xref: /linux/arch/um/drivers/virt-pci.c (revision ad73b9a17d66366d8c9198bc90f1ea99f24a912c)
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/of_platform.h>
12 #include <linux/irqdomain.h>
13 #include <linux/virtio_pcidev.h>
14 #include <linux/virtio-uml.h>
15 #include <linux/delay.h>
16 #include <linux/msi.h>
17 #include <linux/unaligned.h>
18 #include <irq_kern.h>
19 
20 #define MAX_DEVICES 8
21 #define MAX_MSI_VECTORS 32
22 #define CFG_SPACE_SIZE 4096
23 
24 /* for MSI-X we have a 32-bit payload */
25 #define MAX_IRQ_MSG_SIZE (sizeof(struct virtio_pcidev_msg) + sizeof(u32))
26 #define NUM_IRQ_MSGS	10
27 
28 struct um_pci_message_buffer {
29 	struct virtio_pcidev_msg hdr;
30 	u8 data[8];
31 };
32 
33 struct um_pci_device {
34 	struct virtio_device *vdev;
35 
36 	/* for now just standard BARs */
37 	u8 resptr[PCI_STD_NUM_BARS];
38 
39 	struct virtqueue *cmd_vq, *irq_vq;
40 
41 #define UM_PCI_WRITE_BUFS	20
42 	struct um_pci_message_buffer bufs[UM_PCI_WRITE_BUFS + 1];
43 	void *extra_ptrs[UM_PCI_WRITE_BUFS + 1];
44 	DECLARE_BITMAP(used_bufs, UM_PCI_WRITE_BUFS);
45 
46 #define UM_PCI_STAT_WAITING	0
47 	unsigned long status;
48 
49 	int irq;
50 
51 	bool platform;
52 };
53 
54 struct um_pci_device_reg {
55 	struct um_pci_device *dev;
56 	void __iomem *iomem;
57 };
58 
59 static struct pci_host_bridge *bridge;
60 static DEFINE_MUTEX(um_pci_mtx);
61 static struct um_pci_device *um_pci_platform_device;
62 static struct um_pci_device_reg um_pci_devices[MAX_DEVICES];
63 static struct fwnode_handle *um_pci_fwnode;
64 static struct irq_domain *um_pci_inner_domain;
65 static struct irq_domain *um_pci_msi_domain;
66 static unsigned long um_pci_msi_used[BITS_TO_LONGS(MAX_MSI_VECTORS)];
67 
68 static unsigned int um_pci_max_delay_us = 40000;
69 module_param_named(max_delay_us, um_pci_max_delay_us, uint, 0644);
70 
um_pci_get_buf(struct um_pci_device * dev,bool * posted)71 static int um_pci_get_buf(struct um_pci_device *dev, bool *posted)
72 {
73 	int i;
74 
75 	for (i = 0; i < UM_PCI_WRITE_BUFS; i++) {
76 		if (!test_and_set_bit(i, dev->used_bufs))
77 			return i;
78 	}
79 
80 	*posted = false;
81 	return UM_PCI_WRITE_BUFS;
82 }
83 
um_pci_free_buf(struct um_pci_device * dev,void * buf)84 static void um_pci_free_buf(struct um_pci_device *dev, void *buf)
85 {
86 	int i;
87 
88 	if (buf == &dev->bufs[UM_PCI_WRITE_BUFS]) {
89 		kfree(dev->extra_ptrs[UM_PCI_WRITE_BUFS]);
90 		dev->extra_ptrs[UM_PCI_WRITE_BUFS] = NULL;
91 		return;
92 	}
93 
94 	for (i = 0; i < UM_PCI_WRITE_BUFS; i++) {
95 		if (buf == &dev->bufs[i]) {
96 			kfree(dev->extra_ptrs[i]);
97 			dev->extra_ptrs[i] = NULL;
98 			WARN_ON(!test_and_clear_bit(i, dev->used_bufs));
99 			return;
100 		}
101 	}
102 
103 	WARN_ON(1);
104 }
105 
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)106 static int um_pci_send_cmd(struct um_pci_device *dev,
107 			   struct virtio_pcidev_msg *cmd,
108 			   unsigned int cmd_size,
109 			   const void *extra, unsigned int extra_size,
110 			   void *out, unsigned int out_size)
111 {
112 	struct scatterlist out_sg, extra_sg, in_sg;
113 	struct scatterlist *sgs_list[] = {
114 		[0] = &out_sg,
115 		[1] = extra ? &extra_sg : &in_sg,
116 		[2] = extra ? &in_sg : NULL,
117 	};
118 	struct um_pci_message_buffer *buf;
119 	int delay_count = 0;
120 	bool bounce_out;
121 	int ret, len;
122 	int buf_idx;
123 	bool posted;
124 
125 	if (WARN_ON(cmd_size < sizeof(*cmd) || cmd_size > sizeof(*buf)))
126 		return -EINVAL;
127 
128 	switch (cmd->op) {
129 	case VIRTIO_PCIDEV_OP_CFG_WRITE:
130 	case VIRTIO_PCIDEV_OP_MMIO_WRITE:
131 	case VIRTIO_PCIDEV_OP_MMIO_MEMSET:
132 		/* in PCI, writes are posted, so don't wait */
133 		posted = !out;
134 		WARN_ON(!posted);
135 		break;
136 	default:
137 		posted = false;
