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_KERNEL);
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 (dev_data &&
110 	    pci_load_and_free_saved_state(dev, &dev_data->pci_saved_state))
111 		dev_info(&dev->dev, "Could not reload PCI state\n");
112 	else
113 		pci_restore_state(dev);
114 
115 	if (dev->msix_cap) {
116 		struct physdev_pci_device ppdev = {
117 			.seg = pci_domain_nr(dev->bus),
118 			.bus = dev->bus->number,
119 			.devfn = dev->devfn
120 		};
121 		int err = HYPERVISOR_physdev_op(PHYSDEVOP_release_msix,
122 						&ppdev);
123 
124 		if (err && err != -ENOSYS)
125 			dev_warn(&dev->dev, "MSI-X release failed (%d)\n",
126 				 err);
127 	}
128 
129 	/* Disable the device */
130 	xen_pcibk_reset_device(dev);
131 
132 	kfree(dev_data);
133 	pci_set_drvdata(dev, NULL);
134 
135 	/* Clean-up the device */
136 	xen_pcibk_config_free_dyn_fields(dev);
137 	xen_pcibk_config_free_dev(dev);
138 
139 	pci_clear_dev_assigned(dev);
140 	pci_dev_put(dev);
141 
142 	kfree(psdev);
143 }
144 
145 static inline void pcistub_device_get(struct pcistub_device *psdev)
146 {
147 	kref_get(&psdev->kref);
148 }
149 
150 static inline void pcistub_device_put(struct pcistub_device *psdev)
151 {
152 	kref_put(&psdev->kref, pcistub_device_release);
153 }
154 
155 static struct pcistub_device *pcistub_device_find_locked(int domain, int bus,
156 							 int slot, int func)
157 {
158 	struct pcistub_device *psdev;
159 
160 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
161 		if (psdev->dev != NULL
162 		    && domain == pci_domain_nr(psdev->dev->bus)
163 		    && bus == psdev->dev->bus->number
164 		    && slot == PCI_SLOT(psdev->dev->devfn)
165 		    && func == PCI_FUNC(psdev->dev->devfn)) {
166 			return psdev;
167 		}
168 	}
169 
170 	return NULL;
171 }
172 
173 static struct pcistub_device *pcistub_device_find(int domain, int bus,
174 						  int slot, int func)
175 {
176 	struct pcistub_device *psdev;
177 	unsigned long flags;
178 
179 	spin_lock_irqsave(&pcistub_devices_lock, flags);
180 
181 	psdev = pcistub_device_find_locked(domain, bus, slot, func);
182 	if (psdev)
183 		pcistub_device_get(psdev);
184 
185 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
186 	return psdev;
187 }
188 
189 static struct pci_dev *pcistub_device_get_pci_dev(struct xen_pcibk_device *pdev,
190 						  struct pcistub_device *psdev)
191 {
192 	struct pci_dev *pci_dev = NULL;
193 	unsigned long flags;
194 
195 	pcistub_device_get(psdev);
196 
197 	spin_lock_irqsave(&psdev->lock, flags);
198 	if (!psdev->pdev) {
199 		psdev->pdev = pdev;
200 		pci_dev = psdev->dev;
201 	}
202 	spin_unlock_irqrestore(&psdev->lock, flags);
203 
204 	if (!pci_dev)
205 		pcistub_device_put(psdev);
206 
207 	return pci_dev;
208 }
209 
210 struct pci_dev *pcistub_get_pci_dev_by_slot(struct xen_pcibk_device *pdev,
211 					    int domain, int bus,
212 					    int slot, int func)
213 {
214 	struct pcistub_device *psdev;
215 	struct pci_dev *found_dev = NULL;
216 	unsigned long flags;
217 
218 	spin_lock_irqsave(&pcistub_devices_lock, flags);
219 
220 	psdev = pcistub_device_find_locked(domain, bus, slot, func);
221 	if (psdev)
222 		found_dev = pcistub_device_get_pci_dev(pdev, psdev);
223 
224 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
225 	return found_dev;
226 }
227 
228 struct pci_dev *pcistub_get_pci_dev(struct xen_pcibk_device *pdev,
229 				    struct pci_dev *dev)
230 {
231 	struct pcistub_device *psdev;
232 	struct pci_dev *found_dev = NULL;
233 	unsigned long flags;
234 
235 	spin_lock_irqsave(&pcistub_devices_lock, flags);
236 
237 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
238 		if (psdev->dev == dev) {
239 			found_dev = pcistub_device_get_pci_dev(pdev, psdev);
240 			break;
241 		}
242 	}
243 
244 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
245 	return found_dev;
246 }
247 
248 /*
249  * Called when:
250  *  - XenBus state has been reconfigure (pci unplug). See xen_pcibk_remove_device
251  *  - XenBus state has been disconnected (guest shutdown). See xen_pcibk_xenbus_remove
252  *  - 'echo BDF > unbind' on pciback module with no guest attached. See pcistub_remove
253  *  - 'echo BDF > unbind' with a guest still using it. See pcistub_remove
254  *
255  *  As such we have to be careful.
256  *
257  *  To make this easier, the caller has to hold the device lock.
258  */
259 void pcistub_put_pci_dev(struct pci_dev *dev)
260 {
261 	struct pcistub_device *psdev, *found_psdev = NULL;
262 	unsigned long flags;
263 	struct xen_pcibk_dev_data *dev_data;
264 	int ret;
265 
266 	spin_lock_irqsave(&pcistub_devices_lock, flags);
267 
268 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
269 		if (psdev->dev == dev) {
270 			found_psdev = psdev;
271 			break;
272 		}
273 	}
274 
275 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
276 	if (WARN_ON(!found_psdev))
277 		return;
278 
279 	/*hold this lock for avoiding breaking link between
280 	* pcistub and xen_pcibk when AER is in processing
281 	*/
282 	down_write(&pcistub_sem);
283 	/* Cleanup our device
284 	 * (so it's ready for the next domain)
285 	 */
286 	device_lock_assert(&dev->dev);
287 	__pci_reset_function_locked(dev);
288 
289 	dev_data = pci_get_drvdata(dev);
290 	ret = pci_load_saved_state(dev, dev_data->pci_saved_state);
291 	if (!ret) {
292 		/*
293 		 * The usual sequence is pci_save_state & pci_restore_state
294 		 * but the guest might have messed the configuration space up.
295 		 * Use the initial version (when device was bound to us).
296 		 */
297 		pci_restore_state(dev);
298 	} else
299 		dev_info(&dev->dev, "Could not reload PCI state\n");
300 	/* This disables the device. */
301 	xen_pcibk_reset_device(dev);
302 
303 	/* And cleanup up our emulated fields. */
304 	xen_pcibk_config_reset_dev(dev);
305 	xen_pcibk_config_free_dyn_fields(dev);
306 
307 	dev_data->allow_interrupt_control = 0;
308 
309 	xen_unregister_device_domain_owner(dev);
310 
311 	spin_lock_irqsave(&found_psdev->lock, flags);
312 	found_psdev->pdev = NULL;
313 	spin_unlock_irqrestore(&found_psdev->lock, flags);
314 
315 	pcistub_device_put(found_psdev);
316 	up_write(&pcistub_sem);
317 }
318 
319 static int pcistub_match_one(struct pci_dev *dev,
320 			     struct pcistub_device_id *pdev_id)
321 {
322 	/* Match the specified device by domain, bus, slot, func and also if
323 	 * any of the device's parent bridges match.
