xref: /openbmc/linux/drivers/pci/xen-pcifront.c (revision a8a28aff)
1 /*
2  * Xen PCI Frontend.
3  *
4  *   Author: Ryan Wilson <hap9@epoch.ncsc.mil>
5  */
6 #include <linux/module.h>
7 #include <linux/init.h>
8 #include <linux/mm.h>
9 #include <xen/xenbus.h>
10 #include <xen/events.h>
11 #include <xen/grant_table.h>
12 #include <xen/page.h>
13 #include <linux/spinlock.h>
14 #include <linux/pci.h>
15 #include <linux/msi.h>
16 #include <xen/interface/io/pciif.h>
17 #include <asm/xen/pci.h>
18 #include <linux/interrupt.h>
19 #include <linux/atomic.h>
20 #include <linux/workqueue.h>
21 #include <linux/bitops.h>
22 #include <linux/time.h>
23 #include <xen/platform_pci.h>
24 
25 #include <asm/xen/swiotlb-xen.h>
26 #define INVALID_GRANT_REF (0)
27 #define INVALID_EVTCHN    (-1)
28 
29 struct pci_bus_entry {
30 	struct list_head list;
31 	struct pci_bus *bus;
32 };
33 
34 #define _PDEVB_op_active		(0)
35 #define PDEVB_op_active			(1 << (_PDEVB_op_active))
36 
37 struct pcifront_device {
38 	struct xenbus_device *xdev;
39 	struct list_head root_buses;
40 
41 	int evtchn;
42 	int gnt_ref;
43 
44 	int irq;
45 
46 	/* Lock this when doing any operations in sh_info */
47 	spinlock_t sh_info_lock;
48 	struct xen_pci_sharedinfo *sh_info;
49 	struct work_struct op_work;
50 	unsigned long flags;
51 
52 };
53 
54 struct pcifront_sd {
55 	int domain;
56 	struct pcifront_device *pdev;
57 };
58 
59 static inline struct pcifront_device *
60 pcifront_get_pdev(struct pcifront_sd *sd)
61 {
62 	return sd->pdev;
63 }
64 
65 static inline void pcifront_init_sd(struct pcifront_sd *sd,
66 				    unsigned int domain, unsigned int bus,
67 				    struct pcifront_device *pdev)
68 {
69 	sd->domain = domain;
70 	sd->pdev = pdev;
71 }
72 
73 static DEFINE_SPINLOCK(pcifront_dev_lock);
74 static struct pcifront_device *pcifront_dev;
75 
76 static int verbose_request;
77 module_param(verbose_request, int, 0644);
78 
79 static int errno_to_pcibios_err(int errno)
80 {
81 	switch (errno) {
82 	case XEN_PCI_ERR_success:
83 		return PCIBIOS_SUCCESSFUL;
84 
85 	case XEN_PCI_ERR_dev_not_found:
86 		return PCIBIOS_DEVICE_NOT_FOUND;
87 
88 	case XEN_PCI_ERR_invalid_offset:
89 	case XEN_PCI_ERR_op_failed:
90 		return PCIBIOS_BAD_REGISTER_NUMBER;
91 
92 	case XEN_PCI_ERR_not_implemented:
93 		return PCIBIOS_FUNC_NOT_SUPPORTED;
94 
95 	case XEN_PCI_ERR_access_denied:
96 		return PCIBIOS_SET_FAILED;
97 	}
98 	return errno;
99 }
100 
101 static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
102 {
103 	if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
104 		&& !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
105 		dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
106 		schedule_work(&pdev->op_work);
107 	}
108 }
109 
110 static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
111 {
112 	int err = 0;
113 	struct xen_pci_op *active_op = &pdev->sh_info->op;
114 	unsigned long irq_flags;
115 	evtchn_port_t port = pdev->evtchn;
116 	unsigned irq = pdev->irq;
117 	s64 ns, ns_timeout;
118 	struct timeval tv;
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 	do_gettimeofday(&tv);
136 	ns_timeout = timeval_to_ns(&tv) + 2 * (s64)NSEC_PER_SEC;
137 
138 	xen_clear_irq_pending(irq);
139 
140 	while (test_bit(_XEN_PCIF_active,
141 			(unsigned long *)&pdev->sh_info->flags)) {
142 		xen_poll_irq_timeout(irq, jiffies + 3*HZ);
143 		xen_clear_irq_pending(irq);
144 		do_gettimeofday(&tv);
145 		ns = timeval_to_ns(&tv);
146 		if (ns > ns_timeout) {
147 			dev_err(&pdev->xdev->dev,
148 				"pciback not responding!!!\n");
149 			clear_bit(_XEN_PCIF_active,
150 				  (unsigned long *)&pdev->sh_info->flags);
151 			err = XEN_PCI_ERR_dev_not_found;
152 			goto out;
153 		}
154 	}
155 
156 	/*
157 	* We might lose backend service request since we
158 	* reuse same evtchn with pci_conf backend response. So re-schedule
159 	* aer pcifront service.
