1 /*
2  * PCI Stub Driver - Grabs devices in backend to be exported later
3  *
4  * Ryan Wilson <hap9@epoch.ncsc.mil>
5  * Chris Bookholt <hap10@epoch.ncsc.mil>
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/rwsem.h>
13 #include <linux/list.h>
14 #include <linux/spinlock.h>
15 #include <linux/kref.h>
16 #include <linux/pci.h>
17 #include <linux/wait.h>
18 #include <linux/sched.h>
19 #include <linux/atomic.h>
20 #include <xen/events.h>
21 #include <asm/xen/pci.h>
22 #include <asm/xen/hypervisor.h>
23 #include <xen/interface/physdev.h>
24 #include "pciback.h"
25 #include "conf_space.h"
26 #include "conf_space_quirks.h"
27 
28 #define PCISTUB_DRIVER_NAME "pciback"
29 
30 static char *pci_devs_to_hide;
31 wait_queue_head_t xen_pcibk_aer_wait_queue;
32 /*Add sem for sync AER handling and xen_pcibk remove/reconfigue ops,
33 * We want to avoid in middle of AER ops, xen_pcibk devices is being removed
34 */
35 static DECLARE_RWSEM(pcistub_sem);
36 module_param_named(hide, pci_devs_to_hide, charp, 0444);
37 
38 struct pcistub_device_id {
39 	struct list_head slot_list;
40 	int domain;
41 	unsigned char bus;
42 	unsigned int devfn;
43 };
44 static LIST_HEAD(pcistub_device_ids);
45 static DEFINE_SPINLOCK(device_ids_lock);
46 
47 struct pcistub_device {
48 	struct kref kref;
49 	struct list_head dev_list;
50 	spinlock_t lock;
51 
52 	struct pci_dev *dev;
53 	struct xen_pcibk_device *pdev;/* non-NULL if struct pci_dev is in use */
54 };
55 
56 /* Access to pcistub_devices & seized_devices lists and the initialize_devices
57  * flag must be locked with pcistub_devices_lock
58  */
59 static DEFINE_SPINLOCK(pcistub_devices_lock);
60 static LIST_HEAD(pcistub_devices);
61 
62 /* wait for device_initcall before initializing our devices
63  * (see pcistub_init_devices_late)
64  */
65 static int initialize_devices;
66 static LIST_HEAD(seized_devices);
67 
68 static struct pcistub_device *pcistub_device_alloc(struct pci_dev *dev)
69 {
70 	struct pcistub_device *psdev;
71 
72 	dev_dbg(&dev->dev, "pcistub_device_alloc\n");
73 
74 	psdev = kzalloc(sizeof(*psdev), GFP_ATOMIC);
75 	if (!psdev)
76 		return NULL;
77 
78 	psdev->dev = pci_dev_get(dev);
79 	if (!psdev->dev) {
80 		kfree(psdev);
81 		return NULL;
82 	}
83 
84 	kref_init(&psdev->kref);
85 	spin_lock_init(&psdev->lock);
86 
87 	return psdev;
88 }
89 
90 /* Don't call this directly as it's called by pcistub_device_put */
91 static void pcistub_device_release(struct kref *kref)
92 {
93 	struct pcistub_device *psdev;
94 	struct pci_dev *dev;
95 	struct xen_pcibk_dev_data *dev_data;
96 
97 	psdev = container_of(kref, struct pcistub_device, kref);
98 	dev = psdev->dev;
99 	dev_data = pci_get_drvdata(dev);
100 
101 	dev_dbg(&dev->dev, "pcistub_device_release\n");
102 
103 	xen_unregister_device_domain_owner(dev);
104 
105 	/* Call the reset function which does not take lock as this
106 	 * is called from "unbind" which takes a device_lock mutex.
107 	 */
108 	__pci_reset_function_locked(dev);
109 	if (pci_load_and_free_saved_state(dev, &dev_data->pci_saved_state))
110 		dev_info(&dev->dev, "Could not reload PCI state\n");
111 	else
112 		pci_restore_state(dev);
113 
114 	if (dev->msix_cap) {
115 		struct physdev_pci_device ppdev = {
116 			.seg = pci_domain_nr(dev->bus),
117 			.bus = dev->bus->number,
118 			.devfn = dev->devfn
119 		};
120 		int err = HYPERVISOR_physdev_op(PHYSDEVOP_release_msix,
121 						&ppdev);
122 
123 		if (err && err != -ENOSYS)
124 			dev_warn(&dev->dev, "MSI-X release failed (%d)\n",
125 				 err);
126 	}
127 
128 	/* Disable the device */
129 	xen_pcibk_reset_device(dev);
130 
131 	kfree(dev_data);
132 	pci_set_drvdata(dev, NULL);
133 
134 	/* Clean-up the device */
135 	xen_pcibk_config_free_dyn_fields(dev);
136 	xen_pcibk_config_free_dev(dev);
137 
138 	pci_clear_dev_assigned(dev);
139 	pci_dev_put(dev);
140 
141 	kfree(psdev);
142 }
143 
144 static inline void pcistub_device_get(struct pcistub_device *psdev)
145 {
146 	kref_get(&psdev->kref);
147 }
148 
149 static inline void pcistub_device_put(struct pcistub_device *psdev)
150 {
151 	kref_put(&psdev->kref, pcistub_device_release);
152 }
153 
154 static struct pcistub_device *pcistub_device_find_locked(int domain, int bus,
155 							 int slot, int func)
156 {
157 	struct pcistub_device *psdev;
158 
159 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
160 		if (psdev->dev != NULL
161 		    && domain == pci_domain_nr(psdev->dev->bus)
162 		    && bus == psdev->dev->bus->number
163 		    && slot == PCI_SLOT(psdev->dev->devfn)
164 		    && func == PCI_FUNC(psdev->dev->devfn)) {
165 			return psdev;
166 		}
167 	}
168 
169 	return NULL;
170 }
171 
172 static struct pcistub_device *pcistub_device_find(int domain, int bus,
173 						  int slot, int func)
174 {
175 	struct pcistub_device *psdev;
176 	unsigned long flags;
177 
178 	spin_lock_irqsave(&pcistub_devices_lock, flags);
179 
180 	psdev = pcistub_device_find_locked(domain, bus, slot, func);
181 	if (psdev)
182 		pcistub_device_get(psdev);
183 
184 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
185 	return psdev;
186 }
187 
188 static struct pci_dev *pcistub_device_get_pci_dev(struct xen_pcibk_device *pdev,
189 						  struct pcistub_device *psdev)
190 {
191 	struct pci_dev *pci_dev = NULL;
192 	unsigned long flags;
193 
194 	pcistub_device_get(psdev);
195 
196 	spin_lock_irqsave(&psdev->lock, flags);
197 	if (!psdev->pdev) {
198 		psdev->pdev = pdev;
199 		pci_dev = psdev->dev;
200 	}
201 	spin_unlock_irqrestore(&psdev->lock, flags);
202 
203 	if (!pci_dev)
204 		pcistub_device_put(psdev);
205 
206 	return pci_dev;
207 }
208 
209 struct pci_dev *pcistub_get_pci_dev_by_slot(struct xen_pcibk_device *pdev,
210 					    int domain, int bus,
211 					    int slot, int func)
212 {
213 	struct pcistub_device *psdev;
214 	struct pci_dev *found_dev = NULL;
215 	unsigned long flags;
216 
217 	spin_lock_irqsave(&pcistub_devices_lock, flags);
218 
219 	psdev = pcistub_device_find_locked(domain, bus, slot, func);
220 	if (psdev)
221 		found_dev = pcistub_device_get_pci_dev(pdev, psdev);
222 
223 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
224 	return found_dev;
225 }
226 
227 struct pci_dev *pcistub_get_pci_dev(struct xen_pcibk_device *pdev,
228 				    struct pci_dev *dev)
229 {
230 	struct pcistub_device *psdev;
231 	struct pci_dev *found_dev = NULL;
232 	unsigned long flags;
233 
234 	spin_lock_irqsave(&pcistub_devices_lock, flags);
235 
236 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
237 		if (psdev->dev == dev) {
238 			found_dev = pcistub_device_get_pci_dev(pdev, psdev);
239 			break;
240 		}
241 	}
242 
243 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
244 	return found_dev;
245 }
246 
247 /*
248  * Called when:
249  *  - XenBus state has been reconfigure (pci unplug). See xen_pcibk_remove_device
250  *  - XenBus state has been disconnected (guest shutdown). See xen_pcibk_xenbus_remove
251  *  - 'echo BDF > unbind' on pciback module with no guest attached. See pcistub_remove
252  *  - 'echo BDF > unbind' with a guest still using it. See pcistub_remove
253  *
254  *  As such we have to be careful.
