xref: /openbmc/linux/drivers/pci/xen-pcifront.c (revision 800e26b8)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Xen PCI Frontend
4  *
5  * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
6  */
7 #include <linux/module.h>
8 #include <linux/init.h>
9 #include <linux/mm.h>
10 #include <xen/xenbus.h>
11 #include <xen/events.h>
12 #include <xen/grant_table.h>
13 #include <xen/page.h>
14 #include <linux/spinlock.h>
15 #include <linux/pci.h>
16 #include <linux/msi.h>
17 #include <xen/interface/io/pciif.h>
18 #include <asm/xen/pci.h>
19 #include <linux/interrupt.h>
20 #include <linux/atomic.h>
21 #include <linux/workqueue.h>
22 #include <linux/bitops.h>
23 #include <linux/time.h>
24 #include <linux/ktime.h>
25 #include <xen/platform_pci.h>
26 
27 #include <asm/xen/swiotlb-xen.h>
28 #define INVALID_GRANT_REF (0)
29 #define INVALID_EVTCHN    (-1)
30 
31 struct pci_bus_entry {
32 	struct list_head list;
33 	struct pci_bus *bus;
34 };
35 
36 #define _PDEVB_op_active		(0)
37 #define PDEVB_op_active			(1 << (_PDEVB_op_active))
38 
39 struct pcifront_device {
40 	struct xenbus_device *xdev;
41 	struct list_head root_buses;
42 
43 	int evtchn;
44 	int gnt_ref;
45 
46 	int irq;
47 
48 	/* Lock this when doing any operations in sh_info */
49 	spinlock_t sh_info_lock;
50 	struct xen_pci_sharedinfo *sh_info;
51 	struct work_struct op_work;
52 	unsigned long flags;
53 
54 };
55 
56 struct pcifront_sd {
57 	struct pci_sysdata sd;
58 	struct pcifront_device *pdev;
59 };
60 
61 static inline struct pcifront_device *
62 pcifront_get_pdev(struct pcifront_sd *sd)
63 {
64 	return sd->pdev;
65 }
66 
67 static inline void pcifront_init_sd(struct pcifront_sd *sd,
68 				    unsigned int domain, unsigned int bus,
69 				    struct pcifront_device *pdev)
70 {
71 	/* Because we do not expose that information via XenBus. */
72 	sd->sd.node = first_online_node;
73 	sd->sd.domain = domain;
74 	sd->pdev = pdev;
75 }
76 
77 static DEFINE_SPINLOCK(pcifront_dev_lock);
78 static struct pcifront_device *pcifront_dev;
79 
80 static int errno_to_pcibios_err(int errno)
81 {
82 	switch (errno) {
83 	case XEN_PCI_ERR_success:
84 		return PCIBIOS_SUCCESSFUL;
85 
86 	case XEN_PCI_ERR_dev_not_found:
87 		return PCIBIOS_DEVICE_NOT_FOUND;
88 
89 	case XEN_PCI_ERR_invalid_offset:
90 	case XEN_PCI_ERR_op_failed:
91 		return PCIBIOS_BAD_REGISTER_NUMBER;
92 
93 	case XEN_PCI_ERR_not_implemented:
94 		return PCIBIOS_FUNC_NOT_SUPPORTED;
95 
96 	case XEN_PCI_ERR_access_denied:
97 		return PCIBIOS_SET_FAILED;
98 	}
99 	return errno;
100 }
101 
102 static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
103 {
104 	if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
105 		&& !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
106 		dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
107 		schedule_work(&pdev->op_work);
108 	}
109 }
110 
111 static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
112 {
113 	int err = 0;
114 	struct xen_pci_op *active_op = &pdev->sh_info->op;
115 	unsigned long irq_flags;
116 	evtchn_port_t port = pdev->evtchn;
117 	unsigned irq = pdev->irq;
118 	s64 ns, ns_timeout;
119 
120 	spin_lock_irqsave(&pdev->sh_info_lock, irq_flags);
121 
122 	memcpy(active_op, op, sizeof(struct xen_pci_op));
123 
124 	/* Go */
125 	wmb();
126 	set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
127 	notify_remote_via_evtchn(port);
128 
129 	/*
130 	 * We set a poll timeout of 3 seconds but give up on return after
131 	 * 2 seconds. It is better to time out too late rather than too early
132 	 * (in the latter case we end up continually re-executing poll() with a
133 	 * timeout in the past). 1s difference gives plenty of slack for error.
