xref: /linux/drivers/pci/xen-pcifront.c (revision c4ee0af3fa0dc65f690fc908f02b8355f9576ea0)
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 
24 #include <asm/xen/swiotlb-xen.h>
25 #define INVALID_GRANT_REF (0)
26 #define INVALID_EVTCHN    (-1)
27 
28 struct pci_bus_entry {
29 	struct list_head list;
30 	struct pci_bus *bus;
31 };
32 
33 #define _PDEVB_op_active		(0)
34 #define PDEVB_op_active			(1 << (_PDEVB_op_active))
35 
36 struct pcifront_device {
37 	struct xenbus_device *xdev;
38 	struct list_head root_buses;
39 
40 	int evtchn;
41 	int gnt_ref;
42 
43 	int irq;
44 
45 	/* Lock this when doing any operations in sh_info */
46 	spinlock_t sh_info_lock;
47 	struct xen_pci_sharedinfo *sh_info;
48 	struct work_struct op_work;
49 	unsigned long flags;
50 
51 };
52 
53 struct pcifront_sd {
54 	int domain;
55 	struct pcifront_device *pdev;
56 };
57 
58 static inline struct pcifront_device *
59 pcifront_get_pdev(struct pcifront_sd *sd)
60 {
61 	return sd->pdev;
62 }
63 
64 static inline void pcifront_init_sd(struct pcifront_sd *sd,
65 				    unsigned int domain, unsigned int bus,
66 				    struct pcifront_device *pdev)
67 {
68 	sd->domain = domain;
69 	sd->pdev = pdev;
70 }
71 
72 static DEFINE_SPINLOCK(pcifront_dev_lock);
73 static struct pcifront_device *pcifront_dev;
74 
75 static int verbose_request;
76 module_param(verbose_request, int, 0644);
77 
78 static int errno_to_pcibios_err(int errno)
79 {
80 	switch (errno) {
81 	case XEN_PCI_ERR_success:
82 		return PCIBIOS_SUCCESSFUL;
83 
84 	case XEN_PCI_ERR_dev_not_found:
85 		return PCIBIOS_DEVICE_NOT_FOUND;
86 
87 	case XEN_PCI_ERR_invalid_offset:
88 	case XEN_PCI_ERR_op_failed:
89 		return PCIBIOS_BAD_REGISTER_NUMBER;
90 
91 	case XEN_PCI_ERR_not_implemented:
92 		return PCIBIOS_FUNC_NOT_SUPPORTED;
93 
94 	case XEN_PCI_ERR_access_denied:
95 		return PCIBIOS_SET_FAILED;
96 	}
97 	return errno;
98 }
99 
100 static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
101 {
102 	if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
103 		&& !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
104 		dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
105 		schedule_work(&pdev->op_work);
106 	}
107 }
108 
109 static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
110 {
111 	int err = 0;
112 	struct xen_pci_op *active_op = &pdev->sh_info->op;
113 	unsigned long irq_flags;
114 	evtchn_port_t port = pdev->evtchn;
115 	unsigned irq = pdev->irq;
116 	s64 ns, ns_timeout;
117 	struct timeval tv;
118 
119 	spin_lock_irqsave(&pdev->sh_info_lock, irq_flags);
120 
121 	memcpy(active_op, op, sizeof(struct xen_pci_op));
122 
123 	/* Go */
124 	wmb();
125 	set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
126 	notify_remote_via_evtchn(port);
127 
128 	/*
129 	 * We set a poll timeout of 3 seconds but give up on return after
130 	 * 2 seconds. It is better to time out too late rather than too early
131 	 * (in the latter case we end up continually re-executing poll() with a
132 	 * timeout in the past). 1s difference gives plenty of slack for error.
133 	 */
134 	do_gettimeofday(&tv);
135 	ns_timeout = timeval_to_ns(&tv) + 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 		do_gettimeofday(&tv);
144 		ns = timeval_to_ns(&tv);
145 		if (ns > ns_timeout) {
146 			dev_err(&pdev->xdev->dev,
147 				"pciback not responding!!!\n");
148 			clear_bit(_XEN_PCIF_active,
149 				  (unsigned long *)&pdev->sh_info->flags);
150 			err = XEN_PCI_ERR_dev_not_found;
151 			goto out;
152 		}
153 	}
154 
155 	/*
156 	* We might lose backend service request since we
157 	* reuse same evtchn with pci_conf backend response. So re-schedule
158 	* aer pcifront service.
