 	data = buf->data;
194 
195 	memset(buf->data, 0xff, sizeof(buf->data));
196 
197 	switch (size) {
198 	case 1:
199 	case 2:
200 	case 4:
201 #ifdef CONFIG_64BIT
202 	case 8:
203 #endif
204 		break;
205 	default:
206 		WARN(1, "invalid config space read size %d\n", size);
207 		goto out;
208 	}
209 
210 	if (um_pci_send_cmd(dev, &hdr, sizeof(hdr), NULL, 0, data, 8))
211 		goto out;
212 
213 	switch (size) {
214 	case 1:
215 		ret = data[0];
216 		break;
217 	case 2:
218 		ret = le16_to_cpup((void *)data);
219 		break;
220 	case 4:
221 		ret = le32_to_cpup((void *)data);
222 		break;
223 #ifdef CONFIG_64BIT
224 	case 8:
225 		ret = le64_to_cpup((void *)data);
226 		break;
227 #endif
228 	default:
229 		break;
230 	}
231 
232 out:
233 	put_cpu_var(um_pci_msg_bufs);
234 	return ret;
235 }
236 
um_pci_cfgspace_write(void * priv,unsigned int offset,int size,unsigned long val)237 static void um_pci_cfgspace_write(void *priv, unsigned int offset, int size,
238 				  unsigned long val)
239 {
240 	struct um_pci_device_reg *reg = priv;
241 	struct um_pci_device *dev = reg->dev;
242 	struct {
243 		struct virtio_pcidev_msg hdr;
244 		/* maximum size - we may only use parts of it */
245 		u8 data[8];
246 	} msg = {
247 		.hdr = {
248 			.op = VIRTIO_PCIDEV_OP_CFG_WRITE,
249 			.size = size,
250 			.addr = offset,
251 		},
252 	};
253 
254 	if (!dev)
255 		return;
256 
257 	switch (size) {
258 	case 1:
259 		msg.data[0] = (u8)val;
260 		break;
261 	case 2:
262 		put_unaligned_le16(val, (void *)msg.data);
263 		break;
264 	case 4:
265 		put_unaligned_le32(val, (void *)msg.data);
266 		break;
267 #ifdef CONFIG_64BIT
268 	case 8:
269 		put_unaligned_le64(val, (void *)msg.data);
270 		break;
271 #endif
272 	default:
273 		WARN(1, "invalid config space write size %d\n", size);
274 		return;
275 	}
276 
277 	WARN_ON(um_pci_send_cmd(dev, &msg.hdr, sizeof(msg), NULL, 0, NULL, 0));
278 }
279 
280 static const struct logic_iomem_ops um_pci_device_cfgspace_ops = {
281 	.read = um_pci_cfgspace_read,
282 	.write = um_pci_cfgspace_write,
283 };
284 
um_pci_bar_copy_from(void * priv,void * buffer,unsigned int offset,int size)285 static void um_pci_bar_copy_from(void *priv, void *buffer,
286 				 unsigned int offset, int size)
287 {
288 	u8 *resptr = priv;
289 	struct um_pci_device *dev = container_of(resptr - *resptr,
290 						 struct um_pci_device,
291 						 resptr[0]);
292 	struct virtio_pcidev_msg hdr = {
293 		.op = VIRTIO_PCIDEV_OP_MMIO_READ,
294 		.bar = *resptr,
295 		.size = size,
296 		.addr = offset,
297 	};
298 
299 	memset(buffer, 0xff, size);
300 
301 	um_pci_send_cmd(dev, &hdr, sizeof(hdr), NULL, 0, buffer, size);
302 }
303 
um_pci_bar_read(void * priv,unsigned int offset,int size)304 static unsigned long um_pci_bar_read(void *priv, unsigned int offset,
305 				     int size)
306 {
307 	/* buf->data is maximum size - we may only use parts of it */
308 	struct um_pci_message_buffer *buf;
309 	u8 *data;
310 	unsigned long ret = ULONG_MAX;
311 
312 	buf = get_cpu_var(um_pci_msg_bufs);
313 	data = buf->data;
314 
315 	switch (size) {
316 	case 1:
317 	case 2:
318 	case 4:
