xref: /openbmc/linux/drivers/pci/xen-pcifront.c (revision 81de3bf3)
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 verbose_request;
81 module_param(verbose_request, int, 0644);
82 
83 static int errno_to_pcibios_err(int errno)
84 {
85 	switch (errno) {
86 	case XEN_PCI_ERR_success:
87 		return PCIBIOS_SUCCESSFUL;
88 
89 	case XEN_PCI_ERR_dev_not_found:
90 		return PCIBIOS_DEVICE_NOT_FOUND;
91 
92 	case XEN_PCI_ERR_invalid_offset:
93 	case XEN_PCI_ERR_op_failed:
94 		return PCIBIOS_BAD_REGISTER_NUMBER;
95 
96 	case XEN_PCI_ERR_not_implemented:
97 		return PCIBIOS_FUNC_NOT_SUPPORTED;
98 
99 	case XEN_PCI_ERR_access_denied:
100 		return PCIBIOS_SET_FAILED;
101 	}
102 	return errno;
103 }
104 
105 static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
106 {
107 	if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
108 		&& !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
109 		dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
110 		schedule_work(&pdev->op_work);
111 	}
112 }
113 
114 static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
115 {
116 	int err = 0;
117 	struct xen_pci_op *active_op = &pdev->sh_info->op;
118 	unsigned long irq_flags;
119 	evtchn_port_t port = pdev->evtchn;
120 	unsigned irq = pdev->irq;
121 	s64 ns, ns_timeout;
122 
123 	spin_lock_irqsave(&pdev->sh_info_lock, irq_flags);
124 
125 	memcpy(active_op, op, sizeof(struct xen_pci_op));
126 
127 	/* Go */
128 	wmb();
129 	set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
130 	notify_remote_via_evtchn(port);
131 
132 	/*
133 	 * We set a poll timeout of 3 seconds but give up on return after
134 	 * 2 seconds. It is better to time out too late rather than too early
135 	 * (in the latter case we end up continually re-executing poll() with a
136 	 * timeout in the past). 1s difference gives plenty of slack for error.
137 	 */
138 	ns_timeout = ktime_get_ns() + 2 * (s64)NSEC_PER_SEC;
139 
140 	xen_clear_irq_pending(irq);
141 
142 	while (test_bit(_XEN_PCIF_active,
143 			(unsigned long *)&pdev->sh_info->flags)) {
144 		xen_poll_irq_timeout(irq, jiffies + 3*HZ);
145 		xen_clear_irq_pending(irq);
146 		ns = ktime_get_ns();
147 		if (ns > ns_timeout) {
148 			dev_err(&pdev->xdev->dev,
149 				"pciback not responding!!!\n");
150 			clear_bit(_XEN_PCIF_active,
151 				  (unsigned long *)&pdev->sh_info->flags);
152 			err = XEN_PCI_ERR_dev_not_found;
153 			goto out;
154 		}
155 	}
156 
157 	/*
158 	* We might lose backend service request since we
159 	* reuse same evtchn with pci_conf backend response. So re-schedule
160 	* aer pcifront service.