324 	 */
325 	for (; dev != NULL; dev = dev->bus->self) {
326 		if (pci_domain_nr(dev->bus) == pdev_id->domain
327 		    && dev->bus->number == pdev_id->bus
328 		    && dev->devfn == pdev_id->devfn)
329 			return 1;
330 
331 		/* Sometimes topmost bridge links to itself. */
332 		if (dev == dev->bus->self)
333 			break;
334 	}
335 
336 	return 0;
337 }
338 
339 static int pcistub_match(struct pci_dev *dev)
340 {
341 	struct pcistub_device_id *pdev_id;
342 	unsigned long flags;
343 	int found = 0;
344 
345 	spin_lock_irqsave(&device_ids_lock, flags);
346 	list_for_each_entry(pdev_id, &pcistub_device_ids, slot_list) {
347 		if (pcistub_match_one(dev, pdev_id)) {
348 			found = 1;
349 			break;
350 		}
351 	}
352 	spin_unlock_irqrestore(&device_ids_lock, flags);
353 
354 	return found;
355 }
356 
357 static int pcistub_init_device(struct pci_dev *dev)
358 {
359 	struct xen_pcibk_dev_data *dev_data;
360 	int err = 0;
361 
362 	dev_dbg(&dev->dev, "initializing...\n");
363 
364 	/* The PCI backend is not intended to be a module (or to work with
365 	 * removable PCI devices (yet). If it were, xen_pcibk_config_free()
366 	 * would need to be called somewhere to free the memory allocated
367 	 * here and then to call kfree(pci_get_drvdata(psdev->dev)).
368 	 */
369 	dev_data = kzalloc(sizeof(*dev_data) +  strlen(DRV_NAME "[]")
370 				+ strlen(pci_name(dev)) + 1, GFP_KERNEL);
371 	if (!dev_data) {
372 		err = -ENOMEM;
373 		goto out;
374 	}
375 	pci_set_drvdata(dev, dev_data);
376 
377 	/*
378 	 * Setup name for fake IRQ handler. It will only be enabled
379 	 * once the device is turned on by the guest.
380 	 */
381 	sprintf(dev_data->irq_name, DRV_NAME "[%s]", pci_name(dev));
382 
383 	dev_dbg(&dev->dev, "initializing config\n");
384 
385 	init_waitqueue_head(&xen_pcibk_aer_wait_queue);
386 	err = xen_pcibk_config_init_dev(dev);
387 	if (err)
388 		goto out;
389 
390 	/* HACK: Force device (& ACPI) to determine what IRQ it's on - we
391 	 * must do this here because pcibios_enable_device may specify
392 	 * the pci device's true irq (and possibly its other resources)
393 	 * if they differ from what's in the configuration space.
394 	 * This makes the assumption that the device's resources won't
395 	 * change after this point (otherwise this code may break!)
396 	 */
397 	dev_dbg(&dev->dev, "enabling device\n");
398 	err = pci_enable_device(dev);
399 	if (err)
400 		goto config_release;
401 
402 	if (dev->msix_cap) {
403 		struct physdev_pci_device ppdev = {
404 			.seg = pci_domain_nr(dev->bus),
405 			.bus = dev->bus->number,
406 			.devfn = dev->devfn
407 		};
408 
409 		err = HYPERVISOR_physdev_op(PHYSDEVOP_prepare_msix, &ppdev);
410 		if (err && err != -ENOSYS)
411 			dev_err(&dev->dev, "MSI-X preparation failed (%d)\n",
412 				err);
413 	}
414 
415 	/* We need the device active to save the state. */
416 	dev_dbg(&dev->dev, "save state of device\n");
417 	pci_save_state(dev);
418 	dev_data->pci_saved_state = pci_store_saved_state(dev);
419 	if (!dev_data->pci_saved_state)
420 		dev_err(&dev->dev, "Could not store PCI conf saved state!\n");
421 	else {
422 		dev_dbg(&dev->dev, "resetting (FLR, D3, etc) the device\n");
423 		__pci_reset_function_locked(dev);
424 		pci_restore_state(dev);
425 	}
426 	/* Now disable the device (this also ensures some private device
427 	 * data is setup before we export)
428 	 */
429 	dev_dbg(&dev->dev, "reset device\n");
430 	xen_pcibk_reset_device(dev);
431 
432 	pci_set_dev_assigned(dev);
433 	return 0;
434 
435 config_release:
436 	xen_pcibk_config_free_dev(dev);
437 
438 out:
439 	pci_set_drvdata(dev, NULL);
440 	kfree(dev_data);
441 	return err;
442 }
443 
444 /*
445  * Because some initialization still happens on
446  * devices during fs_initcall, we need to defer
447  * full initialization of our devices until
448  * device_initcall.
449  */
450 static int __init pcistub_init_devices_late(void)
451 {
452 	struct pcistub_device *psdev;
453 	unsigned long flags;
454 	int err = 0;
455 
456 	spin_lock_irqsave(&pcistub_devices_lock, flags);
457 
458 	while (!list_empty(&seized_devices)) {
459 		psdev = container_of(seized_devices.next,
460 				     struct pcistub_device, dev_list);
461 		list_del(&psdev->dev_list);
462 
463 		spin_unlock_irqrestore(&pcistub_devices_lock, flags);
464 
465 		err = pcistub_init_device(psdev->dev);
466 		if (err) {
467 			dev_err(&psdev->dev->dev,
468 				"error %d initializing device\n", err);
469 			kfree(psdev);
470 			psdev = NULL;
471 		}
472 
473 		spin_lock_irqsave(&pcistub_devices_lock, flags);
474 
475 		if (psdev)
476 			list_add_tail(&psdev->dev_list, &pcistub_devices);
477 	}
478 
479 	initialize_devices = 1;
480 
481 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
482 
483 	return 0;
484 }
485 
486 static void pcistub_device_id_add_list(struct pcistub_device_id *new,
487 				       int domain, int bus, unsigned int devfn)
488 {
489 	struct pcistub_device_id *pci_dev_id;
490 	unsigned long flags;
491 	int found = 0;
492 
493 	spin_lock_irqsave(&device_ids_lock, flags);
494 
495 	list_for_each_entry(pci_dev_id, &pcistub_device_ids, slot_list) {
496 		if (pci_dev_id->domain == domain && pci_dev_id->bus == bus &&
497 		    pci_dev_id->devfn == devfn) {
498 			found = 1;
499 			break;
500 		}
501 	}
502 
503 	if (!found) {
504 		new->domain = domain;
505 		new->bus = bus;
506 		new->devfn = devfn;
507 		list_add_tail(&new->slot_list, &pcistub_device_ids);
508 	}
509 