134 	 */
135 	ns_timeout = ktime_get_ns() + 2 * (s64)NSEC_PER_SEC;
136 
137 	xen_clear_irq_pending(irq);
138 
139 	while (test_bit(_XEN_PCIF_active,
140 			(unsigned long *)&pdev->sh_info->flags)) {
141 		xen_poll_irq_timeout(irq, jiffies + 3*HZ);
142 		xen_clear_irq_pending(irq);
143 		ns = ktime_get_ns();
144 		if (ns > ns_timeout) {
145 			dev_err(&pdev->xdev->dev,
146 				"pciback not responding!!!\n");
147 			clear_bit(_XEN_PCIF_active,
148 				  (unsigned long *)&pdev->sh_info->flags);
149 			err = XEN_PCI_ERR_dev_not_found;
150 			goto out;
151 		}
152 	}
153 
154 	/*
155 	* We might lose backend service request since we
156 	* reuse same evtchn with pci_conf backend response. So re-schedule
157 	* aer pcifront service.
158 	*/
159 	if (test_bit(_XEN_PCIB_active,
160 			(unsigned long *)&pdev->sh_info->flags)) {
161 		dev_err(&pdev->xdev->dev,
162 			"schedule aer pcifront service\n");
163 		schedule_pcifront_aer_op(pdev);
164 	}
165 
166 	memcpy(op, active_op, sizeof(struct xen_pci_op));
167 
168 	err = op->err;
169 out:
170 	spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags);
171 	return err;
172 }
173 
174 /* Access to this function is spinlocked in drivers/pci/access.c */
175 static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn,
176 			     int where, int size, u32 *val)
177 {
178 	int err = 0;
179 	struct xen_pci_op op = {
180 		.cmd    = XEN_PCI_OP_conf_read,
181 		.domain = pci_domain_nr(bus),
182 		.bus    = bus->number,
183 		.devfn  = devfn,
184 		.offset = where,
185 		.size   = size,
186 	};
187 	struct pcifront_sd *sd = bus->sysdata;
188 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
189 
190 	dev_dbg(&pdev->xdev->dev,
191 		"read dev=%04x:%02x:%02x.%d - offset %x size %d\n",
192 		pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
193 		PCI_FUNC(devfn), where, size);
194 
195 	err = do_pci_op(pdev, &op);
196 
197 	if (likely(!err)) {
198 		dev_dbg(&pdev->xdev->dev, "read got back value %x\n",
199 			op.value);
200 
201 		*val = op.value;
202 	} else if (err == -ENODEV) {
203 		/* No device here, pretend that it just returned 0 */
204 		err = 0;
205 		*val = 0;
206 	}
207 
208 	return errno_to_pcibios_err(err);
209 }
210 
211 /* Access to this function is spinlocked in drivers/pci/access.c */
212 static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
213 			      int where, int size, u32 val)
214 {
215 	struct xen_pci_op op = {
216 		.cmd    = XEN_PCI_OP_conf_write,
217 		.domain = pci_domain_nr(bus),
218 		.bus    = bus->number,
219 		.devfn  = devfn,
220 		.offset = where,
221 		.size   = size,
222 		.value  = val,
223 	};
224 	struct pcifront_sd *sd = bus->sysdata;
225 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
226 
227 	dev_dbg(&pdev->xdev->dev,
228 		"write dev=%04x:%02x:%02x.%d - offset %x size %d val %x\n",
229 		pci_domain_nr(bus), bus->number,
230 		PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
231 
232 	return errno_to_pcibios_err(do_pci_op(pdev, &op));
233 }
234 
235 static struct pci_ops pcifront_bus_ops = {
236 	.read = pcifront_bus_read,
237 	.write = pcifront_bus_write,
238 };
239 
240 #ifdef CONFIG_PCI_MSI
241 static int pci_frontend_enable_msix(struct pci_dev *dev,
242 				    int vector[], int nvec)
243 {
244 	int err;
245 	int i;
246 	struct xen_pci_op op = {
247 		.cmd    = XEN_PCI_OP_enable_msix,
248 		.domain = pci_domain_nr(dev->bus),
249 		.bus = dev->bus->number,
250 		.devfn = dev->devfn,
251 		.value = nvec,
252 	};
253 	struct pcifront_sd *sd = dev->bus->sysdata;
254 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
255 	struct msi_desc *entry;
256 
257 	if (nvec > SH_INFO_MAX_VEC) {
258 		pci_err(dev, "too many vectors (0x%x) for PCI frontend:"
259 				   " Increase SH_INFO_MAX_VEC\n", nvec);
260 		return -EINVAL;
261 	}
262 
263 	i = 0;
264 	for_each_pci_msi_entry(entry, dev) {
265 		op.msix_entries[i].entry = entry->msi_attrib.entry_nr;
266 		/* Vector is useless at this point. */
267 		op.msix_entries[i].vector = -1;
268 		i++;
269 	}
270 
271 	err = do_pci_op(pdev, &op);
272 
273 	if (likely(!err)) {
274 		if (likely(!op.value)) {
275 			/* we get the result */