161 	*/
162 	if (test_bit(_XEN_PCIB_active,
163 			(unsigned long *)&pdev->sh_info->flags)) {
164 		dev_err(&pdev->xdev->dev,
165 			"schedule aer pcifront service\n");
166 		schedule_pcifront_aer_op(pdev);
167 	}
168 
169 	memcpy(op, active_op, sizeof(struct xen_pci_op));
170 
171 	err = op->err;
172 out:
173 	spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags);
174 	return err;
175 }
176 
177 /* Access to this function is spinlocked in drivers/pci/access.c */
178 static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn,
179 			     int where, int size, u32 *val)
180 {
181 	int err = 0;
182 	struct xen_pci_op op = {
183 		.cmd    = XEN_PCI_OP_conf_read,
184 		.domain = pci_domain_nr(bus),
185 		.bus    = bus->number,
186 		.devfn  = devfn,
187 		.offset = where,
188 		.size   = size,
189 	};
190 	struct pcifront_sd *sd = bus->sysdata;
191 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
192 
193 	if (verbose_request)
194 		dev_info(&pdev->xdev->dev,
195 			 "read dev=%04x:%02x:%02x.%d - offset %x size %d\n",
196 			 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
197 			 PCI_FUNC(devfn), where, size);
198 
199 	err = do_pci_op(pdev, &op);
200 
201 	if (likely(!err)) {
202 		if (verbose_request)
203 			dev_info(&pdev->xdev->dev, "read got back value %x\n",
204 				 op.value);
205 
206 		*val = op.value;
207 	} else if (err == -ENODEV) {
208 		/* No device here, pretend that it just returned 0 */
209 		err = 0;
210 		*val = 0;
211 	}
212 
213 	return errno_to_pcibios_err(err);
214 }
215 
216 /* Access to this function is spinlocked in drivers/pci/access.c */
217 static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
218 			      int where, int size, u32 val)
219 {
220 	struct xen_pci_op op = {
221 		.cmd    = XEN_PCI_OP_conf_write,
222 		.domain = pci_domain_nr(bus),
223 		.bus    = bus->number,
224 		.devfn  = devfn,
225 		.offset = where,
226 		.size   = size,
227 		.value  = val,
228 	};
229 	struct pcifront_sd *sd = bus->sysdata;
230 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
231 
232 	if (verbose_request)
233 		dev_info(&pdev->xdev->dev,
234 			 "write dev=%04x:%02x:%02x.%d - "
235 			 "offset %x size %d val %x\n",
236 			 pci_domain_nr(bus), bus->number,
237 			 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
238 
239 	return errno_to_pcibios_err(do_pci_op(pdev, &op));
240 }
241 
242 static struct pci_ops pcifront_bus_ops = {
243 	.read = pcifront_bus_read,
244 	.write = pcifront_bus_write,
245 };
246 
247 #ifdef CONFIG_PCI_MSI
248 static int pci_frontend_enable_msix(struct pci_dev *dev,
249 				    int vector[], int nvec)
250 {
251 	int err;
252 	int i;
253 	struct xen_pci_op op = {
254 		.cmd    = XEN_PCI_OP_enable_msix,
255 		.domain = pci_domain_nr(dev->bus),
256 		.bus = dev->bus->number,
257 		.devfn = dev->devfn,
258 		.value = nvec,
259 	};
260 	struct pcifront_sd *sd = dev->bus->sysdata;
261 	struct pcifront_device *pdev = pcifront_get_pdev(sd);
262 	struct msi_desc *entry;
263 
264 	if (nvec > SH_INFO_MAX_VEC) {
265 		pci_err(dev, "too many vectors (0x%x) for PCI frontend:"
266 				   " Increase SH_INFO_MAX_VEC\n", nvec);
267 		return -EINVAL;
268 	}
269 
270 	i = 0;
271 	for_each_pci_msi_entry(entry, dev) {
272 		op.msix_entries[i].entry = entry->msi_attrib.entry_nr;
273 		/* Vector is useless at this point. */
274 		op.msix_entries[i].vector = -1;
275 		i++;
276 	}
277 
278 	err = do_pci_op(pdev, &op);
279 
280 	if (likely(!err)) {
281 		if (likely(!op.value)) {
282 			/* we get the result */