510 	spin_unlock_irqrestore(&device_ids_lock, flags);
511 
512 	if (found)
513 		kfree(new);
514 }
515 
516 static int pcistub_seize(struct pci_dev *dev,
517 			 struct pcistub_device_id *pci_dev_id)
518 {
519 	struct pcistub_device *psdev;
520 	unsigned long flags;
521 	int err = 0;
522 
523 	psdev = pcistub_device_alloc(dev);
524 	if (!psdev) {
525 		kfree(pci_dev_id);
526 		return -ENOMEM;
527 	}
528 
529 	spin_lock_irqsave(&pcistub_devices_lock, flags);
530 
531 	if (initialize_devices) {
532 		spin_unlock_irqrestore(&pcistub_devices_lock, flags);
533 
534 		/* don't want irqs disabled when calling pcistub_init_device */
535 		err = pcistub_init_device(psdev->dev);
536 
537 		spin_lock_irqsave(&pcistub_devices_lock, flags);
538 
539 		if (!err)
540 			list_add(&psdev->dev_list, &pcistub_devices);
541 	} else {
542 		dev_dbg(&dev->dev, "deferring initialization\n");
543 		list_add(&psdev->dev_list, &seized_devices);
544 	}
545 
546 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
547 
548 	if (err) {
549 		kfree(pci_dev_id);
550 		pcistub_device_put(psdev);
551 	} else if (pci_dev_id)
552 		pcistub_device_id_add_list(pci_dev_id, pci_domain_nr(dev->bus),
553 					   dev->bus->number, dev->devfn);
554 
555 	return err;
556 }
557 
558 /* Called when 'bind'. This means we must _NOT_ call pci_reset_function or
559  * other functions that take the sysfs lock. */
560 static int pcistub_probe(struct pci_dev *dev, const struct pci_device_id *id)
561 {
562 	int err = 0, match;
563 	struct pcistub_device_id *pci_dev_id = NULL;
564 
565 	dev_dbg(&dev->dev, "probing...\n");
566 
567 	match = pcistub_match(dev);
568 
569 	if ((dev->driver_override &&
570 	     !strcmp(dev->driver_override, PCISTUB_DRIVER_NAME)) ||
571 	    match) {
572 
573 		if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL
574 		    && dev->hdr_type != PCI_HEADER_TYPE_BRIDGE) {
575 			dev_err(&dev->dev, "can't export pci devices that "
576 				"don't have a normal (0) or bridge (1) "
577 				"header type!\n");
578 			err = -ENODEV;
579 			goto out;
580 		}
581 
582 		if (!match) {
583 			pci_dev_id = kmalloc(sizeof(*pci_dev_id), GFP_KERNEL);
584 			if (!pci_dev_id) {
585 				err = -ENOMEM;
586 				goto out;
587 			}
588 		}
589 
590 		dev_info(&dev->dev, "seizing device\n");
591 		err = pcistub_seize(dev, pci_dev_id);
592 	} else
593 		/* Didn't find the device */
594 		err = -ENODEV;
595 
596 out:
597 	return err;
598 }
599 
600 /* Called when 'unbind'. This means we must _NOT_ call pci_reset_function or
601  * other functions that take the sysfs lock. */
602 static void pcistub_remove(struct pci_dev *dev)
603 {
604 	struct pcistub_device *psdev, *found_psdev = NULL;
605 	unsigned long flags;
606 
607 	dev_dbg(&dev->dev, "removing\n");
608 
609 	spin_lock_irqsave(&pcistub_devices_lock, flags);
610 
611 	xen_pcibk_config_quirk_release(dev);
612 
613 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
614 		if (psdev->dev == dev) {
615 			found_psdev = psdev;
616 			break;
617 		}
618 	}
619 
620 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
621 
622 	if (found_psdev) {
623 		dev_dbg(&dev->dev, "found device to remove %s\n",
624 			found_psdev->pdev ? "- in-use" : "");
625 
626 		if (found_psdev->pdev) {
627 			int domid = xen_find_device_domain_owner(dev);
628 
629 			pr_warn("****** removing device %s while still in-use by domain %d! ******\n",
630 			       pci_name(found_psdev->dev), domid);
631 			pr_warn("****** driver domain may still access this device's i/o resources!\n");
632 			pr_warn("****** shutdown driver domain before binding device\n");
633 			pr_warn("****** to other drivers or domains\n");
634 
635 			/* N.B. This ends up calling pcistub_put_pci_dev which ends up
636 			 * doing the FLR. */
637 			xen_pcibk_release_pci_dev(found_psdev->pdev,
638 						found_psdev->dev,
639 						false /* caller holds the lock. */);
640 		}
641 
642 		spin_lock_irqsave(&pcistub_devices_lock, flags);
643 		list_del(&found_psdev->dev_list);
644 		spin_unlock_irqrestore(&pcistub_devices_lock, flags);
645 
646 		/* the final put for releasing from the list */
647 		pcistub_device_put(found_psdev);
648 	}
649 }
650 
651 static const struct pci_device_id pcistub_ids[] = {
652 	{
653 	 .vendor = PCI_ANY_ID,
654 	 .device = PCI_ANY_ID,
655 	 .subvendor = PCI_ANY_ID,
656 	 .subdevice = PCI_ANY_ID,
657 	 },
658 	{0,},
659 };
660 
661 #define PCI_NODENAME_MAX 40
662 static void kill_domain_by_device(struct pcistub_device *psdev)
663 {
664 	struct xenbus_transaction xbt;
665 	int err;
666 	char nodename[PCI_NODENAME_MAX];
667 
668 	BUG_ON(!psdev);
669 	snprintf(nodename, PCI_NODENAME_MAX, "/local/domain/0/backend/pci/%d/0",
670 		psdev->pdev->xdev->otherend_id);
671 
672 again:
673 	err = xenbus_transaction_start(&xbt);
674 	if (err) {
675 		dev_err(&psdev->dev->dev,
676 			"error %d when start xenbus transaction\n", err);
677 		return;
678 	}
679 	/*PV AER handlers will set this flag*/
680 	xenbus_printf(xbt, nodename, "aerState" , "aerfail");
681 	err = xenbus_transaction_end(xbt, 0);
682 	if (err) {
683 		if (err == -EAGAIN)
684 			goto again;
685 		dev_err(&psdev->dev->dev,
686 			"error %d when end xenbus transaction\n", err);
687 		return;
688 	}
689 }
690 
691 /* For each aer recovery step error_detected, mmio_enabled, etc, front_end and
692  * backend need to have cooperation. In xen_pcibk, those steps will do similar
693  * jobs: send service request and waiting for front_end response.