255  *
256  *  To make this easier, the caller has to hold the device lock.
257  */
258 void pcistub_put_pci_dev(struct pci_dev *dev)
259 {
260 	struct pcistub_device *psdev, *found_psdev = NULL;
261 	unsigned long flags;
262 	struct xen_pcibk_dev_data *dev_data;
263 	int ret;
264 
265 	spin_lock_irqsave(&pcistub_devices_lock, flags);
266 
267 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
268 		if (psdev->dev == dev) {
269 			found_psdev = psdev;
270 			break;
271 		}
272 	}
273 
274 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
275 	if (WARN_ON(!found_psdev))
276 		return;
277 
278 	/*hold this lock for avoiding breaking link between
279 	* pcistub and xen_pcibk when AER is in processing
280 	*/
281 	down_write(&pcistub_sem);
282 	/* Cleanup our device
283 	 * (so it's ready for the next domain)
284 	 */
285 	device_lock_assert(&dev->dev);
286 	__pci_reset_function_locked(dev);
287 
288 	dev_data = pci_get_drvdata(dev);
289 	ret = pci_load_saved_state(dev, dev_data->pci_saved_state);
290 	if (!ret) {
291 		/*
292 		 * The usual sequence is pci_save_state & pci_restore_state
293 		 * but the guest might have messed the configuration space up.
294 		 * Use the initial version (when device was bound to us).
295 		 */
296 		pci_restore_state(dev);
297 	} else
298 		dev_info(&dev->dev, "Could not reload PCI state\n");
299 	/* This disables the device. */
300 	xen_pcibk_reset_device(dev);
301 
302 	/* And cleanup up our emulated fields. */
303 	xen_pcibk_config_reset_dev(dev);
304 	xen_pcibk_config_free_dyn_fields(dev);
305 
306 	xen_unregister_device_domain_owner(dev);
307 
308 	spin_lock_irqsave(&found_psdev->lock, flags);
309 	found_psdev->pdev = NULL;
310 	spin_unlock_irqrestore(&found_psdev->lock, flags);
311 
312 	pcistub_device_put(found_psdev);
313 	up_write(&pcistub_sem);
314 }
315 
316 static int pcistub_match_one(struct pci_dev *dev,
317 			     struct pcistub_device_id *pdev_id)
318 {
319 	/* Match the specified device by domain, bus, slot, func and also if
320 	 * any of the device's parent bridges match.
321 	 */
322 	for (; dev != NULL; dev = dev->bus->self) {
323 		if (pci_domain_nr(dev->bus) == pdev_id->domain
324 		    && dev->bus->number == pdev_id->bus
325 		    && dev->devfn == pdev_id->devfn)
326 			return 1;
327 
328 		/* Sometimes topmost bridge links to itself. */
329 		if (dev == dev->bus->self)
330 			break;
331 	}
332 
333 	return 0;
334 }
335 
336 static int pcistub_match(struct pci_dev *dev)
337 {
338 	struct pcistub_device_id *pdev_id;
339 	unsigned long flags;
340 	int found = 0;
341 
342 	spin_lock_irqsave(&device_ids_lock, flags);
343 	list_for_each_entry(pdev_id, &pcistub_device_ids, slot_list) {
344 		if (pcistub_match_one(dev, pdev_id)) {
345 			found = 1;
346 			break;
347 		}
348 	}
349 	spin_unlock_irqrestore(&device_ids_lock, flags);
350 
351 	return found;
352 }
353 
354 static int pcistub_init_device(struct pci_dev *dev)
355 {
356 	struct xen_pcibk_dev_data *dev_data;
357 	int err = 0;
358 
359 	dev_dbg(&dev->dev, "initializing...\n");
360 
361 	/* The PCI backend is not intended to be a module (or to work with
362 	 * removable PCI devices (yet). If it were, xen_pcibk_config_free()
363 	 * would need to be called somewhere to free the memory allocated
364 	 * here and then to call kfree(pci_get_drvdata(psdev->dev)).
365 	 */
366 	dev_data = kzalloc(sizeof(*dev_data) +  strlen(DRV_NAME "[]")
367 				+ strlen(pci_name(dev)) + 1, GFP_ATOMIC);
368 	if (!dev_data) {
369 		err = -ENOMEM;
370 		goto out;
371 	}
372 	pci_set_drvdata(dev, dev_data);
373 
374 	/*
375 	 * Setup name for fake IRQ handler. It will only be enabled
376 	 * once the device is turned on by the guest.
377 	 */
378 	sprintf(dev_data->irq_name, DRV_NAME "[%s]", pci_name(dev));
379 
380 	dev_dbg(&dev->dev, "initializing config\n");
381 
382 	init_waitqueue_head(&xen_pcibk_aer_wait_queue);
383 	err = xen_pcibk_config_init_dev(dev);
384 	if (err)
385 		goto out;
386 
387 	/* HACK: Force device (& ACPI) to determine what IRQ it's on - we
388 	 * must do this here because pcibios_enable_device may specify
389 	 * the pci device's true irq (and possibly its other resources)
390 	 * if they differ from what's in the configuration space.
391 	 * This makes the assumption that the device's resources won't
392 	 * change after this point (otherwise this code may break!)
393 	 */
394 	dev_dbg(&dev->dev, "enabling device\n");
395 	err = pci_enable_device(dev);
396 	if (err)
397 		goto config_release;
398 
399 	if (dev->msix_cap) {
400 		struct physdev_pci_device ppdev = {
401 			.seg = pci_domain_nr(dev->bus),
402 			.bus = dev->bus->number,
403 			.devfn = dev->devfn
404 		};
405 
406 		err = HYPERVISOR_physdev_op(PHYSDEVOP_prepare_msix, &ppdev);
407 		if (err && err != -ENOSYS)
408 			dev_err(&dev->dev, "MSI-X preparation failed (%d)\n",
409 				err);
410 	}
411 
412 	/* We need the device active to save the state. */
413 	dev_dbg(&dev->dev, "save state of device\n");
414 	pci_save_state(dev);
415 	dev_data->pci_saved_state = pci_store_saved_state(dev);
416 	if (!dev_data->pci_saved_state)
417 		dev_err(&dev->dev, "Could not store PCI conf saved state!\n");
418 	else {
419 		dev_dbg(&dev->dev, "resetting (FLR, D3, etc) the device\n");
420 		__pci_reset_function_locked(dev);
421 		pci_restore_state(dev);
422 	}
423 	/* Now disable the device (this also ensures some private device
424 	 * data is setup before we export)
425 	 */
426 	dev_dbg(&dev->dev, "reset device\n");
427 	xen_pcibk_reset_device(dev);
428 
429 	pci_set_dev_assigned(dev);
430 	return 0;
431 
432 config_release:
433 	xen_pcibk_config_free_dev(dev);
434 
435 out:
436 	pci_set_drvdata(dev, NULL);
437 	kfree(dev_data);
438 	return err;
439 }
440 
441 /*
442  * Because some initialization still happens on
443  * devices during fs_initcall, we need to defer
444  * full initialization of our devices until
445  * device_initcall.