276 			for (i = 0; i < nvec; i++) {
277 				if (op.msix_entries[i].vector <= 0) {
278 					pci_warn(dev, "MSI-X entry %d is invalid: %d!\n",
279 						i, op.msix_entries[i].vector);
280 					err = -EINVAL;
281 					vector[i] = -1;
282 					continue;
283 				}
284 				vector[i] = op.msix_entries[i].vector;
285 			}
286 		} else {
287 			pr_info("enable msix get value %x\n", op.value);
288 			err = op.value;
289 		}
290 	} else {
291 		pci_err(dev, "enable msix get err %x\n", err);
292 	}
293 	return err;
294 }
295 
296 static void pci_frontend_disable_msix(struct pci_dev *dev)
297 {
298 	int err;
299 	struct xen_pci_op op = {
300 		.cmd    = XEN_PCI_OP_disable_msix,
301 		.domain = pci_domain_nr(dev->bus),
302 		.bus = dev->bus->number,
303 		.devfn = dev->devfn,
304 	};
305 	struct pcifront_sd *sd = dev->bus->sysdata;
306 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
307 
308 	err = do_pci_op(pdev, &op);
309 
310 	/* What should do for error ? */
311 	if (err)
312 		pci_err(dev, "pci_disable_msix get err %x\n", err);
313 }
314 
315 static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[])
316 {
317 	int err;
318 	struct xen_pci_op op = {
319 		.cmd    = XEN_PCI_OP_enable_msi,
320 		.domain = pci_domain_nr(dev->bus),
321 		.bus = dev->bus->number,
322 		.devfn = dev->devfn,
323 	};
324 	struct pcifront_sd *sd = dev->bus->sysdata;
325 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
326 
327 	err = do_pci_op(pdev, &op);
328 	if (likely(!err)) {
329 		vector[0] = op.value;
330 		if (op.value <= 0) {
331 			pci_warn(dev, "MSI entry is invalid: %d!\n",
332 				op.value);
333 			err = -EINVAL;
334 			vector[0] = -1;
335 		}
336 	} else {
337 		pci_err(dev, "pci frontend enable msi failed for dev "
338 				    "%x:%x\n", op.bus, op.devfn);
339 		err = -EINVAL;
340 	}
341 	return err;
342 }
343 
344 static void pci_frontend_disable_msi(struct pci_dev *dev)
345 {
346 	int err;
347 	struct xen_pci_op op = {
348 		.cmd    = XEN_PCI_OP_disable_msi,
349 		.domain = pci_domain_nr(dev->bus),
350 		.bus = dev->bus->number,
351 		.devfn = dev->devfn,
352 	};
353 	struct pcifront_sd *sd = dev->bus->sysdata;
354 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
355 
356 	err = do_pci_op(pdev, &op);
357 	if (err == XEN_PCI_ERR_dev_not_found) {
358 		/* XXX No response from backend, what shall we do? */
359 		pr_info("get no response from backend for disable MSI\n");
360 		return;
361 	}
362 	if (err)
363 		/* how can pciback notify us fail? */
364 		pr_info("get fake response from backend\n");
365 }
366 
367 static struct xen_pci_frontend_ops pci_frontend_ops = {
368 	.enable_msi = pci_frontend_enable_msi,
369 	.disable_msi = pci_frontend_disable_msi,
370 	.enable_msix = pci_frontend_enable_msix,
371 	.disable_msix = pci_frontend_disable_msix,
372 };
373 
374 static void pci_frontend_registrar(int enable)
375 {
376 	if (enable)
377 		xen_pci_frontend = &pci_frontend_ops;
378 	else
379 		xen_pci_frontend = NULL;
380 };
381 #else
382 static inline void pci_frontend_registrar(int enable) { };
383 #endif /* CONFIG_PCI_MSI */
384 
385 /* Claim resources for the PCI frontend as-is, backend won't allow changes */
386 static int pcifront_claim_resource(struct pci_dev *dev, void *data)
387 {
388 	struct pcifront_device *pdev = data;
389 	int i;
390 	struct resource *r;
391 
392 	for (i = 0; i < PCI_NUM_RESOURCES; i++) {
393 		r = &dev->resource[i];
394 
395 		if (!r->parent && r->start && r->flags) {
396 			dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n",
397 				pci_name(dev), i);
398 			if (pci_claim_resource(dev, i)) {
399 				dev_err(&pdev->xdev->dev, "Could not claim resource %s/%d! "
400 					"Device offline. Try using e820_host=1 in the guest config.\n",
401 					pci_name(dev), i);
402 			}
403 		}
404 	}
405 
406 	return 0;
407 }
408 
409 static int pcifront_scan_bus(struct pcifront_device *pdev,
410 				unsigned int domain, unsigned int bus,
411 				struct pci_bus *b)
412 {
413 	struct pci_dev *d;
414 	unsigned int devfn;
415 
416 	/* Scan the bus for functions and add.
417 	 * We omit handling of PCI bridge attachment because pciback prevents
418 	 * bridges from being exported.