694 */
695 static pci_ers_result_t common_process(struct pcistub_device *psdev,
696 				       pci_channel_state_t state, int aer_cmd,
697 				       pci_ers_result_t result)
698 {
699 	pci_ers_result_t res = result;
700 	struct xen_pcie_aer_op *aer_op;
701 	struct xen_pcibk_device *pdev = psdev->pdev;
702 	struct xen_pci_sharedinfo *sh_info = pdev->sh_info;
703 	int ret;
704 
705 	/*with PV AER drivers*/
706 	aer_op = &(sh_info->aer_op);
707 	aer_op->cmd = aer_cmd ;
708 	/*useful for error_detected callback*/
709 	aer_op->err = state;
710 	/*pcifront_end BDF*/
711 	ret = xen_pcibk_get_pcifront_dev(psdev->dev, psdev->pdev,
712 		&aer_op->domain, &aer_op->bus, &aer_op->devfn);
713 	if (!ret) {
714 		dev_err(&psdev->dev->dev,
715 			DRV_NAME ": failed to get pcifront device\n");
716 		return PCI_ERS_RESULT_NONE;
717 	}
718 	wmb();
719 
720 	dev_dbg(&psdev->dev->dev,
721 			DRV_NAME ": aer_op %x dom %x bus %x devfn %x\n",
722 			aer_cmd, aer_op->domain, aer_op->bus, aer_op->devfn);
723 	/*local flag to mark there's aer request, xen_pcibk callback will use
724 	* this flag to judge whether we need to check pci-front give aer
725 	* service ack signal
726 	*/
727 	set_bit(_PCIB_op_pending, (unsigned long *)&pdev->flags);
728 
729 	/*It is possible that a pcifront conf_read_write ops request invokes
730 	* the callback which cause the spurious execution of wake_up.
731 	* Yet it is harmless and better than a spinlock here
732 	*/
733 	set_bit(_XEN_PCIB_active,
734 		(unsigned long *)&sh_info->flags);
735 	wmb();
736 	notify_remote_via_irq(pdev->evtchn_irq);
737 
738 	ret = wait_event_timeout(xen_pcibk_aer_wait_queue,
739 				 !(test_bit(_XEN_PCIB_active, (unsigned long *)
740 				 &sh_info->flags)), 300*HZ);
741 
742 	if (!ret) {
743 		if (test_bit(_XEN_PCIB_active,
744 			(unsigned long *)&sh_info->flags)) {
745 			dev_err(&psdev->dev->dev,
746 				"pcifront aer process not responding!\n");
747 			clear_bit(_XEN_PCIB_active,
748 			  (unsigned long *)&sh_info->flags);
749 			aer_op->err = PCI_ERS_RESULT_NONE;
750 			return res;
751 		}
752 	}
753 	clear_bit(_PCIB_op_pending, (unsigned long *)&pdev->flags);
754 
755 	if (test_bit(_XEN_PCIF_active,
756 		(unsigned long *)&sh_info->flags)) {
757 		dev_dbg(&psdev->dev->dev,
758 			"schedule pci_conf service in " DRV_NAME "\n");
759 		xen_pcibk_test_and_schedule_op(psdev->pdev);
760 	}
761 
762 	res = (pci_ers_result_t)aer_op->err;
763 	return res;
764 }
765 
766 /*
767 * xen_pcibk_slot_reset: it will send the slot_reset request to  pcifront in case
768 * of the device driver could provide this service, and then wait for pcifront
769 * ack.
770 * @dev: pointer to PCI devices
771 * return value is used by aer_core do_recovery policy
772 */
773 static pci_ers_result_t xen_pcibk_slot_reset(struct pci_dev *dev)
774 {
775 	struct pcistub_device *psdev;
776 	pci_ers_result_t result;
777 
778 	result = PCI_ERS_RESULT_RECOVERED;
779 	dev_dbg(&dev->dev, "xen_pcibk_slot_reset(bus:%x,devfn:%x)\n",
780 		dev->bus->number, dev->devfn);
781 
782 	down_write(&pcistub_sem);
783 	psdev = pcistub_device_find(pci_domain_nr(dev->bus),
784 				dev->bus->number,
785 				PCI_SLOT(dev->devfn),
786 				PCI_FUNC(dev->devfn));
787 
788 	if (!psdev || !psdev->pdev) {
789 		dev_err(&dev->dev,
790 			DRV_NAME " device is not found/assigned\n");
791 		goto end;
792 	}
793 
794 	if (!psdev->pdev->sh_info) {
795 		dev_err(&dev->dev, DRV_NAME " device is not connected or owned"
796 			" by HVM, kill it\n");
797 		kill_domain_by_device(psdev);
798 		goto end;
799 	}
800 
801 	if (!test_bit(_XEN_PCIB_AERHANDLER,
802 		(unsigned long *)&psdev->pdev->sh_info->flags)) {
803 		dev_err(&dev->dev,
804 			"guest with no AER driver should have been killed\n");
805 		goto end;
806 	}
807 	result = common_process(psdev, 1, XEN_PCI_OP_aer_slotreset, result);
808 
809 	if (result == PCI_ERS_RESULT_NONE ||
810 		result == PCI_ERS_RESULT_DISCONNECT) {
811 		dev_dbg(&dev->dev,
812 			"No AER slot_reset service or disconnected!\n");
813 		kill_domain_by_device(psdev);
814 	}
815 end:
816 	if (psdev)
817 		pcistub_device_put(psdev);
818 	up_write(&pcistub_sem);
819 	return result;
820 
821 }
822 
823 
824 /*xen_pcibk_mmio_enabled: it will send the mmio_enabled request to  pcifront
825 * in case of the device driver could provide this service, and then wait
826 * for pcifront ack
827 * @dev: pointer to PCI devices
828 * return value is used by aer_core do_recovery policy
829 */
830 
831 static pci_ers_result_t xen_pcibk_mmio_enabled(struct pci_dev *dev)
832 {
833 	struct pcistub_device *psdev;
834 	pci_ers_result_t result;
835 
836 	result = PCI_ERS_RESULT_RECOVERED;
837 	dev_dbg(&dev->dev, "xen_pcibk_mmio_enabled(bus:%x,devfn:%x)\n",
838 		dev->bus->number, dev->devfn);
839 
840 	down_write(&pcistub_sem);
841 	psdev = pcistub_device_find(pci_domain_nr(dev->bus),
842 				dev->bus->number,
843 				PCI_SLOT(dev->devfn),
844 				PCI_FUNC(dev->devfn));
845 
846 	if (!psdev || !psdev->pdev) {
847 		dev_err(&dev->dev,
848 			DRV_NAME " device is not found/assigned\n");
849 		goto end;
850 	}
851 
852 	if (!psdev->pdev->sh_info) {
853 		dev_err(&dev->dev, DRV_NAME " device is not connected or owned"
854 			" by HVM, kill it\n");
855 		kill_domain_by_device(psdev);
856 		goto end;
857 	}
858 
859 	if (!test_bit(_XEN_PCIB_AERHANDLER,
860 		(unsigned long *)&psdev->pdev->sh_info->flags)) {
861 		dev_err(&dev->dev,
862 			"guest with no AER driver should have been killed\n");
863 		goto end;
864 	}
865 	result = common_process(psdev, 1, XEN_PCI_OP_aer_mmio, result);
866 
867 	if (result == PCI_ERS_RESULT_NONE ||
868 		result == PCI_ERS_RESULT_DISCONNECT) {
869 		dev_dbg(&dev->dev,
870 			"No AER mmio_enabled service or disconnected!\n");
871 		kill_domain_by_device(psdev);
872 	}
873 end:
874 	if (psdev)
875 		pcistub_device_put(psdev);
876 	up_write(&pcistub_sem);
877 	return result;
878 }
879 
880 /*xen_pcibk_error_detected: it will send the error_detected request to  pcifront
881 * in case of the device driver could provide this service, and then wait
882 * for pcifront ack.