159 	*/
160 	if (test_bit(_XEN_PCIB_active,
161 			(unsigned long *)&pdev->sh_info->flags)) {
162 		dev_err(&pdev->xdev->dev,
163 			"schedule aer pcifront service\n");
164 		schedule_pcifront_aer_op(pdev);
165 	}
166 
167 	memcpy(op, active_op, sizeof(struct xen_pci_op));
168 
169 	err = op->err;
170 out:
171 	spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags);
172 	return err;
173 }
174 
175 /* Access to this function is spinlocked in drivers/pci/access.c */
176 static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn,
177 			     int where, int size, u32 *val)
178 {
179 	int err = 0;
180 	struct xen_pci_op op = {
181 		.cmd    = XEN_PCI_OP_conf_read,
182 		.domain = pci_domain_nr(bus),
183 		.bus    = bus->number,
184 		.devfn  = devfn,
185 		.offset = where,
186 		.size   = size,
187 	};
188 	struct pcifront_sd *sd = bus->sysdata;
189 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
190 
191 	if (verbose_request)
192 		dev_info(&pdev->xdev->dev,
193 			 "read dev=%04x:%02x:%02x.%d - offset %x size %d\n",
194 			 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
195 			 PCI_FUNC(devfn), where, size);
196 
197 	err = do_pci_op(pdev, &op);
198 
199 	if (likely(!err)) {
200 		if (verbose_request)
201 			dev_info(&pdev->xdev->dev, "read got back value %x\n",
202 				 op.value);
203 
204 		*val = op.value;
205 	} else if (err == -ENODEV) {
206 		/* No device here, pretend that it just returned 0 */
207 		err = 0;
208 		*val = 0;
209 	}
210 
211 	return errno_to_pcibios_err(err);
212 }
213 
214 /* Access to this function is spinlocked in drivers/pci/access.c */
215 static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
216 			      int where, int size, u32 val)
217 {
218 	struct xen_pci_op op = {
219 		.cmd    = XEN_PCI_OP_conf_write,
220 		.domain = pci_domain_nr(bus),
221 		.bus    = bus->number,
222 		.devfn  = devfn,
223 		.offset = where,
224 		.size   = size,
225 		.value  = val,
226 	};
227 	struct pcifront_sd *sd = bus->sysdata;
228 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
229 
230 	if (verbose_request)
231 		dev_info(&pdev->xdev->dev,
232 			 "write dev=%04x:%02x:%02x.%d - "
233 			 "offset %x size %d val %x\n",
234 			 pci_domain_nr(bus), bus->number,
235 			 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
236 
237 	return errno_to_pcibios_err(do_pci_op(pdev, &op));
238 }
239 
240 static struct pci_ops pcifront_bus_ops = {
241 	.read = pcifront_bus_read,
242 	.write = pcifront_bus_write,
243 };
244 
245 #ifdef CONFIG_PCI_MSI
246 static int pci_frontend_enable_msix(struct pci_dev *dev,
247 				    int vector[], int nvec)
248 {
249 	int err;
250 	int i;
251 	struct xen_pci_op op = {
252 		.cmd    = XEN_PCI_OP_enable_msix,
253 		.domain = pci_domain_nr(dev->bus),
254 		.bus = dev->bus->number,
255 		.devfn = dev->devfn,
256 		.value = nvec,
257 	};
258 	struct pcifront_sd *sd = dev->bus->sysdata;
259 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
260 	struct msi_desc *entry;
261 
262 	if (nvec > SH_INFO_MAX_VEC) {
263 		dev_err(&dev->dev, "too much vector for pci frontend: %x."