138 		break;
139 	}
140 
141 	bounce_out = !posted && cmd_size <= sizeof(*cmd) &&
142 		     out && out_size <= sizeof(buf->data);
143 
144 	buf_idx = um_pci_get_buf(dev, &posted);
145 	buf = &dev->bufs[buf_idx];
146 	memcpy(buf, cmd, cmd_size);
147 
148 	if (posted && extra && extra_size > sizeof(buf) - cmd_size) {
149 		dev->extra_ptrs[buf_idx] = kmemdup(extra, extra_size,
150 						   GFP_ATOMIC);
151 
152 		if (!dev->extra_ptrs[buf_idx]) {
153 			um_pci_free_buf(dev, buf);
154 			return -ENOMEM;
155 		}
156 		extra = dev->extra_ptrs[buf_idx];
157 	} else if (extra && extra_size <= sizeof(buf) - cmd_size) {
158 		memcpy((u8 *)buf + cmd_size, extra, extra_size);
159 		cmd_size += extra_size;
160 		extra_size = 0;
161 		extra = NULL;
162 		cmd = (void *)buf;
163 	} else {
164 		cmd = (void *)buf;
165 	}
166 
167 	sg_init_one(&out_sg, cmd, cmd_size);
168 	if (extra)
169 		sg_init_one(&extra_sg, extra, extra_size);
170 	/* allow stack for small buffers */
171 	if (bounce_out)
172 		sg_init_one(&in_sg, buf->data, out_size);
173 	else if (out)
174 		sg_init_one(&in_sg, out, out_size);
175 
176 	/* add to internal virtio queue */
177 	ret = virtqueue_add_sgs(dev->cmd_vq, sgs_list,
178 				extra ? 2 : 1,
179 				out ? 1 : 0,
180 				cmd, GFP_ATOMIC);
181 	if (ret) {
182 		um_pci_free_buf(dev, buf);
183 		return ret;
184 	}
185 
186 	if (posted) {
187 		virtqueue_kick(dev->cmd_vq);
188 		return 0;
189 	}
190 
191 	/* kick and poll for getting a response on the queue */
192 	set_bit(UM_PCI_STAT_WAITING, &dev->status);
193 	virtqueue_kick(dev->cmd_vq);
194 	ret = 0;
195 
196 	while (1) {
197 		void *completed = virtqueue_get_buf(dev->cmd_vq, &len);
198 
199 		if (completed == buf)
200 			break;
201 
202 		if (completed)
203 			um_pci_free_buf(dev, completed);
204 
205 		if (WARN_ONCE(virtqueue_is_broken(dev->cmd_vq) ||
206 			      ++delay_count > um_pci_max_delay_us,
207 			      "um virt-pci delay: %d", delay_count)) {
208 			ret = -EIO;
209 			break;
210 		}
211 		udelay(1);
212 	}
213 	clear_bit(UM_PCI_STAT_WAITING, &dev->status);
214 
215 	if (bounce_out)
216 		memcpy(out, buf->data, out_size);
217 
218 	um_pci_free_buf(dev, buf);
219 
220 	return ret;
221 }
222 
um_pci_cfgspace_read(void * priv,unsigned int offset,int size)223 static unsigned long um_pci_cfgspace_read(void *priv, unsigned int offset,
224 					  int size)
225 {
226 	struct um_pci_device_reg *reg = priv;
227 	struct um_pci_device *dev = reg->dev;
228 	struct virtio_pcidev_msg hdr = {
229 		.op = VIRTIO_PCIDEV_OP_CFG_READ,
230 		.size = size,
231 		.addr = offset,
232 	};
233 	/* max 8, we might not use it all */
234 	u8 data[8];
235 
236 	if (!dev)
237 		return ULONG_MAX;
238 
239 	memset(data, 0xff, sizeof(data));
240 
241 	switch (size) {
242 	case 1:
243 	case 2:
244 	case 4:
245 #ifdef CONFIG_64BIT
246 	case 8:
247 #endif
248 		break;
249 	default:
250 		WARN(1, "invalid config space read size %d\n", size);
251 		return ULONG_MAX;
252 	}
253 
254 	if (um_pci_send_cmd(dev, &hdr, sizeof(hdr), NULL, 0, data, size))
255 		return ULONG_MAX;
256 
257 	switch (size) {
258 	case 1:
259 		return data[0];
260 	case 2:
261 		return le16_to_cpup((void *)data);
262 	case 4:
263 		return le32_to_cpup((void *)data);
264 #ifdef CONFIG_64BIT
265 	case 8:
266 		return le64_to_cpup((void *)data);
267 #endif
268 	default:
269 		return ULONG_MAX;
270 	}
271 }
272 
um_pci_cfgspace_write(void * priv,unsigned int offset,int size,unsigned long val)273 static void um_pci_cfgspace_write(void *priv, unsigned int offset, int size,
274 				  unsigned long val)
275 {
276 	struct um_pci_device_reg *reg = priv;
277 	struct um_pci_device *dev = reg->dev;
278 	struct {
279 		struct virtio_pcidev_msg hdr;
280 		/* maximum size - we may only use parts of it */