283 			for (i = 0; i < nvec; i++) {
284 				if (op.msix_entries[i].vector <= 0) {
285 					pci_warn(dev, "MSI-X entry %d is invalid: %d!\n",
286 						i, op.msix_entries[i].vector);
287 					err = -EINVAL;
288 					vector[i] = -1;
289 					continue;
290 				}
291 				vector[i] = op.msix_entries[i].vector;
292 			}
293 		} else {
294 			pr_info("enable msix get value %x\n", op.value);
295 			err = op.value;
296 		}
297 	} else {
298 		pci_err(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 		pci_err(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 			pci_warn(dev, "MSI entry is invalid: %d!\n",
339 				op.value);
340 			err = -EINVAL;
341 			vector[0] = -1;
342 		}
343 	} else {
344 		pci_err(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 		pr_info("get no response from backend for disable MSI\n");
367 		return;
368 	}
369 	if (err)
370 		/* how can pciback notify us fail? */
371 		pr_info("get fake response from backend\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 	LIST_HEAD(resources);
450 	struct pcifront_sd *sd = NULL;
451 	struct pci_bus_entry *bus_entry = NULL;
452 	int err = 0;
453 	static struct resource busn_res = {
454 		.start = 0,
455 		.end = 255,
456 		.flags = IORESOURCE_BUS,
457 	};
458 
459 #ifndef CONFIG_PCI_DOMAINS
460 	if (domain != 0) {
461 		dev_err(&pdev->xdev->dev,
462 			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
463 		dev_err(&pdev->xdev->dev,
464 			"Please compile with CONFIG_PCI_DOMAINS\n");
465 		err = -EINVAL;
466 		goto err_out;
467 	}
468 #endif
469 
470 	dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n",
471 		 domain, bus);
472 
473 	bus_entry = kzalloc(sizeof(*bus_entry), GFP_KERNEL);
474 	sd = kzalloc(sizeof(*sd), GFP_KERNEL);
475 	if (!bus_entry || !sd) {
476 		err = -ENOMEM;
477 		goto err_out;
478 	}
479 	pci_add_resource(&resources, &ioport_resource);
480 	pci_add_resource(&resources, &iomem_resource);
481 	pci_add_resource(&resources, &busn_res);
482 	pcifront_init_sd(sd, domain, bus, pdev);
483 
484 	pci_lock_rescan_remove();
485 
486 	b = pci_scan_root_bus(&pdev->xdev->dev, bus,
487 				  &pcifront_bus_ops, sd, &resources);
488 	if (!b) {
489 		dev_err(&pdev->xdev->dev,
490 			"Error creating PCI Frontend Bus!\n");
491 		err = -ENOMEM;
492 		pci_unlock_rescan_remove();
493 		pci_free_resource_list(&resources);
494 		goto err_out;
495 	}
496 
497 	bus_entry->bus = b;
498 
499 	list_add(&bus_entry->list, &pdev->root_buses);
500 
501 	/* pci_scan_root_bus skips devices which do not have a
502 	* devfn==0. The pcifront_scan_bus enumerates all devfn. */
503 	err = pcifront_scan_bus(pdev, domain, bus, b);
504 
505 	/* Claim resources before going "live" with our devices */
506 	pci_walk_bus(b, pcifront_claim_resource, pdev);
507 
508 	/* Create SysFS and notify udev of the devices. Aka: "going live" */
509 	pci_bus_add_devices(b);
510 
511 	pci_unlock_rescan_remove();
512 	return err;
513 
514 err_out:
515 	kfree(bus_entry);
516 	kfree(sd);
517 
518 	return err;
519 }
520 
521 static int pcifront_rescan_root(struct pcifront_device *pdev,
522 				   unsigned int domain, unsigned int bus)
523 {
524 	int err;
525 	struct pci_bus *b;
526 
527 #ifndef CONFIG_PCI_DOMAINS
528 	if (domain != 0) {
529 		dev_err(&pdev->xdev->dev,
530 			"PCI Root in non-zero PCI Domain! domain=%d\n", domain);
531 		dev_err(&pdev->xdev->dev,
532 			"Please compile with CONFIG_PCI_DOMAINS\n");
533 		return -EINVAL;
534 	}
535 #endif
536 
537 	dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