319 #ifdef CONFIG_64BIT
320 	case 8:
321 #endif
322 		break;
323 	default:
324 		WARN(1, "invalid config space read size %d\n", size);
325 		goto out;
326 	}
327 
328 	um_pci_bar_copy_from(priv, data, offset, size);
329 
330 	switch (size) {
331 	case 1:
332 		ret = data[0];
333 		break;
334 	case 2:
335 		ret = le16_to_cpup((void *)data);
336 		break;
337 	case 4:
338 		ret = le32_to_cpup((void *)data);
339 		break;
340 #ifdef CONFIG_64BIT
341 	case 8:
342 		ret = le64_to_cpup((void *)data);
343 		break;
344 #endif
345 	default:
346 		break;
347 	}
348 
349 out:
350 	put_cpu_var(um_pci_msg_bufs);
351 	return ret;
352 }
353 
um_pci_bar_copy_to(void * priv,unsigned int offset,const void * buffer,int size)354 static void um_pci_bar_copy_to(void *priv, unsigned int offset,
355 			       const void *buffer, int size)
356 {
357 	u8 *resptr = priv;
358 	struct um_pci_device *dev = container_of(resptr - *resptr,
359 						 struct um_pci_device,
360 						 resptr[0]);
361 	struct virtio_pcidev_msg hdr = {
362 		.op = VIRTIO_PCIDEV_OP_MMIO_WRITE,
363 		.bar = *resptr,
364 		.size = size,
365 		.addr = offset,
366 	};
367 
368 	um_pci_send_cmd(dev, &hdr, sizeof(hdr), buffer, size, NULL, 0);
369 }
370 
um_pci_bar_write(void * priv,unsigned int offset,int size,unsigned long val)371 static void um_pci_bar_write(void *priv, unsigned int offset, int size,
372 			     unsigned long val)
373 {
374 	/* maximum size - we may only use parts of it */
375 	u8 data[8];
376 
377 	switch (size) {
378 	case 1:
379 		data[0] = (u8)val;
380 		break;
381 	case 2:
382 		put_unaligned_le16(val, (void *)data);
383 		break;
384 	case 4:
385 		put_unaligned_le32(val, (void *)data);
386 		break;
387 #ifdef CONFIG_64BIT
388 	case 8:
389 		put_unaligned_le64(val, (void *)data);
390 		break;
391 #endif
392 	default:
393 		WARN(1, "invalid config space write size %d\n", size);
394 		return;
395 	}
396 
397 	um_pci_bar_copy_to(priv, offset, data, size);
398 }
399 
um_pci_bar_set(void * priv,unsigned int offset,u8 value,int size)400 static void um_pci_bar_set(void *priv, unsigned int offset, u8 value, int size)
401 {
402 	u8 *resptr = priv;
403 	struct um_pci_device *dev = container_of(resptr - *resptr,
404 						 struct um_pci_device,
405 						 resptr[0]);
406 	struct {
407 		struct virtio_pcidev_msg hdr;
408 		u8 data;
409 	} msg = {
410 		.hdr = {
411 			.op = VIRTIO_PCIDEV_OP_CFG_WRITE,
412 			.bar = *resptr,
413 			.size = size,
414 			.addr = offset,
415 		},
416 		.data = value,
417 	};
418 
419 	um_pci_send_cmd(dev, &msg.hdr, sizeof(msg), NULL, 0, NULL, 0);
420 }
421 
422 static const struct logic_iomem_ops um_pci_device_bar_ops = {
423 	.read = um_pci_bar_read,
424 	.write = um_pci_bar_write,
425 	.set = um_pci_bar_set,
426 	.copy_from = um_pci_bar_copy_from,
427 	.copy_to = um_pci_bar_copy_to,
428 };
429 
um_pci_map_bus(struct pci_bus * bus,unsigned int devfn,int where)430 static void __iomem *um_pci_map_bus(struct pci_bus *bus, unsigned int devfn,
431 				    int where)
432 {
433 	struct um_pci_device_reg *dev;
434 	unsigned int busn = bus->number;
435 
436 	if (busn > 0)
437 		return NULL;
438 