264 				   " Increase SH_INFO_MAX_VEC.\n", nvec);
265 		return -EINVAL;
266 	}
267 
268 	i = 0;
269 	list_for_each_entry(entry, &dev->msi_list, list) {
270 		op.msix_entries[i].entry = entry->msi_attrib.entry_nr;
271 		/* Vector is useless at this point. */
272 		op.msix_entries[i].vector = -1;
273 		i++;
274 	}
275 
276 	err = do_pci_op(pdev, &op);
277 
278 	if (likely(!err)) {
279 		if (likely(!op.value)) {
280 			/* we get the result */
281 			for (i = 0; i < nvec; i++) {
282 				if (op.msix_entries[i].vector <= 0) {
283 					dev_warn(&dev->dev, "MSI-X entry %d is invalid: %d!\n",
284 						i, op.msix_entries[i].vector);
285 					err = -EINVAL;
286 					vector[i] = -1;
287 					continue;
288 				}
289 				vector[i] = op.msix_entries[i].vector;
290 			}
291 		} else {
292 			printk(KERN_DEBUG "enable msix get value %x\n",
293 				op.value);
294 			err = op.value;
295 		}
296 	} else {
297 		dev_err(&dev->dev, "enable msix get err %x\n", err);
298 	}
299 	return err;
300 }
301 
302 static void pci_frontend_disable_msix(struct pci_dev *dev)
303 {
304 	int err;
305 	struct xen_pci_op op = {
306 		.cmd    = XEN_PCI_OP_disable_msix,
307 		.domain = pci_domain_nr(dev->bus),
308 		.bus = dev->bus->number,
309 		.devfn = dev->devfn,
310 	};
311 	struct pcifront_sd *sd = dev->bus->sysdata;
312 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
313 
314 	err = do_pci_op(pdev, &op);
315 
316 	/* What should do for error ? */
317 	if (err)
318 		dev_err(&dev->dev, "pci_disable_msix get err %x\n", err);
319 }
320 
321 static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[])
322 {
323 	int err;
324 	struct xen_pci_op op = {
325 		.cmd    = XEN_PCI_OP_enable_msi,
326 		.domain = pci_domain_nr(dev->bus),
327 		.bus = dev->bus->number,
328 		.devfn = dev->devfn,
329 	};
330 	struct pcifront_sd *sd = dev->bus->sysdata;
331 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
332 
333 	err = do_pci_op(pdev, &op);
334 	if (likely(!err)) {
335 		vector[0] = op.value;
336 		if (op.value <= 0) {
337 			dev_warn(&dev->dev, "MSI entry is invalid: %d!\n",
338 				op.value);
339 			err = -EINVAL;
340 			vector[0] = -1;
341 		}
342 	} else {
343 		dev_err(&dev->dev, "pci frontend enable msi failed for dev "
344 				    "%x:%x\n", op.bus, op.devfn);
345 		err = -EINVAL;
346 	}
347 	return err;
348 }
349 
350 static void pci_frontend_disable_msi(struct pci_dev *dev)
351 {
352 	int err;
353 	struct xen_pci_op op = {
354 		.cmd    = XEN_PCI_OP_disable_msi,
355 		.domain = pci_domain_nr(dev->bus),
356 		.bus = dev->bus->number,
357 		.devfn = dev->devfn,
358 	};
359 	struct pcifront_sd *sd = dev->bus->sysdata;
360 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
361 
362 	err = do_pci_op(pdev, &op);
363 	if (err == XEN_PCI_ERR_dev_not_found) {
364 		/* XXX No response from backend, what shall we do? */
365 		printk(KERN_DEBUG "get no response from backend for disable MSI\n");
366 		return;
367 	}
368 	if (err)
369 		/* how can pciback notify us fail? */
370 		printk(KERN_DEBUG "get fake response frombackend\n");
371 }
372 
373 static struct xen_pci_frontend_ops pci_frontend_ops = {
374 	.enable_msi = pci_frontend_enable_msi,
375 	.disable_msi = pci_frontend_disable_msi,
376 	.enable_msix = pci_frontend_enable_msix,
377 	.disable_msix = pci_frontend_disable_msix,
378 };
379 
380 static void pci_frontend_registrar(int enable)
381 {
382 	if (enable)
383 		xen_pci_frontend = &pci_frontend_ops;
384 	else
385 		xen_pci_frontend = NULL;
386 };
387 #else
388 static inline void pci_frontend_registrar(int enable) { };
389 #endif /* CONFIG_PCI_MSI */
390 
391 /* Claim resources for the PCI frontend as-is, backend won't allow changes */
392 static int pcifront_claim_resource(struct pci_dev *dev, void *data)
393 {
394 	struct pcifront_device *pdev = data;
395 	int i;
396 	struct resource *r;
397 
398 	for (i = 0; i < PCI_NUM_RESOURCES; i++) {
399 		r = &dev->resource[i];
400 
401 		if (!r->parent && r->start && r->flags) {
402 			dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n",
403 				pci_name(dev), i);
404 			if (pci_claim_resource(dev, i)) {
405 				dev_err(&pdev->xdev->dev, "Could not claim resource %s/%d! "
406 					"Device offline. Try using e820_host=1 in the guest config.\n",
407 					pci_name(dev), i);
408 			}
409 		}
410 	}
411 
412 	return 0;
413 }
414 
415 static int pcifront_scan_bus(struct pcifront_device *pdev,
416 				unsigned int domain, unsigned int bus,
417 				struct pci_bus *b)
418 {
419 	struct pci_dev *d;
420 	unsigned int devfn;
421 
422 	/* Scan the bus for functions and add.
423 	 * We omit handling of PCI bridge attachment because pciback prevents
424 	 * bridges from being exported.