281 		u8 data[8];
282 	} msg = {
283 		.hdr = {
284 			.op = VIRTIO_PCIDEV_OP_CFG_WRITE,
285 			.size = size,
286 			.addr = offset,
287 		},
288 	};
289 
290 	if (!dev)
291 		return;
292 
293 	switch (size) {
294 	case 1:
295 		msg.data[0] = (u8)val;
296 		break;
297 	case 2:
298 		put_unaligned_le16(val, (void *)msg.data);
299 		break;
300 	case 4:
301 		put_unaligned_le32(val, (void *)msg.data);
302 		break;
303 #ifdef CONFIG_64BIT
304 	case 8:
305 		put_unaligned_le64(val, (void *)msg.data);
306 		break;
307 #endif
308 	default:
309 		WARN(1, "invalid config space write size %d\n", size);
310 		return;
311 	}
312 
313 	WARN_ON(um_pci_send_cmd(dev, &msg.hdr, sizeof(msg), NULL, 0, NULL, 0));
314 }
315 
316 static const struct logic_iomem_ops um_pci_device_cfgspace_ops = {
317 	.read = um_pci_cfgspace_read,
318 	.write = um_pci_cfgspace_write,
319 };
320 
um_pci_bar_copy_from(void * priv,void * buffer,unsigned int offset,int size)321 static void um_pci_bar_copy_from(void *priv, void *buffer,
322 				 unsigned int offset, int size)
323 {
324 	u8 *resptr = priv;
325 	struct um_pci_device *dev = container_of(resptr - *resptr,
326 						 struct um_pci_device,
327 						 resptr[0]);
328 	struct virtio_pcidev_msg hdr = {
329 		.op = VIRTIO_PCIDEV_OP_MMIO_READ,
330 		.bar = *resptr,
331 		.size = size,
332 		.addr = offset,
333 	};
334 
335 	memset(buffer, 0xff, size);
336 
337 	um_pci_send_cmd(dev, &hdr, sizeof(hdr), NULL, 0, buffer, size);
338 }
339 
um_pci_bar_read(void * priv,unsigned int offset,int size)340 static unsigned long um_pci_bar_read(void *priv, unsigned int offset,
341 				     int size)
342 {
343 	/* 8 is maximum size - we may only use parts of it */
344 	u8 data[8];
345 
346 	switch (size) {
347 	case 1:
348 	case 2:
349 	case 4:
350 #ifdef CONFIG_64BIT
351 	case 8:
352 #endif
353 		break;
354 	default:
355 		WARN(1, "invalid config space read size %d\n", size);
356 		return ULONG_MAX;
357 	}
358 
359 	um_pci_bar_copy_from(priv, data, offset, size);
360 
361 	switch (size) {
362 	case 1:
363 		return data[0];
364 	case 2:
365 		return le16_to_cpup((void *)data);
366 	case 4:
367 		return le32_to_cpup((void *)data);
368 #ifdef CONFIG_64BIT
369 	case 8:
370 		return le64_to_cpup((void *)data);
371 #endif
372 	default:
373 		return ULONG_MAX;
374 	}
375 }
376 
um_pci_bar_copy_to(void * priv,unsigned int offset,const void * buffer,int size)377 static void um_pci_bar_copy_to(void *priv, unsigned int offset,
378 			       const void *buffer, int size)
379 {
380 	u8 *resptr = priv;
381 	struct um_pci_device *dev = container_of(resptr - *resptr,
382 						 struct um_pci_device,
383 						 resptr[0]);
384 	struct virtio_pcidev_msg hdr = {
385 		.op = VIRTIO_PCIDEV_OP_MMIO_WRITE,
386 		.bar = *resptr,
387 		.size = size,
388 		.addr = offset,
389 	};
390 
391 	um_pci_send_cmd(dev, &hdr, sizeof(hdr), buffer, size, NULL, 0);
392 }
393 
um_pci_bar_write(void * priv,unsigned int offset,int size,unsigned long val)394 static void um_pci_bar_write(void *priv, unsigned int offset, int size,
395 			     unsigned long val)
396 {
397 	/* maximum size - we may only use parts of it */
398 	u8 data[8];
399 
400 	switch (size) {
401 	case 1:
402 		data[0] = (u8)val;
403 		break;
404 	case 2:
405 		put_unaligned_le16(val, (void *)data);
406 		break;
407 	case 4:
408 		put_unaligned_le32(val, (void *)data);
409 		break;
410 #ifdef CONFIG_64BIT
411 	case 8:
412 		put_unaligned_le64(val, (void *)data);
413 		break;
414 #endif
415 	default:
416 		WARN(1, "invalid config space write size %d\n", size);
417 		return;
418 	}
419 
420 	um_pci_bar_copy_to(priv, offset, data, size);
421 }
422 
um_pci_bar_set(void * priv,unsigned int offset,u8 value,int size)423 static void um_pci_bar_set(void *priv, unsigned int offset, u8 value, int size)
