2 * PCI Stub Driver - Grabs devices in backend to be exported later
4 * Ryan Wilson <hap9@epoch.ncsc.mil>
5 * Chris Bookholt <hap10@epoch.ncsc.mil>
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
25 #include "conf_space.h"
26 #include "conf_space_quirks.h"
28 #define PCISTUB_DRIVER_NAME "pciback"
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
35 static DECLARE_RWSEM(pcistub_sem
);
36 module_param_named(hide
, pci_devs_to_hide
, charp
, 0444);
38 struct pcistub_device_id
{
39 struct list_head slot_list
;
44 static LIST_HEAD(pcistub_device_ids
);
45 static DEFINE_SPINLOCK(device_ids_lock
);
47 struct pcistub_device
{
49 struct list_head dev_list
;
53 struct xen_pcibk_device
*pdev
;/* non-NULL if struct pci_dev is in use */
56 /* Access to pcistub_devices & seized_devices lists and the initialize_devices
57 * flag must be locked with pcistub_devices_lock
59 static DEFINE_SPINLOCK(pcistub_devices_lock
);
60 static LIST_HEAD(pcistub_devices
);
62 /* wait for device_initcall before initializing our devices
63 * (see pcistub_init_devices_late)
65 static int initialize_devices
;
66 static LIST_HEAD(seized_devices
);
68 static struct pcistub_device
*pcistub_device_alloc(struct pci_dev
*dev
)
70 struct pcistub_device
*psdev
;
72 dev_dbg(&dev
->dev
, "pcistub_device_alloc\n");
74 psdev
= kzalloc(sizeof(*psdev
), GFP_KERNEL
);
78 psdev
->dev
= pci_dev_get(dev
);
84 kref_init(&psdev
->kref
);
85 spin_lock_init(&psdev
->lock
);
90 /* Don't call this directly as it's called by pcistub_device_put */
91 static void pcistub_device_release(struct kref
*kref
)
93 struct pcistub_device
*psdev
;
95 struct xen_pcibk_dev_data
*dev_data
;
97 psdev
= container_of(kref
, struct pcistub_device
, kref
);
99 dev_data
= pci_get_drvdata(dev
);
101 dev_dbg(&dev
->dev
, "pcistub_device_release\n");
103 xen_unregister_device_domain_owner(dev
);
105 /* Call the reset function which does not take lock as this
106 * is called from "unbind" which takes a device_lock mutex.
108 __pci_reset_function_locked(dev
);
110 pci_load_and_free_saved_state(dev
, &dev_data
->pci_saved_state
))
111 dev_info(&dev
->dev
, "Could not reload PCI state\n");
113 pci_restore_state(dev
);
116 struct physdev_pci_device ppdev
= {
117 .seg
= pci_domain_nr(dev
->bus
),
118 .bus
= dev
->bus
->number
,
121 int err
= HYPERVISOR_physdev_op(PHYSDEVOP_release_msix
,
124 if (err
&& err
!= -ENOSYS
)
125 dev_warn(&dev
->dev
, "MSI-X release failed (%d)\n",
129 /* Disable the device */
130 xen_pcibk_reset_device(dev
);
133 pci_set_drvdata(dev
, NULL
);
135 /* Clean-up the device */
136 xen_pcibk_config_free_dyn_fields(dev
);
137 xen_pcibk_config_free_dev(dev
);
139 pci_clear_dev_assigned(dev
);
145 static inline void pcistub_device_get(struct pcistub_device
*psdev
)
147 kref_get(&psdev
->kref
);
150 static inline void pcistub_device_put(struct pcistub_device
*psdev
)
152 kref_put(&psdev
->kref
, pcistub_device_release
);
155 static struct pcistub_device
*pcistub_device_find_locked(int domain
, int bus
,
158 struct pcistub_device
*psdev
;
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
)) {
173 static struct pcistub_device
*pcistub_device_find(int domain
, int bus
,
176 struct pcistub_device
*psdev
;
179 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
181 psdev
= pcistub_device_find_locked(domain
, bus
, slot
, func
);
183 pcistub_device_get(psdev
);
185 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
189 static struct pci_dev
*pcistub_device_get_pci_dev(struct xen_pcibk_device
*pdev
,
190 struct pcistub_device
*psdev
)
192 struct pci_dev
*pci_dev
= NULL
;
195 pcistub_device_get(psdev
);
197 spin_lock_irqsave(&psdev
->lock
, flags
);
200 pci_dev
= psdev
->dev
;
202 spin_unlock_irqrestore(&psdev
->lock
, flags
);
205 pcistub_device_put(psdev
);
210 struct pci_dev
*pcistub_get_pci_dev_by_slot(struct xen_pcibk_device
*pdev
,
214 struct pcistub_device
*psdev
;
215 struct pci_dev
*found_dev
= NULL
;
218 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
220 psdev
= pcistub_device_find_locked(domain
, bus
, slot
, func
);
222 found_dev
= pcistub_device_get_pci_dev(pdev
, psdev
);
224 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
228 struct pci_dev
*pcistub_get_pci_dev(struct xen_pcibk_device
*pdev
,
231 struct pcistub_device
*psdev
;
232 struct pci_dev
*found_dev
= NULL
;
235 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
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
);
244 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
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
255 * As such we have to be careful.
257 * To make this easier, the caller has to hold the device lock.
259 void pcistub_put_pci_dev(struct pci_dev
*dev
)
261 struct pcistub_device
*psdev
, *found_psdev
= NULL
;
263 struct xen_pcibk_dev_data
*dev_data
;
266 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
268 list_for_each_entry(psdev
, &pcistub_devices
, dev_list
) {
269 if (psdev
->dev
== dev
) {
275 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
276 if (WARN_ON(!found_psdev
))
279 /*hold this lock for avoiding breaking link between
280 * pcistub and xen_pcibk when AER is in processing
282 down_write(&pcistub_sem
);
283 /* Cleanup our device
284 * (so it's ready for the next domain)
286 device_lock_assert(&dev
->dev
);
287 __pci_reset_function_locked(dev
);
289 dev_data
= pci_get_drvdata(dev
);
290 ret
= pci_load_saved_state(dev
, dev_data
->pci_saved_state
);
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).