538 		 domain, bus);
539 
540 	b = pci_find_bus(domain, bus);
541 	if (!b)
542 		/* If the bus is unknown, create it. */
543 		return pcifront_scan_root(pdev, domain, bus);
544 
545 	err = pcifront_scan_bus(pdev, domain, bus, b);
546 
547 	/* Claim resources before going "live" with our devices */
548 	pci_walk_bus(b, pcifront_claim_resource, pdev);
549 
550 	/* Create SysFS and notify udev of the devices. Aka: "going live" */
551 	pci_bus_add_devices(b);
552 
553 	return err;
554 }
555 
556 static void free_root_bus_devs(struct pci_bus *bus)
557 {
558 	struct pci_dev *dev;
559 
560 	while (!list_empty(&bus->devices)) {
561 		dev = container_of(bus->devices.next, struct pci_dev,
562 				   bus_list);
563 		pci_dbg(dev, "removing device\n");
564 		pci_stop_and_remove_bus_device(dev);
565 	}
566 }
567 
568 static void pcifront_free_roots(struct pcifront_device *pdev)
569 {
570 	struct pci_bus_entry *bus_entry, *t;
571 
572 	dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
573 
574 	pci_lock_rescan_remove();
575 	list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
576 		list_del(&bus_entry->list);
577 
578 		free_root_bus_devs(bus_entry->bus);
579 
580 		kfree(bus_entry->bus->sysdata);
581 
582 		device_unregister(bus_entry->bus->bridge);
583 		pci_remove_bus(bus_entry->bus);
584 
585 		kfree(bus_entry);
586 	}
587 	pci_unlock_rescan_remove();
588 }
589 
590 static pci_ers_result_t pcifront_common_process(int cmd,
591 						struct pcifront_device *pdev,
592 						pci_channel_state_t state)
593 {
594 	pci_ers_result_t result;
595 	struct pci_driver *pdrv;
596 	int bus = pdev->sh_info->aer_op.bus;
597 	int devfn = pdev->sh_info->aer_op.devfn;
598 	int domain = pdev->sh_info->aer_op.domain;
599 	struct pci_dev *pcidev;
600 	int flag = 0;
601 
602 	dev_dbg(&pdev->xdev->dev,
603 		"pcifront AER process: cmd %x (bus:%x, devfn%x)",
604 		cmd, bus, devfn);
605 	result = PCI_ERS_RESULT_NONE;
606 
607 	pcidev = pci_get_domain_bus_and_slot(domain, bus, devfn);
608 	if (!pcidev || !pcidev->driver) {
609 		dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
610 		pci_dev_put(pcidev);
611 		return result;
612 	}
613 	pdrv = pcidev->driver;
614 
615 	if (pdrv) {
616 		if (pdrv->err_handler && pdrv->err_handler->error_detected) {
617 			pci_dbg(pcidev, "trying to call AER service\n");
618 			if (pcidev) {
619 				flag = 1;
620 				switch (cmd) {
621 				case XEN_PCI_OP_aer_detected:
622 					result = pdrv->err_handler->
623 						 error_detected(pcidev, state);
624 					break;
625 				case XEN_PCI_OP_aer_mmio:
626 					result = pdrv->err_handler->
627 						 mmio_enabled(pcidev);
628 					break;
629 				case XEN_PCI_OP_aer_slotreset:
630 					result = pdrv->err_handler->
631 						 slot_reset(pcidev);
632 					break;
633 				case XEN_PCI_OP_aer_resume:
634 					pdrv->err_handler->resume(pcidev);
635 					break;
636 				default:
637 					dev_err(&pdev->xdev->dev,
638 						"bad request in aer recovery "
639 						"operation!\n");
640 
641 				}
642 			}
643 		}
644 	}
645 	if (!flag)
646 		result = PCI_ERS_RESULT_NONE;
647 
648 	return result;
649 }
650 
651 
652 static void pcifront_do_aer(struct work_struct *data)
653 {
654 	struct pcifront_device *pdev =
655 		container_of(data, struct pcifront_device, op_work);
656 	int cmd = pdev->sh_info->aer_op.cmd;
657 	pci_channel_state_t state =
658 		(pci_channel_state_t)pdev->sh_info->aer_op.err;
659 
660 	/*If a pci_conf op is in progress,
661 		we have to wait until it is done before service aer op*/