160 	*/
161 	if (test_bit(_XEN_PCIB_active,
162 			(unsigned long *)&pdev->sh_info->flags)) {
163 		dev_err(&pdev->xdev->dev,
164 			"schedule aer pcifront service\n");
165 		schedule_pcifront_aer_op(pdev);
166 	}
167 
168 	memcpy(op, active_op, sizeof(struct xen_pci_op));
169 
170 	err = op->err;
171 out:
172 	spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags);
173 	return err;
174 }
175 
176 /* Access to this function is spinlocked in drivers/pci/access.c */
177 static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn,
178 			     int where, int size, u32 *val)
179 {
180 	int err = 0;
181 	struct xen_pci_op op = {
182 		.cmd    = XEN_PCI_OP_conf_read,
183 		.domain = pci_domain_nr(bus),
184 		.bus    = bus->number,
185 		.devfn  = devfn,
186 		.offset = where,
187 		.size   = size,
188 	};
189 	struct pcifront_sd *sd = bus->sysdata;
190 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
191 
192 	if (verbose_request)
193 		dev_info(&pdev->xdev->dev,
194 			 "read dev=%04x:%02x:%02x.%d - offset %x size %d\n",
195 			 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
196 			 PCI_FUNC(devfn), where, size);
197 
198 	err = do_pci_op(pdev, &op);
199 
200 	if (likely(!err)) {
201 		if (verbose_request)
202 			dev_info(&pdev->xdev->dev, "read got back value %x\n",
203 				 op.value);
204 
205 		*val = op.value;
206 	} else if (err == -ENODEV) {
207 		/* No device here, pretend that it just returned 0 */
208 		err = 0;
209 		*val = 0;
210 	}
211 
212 	return errno_to_pcibios_err(err);
213 }
214 
215 /* Access to this function is spinlocked in drivers/pci/access.c */
216 static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
217 			      int where, int size, u32 val)
218 {
219 	struct xen_pci_op op = {
220 		.cmd    = XEN_PCI_OP_conf_write,
221 		.domain = pci_domain_nr(bus),
222 		.bus    = bus->number,
223 		.devfn  = devfn,
224 		.offset = where,
225 		.size   = size,
226 		.value  = val,
227 	};
228 	struct pcifront_sd *sd = bus->sysdata;
229 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
230 
231 	if (verbose_request)
232 		dev_info(&pdev->xdev->dev,
233 			 "write dev=%04x:%02x:%02x.%d - "
234 			 "offset %x size %d val %x\n",
235 			 pci_domain_nr(bus), bus->number,
236 			 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
237 
238 	return errno_to_pcibios_err(do_pci_op(pdev, &op));
239 }
240 
241 static struct pci_ops pcifront_bus_ops = {
242 	.read = pcifront_bus_read,
243 	.write = pcifront_bus_write,
244 };
245 
246 #ifdef CONFIG_PCI_MSI
247 static int pci_frontend_enable_msix(struct pci_dev *dev,
248 				    int vector[], int nvec)
249 {
250 	int err;
251 	int i;
252 	struct xen_pci_op op = {
253 		.cmd    = XEN_PCI_OP_enable_msix,
254 		.domain = pci_domain_nr(dev->bus),
255 		.bus = dev->bus->number,
256 		.devfn = dev->devfn,
257 		.value = nvec,
258 	};
259 	struct pcifront_sd *sd = dev->bus->sysdata;
260 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
261 	struct msi_desc *entry;
262 
263 	if (nvec > SH_INFO_MAX_VEC) {
264 		dev_err(&dev->dev, "too much vector for pci frontend: %x."