439 	/* not allowing functions for now ... */
440 	if (devfn % 8)
441 		return NULL;
442 
443 	if (devfn / 8 >= ARRAY_SIZE(um_pci_devices))
444 		return NULL;
445 
446 	dev = &um_pci_devices[devfn / 8];
447 	if (!dev)
448 		return NULL;
449 
450 	return (void __iomem *)((unsigned long)dev->iomem + where);
451 }
452 
453 static struct pci_ops um_pci_ops = {
454 	.map_bus = um_pci_map_bus,
455 	.read = pci_generic_config_read,
456 	.write = pci_generic_config_write,
457 };
458 
um_pci_rescan(void)459 static void um_pci_rescan(void)
460 {
461 	pci_lock_rescan_remove();
462 	pci_rescan_bus(bridge->bus);
463 	pci_unlock_rescan_remove();
464 }
465 
um_pci_irq_vq_addbuf(struct virtqueue * vq,void * buf,bool kick)466 static void um_pci_irq_vq_addbuf(struct virtqueue *vq, void *buf, bool kick)
467 {
468 	struct scatterlist sg[1];
469 
470 	sg_init_one(sg, buf, MAX_IRQ_MSG_SIZE);
471 	if (virtqueue_add_inbuf(vq, sg, 1, buf, GFP_ATOMIC))
472 		kfree(buf);
473 	else if (kick)
474 		virtqueue_kick(vq);
475 }
476 
um_pci_handle_irq_message(struct virtqueue * vq,struct virtio_pcidev_msg * msg)477 static void um_pci_handle_irq_message(struct virtqueue *vq,
478 				      struct virtio_pcidev_msg *msg)
479 {
480 	struct virtio_device *vdev = vq->vdev;
481 	struct um_pci_device *dev = vdev->priv;
482 
483 	/* we should properly chain interrupts, but on ARCH=um we don't care */
484 
485 	switch (msg->op) {
486 	case VIRTIO_PCIDEV_OP_INT:
487 		generic_handle_irq(dev->irq);
488 		break;
489 	case VIRTIO_PCIDEV_OP_MSI:
490 		/* our MSI message is just the interrupt number */
491 		if (msg->size == sizeof(u32))
492 			generic_handle_irq(le32_to_cpup((void *)msg->data));
493 		else
494 			generic_handle_irq(le16_to_cpup((void *)msg->data));
495 		break;
496 	case VIRTIO_PCIDEV_OP_PME:
497 		/* nothing to do - we already woke up due to the message */
498 		break;
499 	default:
500 		dev_err(&vdev->dev, "unexpected virt-pci message %d\n", msg->op);
501 		break;
502 	}
503 }
504 
um_pci_cmd_vq_cb(struct virtqueue * vq)505 static void um_pci_cmd_vq_cb(struct virtqueue *vq)
506 {
507 	struct virtio_device *vdev = vq->vdev;
508 	struct um_pci_device *dev = vdev->priv;
509 	void *cmd;
510 	int len;
511 
512 	if (test_bit(UM_PCI_STAT_WAITING, &dev->status))
513 		return;
514 
515 	while ((cmd = virtqueue_get_buf(vq, &len))) {
516 		if (WARN_ON(HANDLE_IS_NO_FREE(cmd)))
517 			continue;
518 		kfree(cmd);
519 	}
520 }
521 
um_pci_irq_vq_cb(struct virtqueue * vq)522 static void um_pci_irq_vq_cb(struct virtqueue *vq)
523 {
524 	struct virtio_pcidev_msg *msg;
525 	int len;
526 
527 	while ((msg = virtqueue_get_buf(vq, &len))) {
528 		if (len >= sizeof(*msg))
529 			um_pci_handle_irq_message(vq, msg);
530 
531 		/* recycle the message buffer */
532 		um_pci_irq_vq_addbuf(vq, msg, true);
533 	}
534 }
535 
um_pci_init_vqs(struct um_pci_device * dev)536 static int um_pci_init_vqs(struct um_pci_device *dev)
537 {
538 	struct virtqueue *vqs[2];
539 	static const char *const names[2] = { "cmd", "irq" };
540 	vq_callback_t *cbs[2] = { um_pci_cmd_vq_cb, um_pci_irq_vq_cb };
541 	int err, i;
542 