424 {
425 	u8 *resptr = priv;
426 	struct um_pci_device *dev = container_of(resptr - *resptr,
427 						 struct um_pci_device,
428 						 resptr[0]);
429 	struct {
430 		struct virtio_pcidev_msg hdr;
431 		u8 data;
432 	} msg = {
433 		.hdr = {
434 			.op = VIRTIO_PCIDEV_OP_CFG_WRITE,
435 			.bar = *resptr,
436 			.size = size,
437 			.addr = offset,
438 		},
439 		.data = value,
440 	};
441 
442 	um_pci_send_cmd(dev, &msg.hdr, sizeof(msg), NULL, 0, NULL, 0);
443 }
444 
445 static const struct logic_iomem_ops um_pci_device_bar_ops = {
446 	.read = um_pci_bar_read,
447 	.write = um_pci_bar_write,
448 	.set = um_pci_bar_set,
449 	.copy_from = um_pci_bar_copy_from,
450 	.copy_to = um_pci_bar_copy_to,
451 };
452 
um_pci_map_bus(struct pci_bus * bus,unsigned int devfn,int where)453 static void __iomem *um_pci_map_bus(struct pci_bus *bus, unsigned int devfn,
454 				    int where)
455 {
456 	struct um_pci_device_reg *dev;
457 	unsigned int busn = bus->number;
458 
459 	if (busn > 0)
460 		return NULL;
461 
462 	/* not allowing functions for now ... */
463 	if (devfn % 8)
464 		return NULL;
465 
466 	if (devfn / 8 >= ARRAY_SIZE(um_pci_devices))
467 		return NULL;
468 
469 	dev = &um_pci_devices[devfn / 8];
470 	if (!dev)
471 		return NULL;
472 
473 	return (void __iomem *)((unsigned long)dev->iomem + where);
474 }
475 
476 static struct pci_ops um_pci_ops = {
477 	.map_bus = um_pci_map_bus,
478 	.read = pci_generic_config_read,
479 	.write = pci_generic_config_write,
480 };
481 
um_pci_rescan(void)482 static void um_pci_rescan(void)
483 {
484 	pci_lock_rescan_remove();
485 	pci_rescan_bus(bridge->bus);
486 	pci_unlock_rescan_remove();
487 }
488 
um_pci_irq_vq_addbuf(struct virtqueue * vq,void * buf,bool kick)489 static void um_pci_irq_vq_addbuf(struct virtqueue *vq, void *buf, bool kick)
490 {
491 	struct scatterlist sg[1];
492 
493 	sg_init_one(sg, buf, MAX_IRQ_MSG_SIZE);
494 	if (virtqueue_add_inbuf(vq, sg, 1, buf, GFP_ATOMIC))
495 		kfree(buf);
496 	else if (kick)
497 		virtqueue_kick(vq);
498 }
499 
um_pci_handle_irq_message(struct virtqueue * vq,struct virtio_pcidev_msg * msg)500 static void um_pci_handle_irq_message(struct virtqueue *vq,
501 				      struct virtio_pcidev_msg *msg)
502 {
503 	struct virtio_device *vdev = vq->vdev;
504 	struct um_pci_device *dev = vdev->priv;
505 
506 	if (!dev->irq)
507 		return;
508 
509 	/* we should properly chain interrupts, but on ARCH=um we don't care */
510 
511 	switch (msg->op) {
512 	case VIRTIO_PCIDEV_OP_INT:
513 		generic_handle_irq(dev->irq);
514 		break;
515 	case VIRTIO_PCIDEV_OP_MSI:
516 		/* our MSI message is just the interrupt number */
517 		if (msg->size == sizeof(u32))
518 			generic_handle_irq(le32_to_cpup((void *)msg->data));
519 		else
520 			generic_handle_irq(le16_to_cpup((void *)msg->data));
521 		break;
522 	case VIRTIO_PCIDEV_OP_PME:
523 		/* nothing to do - we already woke up due to the message */
524 		break;
525 	default:
526 		dev_err(&vdev->dev, "unexpected virt-pci message %d\n", msg->op);
527 		break;
528 	}
529 }
530 
um_pci_cmd_vq_cb(struct virtqueue * vq)531 static void um_pci_cmd_vq_cb(struct virtqueue *vq)
532 {
533 	struct virtio_device *vdev = vq->vdev;
534 	struct um_pci_device *dev = vdev->priv;
535 	void *cmd;
536 	int len;
537 
538 	if (test_bit(UM_PCI_STAT_WAITING, &dev->status))
539 		return;
540 
541 	while ((cmd = virtqueue_get_buf(vq, &len)))
542 		um_pci_free_buf(dev, cmd);
543 }
544 
um_pci_irq_vq_cb(struct virtqueue * vq)545 static void um_pci_irq_vq_cb(struct virtqueue *vq)
546 {
547 	struct virtio_pcidev_msg *msg;
548 	int len;
549 
550 	while ((msg = virtqueue_get_buf(vq, &len))) {
551 		if (len >= sizeof(*msg))
552 			um_pci_handle_irq_message(vq, msg);