297 pci_restore_state(dev
);
299 dev_info(&dev
->dev
, "Could not reload PCI state\n");
300 /* This disables the device. */
301 xen_pcibk_reset_device(dev
);
303 /* And cleanup up our emulated fields. */
304 xen_pcibk_config_reset_dev(dev
);
305 xen_pcibk_config_free_dyn_fields(dev
);
307 xen_unregister_device_domain_owner(dev
);
309 spin_lock_irqsave(&found_psdev
->lock
, flags
);
310 found_psdev
->pdev
= NULL
;
311 spin_unlock_irqrestore(&found_psdev
->lock
, flags
);
313 pcistub_device_put(found_psdev
);
314 up_write(&pcistub_sem
);
317 static int pcistub_match_one(struct pci_dev
*dev
,
318 struct pcistub_device_id
*pdev_id
)
320 /* Match the specified device by domain, bus, slot, func and also if
321 * any of the device's parent bridges match.
323 for (; dev
!= NULL
; dev
= dev
->bus
->self
) {
324 if (pci_domain_nr(dev
->bus
) == pdev_id
->domain
325 && dev
->bus
->number
== pdev_id
->bus
326 && dev
->devfn
== pdev_id
->devfn
)
329 /* Sometimes topmost bridge links to itself. */
330 if (dev
== dev
->bus
->self
)
337 static int pcistub_match(struct pci_dev
*dev
)
339 struct pcistub_device_id
*pdev_id
;
343 spin_lock_irqsave(&device_ids_lock
, flags
);
344 list_for_each_entry(pdev_id
, &pcistub_device_ids
, slot_list
) {
345 if (pcistub_match_one(dev
, pdev_id
)) {
350 spin_unlock_irqrestore(&device_ids_lock
, flags
);
355 static int pcistub_init_device(struct pci_dev
*dev
)
357 struct xen_pcibk_dev_data
*dev_data
;
360 dev_dbg(&dev
->dev
, "initializing...\n");
362 /* The PCI backend is not intended to be a module (or to work with
363 * removable PCI devices (yet). If it were, xen_pcibk_config_free()
364 * would need to be called somewhere to free the memory allocated
365 * here and then to call kfree(pci_get_drvdata(psdev->dev)).
367 dev_data
= kzalloc(sizeof(*dev_data
) + strlen(DRV_NAME
"[]")
368 + strlen(pci_name(dev
)) + 1, GFP_KERNEL
);
373 pci_set_drvdata(dev
, dev_data
);
376 * Setup name for fake IRQ handler. It will only be enabled
377 * once the device is turned on by the guest.
379 sprintf(dev_data
->irq_name
, DRV_NAME
"[%s]", pci_name(dev
));
381 dev_dbg(&dev
->dev
, "initializing config\n");
383 init_waitqueue_head(&xen_pcibk_aer_wait_queue
);
384 err
= xen_pcibk_config_init_dev(dev
);
388 /* HACK: Force device (& ACPI) to determine what IRQ it's on - we
389 * must do this here because pcibios_enable_device may specify
390 * the pci device's true irq (and possibly its other resources)
391 * if they differ from what's in the configuration space.
392 * This makes the assumption that the device's resources won't
393 * change after this point (otherwise this code may break!)
395 dev_dbg(&dev
->dev
, "enabling device\n");
396 err
= pci_enable_device(dev
);
401 struct physdev_pci_device ppdev
= {
402 .seg
= pci_domain_nr(dev
->bus
),
403 .bus
= dev
->bus
->number
,
407 err
= HYPERVISOR_physdev_op(PHYSDEVOP_prepare_msix
, &ppdev
);
408 if (err
&& err
!= -ENOSYS
)
409 dev_err(&dev
->dev
, "MSI-X preparation failed (%d)\n",
413 /* We need the device active to save the state. */
414 dev_dbg(&dev
->dev
, "save state of device\n");
416 dev_data
->pci_saved_state
= pci_store_saved_state(dev
);
417 if (!dev_data
->pci_saved_state
)
418 dev_err(&dev
->dev
, "Could not store PCI conf saved state!\n");
420 dev_dbg(&dev
->dev
, "resetting (FLR, D3, etc) the device\n");
421 __pci_reset_function_locked(dev
);
422 pci_restore_state(dev
);
424 /* Now disable the device (this also ensures some private device
425 * data is setup before we export)
427 dev_dbg(&dev
->dev
, "reset device\n");
428 xen_pcibk_reset_device(dev
);
430 pci_set_dev_assigned(dev
);
434 xen_pcibk_config_free_dev(dev
);
437 pci_set_drvdata(dev
, NULL
);
443 * Because some initialization still happens on
444 * devices during fs_initcall, we need to defer
445 * full initialization of our devices until
448 static int __init
pcistub_init_devices_late(void)
450 struct pcistub_device
*psdev
;
454 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
456 while (!list_empty(&seized_devices
)) {
457 psdev
= container_of(seized_devices
.next
,
458 struct pcistub_device
, dev_list
);
459 list_del(&psdev
->dev_list
);
461 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
463 err
= pcistub_init_device(psdev
->dev
);
465 dev_err(&psdev
->dev
->dev
,
466 "error %d initializing device\n", err
);
471 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
474 list_add_tail(&psdev
->dev_list
, &pcistub_devices
);
477 initialize_devices
= 1;
479 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
484 static void pcistub_device_id_add_list(struct pcistub_device_id
*new,
485 int domain
, int bus
, unsigned int devfn