425 	 */
426 	for (devfn = 0; devfn < 0x100; devfn++) {
427 		d = pci_get_slot(b, devfn);
428 		if (d) {
429 			/* Device is already known. */
430 			pci_dev_put(d);
431 			continue;
432 		}
433 
434 		d = pci_scan_single_device(b, devfn);
435 		if (d)
436 			dev_info(&pdev->xdev->dev, "New device on "
437 				 "%04x:%02x:%02x.%d found.\n", domain, bus,
438 				 PCI_SLOT(devfn), PCI_FUNC(devfn));
439 	}
440 
441 	return 0;
442 }
443 
444 static int pcifront_scan_root(struct pcifront_device *pdev,
445 				 unsigned int domain, unsigned int bus)
446 {
447 	struct pci_bus *b;
448 	struct pcifront_sd *sd = NULL;
449 	struct pci_bus_entry *bus_entry = NULL;
450 	int err = 0;
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 = kmalloc(sizeof(*bus_entry), GFP_KERNEL);
467 	sd = kmalloc(sizeof(*sd), GFP_KERNEL);
468 	if (!bus_entry || !sd) {
469 		err = -ENOMEM;
470 		goto err_out;
471 	}
472 	pcifront_init_sd(sd, domain, bus, pdev);
473 
474 	b = pci_scan_bus_parented(&pdev->xdev->dev, bus,
475 				  &pcifront_bus_ops, sd);
476 	if (!b) {
477 		dev_err(&pdev->xdev->dev,
478 			"Error creating PCI Frontend Bus!\n");
479 		err = -ENOMEM;
480 		goto err_out;
481 	}
482 
483 	bus_entry->bus = b;
484 
485 	list_add(&bus_entry->list, &pdev->root_buses);
486 
487 	/* pci_scan_bus_parented skips devices which do not have a have
488 	* devfn==0. The pcifront_scan_bus enumerates all devfn. */
489 	err = pcifront_scan_bus(pdev, domain, bus, b);
490 
491 	/* Claim resources before going "live" with our devices */
492 	pci_walk_bus(b, pcifront_claim_resource, pdev);
493 
494 	/* Create SysFS and notify udev of the devices. Aka: "going live" */
495 	pci_bus_add_devices(b);
496 
497 	return err;
498 
499 err_out:
500 	kfree(bus_entry);
501 	kfree(sd);
502 
503 	return err;
504 }
505 
506 static int pcifront_rescan_root(struct pcifront_device *pdev,
507 				   unsigned int domain, unsigned int bus)
508 {
509 	int err;
510 	struct pci_bus *b;
511 
512 #ifndef CONFIG_PCI_DOMAINS
513 	if (domain != 0) {
514 		dev_err(&pdev->xdev->dev,
515 			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
516 		dev_err(&pdev->xdev->dev,
517 			"Please compile with CONFIG_PCI_DOMAINS\n");
518 		return -EINVAL;
519 	}
520 #endif
521 
522 	dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
523 		 domain, bus);
524 
525 	b = pci_find_bus(domain, bus);
526 	if (!b)
527 		/* If the bus is unknown, create it. */
528 		return pcifront_scan_root(pdev, domain, bus);
529 
530 	err = pcifront_scan_bus(pdev, domain, bus, b);
531 
532 	/* Claim resources before going "live" with our devices */
533 	pci_walk_bus(b, pcifront_claim_resource, pdev);
534 
535 	/* Create SysFS and notify udev of the devices. Aka: "going live" */
536 	pci_bus_add_devices(b);
537 
538 	return err;
539 }
540 
541 static void free_root_bus_devs(struct pci_bus *bus)
542 {
543 	struct pci_dev *dev;
544 
545 	while (!list_empty(&bus->devices)) {
546 		dev = container_of(bus->devices.next, struct pci_dev,
547 				   bus_list);
548 		dev_dbg(&dev->dev, "removing device\n");
549 		pci_stop_and_remove_bus_device(dev);
550 	}
551 }
552 
553 static void pcifront_free_roots(struct pcifront_device *pdev)
554 {
555 	struct pci_bus_entry *bus_entry, *t;
556 
557 	dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
558 
559 	list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
560 		list_del(&bus_entry->list);
561 
562 		free_root_bus_devs(bus_entry->bus);
563 
564 		kfree(bus_entry->bus->sysdata);
565 
566 		device_unregister(bus_entry->bus->bridge);
567 		pci_remove_bus(bus_entry->bus);
568 
569 		kfree(bus_entry);
570 	}
571 }
572 
573 static pci_ers_result_t pcifront_common_process(int cmd,
574 						struct pcifront_device *pdev,
575 						pci_channel_state_t state)
576 {
577 	pci_ers_result_t result;
578 	struct pci_driver *pdrv;