543 	err = virtio_find_vqs(dev->vdev, 2, vqs, cbs, names, NULL);
544 	if (err)
545 		return err;
546 
547 	dev->cmd_vq = vqs[0];
548 	dev->irq_vq = vqs[1];
549 
550 	for (i = 0; i < NUM_IRQ_MSGS; i++) {
551 		void *msg = kzalloc(MAX_IRQ_MSG_SIZE, GFP_KERNEL);
552 
553 		if (msg)
554 			um_pci_irq_vq_addbuf(dev->irq_vq, msg, false);
555 	}
556 
557 	virtqueue_kick(dev->irq_vq);
558 
559 	return 0;
560 }
561 
um_pci_virtio_probe(struct virtio_device * vdev)562 static int um_pci_virtio_probe(struct virtio_device *vdev)
563 {
564 	struct um_pci_device *dev;
565 	int i, free = -1;
566 	int err = -ENOSPC;
567 
568 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
569 	if (!dev)
570 		return -ENOMEM;
571 
572 	dev->vdev = vdev;
573 	vdev->priv = dev;
574 
575 	mutex_lock(&um_pci_mtx);
576 	for (i = 0; i < MAX_DEVICES; i++) {
577 		if (um_pci_devices[i].dev)
578 			continue;
579 		free = i;
580 		break;
581 	}
582 
583 	if (free < 0)
584 		goto error;
585 
586 	err = um_pci_init_vqs(dev);
587 	if (err)
588 		goto error;
589 
590 	dev->irq = irq_alloc_desc(numa_node_id());
591 	if (dev->irq < 0) {
592 		err = dev->irq;
593 		goto error;
594 	}
595 	um_pci_devices[free].dev = dev;
596 	vdev->priv = dev;
597 
598 	mutex_unlock(&um_pci_mtx);
599 
600 	device_set_wakeup_enable(&vdev->dev, true);
601 
602 	/*
603 	 * In order to do suspend-resume properly, don't allow VQs
604 	 * to be suspended.
605 	 */
606 	virtio_uml_set_no_vq_suspend(vdev, true);
607 
608 	um_pci_rescan();
609 	return 0;
610 error:
611 	mutex_unlock(&um_pci_mtx);
612 	kfree(dev);
613 	return err;
614 }
615 
um_pci_virtio_remove(struct virtio_device * vdev)616 static void um_pci_virtio_remove(struct virtio_device *vdev)
617 {
618 	struct um_pci_device *dev = vdev->priv;
619 	int i;
620 
621 	device_set_wakeup_enable(&vdev->dev, false);
622 
623 	mutex_lock(&um_pci_mtx);
624 	for (i = 0; i < MAX_DEVICES; i++) {
625 		if (um_pci_devices[i].dev != dev)
626 			continue;
627 
628 		um_pci_devices[i].dev = NULL;
629 		irq_free_desc(dev->irq);
630 
631 		break;
632 	}
633 	mutex_unlock(&um_pci_mtx);
634 
635 	if (i < MAX_DEVICES) {
636 		struct pci_dev *pci_dev;
637 
638 		pci_dev = pci_get_slot(bridge->bus, i);
639 		if (pci_dev)
640 			pci_stop_and_remove_bus_device_locked(pci_dev);
641 	}
642 
643 	/* Stop all virtqueues */
644 	virtio_reset_device(vdev);
645 	dev->cmd_vq = NULL;
646 	dev->irq_vq = NULL;
647 	vdev->config->del_vqs(vdev);
648 
649 	kfree(dev);
650 }
651 
652 static struct virtio_device_id id_table[] = {
653 	{ CONFIG_UML_PCI_OVER_VIRTIO_DEVICE_ID, VIRTIO_DEV_ANY_ID },
654 	{ 0 },
655 };
656 MODULE_DEVICE_TABLE(virtio, id_table);
657 
658 static struct virtio_driver um_pci_virtio_driver = {
659 	.driver.name = "virtio-pci",
660 	.driver.owner = THIS_MODULE,
661 	.id_table = id_table,
662 	.probe = um_pci_virtio_probe,
663 	.remove = um_pci_virtio_remove,
664 };
665 
666 static struct resource virt_cfgspace_resource = {
667 	.name = "PCI config space",
668 	.start = 0xf0000000 - MAX_DEVICES * CFG_SPACE_SIZE,
669 	.end = 0xf0000000 - 1,
670 	.flags = IORESOURCE_MEM,
671 };
672 