662 	dev_dbg(&pdev->xdev->dev,
663 		"pcifront service aer bus %x devfn %x\n",
664 		pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
665 
666 	pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
667 
668 	/* Post the operation to the guest. */
669 	wmb();
670 	clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
671 	notify_remote_via_evtchn(pdev->evtchn);
672 
673 	/*in case of we lost an aer request in four lines time_window*/
674 	smp_mb__before_atomic();
675 	clear_bit(_PDEVB_op_active, &pdev->flags);
676 	smp_mb__after_atomic();
677 
678 	schedule_pcifront_aer_op(pdev);
679 
680 }
681 
682 static irqreturn_t pcifront_handler_aer(int irq, void *dev)
683 {
684 	struct pcifront_device *pdev = dev;
685 	schedule_pcifront_aer_op(pdev);
686 	return IRQ_HANDLED;
687 }
688 static int pcifront_connect_and_init_dma(struct pcifront_device *pdev)
689 {
690 	int err = 0;
691 
692 	spin_lock(&pcifront_dev_lock);
693 
694 	if (!pcifront_dev) {
695 		dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
696 		pcifront_dev = pdev;
697 	} else
698 		err = -EEXIST;
699 
700 	spin_unlock(&pcifront_dev_lock);
701 
702 	if (!err && !swiotlb_nr_tbl()) {
703 		err = pci_xen_swiotlb_init_late();
704 		if (err)
705 			dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n");
706 	}
707 	return err;
708 }
709 
710 static void pcifront_disconnect(struct pcifront_device *pdev)
711 {
712 	spin_lock(&pcifront_dev_lock);
713 
714 	if (pdev == pcifront_dev) {
715 		dev_info(&pdev->xdev->dev,
716 			 "Disconnecting PCI Frontend Buses\n");
717 		pcifront_dev = NULL;
718 	}
719 
720 	spin_unlock(&pcifront_dev_lock);
721 }
722 static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
723 {
724 	struct pcifront_device *pdev;
725 
726 	pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
727 	if (pdev == NULL)
728 		goto out;
729 
730 	pdev->sh_info =
731 	    (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL);
732 	if (pdev->sh_info == NULL) {
733 		kfree(pdev);
734 		pdev = NULL;
735 		goto out;
736 	}
737 	pdev->sh_info->flags = 0;
738 
739 	/*Flag for registering PV AER handler*/
740 	set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
741 
742 	dev_set_drvdata(&xdev->dev, pdev);
743 	pdev->xdev = xdev;
744 
745 	INIT_LIST_HEAD(&pdev->root_buses);
746 
747 	spin_lock_init(&pdev->sh_info_lock);
748 
749 	pdev->evtchn = INVALID_EVTCHN;
750 	pdev->gnt_ref = INVALID_GRANT_REF;
751 	pdev->irq = -1;
752 
753 	INIT_WORK(&pdev->op_work, pcifront_do_aer);
754 
755 	dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
756 		pdev, pdev->sh_info);
757 out:
758 	return pdev;
759 }
760 
761 static void free_pdev(struct pcifront_device *pdev)
762 {
763 	dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
764 
765 	pcifront_free_roots(pdev);
766 
767 	cancel_work_sync(&pdev->op_work);
768 
769 	if (pdev->irq >= 0)
770 		unbind_from_irqhandler(pdev->irq, pdev);
771 
772 	if (pdev->evtchn != INVALID_EVTCHN)
773 		xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
774 
775 	if (pdev->gnt_ref != INVALID_GRANT_REF)
776 		gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */,
777 					  (unsigned long)pdev->sh_info);
778 	else
779 		free_page((unsigned long)pdev->sh_info);
780 
781 	dev_set_drvdata(&pdev->xdev->dev, NULL);
782 
783 	kfree(pdev);
784 }
785 
786 static int pcifront_publish_info(struct pcifront_device *pdev)
787 {
788 	int err = 0;
789 	struct xenbus_transaction trans;
790 	grant_ref_t gref;
791 