446  */
447 static int __init pcistub_init_devices_late(void)
448 {
449 	struct pcistub_device *psdev;
450 	unsigned long flags;
451 	int err = 0;
452 
453 	spin_lock_irqsave(&pcistub_devices_lock, flags);
454 
455 	while (!list_empty(&seized_devices)) {
456 		psdev = container_of(seized_devices.next,
457 				     struct pcistub_device, dev_list);
458 		list_del(&psdev->dev_list);
459 
460 		spin_unlock_irqrestore(&pcistub_devices_lock, flags);
461 
462 		err = pcistub_init_device(psdev->dev);
463 		if (err) {
464 			dev_err(&psdev->dev->dev,
465 				"error %d initializing device\n", err);
466 			kfree(psdev);
467 			psdev = NULL;
468 		}
469 
470 		spin_lock_irqsave(&pcistub_devices_lock, flags);
471 
472 		if (psdev)
473 			list_add_tail(&psdev->dev_list, &pcistub_devices);
474 	}
475 
476 	initialize_devices = 1;
477 
478 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
479 
480 	return 0;
481 }
482 
483 static void pcistub_device_id_add_list(struct pcistub_device_id *new,
484 				       int domain, int bus, unsigned int devfn)
485 {
486 	struct pcistub_device_id *pci_dev_id;
487 	unsigned long flags;
488 	int found = 0;
489 
490 	spin_lock_irqsave(&device_ids_lock, flags);
491 
492 	list_for_each_entry(pci_dev_id, &pcistub_device_ids, slot_list) {
493 		if (pci_dev_id->domain == domain && pci_dev_id->bus == bus &&
494 		    pci_dev_id->devfn == devfn) {
495 			found = 1;
496 			break;
497 		}
498 	}
499 
500 	if (!found) {
501 		new->domain = domain;
502 		new->bus = bus;
503 		new->devfn = devfn;
504 		list_add_tail(&new->slot_list, &pcistub_device_ids);
505 	}
506 
507 	spin_unlock_irqrestore(&device_ids_lock, flags);
508 
509 	if (found)
510 		kfree(new);
511 }
512 
513 static int pcistub_seize(struct pci_dev *dev,
514 			 struct pcistub_device_id *pci_dev_id)
515 {
516 	struct pcistub_device *psdev;
517 	unsigned long flags;
518 	int err = 0;
519 
520 	psdev = pcistub_device_alloc(dev);
521 	if (!psdev) {
522 		kfree(pci_dev_id);
523 		return -ENOMEM;
524 	}
525 
526 	spin_lock_irqsave(&pcistub_devices_lock, flags);
527 
528 	if (initialize_devices) {
529 		spin_unlock_irqrestore(&pcistub_devices_lock, flags);
530 
531 		/* don't want irqs disabled when calling pcistub_init_device */
532 		err = pcistub_init_device(psdev->dev);
533 
534 		spin_lock_irqsave(&pcistub_devices_lock, flags);
535 
536 		if (!err)
537 			list_add(&psdev->dev_list, &pcistub_devices);
538 	} else {
539 		dev_dbg(&dev->dev, "deferring initialization\n");
540 		list_add(&psdev->dev_list, &seized_devices);
541 	}
542 
543 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
544 
545 	if (err) {
546 		kfree(pci_dev_id);
547 		pcistub_device_put(psdev);
548 	} else if (pci_dev_id)
549 		pcistub_device_id_add_list(pci_dev_id, pci_domain_nr(dev->bus),
550 					   dev->bus->number, dev->devfn);
551 
552 	return err;
553 }
554 
555 /* Called when 'bind'. This means we must _NOT_ call pci_reset_function or
556  * other functions that take the sysfs lock. */
557 static int pcistub_probe(struct pci_dev *dev, const struct pci_device_id *id)
558 {
559 	int err = 0, match;
560 	struct pcistub_device_id *pci_dev_id = NULL;
561 
562 	dev_dbg(&dev->dev, "probing...\n");
563 
564 	match = pcistub_match(dev);
565 
566 	if ((dev->driver_override &&
567 	     !strcmp(dev->driver_override, PCISTUB_DRIVER_NAME)) ||
568 	    match) {
569 
570 		if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL
571 		    && dev->hdr_type != PCI_HEADER_TYPE_BRIDGE) {
572 			dev_err(&dev->dev, "can't export pci devices that "
573 				"don't have a normal (0) or bridge (1) "
574 				"header type!\n");
575 			err = -ENODEV;
576 			goto out;
577 		}
578 
579 		if (!match) {
580 			pci_dev_id = kmalloc(sizeof(*pci_dev_id), GFP_ATOMIC);
581 			if (!pci_dev_id) {
582 				err = -ENOMEM;
583 				goto out;
584 			}
585 		}
586 
587 		dev_info(&dev->dev, "seizing device\n");
588 		err = pcistub_seize(dev, pci_dev_id);
589 	} else
590 		/* Didn't find the device */
591 		err = -ENODEV;
592 
593 out:
594 	return err;
595 }
596 
597 /* Called when 'unbind'. This means we must _NOT_ call pci_reset_function or
598  * other functions that take the sysfs lock. */
599 static void pcistub_remove(struct pci_dev *dev)
600 {
601 	struct pcistub_device *psdev, *found_psdev = NULL;
602 	unsigned long flags;
603 
604 	dev_dbg(&dev->dev, "removing\n");
605 
606 	spin_lock_irqsave(&pcistub_devices_lock, flags);
607 
608 	xen_pcibk_config_quirk_release(dev);
609 
610 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
611 		if (psdev->dev == dev) {
612 			found_psdev = psdev;
613 			break;
614 		}
615 	}
616 
617 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
618 
619 	if (found_psdev) {
620 		dev_dbg(&dev->dev, "found device to remove %s\n",
621 			found_psdev->pdev ? "- in-use" : "");
622 
623 		if (found_psdev->pdev) {
624 			int domid = xen_find_device_domain_owner(dev);
625 
626 			pr_warn("****** removing device %s while still in-use by domain %d! ******\n",
627 			       pci_name(found_psdev->dev), domid);
628 			pr_warn("****** driver domain may still access this device's i/o resources!\n");
629 			pr_warn("****** shutdown driver domain before binding device\n");
630 			pr_warn("****** to other drivers or domains\n");
631 
632 			/* N.B. This ends up calling pcistub_put_pci_dev which ends up
633 			 * doing the FLR. */
634 			xen_pcibk_release_pci_dev(found_psdev->pdev,
635 						found_psdev->dev,
636 						false /* caller holds the lock. */);
637 		}
638 
639 		spin_lock_irqsave(&pcistub_devices_lock, flags);
640 		list_del(&found_psdev->dev_list);
641 		spin_unlock_irqrestore(&pcistub_devices_lock, flags);
642 
643 		/* the final put for releasing from the list */
644 		pcistub_device_put(found_psdev);
645 	}
646 }
647 
648 static const struct pci_device_id pcistub_ids[] = {
649 	{
650 	 .vendor = PCI_ANY_ID,
651 	 .device = PCI_ANY_ID,
652 	 .subvendor = PCI_ANY_ID,
653 	 .subdevice = PCI_ANY_ID,
654 	 },
655 	{0,},
656 };
657 
658 #define PCI_NODENAME_MAX 40
659 static void kill_domain_by_device(struct pcistub_device *psdev)
660 {
661 	struct xenbus_transaction xbt;
662 	int err;
663 	char nodename[PCI_NODENAME_MAX];
664 
665 	BUG_ON(!psdev);
666 	snprintf(nodename, PCI_NODENAME_MAX, "/local/domain/0/backend/pci/%d/0",
667 		psdev->pdev->xdev->otherend_id);
668 
669 again:
670 	err = xenbus_transaction_start(&xbt);
671 	if (err) {
672 		dev_err(&psdev->dev->dev,
673 			"error %d when start xenbus transaction\n", err);
674 		return;
675 	}
676 	/*PV AER handlers will set this flag*/
677 	xenbus_printf(xbt, nodename, "aerState" , "aerfail");
678 	err = xenbus_transaction_end(xbt, 0);
679 	if (err) {
680 		if (err == -EAGAIN)
681 			goto again;
682 		dev_err(&psdev->dev->dev,
683 			"error %d when end xenbus transaction\n", err);
684 		return;
685 	}
686 }
687 
688 /* For each aer recovery step error_detected, mmio_enabled, etc, front_end and
689  * backend need to have cooperation. In xen_pcibk, those steps will do similar
690  * jobs: send service request and waiting for front_end response.