579 	int bus = pdev->sh_info->aer_op.bus;
580 	int devfn = pdev->sh_info->aer_op.devfn;
581 	struct pci_dev *pcidev;
582 	int flag = 0;
583 
584 	dev_dbg(&pdev->xdev->dev,
585 		"pcifront AER process: cmd %x (bus:%x, devfn%x)",
586 		cmd, bus, devfn);
587 	result = PCI_ERS_RESULT_NONE;
588 
589 	pcidev = pci_get_bus_and_slot(bus, devfn);
590 	if (!pcidev || !pcidev->driver) {
591 		dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
592 		if (pcidev)
593 			pci_dev_put(pcidev);
594 		return result;
595 	}
596 	pdrv = pcidev->driver;
597 
598 	if (pdrv) {
599 		if (pdrv->err_handler && pdrv->err_handler->error_detected) {
600 			dev_dbg(&pcidev->dev,
601 				"trying to call AER service\n");
602 			if (pcidev) {
603 				flag = 1;
604 				switch (cmd) {
605 				case XEN_PCI_OP_aer_detected:
606 					result = pdrv->err_handler->
607 						 error_detected(pcidev, state);
608 					break;
609 				case XEN_PCI_OP_aer_mmio:
610 					result = pdrv->err_handler->
611 						 mmio_enabled(pcidev);
612 					break;
613 				case XEN_PCI_OP_aer_slotreset:
614 					result = pdrv->err_handler->
615 						 slot_reset(pcidev);
616 					break;
617 				case XEN_PCI_OP_aer_resume:
618 					pdrv->err_handler->resume(pcidev);
619 					break;
620 				default:
621 					dev_err(&pdev->xdev->dev,
622 						"bad request in aer recovery "
623 						"operation!\n");
624 
625 				}
626 			}
627 		}
628 	}
629 	if (!flag)
630 		result = PCI_ERS_RESULT_NONE;
631 
632 	return result;
633 }
634 
635 
636 static void pcifront_do_aer(struct work_struct *data)
637 {
638 	struct pcifront_device *pdev =
639 		container_of(data, struct pcifront_device, op_work);
640 	int cmd = pdev->sh_info->aer_op.cmd;
641 	pci_channel_state_t state =
642 		(pci_channel_state_t)pdev->sh_info->aer_op.err;
643 
644 	/*If a pci_conf op is in progress,
645 		we have to wait until it is done before service aer op*/
646 	dev_dbg(&pdev->xdev->dev,
647 		"pcifront service aer bus %x devfn %x\n",
648 		pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
649 
650 	pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
651 
652 	/* Post the operation to the guest. */
653 	wmb();
654 	clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
655 	notify_remote_via_evtchn(pdev->evtchn);
656 
657 	/*in case of we lost an aer request in four lines time_window*/
658 	smp_mb__before_clear_bit();
659 	clear_bit(_PDEVB_op_active, &pdev->flags);
660 	smp_mb__after_clear_bit();
661 
662 	schedule_pcifront_aer_op(pdev);
663 
664 }
665 
666 static irqreturn_t pcifront_handler_aer(int irq, void *dev)
667 {
668 	struct pcifront_device *pdev = dev;
669 	schedule_pcifront_aer_op(pdev);
670 	return IRQ_HANDLED;
671 }
672 static int pcifront_connect_and_init_dma(struct pcifront_device *pdev)
673 {
674 	int err = 0;
675 
676 	spin_lock(&pcifront_dev_lock);
677 
678 	if (!pcifront_dev) {
679 		dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
680 		pcifront_dev = pdev;
681 	} else
682 		err = -EEXIST;
683 
684 	spin_unlock(&pcifront_dev_lock);
685 
686 	if (!err && !swiotlb_nr_tbl()) {
687 		err = pci_xen_swiotlb_init_late();
688 		if (err)
689 			dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n");
690 	}
691 	return err;
692 }
693 
694 static void pcifront_disconnect(struct pcifront_device *pdev)
695 {
696 	spin_lock(&pcifront_dev_lock);
697 
698 	if (pdev == pcifront_dev) {
699 		dev_info(&pdev->xdev->dev,
700 			 "Disconnecting PCI Frontend Buses\n");
701 		pcifront_dev = NULL;
702 	}
703 
704 	spin_unlock(&pcifront_dev_lock);
705 }
706 static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
707 {
708 	struct pcifront_device *pdev;
709 
710 	pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
711 	if (pdev == NULL)
712 		goto out;
713 
714 	pdev->sh_info =
715 	    (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL);
716 	if (pdev->sh_info == NULL) {