792 	err = xenbus_grant_ring(pdev->xdev, pdev->sh_info, 1, &gref);
793 	if (err < 0)
794 		goto out;
795 
796 	pdev->gnt_ref = gref;
797 
798 	err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
799 	if (err)
800 		goto out;
801 
802 	err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
803 		0, "pcifront", pdev);
804 
805 	if (err < 0)
806 		return err;
807 
808 	pdev->irq = err;
809 
810 do_publish:
811 	err = xenbus_transaction_start(&trans);
812 	if (err) {
813 		xenbus_dev_fatal(pdev->xdev, err,
814 				 "Error writing configuration for backend "
815 				 "(start transaction)");
816 		goto out;
817 	}
818 
819 	err = xenbus_printf(trans, pdev->xdev->nodename,
820 			    "pci-op-ref", "%u", pdev->gnt_ref);
821 	if (!err)
822 		err = xenbus_printf(trans, pdev->xdev->nodename,
823 				    "event-channel", "%u", pdev->evtchn);
824 	if (!err)
825 		err = xenbus_printf(trans, pdev->xdev->nodename,
826 				    "magic", XEN_PCI_MAGIC);
827 
828 	if (err) {
829 		xenbus_transaction_end(trans, 1);
830 		xenbus_dev_fatal(pdev->xdev, err,
831 				 "Error writing configuration for backend");
832 		goto out;
833 	} else {
834 		err = xenbus_transaction_end(trans, 0);
835 		if (err == -EAGAIN)
836 			goto do_publish;
837 		else if (err) {
838 			xenbus_dev_fatal(pdev->xdev, err,
839 					 "Error completing transaction "
840 					 "for backend");
841 			goto out;
842 		}
843 	}
844 
845 	xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
846 
847 	dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
848 
849 out:
850 	return err;
851 }
852 
853 static int pcifront_try_connect(struct pcifront_device *pdev)
854 {
855 	int err = -EFAULT;
856 	int i, num_roots, len;
857 	char str[64];
858 	unsigned int domain, bus;
859 
860 
861 	/* Only connect once */
862 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
863 	    XenbusStateInitialised)
864 		goto out;
865 
866 	err = pcifront_connect_and_init_dma(pdev);
867 	if (err && err != -EEXIST) {
868 		xenbus_dev_fatal(pdev->xdev, err,
869 				 "Error setting up PCI Frontend");
870 		goto out;
871 	}
872 
873 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
874 			   "root_num", "%d", &num_roots);
875 	if (err == -ENOENT) {
876 		xenbus_dev_error(pdev->xdev, err,
877 				 "No PCI Roots found, trying 0000:00");
878 		err = pcifront_scan_root(pdev, 0, 0);
879 		if (err) {
880 			xenbus_dev_fatal(pdev->xdev, err,
881 					 "Error scanning PCI root 0000:00");
882 			goto out;
883 		}
884 		num_roots = 0;
885 	} else if (err != 1) {
886 		if (err == 0)
887 			err = -EINVAL;
888 		xenbus_dev_fatal(pdev->xdev, err,
889 				 "Error reading number of PCI roots");
890 		goto out;
891 	}
892 
893 	for (i = 0; i < num_roots; i++) {
894 		len = snprintf(str, sizeof(str), "root-%d", i);
895 		if (unlikely(len >= (sizeof(str) - 1))) {
896 			err = -ENOMEM;
897 			goto out;
898 		}
899 
900 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
901 				   "%x:%x", &domain, &bus);
902 		if (err != 2) {
903 			if (err >= 0)
904 				err = -EINVAL;
905 			xenbus_dev_fatal(pdev->xdev, err,
906 					 "Error reading PCI root %d", i);
907 			goto out;
908 		}
909 
910 		err = pcifront_scan_root(pdev, domain, bus);
911 		if (err) {
912 			xenbus_dev_fatal(pdev->xdev, err,
913 					 "Error scanning PCI root %04x:%02x",
914 					 domain, bus);
915 			goto out;
916 		}
917 	}
918 
919 	err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
920 
921 out:
922 	return err;
923 }
924 
925 static int pcifront_try_disconnect(struct pcifront_device *pdev)