)
487 struct pcistub_device_id
*pci_dev_id
;
491 spin_lock_irqsave(&device_ids_lock
, flags
);
493 list_for_each_entry(pci_dev_id
, &pcistub_device_ids
, slot_list
) {
494 if (pci_dev_id
->domain
== domain
&& pci_dev_id
->bus
== bus
&&
495 pci_dev_id
->devfn
== devfn
) {
502 new->domain
= domain
;
505 list_add_tail(&new->slot_list
, &pcistub_device_ids
);
508 spin_unlock_irqrestore(&device_ids_lock
, flags
);
514 static int pcistub_seize(struct pci_dev
*dev
,
515 struct pcistub_device_id
*pci_dev_id
)
517 struct pcistub_device
*psdev
;
521 psdev
= pcistub_device_alloc(dev
);
527 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
529 if (initialize_devices
) {
530 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
532 /* don't want irqs disabled when calling pcistub_init_device */
533 err
= pcistub_init_device(psdev
->dev
);
535 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
538 list_add(&psdev
->dev_list
, &pcistub_devices
);
540 dev_dbg(&dev
->dev
, "deferring initialization\n");
541 list_add(&psdev
->dev_list
, &seized_devices
);
544 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
548 pcistub_device_put(psdev
);
549 } else if (pci_dev_id
)
550 pcistub_device_id_add_list(pci_dev_id
, pci_domain_nr(dev
->bus
),
551 dev
->bus
->number
, dev
->devfn
);
556 /* Called when 'bind'. This means we must _NOT_ call pci_reset_function or
557 * other functions that take the sysfs lock. */
558 static int pcistub_probe(struct pci_dev
*dev
, const struct pci_device_id
*id
)
561 struct pcistub_device_id
*pci_dev_id
= NULL
;
563 dev_dbg(&dev
->dev
, "probing...\n");
565 match
= pcistub_match(dev
);
567 if ((dev
->driver_override
&&
568 !strcmp(dev
->driver_override
, PCISTUB_DRIVER_NAME
)) ||
571 if (dev
->hdr_type
!= PCI_HEADER_TYPE_NORMAL
572 && dev
->hdr_type
!= PCI_HEADER_TYPE_BRIDGE
) {
573 dev_err(&dev
->dev
, "can't export pci devices that "
574 "don't have a normal (0) or bridge (1) "
581 pci_dev_id
= kmalloc(sizeof(*pci_dev_id
), GFP_KERNEL
);
588 dev_info(&dev
->dev
, "seizing device\n");
589 err
= pcistub_seize(dev
, pci_dev_id
);
591 /* Didn't find the device */
598 /* Called when 'unbind'. This means we must _NOT_ call pci_reset_function or
599 * other functions that take the sysfs lock. */
600 static void pcistub_remove(struct pci_dev
*dev
)
602 struct pcistub_device
*psdev
, *found_psdev
= NULL
;
605 dev_dbg(&dev
->dev
, "removing\n");
607 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
609 xen_pcibk_config_quirk_release(dev
);
611 list_for_each_entry(psdev
, &pcistub_devices
, dev_list
) {
612 if (psdev
->dev
== dev
) {
618 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
621 dev_dbg(&dev
->dev
, "found device to remove %s\n",
622 found_psdev
->pdev
? "- in-use" : "");
624 if (found_psdev
->pdev
) {
625 int domid
= xen_find_device_domain_owner(dev
);
627 pr_warn("****** removing device %s while still in-use by domain %d! ******\n",
628 pci_name(found_psdev
->dev
), domid
);
629 pr_warn("****** driver domain may still access this device's i/o resources!\n");
630 pr_warn("****** shutdown driver domain before binding device\n");
631 pr_warn("****** to other drivers or domains\n");
633 /* N.B. This ends up calling pcistub_put_pci_dev which ends up
635 xen_pcibk_release_pci_dev(found_psdev
->pdev
,
637 false /* caller holds the lock. */);
640 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
641 list_del(&found_psdev
->dev_list
);
642 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
644 /* the final put for releasing from the list */
645 pcistub_device_put(found_psdev
);
649 static const struct pci_device_id pcistub_ids
[] = {
651 .vendor
= PCI_ANY_ID
,
652 .device
= PCI_ANY_ID
,
653 .subvendor
= PCI_ANY_ID
,
654 .subdevice
= PCI_ANY_ID
,
659 #define PCI_NODENAME_MAX 40
660 static void kill_domain_by_device(struct pcistub_device
*psdev
)
662 struct xenbus_transaction xbt
;
664 char nodename
[PCI_NODENAME_MAX
];
667 snprintf(nodename
, PCI_NODENAME_MAX
, "/local/domain/0/backend/pci/%d/0",
668 psdev
->pdev
->xdev
->otherend_id
);
671 err
= xenbus_transaction_start(&xbt
);
673 dev_err(&psdev
->dev
->dev
,
674 "error %d when start xenbus transaction\n", err
);
677 /*PV AER handlers will set this flag*/
678 xenbus_printf(xbt
, nodename
, "aerState" , "aerfail");
679 err
= xenbus_transaction_end(xbt
, 0);
683 dev_err(&psdev
->dev
->dev
,
684 "error %d when end xenbus transaction\n", err
);
689 /* For each aer recovery step error_detected, mmio_enabled, etc, front_end and
690 * backend need to have cooperation. In xen_pcibk, those steps will do similar
691 * jobs: send service request and waiting for front_end response.