717 		kfree(pdev);
718 		pdev = NULL;
719 		goto out;
720 	}
721 	pdev->sh_info->flags = 0;
722 
723 	/*Flag for registering PV AER handler*/
724 	set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
725 
726 	dev_set_drvdata(&xdev->dev, pdev);
727 	pdev->xdev = xdev;
728 
729 	INIT_LIST_HEAD(&pdev->root_buses);
730 
731 	spin_lock_init(&pdev->sh_info_lock);
732 
733 	pdev->evtchn = INVALID_EVTCHN;
734 	pdev->gnt_ref = INVALID_GRANT_REF;
735 	pdev->irq = -1;
736 
737 	INIT_WORK(&pdev->op_work, pcifront_do_aer);
738 
739 	dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
740 		pdev, pdev->sh_info);
741 out:
742 	return pdev;
743 }
744 
745 static void free_pdev(struct pcifront_device *pdev)
746 {
747 	dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
748 
749 	pcifront_free_roots(pdev);
750 
751 	cancel_work_sync(&pdev->op_work);
752 
753 	if (pdev->irq >= 0)
754 		unbind_from_irqhandler(pdev->irq, pdev);
755 
756 	if (pdev->evtchn != INVALID_EVTCHN)
757 		xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
758 
759 	if (pdev->gnt_ref != INVALID_GRANT_REF)
760 		gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */,
761 					  (unsigned long)pdev->sh_info);
762 	else
763 		free_page((unsigned long)pdev->sh_info);
764 
765 	dev_set_drvdata(&pdev->xdev->dev, NULL);
766 
767 	kfree(pdev);
768 }
769 
770 static int pcifront_publish_info(struct pcifront_device *pdev)
771 {
772 	int err = 0;
773 	struct xenbus_transaction trans;
774 
775 	err = xenbus_grant_ring(pdev->xdev, virt_to_mfn(pdev->sh_info));
776 	if (err < 0)
777 		goto out;
778 
779 	pdev->gnt_ref = err;
780 
781 	err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
782 	if (err)
783 		goto out;
784 
785 	err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
786 		0, "pcifront", pdev);
787 
788 	if (err < 0)
789 		return err;
790 
791 	pdev->irq = err;
792 
793 do_publish:
794 	err = xenbus_transaction_start(&trans);
795 	if (err) {
796 		xenbus_dev_fatal(pdev->xdev, err,
797 				 "Error writing configuration for backend "
798 				 "(start transaction)");
799 		goto out;
800 	}
801 
802 	err = xenbus_printf(trans, pdev->xdev->nodename,
803 			    "pci-op-ref", "%u", pdev->gnt_ref);
804 	if (!err)
805 		err = xenbus_printf(trans, pdev->xdev->nodename,
806 				    "event-channel", "%u", pdev->evtchn);
807 	if (!err)
808 		err = xenbus_printf(trans, pdev->xdev->nodename,
809 				    "magic", XEN_PCI_MAGIC);
810 
811 	if (err) {
812 		xenbus_transaction_end(trans, 1);
813 		xenbus_dev_fatal(pdev->xdev, err,
814 				 "Error writing configuration for backend");
815 		goto out;
816 	} else {
817 		err = xenbus_transaction_end(trans, 0);
818 		if (err == -EAGAIN)
819 			goto do_publish;
820 		else if (err) {
821 			xenbus_dev_fatal(pdev->xdev, err,
822 					 "Error completing transaction "
823 					 "for backend");
824 			goto out;
825 		}
826 	}
827 
828 	xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
829 
830 	dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
831 
832 out:
833 	return err;
834 }
835 
836 static int pcifront_try_connect(struct pcifront_device *pdev)
837 {
838 	int err = -EFAULT;
839 	int i, num_roots, len;
840 	char str[64];
841 	unsigned int domain, bus;
842 
843 
844 	/* Only connect once */
845 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
846 	    XenbusStateInitialised)
847 		goto out;
848 
849 	err = pcifront_connect_and_init_dma(pdev);
850 	if (err && err != -EEXIST) {
851 		xenbus_dev_fatal(pdev->xdev, err,
852 				 "Error setting up PCI Frontend");
853 		goto out;
854 	}
855 
856 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
857 			   "root_num", "%d", &num_roots);
858 	if (err == -ENOENT) {
859 		xenbus_dev_error(pdev->xdev, err,
860 				 "No PCI Roots found, trying 0000:00");
861 		err = pcifront_scan_root(pdev, 0, 0);
862 		num_roots = 0;
863 	} else if (err != 1) {