265 				   " Increase SH_INFO_MAX_VEC.\n", nvec);
266 		return -EINVAL;
267 	}
268 
269 	i = 0;
270 	list_for_each_entry(entry, &dev->msi_list, list) {
271 		op.msix_entries[i].entry = entry->msi_attrib.entry_nr;
272 		/* Vector is useless at this point. */
273 		op.msix_entries[i].vector = -1;
274 		i++;
275 	}
276 
277 	err = do_pci_op(pdev, &op);
278 
279 	if (likely(!err)) {
280 		if (likely(!op.value)) {
281 			/* we get the result */
282 			for (i = 0; i < nvec; i++) {
283 				if (op.msix_entries[i].vector <= 0) {
284 					dev_warn(&dev->dev, "MSI-X entry %d is invalid: %d!\n",
285 						i, op.msix_entries[i].vector);
286 					err = -EINVAL;
287 					vector[i] = -1;
288 					continue;
289 				}
290 				vector[i] = op.msix_entries[i].vector;
291 			}
292 		} else {
293 			printk(KERN_DEBUG "enable msix get value %x\n",
294 				op.value);
295 			err = op.value;
296 		}
297 	} else {
298 		dev_err(&dev->dev, "enable msix get err %x\n", err);
299 	}
300 	return err;
301 }
302 
303 static void pci_frontend_disable_msix(struct pci_dev *dev)
304 {
305 	int err;
306 	struct xen_pci_op op = {
307 		.cmd    = XEN_PCI_OP_disable_msix,
308 		.domain = pci_domain_nr(dev->bus),
309 		.bus = dev->bus->number,
310 		.devfn = dev->devfn,
311 	};
312 	struct pcifront_sd *sd = dev->bus->sysdata;
313 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
314 
315 	err = do_pci_op(pdev, &op);
316 
317 	/* What should do for error ? */
318 	if (err)
319 		dev_err(&dev->dev, "pci_disable_msix get err %x\n", err);
320 }
321 
322 static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[])
323 {
324 	int err;
325 	struct xen_pci_op op = {
326 		.cmd    = XEN_PCI_OP_enable_msi,
327 		.domain = pci_domain_nr(dev->bus),
328 		.bus = dev->bus->number,
329 		.devfn = dev->devfn,
330 	};
331 	struct pcifront_sd *sd = dev->bus->sysdata;
332 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
333 
334 	err = do_pci_op(pdev, &op);
335 	if (likely(!err)) {
336 		vector[0] = op.value;
337 		if (op.value <= 0) {
338 			dev_warn(&dev->dev, "MSI entry is invalid: %d!\n",
339 				op.value);
340 			err = -EINVAL;
341 			vector[0] = -1;
342 		}
343 	} else {
344 		dev_err(&dev->dev, "pci frontend enable msi failed for dev "
345 				    "%x:%x\n", op.bus, op.devfn);
346 		err = -EINVAL;
347 	}
348 	return err;
349 }
350 
351 static void pci_frontend_disable_msi(struct pci_dev *dev)
352 {
353 	int err;
354 	struct xen_pci_op op = {
355 		.cmd    = XEN_PCI_OP_disable_msi,
356 		.domain = pci_domain_nr(dev->bus),
357 		.bus = dev->bus->number,
358 		.devfn = dev->devfn,
359 	};
360 	struct pcifront_sd *sd = dev->bus->sysdata;
361 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
362 
363 	err = do_pci_op(pdev, &op);
364 	if (err == XEN_PCI_ERR_dev_not_found) {
365 		/* XXX No response from backend, what shall we do? */
366 		printk(KERN_DEBUG "get no response from backend for disable MSI\n");
367 		return;
368 	}
369 	if (err)
370 		/* how can pciback notify us fail? */
371 		printk(KERN_DEBUG "get fake response frombackend\n");
372 }
373 
374 static struct xen_pci_frontend_ops pci_frontend_ops = {
375 	.enable_msi = pci_frontend_enable_msi,
376 	.disable_msi = pci_frontend_disable_msi,
377 	.enable_msix = pci_frontend_enable_msix,
378 	.disable_msix = pci_frontend_disable_msix,
379 };
380 
381 static void pci_frontend_registrar(int enable)
382 {
383 	if (enable)
384 		xen_pci_frontend = &pci_frontend_ops;
385 	else
386 		xen_pci_frontend = NULL;
387 };
388 #else
389 static inline void pci_frontend_registrar(int enable) { };
390 #endif /* CONFIG_PCI_MSI */
391 
392 /* Claim resources for the PCI frontend as-is, backend won't allow changes */
393 static int pcifront_claim_resource(struct pci_dev *dev, void *data)
394 {
395 	struct pcifront_device *pdev = data;
396 	int i;
397 	struct resource *r;
398 
399 	for (i = 0; i < PCI_NUM_RESOURCES; i++) {
400 		r = &dev->resource[i];
401 
402 		if (!r->parent && r->start && r->flags) {
403 			dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n",
404 				pci_name(dev), i);
405 			if (pci_claim_resource(dev, i)) {
406 				dev_err(&pdev->xdev->dev, "Could not claim resource %s/%d! "
407 					"Device offline. Try using e820_host=1 in the guest config.\n",
408 					pci_name(dev), i);
409 			}
410 		}
411 	}
412 
413 	return 0;
414 }
415 
416 static int pcifront_scan_bus(struct pcifront_device *pdev,
417 				unsigned int domain, unsigned int bus,
418 				struct pci_bus *b)
419 {
420 	struct pci_dev *d;
421 	unsigned int devfn;
422 
423 	/* Scan the bus for functions and add.
424 	 * We omit handling of PCI bridge attachment because pciback prevents
425 	 * bridges from being exported.