883 * @dev: pointer to PCI devices
884 * @error: the current PCI connection state
885 * return value is used by aer_core do_recovery policy
886 */
887 
888 static pci_ers_result_t xen_pcibk_error_detected(struct pci_dev *dev,
889 	pci_channel_state_t error)
890 {
891 	struct pcistub_device *psdev;
892 	pci_ers_result_t result;
893 
894 	result = PCI_ERS_RESULT_CAN_RECOVER;
895 	dev_dbg(&dev->dev, "xen_pcibk_error_detected(bus:%x,devfn:%x)\n",
896 		dev->bus->number, dev->devfn);
897 
898 	down_write(&pcistub_sem);
899 	psdev = pcistub_device_find(pci_domain_nr(dev->bus),
900 				dev->bus->number,
901 				PCI_SLOT(dev->devfn),
902 				PCI_FUNC(dev->devfn));
903 
904 	if (!psdev || !psdev->pdev) {
905 		dev_err(&dev->dev,
906 			DRV_NAME " device is not found/assigned\n");
907 		goto end;
908 	}
909 
910 	if (!psdev->pdev->sh_info) {
911 		dev_err(&dev->dev, DRV_NAME " device is not connected or owned"
912 			" by HVM, kill it\n");
913 		kill_domain_by_device(psdev);
914 		goto end;
915 	}
916 
917 	/*Guest owns the device yet no aer handler regiested, kill guest*/
918 	if (!test_bit(_XEN_PCIB_AERHANDLER,
919 		(unsigned long *)&psdev->pdev->sh_info->flags)) {
920 		dev_dbg(&dev->dev, "guest may have no aer driver, kill it\n");
921 		kill_domain_by_device(psdev);
922 		goto end;
923 	}
924 	result = common_process(psdev, error, XEN_PCI_OP_aer_detected, result);
925 
926 	if (result == PCI_ERS_RESULT_NONE ||
927 		result == PCI_ERS_RESULT_DISCONNECT) {
928 		dev_dbg(&dev->dev,
929 			"No AER error_detected service or disconnected!\n");
930 		kill_domain_by_device(psdev);
931 	}
932 end:
933 	if (psdev)
934 		pcistub_device_put(psdev);
935 	up_write(&pcistub_sem);
936 	return result;
937 }
938 
939 /*xen_pcibk_error_resume: it will send the error_resume request to  pcifront
940 * in case of the device driver could provide this service, and then wait
941 * for pcifront ack.
942 * @dev: pointer to PCI devices
943 */
944 
945 static void xen_pcibk_error_resume(struct pci_dev *dev)
946 {
947 	struct pcistub_device *psdev;
948 
949 	dev_dbg(&dev->dev, "xen_pcibk_error_resume(bus:%x,devfn:%x)\n",
950 		dev->bus->number, dev->devfn);
951 
952 	down_write(&pcistub_sem);
953 	psdev = pcistub_device_find(pci_domain_nr(dev->bus),
954 				dev->bus->number,
955 				PCI_SLOT(dev->devfn),
956 				PCI_FUNC(dev->devfn));
957 
958 	if (!psdev || !psdev->pdev) {
959 		dev_err(&dev->dev,
960 			DRV_NAME " device is not found/assigned\n");
961 		goto end;
962 	}
963 
964 	if (!psdev->pdev->sh_info) {
965 		dev_err(&dev->dev, DRV_NAME " device is not connected or owned"
966 			" by HVM, kill it\n");
967 		kill_domain_by_device(psdev);
968 		goto end;
969 	}
970 
971 	if (!test_bit(_XEN_PCIB_AERHANDLER,
972 		(unsigned long *)&psdev->pdev->sh_info->flags)) {
973 		dev_err(&dev->dev,
974 			"guest with no AER driver should have been killed\n");
975 		kill_domain_by_device(psdev);
976 		goto end;
977 	}
978 	common_process(psdev, 1, XEN_PCI_OP_aer_resume,
979 		       PCI_ERS_RESULT_RECOVERED);
980 end:
981 	if (psdev)
982 		pcistub_device_put(psdev);
983 	up_write(&pcistub_sem);
984 	return;
985 }
986 
987 /*add xen_pcibk AER handling*/
988 static const struct pci_error_handlers xen_pcibk_error_handler = {
989 	.error_detected = xen_pcibk_error_detected,
990 	.mmio_enabled = xen_pcibk_mmio_enabled,
991 	.slot_reset = xen_pcibk_slot_reset,
992 	.resume = xen_pcibk_error_resume,
993 };
994 
995 /*
996  * Note: There is no MODULE_DEVICE_TABLE entry here because this isn't
997  * for a normal device. I don't want it to be loaded automatically.
998  */
999 
1000 static struct pci_driver xen_pcibk_pci_driver = {
1001 	/* The name should be xen_pciback, but until the tools are updated
1002 	 * we will keep it as pciback. */
1003 	.name = PCISTUB_DRIVER_NAME,
1004 	.id_table = pcistub_ids,
1005 	.probe = pcistub_probe,
1006 	.remove = pcistub_remove,
1007 	.err_handler = &xen_pcibk_error_handler,
1008 };
1009 
1010 static inline int str_to_slot(const char *buf, int *domain, int *bus,
1011 			      int *slot, int *func)
1012 {
1013 	int parsed = 0;
1014 
1015 	switch (sscanf(buf, " %x:%x:%x.%x %n", domain, bus, slot, func,
1016 		       &parsed)) {
1017 	case 3:
1018 		*func = -1;
1019 		sscanf(buf, " %x:%x:%x.* %n", domain, bus, slot, &parsed);
1020 		break;
1021 	case 2:
1022 		*slot = *func = -1;
1023 		sscanf(buf, " %x:%x:*.* %n", domain, bus, &parsed);
1024 		break;
1025 	}
1026 	if (parsed && !buf[parsed])
1027 		return 0;
1028 
1029 	/* try again without domain */
1030 	*domain = 0;
1031 	switch (sscanf(buf, " %x:%x.%x %n", bus, slot, func, &parsed)) {
1032 	case 2:
1033 		*func = -1;
1034 		sscanf(buf, " %x:%x.* %n", bus, slot, &parsed);
1035 		break;
1036 	case 1:
1037 		*slot = *func = -1;
1038 		sscanf(buf, " %x:*.* %n", bus, &parsed);
1039 		break;
1040 	}
1041 	if (parsed && !buf[parsed])
1042 		return 0;
1043 
1044 	return -EINVAL;
1045 }
1046 
1047 static inline int str_to_quirk(const char *buf, int *domain, int *bus, int
1048 			       *slot, int *func, int *reg, int *size, int *mask)
1049 {
1050 	int parsed = 0;
1051 
1052 	sscanf(buf, " %x:%x:%x.%x-%x:%x:%x %n", domain, bus, slot, func,
1053 	       reg, size, mask, &parsed);
1054 	if (parsed && !buf[parsed])
1055 		return 0;
1056 
1057 	/* try again without domain */
1058 	*domain = 0;
1059 	sscanf(buf, " %x:%x.%x-%x:%x:%x %n", bus, slot, func, reg, size,
1060 	       mask, &parsed);
1061 	if (parsed && !buf[parsed])
1062 		return 0;
1063 
1064 	return -EINVAL;
1065 }
1066 
1067 static int pcistub_device_id_add(int domain, int bus, int slot, int func)