691 */
692 static pci_ers_result_t common_process(struct pcistub_device *psdev,
693 				       pci_channel_state_t state, int aer_cmd,
694 				       pci_ers_result_t result)
695 {
696 	pci_ers_result_t res = result;
697 	struct xen_pcie_aer_op *aer_op;
698 	struct xen_pcibk_device *pdev = psdev->pdev;
699 	struct xen_pci_sharedinfo *sh_info = pdev->sh_info;
700 	int ret;
701 
702 	/*with PV AER drivers*/
703 	aer_op = &(sh_info->aer_op);
704 	aer_op->cmd = aer_cmd ;
705 	/*useful for error_detected callback*/
706 	aer_op->err = state;
707 	/*pcifront_end BDF*/
708 	ret = xen_pcibk_get_pcifront_dev(psdev->dev, psdev->pdev,
709 		&aer_op->domain, &aer_op->bus, &aer_op->devfn);
710 	if (!ret) {
711 		dev_err(&psdev->dev->dev,
712 			DRV_NAME ": failed to get pcifront device\n");
713 		return PCI_ERS_RESULT_NONE;
714 	}
715 	wmb();
716 
717 	dev_dbg(&psdev->dev->dev,
718 			DRV_NAME ": aer_op %x dom %x bus %x devfn %x\n",
719 			aer_cmd, aer_op->domain, aer_op->bus, aer_op->devfn);
720 	/*local flag to mark there's aer request, xen_pcibk callback will use
721 	* this flag to judge whether we need to check pci-front give aer
722 	* service ack signal
723 	*/
724 	set_bit(_PCIB_op_pending, (unsigned long *)&pdev->flags);
725 
726 	/*It is possible that a pcifront conf_read_write ops request invokes
727 	* the callback which cause the spurious execution of wake_up.
728 	* Yet it is harmless and better than a spinlock here
729 	*/
730 	set_bit(_XEN_PCIB_active,
731 		(unsigned long *)&sh_info->flags);
732 	wmb();
733 	notify_remote_via_irq(pdev->evtchn_irq);
734 
735 	ret = wait_event_timeout(xen_pcibk_aer_wait_queue,
736 				 !(test_bit(_XEN_PCIB_active, (unsigned long *)
737 				 &sh_info->flags)), 300*HZ);
738 
739 	if (!ret) {
740 		if (test_bit(_XEN_PCIB_active,
741 			(unsigned long *)&sh_info->flags)) {
742 			dev_err(&psdev->dev->dev,
743 				"pcifront aer process not responding!\n");
744 			clear_bit(_XEN_PCIB_active,
745 			  (unsigned long *)&sh_info->flags);
746 			aer_op->err = PCI_ERS_RESULT_NONE;
747 			return res;
748 		}
749 	}
750 	clear_bit(_PCIB_op_pending, (unsigned long *)&pdev->flags);
751 
752 	if (test_bit(_XEN_PCIF_active,
753 		(unsigned long *)&sh_info->flags)) {
754 		dev_dbg(&psdev->dev->dev,
755 			"schedule pci_conf service in " DRV_NAME "\n");
756 		xen_pcibk_test_and_schedule_op(psdev->pdev);
757 	}
758 
759 	res = (pci_ers_result_t)aer_op->err;
760 	return res;
761 }
762 
763 /*
764 * xen_pcibk_slot_reset: it will send the slot_reset request to  pcifront in case
765 * of the device driver could provide this service, and then wait for pcifront
766 * ack.
767 * @dev: pointer to PCI devices
768 * return value is used by aer_core do_recovery policy
769 */
770 static pci_ers_result_t xen_pcibk_slot_reset(struct pci_dev *dev)
771 {
772 	struct pcistub_device *psdev;
773 	pci_ers_result_t result;
774 
775 	result = PCI_ERS_RESULT_RECOVERED;
776 	dev_dbg(&dev->dev, "xen_pcibk_slot_reset(bus:%x,devfn:%x)\n",
777 		dev->bus->number, dev->devfn);
778 
779 	down_write(&pcistub_sem);
780 	psdev = pcistub_device_find(pci_domain_nr(dev->bus),
781 				dev->bus->number,
782 				PCI_SLOT(dev->devfn),
783 				PCI_FUNC(dev->devfn));
784 
785 	if (!psdev || !psdev->pdev) {
786 		dev_err(&dev->dev,
787 			DRV_NAME " device is not found/assigned\n");
788 		goto end;
789 	}
790 
791 	if (!psdev->pdev->sh_info) {
792 		dev_err(&dev->dev, DRV_NAME " device is not connected or owned"
793 			" by HVM, kill it\n");
794 		kill_domain_by_device(psdev);
795 		goto end;
796 	}
797 
798 	if (!test_bit(_XEN_PCIB_AERHANDLER,
799 		(unsigned long *)&psdev->pdev->sh_info->flags)) {
800 		dev_err(&dev->dev,
801 			"guest with no AER driver should have been killed\n");
802 		goto end;
803 	}
804 	result = common_process(psdev, 1, XEN_PCI_OP_aer_slotreset, result);
805 
806 	if (result == PCI_ERS_RESULT_NONE ||
807 		result == PCI_ERS_RESULT_DISCONNECT) {
808 		dev_dbg(&dev->dev,
809 			"No AER slot_reset service or disconnected!\n");
810 		kill_domain_by_device(psdev);
811 	}
812 end:
813 	if (psdev)
814 		pcistub_device_put(psdev);
815 	up_write(&pcistub_sem);
816 	return result;
817 
818 }
819 
820 
821 /*xen_pcibk_mmio_enabled: it will send the mmio_enabled request to  pcifront
822 * in case of the device driver could provide this service, and then wait
823 * for pcifront ack
824 * @dev: pointer to PCI devices
825 * return value is used by aer_core do_recovery policy
826 */
827 
828 static pci_ers_result_t xen_pcibk_mmio_enabled(struct pci_dev *dev)
829 {
830 	struct pcistub_device *psdev;
831 	pci_ers_result_t result;
832 
833 	result = PCI_ERS_RESULT_RECOVERED;
834 	dev_dbg(&dev->dev, "xen_pcibk_mmio_enabled(bus:%x,devfn:%x)\n",
835 		dev->bus->number, dev->devfn);
836 
837 	down_write(&pcistub_sem);
838 	psdev = pcistub_device_find(pci_domain_nr(dev->bus),
839 				dev->bus->number,
840 				PCI_SLOT(dev->devfn),
841 				PCI_FUNC(dev->devfn));
842 
843 	if (!psdev || !psdev->pdev) {
844 		dev_err(&dev->dev,
845 			DRV_NAME " device is not found/assigned\n");
846 		goto end;
847 	}
848 
849 	if (!psdev->pdev->sh_info) {
850 		dev_err(&dev->dev, DRV_NAME " device is not connected or owned"
851 			" by HVM, kill it\n");
852 		kill_domain_by_device(psdev);
853 		goto end;
854 	}
855 
856 	if (!test_bit(_XEN_PCIB_AERHANDLER,
857 		(unsigned long *)&psdev->pdev->sh_info->flags)) {
858 		dev_err(&dev->dev,
859 			"guest with no AER driver should have been killed\n");
860 		goto end;
861 	}
862 	result = common_process(psdev, 1, XEN_PCI_OP_aer_mmio, result);
863 
864 	if (result == PCI_ERS_RESULT_NONE ||
865 		result == PCI_ERS_RESULT_DISCONNECT) {
866 		dev_dbg(&dev->dev,
867 			"No AER mmio_enabled service or disconnected!\n");
868 		kill_domain_by_device(psdev);
869 	}
870 end:
871 	if (psdev)
872 		pcistub_device_put(psdev);
873 	up_write(&pcistub_sem);
874 	return result;
875 }
876 
877 /*xen_pcibk_error_detected: it will send the error_detected request to  pcifront
878 * in case of the device driver could provide this service, and then wait
879 * for pcifront ack.