693 static pci_ers_result_t
common_process(struct pcistub_device
*psdev
,
694 pci_channel_state_t state
, int aer_cmd
,
695 pci_ers_result_t result
)
697 pci_ers_result_t res
= result
;
698 struct xen_pcie_aer_op
*aer_op
;
699 struct xen_pcibk_device
*pdev
= psdev
->pdev
;
700 struct xen_pci_sharedinfo
*sh_info
= pdev
->sh_info
;
703 /*with PV AER drivers*/
704 aer_op
= &(sh_info
->aer_op
);
705 aer_op
->cmd
= aer_cmd
;
706 /*useful for error_detected callback*/
709 ret
= xen_pcibk_get_pcifront_dev(psdev
->dev
, psdev
->pdev
,
710 &aer_op
->domain
, &aer_op
->bus
, &aer_op
->devfn
);
712 dev_err(&psdev
->dev
->dev
,
713 DRV_NAME
": failed to get pcifront device\n");
714 return PCI_ERS_RESULT_NONE
;
718 dev_dbg(&psdev
->dev
->dev
,
719 DRV_NAME
": aer_op %x dom %x bus %x devfn %x\n",
720 aer_cmd
, aer_op
->domain
, aer_op
->bus
, aer_op
->devfn
);
721 /*local flag to mark there's aer request, xen_pcibk callback will use
722 * this flag to judge whether we need to check pci-front give aer
725 set_bit(_PCIB_op_pending
, (unsigned long *)&pdev
->flags
);
727 /*It is possible that a pcifront conf_read_write ops request invokes
728 * the callback which cause the spurious execution of wake_up.
729 * Yet it is harmless and better than a spinlock here
731 set_bit(_XEN_PCIB_active
,
732 (unsigned long *)&sh_info
->flags
);
734 notify_remote_via_irq(pdev
->evtchn_irq
);
736 ret
= wait_event_timeout(xen_pcibk_aer_wait_queue
,
737 !(test_bit(_XEN_PCIB_active
, (unsigned long *)
738 &sh_info
->flags
)), 300*HZ
);
741 if (test_bit(_XEN_PCIB_active
,
742 (unsigned long *)&sh_info
->flags
)) {
743 dev_err(&psdev
->dev
->dev
,
744 "pcifront aer process not responding!\n");
745 clear_bit(_XEN_PCIB_active
,
746 (unsigned long *)&sh_info
->flags
);
747 aer_op
->err
= PCI_ERS_RESULT_NONE
;
751 clear_bit(_PCIB_op_pending
, (unsigned long *)&pdev
->flags
);
753 if (test_bit(_XEN_PCIF_active
,
754 (unsigned long *)&sh_info
->flags
)) {
755 dev_dbg(&psdev
->dev
->dev
,
756 "schedule pci_conf service in " DRV_NAME
"\n");
757 xen_pcibk_test_and_schedule_op(psdev
->pdev
);
760 res
= (pci_ers_result_t
)aer_op
->err
;
765 * xen_pcibk_slot_reset: it will send the slot_reset request to pcifront in case
766 * of the device driver could provide this service, and then wait for pcifront
768 * @dev: pointer to PCI devices
769 * return value is used by aer_core do_recovery policy
771 static pci_ers_result_t
xen_pcibk_slot_reset(struct pci_dev
*dev
)
773 struct pcistub_device
*psdev
;
774 pci_ers_result_t result
;
776 result
= PCI_ERS_RESULT_RECOVERED
;
777 dev_dbg(&dev
->dev
, "xen_pcibk_slot_reset(bus:%x,devfn:%x)\n",
778 dev
->bus
->number
, dev
->devfn
);
780 down_write(&pcistub_sem
);
781 psdev
= pcistub_device_find(pci_domain_nr(dev
->bus
),
783 PCI_SLOT(dev
->devfn
),
784 PCI_FUNC(dev
->devfn
));
786 if (!psdev
|| !psdev
->pdev
) {
788 DRV_NAME
" device is not found/assigned\n");
792 if (!psdev
->pdev
->sh_info
) {
793 dev_err(&dev
->dev
, DRV_NAME
" device is not connected or owned"
794 " by HVM, kill it\n");
795 kill_domain_by_device(psdev
);
799 if (!test_bit(_XEN_PCIB_AERHANDLER
,
800 (unsigned long *)&psdev
->pdev
->sh_info
->flags
)) {
802 "guest with no AER driver should have been killed\n");
805 result
= common_process(psdev
, 1, XEN_PCI_OP_aer_slotreset
, result
);
807 if (result
== PCI_ERS_RESULT_NONE
||
808 result
== PCI_ERS_RESULT_DISCONNECT
) {
810 "No AER slot_reset service or disconnected!\n");
811 kill_domain_by_device(psdev
);
815 pcistub_device_put(psdev
);
816 up_write(&pcistub_sem
);
822 /*xen_pcibk_mmio_enabled: it will send the mmio_enabled request to pcifront
823 * in case of the device driver could provide this service, and then wait
825 * @dev: pointer to PCI devices
826 * return value is used by aer_core do_recovery policy
829 static pci_ers_result_t
xen_pcibk_mmio_enabled(struct pci_dev
*dev
)
831 struct pcistub_device
*psdev
;
832 pci_ers_result_t result
;
834 result
= PCI_ERS_RESULT_RECOVERED
;
835 dev_dbg(&dev
->dev
, "xen_pcibk_mmio_enabled(bus:%x,devfn:%x)\n",
836 dev
->bus
->number
, dev