426 	 */
427 	for (devfn = 0; devfn < 0x100; devfn++) {
428 		d = pci_get_slot(b, devfn);
429 		if (d) {
430 			/* Device is already known. */
431 			pci_dev_put(d);
432 			continue;
433 		}
434 
435 		d = pci_scan_single_device(b, devfn);
436 		if (d)
437 			dev_info(&pdev->xdev->dev, "New device on "
438 				 "%04x:%02x:%02x.%d found.\n", domain, bus,
439 				 PCI_SLOT(devfn), PCI_FUNC(devfn));
440 	}
441 
442 	return 0;
443 }
444 
445 static int pcifront_scan_root(struct pcifront_device *pdev,
446 				 unsigned int domain, unsigned int bus)
447 {
448 	struct pci_bus *b;
449 	struct pcifront_sd *sd = NULL;
450 	struct pci_bus_entry *bus_entry = NULL;
451 	int err = 0;
452 
453 #ifndef CONFIG_PCI_DOMAINS
454 	if (domain != 0) {
455 		dev_err(&pdev->xdev->dev,
456 			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
457 		dev_err(&pdev->xdev->dev,
458 			"Please compile with CONFIG_PCI_DOMAINS\n");
459 		err = -EINVAL;
460 		goto err_out;
461 	}
462 #endif
463 
464 	dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n",
465 		 domain, bus);
466 
467 	bus_entry = kmalloc(sizeof(*bus_entry), GFP_KERNEL);
468 	sd = kmalloc(sizeof(*sd), GFP_KERNEL);
469 	if (!bus_entry || !sd) {
470 		err = -ENOMEM;
471 		goto err_out;
472 	}
473 	pcifront_init_sd(sd, domain, bus, pdev);
474 
475 	pci_lock_rescan_remove();
476 
477 	b = pci_scan_bus_parented(&pdev->xdev->dev, bus,
478 				  &pcifront_bus_ops, sd);
479 	if (!b) {
480 		dev_err(&pdev->xdev->dev,
481 			"Error creating PCI Frontend Bus!\n");
482 		err = -ENOMEM;
483 		pci_unlock_rescan_remove();
484 		goto err_out;
485 	}
486 
487 	bus_entry->bus = b;
488 
489 	list_add(&bus_entry->list, &pdev->root_buses);
490 
491 	/* pci_scan_bus_parented skips devices which do not have a have
492 	* devfn==0. The pcifront_scan_bus enumerates all devfn. */
493 	err = pcifront_scan_bus(pdev, domain, bus, b);
494 
495 	/* Claim resources before going "live" with our devices */
496 	pci_walk_bus(b, pcifront_claim_resource, pdev);
497 
498 	/* Create SysFS and notify udev of the devices. Aka: "going live" */
499 	pci_bus_add_devices(b);
500 
501 	pci_unlock_rescan_remove();
502 	return err;
503 
504 err_out:
505 	kfree(bus_entry);
506 	kfree(sd);
507 
508 	return err;
509 }
510 
511 static int pcifront_rescan_root(struct pcifront_device *pdev,
512 				   unsigned int domain, unsigned int bus)
513 {
514 	int err;
515 	struct pci_bus *b;
516 
517 #ifndef CONFIG_PCI_DOMAINS
518 	if (domain != 0) {
519 		dev_err(&pdev->xdev->dev,
520 			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
521 		dev_err(&pdev->xdev->dev,
522 			"Please compile with CONFIG_PCI_DOMAINS\n");
523 		return -EINVAL;
524 	}
525 #endif
526 
527 	dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
528 		 domain, bus);
529 
530 	b = pci_find_bus(domain, bus);
531 	if (!b)
532 		/* If the bus is unknown, create it. */
533 		return pcifront_scan_root(pdev, domain, bus);
534 
535 	err = pcifront_scan_bus(pdev, domain, bus, b);
536 
537 	/* Claim resources before going "live" with our devices */
538 	pci_walk_bus(b, pcifront_claim_resource, pdev);
539 
540 	/* Create SysFS and notify udev of the devices. Aka: "going live" */
541 	pci_bus_add_devices(b);
542 
543 	return err;
544 }
545 
546 static void free_root_bus_devs(struct pci_bus *bus)
547 {
548 	struct pci_dev *dev;
549 
550 	while (!list_empty(&bus->devices)) {
551 		dev = container_of(bus->devices.next, struct pci_dev,
552 				   bus_list);
553 		dev_dbg(&dev->dev, "removing device\n");
554 		pci_stop_and_remove_bus_device(dev);
555 	}
556 }
557 
558 static void pcifront_free_roots(struct pcifront_device *pdev)
559 {
560 	struct pci_bus_entry *bus_entry, *t;
561 
562 	dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
563 
564 	pci_lock_rescan_remove();
565 	list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
566 		list_del(&bus_entry->list);
567 
568 		free_root_bus_devs(bus_entry->bus);
569 
570 		kfree(bus_entry->bus->sysdata);
571 