419 	 */
420 	for (devfn = 0; devfn < 0x100; devfn++) {
421 		d = pci_get_slot(b, devfn);
422 		if (d) {
423 			/* Device is already known. */
424 			pci_dev_put(d);
425 			continue;
426 		}
427 
428 		d = pci_scan_single_device(b, devfn);
429 		if (d)
430 			dev_info(&pdev->xdev->dev, "New device on "
431 				 "%04x:%02x:%02x.%d found.\n", domain, bus,
432 				 PCI_SLOT(devfn), PCI_FUNC(devfn));
433 	}
434 
435 	return 0;
436 }
437 
438 static int pcifront_scan_root(struct pcifront_device *pdev,
439 				 unsigned int domain, unsigned int bus)
440 {
441 	struct pci_bus *b;
442 	LIST_HEAD(resources);
443 	struct pcifront_sd *sd = NULL;
444 	struct pci_bus_entry *bus_entry = NULL;
445 	int err = 0;
446 	static struct resource busn_res = {
447 		.start = 0,
448 		.end = 255,
449 		.flags = IORESOURCE_BUS,
450 	};
451 
452 #ifndef CONFIG_PCI_DOMAINS
453 	if (domain != 0) {
454 		dev_err(&pdev->xdev->dev,
455 			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
456 		dev_err(&pdev->xdev->dev,
457 			"Please compile with CONFIG_PCI_DOMAINS\n");
458 		err = -EINVAL;
459 		goto err_out;
460 	}
461 #endif
462 
463 	dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n",
464 		 domain, bus);
465 
466 	bus_entry = kzalloc(sizeof(*bus_entry), GFP_KERNEL);
467 	sd = kzalloc(sizeof(*sd), GFP_KERNEL);
468 	if (!bus_entry || !sd) {
469 		err = -ENOMEM;
470 		goto err_out;
471 	}
472 	pci_add_resource(&resources, &ioport_resource);
473 	pci_add_resource(&resources, &iomem_resource);
474 	pci_add_resource(&resources, &busn_res);
475 	pcifront_init_sd(sd, domain, bus, pdev);
476 
477 	pci_lock_rescan_remove();
478 
479 	b = pci_scan_root_bus(&pdev->xdev->dev, bus,
480 				  &pcifront_bus_ops, sd, &resources);
481 	if (!b) {
482 		dev_err(&pdev->xdev->dev,
483 			"Error creating PCI Frontend Bus!\n");
484 		err = -ENOMEM;
485 		pci_unlock_rescan_remove();
486 		pci_free_resource_list(&resources);
487 		goto err_out;
488 	}
489 
490 	bus_entry->bus = b;
491 
492 	list_add(&bus_entry->list, &pdev->root_buses);
493 
494 	/* pci_scan_root_bus skips devices which do not have a
495 	* devfn==0. The pcifront_scan_bus enumerates all devfn. */
496 	err = pcifront_scan_bus(pdev, domain, bus, b);
497 
498 	/* Claim resources before going "live" with our devices */
499 	pci_walk_bus(b, pcifront_claim_resource, pdev);
500 
501 	/* Create SysFS and notify udev of the devices. Aka: "going live" */
502 	pci_bus_add_devices(b);
503 
504 	pci_unlock_rescan_remove();
505 	return err;
506 
507 err_out:
508 	kfree(bus_entry);
509 	kfree(sd);
510 
511 	return err;
512 }
513 
514 static int pcifront_rescan_root(struct pcifront_device *pdev,
515 				   unsigned int domain, unsigned int bus)
516 {
517 	int err;
518 	struct pci_bus *b;
519 
520 #ifndef CONFIG_PCI_DOMAINS
521 	if (domain != 0) {
522 		dev_err(&pdev->xdev->dev,
523 			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
524 		dev_err(&pdev->xdev->dev,
525 			"Please compile with CONFIG_PCI_DOMAINS\n");
526 		return -EINVAL;
527 	}
528 #endif
529 
530 	dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
531 		 domain, bus);
532 
533 	b = pci_find_bus(domain, bus);
534 	if (!b)
535 		/* If the bus is unknown, create it. */
536 		return pcifront_scan_root(pdev, domain, bus);
537 
538 	err = pcifront_scan_bus(pdev, domain, bus, b);
539 
540 	/* Claim resources before going "live" with our devices */
541 	pci_walk_bus(b, pcifront_claim_resource, pdev);
542 
543 	/* Create SysFS and notify udev of the devices. Aka: "going live" */
544 	pci_bus_add_devices(b);
545 
546 	return err;
547 }
548 
549 static void free_root_bus_devs(struct pci_bus *bus)
550 {
551 	struct pci_dev *dev;
552 
553 	while (!list_empty(&bus->devices)) {
554 		dev = container_of(bus->devices.next, struct pci_dev,
555 				   bus_list);
556 		pci_dbg(dev, "removing device\n");
557 		pci_stop_and_remove_bus_device(dev);
558 	}
559 }
560 
561 static void pcifront_free_roots(struct pcifront_device *pdev)
562 {
563 	struct pci_bus_entry *bus_entry, *t;
564 
565 	dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
566 
567 	pci_lock_rescan_remove();