->devfn
);
838 down_write(&pcistub_sem
);
839 psdev
= pcistub_device_find(pci_domain_nr(dev
->bus
),
841 PCI_SLOT(dev
->devfn
),
842 PCI_FUNC(dev
->devfn
));
844 if (!psdev
|| !psdev
->pdev
) {
846 DRV_NAME
" device is not found/assigned\n");
850 if (!psdev
->pdev
->sh_info
) {
851 dev_err(&dev
->dev
, DRV_NAME
" device is not connected or owned"
852 " by HVM, kill it\n");
853 kill_domain_by_device(psdev
);
857 if (!test_bit(_XEN_PCIB_AERHANDLER
,
858 (unsigned long *)&psdev
->pdev
->sh_info
->flags
)) {
860 "guest with no AER driver should have been killed\n");
863 result
= common_process(psdev
, 1, XEN_PCI_OP_aer_mmio
, result
);
865 if (result
== PCI_ERS_RESULT_NONE
||
866 result
== PCI_ERS_RESULT_DISCONNECT
) {
868 "No AER mmio_enabled service or disconnected!\n");
869 kill_domain_by_device(psdev
);
873 pcistub_device_put(psdev
);
874 up_write(&pcistub_sem
);
878 /*xen_pcibk_error_detected: it will send the error_detected request to pcifront
879 * in case of the device driver could provide this service, and then wait
881 * @dev: pointer to PCI devices
882 * @error: the current PCI connection state
883 * return value is used by aer_core do_recovery policy
886 static pci_ers_result_t
xen_pcibk_error_detected(struct pci_dev
*dev
,
887 pci_channel_state_t error
)
889 struct pcistub_device
*psdev
;
890 pci_ers_result_t result
;
892 result
= PCI_ERS_RESULT_CAN_RECOVER
;
893 dev_dbg(&dev
->dev
, "xen_pcibk_error_detected(bus:%x,devfn:%x)\n",
894 dev
->bus
->number
, dev
->devfn
);
896 down_write(&pcistub_sem
);
897 psdev
= pcistub_device_find(pci_domain_nr(dev
->bus
),
899 PCI_SLOT(dev
->devfn
),
900 PCI_FUNC(dev
->devfn
));
902 if (!psdev
|| !psdev
->pdev
) {
904 DRV_NAME
" device is not found/assigned\n");
908 if (!psdev
->pdev
->sh_info
) {
909 dev_err(&dev
->dev
, DRV_NAME
" device is not connected or owned"
910 " by HVM, kill it\n");
911 kill_domain_by_device(psdev
);
915 /*Guest owns the device yet no aer handler regiested, kill guest*/
916 if (!test_bit(_XEN_PCIB_AERHANDLER
,
917 (unsigned long *)&psdev
->pdev
->sh_info
->flags
)) {
918 dev_dbg(&dev
->dev
, "guest may have no aer driver, kill it\n");
919 kill_domain_by_device(psdev
);
922 result
= common_process(psdev
, error
, XEN_PCI_OP_aer_detected
, result
);
924 if (result
== PCI_ERS_RESULT_NONE
||
925 result
== PCI_ERS_RESULT_DISCONNECT
) {
927 "No AER error_detected service or disconnected!\n");
928 kill_domain_by_device(psdev
);
932 pcistub_device_put(psdev
);
933 up_write(&pcistub_sem
);
937 /*xen_pcibk_error_resume: it will send the error_resume request to pcifront
938 * in case of the device driver could provide this service, and then wait
940 * @dev: pointer to PCI devices
943 static void xen_pcibk_error_resume(struct pci_dev
*dev
)
945 struct pcistub_device
*psdev
;
947 dev_dbg(&dev
->dev
, "xen_pcibk_error_resume(bus:%x,devfn:%x)\n",
948 dev
->bus
->number
, dev
->devfn
);
950 down_write(&pcistub_sem
);
951 psdev
= pcistub_device_find(pci_domain_nr(dev
->bus
),
953 PCI_SLOT(dev
->devfn
),
954 PCI_FUNC(dev
->devfn
));
956 if (!psdev
|| !psdev
->pdev
) {
958 DRV_NAME
" device is not found/assigned\n");
962 if (!psdev
->pdev
->sh_info
) {
963 dev_err(&dev
->dev
, DRV_NAME
" device is not connected or owned"
964 " by HVM, kill it\n");
965 kill_domain_by_device(psdev
);
969 if (!test_bit(_XEN_PCIB_AERHANDLER
,
970 (unsigned long *)&psdev
->pdev
->sh_info
->flags
)) {
972 "guest with no AER driver should have been killed\n");
973 kill_domain_by_device(psdev
);
976 common_process(psdev
, 1, XEN_PCI_OP_aer_resume
,
977 PCI_ERS_RESULT_RECOVERED
);
980 pcistub_device_put(psdev
);
981 up_write(&pcistub_sem
);
985 /*add xen_pcibk AER handling*/
986 static const struct pci_error_handlers xen_pcibk_error_handler
= {
987 .error_detected
= xen_pcibk_error_detected
,
988 .mmio_enabled
= xen_pcibk_mmio_enabled
,
989 .slot_reset
= xen_pcibk_slot_reset
,
990 .resume
= xen_pcibk_error_resume
,
994 * Note: There is no MODULE_DEVICE_TABLE entry here because this isn't
995 * for a normal device. I don't want it to be loaded automatically.