864 		if (err == 0)
865 			err = -EINVAL;
866 		xenbus_dev_fatal(pdev->xdev, err,
867 				 "Error reading number of PCI roots");
868 		goto out;
869 	}
870 
871 	for (i = 0; i < num_roots; i++) {
872 		len = snprintf(str, sizeof(str), "root-%d", i);
873 		if (unlikely(len >= (sizeof(str) - 1))) {
874 			err = -ENOMEM;
875 			goto out;
876 		}
877 
878 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
879 				   "%x:%x", &domain, &bus);
880 		if (err != 2) {
881 			if (err >= 0)
882 				err = -EINVAL;
883 			xenbus_dev_fatal(pdev->xdev, err,
884 					 "Error reading PCI root %d", i);
885 			goto out;
886 		}
887 
888 		err = pcifront_scan_root(pdev, domain, bus);
889 		if (err) {
890 			xenbus_dev_fatal(pdev->xdev, err,
891 					 "Error scanning PCI root %04x:%02x",
892 					 domain, bus);
893 			goto out;
894 		}
895 	}
896 
897 	err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
898 
899 out:
900 	return err;
901 }
902 
903 static int pcifront_try_disconnect(struct pcifront_device *pdev)
904 {
905 	int err = 0;
906 	enum xenbus_state prev_state;
907 
908 
909 	prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
910 
911 	if (prev_state >= XenbusStateClosing)
912 		goto out;
913 
914 	if (prev_state == XenbusStateConnected) {
915 		pcifront_free_roots(pdev);
916 		pcifront_disconnect(pdev);
917 	}
918 
919 	err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
920 
921 out:
922 
923 	return err;
924 }
925 
926 static int pcifront_attach_devices(struct pcifront_device *pdev)
927 {
928 	int err = -EFAULT;
929 	int i, num_roots, len;
930 	unsigned int domain, bus;
931 	char str[64];
932 
933 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
934 	    XenbusStateReconfiguring)
935 		goto out;
936 
937 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
938 			   "root_num", "%d", &num_roots);
939 	if (err == -ENOENT) {
940 		xenbus_dev_error(pdev->xdev, err,
941 				 "No PCI Roots found, trying 0000:00");
942 		err = pcifront_rescan_root(pdev, 0, 0);
943 		num_roots = 0;
944 	} else if (err != 1) {
945 		if (err == 0)
946 			err = -EINVAL;
947 		xenbus_dev_fatal(pdev->xdev, err,
948 				 "Error reading number of PCI roots");
949 		goto out;
950 	}
951 
952 	for (i = 0; i < num_roots; i++) {
953 		len = snprintf(str, sizeof(str), "root-%d", i);
954 		if (unlikely(len >= (sizeof(str) - 1))) {
955 			err = -ENOMEM;
956 			goto out;
957 		}
958 
959 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
960 				   "%x:%x", &domain, &bus);
961 		if (err != 2) {
962 			if (err >= 0)
963 				err = -EINVAL;
964 			xenbus_dev_fatal(pdev->xdev, err,
965 					 "Error reading PCI root %d", i);
966 			goto out;
967 		}
968 
969 		err = pcifront_rescan_root(pdev, domain, bus);
970 		if (err) {
971 			xenbus_dev_fatal(pdev->xdev, err,
972 					 "Error scanning PCI root %04x:%02x",
973 					 domain, bus);
974 			goto out;
975 		}
976 	}
977 
978 	xenbus_switch_state(pdev->xdev, XenbusStateConnected);
979 
980 out:
981 	return err;
982 }
983 
984 static int pcifront_detach_devices(struct pcifront_device *pdev)
985 {
986 	int err = 0;
987 	int i, num_devs;
988 	unsigned int domain, bus, slot, func;
989 	struct pci_dev *pci_dev;
990 	char str[64];
991 
992 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
993 	    XenbusStateConnected)
994 		goto out;
995 
996 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
997 			   &num_devs);
998 	if (err != 1) {
999 		if (err >= 0)
1000 			err = -EINVAL;
1001 		xenbus_dev_fatal(pdev->xdev, err,
1002 				 "Error reading number of PCI devices");
1003 		goto out;
1004 	}
1005 
1006 	/* Find devices being detached and remove them. */
1007 	for (i = 0; i < num_devs; i++) {
1008 		int l, state;
1009 		l = snprintf(str, sizeof(str), "state-%d", i);
1010 		if (unlikely(l >= (sizeof(str) - 1))) {
1011 			err = -ENOMEM;
1012 			goto out;