926 {
927 	int err = 0;
928 	enum xenbus_state prev_state;
929 
930 
931 	prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
932 
933 	if (prev_state >= XenbusStateClosing)
934 		goto out;
935 
936 	if (prev_state == XenbusStateConnected) {
937 		pcifront_free_roots(pdev);
938 		pcifront_disconnect(pdev);
939 	}
940 
941 	err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
942 
943 out:
944 
945 	return err;
946 }
947 
948 static int pcifront_attach_devices(struct pcifront_device *pdev)
949 {
950 	int err = -EFAULT;
951 	int i, num_roots, len;
952 	unsigned int domain, bus;
953 	char str[64];
954 
955 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
956 	    XenbusStateReconfiguring)
957 		goto out;
958 
959 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
960 			   "root_num", "%d", &num_roots);
961 	if (err == -ENOENT) {
962 		xenbus_dev_error(pdev->xdev, err,
963 				 "No PCI Roots found, trying 0000:00");
964 		err = pcifront_rescan_root(pdev, 0, 0);
965 		if (err) {
966 			xenbus_dev_fatal(pdev->xdev, err,
967 					 "Error scanning PCI root 0000:00");
968 			goto out;
969 		}
970 		num_roots = 0;
971 	} else if (err != 1) {
972 		if (err == 0)
973 			err = -EINVAL;
974 		xenbus_dev_fatal(pdev->xdev, err,
975 				 "Error reading number of PCI roots");
976 		goto out;
977 	}
978 
979 	for (i = 0; i < num_roots; i++) {
980 		len = snprintf(str, sizeof(str), "root-%d", i);
981 		if (unlikely(len >= (sizeof(str) - 1))) {
982 			err = -ENOMEM;
983 			goto out;
984 		}
985 
986 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
987 				   "%x:%x", &domain, &bus);
988 		if (err != 2) {
989 			if (err >= 0)
990 				err = -EINVAL;
991 			xenbus_dev_fatal(pdev->xdev, err,
992 					 "Error reading PCI root %d", i);
993 			goto out;
994 		}
995 
996 		err = pcifront_rescan_root(pdev, domain, bus);
997 		if (err) {
998 			xenbus_dev_fatal(pdev->xdev, err,
999 					 "Error scanning PCI root %04x:%02x",
1000 					 domain, bus);
1001 			goto out;
1002 		}
1003 	}
1004 
1005 	xenbus_switch_state(pdev->xdev, XenbusStateConnected);
1006 
1007 out:
1008 	return err;
1009 }
1010 
1011 static int pcifront_detach_devices(struct pcifront_device *pdev)
1012 {
1013 	int err = 0;
1014 	int i, num_devs;
1015 	unsigned int domain, bus, slot, func;
1016 	struct pci_dev *pci_dev;
1017 	char str[64];
1018 
1019 	if (xenbus_read_driver_state(pdev->xdev->nodename) !=
1020 	    XenbusStateConnected)
1021 		goto out;
1022 
1023 	err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
1024 			   &num_devs);
1025 	if (err != 1) {
1026 		if (err >= 0)
1027 			err = -EINVAL;
1028 		xenbus_dev_fatal(pdev->xdev, err,
1029 				 "Error reading number of PCI devices");
1030 		goto out;
1031 	}
1032 
1033 	/* Find devices being detached and remove them. */
1034 	for (i = 0; i < num_devs; i++) {
1035 		int l, state;
1036 		l = snprintf(str, sizeof(str), "state-%d", i);
1037 		if (unlikely(l >= (sizeof(str) - 1))) {
1038 			err = -ENOMEM;
1039 			goto out;
1040 		}
1041 		state = xenbus_read_unsigned(pdev->xdev->otherend, str,
1042 					     XenbusStateUnknown);
1043 
1044 		if (state != XenbusStateClosing)
1045 			continue;
1046 
1047 		/* Remove device. */
1048 		l = snprintf(str, sizeof(str), "vdev-%d", i);
1049 		if (unlikely(l >= (sizeof(str) - 1))) {
1050 			err = -ENOMEM;
1051 			goto out;
1052 		}
1053 		err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
1054 				   "%x:%x:%x.%x", &domain, &bus, &slot, &func);
1055 		if (err != 4) {