um_pci_map_cfgspace(unsigned long offset,size_t size,const struct logic_iomem_ops ** ops,void ** priv)673 static long um_pci_map_cfgspace(unsigned long offset, size_t size,
674 				const struct logic_iomem_ops **ops,
675 				void **priv)
676 {
677 	if (WARN_ON(size > CFG_SPACE_SIZE || offset % CFG_SPACE_SIZE))
678 		return -EINVAL;
679 
680 	if (offset / CFG_SPACE_SIZE < MAX_DEVICES) {
681 		*ops = &um_pci_device_cfgspace_ops;
682 		*priv = &um_pci_devices[offset / CFG_SPACE_SIZE];
683 		return 0;
684 	}
685 
686 	WARN(1, "cannot map offset 0x%lx/0x%zx\n", offset, size);
687 	return -ENOENT;
688 }
689 
690 static const struct logic_iomem_region_ops um_pci_cfgspace_ops = {
691 	.map = um_pci_map_cfgspace,
692 };
693 
694 static struct resource virt_iomem_resource = {
695 	.name = "PCI iomem",
696 	.start = 0xf0000000,
697 	.end = 0xffffffff,
698 	.flags = IORESOURCE_MEM,
699 };
700 
701 struct um_pci_map_iomem_data {
702 	unsigned long offset;
703 	size_t size;
704 	const struct logic_iomem_ops **ops;
705 	void **priv;
706 	long ret;
707 };
708 
um_pci_map_iomem_walk(struct pci_dev * pdev,void * _data)709 static int um_pci_map_iomem_walk(struct pci_dev *pdev, void *_data)
710 {
711 	struct um_pci_map_iomem_data *data = _data;
712 	struct um_pci_device_reg *reg = &um_pci_devices[pdev->devfn / 8];
713 	struct um_pci_device *dev;
714 	int i;
715 
716 	if (!reg->dev)
717 		return 0;
718 
719 	for (i = 0; i < ARRAY_SIZE(dev->resptr); i++) {
720 		struct resource *r = &pdev->resource[i];
721 
722 		if ((r->flags & IORESOURCE_TYPE_BITS) != IORESOURCE_MEM)
723 			continue;
724 
725 		/*
726 		 * must be the whole or part of the resource,
727 		 * not allowed to only overlap
728 		 */
729 		if (data->offset < r->start || data->offset > r->end)
730 			continue;
731 		if (data->offset + data->size - 1 > r->end)
732 			continue;
733 
734 		dev = reg->dev;
735 		*data->ops = &um_pci_device_bar_ops;
736 		dev->resptr[i] = i;
737 		*data->priv = &dev->resptr[i];
738 		data->ret = data->offset - r->start;
739 
740 		/* no need to continue */
741 		return 1;
742 	}
743 
744 	return 0;
745 }
746 
um_pci_map_iomem(unsigned long offset,size_t size,const struct logic_iomem_ops ** ops,void ** priv)747 static long um_pci_map_iomem(unsigned long offset, size_t size,
748 			     const struct logic_iomem_ops **ops,
749 			     void **priv)
750 {
751 	struct um_pci_map_iomem_data data = {
752 		/* we want the full address here */
753 		.offset = offset + virt_iomem_resource.start,
754 		.size = size,
755 		.ops = ops,
756 		.priv = priv,
757 		.ret = -ENOENT,
758 	};
759 
760 	pci_walk_bus(bridge->bus, um_pci_map_iomem_walk, &data);
761 	return data.ret;
762 }
763 
764 static const struct logic_iomem_region_ops um_pci_iomem_ops = {
765 	.map = um_pci_map_iomem,
766 };
767 
um_pci_compose_msi_msg(struct irq_data * data,struct msi_msg * msg)768 static void um_pci_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
769 {
770 	/*
771 	 * This is a very low address and not actually valid 'physical' memory
772 	 * in UML, so we can simply map MSI(-X) vectors to there, it cannot be
773 	 * legitimately written to by the device in any other way.