572 		device_unregister(bus_entry->bus->bridge);
573 		pci_remove_bus(bus_entry->bus);
574 
575 		kfree(bus_entry);
576 	}
577 	pci_unlock_rescan_remove();
578 }
579 
580 static pci_ers_result_t pcifront_common_process(int cmd,
581 						struct pcifront_device *pdev,
582 						pci_channel_state_t state)
583 {
584 	pci_ers_result_t result;
585 	struct pci_driver *pdrv;
586 	int bus = pdev->sh_info->aer_op.bus;
587 	int devfn = pdev->sh_info->aer_op.devfn;
588 	struct pci_dev *pcidev;
589 	int flag = 0;
590 
591 	dev_dbg(&pdev->xdev->dev,
592 		"pcifront AER process: cmd %x (bus:%x, devfn%x)",
593 		cmd, bus, devfn);
594 	result = PCI_ERS_RESULT_NONE;
595 
596 	pcidev = pci_get_bus_and_slot(bus, devfn);
597 	if (!pcidev || !pcidev->driver) {
598 		dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
599 		if (pcidev)
600 			pci_dev_put(pcidev);
601 		return result;
602 	}
603 	pdrv = pcidev->driver;
604 
605 	if (pdrv) {
606 		if (pdrv->err_handler && pdrv->err_handler->error_detected) {
607 			dev_dbg(&pcidev->dev,
608 				"trying to call AER service\n");
609 			if (pcidev) {
610 				flag = 1;
611 				switch (cmd) {
612 				case XEN_PCI_OP_aer_detected:
613 					result = pdrv->err_handler->
614 						 error_detected(pcidev, state);
615 					break;
616 				case XEN_PCI_OP_aer_mmio:
617 					result = pdrv->err_handler->
618 						 mmio_enabled(pcidev);
619 					break;
620 				case XEN_PCI_OP_aer_slotreset:
621 					result = pdrv->err_handler->
622 						 slot_reset(pcidev);
623 					break;
624 				case XEN_PCI_OP_aer_resume:
625 					pdrv->err_handler->resume(pcidev);
626 					break;
627 				default:
628 					dev_err(&pdev->xdev->dev,
629 						"bad request in aer recovery "
630 						"operation!\n");
631 
632 				}
633 			}
634 		}
635 	}
636 	if (!flag)
637 		result = PCI_ERS_RESULT_NONE;
638 
639 	return result;
640 }
641 
642 
643 static void pcifront_do_aer(struct work_struct *data)
644 {
645 	struct pcifront_device *pdev =
646 		container_of(data, struct pcifront_device, op_work);
647 	int cmd = pdev->sh_info->aer_op.cmd;
648 	pci_channel_state_t state =
649 		(pci_channel_state_t)pdev->sh_info->aer_op.err;
650 
651 	/*If a pci_conf op is in progress,
652 		we have to wait until it is done before service aer op*/
653 	dev_dbg(&pdev->xdev->dev,
654 		"pcifront service aer bus %x devfn %x\n",
655 		pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
656 
657 	pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
658 
659 	/* Post the operation to the guest. */
660 	wmb();
661 	clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
662 	notify_remote_via_evtchn(pdev->evtchn);
663 
664 	/*in case of we lost an aer request in four lines time_window*/
665 	smp_mb__before_atomic();
666 	clear_bit(_PDEVB_op_active, &pdev->flags);
667 	smp_mb__after_atomic();
668 
669 	schedule_pcifront_aer_op(pdev);
670 
671 }
672 
673 static irqreturn_t pcifront_handler_aer(int irq, void *dev)
674 {
675 	struct pcifront_device *pdev = dev;
676 	schedule_pcifront_aer_op(pdev);
677 	return IRQ_HANDLED;
678 }
679 static int pcifront_connect_and_init_dma(struct pcifront_device *pdev)
680 {
681 	int err = 0;
682 
683 	spin_lock(&pcifront_dev_lock);
684 
685 	if (!pcifront_dev) {
686 		dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
687 		pcifront_dev = pdev;
688 	} else
689 		err = -EEXIST;
690 
691 	spin_unlock(&pcifront_dev_lock);
692 
693 	if (!err && !swiotlb_nr_tbl()) {
694 		err = pci_xen_swiotlb_init_late();
695 		if (err)
696 			dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n");
697 	}
698 	return err;
699 }
700 
701 static void pcifront_disconnect(struct pcifront_device *pdev)
702 {
703 	spin_lock(&pcifront_dev_lock);
704 
705 	if (pdev == pcifront_dev) {
706 		dev_info(&pdev->xdev->dev,
707 			 "Disconnecting PCI Frontend Buses\n");
708 		pcifront_dev = NULL;
709 	}
710 
711 	spin_unlock(&pcifront_dev_lock);
712 }
713 static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
714 {
715 	struct pcifront_device *pdev;