568 	list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
569 		list_del(&bus_entry->list);
570 
571 		free_root_bus_devs(bus_entry->bus);
572 
573 		kfree(bus_entry->bus->sysdata);
574 
575 		device_unregister(bus_entry->bus->bridge);
576 		pci_remove_bus(bus_entry->bus);
577 
578 		kfree(bus_entry);
579 	}
580 	pci_unlock_rescan_remove();
581 }
582 
583 static pci_ers_result_t pcifront_common_process(int cmd,
584 						struct pcifront_device *pdev,
585 						pci_channel_state_t state)
586 {
587 	pci_ers_result_t result;
588 	struct pci_driver *pdrv;
589 	int bus = pdev->sh_info->aer_op.bus;
590 	int devfn = pdev->sh_info->aer_op.devfn;
591 	int domain = pdev->sh_info->aer_op.domain;
592 	struct pci_dev *pcidev;
593 	int flag = 0;
594 
595 	dev_dbg(&pdev->xdev->dev,
596 		"pcifront AER process: cmd %x (bus:%x, devfn%x)",
597 		cmd, bus, devfn);
598 	result = PCI_ERS_RESULT_NONE;
599 
600 	pcidev = pci_get_domain_bus_and_slot(domain, bus, devfn);
601 	if (!pcidev || !pcidev->driver) {
602 		dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
603 		pci_dev_put(pcidev);
604 		return result;
605 	}
606 	pdrv = pcidev->driver;
607 
608 	if (pdrv) {
609 		if (pdrv->err_handler && pdrv->err_handler->error_detected) {
610 			pci_dbg(pcidev, "trying to call AER service\n");
611 			if (pcidev) {
612 				flag = 1;
613 				switch (cmd) {
614 				case XEN_PCI_OP_aer_detected:
615 					result = pdrv->err_handler->
616 						 error_detected(pcidev, state);
617 					break;
618 				case XEN_PCI_OP_aer_mmio:
619 					result = pdrv->err_handler->
620 						 mmio_enabled(pcidev);
621 					break;
622 				case XEN_PCI_OP_aer_slotreset:
623 					result = pdrv->err_handler->
624 						 slot_reset(pcidev);
625 					break;
626 				case XEN_PCI_OP_aer_resume:
627 					pdrv->err_handler->resume(pcidev);
628 					break;
629 				default:
630 					dev_err(&pdev->xdev->dev,
631 						"bad request in aer recovery "
632 						"operation!\n");
633 
634 				}
635 			}
636 		}
637 	}
638 	if (!flag)
639 		result = PCI_ERS_RESULT_NONE;
640 
641 	return result;
642 }
643 
644 
645 static void pcifront_do_aer(struct work_struct *data)
646 {
647 	struct pcifront_device *pdev =
648 		container_of(data, struct pcifront_device, op_work);
649 	int cmd = pdev->sh_info->aer_op.cmd;
650 	pci_channel_state_t state =
651 		(pci_channel_state_t)pdev->sh_info->aer_op.err;
652 
653 	/*If a pci_conf op is in progress,
654 		we have to wait until it is done before service aer op*/
655 	dev_dbg(&pdev->xdev->dev,
656 		"pcifront service aer bus %x devfn %x\n",
657 		pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
658 
659 	pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
660 
661 	/* Post the operation to the guest. */
662 	wmb();
663 	clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
664 	notify_remote_via_evtchn(pdev->evtchn);
665 
666 	/*in case of we lost an aer request in four lines time_window*/
667 	smp_mb__before_atomic();
668 	clear_bit(_PDEVB_op_active, &pdev->flags);
669 	smp_mb__after_atomic();
670 
671 	schedule_pcifront_aer_op(pdev);
672 
673 }
674 
675 static irqreturn_t pcifront_handler_aer(int irq, void *dev)
676 {
677 	struct pcifront_device *pdev = dev;
678 	schedule_pcifront_aer_op(pdev);
679 	return IRQ_HANDLED;
680 }
681 static int pcifront_connect_and_init_dma(struct pcifront_device *pdev)
682 {
683 	int err = 0;
684 
685 	spin_lock(&pcifront_dev_lock);
686 
687 	if (!pcifront_dev) {
688 		dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
689 		pcifront_dev = pdev;
690 	} else
691 		err = -EEXIST;
692 
693 	spin_unlock(&pcifront_dev_lock);
694 
695 	if (!err && !swiotlb_nr_tbl()) {
696 		err = pci_xen_swiotlb_init_late();
697 		if (err)
698 			dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n");
699 	}
700 	return err;
701 }
702 
703 static void pcifront_disconnect(struct pcifront_device *pdev)
704 {
705 	spin_lock(&pcifront_dev_lock);
706 
707 	if (pdev == pcifront_dev) {
708 		dev_info(&pdev->xdev->dev,
709 			 "Disconnecting PCI Frontend Buses\n");
710 		pcifront_dev = NULL;
711 	}
712 
713 	spin_unlock(&pcifront_dev_lock);
714 }