774 	 * We use the (virtual) IRQ number here as the message to simplify the
775 	 * code that receives the message, where for now we simply trust the
776 	 * device to send the correct message.
777 	 */
778 	msg->address_hi = 0;
779 	msg->address_lo = 0xa0000;
780 	msg->data = data->irq;
781 }
782 
783 static struct irq_chip um_pci_msi_bottom_irq_chip = {
784 	.name = "UM virtio MSI",
785 	.irq_compose_msi_msg = um_pci_compose_msi_msg,
786 };
787 
um_pci_inner_domain_alloc(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs,void * args)788 static int um_pci_inner_domain_alloc(struct irq_domain *domain,
789 				     unsigned int virq, unsigned int nr_irqs,
790 				     void *args)
791 {
792 	unsigned long bit;
793 
794 	WARN_ON(nr_irqs != 1);
795 
796 	mutex_lock(&um_pci_mtx);
797 	bit = find_first_zero_bit(um_pci_msi_used, MAX_MSI_VECTORS);
798 	if (bit >= MAX_MSI_VECTORS) {
799 		mutex_unlock(&um_pci_mtx);
800 		return -ENOSPC;
801 	}
802 
803 	set_bit(bit, um_pci_msi_used);
804 	mutex_unlock(&um_pci_mtx);
805 
806 	irq_domain_set_info(domain, virq, bit, &um_pci_msi_bottom_irq_chip,
807 			    domain->host_data, handle_simple_irq,
808 			    NULL, NULL);
809 
810 	return 0;
811 }
812 
um_pci_inner_domain_free(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs)813 static void um_pci_inner_domain_free(struct irq_domain *domain,
814 				     unsigned int virq, unsigned int nr_irqs)
815 {
816 	struct irq_data *d = irq_domain_get_irq_data(domain, virq);
817 
818 	mutex_lock(&um_pci_mtx);
819 
820 	if (!test_bit(d->hwirq, um_pci_msi_used))
821 		pr_err("trying to free unused MSI#%lu\n", d->hwirq);
822 	else
823 		__clear_bit(d->hwirq, um_pci_msi_used);
824 
825 	mutex_unlock(&um_pci_mtx);
826 }
827 
828 static const struct irq_domain_ops um_pci_inner_domain_ops = {
829 	.alloc = um_pci_inner_domain_alloc,
830 	.free = um_pci_inner_domain_free,
831 };
832 
833 static struct irq_chip um_pci_msi_irq_chip = {
834 	.name = "UM virtio PCIe MSI",
835 	.irq_mask = pci_msi_mask_irq,
836 	.irq_unmask = pci_msi_unmask_irq,
837 };
838 
839 static struct msi_domain_info um_pci_msi_domain_info = {
840 	.flags	= MSI_FLAG_USE_DEF_DOM_OPS |
841 		  MSI_FLAG_USE_DEF_CHIP_OPS |
842 		  MSI_FLAG_PCI_MSIX,
843 	.chip	= &um_pci_msi_irq_chip,
844 };
845 
846 static struct resource busn_resource = {
847 	.name	= "PCI busn",
848 	.start	= 0,
849 	.end	= 0,
850 	.flags	= IORESOURCE_BUS,
851 };
852 
um_pci_map_irq(const struct pci_dev * pdev,u8 slot,u8 pin)853 static int um_pci_map_irq(const struct pci_dev *pdev, u8 slot, u8 pin)
854 {
855 	struct um_pci_device_reg *reg = &um_pci_devices[pdev->devfn / 8];
856 
857 	if (WARN_ON(!reg->dev))
858 		return -EINVAL;
859 
860 	/* Yes, we map all pins to the same IRQ ... doesn't matter for now. */
861 	return reg->dev->irq;
862 }
863 