716 
717 	pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
718 	if (pdev == NULL)
719 		goto out;
720 
721 	pdev->sh_info =
722 	    (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL);
723 	if (pdev->sh_info == NULL) {
724 		kfree(pdev);
725 		pdev = NULL;
726 		goto out;
727 	}
728 	pdev->sh_info->flags = 0;
729 
730 	/*Flag for registering PV AER handler*/
731 	set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
732 
733 	dev_set_drvdata(&xdev->dev, pdev);
734 	pdev->xdev = xdev;
735 
736 	INIT_LIST_HEAD(&pdev->root_buses);
737 
738 	spin_lock_init(&pdev->sh_info_lock);
739 
740 	pdev->evtchn = INVALID_EVTCHN;
741 	pdev->gnt_ref = INVALID_GRANT_REF;
742 	pdev->irq = -1;
743 
744 	INIT_WORK(&pdev->op_work, pcifront_do_aer);
745 
746 	dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
747 		pdev, pdev->sh_info);
748 out:
749 	return pdev;
750 }
751 
752 static void free_pdev(struct pcifront_device *pdev)
753 {
754 	dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
755 
756 	pcifront_free_roots(pdev);
757 
758 	cancel_work_sync(&pdev->op_work);
759 
760 	if (pdev->irq >= 0)
761 		unbind_from_irqhandler(pdev->irq, pdev);
762 
763 	if (pdev->evtchn != INVALID_EVTCHN)
764 		xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
765 
766 	if (pdev->gnt_ref != INVALID_GRANT_REF)
767 		gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */,
768 					  (unsigned long)pdev->sh_info);
769 	else
770 		free_page((unsigned long)pdev->sh_info);
771 
772 	dev_set_drvdata(&pdev->xdev->dev, NULL);
773 
774 	kfree(pdev);
775 }
776 
777 static int pcifront_publish_info(struct pcifront_device *pdev)
778 {
779 	int err = 0;
780 	struct xenbus_transaction trans;
781 
782 	err = xenbus_grant_ring(pdev->xdev, virt_to_mfn(pdev->sh_info));
783 	if (err < 0)
784 		goto out;
785 
786 	pdev->gnt_ref = err;
787 
788 	err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
789 	if (err)
790 		goto out;
791 
792 	err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
793 		0, "pcifront", pdev);
794 
795 	if (err < 0)
796 		return err;
797 
798 	pdev->irq = err;
799 
800 do_publish:
801 	err = xenbus_transaction_start(&trans);
802 	if (err) {
803 		xenbus_dev_fatal(pdev->xdev, err,
804 				 "Error writing configuration for backend "
805 				 "(start transaction)");
806 		goto out;
807 	}
808 
809 	err = xenbus_printf(trans, pdev->xdev->nodename,
810 			    "pci-op-ref", "%u", pdev->gnt_ref);
811 	if (!err)
812 		err = xenbus_printf(trans, pdev->xdev->nodename,
813 				    "event-channel", "%u", pdev->evtchn);
814 	if (!err)
815 		err = xenbus_printf(trans, pdev->xdev->nodename,
816 				    "magic", XEN_PCI_MAGIC);
817 
818 	if (err) {
819 		xenbus_transaction_end(trans, 1);
820 		xenbus_dev_fatal(pdev->xdev, err,
821 				 "Error writing configuration for backend");
822 		goto out;
823 	} else {
824 		err = xenbus_transaction_end(trans, 0);
825 		if (err == -EAGAIN)
826 			goto do_publish;
827 		else if (err) {
828 			xenbus_dev_fatal(pdev->xdev, err,
829 					 "Error completing transaction "
830 					 "for backend");
831 			goto out;
832 		}
833 	}
834 
835 	xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
836 
837 	dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
838 
839 out:
840 	return err;
841 }
842 
843 static int pcifront_try_connect(struct pcifront_device *pdev)
844 {
845 	int err = -EFAULT;
846 	int i, num_roots, len;
847 	char str[64];
848 	unsigned int domain, bus;
849 
850 
851 	/* Only connect once */
852 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
853 	    XenbusStateInitialised)
854 		goto out;
855 
856 	err = pcifront_connect_and_init_dma(pdev);
857 	if (err && err != -EEXIST) {
858 		xenbus_dev_fatal(pdev->xdev, err,
859 				 "Error setting up PCI Frontend");
860 		goto out;
861 	}
862 
863 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
864 			   "root_num", "%d", &num_roots);
865 	if (err == -ENOENT) {
866 		xenbus_dev_error(pdev->xdev, err,