998 static struct pci_driver xen_pcibk_pci_driver
= {
999 /* The name should be xen_pciback, but until the tools are updated
1000 * we will keep it as pciback. */
1001 .name
= PCISTUB_DRIVER_NAME
,
1002 .id_table
= pcistub_ids
,
1003 .probe
= pcistub_probe
,
1004 .remove
= pcistub_remove
,
1005 .err_handler
= &xen_pcibk_error_handler
,
1008 static inline int str_to_slot(const char *buf
, int *domain
, int *bus
,
1009 int *slot
, int *func
)
1013 switch (sscanf(buf
, " %x:%x:%x.%x %n", domain
, bus
, slot
, func
,
1017 sscanf(buf
, " %x:%x:%x.* %n", domain
, bus
, slot
, &parsed
);
1021 sscanf(buf
, " %x:%x:*.* %n", domain
, bus
, &parsed
);
1024 if (parsed
&& !buf
[parsed
])
1027 /* try again without domain */
1029 switch (sscanf(buf
, " %x:%x.%x %n", bus
, slot
, func
, &parsed
)) {
1032 sscanf(buf
, " %x:%x.* %n", bus
, slot
, &parsed
);
1036 sscanf(buf
, " %x:*.* %n", bus
, &parsed
);
1039 if (parsed
&& !buf
[parsed
])
1045 static inline int str_to_quirk(const char *buf
, int *domain
, int *bus
, int
1046 *slot
, int *func
, int *reg
, int *size
, int *mask
)
1050 sscanf(buf
, " %x:%x:%x.%x-%x:%x:%x %n", domain
, bus
, slot
, func
,
1051 reg
, size
, mask
, &parsed
);
1052 if (parsed
&& !buf
[parsed
])
1055 /* try again without domain */
1057 sscanf(buf
, " %x:%x.%x-%x:%x:%x %n", bus
, slot
, func
, reg
, size
,
1059 if (parsed
&& !buf
[parsed
])
1065 static int pcistub_device_id_add(int domain
, int bus
, int slot
, int func
)
1067 struct pcistub_device_id
*pci_dev_id
;
1068 int rc
= 0, devfn
= PCI_DEVFN(slot
, func
);
1071 for (slot
= 0; !rc
&& slot
< 32; ++slot
)
1072 rc
= pcistub_device_id_add(domain
, bus
, slot
, func
);
1077 for (func
= 0; !rc
&& func
< 8; ++func
)
1078 rc
= pcistub_device_id_add(domain
, bus
, slot
, func
);
1083 #if !defined(MODULE) /* pci_domains_supported is not being exported */ \
1084 || !defined(CONFIG_PCI_DOMAINS)
1085 !pci_domains_supported
? domain
:
1087 domain
< 0 || domain
> 0xffff)
1088 || bus
< 0 || bus
> 0xff
1089 || PCI_SLOT(devfn
) != slot
1090 || PCI_FUNC(devfn
) != func
)
1093 pci_dev_id
= kmalloc(sizeof(*pci_dev_id
), GFP_KERNEL
);
1097 pr_debug("wants to seize %04x:%02x:%02x.%d\n",
1098 domain
, bus
, slot
, func
);
1100 pcistub_device_id_add_list(pci_dev_id
, domain
, bus
, devfn
);
1105 static int pcistub_device_id_remove(int domain
, int bus
, int slot
, int func
)
1107 struct pcistub_device_id
*pci_dev_id
, *t
;
1109 unsigned long flags
;
1111 spin_lock_irqsave(&device_ids_lock
, flags
);
1112 list_for_each_entry_safe(pci_dev_id
, t
, &pcistub_device_ids
,
1114 if (pci_dev_id
->domain
== domain
&& pci_dev_id
->bus
== bus
1115 && (slot
< 0 || PCI_SLOT(pci_dev_id
->devfn
) == slot
)
1116 && (func
< 0 || PCI_FUNC(pci_dev_id
->devfn
) == func
)) {
1117 /* Don't break; here because it's possible the same
1118 * slot could be in the list more than once
1120 list_del(&pci_dev_id
->slot_list
);
1125 pr_debug("removed %04x:%02x:%02x.%d from seize list\n",
1126 domain
, bus
, slot
, func
);
1129 spin_unlock_irqrestore(&device_ids_lock
, flags
);
1134 static int pcistub_reg_add(int domain
, int bus
, int slot
, int func
,
1135 unsigned int reg
, unsigned int size
,
1139 struct pcistub_device
*psdev
;
1140 struct pci_dev
*dev
;
1141 struct config_field
*field
;
1143 if (reg
> 0xfff || (size
< 4 && (mask
>> (size
* 8))))
1146 psdev
= pcistub_device_find(domain
, bus
, slot
, func
);
1153 field
= kzalloc(sizeof(*field
), GFP_KERNEL
);
1159 field
->offset
= reg
;
1163 field
->reset
= NULL
;
1164 field
->release
= NULL
;
1165 field
->clean
= xen_pcibk_config_field_free
;
1167 err
= xen_pcibk_config_quirks_add_field(dev
, field
);
1172 pcistub_device_put(psdev
);
1176 static ssize_t
new_slot_store(struct device_driver
*drv
, const char *buf
,
1179 int domain
, bus
, slot
, func
;
1182 err
= str_to_slot(buf
, &domain
, &bus
, &slot
, &func
);
1186 err
= pcistub_device_id_add(domain
, bus
, slot
, func
);
1193 static DRIVER_ATTR_WO(new_slot
);
1195 static ssize_t
remove_slot_store(struct device_driver
*drv
, const char *buf
,
1198 int domain
, bus
, slot
, func
;
1201 err
= str_to_slot(buf
, &domain
, &bus
, &slot
, &func
);
1205 err
= pcistub_device_id_remove(domain
, bus
, slot
, func
);
1212 static DRIVER_ATTR_WO(remove_slot
);
1214 static ssize_t
slots_show(struct device_driver
*drv
, char *buf
)
1216 struct pcistub_device_id
*pci_dev_id
;
1218 unsigned long flags
;
1220 spin_lock_irqsave(&device_ids_lock