880 * @dev: pointer to PCI devices
881 * @error: the current PCI connection state
882 * return value is used by aer_core do_recovery policy
883 */
884 
885 static pci_ers_result_t xen_pcibk_error_detected(struct pci_dev *dev,
886 	pci_channel_state_t error)
887 {
888 	struct pcistub_device *psdev;
889 	pci_ers_result_t result;
890 
891 	result = PCI_ERS_RESULT_CAN_RECOVER;
892 	dev_dbg(&dev->dev, "xen_pcibk_error_detected(bus:%x,devfn:%x)\n",
893 		dev->bus->number, dev->devfn);
894 
895 	down_write(&pcistub_sem);
896 	psdev = pcistub_device_find(pci_domain_nr(dev->bus),
897 				dev->bus->number,
898 				PCI_SLOT(dev->devfn),
899 				PCI_FUNC(dev->devfn));
900 
901 	if (!psdev || !psdev->pdev) {
902 		dev_err(&dev->dev,
903 			DRV_NAME " device is not found/assigned\n");
904 		goto end;
905 	}
906 
907 	if (!psdev->pdev->sh_info) {
908 		dev_err(&dev->dev, DRV_NAME " device is not connected or owned"
909 			" by HVM, kill it\n");
910 		kill_domain_by_device(psdev);
911 		goto end;
912 	}
913 
914 	/*Guest owns the device yet no aer handler regiested, kill guest*/
915 	if (!test_bit(_XEN_PCIB_AERHANDLER,
916 		(unsigned long *)&psdev->pdev->sh_info->flags)) {
917 		dev_dbg(&dev->dev, "guest may have no aer driver, kill it\n");
918 		kill_domain_by_device(psdev);
919 		goto end;
920 	}
921 	result = common_process(psdev, error, XEN_PCI_OP_aer_detected, result);
922 
923 	if (result == PCI_ERS_RESULT_NONE ||
924 		result == PCI_ERS_RESULT_DISCONNECT) {
925 		dev_dbg(&dev->dev,
926 			"No AER error_detected service or disconnected!\n");
927 		kill_domain_by_device(psdev);
928 	}
929 end:
930 	if (psdev)
931 		pcistub_device_put(psdev);
932 	up_write(&pcistub_sem);
933 	return result;
934 }
935 
936 /*xen_pcibk_error_resume: it will send the error_resume request to  pcifront
937 * in case of the device driver could provide this service, and then wait
938 * for pcifront ack.
939 * @dev: pointer to PCI devices
940 */
941 
942 static void xen_pcibk_error_resume(struct pci_dev *dev)
943 {
944 	struct pcistub_device *psdev;
945 
946 	dev_dbg(&dev->dev, "xen_pcibk_error_resume(bus:%x,devfn:%x)\n",
947 		dev->bus->number, dev->devfn);
948 
949 	down_write(&pcistub_sem);
950 	psdev = pcistub_device_find(pci_domain_nr(dev->bus),
951 				dev->bus->number,
952 				PCI_SLOT(dev->devfn),
953 				PCI_FUNC(dev->devfn));
954 
955 	if (!psdev || !psdev->pdev) {
956 		dev_err(&dev->dev,
957 			DRV_NAME " device is not found/assigned\n");
958 		goto end;
959 	}
960 
961 	if (!psdev->pdev->sh_info) {
962 		dev_err(&dev->dev, DRV_NAME " device is not connected or owned"
963 			" by HVM, kill it\n");
964 		kill_domain_by_device(psdev);
965 		goto end;
966 	}
967 
968 	if (!test_bit(_XEN_PCIB_AERHANDLER,
969 		(unsigned long *)&psdev->pdev->sh_info->flags)) {
970 		dev_err(&dev->dev,
971 			"guest with no AER driver should have been killed\n");
972 		kill_domain_by_device(psdev);
973 		goto end;
974 	}
975 	common_process(psdev, 1, XEN_PCI_OP_aer_resume,
976 		       PCI_ERS_RESULT_RECOVERED);
977 end:
978 	if (psdev)
979 		pcistub_device_put(psdev);
980 	up_write(&pcistub_sem);
981 	return;
982 }
983 
984 /*add xen_pcibk AER handling*/
985 static const struct pci_error_handlers xen_pcibk_error_handler = {
986 	.error_detected = xen_pcibk_error_detected,
987 	.mmio_enabled = xen_pcibk_mmio_enabled,
988 	.slot_reset = xen_pcibk_slot_reset,
989 	.resume = xen_pcibk_error_resume,
990 };
991 
992 /*
993  * Note: There is no MODULE_DEVICE_TABLE entry here because this isn't
994  * for a normal device. I don't want it to be loaded automatically.