867 				 "No PCI Roots found, trying 0000:00");
868 		err = pcifront_scan_root(pdev, 0, 0);
869 		num_roots = 0;
870 	} else if (err != 1) {
871 		if (err == 0)
872 			err = -EINVAL;
873 		xenbus_dev_fatal(pdev->xdev, err,
874 				 "Error reading number of PCI roots");
875 		goto out;
876 	}
877 
878 	for (i = 0; i < num_roots; i++) {
879 		len = snprintf(str, sizeof(str), "root-%d", i);
880 		if (unlikely(len >= (sizeof(str) - 1))) {
881 			err = -ENOMEM;
882 			goto out;
883 		}
884 
885 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
886 				   "%x:%x", &domain, &bus);
887 		if (err != 2) {
888 			if (err >= 0)
889 				err = -EINVAL;
890 			xenbus_dev_fatal(pdev->xdev, err,
891 					 "Error reading PCI root %d", i);
892 			goto out;
893 		}
894 
895 		err = pcifront_scan_root(pdev, domain, bus);
896 		if (err) {
897 			xenbus_dev_fatal(pdev->xdev, err,
898 					 "Error scanning PCI root %04x:%02x",
899 					 domain, bus);
900 			goto out;
901 		}
902 	}
903 
904 	err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
905 
906 out:
907 	return err;
908 }
909 
910 static int pcifront_try_disconnect(struct pcifront_device *pdev)
911 {
912 	int err = 0;
913 	enum xenbus_state prev_state;
914 
915 
916 	prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
917 
918 	if (prev_state >= XenbusStateClosing)
919 		goto out;
920 
921 	if (prev_state == XenbusStateConnected) {
922 		pcifront_free_roots(pdev);
923 		pcifront_disconnect(pdev);
924 	}
925 
926 	err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
927 
928 out:
929 
930 	return err;
931 }
932 
933 static int pcifront_attach_devices(struct pcifront_device *pdev)
934 {
935 	int err = -EFAULT;
936 	int i, num_roots, len;
937 	unsigned int domain, bus;
938 	char str[64];
939 
940 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
941 	    XenbusStateReconfiguring)
942 		goto out;
943 
944 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
945 			   "root_num", "%d", &num_roots);
946 	if (err == -ENOENT) {
947 		xenbus_dev_error(pdev->xdev, err,
948 				 "No PCI Roots found, trying 0000:00");
949 		err = pcifront_rescan_root(pdev, 0, 0);
950 		num_roots = 0;
951 	} else if (err != 1) {
952 		if (err == 0)
953 			err = -EINVAL;
954 		xenbus_dev_fatal(pdev->xdev, err,
955 				 "Error reading number of PCI roots");
956 		goto out;
957 	}
958 
959 	for (i = 0; i < num_roots; i++) {
960 		len = snprintf(str, sizeof(str), "root-%d", i);
961 		if (unlikely(len >= (sizeof(str) - 1))) {
962 			err = -ENOMEM;
963 			goto out;
964 		}
965 
966 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
967 				   "%x:%x", &domain, &bus);
968 		if (err != 2) {
969 			if (err >= 0)
970 				err = -EINVAL;
971 			xenbus_dev_fatal(pdev->xdev, err,
972 					 "Error reading PCI root %d", i);
973 			goto out;
974 		}
975 
976 		err = pcifront_rescan_root(pdev, domain, bus);
977 		if (err) {
978 			xenbus_dev_fatal(pdev->xdev, err,
979 					 "Error scanning PCI root %04x:%02x",
980 					 domain, bus);
981 			goto out;
982 		}
983 	}
984 
985 	xenbus_switch_state(pdev->xdev, XenbusStateConnected);
986 
987 out:
988 	return err;
989 }
990 
991 static int pcifront_detach_devices(struct pcifront_device *pdev)
992 {
993 	int err = 0;
994 	int i, num_devs;
995 	unsigned int domain, bus, slot, func;
996 	struct pci_dev *pci_dev;
997 	char str[64];
998 
999 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
1000 	    XenbusStateConnected)
1001 		goto out;
1002 
1003 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
1004 			   &num_devs);
1005 	if (err != 1) {
1006 		if (err >= 0)
1007 			err = -EINVAL;
1008 		xenbus_dev_fatal(pdev->xdev, err,
1009 				 "Error reading number of PCI devices");
1010 		goto out;
1011 	}
1012 
1013 	/* Find devices being detached and remove them. */
1014 	for (i = 0; i < num_devs; i++) {
1015 		int l, state;
1016 		l = snprintf(str, sizeof(str), "state-%d", i);
1017 		if (unlikely(l >= (sizeof(str) - 1))) {
1018 			err = -ENOMEM;
1019 			goto out;