715 static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
716 {
717 	struct pcifront_device *pdev;
718 
719 	pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
720 	if (pdev == NULL)
721 		goto out;
722 
723 	pdev->sh_info =
724 	    (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL);
725 	if (pdev->sh_info == NULL) {
726 		kfree(pdev);
727 		pdev = NULL;
728 		goto out;
729 	}
730 	pdev->sh_info->flags = 0;
731 
732 	/*Flag for registering PV AER handler*/
733 	set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
734 
735 	dev_set_drvdata(&xdev->dev, pdev);
736 	pdev->xdev = xdev;
737 
738 	INIT_LIST_HEAD(&pdev->root_buses);
739 
740 	spin_lock_init(&pdev->sh_info_lock);
741 
742 	pdev->evtchn = INVALID_EVTCHN;
743 	pdev->gnt_ref = INVALID_GRANT_REF;
744 	pdev->irq = -1;
745 
746 	INIT_WORK(&pdev->op_work, pcifront_do_aer);
747 
748 	dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
749 		pdev, pdev->sh_info);
750 out:
751 	return pdev;
752 }
753 
754 static void free_pdev(struct pcifront_device *pdev)
755 {
756 	dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
757 
758 	pcifront_free_roots(pdev);
759 
760 	cancel_work_sync(&pdev->op_work);
761 
762 	if (pdev->irq >= 0)
763 		unbind_from_irqhandler(pdev->irq, pdev);
764 
765 	if (pdev->evtchn != INVALID_EVTCHN)
766 		xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
767 
768 	if (pdev->gnt_ref != INVALID_GRANT_REF)
769 		gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */,
770 					  (unsigned long)pdev->sh_info);
771 	else
772 		free_page((unsigned long)pdev->sh_info);
773 
774 	dev_set_drvdata(&pdev->xdev->dev, NULL);
775 
776 	kfree(pdev);
777 }
778 
779 static int pcifront_publish_info(struct pcifront_device *pdev)
780 {
781 	int err = 0;
782 	struct xenbus_transaction trans;
783 	grant_ref_t gref;
784 
785 	err = xenbus_grant_ring(pdev->xdev, pdev->sh_info, 1, &gref);
786 	if (err < 0)
787 		goto out;
788 
789 	pdev->gnt_ref = gref;
790 
791 	err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
792 	if (err)
793 		goto out;
794 
795 	err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
796 		0, "pcifront", pdev);
797 
798 	if (err < 0)
799 		return err;
800 
801 	pdev->irq = err;
802 
803 do_publish:
804 	err = xenbus_transaction_start(&trans);
805 	if (err) {
806 		xenbus_dev_fatal(pdev->xdev, err,
807 				 "Error writing configuration for backend "
808 				 "(start transaction)");
809 		goto out;
810 	}
811 
812 	err = xenbus_printf(trans, pdev->xdev->nodename,
813 			    "pci-op-ref", "%u", pdev->gnt_ref);
814 	if (!err)
815 		err = xenbus_printf(trans, pdev->xdev->nodename,
816 				    "event-channel", "%u", pdev->evtchn);
817 	if (!err)
818 		err = xenbus_printf(trans, pdev->xdev->nodename,
819 				    "magic", XEN_PCI_MAGIC);
820 
821 	if (err) {
822 		xenbus_transaction_end(trans, 1);
823 		xenbus_dev_fatal(pdev->xdev, err,
824 				 "Error writing configuration for backend");
825 		goto out;
826 	} else {
827 		err = xenbus_transaction_end(trans, 0);
828 		if (err == -EAGAIN)
829 			goto do_publish;
830 		else if (err) {
831 			xenbus_dev_fatal(pdev->xdev, err,
832 					 "Error completing transaction "
833 					 "for backend");
834 			goto out;
835 		}
836 	}
837 
838 	xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
839 
840 	dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
841 
842 out:
843 	return err;
844 }
845 
846 static int pcifront_try_connect(struct pcifront_device *pdev)
847 {
848 	int err = -EFAULT;
849 	int i, num_roots, len;
850 	char str[64];
851 	unsigned int domain, bus;
852 
853 
854 	/* Only connect once */
855 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
856 	    XenbusStateInitialised)
857 		goto out;
858 
859 	err = pcifront_connect_and_init_dma(pdev);
860 	if (err && err != -EEXIST) {
861 		xenbus_dev_fatal(pdev->xdev, err,
862 				 "Error setting up PCI Frontend");
863 		goto out;
864 	}
865 
866 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
867 			   "root_num", "%d", &num_roots);
868 	if (err == -ENOENT) {
869 		xenbus_dev_error(pdev->xdev, err,
870 				 "No PCI Roots found, trying 0000:00");