, flags
);
1221 list_for_each_entry(pci_dev_id
, &pcistub_device_ids
, slot_list
) {
1222 if (count
>= PAGE_SIZE
)
1225 count
+= scnprintf(buf
+ count
, PAGE_SIZE
- count
,
1226 "%04x:%02x:%02x.%d\n",
1227 pci_dev_id
->domain
, pci_dev_id
->bus
,
1228 PCI_SLOT(pci_dev_id
->devfn
),
1229 PCI_FUNC(pci_dev_id
->devfn
));
1231 spin_unlock_irqrestore(&device_ids_lock
, flags
);
1235 static DRIVER_ATTR_RO(slots
);
1237 static ssize_t
irq_handlers_show(struct device_driver
*drv
, char *buf
)
1239 struct pcistub_device
*psdev
;
1240 struct xen_pcibk_dev_data
*dev_data
;
1242 unsigned long flags
;
1244 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
1245 list_for_each_entry(psdev
, &pcistub_devices
, dev_list
) {
1246 if (count
>= PAGE_SIZE
)
1250 dev_data
= pci_get_drvdata(psdev
->dev
);
1254 scnprintf(buf
+ count
, PAGE_SIZE
- count
,
1255 "%s:%s:%sing:%ld\n",
1256 pci_name(psdev
->dev
),
1257 dev_data
->isr_on
? "on" : "off",
1258 dev_data
->ack_intr
? "ack" : "not ack",
1261 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
1264 static DRIVER_ATTR_RO(irq_handlers
);
1266 static ssize_t
irq_handler_state_store(struct device_driver
*drv
,
1267 const char *buf
, size_t count
)
1269 struct pcistub_device
*psdev
;
1270 struct xen_pcibk_dev_data
*dev_data
;
1271 int domain
, bus
, slot
, func
;
1274 err
= str_to_slot(buf
, &domain
, &bus
, &slot
, &func
);
1278 psdev
= pcistub_device_find(domain
, bus
, slot
, func
);
1284 dev_data
= pci_get_drvdata(psdev
->dev
);
1290 dev_dbg(&psdev
->dev
->dev
, "%s fake irq handler: %d->%d\n",
1291 dev_data
->irq_name
, dev_data
->isr_on
,
1294 dev_data
->isr_on
= !(dev_data
->isr_on
);
1295 if (dev_data
->isr_on
)
1296 dev_data
->ack_intr
= 1;
1299 pcistub_device_put(psdev
);
1304 static DRIVER_ATTR_WO(irq_handler_state
);
1306 static ssize_t
quirks_store(struct device_driver
*drv
, const char *buf
,
1309 int domain
, bus
, slot
, func
, reg
, size
, mask
;
1312 err
= str_to_quirk(buf
, &domain
, &bus
, &slot
, &func
, ®
, &size
,
1317 err
= pcistub_reg_add(domain
, bus
, slot
, func
, reg
, size
, mask
);
1325 static ssize_t
quirks_show(struct device_driver
*drv
, char *buf
)
1328 unsigned long flags
;
1329 struct xen_pcibk_config_quirk
*quirk
;
1330 struct xen_pcibk_dev_data
*dev_data
;
1331 const struct config_field
*field
;
1332 const struct config_field_entry
*cfg_entry
;
1334 spin_lock_irqsave(&device_ids_lock
, flags
);
1335 list_for_each_entry(quirk
, &xen_pcibk_quirks
, quirks_list
) {
1336 if (count
>= PAGE_SIZE
)
1339 count
+= scnprintf(buf
+ count
, PAGE_SIZE
- count
,
1340 "%02x:%02x.%01x\n\t%04x:%04x:%04x:%04x\n",
1341 quirk
->pdev
->bus
->number
,
1342 PCI_SLOT(quirk
->pdev
->devfn
),
1343 PCI_FUNC(quirk
->pdev
->devfn
),
1344 quirk
->devid
.vendor
, quirk
->devid
.device
,
1345 quirk
->devid
.subvendor
,
1346 quirk
->devid
.subdevice
);
1348 dev_data
= pci_get_drvdata(quirk
->pdev
);
1350 list_for_each_entry(cfg_entry
, &dev_data
->config_fields
, list
) {
1351 field
= cfg_entry
->field
;
1352 if (count
>= PAGE_SIZE
)
1355 count
+= scnprintf(buf
+ count
, PAGE_SIZE
- count
,
1356 "\t\t%08x:%01x:%08x\n",
1357 cfg_entry
->base_offset
+
1358 field
->offset
, field
->size
,
1364 spin_unlock_irqrestore(&device_ids_lock
, flags
);
1368 static DRIVER_ATTR_RW(quirks
);
1370 static ssize_t
permissive_store(struct device_driver
*drv
, const char *buf
,
1373 int domain
, bus
, slot
, func
;
1375 struct pcistub_device
*psdev
;
1376 struct xen_pcibk_dev_data
*dev_data
;
1378 err
= str_to_slot(buf
, &domain
, &bus
, &slot
, &func
);
1382 psdev
= pcistub_device_find(domain
, bus
, slot
, func
);
1388 dev_data
= pci_get_drvdata(psdev
->dev
);
1389 /* the driver data for a device should never be null at this point */
1394 if (!dev_data
->permissive
) {
1395 dev_data
->permissive
= 1;
1396 /* Let user know that what they're doing could be unsafe */
1397 dev_warn(&psdev
->dev
->dev
, "enabling permissive mode "
1398 "configuration space accesses!\n");
1399 dev_warn(&psdev
->dev
->dev
,
1400 "permissive mode is potentially unsafe!\n");
1403 pcistub_device_put(psdev
);
1410 static ssize_t
permissive_show(struct device_driver
*drv
, char *buf
)
1412 struct pcistub_device
*psdev
;
1413 struct xen_pcibk_dev_data
*dev_data
;
1415 unsigned long flags
;
1416 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
1417 list_for_each_entry(psdev
, &pcistub_devices
, dev_list
) {
1418 if (count
>= PAGE_SIZE
)
1422 dev_data