553 
554 		/* recycle the message buffer */
555 		um_pci_irq_vq_addbuf(vq, msg, true);
556 	}
557 }
558 
559 #ifdef CONFIG_OF
560 /* Copied from arch/x86/kernel/devicetree.c */
pcibios_get_phb_of_node(struct pci_bus * bus)561 struct device_node *pcibios_get_phb_of_node(struct pci_bus *bus)
562 {
563 	struct device_node *np;
564 
565 	for_each_node_by_type(np, "pci") {
566 		const void *prop;
567 		unsigned int bus_min;
568 
569 		prop = of_get_property(np, "bus-range", NULL);
570 		if (!prop)
571 			continue;
572 		bus_min = be32_to_cpup(prop);
573 		if (bus->number == bus_min)
574 			return np;
575 	}
576 	return NULL;
577 }
578 #endif
579 
um_pci_init_vqs(struct um_pci_device * dev)580 static int um_pci_init_vqs(struct um_pci_device *dev)
581 {
582 	struct virtqueue_info vqs_info[] = {
583 		{ "cmd", um_pci_cmd_vq_cb },
584 		{ "irq", um_pci_irq_vq_cb },
585 	};
586 	struct virtqueue *vqs[2];
587 	int err, i;
588 
589 	err = virtio_find_vqs(dev->vdev, 2, vqs, vqs_info, NULL);
590 	if (err)
591 		return err;
592 
593 	dev->cmd_vq = vqs[0];
594 	dev->irq_vq = vqs[1];
595 
596 	virtio_device_ready(dev->vdev);
597 
598 	for (i = 0; i < NUM_IRQ_MSGS; i++) {
599 		void *msg = kzalloc(MAX_IRQ_MSG_SIZE, GFP_KERNEL);
600 
601 		if (msg)
602 			um_pci_irq_vq_addbuf(dev->irq_vq, msg, false);
603 	}
604 
605 	virtqueue_kick(dev->irq_vq);
606 
607 	return 0;
608 }
609 
__um_pci_virtio_platform_remove(struct virtio_device * vdev,struct um_pci_device * dev)610 static void __um_pci_virtio_platform_remove(struct virtio_device *vdev,
611 					    struct um_pci_device *dev)
612 {
613 	virtio_reset_device(vdev);
614 	vdev->config->del_vqs(vdev);
615 
616 	mutex_lock(&um_pci_mtx);
617 	um_pci_platform_device = NULL;
618 	mutex_unlock(&um_pci_mtx);
619 
620 	kfree(dev);
621 }
622 
um_pci_virtio_platform_probe(struct virtio_device * vdev,struct um_pci_device * dev)623 static int um_pci_virtio_platform_probe(struct virtio_device *vdev,
624 					struct um_pci_device *dev)
625 {
626 	int ret;
627 
628 	dev->platform = true;
629 
630 	mutex_lock(&um_pci_mtx);
631 
632 	if (um_pci_platform_device) {
633 		mutex_unlock(&um_pci_mtx);
634 		ret = -EBUSY;
635 		goto out_free;
636 	}
637 
638 	ret = um_pci_init_vqs(dev);
639 	if (ret) {
640 		mutex_unlock(&um_pci_mtx);
641 		goto out_free;
642 	}
643 
644 	um_pci_platform_device = dev;
645 
646 	mutex_unlock(&um_pci_mtx);
647 
648 	ret = of_platform_default_populate(vdev->dev.of_node, NULL, &vdev->dev);
649 	if (ret)
650 		__um_pci_virtio_platform_remove(vdev, dev);
651 
652 	return ret;
653 
654 out_free:
655 	kfree(dev);
656 	return ret;
657 }
658 
um_pci_virtio_probe(struct virtio_device * vdev)659 static int um_pci_virtio_probe(struct virtio_device *vdev)
660 {
661 	struct um_pci_device *dev;
662 	int i, free = -1;
663 	int err = -ENOSPC;
664 
665 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
666 	if (!dev)
667 		return -ENOMEM;
668 
669 	dev->vdev = vdev;
670 	vdev->priv = dev;
671 
672 	if (of_device_is_compatible(vdev->dev.of_node, "simple-bus"))
673 		return um_pci_virtio_platform_probe(vdev, dev);
674 
675 	mutex_lock(&um_pci_mtx);
676 	for (i = 0; i < MAX_DEVICES; i++) {
677 		if (um_pci_devices[i].dev)
678 			continue;
679 		free = i;
680 		break;
681 	}
682 
683 	if (free < 0)
684 		goto error;
685 
686 	err = um_pci_init_vqs(dev);
687 	if (err)
688 		goto error;
689 
690 	dev->irq = irq_alloc_desc(numa_node_id());
691 	if (dev->irq < 0) {
692 		err = dev->irq;
693 		goto err_reset;
694 	}
695 	um_pci_devices[free].dev = dev;
696 	vdev->priv = dev;
697 
698 	mutex_unlock(&um_pci_mtx);
699 
700 	device_set_wakeup_enable(&vdev->dev, true);
701 
702 	/*
703 	 * In order to do suspend-resume properly, don't allow VQs
704 	 * to be suspended.