995  */
996 
997 static struct pci_driver xen_pcibk_pci_driver = {
998 	/* The name should be xen_pciback, but until the tools are updated
999 	 * we will keep it as pciback. */
1000 	.name = PCISTUB_DRIVER_NAME,
1001 	.id_table = pcistub_ids,
1002 	.probe = pcistub_probe,
1003 	.remove = pcistub_remove,
1004 	.err_handler = &xen_pcibk_error_handler,
1005 };
1006 
1007 static inline int str_to_slot(const char *buf, int *domain, int *bus,
1008 			      int *slot, int *func)
1009 {
1010 	int parsed = 0;
1011 
1012 	switch (sscanf(buf, " %x:%x:%x.%x %n", domain, bus, slot, func,
1013 		       &parsed)) {
1014 	case 3:
1015 		*func = -1;
1016 		sscanf(buf, " %x:%x:%x.* %n", domain, bus, slot, &parsed);
1017 		break;
1018 	case 2:
1019 		*slot = *func = -1;
1020 		sscanf(buf, " %x:%x:*.* %n", domain, bus, &parsed);
1021 		break;
1022 	}
1023 	if (parsed && !buf[parsed])
1024 		return 0;
1025 
1026 	/* try again without domain */
1027 	*domain = 0;
1028 	switch (sscanf(buf, " %x:%x.%x %n", bus, slot, func, &parsed)) {
1029 	case 2:
1030 		*func = -1;
1031 		sscanf(buf, " %x:%x.* %n", bus, slot, &parsed);
1032 		break;
1033 	case 1:
1034 		*slot = *func = -1;
1035 		sscanf(buf, " %x:*.* %n", bus, &parsed);
1036 		break;
1037 	}
1038 	if (parsed && !buf[parsed])
1039 		return 0;
1040 
1041 	return -EINVAL;
1042 }
1043 
1044 static inline int str_to_quirk(const char *buf, int *domain, int *bus, int
1045 			       *slot, int *func, int *reg, int *size, int *mask)
1046 {
1047 	int parsed = 0;
1048 
1049 	sscanf(buf, " %x:%x:%x.%x-%x:%x:%x %n", domain, bus, slot, func,
1050 	       reg, size, mask, &parsed);
1051 	if (parsed && !buf[parsed])
1052 		return 0;
1053 
1054 	/* try again without domain */
1055 	*domain = 0;
1056 	sscanf(buf, " %x:%x.%x-%x:%x:%x %n", bus, slot, func, reg, size,
1057 	       mask, &parsed);
1058 	if (parsed && !buf[parsed])
1059 		return 0;
1060 
1061 	return -EINVAL;
1062 }
1063 
1064 static int pcistub_device_id_add(int domain, int bus, int slot, int func)
1065 {
1066 	struct pcistub_device_id *pci_dev_id;
1067 	int rc = 0, devfn = PCI_DEVFN(slot, func);
1068 
1069 	if (slot < 0) {
1070 		for (slot = 0; !rc && slot < 32; ++slot)
1071 			rc = pcistub_device_id_add(domain, bus, slot, func);
1072 		return rc;
1073 	}
1074 
1075 	if (func < 0) {
1076 		for (func = 0; !rc && func < 8; ++func)
1077 			rc = pcistub_device_id_add(domain, bus, slot, func);
1078 		return rc;
1079 	}
1080 
1081 	if ((
1082 #if !defined(MODULE) /* pci_domains_supported is not being exported */ \
1083     || !defined(CONFIG_PCI_DOMAINS)
1084 	     !pci_domains_supported ? domain :
1085 #endif
1086 	     domain < 0 || domain > 0xffff)
1087 	    || bus < 0 || bus > 0xff
1088 	    || PCI_SLOT(devfn) != slot
1089 	    || PCI_FUNC(devfn) != func)
1090 		return -EINVAL;
1091 
1092 	pci_dev_id = kmalloc(sizeof(*pci_dev_id), GFP_KERNEL);
1093 	if (!pci_dev_id)
1094 		return -ENOMEM;
1095 
1096 	pr_debug("wants to seize %04x:%02x:%02x.%d\n",
1097 		 domain, bus, slot, func);
1098 
1099 	pcistub_device_id_add_list(pci_dev_id, domain, bus, devfn);
1100 
1101 	return 0;
1102 }
1103 
1104 static int pcistub_device_id_remove(int domain, int bus, int slot, int func)
1105 {
1106 	struct pcistub_device_id *pci_dev_id, *t;
1107 	int err = -ENOENT;
1108 	unsigned long flags;
1109 
1110 	spin_lock_irqsave(&device_ids_lock, flags);
1111 	list_for_each_entry_safe(pci_dev_id, t, &pcistub_device_ids,
1112 				 slot_list) {
1113 		if (pci_dev_id->domain == domain && pci_dev_id->bus == bus
1114 		    && (slot < 0 || PCI_SLOT(pci_dev_id->devfn) == slot)
1115 		    && (func < 0 || PCI_FUNC(pci_dev_id->devfn) == func)) {
1116 			/* Don't break; here because it's possible the same
1117 			 * slot could be in the list more than once
1118 			 */
1119 			list_del(&pci_dev_id->slot_list);
1120 			kfree(pci_dev_id);
1121 
1122 			err = 0;
1123 
1124 			pr_debug("removed %04x:%02x:%02x.%d from seize list\n",
1125 				 domain, bus, slot, func);
1126 		}
1127 	}
1128 	spin_unlock_irqrestore(&device_ids_lock, flags);
1129 
1130 	return err;
1131 }
1132 
1133 static int pcistub_reg_add(int domain, int bus, int slot, int func,
1134 			   unsigned int reg, unsigned int size,
1135 			   unsigned int mask)
1136 {
1137 	int err = 0;
1138 	struct pcistub_device *psdev;
1139 	struct pci_dev *dev;
1140 	struct config_field *field;
1141 
1142 	if (reg > 0xfff || (size < 4 && (mask >> (size * 8))))
1143 		return -EINVAL;
1144 
1145 	psdev = pcistub_device_find(domain, bus, slot, func);
1146 	if (!psdev) {
1147 		err = -ENODEV;
1148 		goto out;
1149 	}
1150 	dev = psdev->dev;
1151 
1152 	field = kzalloc(sizeof(*field), GFP_ATOMIC);
1153 	if (!field) {
1154 		err = -ENOMEM;
1155 		goto out;
1156 	}
1157 
1158 	field->offset = reg;
1159 	field->size = size;
1160 	field->mask = mask;
1161 	field->init = NULL;
1162 	field->reset = NULL;
1163 	field->release = NULL;
1164 	field->clean = xen_pcibk_config_field_free;
1165 
1166 	err = xen_pcibk_config_quirks_add_field(dev, field);
1167 	if (err)
1168 		kfree(field);
1169 out:
1170 	if (psdev)
1171 		pcistub_device_put(psdev);
1172 	return err;
1173 }
1174 
1175 static ssize_t pcistub_slot_add(struct device_driver *drv, const char *buf,
1176 				size_t count)
1177 {
1178 	int domain, bus, slot, func;
1179 	int err;
1180 
1181 	err = str_to_slot(buf, &domain, &bus, &slot, &func);
1182 	if (err)
1183 		goto out;
1184 
1185 	err = pcistub_device_id_add(domain, bus, slot, func);
1186 
1187 out:
1188 	if (!err)
1189 		err = count;
1190 	return err;
1191 }
1192 static DRIVER_ATTR(new_slot, S_IWUSR, NULL, pcistub_slot_add);
1193 
1194 static ssize_t pcistub_slot_remove(struct device_driver *drv, const char *buf,
1195 				   size_t count)
1196 {
1197 	int domain, bus, slot, func;
1198 	int err;
1199 
1200 	err = str_to_slot(buf, &domain, &bus, &slot, &func);
1201 	if (err)
1202 		goto out;
1203 
1204 	err = pcistub_device_id_remove(domain, bus, slot, func);
1205 
1206 out:
1207 	if (!err)
1208 		err = count;
1209 	return err;
1210 }
1211 static DRIVER_ATTR(remove_slot, S_IWUSR, NULL, pcistub_slot_remove);
1212 
1213 static ssize_t pcistub_slot_show(struct device_driver *drv, char *buf)
1214 {
1215 	struct pcistub_device_id *pci_dev_id;
1216 	size_t count = 0;
1217 	unsigned long flags;
1218 
1219 	spin_lock_irqsave(&device_ids_lock, flags);