1056 			if (err >= 0)
1057 				err = -EINVAL;
1058 			xenbus_dev_fatal(pdev->xdev, err,
1059 					 "Error reading PCI device %d", i);
1060 			goto out;
1061 		}
1062 
1063 		pci_dev = pci_get_domain_bus_and_slot(domain, bus,
1064 				PCI_DEVFN(slot, func));
1065 		if (!pci_dev) {
1066 			dev_dbg(&pdev->xdev->dev,
1067 				"Cannot get PCI device %04x:%02x:%02x.%d\n",
1068 				domain, bus, slot, func);
1069 			continue;
1070 		}
1071 		pci_lock_rescan_remove();
1072 		pci_stop_and_remove_bus_device(pci_dev);
1073 		pci_dev_put(pci_dev);
1074 		pci_unlock_rescan_remove();
1075 
1076 		dev_dbg(&pdev->xdev->dev,
1077 			"PCI device %04x:%02x:%02x.%d removed.\n",
1078 			domain, bus, slot, func);
1079 	}
1080 
1081 	err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
1082 
1083 out:
1084 	return err;
1085 }
1086 
1087 static void __ref pcifront_backend_changed(struct xenbus_device *xdev,
1088 						  enum xenbus_state be_state)
1089 {
1090 	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1091 
1092 	switch (be_state) {
1093 	case XenbusStateUnknown:
1094 	case XenbusStateInitialising:
1095 	case XenbusStateInitWait:
1096 	case XenbusStateInitialised:
1097 		break;
1098 
1099 	case XenbusStateConnected:
1100 		pcifront_try_connect(pdev);
1101 		break;
1102 
1103 	case XenbusStateClosed:
1104 		if (xdev->state == XenbusStateClosed)
1105 			break;
1106 		/* fall through - Missed the backend's CLOSING state. */
1107 	case XenbusStateClosing:
1108 		dev_warn(&xdev->dev, "backend going away!\n");
1109 		pcifront_try_disconnect(pdev);
1110 		break;
1111 
1112 	case XenbusStateReconfiguring:
1113 		pcifront_detach_devices(pdev);
1114 		break;
1115 
1116 	case XenbusStateReconfigured:
1117 		pcifront_attach_devices(pdev);
1118 		break;
1119 	}
1120 }
1121 
1122 static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1123 				 const struct xenbus_device_id *id)
1124 {
1125 	int err = 0;
1126 	struct pcifront_device *pdev = alloc_pdev(xdev);
1127 
1128 	if (pdev == NULL) {
1129 		err = -ENOMEM;
1130 		xenbus_dev_fatal(xdev, err,
1131 				 "Error allocating pcifront_device struct");
1132 		goto out;
1133 	}
1134 
1135 	err = pcifront_publish_info(pdev);
1136 	if (err)
1137 		free_pdev(pdev);
1138 
1139 out:
1140 	return err;
1141 }
1142 
1143 static int pcifront_xenbus_remove(struct xenbus_device *xdev)
1144 {
1145 	struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1146 	if (pdev)
1147 		free_pdev(pdev);
1148 
1149 	return 0;
1150 }
1151 
1152 static const struct xenbus_device_id xenpci_ids[] = {
1153 	{"pci"},
1154 	{""},
1155 };
1156 
1157 static struct xenbus_driver xenpci_driver = {
1158 	.name			= "pcifront",
1159 	.ids			= xenpci_ids,
1160 	.probe			= pcifront_xenbus_probe,
1161 	.remove			= pcifront_xenbus_remove,
1162 	.otherend_changed	= pcifront_backend_changed,
1163 };
1164 
1165 static int __init pcifront_init(void)
1166 {
1167 	if (!xen_pv_domain() || xen_initial_domain())
1168 		return -ENODEV;
1169 
1170 	if (!xen_has_pv_devices())
1171 		return -ENODEV;
1172 
1173 	pci_frontend_registrar(1 /* enable */);
1174 
1175 	return xenbus_register_frontend(&xenpci_driver);
1176 }
1177 
1178 static void __exit pcifront_cleanup(void)
1179 {
1180 	xenbus_unregister_driver(&xenpci_driver);
1181 	pci_frontend_registrar(0 /* disable */);
1182 }
1183 module_init(pcifront_init);
1184 module_exit(pcifront_cleanup);
1185 
1186 MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1187 MODULE_LICENSE("GPL");
1188 MODULE_ALIAS("xen:pci");
1189