1068 {
1069 	struct pcistub_device_id *pci_dev_id;
1070 	int rc = 0, devfn = PCI_DEVFN(slot, func);
1071 
1072 	if (slot < 0) {
1073 		for (slot = 0; !rc && slot < 32; ++slot)
1074 			rc = pcistub_device_id_add(domain, bus, slot, func);
1075 		return rc;
1076 	}
1077 
1078 	if (func < 0) {
1079 		for (func = 0; !rc && func < 8; ++func)
1080 			rc = pcistub_device_id_add(domain, bus, slot, func);
1081 		return rc;
1082 	}
1083 
1084 	if ((
1085 #if !defined(MODULE) /* pci_domains_supported is not being exported */ \
1086     || !defined(CONFIG_PCI_DOMAINS)
1087 	     !pci_domains_supported ? domain :
1088 #endif
1089 	     domain < 0 || domain > 0xffff)
1090 	    || bus < 0 || bus > 0xff
1091 	    || PCI_SLOT(devfn) != slot
1092 	    || PCI_FUNC(devfn) != func)
1093 		return -EINVAL;
1094 
1095 	pci_dev_id = kmalloc(sizeof(*pci_dev_id), GFP_KERNEL);
1096 	if (!pci_dev_id)
1097 		return -ENOMEM;
1098 
1099 	pr_debug("wants to seize %04x:%02x:%02x.%d\n",
1100 		 domain, bus, slot, func);
1101 
1102 	pcistub_device_id_add_list(pci_dev_id, domain, bus, devfn);
1103 
1104 	return 0;
1105 }
1106 
1107 static int pcistub_device_id_remove(int domain, int bus, int slot, int func)
1108 {
1109 	struct pcistub_device_id *pci_dev_id, *t;
1110 	int err = -ENOENT;
1111 	unsigned long flags;
1112 
1113 	spin_lock_irqsave(&device_ids_lock, flags);
1114 	list_for_each_entry_safe(pci_dev_id, t, &pcistub_device_ids,
1115 				 slot_list) {
1116 		if (pci_dev_id->domain == domain && pci_dev_id->bus == bus
1117 		    && (slot < 0 || PCI_SLOT(pci_dev_id->devfn) == slot)
1118 		    && (func < 0 || PCI_FUNC(pci_dev_id->devfn) == func)) {
1119 			/* Don't break; here because it's possible the same
1120 			 * slot could be in the list more than once
1121 			 */
1122 			list_del(&pci_dev_id->slot_list);
1123 			kfree(pci_dev_id);
1124 
1125 			err = 0;
1126 
1127 			pr_debug("removed %04x:%02x:%02x.%d from seize list\n",
1128 				 domain, bus, slot, func);
1129 		}
1130 	}
1131 	spin_unlock_irqrestore(&device_ids_lock, flags);
1132 
1133 	return err;
1134 }
1135 
1136 static int pcistub_reg_add(int domain, int bus, int slot, int func,
1137 			   unsigned int reg, unsigned int size,
1138 			   unsigned int mask)
1139 {
1140 	int err = 0;
1141 	struct pcistub_device *psdev;
1142 	struct pci_dev *dev;
1143 	struct config_field *field;
1144 
1145 	if (reg > 0xfff || (size < 4 && (mask >> (size * 8))))
1146 		return -EINVAL;
1147 
1148 	psdev = pcistub_device_find(domain, bus, slot, func);
1149 	if (!psdev) {
1150 		err = -ENODEV;
1151 		goto out;
1152 	}
1153 	dev = psdev->dev;
1154 
1155 	field = kzalloc(sizeof(*field), GFP_KERNEL);
1156 	if (!field) {
1157 		err = -ENOMEM;
1158 		goto out;
1159 	}
1160 
1161 	field->offset = reg;
1162 	field->size = size;
1163 	field->mask = mask;
1164 	field->init = NULL;
1165 	field->reset = NULL;
1166 	field->release = NULL;
1167 	field->clean = xen_pcibk_config_field_free;
1168 
1169 	err = xen_pcibk_config_quirks_add_field(dev, field);
1170 	if (err)
1171 		kfree(field);
1172 out:
1173 	if (psdev)
1174 		pcistub_device_put(psdev);
1175 	return err;
1176 }
1177 
1178 static ssize_t new_slot_store(struct device_driver *drv, const char *buf,
1179 			      size_t count)
1180 {
1181 	int domain, bus, slot, func;
1182 	int err;
1183 
1184 	err = str_to_slot(buf, &domain, &bus, &slot, &func);
1185 	if (err)
1186 		goto out;
1187 
1188 	err = pcistub_device_id_add(domain, bus, slot, func);
1189 
1190 out:
1191 	if (!err)
1192 		err = count;
1193 	return err;
1194 }
1195 static DRIVER_ATTR_WO(new_slot);
1196 
1197 static ssize_t remove_slot_store(struct device_driver *drv, const char *buf,
1198 				 size_t count)
1199 {
1200 	int domain, bus, slot, func;
1201 	int err;
1202 
1203 	err = str_to_slot(buf, &domain, &bus, &slot, &func);
1204 	if (err)
1205 		goto out;
1206 
1207 	err = pcistub_device_id_remove(domain, bus, slot, func);
1208 
1209 out:
1210 	if (!err)
1211 		err = count;
1212 	return err;
1213 }
1214 static DRIVER_ATTR_WO(remove_slot);
1215 
1216 static ssize_t slots_show(struct device_driver *drv, char *buf)
1217 {
1218 	struct pcistub_device_id *pci_dev_id;
1219 	size_t count = 0;
1220 	unsigned long flags;
1221 
1222 	spin_lock_irqsave(&device_ids_lock, flags);
1223 	list_for_each_entry(pci_dev_id, &pcistub_device_ids, slot_list) {
1224 		if (count >= PAGE_SIZE)
1225 			break;
1226 
1227 		count += scnprintf(buf + count, PAGE_SIZE - count,
1228 				   "%04x:%02x:%02x.%d\n",
1229 				   pci_dev_id->domain, pci_dev_id->bus,
1230 				   PCI_SLOT(pci_dev_id->devfn),
1231 				   PCI_FUNC(pci_dev_id->devfn));
1232 	}
1233 	spin_unlock_irqrestore(&device_ids_lock, flags);
1234 
1235 	return count;
1236 }
1237 static DRIVER_ATTR_RO(slots);
1238 
1239 static ssize_t irq_handlers_show(struct device_driver *drv, char *buf)
1240 {
1241 	struct pcistub_device *psdev;
1242 	struct xen_pcibk_dev_data *dev_data;
1243 	size_t count = 0;
1244 	unsigned long flags;
1245 
1246 	spin_lock_irqsave(&pcistub_devices_lock, flags);
1247 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
1248 		if (count >= PAGE_SIZE)
1249 			break;
1250 		if (!psdev->dev)
1251 			continue;
1252 		dev_data = pci_get_drvdata(psdev->dev);
1253 		if (!dev_data)
1254 			continue;
1255 		count +=
1256 		    scnprintf(buf + count, PAGE_SIZE - count,
1257 			      "%s:%s:%sing:%ld\n",
1258 			      pci_name(psdev->dev),
1259 			      dev_data->isr_on ? "on" : "off",
1260 			      dev_data->ack_intr ? "ack" : "not ack",
1261 			      dev_data->handled);
1262 	}
1263 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