1220 	list_for_each_entry(pci_dev_id, &pcistub_device_ids, slot_list) {
1221 		if (count >= PAGE_SIZE)
1222 			break;
1223 
1224 		count += scnprintf(buf + count, PAGE_SIZE - count,
1225 				   "%04x:%02x:%02x.%d\n",
1226 				   pci_dev_id->domain, pci_dev_id->bus,
1227 				   PCI_SLOT(pci_dev_id->devfn),
1228 				   PCI_FUNC(pci_dev_id->devfn));
1229 	}
1230 	spin_unlock_irqrestore(&device_ids_lock, flags);
1231 
1232 	return count;
1233 }
1234 static DRIVER_ATTR(slots, S_IRUSR, pcistub_slot_show, NULL);
1235 
1236 static ssize_t pcistub_irq_handler_show(struct device_driver *drv, char *buf)
1237 {
1238 	struct pcistub_device *psdev;
1239 	struct xen_pcibk_dev_data *dev_data;
1240 	size_t count = 0;
1241 	unsigned long flags;
1242 
1243 	spin_lock_irqsave(&pcistub_devices_lock, flags);
1244 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
1245 		if (count >= PAGE_SIZE)
1246 			break;
1247 		if (!psdev->dev)
1248 			continue;
1249 		dev_data = pci_get_drvdata(psdev->dev);
1250 		if (!dev_data)
1251 			continue;
1252 		count +=
1253 		    scnprintf(buf + count, PAGE_SIZE - count,
1254 			      "%s:%s:%sing:%ld\n",
1255 			      pci_name(psdev->dev),
1256 			      dev_data->isr_on ? "on" : "off",
1257 			      dev_data->ack_intr ? "ack" : "not ack",
1258 			      dev_data->handled);
1259 	}
1260 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
1261 	return count;
1262 }
1263 static DRIVER_ATTR(irq_handlers, S_IRUSR, pcistub_irq_handler_show, NULL);
1264 
1265 static ssize_t pcistub_irq_handler_switch(struct device_driver *drv,
1266 					  const char *buf,
1267 					  size_t count)
1268 {
1269 	struct pcistub_device *psdev;
1270 	struct xen_pcibk_dev_data *dev_data;
1271 	int domain, bus, slot, func;
1272 	int err;
1273 
1274 	err = str_to_slot(buf, &domain, &bus, &slot, &func);
1275 	if (err)
1276 		return err;
1277 
1278 	psdev = pcistub_device_find(domain, bus, slot, func);
1279 	if (!psdev) {
1280 		err = -ENOENT;
1281 		goto out;
1282 	}
1283 
1284 	dev_data = pci_get_drvdata(psdev->dev);
1285 	if (!dev_data) {
1286 		err = -ENOENT;
1287 		goto out;
1288 	}
1289 
1290 	dev_dbg(&psdev->dev->dev, "%s fake irq handler: %d->%d\n",
1291 		dev_data->irq_name, dev_data->isr_on,
1292 		!dev_data->isr_on);
1293 
1294 	dev_data->isr_on = !(dev_data->isr_on);
1295 	if (dev_data->isr_on)
1296 		dev_data->ack_intr = 1;
1297 out:
1298 	if (psdev)
1299 		pcistub_device_put(psdev);
1300 	if (!err)
1301 		err = count;
1302 	return err;
1303 }
1304 static DRIVER_ATTR(irq_handler_state, S_IWUSR, NULL,
1305 		   pcistub_irq_handler_switch);
1306 
1307 static ssize_t pcistub_quirk_add(struct device_driver *drv, const char *buf,
1308 				 size_t count)
1309 {
1310 	int domain, bus, slot, func, reg, size, mask;
1311 	int err;
1312 
1313 	err = str_to_quirk(buf, &domain, &bus, &slot, &func, &reg, &size,
1314 			   &mask);
1315 	if (err)
1316 		goto out;
1317 
1318 	err = pcistub_reg_add(domain, bus, slot, func, reg, size, mask);
1319 
1320 out:
1321 	if (!err)
1322 		err = count;
1323 	return err;
1324 }
1325 
1326 static ssize_t pcistub_quirk_show(struct device_driver *drv, char *buf)
1327 {
1328 	int count = 0;
1329 	unsigned long flags;
1330 	struct xen_pcibk_config_quirk *quirk;
1331 	struct xen_pcibk_dev_data *dev_data;
1332 	const struct config_field *field;
1333 	const struct config_field_entry *cfg_entry;
1334 
1335 	spin_lock_irqsave(&device_ids_lock, flags);
1336 	list_for_each_entry(quirk, &xen_pcibk_quirks, quirks_list) {
1337 		if (count >= PAGE_SIZE)
1338 			goto out;
1339 
1340 		count += scnprintf(buf + count, PAGE_SIZE - count,
1341 				   "%02x:%02x.%01x\n\t%04x:%04x:%04x:%04x\n",
1342 				   quirk->pdev->bus->number,
1343 				   PCI_SLOT(quirk->pdev->devfn),
1344 				   PCI_FUNC(quirk->pdev->devfn),
1345 				   quirk->devid.vendor, quirk->devid.device,
1346 				   quirk->devid.subvendor,
1347 				   quirk->devid.subdevice);
1348 
1349 		dev_data = pci_get_drvdata(quirk->pdev);
1350 
1351 		list_for_each_entry(cfg_entry, &dev_data->config_fields, list) {
1352 			field = cfg_entry->field;
1353 			if (count >= PAGE_SIZE)
1354 				goto out;
1355 
1356 			count += scnprintf(buf + count, PAGE_SIZE - count,
1357 					   "\t\t%08x:%01x:%08x\n",
1358 					   cfg_entry->base_offset +
1359 					   field->offset, field->size,
1360 					   field->mask);
1361 		}
1362 	}
1363 
1364 out:
1365 	spin_unlock_irqrestore(&device_ids_lock, flags);
1366 
1367 	return count;
1368 }
1369 static DRIVER_ATTR(quirks, S_IRUSR | S_IWUSR, pcistub_quirk_show,
1370 		   pcistub_quirk_add);
1371 
1372 static ssize_t permissive_add(struct device_driver *drv, const char *buf,
1373 			      size_t count)
1374 {
1375 	int domain, bus, slot, func;
1376 	int err;
1377 	struct pcistub_device *psdev;
1378 	struct xen_pcibk_dev_data *dev_data;
1379 
1380 	err = str_to_slot(buf, &domain, &bus, &slot, &func);
1381 	if (err)
1382 		goto out;
1383 
1384 	psdev = pcistub_device_find(domain, bus, slot, func);
1385 	if (!psdev) {
1386 		err = -ENODEV;
1387 		goto out;
1388 	}
1389 
1390 	dev_data = pci_get_drvdata(psdev->dev);
1391 	/* the driver data for a device should never be null at this point */
1392 	if (!dev_data) {
1393 		err = -ENXIO;
1394 		goto release;
1395 	}
1396 	if (!dev_data->permissive) {
1397 		dev_data->permissive = 1;
1398 		/* Let user know that what they're doing could be unsafe */
1399 		dev_warn(&psdev->dev->dev, "enabling permissive mode "
1400 			 "configuration space accesses!\n");
1401 		dev_warn(&psdev->dev->dev,
1402 			 "permissive mode is potentially unsafe!\n");
1403 	}
1404 release:
1405 	pcistub_device_put(psdev);
1406 out:
1407 	if (!err)
1408 		err = count;
1409 	return err;
1410 }
1411 
1412 static ssize_t permissive_show(struct device_driver *drv, char *buf)
1413 {
1414 	struct pcistub_device *psdev;
1415 	struct xen_pcibk_dev_data *dev_data;
1416 	size_t count = 0;
1417 	unsigned long flags;
1418 	spin_lock_irqsave(&pcistub_devices_lock, flags);
1419 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
1420 		if (count >= PAGE_SIZE)
1421 			break;
1422 		if (!psdev->dev)
1423 			continue;
1424 		dev_data = pci_get_drvdata(psdev->dev);
1425 		if (!dev_data || !dev_data->permissive)
1426 			continue;
1427 		count +=
1428 		    scnprintf(buf + count, PAGE_SIZE - count, "%s\n",
1429 			      pci_name(psdev->dev));