= pci_get_drvdata(psdev
->dev
);
1423 if (!dev_data
|| !dev_data
->permissive
)
1426 scnprintf(buf
+ count
, PAGE_SIZE
- count
, "%s\n",
1427 pci_name(psdev
->dev
));
1429 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
1432 static DRIVER_ATTR_RW(permissive
);
1434 static void pcistub_exit(void)
1436 driver_remove_file(&xen_pcibk_pci_driver
.driver
, &driver_attr_new_slot
);
1437 driver_remove_file(&xen_pcibk_pci_driver
.driver
,
1438 &driver_attr_remove_slot
);
1439 driver_remove_file(&xen_pcibk_pci_driver
.driver
, &driver_attr_slots
);
1440 driver_remove_file(&xen_pcibk_pci_driver
.driver
, &driver_attr_quirks
);
1441 driver_remove_file(&xen_pcibk_pci_driver
.driver
,
1442 &driver_attr_permissive
);
1443 driver_remove_file(&xen_pcibk_pci_driver
.driver
,
1444 &driver_attr_irq_handlers
);
1445 driver_remove_file(&xen_pcibk_pci_driver
.driver
,
1446 &driver_attr_irq_handler_state
);
1447 pci_unregister_driver(&xen_pcibk_pci_driver
);
1450 static int __init
pcistub_init(void)
1454 int domain
, bus
, slot
, func
;
1457 if (pci_devs_to_hide
&& *pci_devs_to_hide
) {
1461 err
= sscanf(pci_devs_to_hide
+ pos
,
1462 " (%x:%x:%x.%x) %n",
1463 &domain
, &bus
, &slot
, &func
, &parsed
);
1467 sscanf(pci_devs_to_hide
+ pos
,
1469 &domain
, &bus
, &slot
, &parsed
);
1473 sscanf(pci_devs_to_hide
+ pos
,
1475 &domain
, &bus
, &parsed
);
1481 err
= sscanf(pci_devs_to_hide
+ pos
,
1483 &bus
, &slot
, &func
, &parsed
);
1487 sscanf(pci_devs_to_hide
+ pos
,
1489 &bus
, &slot
, &parsed
);
1493 sscanf(pci_devs_to_hide
+ pos
,
1503 err
= pcistub_device_id_add(domain
, bus
, slot
, func
);
1508 } while (pci_devs_to_hide
[pos
]);
1511 /* If we're the first PCI Device Driver to register, we're the
1512 * first one to get offered PCI devices as they become
1513 * available (and thus we can be the first to grab them)
1515 err
= pci_register_driver(&xen_pcibk_pci_driver
);
1519 err
= driver_create_file(&xen_pcibk_pci_driver
.driver
,
1520 &driver_attr_new_slot
);
1522 err
= driver_create_file(&xen_pcibk_pci_driver
.driver
,
1523 &driver_attr_remove_slot
);
1525 err
= driver_create_file(&xen_pcibk_pci_driver
.driver
,
1526 &driver_attr_slots
);
1528 err
= driver_create_file(&xen_pcibk_pci_driver
.driver
,
1529 &driver_attr_quirks
);
1531 err
= driver_create_file(&xen_pcibk_pci_driver
.driver
,
1532 &driver_attr_permissive
);
1535 err
= driver_create_file(&xen_pcibk_pci_driver
.driver
,
1536 &driver_attr_irq_handlers
);
1538 err
= driver_create_file(&xen_pcibk_pci_driver
.driver
,
1539 &driver_attr_irq_handler_state
);
1547 pr_err("Error parsing pci_devs_to_hide at \"%s\"\n",
1548 pci_devs_to_hide
+ pos
);
1554 * fs_initcall happens before device_initcall
1555 * so xen_pcibk *should* get called first (b/c we
1556 * want to suck up any device before other drivers
1557 * get a chance by being the first pci device
1558 * driver to register)
1560 fs_initcall(pcistub_init
);
1563 #ifdef CONFIG_PCI_IOV
1564 static struct pcistub_device
*find_vfs(const struct pci_dev
*pdev
)
1566 struct pcistub_device
*psdev
= NULL
;
1567 unsigned long flags
;
1570 spin_lock_irqsave(&pcistub_devices_lock
, flags
);
1571 list_for_each_entry(psdev
, &pcistub_devices
, dev_list
) {
1572 if (!psdev
->pdev
&& psdev
->dev
!= pdev
1573 && pci_physfn(psdev
->dev
) == pdev
) {
1578 spin_unlock_irqrestore(&pcistub_devices_lock
, flags
);
1584 static int pci_stub_notifier(struct notifier_block
*nb
,
1585 unsigned long action
, void *data
)
1587 struct device
*dev
= data
;
1588 const struct pci_dev
*pdev
= to_pci_dev(dev
);
1590 if (action
!= BUS_NOTIFY_UNBIND_DRIVER
)
1593 if (!pdev
->is_physfn
)
1597 struct pcistub_device
*psdev
= find_vfs(pdev
);
1600 device_release_driver(&psdev
->dev
->dev
);
1605 static struct notifier_block pci_stub_nb
= {
1606 .notifier_call
= pci_stub_notifier
,
1610 static int __init
xen_pcibk_init(void)
1614 if (!xen_initial_domain())
1617 err
= xen_pcibk_config_init();
1622 err
= pcistub_init();
1627 pcistub_init_devices_late();
1628 err
= xen_pcibk_xenbus_register();
1631 #ifdef CONFIG_PCI_IOV
1633 bus_register_notifier(&pci_bus_type
, &pci_stub_nb
);
1639 static void __exit
xen_pcibk_cleanup(void)
1641 #ifdef CONFIG_PCI_IOV
1642 bus_unregister_notifier(&pci_bus_type
, &pci_stub_nb
);
1644 xen_pcibk_xenbus_unregister();
1648 module_init(xen_pcibk_init
);
1649 module_exit(xen_pcibk_cleanup
);
1651 MODULE_LICENSE("Dual BSD/GPL");
1652 MODULE_ALIAS("xen-backend:pci");