1264 	return count;
1265 }
1266 static DRIVER_ATTR_RO(irq_handlers);
1267 
1268 static ssize_t irq_handler_state_store(struct device_driver *drv,
1269 				       const char *buf, size_t count)
1270 {
1271 	struct pcistub_device *psdev;
1272 	struct xen_pcibk_dev_data *dev_data;
1273 	int domain, bus, slot, func;
1274 	int err;
1275 
1276 	err = str_to_slot(buf, &domain, &bus, &slot, &func);
1277 	if (err)
1278 		return err;
1279 
1280 	psdev = pcistub_device_find(domain, bus, slot, func);
1281 	if (!psdev) {
1282 		err = -ENOENT;
1283 		goto out;
1284 	}
1285 
1286 	dev_data = pci_get_drvdata(psdev->dev);
1287 	if (!dev_data) {
1288 		err = -ENOENT;
1289 		goto out;
1290 	}
1291 
1292 	dev_dbg(&psdev->dev->dev, "%s fake irq handler: %d->%d\n",
1293 		dev_data->irq_name, dev_data->isr_on,
1294 		!dev_data->isr_on);
1295 
1296 	dev_data->isr_on = !(dev_data->isr_on);
1297 	if (dev_data->isr_on)
1298 		dev_data->ack_intr = 1;
1299 out:
1300 	if (psdev)
1301 		pcistub_device_put(psdev);
1302 	if (!err)
1303 		err = count;
1304 	return err;
1305 }
1306 static DRIVER_ATTR_WO(irq_handler_state);
1307 
1308 static ssize_t quirks_store(struct device_driver *drv, const char *buf,
1309 			    size_t count)
1310 {
1311 	int domain, bus, slot, func, reg, size, mask;
1312 	int err;
1313 
1314 	err = str_to_quirk(buf, &domain, &bus, &slot, &func, &reg, &size,
1315 			   &mask);
1316 	if (err)
1317 		goto out;
1318 
1319 	err = pcistub_reg_add(domain, bus, slot, func, reg, size, mask);
1320 
1321 out:
1322 	if (!err)
1323 		err = count;
1324 	return err;
1325 }
1326 
1327 static ssize_t quirks_show(struct device_driver *drv, char *buf)
1328 {
1329 	int count = 0;
1330 	unsigned long flags;
1331 	struct xen_pcibk_config_quirk *quirk;
1332 	struct xen_pcibk_dev_data *dev_data;
1333 	const struct config_field *field;
1334 	const struct config_field_entry *cfg_entry;
1335 
1336 	spin_lock_irqsave(&device_ids_lock, flags);
1337 	list_for_each_entry(quirk, &xen_pcibk_quirks, quirks_list) {
1338 		if (count >= PAGE_SIZE)
1339 			goto out;
1340 
1341 		count += scnprintf(buf + count, PAGE_SIZE - count,
1342 				   "%02x:%02x.%01x\n\t%04x:%04x:%04x:%04x\n",
1343 				   quirk->pdev->bus->number,
1344 				   PCI_SLOT(quirk->pdev->devfn),
1345 				   PCI_FUNC(quirk->pdev->devfn),
1346 				   quirk->devid.vendor, quirk->devid.device,
1347 				   quirk->devid.subvendor,
1348 				   quirk->devid.subdevice);
1349 
1350 		dev_data = pci_get_drvdata(quirk->pdev);
1351 
1352 		list_for_each_entry(cfg_entry, &dev_data->config_fields, list) {
1353 			field = cfg_entry->field;
1354 			if (count >= PAGE_SIZE)
1355 				goto out;
1356 
1357 			count += scnprintf(buf + count, PAGE_SIZE - count,
1358 					   "\t\t%08x:%01x:%08x\n",
1359 					   cfg_entry->base_offset +
1360 					   field->offset, field->size,
1361 					   field->mask);
1362 		}
1363 	}
1364 
1365 out:
1366 	spin_unlock_irqrestore(&device_ids_lock, flags);
1367 
1368 	return count;
1369 }
1370 static DRIVER_ATTR_RW(quirks);
1371 
1372 static ssize_t permissive_store(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_RW(permissive);
1435 
1436 static ssize_t allow_interrupt_control_store(struct device_driver *drv,
1437 					     const char *buf, size_t count)
1438 {
1439 	int domain, bus, slot, func;
1440 	int err;
1441 	struct pcistub_device *psdev;
1442 	struct xen_pcibk_dev_data *dev_data;
1443 
1444 	err = str_to_slot(buf, &domain, &bus, &slot, &func);
1445 	if (err)
1446 		goto out;
1447 
1448 	psdev = pcistub_device_find(domain, bus, slot, func);
1449 	if (!psdev) {
1450 		err = -ENODEV;
1451 		goto out;
1452 	}
1453 
1454 	dev_data = pci_get_drvdata(psdev->dev);
1455 	/* the driver data for a device should never be null at this point */
1456 	if (!dev_data) {
1457 		err = -ENXIO;
1458 		goto release;
1459 	}
1460 	dev_data->allow_interrupt_control = 1;
1461 release:
1462 	pcistub_device_put(psdev);
1463 out:
1464 	if (!err)
1465 		err = count;
1466 	return err;
1467 }
1468 
1469 static ssize_t allow_interrupt_control_show(struct device_driver *drv,
1470 					    char *buf)
1471 {
1472 	struct pcistub_device *psdev;
1473 	struct xen_pcibk_dev_data *dev_data;
1474 	size_t count = 0;
1475 	unsigned long flags;
1476 
1477 	spin_lock_irqsave(&pcistub_devices_lock, flags);
1478 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
1479 		if (count >= PAGE_SIZE)
1480 			break;
1481 		if (!psdev->dev)
1482 			continue;
1483 		dev_data = pci_get_drvdata(psdev->dev);
1484 		if (!dev_data || !dev_data->allow_interrupt_control)
1485 			continue;
1486 		count +=
1487 		    scnprintf(buf + count, PAGE_SIZE - count, "%s\n",
1488 			      pci_name(psdev->dev));
1489 	}
1490 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
1491 	return count;
1492 }
1493 static DRIVER_ATTR_RW(allow_interrupt_control);
1494 
1495 static void pcistub_exit(void)
1496 {
1497 	driver_remove_file(&xen_pcibk_pci_driver.driver, &driver_attr_new_slot);
1498 	driver_remove_file(&xen_pcibk_pci_driver.driver,
1499 			   &driver_attr_remove_slot);
1500 	driver_remove_file(&xen_pcibk_pci_driver.driver, &driver_attr_slots);
1501 	driver_remove_file(&xen_pcibk_pci_driver.driver, &driver_attr_quirks);
1502 	driver_remove_file(&xen_pcibk_pci_driver.driver,
1503 			   &driver_attr_permissive);
1504 	driver_remove_file(&xen_pcibk_pci_driver.driver,
1505 			   &driver_attr_allow_interrupt_control);
1506 	driver_remove_file(&xen_pcibk_pci_driver.driver,
1507 			   &driver_attr_irq_handlers);
1508 	driver_remove_file(&xen_pcibk_pci_driver.driver,
1509 			   &driver_attr_irq_handler_state);