1430 	}
1431 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
1432 	return count;
1433 }
1434 static DRIVER_ATTR(permissive, S_IRUSR | S_IWUSR, permissive_show,
1435 		   permissive_add);
1436 
1437 static void pcistub_exit(void)
1438 {
1439 	driver_remove_file(&xen_pcibk_pci_driver.driver, &driver_attr_new_slot);
1440 	driver_remove_file(&xen_pcibk_pci_driver.driver,
1441 			   &driver_attr_remove_slot);
1442 	driver_remove_file(&xen_pcibk_pci_driver.driver, &driver_attr_slots);
1443 	driver_remove_file(&xen_pcibk_pci_driver.driver, &driver_attr_quirks);
1444 	driver_remove_file(&xen_pcibk_pci_driver.driver,
1445 			   &driver_attr_permissive);
1446 	driver_remove_file(&xen_pcibk_pci_driver.driver,
1447 			   &driver_attr_irq_handlers);
1448 	driver_remove_file(&xen_pcibk_pci_driver.driver,
1449 			   &driver_attr_irq_handler_state);
1450 	pci_unregister_driver(&xen_pcibk_pci_driver);
1451 }
1452 
1453 static int __init pcistub_init(void)
1454 {
1455 	int pos = 0;
1456 	int err = 0;
1457 	int domain, bus, slot, func;
1458 	int parsed;
1459 
1460 	if (pci_devs_to_hide && *pci_devs_to_hide) {
1461 		do {
1462 			parsed = 0;
1463 
1464 			err = sscanf(pci_devs_to_hide + pos,
1465 				     " (%x:%x:%x.%x) %n",
1466 				     &domain, &bus, &slot, &func, &parsed);
1467 			switch (err) {
1468 			case 3:
1469 				func = -1;
1470 				sscanf(pci_devs_to_hide + pos,
1471 				       " (%x:%x:%x.*) %n",
1472 				       &domain, &bus, &slot, &parsed);
1473 				break;
1474 			case 2:
1475 				slot = func = -1;
1476 				sscanf(pci_devs_to_hide + pos,
1477 				       " (%x:%x:*.*) %n",
1478 				       &domain, &bus, &parsed);
1479 				break;
1480 			}
1481 
1482 			if (!parsed) {
1483 				domain = 0;
1484 				err = sscanf(pci_devs_to_hide + pos,
1485 					     " (%x:%x.%x) %n",
1486 					     &bus, &slot, &func, &parsed);
1487 				switch (err) {
1488 				case 2:
1489 					func = -1;
1490 					sscanf(pci_devs_to_hide + pos,
1491 					       " (%x:%x.*) %n",
1492 					       &bus, &slot, &parsed);
1493 					break;
1494 				case 1:
1495 					slot = func = -1;
1496 					sscanf(pci_devs_to_hide + pos,
1497 					       " (%x:*.*) %n",
1498 					       &bus, &parsed);
1499 					break;
1500 				}
1501 			}
1502 
1503 			if (parsed <= 0)
1504 				goto parse_error;
1505 
1506 			err = pcistub_device_id_add(domain, bus, slot, func);
1507 			if (err)
1508 				goto out;
1509 
1510 			pos += parsed;
1511 		} while (pci_devs_to_hide[pos]);
1512 	}
1513 
1514 	/* If we're the first PCI Device Driver to register, we're the
1515 	 * first one to get offered PCI devices as they become
1516 	 * available (and thus we can be the first to grab them)
1517 	 */
1518 	err = pci_register_driver(&xen_pcibk_pci_driver);
1519 	if (err < 0)
1520 		goto out;
1521 
1522 	err = driver_create_file(&xen_pcibk_pci_driver.driver,
1523 				 &driver_attr_new_slot);
1524 	if (!err)
1525 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1526 					 &driver_attr_remove_slot);
1527 	if (!err)
1528 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1529 					 &driver_attr_slots);
1530 	if (!err)
1531 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1532 					 &driver_attr_quirks);
1533 	if (!err)
1534 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1535 					 &driver_attr_permissive);
1536 
1537 	if (!err)
1538 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1539 					 &driver_attr_irq_handlers);
1540 	if (!err)
1541 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1542 					&driver_attr_irq_handler_state);
1543 	if (err)
1544 		pcistub_exit();
1545 
1546 out:
1547 	return err;
1548 
1549 parse_error:
1550 	pr_err("Error parsing pci_devs_to_hide at \"%s\"\n",
1551 	       pci_devs_to_hide + pos);
1552 	return -EINVAL;
1553 }
1554 
1555 #ifndef MODULE
1556 /*
1557  * fs_initcall happens before device_initcall
1558  * so xen_pcibk *should* get called first (b/c we
1559  * want to suck up any device before other drivers
1560  * get a chance by being the first pci device
1561  * driver to register)
1562  */
1563 fs_initcall(pcistub_init);
1564 #endif
1565 
1566 #ifdef CONFIG_PCI_IOV
1567 static struct pcistub_device *find_vfs(const struct pci_dev *pdev)
1568 {
1569 	struct pcistub_device *psdev = NULL;
1570 	unsigned long flags;
1571 	bool found = false;
1572 
1573 	spin_lock_irqsave(&pcistub_devices_lock, flags);
1574 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
1575 		if (!psdev->pdev && psdev->dev != pdev
1576 		    && pci_physfn(psdev->dev) == pdev) {
1577 			found = true;
1578 			break;
1579 		}
1580 	}
1581 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
1582 	if (found)
1583 		return psdev;
1584 	return NULL;
1585 }
1586 
1587 static int pci_stub_notifier(struct notifier_block *nb,
1588 			     unsigned long action, void *data)
1589 {
1590 	struct device *dev = data;
1591 	const struct pci_dev *pdev = to_pci_dev(dev);
1592 
1593 	if (action != BUS_NOTIFY_UNBIND_DRIVER)
1594 		return NOTIFY_DONE;
1595 
1596 	if (!pdev->is_physfn)
1597 		return NOTIFY_DONE;
1598 
1599 	for (;;) {
1600 		struct pcistub_device *psdev = find_vfs(pdev);
1601 		if (!psdev)
1602 			break;
1603 		device_release_driver(&psdev->dev->dev);
1604 	}
1605 	return NOTIFY_DONE;
1606 }
1607 
1608 static struct notifier_block pci_stub_nb = {
1609 	.notifier_call = pci_stub_notifier,
1610 };
1611 #endif
1612 
1613 static int __init xen_pcibk_init(void)
1614 {
1615 	int err;
1616 
1617 	if (!xen_initial_domain())
1618 		return -ENODEV;
1619 
1620 	err = xen_pcibk_config_init();
1621 	if (err)
1622 		return err;
1623 
1624 #ifdef MODULE
1625 	err = pcistub_init();
1626 	if (err < 0)
1627 		return err;
1628 #endif
1629 
1630 	pcistub_init_devices_late();
1631 	err = xen_pcibk_xenbus_register();
1632 	if (err)
1633 		pcistub_exit();
1634 #ifdef CONFIG_PCI_IOV
1635 	else
1636 		bus_register_notifier(&pci_bus_type, &pci_stub_nb);
1637 #endif
1638 
1639 	return err;
1640 }
1641 
1642 static void __exit xen_pcibk_cleanup(void)
1643 {
1644 #ifdef CONFIG_PCI_IOV
1645 	bus_unregister_notifier(&pci_bus_type, &pci_stub_nb);
1646 #endif
1647 	xen_pcibk_xenbus_unregister();
1648 	pcistub_exit();
1649 }
1650 
1651 module_init(xen_pcibk_init);
1652 module_exit(xen_pcibk_cleanup);
1653 
1654 MODULE_LICENSE("Dual BSD/GPL");
1655 MODULE_ALIAS("xen-backend:pci");
1656