pci_root_bus_fwnode(struct pci_bus * bus)864 void *pci_root_bus_fwnode(struct pci_bus *bus)
865 {
866 	return um_pci_fwnode;
867 }
868 
um_pci_init(void)869 static int um_pci_init(void)
870 {
871 	int err, i;
872 
873 	WARN_ON(logic_iomem_add_region(&virt_cfgspace_resource,
874 				       &um_pci_cfgspace_ops));
875 	WARN_ON(logic_iomem_add_region(&virt_iomem_resource,
876 				       &um_pci_iomem_ops));
877 
878 	if (WARN(CONFIG_UML_PCI_OVER_VIRTIO_DEVICE_ID < 0,
879 		 "No virtio device ID configured for PCI - no PCI support\n"))
880 		return 0;
881 
882 	um_pci_msg_bufs = alloc_percpu(struct um_pci_message_buffer);
883 	if (!um_pci_msg_bufs)
884 		return -ENOMEM;
885 
886 	bridge = pci_alloc_host_bridge(0);
887 	if (!bridge) {
888 		err = -ENOMEM;
889 		goto free;
890 	}
891 
892 	um_pci_fwnode = irq_domain_alloc_named_fwnode("um-pci");
893 	if (!um_pci_fwnode) {
894 		err = -ENOMEM;
895 		goto free;
896 	}
897 
898 	um_pci_inner_domain = __irq_domain_add(um_pci_fwnode, MAX_MSI_VECTORS,
899 					       MAX_MSI_VECTORS, 0,
900 					       &um_pci_inner_domain_ops, NULL);
901 	if (!um_pci_inner_domain) {
902 		err = -ENOMEM;
903 		goto free;
904 	}
905 
906 	um_pci_msi_domain = pci_msi_create_irq_domain(um_pci_fwnode,
907 						      &um_pci_msi_domain_info,
908 						      um_pci_inner_domain);
909 	if (!um_pci_msi_domain) {
910 		err = -ENOMEM;
911 		goto free;
912 	}
913 
914 	pci_add_resource(&bridge->windows, &virt_iomem_resource);
915 	pci_add_resource(&bridge->windows, &busn_resource);
916 	bridge->ops = &um_pci_ops;
917 	bridge->map_irq = um_pci_map_irq;
918 
919 	for (i = 0; i < MAX_DEVICES; i++) {
920 		resource_size_t start;
921 
922 		start = virt_cfgspace_resource.start + i * CFG_SPACE_SIZE;
923 		um_pci_devices[i].iomem = ioremap(start, CFG_SPACE_SIZE);
924 		if (WARN(!um_pci_devices[i].iomem, "failed to map %d\n", i)) {
925 			err = -ENOMEM;
926 			goto free;
927 		}
928 	}
929 
930 	err = pci_host_probe(bridge);
931 	if (err)
932 		goto free;
933 
934 	err = register_virtio_driver(&um_pci_virtio_driver);
935 	if (err)
936 		goto free;
937 	return 0;
938 free:
939 	if (um_pci_inner_domain)
940 		irq_domain_remove(um_pci_inner_domain);
941 	if (um_pci_fwnode)
942 		irq_domain_free_fwnode(um_pci_fwnode);
943 	if (bridge) {
944 		pci_free_resource_list(&bridge->windows);
945 		pci_free_host_bridge(bridge);
946 	}
947 	free_percpu(um_pci_msg_bufs);
948 	return err;
949 }
950 module_init(um_pci_init);
951 
um_pci_exit(void)952 static void um_pci_exit(void)
953 {
954 	unregister_virtio_driver(&um_pci_virtio_driver);
955 	irq_domain_remove(um_pci_msi_domain);
956 	irq_domain_remove(um_pci_inner_domain);
957 	pci_free_resource_list(&bridge->windows);
958 	pci_free_host_bridge(bridge);
959 	free_percpu(um_pci_msg_bufs);
960 }
961 module_exit(um_pci_exit);
962