1510 	pci_unregister_driver(&xen_pcibk_pci_driver);
1511 }
1512 
1513 static int __init pcistub_init(void)
1514 {
1515 	int pos = 0;
1516 	int err = 0;
1517 	int domain, bus, slot, func;
1518 	int parsed;
1519 
1520 	if (pci_devs_to_hide && *pci_devs_to_hide) {
1521 		do {
1522 			parsed = 0;
1523 
1524 			err = sscanf(pci_devs_to_hide + pos,
1525 				     " (%x:%x:%x.%x) %n",
1526 				     &domain, &bus, &slot, &func, &parsed);
1527 			switch (err) {
1528 			case 3:
1529 				func = -1;
1530 				sscanf(pci_devs_to_hide + pos,
1531 				       " (%x:%x:%x.*) %n",
1532 				       &domain, &bus, &slot, &parsed);
1533 				break;
1534 			case 2:
1535 				slot = func = -1;
1536 				sscanf(pci_devs_to_hide + pos,
1537 				       " (%x:%x:*.*) %n",
1538 				       &domain, &bus, &parsed);
1539 				break;
1540 			}
1541 
1542 			if (!parsed) {
1543 				domain = 0;
1544 				err = sscanf(pci_devs_to_hide + pos,
1545 					     " (%x:%x.%x) %n",
1546 					     &bus, &slot, &func, &parsed);
1547 				switch (err) {
1548 				case 2:
1549 					func = -1;
1550 					sscanf(pci_devs_to_hide + pos,
1551 					       " (%x:%x.*) %n",
1552 					       &bus, &slot, &parsed);
1553 					break;
1554 				case 1:
1555 					slot = func = -1;
1556 					sscanf(pci_devs_to_hide + pos,
1557 					       " (%x:*.*) %n",
1558 					       &bus, &parsed);
1559 					break;
1560 				}
1561 			}
1562 
1563 			if (parsed <= 0)
1564 				goto parse_error;
1565 
1566 			err = pcistub_device_id_add(domain, bus, slot, func);
1567 			if (err)
1568 				goto out;
1569 
1570 			pos += parsed;
1571 		} while (pci_devs_to_hide[pos]);
1572 	}
1573 
1574 	/* If we're the first PCI Device Driver to register, we're the
1575 	 * first one to get offered PCI devices as they become
1576 	 * available (and thus we can be the first to grab them)
1577 	 */
1578 	err = pci_register_driver(&xen_pcibk_pci_driver);
1579 	if (err < 0)
1580 		goto out;
1581 
1582 	err = driver_create_file(&xen_pcibk_pci_driver.driver,
1583 				 &driver_attr_new_slot);
1584 	if (!err)
1585 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1586 					 &driver_attr_remove_slot);
1587 	if (!err)
1588 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1589 					 &driver_attr_slots);
1590 	if (!err)
1591 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1592 					 &driver_attr_quirks);
1593 	if (!err)
1594 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1595 					 &driver_attr_permissive);
1596 	if (!err)
1597 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1598 					 &driver_attr_allow_interrupt_control);
1599 
1600 	if (!err)
1601 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1602 					 &driver_attr_irq_handlers);
1603 	if (!err)
1604 		err = driver_create_file(&xen_pcibk_pci_driver.driver,
1605 					&driver_attr_irq_handler_state);
1606 	if (err)
1607 		pcistub_exit();
1608 
1609 out:
1610 	return err;
1611 
1612 parse_error:
1613 	pr_err("Error parsing pci_devs_to_hide at \"%s\"\n",
1614 	       pci_devs_to_hide + pos);
1615 	return -EINVAL;
1616 }
1617 
1618 #ifndef MODULE
1619 /*
1620  * fs_initcall happens before device_initcall
1621  * so xen_pcibk *should* get called first (b/c we
1622  * want to suck up any device before other drivers
1623  * get a chance by being the first pci device
1624  * driver to register)
1625  */
1626 fs_initcall(pcistub_init);
1627 #endif
1628 
1629 #ifdef CONFIG_PCI_IOV
1630 static struct pcistub_device *find_vfs(const struct pci_dev *pdev)
1631 {
1632 	struct pcistub_device *psdev = NULL;
1633 	unsigned long flags;
1634 	bool found = false;
1635 
1636 	spin_lock_irqsave(&pcistub_devices_lock, flags);
1637 	list_for_each_entry(psdev, &pcistub_devices, dev_list) {
1638 		if (!psdev->pdev && psdev->dev != pdev
1639 		    && pci_physfn(psdev->dev) == pdev) {
1640 			found = true;
1641 			break;
1642 		}
1643 	}
1644 	spin_unlock_irqrestore(&pcistub_devices_lock, flags);
1645 	if (found)
1646 		return psdev;
1647 	return NULL;
1648 }
1649 
1650 static int pci_stub_notifier(struct notifier_block *nb,
1651 			     unsigned long action, void *data)
1652 {
1653 	struct device *dev = data;
1654 	const struct pci_dev *pdev = to_pci_dev(dev);
1655 
1656 	if (action != BUS_NOTIFY_UNBIND_DRIVER)
1657 		return NOTIFY_DONE;
1658 
1659 	if (!pdev->is_physfn)
1660 		return NOTIFY_DONE;
1661 
1662 	for (;;) {
1663 		struct pcistub_device *psdev = find_vfs(pdev);
1664 		if (!psdev)
1665 			break;
1666 		device_release_driver(&psdev->dev->dev);
1667 	}
1668 	return NOTIFY_DONE;
1669 }
1670 
1671 static struct notifier_block pci_stub_nb = {
1672 	.notifier_call = pci_stub_notifier,
1673 };
1674 #endif
1675 
1676 static int __init xen_pcibk_init(void)
1677 {
1678 	int err;
1679 
1680 	if (!xen_initial_domain())
1681 		return -ENODEV;
1682 
1683 	err = xen_pcibk_config_init();
1684 	if (err)
1685 		return err;
1686 
1687 #ifdef MODULE
1688 	err = pcistub_init();
1689 	if (err < 0)
1690 		return err;
1691 #endif
1692 
1693 	pcistub_init_devices_late();
1694 	err = xen_pcibk_xenbus_register();
1695 	if (err)
1696 		pcistub_exit();
1697 #ifdef CONFIG_PCI_IOV
1698 	else
1699 		bus_register_notifier(&pci_bus_type, &pci_stub_nb);
1700 #endif
1701 
1702 	return err;
1703 }
1704 
1705 static void __exit xen_pcibk_cleanup(void)
1706 {
1707 #ifdef CONFIG_PCI_IOV
1708 	bus_unregister_notifier(&pci_bus_type, &pci_stub_nb);
1709 #endif
1710 	xen_pcibk_xenbus_unregister();
1711 	pcistub_exit();
1712 }
1713 
1714 module_init(xen_pcibk_init);
1715 module_exit(xen_pcibk_cleanup);
1716 
1717 MODULE_LICENSE("Dual BSD/GPL");
1718 MODULE_ALIAS("xen-backend:pci");
1719