1020 		}
1021 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d",
1022 				   &state);
1023 		if (err != 1)
1024 			state = XenbusStateUnknown;
1025 
1026 		if (state != XenbusStateClosing)
1027 			continue;
1028 
1029 		/* Remove device. */
1030 		l = snprintf(str, sizeof(str), "vdev-%d", i);
1031 		if (unlikely(l >= (sizeof(str) - 1))) {
1032 			err = -ENOMEM;
1033 			goto out;
1034 		}
1035 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
1036 				   "%x:%x:%x.%x", &domain, &bus, &slot, &func);
1037 		if (err != 4) {
1038 			if (err >= 0)
1039 				err = -EINVAL;
1040 			xenbus_dev_fatal(pdev->xdev, err,
1041 					 "Error reading PCI device %d", i);
1042 			goto out;
1043 		}
1044 
1045 		pci_dev = pci_get_domain_bus_and_slot(domain, bus,
1046 				PCI_DEVFN(slot, func));
1047 		if (!pci_dev) {
1048 			dev_dbg(&pdev->xdev->dev,
1049 				"Cannot get PCI device %04x:%02x:%02x.%d\n",
1050 				domain, bus, slot, func);
1051 			continue;
1052 		}
1053 		pci_lock_rescan_remove();
1054 		pci_stop_and_remove_bus_device(pci_dev);
1055 		pci_dev_put(pci_dev);
1056 		pci_unlock_rescan_remove();
1057 
1058 		dev_dbg(&pdev->xdev->dev,
1059 			"PCI device %04x:%02x:%02x.%d removed.\n",
1060 			domain, bus, slot, func);
1061 	}
1062 
1063 	err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
1064 
1065 out:
1066 	return err;
1067 }
1068 
1069 static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev,
1070 						  enum xenbus_state be_state)
1071 {
1072 	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1073 
1074 	switch (be_state) {
1075 	case XenbusStateUnknown:
1076 	case XenbusStateInitialising:
1077 	case XenbusStateInitWait:
1078 	case XenbusStateInitialised:
1079 		break;
1080 
1081 	case XenbusStateConnected:
1082 		pcifront_try_connect(pdev);
1083 		break;
1084 
1085 	case XenbusStateClosed:
1086 		if (xdev->state == XenbusStateClosed)
1087 			break;
1088 		/* Missed the backend's CLOSING state -- fallthrough */
1089 	case XenbusStateClosing:
1090 		dev_warn(&xdev->dev, "backend going away!\n");
1091 		pcifront_try_disconnect(pdev);
1092 		break;
1093 
1094 	case XenbusStateReconfiguring:
1095 		pcifront_detach_devices(pdev);
1096 		break;
1097 
1098 	case XenbusStateReconfigured:
1099 		pcifront_attach_devices(pdev);
1100 		break;
1101 	}
1102 }
1103 
1104 static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1105 				 const struct xenbus_device_id *id)
1106 {
1107 	int err = 0;
1108 	struct pcifront_device *pdev = alloc_pdev(xdev);
1109 
1110 	if (pdev == NULL) {
1111 		err = -ENOMEM;
1112 		xenbus_dev_fatal(xdev, err,
1113 				 "Error allocating pcifront_device struct");
1114 		goto out;
1115 	}
1116 
1117 	err = pcifront_publish_info(pdev);
1118 	if (err)
1119 		free_pdev(pdev);
1120 
1121 out:
1122 	return err;
1123 }
1124 
1125 static int pcifront_xenbus_remove(struct xenbus_device *xdev)
1126 {
1127 	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1128 	if (pdev)
1129 		free_pdev(pdev);
1130 
1131 	return 0;
1132 }
1133 
1134 static const struct xenbus_device_id xenpci_ids[] = {
1135 	{"pci"},
1136 	{""},
1137 };
1138 
1139 static DEFINE_XENBUS_DRIVER(xenpci, "pcifront",
1140 	.probe			= pcifront_xenbus_probe,
1141 	.remove			= pcifront_xenbus_remove,
1142 	.otherend_changed	= pcifront_backend_changed,
1143 );
1144 
1145 static int __init pcifront_init(void)
1146 {
1147 	if (!xen_pv_domain() || xen_initial_domain())
1148 		return -ENODEV;
1149 
1150 	if (!xen_has_pv_devices())
1151 		return -ENODEV;
1152 
1153 	pci_frontend_registrar(1 /* enable */);
1154 
1155 	return xenbus_register_frontend(&xenpci_driver);
1156 }
1157 
1158 static void __exit pcifront_cleanup(void)
1159 {
1160 	xenbus_unregister_driver(&xenpci_driver);
1161 	pci_frontend_registrar(0 /* disable */);
1162 }
1163 module_init(pcifront_init);
1164 module_exit(pcifront_cleanup);
1165 
1166 MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1167 MODULE_LICENSE("GPL");
1168 MODULE_ALIAS("xen:pci");
1169