705 	 */
706 	virtio_uml_set_no_vq_suspend(vdev, true);
707 
708 	um_pci_rescan();
709 	return 0;
710 err_reset:
711 	virtio_reset_device(vdev);
712 	vdev->config->del_vqs(vdev);
713 error:
714 	mutex_unlock(&um_pci_mtx);
715 	kfree(dev);
716 	return err;
717 }
718 
um_pci_virtio_remove(struct virtio_device * vdev)719 static void um_pci_virtio_remove(struct virtio_device *vdev)
720 {
721 	struct um_pci_device *dev = vdev->priv;
722 	int i;
723 
724 	if (dev->platform) {
725 		of_platform_depopulate(&vdev->dev);
726 		__um_pci_virtio_platform_remove(vdev, dev);
727 		return;
728 	}
729 
730 	device_set_wakeup_enable(&vdev->dev, false);
731 
732 	mutex_lock(&um_pci_mtx);
733 	for (i = 0; i < MAX_DEVICES; i++) {
734 		if (um_pci_devices[i].dev != dev)
735 			continue;
736 
737 		um_pci_devices[i].dev = NULL;
738 		irq_free_desc(dev->irq);
739 
740 		break;
741 	}
742 	mutex_unlock(&um_pci_mtx);
743 
744 	if (i < MAX_DEVICES) {
745 		struct pci_dev *pci_dev;
746 
747 		pci_dev = pci_get_slot(bridge->bus, i);
748 		if (pci_dev)
749 			pci_stop_and_remove_bus_device_locked(pci_dev);
750 	}
751 
752 	/* Stop all virtqueues */
753 	virtio_reset_device(vdev);
754 	dev->cmd_vq = NULL;
755 	dev->irq_vq = NULL;
756 	vdev->config->del_vqs(vdev);
757 
758 	kfree(dev);
759 }
760 
761 static struct virtio_device_id id_table[] = {
762 	{ CONFIG_UML_PCI_OVER_VIRTIO_DEVICE_ID, VIRTIO_DEV_ANY_ID },
763 	{ 0 },
764 };
765 MODULE_DEVICE_TABLE(virtio, id_table);
766 
767 static struct virtio_driver um_pci_virtio_driver = {
768 	.driver.name = "virtio-pci",
769 	.id_table = id_table,
770 	.probe = um_pci_virtio_probe,
771 	.remove = um_pci_virtio_remove,
772 };
773 
774 static struct resource virt_cfgspace_resource = {
775 	.name = "PCI config space",
776 	.start = 0xf0000000 - MAX_DEVICES * CFG_SPACE_SIZE,
777 	.end = 0xf0000000 - 1,
778 	.flags = IORESOURCE_MEM,
779 };
780 
um_pci_map_cfgspace(unsigned long offset,size_t size,const struct logic_iomem_ops ** ops,void ** priv)781 static long um_pci_map_cfgspace(unsigned long offset, size_t size,
782 				const struct logic_iomem_ops **ops,
783 				void **priv)
784 {
785 	if (WARN_ON(size > CFG_SPACE_SIZE || offset % CFG_SPACE_SIZE))
786 		return -EINVAL;
787 
788 	if (offset / CFG_SPACE_SIZE < MAX_DEVICES) {
789 		*ops = &um_pci_device_cfgspace_ops;
790 		*priv = &um_pci_devices[offset / CFG_SPACE_SIZE];
791 		return 0;
792 	}
793 
794 	WARN(1, "cannot map offset 0x%lx/0x%zx\n", offset, size);
795 	return -ENOENT;
796 }
797 
798 static const struct logic_iomem_region_ops um_pci_cfgspace_ops = {
799 	.map = um_pci_map_cfgspace,
800 };
801 
802 static struct resource virt_iomem_resource = {
803 	.name = "PCI iomem",
804 	.start = 0xf0000000,
805 	.end = 0xffffffff,
806 	.flags = IORESOURCE_MEM,
807 };
808 
809 struct um_pci_map_iomem_data {
810 	unsigned long offset;
811 	size_t size;
812 	const struct logic_iomem_ops **ops;
813 	void **priv;
814 	long ret;
815 };
816 
um_pci_map_iomem_walk(struct pci_dev * pdev,void * _data)817 static int um_pci_map_iomem_walk(struct pci_dev *pdev, void *_data)
818 {
819 	struct um_pci_map_iomem_data *data = _data;
820 	struct um_pci_device_reg *reg = &um_pci_devices[pdev->devfn / 8];
821 	struct um_pci_device *dev;
822 	int i;
823 
824 	if (!reg->dev)
825 		return 0;
826 
827 	for (i = 0; i < ARRAY_SIZE(dev->resptr); i++) {
828 		struct resource *r = &pdev->resource[i];
829 
830 		if ((r->flags & IORESOURCE_TYPE_BITS) != IORESOURCE_MEM)
831 			continue;
832 
833 		/*
834 		 * must be the whole or part of the resource,
835 		 * not allowed to only overlap
836 		 */
837 		if (data->offset < r->start || data->offset > r->end)
838 			continue;
839 		if (data->offset + data->size - 1 > r->end)
840 			continue;
841 
842 		dev = reg->dev;
843 		*data->ops = &um_pci_device_bar_ops;
844 		dev->resptr[i] = i;
845 		*data->priv = &dev->resptr[i];
846 		data->ret = data->offset - r->start;
847 
848 		/* no need to continue */
849 		return 1;
850 	}
851 
852 	return 0;
853 }
854 
um_pci_map_iomem(unsigned long offset,size_t size,const struct logic_iomem_ops ** ops,void ** priv)855 static long um_pci_map_iomem(unsigned long offset, size_t size,
856 			     const struct logic_iomem_ops **ops,
857 			     void **priv)
858 {
859 	struct um_pci_map_iomem_data data = {
860 		/* we want the full address here */
861 		.offset = offset + virt_iomem_resource.start,
862 		.size = size,
863 		.ops = ops,
864 		.priv = priv,
865 		.ret = -ENOENT,
866 	};
867 
868 	pci_walk_bus(bridge->bus, um_pci_map_iomem_walk, &data);
869 	return data.ret;
870 }
871 
872 static const struct logic_iomem_region_ops um_pci_iomem_ops = {
873 	.map = um_pci_map_iomem,
874 };
875 
um_pci_compose_msi_msg(struct irq_data * data,struct msi_msg * msg)876 static void um_pci_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
877 {
878 	/*
879 	 * This is a very low address and not actually valid 'physical' memory
880 	 * in UML, so we can simply map MSI(-X) vectors to there, it cannot be
881 	 * legitimately written to by the device in any other way.