1013 		}
1014 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d",
1015 				   &state);
1016 		if (err != 1)
1017 			state = XenbusStateUnknown;
1018 
1019 		if (state != XenbusStateClosing)
1020 			continue;
1021 
1022 		/* Remove device. */
1023 		l = snprintf(str, sizeof(str), "vdev-%d", i);
1024 		if (unlikely(l >= (sizeof(str) - 1))) {
1025 			err = -ENOMEM;
1026 			goto out;
1027 		}
1028 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
1029 				   "%x:%x:%x.%x", &domain, &bus, &slot, &func);
1030 		if (err != 4) {
1031 			if (err >= 0)
1032 				err = -EINVAL;
1033 			xenbus_dev_fatal(pdev->xdev, err,
1034 					 "Error reading PCI device %d", i);
1035 			goto out;
1036 		}
1037 
1038 		pci_dev = pci_get_domain_bus_and_slot(domain, bus,
1039 				PCI_DEVFN(slot, func));
1040 		if (!pci_dev) {
1041 			dev_dbg(&pdev->xdev->dev,
1042 				"Cannot get PCI device %04x:%02x:%02x.%d\n",
1043 				domain, bus, slot, func);
1044 			continue;
1045 		}
1046 		pci_stop_and_remove_bus_device(pci_dev);
1047 		pci_dev_put(pci_dev);
1048 
1049 		dev_dbg(&pdev->xdev->dev,
1050 			"PCI device %04x:%02x:%02x.%d removed.\n",
1051 			domain, bus, slot, func);
1052 	}
1053 
1054 	err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
1055 
1056 out:
1057 	return err;
1058 }
1059 
1060 static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev,
1061 						  enum xenbus_state be_state)
1062 {
1063 	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1064 
1065 	switch (be_state) {
1066 	case XenbusStateUnknown:
1067 	case XenbusStateInitialising:
1068 	case XenbusStateInitWait:
1069 	case XenbusStateInitialised:
1070 		break;
1071 
1072 	case XenbusStateConnected:
1073 		pcifront_try_connect(pdev);
1074 		break;
1075 
1076 	case XenbusStateClosed:
1077 		if (xdev->state == XenbusStateClosed)
1078 			break;
1079 		/* Missed the backend's CLOSING state -- fallthrough */
1080 	case XenbusStateClosing:
1081 		dev_warn(&xdev->dev, "backend going away!\n");
1082 		pcifront_try_disconnect(pdev);
1083 		break;
1084 
1085 	case XenbusStateReconfiguring:
1086 		pcifront_detach_devices(pdev);
1087 		break;
1088 
1089 	case XenbusStateReconfigured:
1090 		pcifront_attach_devices(pdev);
1091 		break;
1092 	}
1093 }
1094 
1095 static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1096 				 const struct xenbus_device_id *id)
1097 {
1098 	int err = 0;
1099 	struct pcifront_device *pdev = alloc_pdev(xdev);
1100 
1101 	if (pdev == NULL) {
1102 		err = -ENOMEM;
1103 		xenbus_dev_fatal(xdev, err,
1104 				 "Error allocating pcifront_device struct");
1105 		goto out;
1106 	}
1107 
1108 	err = pcifront_publish_info(pdev);
1109 	if (err)
1110 		free_pdev(pdev);
1111 
1112 out:
1113 	return err;
1114 }
1115 
1116 static int pcifront_xenbus_remove(struct xenbus_device *xdev)
1117 {
1118 	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1119 	if (pdev)
1120 		free_pdev(pdev);
1121 
1122 	return 0;
1123 }
1124 
1125 static const struct xenbus_device_id xenpci_ids[] = {
1126 	{"pci"},
1127 	{""},
1128 };
1129 
1130 static DEFINE_XENBUS_DRIVER(xenpci, "pcifront",
1131 	.probe			= pcifront_xenbus_probe,
1132 	.remove			= pcifront_xenbus_remove,
1133 	.otherend_changed	= pcifront_backend_changed,
1134 );
1135 
1136 static int __init pcifront_init(void)
1137 {
1138 	if (!xen_pv_domain() || xen_initial_domain())
1139 		return -ENODEV;
1140 
1141 	pci_frontend_registrar(1 /* enable */);
1142 
1143 	return xenbus_register_frontend(&xenpci_driver);
1144 }
1145 
1146 static void __exit pcifront_cleanup(void)
1147 {
1148 	xenbus_unregister_driver(&xenpci_driver);
1149 	pci_frontend_registrar(0 /* disable */);
1150 }
1151 module_init(pcifront_init);
1152 module_exit(pcifront_cleanup);
1153 
1154 MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1155 MODULE_LICENSE("GPL");
1156 MODULE_ALIAS("xen:pci");
1157