871 		err = pcifront_scan_root(pdev, 0, 0);
872 		if (err) {
873 			xenbus_dev_fatal(pdev->xdev, err,
874 					 "Error scanning PCI root 0000:00");
875 			goto out;
876 		}
877 		num_roots = 0;
878 	} else if (err != 1) {
879 		if (err == 0)
880 			err = -EINVAL;
881 		xenbus_dev_fatal(pdev->xdev, err,
882 				 "Error reading number of PCI roots");
883 		goto out;
884 	}
885 
886 	for (i = 0; i < num_roots; i++) {
887 		len = snprintf(str, sizeof(str), "root-%d", i);
888 		if (unlikely(len >= (sizeof(str) - 1))) {
889 			err = -ENOMEM;
890 			goto out;
891 		}
892 
893 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
894 				   "%x:%x", &domain, &bus);
895 		if (err != 2) {
896 			if (err >= 0)
897 				err = -EINVAL;
898 			xenbus_dev_fatal(pdev->xdev, err,
899 					 "Error reading PCI root %d", i);
900 			goto out;
901 		}
902 
903 		err = pcifront_scan_root(pdev, domain, bus);
904 		if (err) {
905 			xenbus_dev_fatal(pdev->xdev, err,
906 					 "Error scanning PCI root %04x:%02x",
907 					 domain, bus);
908 			goto out;
909 		}
910 	}
911 
912 	err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
913 
914 out:
915 	return err;
916 }
917 
918 static int pcifront_try_disconnect(struct pcifront_device *pdev)
919 {
920 	int err = 0;
921 	enum xenbus_state prev_state;
922 
923 
924 	prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
925 
926 	if (prev_state >= XenbusStateClosing)
927 		goto out;
928 
929 	if (prev_state == XenbusStateConnected) {
930 		pcifront_free_roots(pdev);
931 		pcifront_disconnect(pdev);
932 	}
933 
934 	err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
935 
936 out:
937 
938 	return err;
939 }
940 
941 static int pcifront_attach_devices(struct pcifront_device *pdev)
942 {
943 	int err = -EFAULT;
944 	int i, num_roots, len;
945 	unsigned int domain, bus;
946 	char str[64];
947 
948 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
949 	    XenbusStateReconfiguring)
950 		goto out;
951 
952 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
953 			   "root_num", "%d", &num_roots);
954 	if (err == -ENOENT) {
955 		xenbus_dev_error(pdev->xdev, err,
956 				 "No PCI Roots found, trying 0000:00");
957 		err = pcifront_rescan_root(pdev, 0, 0);
958 		if (err) {
959 			xenbus_dev_fatal(pdev->xdev, err,
960 					 "Error scanning PCI root 0000:00");
961 			goto out;
962 		}
963 		num_roots = 0;
964 	} else if (err != 1) {
965 		if (err == 0)
966 			err = -EINVAL;
967 		xenbus_dev_fatal(pdev->xdev, err,
968 				 "Error reading number of PCI roots");
969 		goto out;
970 	}
971 
972 	for (i = 0; i < num_roots; i++) {
973 		len = snprintf(str, sizeof(str), "root-%d", i);
974 		if (unlikely(len >= (sizeof(str) - 1))) {
975 			err = -ENOMEM;
976 			goto out;
977 		}
978 
979 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
980 				   "%x:%x", &domain, &bus);
981 		if (err != 2) {
982 			if (err >= 0)
983 				err = -EINVAL;
984 			xenbus_dev_fatal(pdev->xdev, err,
985 					 "Error reading PCI root %d", i);
986 			goto out;
987 		}
988 
989 		err = pcifront_rescan_root(pdev, domain, bus);
990 		if (err) {
991 			xenbus_dev_fatal(pdev->xdev, err,
992 					 "Error scanning PCI root %04x:%02x",
993 					 domain, bus);
994 			goto out;
995 		}
996 	}
997 
998 	xenbus_switch_state(pdev->xdev, XenbusStateConnected);
999 
1000 out:
1001 	return err;
1002 }
1003 
1004 static int pcifront_detach_devices(struct pcifront_device *pdev)
1005 {
1006 	int err = 0;
1007 	int i, num_devs;
1008 	unsigned int domain, bus, slot, func;
1009 	struct pci_dev *pci_dev;
1010 	char str[64];
1011 
1012 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
1013 	    XenbusStateConnected)
1014 		goto out;
1015 
1016 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
1017 			   &num_devs);
1018 	if (err != 1) {
1019 		if (err >= 0)
1020 			err = -EINVAL;
1021 		xenbus_dev_fatal(pdev->xdev, err,
1022 				 "Error reading number of PCI devices");
1023 		goto out;
1024 	}
1025 
1026 	/* Find devices being detached and remove them. */
1027 	for (i = 0; i < num_devs; i++) {
1028 		int l, state;