882 	 * We use the (virtual) IRQ number here as the message to simplify the
883 	 * code that receives the message, where for now we simply trust the
884 	 * device to send the correct message.
885 	 */
886 	msg->address_hi = 0;
887 	msg->address_lo = 0xa0000;
888 	msg->data = data->irq;
889 }
890 
891 static struct irq_chip um_pci_msi_bottom_irq_chip = {
892 	.name = "UM virtio MSI",
893 	.irq_compose_msi_msg = um_pci_compose_msi_msg,
894 };
895 
um_pci_inner_domain_alloc(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs,void * args)896 static int um_pci_inner_domain_alloc(struct irq_domain *domain,
897 				     unsigned int virq, unsigned int nr_irqs,
898 				     void *args)
899 {
900 	unsigned long bit;
901 
902 	WARN_ON(nr_irqs != 1);
903 
904 	mutex_lock(&um_pci_mtx);
905 	bit = find_first_zero_bit(um_pci_msi_used, MAX_MSI_VECTORS);
906 	if (bit >= MAX_MSI_VECTORS) {
907 		mutex_unlock(&um_pci_mtx);
908 		return -ENOSPC;
909 	}
910 
911 	set_bit(bit, um_pci_msi_used);
912 	mutex_unlock(&um_pci_mtx);
913 
914 	irq_domain_set_info(domain, virq, bit, &um_pci_msi_bottom_irq_chip,
915 			    domain->host_data, handle_simple_irq,
916 			    NULL, NULL);
917 
918 	return 0;
919 }
920 
um_pci_inner_domain_free(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs)921 static void um_pci_inner_domain_free(struct irq_domain *domain,
922 				     unsigned int virq, unsigned int nr_irqs)
923 {
924 	struct irq_data *d = irq_domain_get_irq_data(domain, virq);
925 
926 	mutex_lock(&um_pci_mtx);
927 
928 	if (!test_bit(d->hwirq, um_pci_msi_used))
929 		pr_err("trying to free unused MSI#%lu\n", d->hwirq);
930 	else
931 		__clear_bit(d->hwirq, um_pci_msi_used);
932 
933 	mutex_unlock(&um_pci_mtx);
934 }
935 
936 static const struct irq_domain_ops um_pci_inner_domain_ops = {
937 	.alloc = um_pci_inner_domain_alloc,
938 	.free = um_pci_inner_domain_free,
939 };
940 
941 static struct irq_chip um_pci_msi_irq_chip = {
942 	.name = "UM virtio PCIe MSI",
943 	.irq_mask = pci_msi_mask_irq,
944 	.irq_unmask = pci_msi_unmask_irq,
945 };
946 
947 static struct msi_domain_info um_pci_msi_domain_info = {
948 	.flags	= MSI_FLAG_USE_DEF_DOM_OPS |
949 		  MSI_FLAG_USE_DEF_CHIP_OPS |
950 		  MSI_FLAG_PCI_MSIX,
951 	.chip	= &um_pci_msi_irq_chip,
952 };
953 
954 static struct resource busn_resource = {
955 	.name	= "PCI busn",
956 	.start	= 0,
957 	.end	= 0,
958 	.flags	= IORESOURCE_BUS,
959 };
960 
um_pci_map_irq(const struct pci_dev * pdev,u8 slot,u8 pin)961 static int um_pci_map_irq(const struct pci_dev *pdev, u8 slot, u8 pin)
962 {
963 	struct um_pci_device_reg *reg = &um_pci_devices[pdev->devfn / 8];
964 
965 	if (WARN_ON(!reg->dev))
966 		return -EINVAL;
967 
968 	/* Yes, we map all pins to the same IRQ ... doesn't matter for now. */
969 	return reg->dev->irq;
970 }
971 
pci_root_bus_fwnode(struct pci_bus * bus)972 void *pci_root_bus_fwnode(struct pci_bus *bus)
973 {
974 	return um_pci_fwnode;
975 }
976 
um_pci_map_platform(unsigned long offset,size_t size,const struct logic_iomem_ops ** ops,void ** priv)977 static long um_pci_map_platform(unsigned long offset, size_t size,
978 				const struct logic_iomem_ops **ops,
979 				void **priv)
980 {
981 	if (!um_pci_platform_device)
982 		return -ENOENT;
983 
984 	*ops = &um_pci_device_bar_ops;
985 	*priv = &um_pci_platform_device->resptr[0];
986 
987 	return offset;
988 }
989 
990 static const struct logic_iomem_region_ops um_pci_platform_ops = {
991 	.map = um_pci_map_platform,
992 };
993 
994 static struct resource virt_platform_resource = {
995 	.name = "platform",
996 	.start = 0x10000000,
997 	.end = 0x1fffffff,
998 	.flags = IORESOURCE_MEM,
999 };
1000 
um_pci_init(void)1001 static int __init um_pci_init(void)
1002 {
1003 	struct irq_domain_info inner_domain_info = {
1004 		.size		= MAX_MSI_VECTORS,
1005 		.hwirq_max	= MAX_MSI_VECTORS,
1006 		.ops		= &um_pci_inner_domain_ops,
1007 	};
1008 	int err, i;
1009 
1010 	WARN_ON(logic_iomem_add_region(&virt_cfgspace_resource,
1011 				       &um_pci_cfgspace_ops));
1012 	WARN_ON(logic_iomem_add_region(&virt_iomem_resource,
1013 				       &um_pci_iomem_ops));
1014 	WARN_ON(logic_iomem_add_region(&virt_platform_resource,
1015 				       &um_pci_platform_ops));
1016 
1017 	if (WARN(CONFIG_UML_PCI_OVER_VIRTIO_DEVICE_ID < 0,
1018 		 "No virtio device ID configured for PCI - no PCI support\n"))
1019 		return 0;
1020 
1021 	bridge = pci_alloc_host_bridge(0);
1022 	if (!bridge) {
1023 		err = -ENOMEM;
1024 		goto free;
1025 	}
1026 
1027 	um_pci_fwnode = irq_domain_alloc_named_fwnode("um-pci");
1028 	if (!um_pci_fwnode) {
1029 		err = -ENOMEM;
1030 		goto free;
1031 	}
1032 
1033 	inner_domain_info.fwnode = um_pci_fwnode;
1034 	um_pci_inner_domain = irq_domain_instantiate(&inner_domain_info);
1035 	if (IS_ERR(um_pci_inner_domain)) {
1036 		err = PTR_ERR(um_pci_inner_domain);
1037 		goto free;
1038 	}
1039 
1040 	um_pci_msi_domain = pci_msi_create_irq_domain(um_pci_fwnode,
1041 						      &um_pci_msi_domain_info,
1042 						      um_pci_inner_domain);
1043 	if (!um_pci_msi_domain) {
1044 		err = -ENOMEM;
1045 		goto free;
1046 	}
1047 
1048 	pci_add_resource(&bridge->windows, &virt_iomem_resource);
1049 	pci_add_resource(&bridge->windows, &busn_resource);
1050 	bridge->ops = &um_pci_ops;
1051 	bridge->map_irq = um_pci_map_irq;
1052 
1053 	for (i = 0; i < MAX_DEVICES; i++) {
1054 		resource_size_t start;
1055 
1056 		start = virt_cfgspace_resource.start + i * CFG_SPACE_SIZE;
1057 		um_pci_devices[i].iomem = ioremap(start, CFG_SPACE_SIZE);
1058 		if (WARN(!um_pci_devices[i].iomem, "failed to map %d\n", i)) {
1059 			err = -ENOMEM;
1060 			goto free;
1061 		}
1062 	}
1063 
1064 	err = pci_host_probe(bridge);
1065 	if (err)
1066 		goto free;
1067 
1068 	err = register_virtio_driver(&um_pci_virtio_driver);
1069 	if (err)
1070 		goto free;
1071 	return 0;
1072 free:
1073 	if (!IS_ERR_OR_NULL(um_pci_inner_domain))
1074 		irq_domain_remove(um_pci_inner_domain);
1075 	if (um_pci_fwnode)
1076 		irq_domain_free_fwnode(um_pci_fwnode);
1077 	if (bridge) {
1078 		pci_free_resource_list(&bridge->windows);
1079 		pci_free_host_bridge(bridge);
1080 	}
1081 	return err;
1082 }
1083 module_init(um_pci_init);
1084 
um_pci_exit(void)1085 static void __exit um_pci_exit(void)
1086 {
1087 	unregister_virtio_driver(&um_pci_virtio_driver);
1088 	irq_domain_remove(um_pci_msi_domain);
1089 	irq_domain_remove(um_pci_inner_domain);
1090 	pci_free_resource_list(&bridge->windows);
1091 	pci_free_host_bridge(bridge);
1092 }
1093 module_exit(um_pci_exit);
1094