1029 		l = snprintf(str, sizeof(str), "state-%d", i);
1030 		if (unlikely(l >= (sizeof(str) - 1))) {
1031 			err = -ENOMEM;
1032 			goto out;
1033 		}
1034 		state = xenbus_read_unsigned(pdev->xdev->otherend, str,
1035 					     XenbusStateUnknown);
1036 
1037 		if (state != XenbusStateClosing)
1038 			continue;
1039 
1040 		/* Remove device. */
1041 		l = snprintf(str, sizeof(str), "vdev-%d", i);
1042 		if (unlikely(l >= (sizeof(str) - 1))) {
1043 			err = -ENOMEM;
1044 			goto out;
1045 		}
1046 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
1047 				   "%x:%x:%x.%x", &domain, &bus, &slot, &func);
1048 		if (err != 4) {
1049 			if (err >= 0)
1050 				err = -EINVAL;
1051 			xenbus_dev_fatal(pdev->xdev, err,
1052 					 "Error reading PCI device %d", i);
1053 			goto out;
1054 		}
1055 
1056 		pci_dev = pci_get_domain_bus_and_slot(domain, bus,
1057 				PCI_DEVFN(slot, func));
1058 		if (!pci_dev) {
1059 			dev_dbg(&pdev->xdev->dev,
1060 				"Cannot get PCI device %04x:%02x:%02x.%d\n",
1061 				domain, bus, slot, func);
1062 			continue;
1063 		}
1064 		pci_lock_rescan_remove();
1065 		pci_stop_and_remove_bus_device(pci_dev);
1066 		pci_dev_put(pci_dev);
1067 		pci_unlock_rescan_remove();
1068 
1069 		dev_dbg(&pdev->xdev->dev,
1070 			"PCI device %04x:%02x:%02x.%d removed.\n",
1071 			domain, bus, slot, func);
1072 	}
1073 
1074 	err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
1075 
1076 out:
1077 	return err;
1078 }
1079 
1080 static void __ref pcifront_backend_changed(struct xenbus_device *xdev,
1081 						  enum xenbus_state be_state)
1082 {
1083 	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1084 
1085 	switch (be_state) {
1086 	case XenbusStateUnknown:
1087 	case XenbusStateInitialising:
1088 	case XenbusStateInitWait:
1089 	case XenbusStateInitialised:
1090 		break;
1091 
1092 	case XenbusStateConnected:
1093 		pcifront_try_connect(pdev);
1094 		break;
1095 
1096 	case XenbusStateClosed:
1097 		if (xdev->state == XenbusStateClosed)
1098 			break;
1099 		/* fall through - Missed the backend's CLOSING state. */
1100 	case XenbusStateClosing:
1101 		dev_warn(&xdev->dev, "backend going away!\n");
1102 		pcifront_try_disconnect(pdev);
1103 		break;
1104 
1105 	case XenbusStateReconfiguring:
1106 		pcifront_detach_devices(pdev);
1107 		break;
1108 
1109 	case XenbusStateReconfigured:
1110 		pcifront_attach_devices(pdev);
1111 		break;
1112 	}
1113 }
1114 
1115 static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1116 				 const struct xenbus_device_id *id)
1117 {
1118 	int err = 0;
1119 	struct pcifront_device *pdev = alloc_pdev(xdev);
1120 
1121 	if (pdev == NULL) {
1122 		err = -ENOMEM;
1123 		xenbus_dev_fatal(xdev, err,
1124 				 "Error allocating pcifront_device struct");
1125 		goto out;
1126 	}
1127 
1128 	err = pcifront_publish_info(pdev);
1129 	if (err)
1130 		free_pdev(pdev);
1131 
1132 out:
1133 	return err;
1134 }
1135 
1136 static int pcifront_xenbus_remove(struct xenbus_device *xdev)
1137 {
1138 	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1139 	if (pdev)
1140 		free_pdev(pdev);
1141 
1142 	return 0;
1143 }
1144 
1145 static const struct xenbus_device_id xenpci_ids[] = {
1146 	{"pci"},
1147 	{""},
1148 };
1149 
1150 static struct xenbus_driver xenpci_driver = {
1151 	.name			= "pcifront",
1152 	.ids			= xenpci_ids,
1153 	.probe			= pcifront_xenbus_probe,
1154 	.remove			= pcifront_xenbus_remove,
1155 	.otherend_changed	= pcifront_backend_changed,
1156 };
1157 
1158 static int __init pcifront_init(void)
1159 {
1160 	if (!xen_pv_domain() || xen_initial_domain())
1161 		return -ENODEV;
1162 
1163 	if (!xen_has_pv_devices())
1164 		return -ENODEV;
1165 
1166 	pci_frontend_registrar(1 /* enable */);
1167 
1168 	return xenbus_register_frontend(&xenpci_driver);
1169 }
1170 
1171 static void __exit pcifront_cleanup(void)
1172 {
1173 	xenbus_unregister_driver(&xenpci_driver);
1174 	pci_frontend_registrar(0 /* disable */);
1175 }
1176 module_init(pcifront_init);
1177 module_exit(pcifront_cleanup);
1178 
1179 MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1180 MODULE_LICENSE("GPL");
1181 MODULE_ALIAS("xen:pci");
1182