2 * PCI Backend Xenbus Setup - handles setup with frontend and xend
4 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
6 #include <linux/module.h>
7 #include <linux/init.h>
8 #include <linux/list.h>
9 #include <linux/vmalloc.h>
10 #include <linux/workqueue.h>
11 #include <xen/xenbus.h>
12 #include <xen/events.h>
13 #include <asm/xen/pci.h>
16 #define INVALID_EVTCHN_IRQ (-1)
17 struct workqueue_struct
*xen_pcibk_wq
;
19 static bool __read_mostly passthrough
;
20 module_param(passthrough
, bool, S_IRUGO
);
21 MODULE_PARM_DESC(passthrough
,
22 "Option to specify how to export PCI topology to guest:\n"\
23 " 0 - (default) Hide the true PCI topology and makes the frontend\n"\
24 " there is a single PCI bus with only the exported devices on it.\n"\
25 " For example, a device at 03:05.0 will be re-assigned to 00:00.0\n"\
26 " while second device at 02:1a.1 will be re-assigned to 00:01.1.\n"\
27 " 1 - Passthrough provides a real view of the PCI topology to the\n"\
28 " frontend (for example, a device at 06:01.b will still appear at\n"\
29 " 06:01.b to the frontend). This is similar to how Xen 2.0.x\n"\
30 " exposed PCI devices to its driver domains. This may be required\n"\
31 " for drivers which depend on finding their hardward in certain\n"\
32 " bus/slot locations.");
34 static struct xen_pcibk_device
*alloc_pdev(struct xenbus_device
*xdev
)
36 struct xen_pcibk_device
*pdev
;
38 pdev
= kzalloc(sizeof(struct xen_pcibk_device
), GFP_KERNEL
);
41 dev_dbg(&xdev
->dev
, "allocated pdev @ 0x%p\n", pdev
);
44 dev_set_drvdata(&xdev
->dev
, pdev
);
46 mutex_init(&pdev
->dev_lock
);
49 pdev
->evtchn_irq
= INVALID_EVTCHN_IRQ
;
50 pdev
->be_watching
= 0;
52 INIT_WORK(&pdev
->op_work
, xen_pcibk_do_op
);
54 if (xen_pcibk_init_devices(pdev
)) {
62 static void xen_pcibk_disconnect(struct xen_pcibk_device
*pdev
)
64 mutex_lock(&pdev
->dev_lock
);
65 /* Ensure the guest can't trigger our handler before removing devices */
66 if (pdev
->evtchn_irq
!= INVALID_EVTCHN_IRQ
) {
67 unbind_from_irqhandler(pdev
->evtchn_irq
, pdev
);
68 pdev
->evtchn_irq
= INVALID_EVTCHN_IRQ
;
71 /* If the driver domain started an op, make sure we complete it
72 * before releasing the shared memory */
74 /* Note, the workqueue does not use spinlocks at all.*/
75 flush_workqueue(xen_pcibk_wq
);
77 if (pdev
->sh_info
!= NULL
) {
78 xenbus_unmap_ring_vfree(pdev
->xdev
, pdev
->sh_info
);
81 mutex_unlock(&pdev
->dev_lock
);
84 static void free_pdev(struct xen_pcibk_device
*pdev
)
86 if (pdev
->be_watching
) {
87 unregister_xenbus_watch(&pdev
->be_watch
);
88 pdev
->be_watching
= 0;
91 xen_pcibk_disconnect(pdev
);
93 xen_pcibk_release_devices(pdev
);
95 dev_set_drvdata(&pdev
->xdev
->dev
, NULL
);
101 static int xen_pcibk_do_attach(struct xen_pcibk_device
*pdev
, int gnt_ref
,
107 dev_dbg(&pdev
->xdev
->dev
,
108 "Attaching to frontend resources - gnt_ref=%d evtchn=%d\n",
109 gnt_ref
, remote_evtchn
);
111 err
= xenbus_map_ring_valloc(pdev
->xdev
, gnt_ref
, &vaddr
);
113 xenbus_dev_fatal(pdev
->xdev
, err
,
114 "Error mapping other domain page in ours.");
118 pdev
->sh_info
= vaddr
;
120 err
= bind_interdomain_evtchn_to_irqhandler(
121 pdev
->xdev
->otherend_id
, remote_evtchn
, xen_pcibk_handle_event
,
124 xenbus_dev_fatal(pdev
->xdev
, err
,
125 "Error binding event channel to IRQ");
128 pdev
->evtchn_irq
= err
;
131 dev_dbg(&pdev
->xdev
->dev
, "Attached!\n");
136 static int xen_pcibk_attach(struct xen_pcibk_device
*pdev
)
139 int gnt_ref
, remote_evtchn
;
143 mutex_lock(&pdev
->dev_lock
);
144 /* Make sure we only do this setup once */
145 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
146 XenbusStateInitialised
)
149 /* Wait for frontend to state that it has published the configuration */
150 if (xenbus_read_driver_state(pdev
->xdev
->otherend
) !=
151 XenbusStateInitialised
)
154 dev_dbg(&pdev
->xdev
->dev
, "Reading frontend config\n");
156 err
= xenbus_gather(XBT_NIL
, pdev
->xdev
->otherend
,
157 "pci-op-ref", "%u", &gnt_ref
,
158 "event-channel", "%u", &remote_evtchn
,
159 "magic", NULL
, &magic
, NULL
);
161 /* If configuration didn't get read correctly, wait longer */
162 xenbus_dev_fatal(pdev
->xdev
, err
,
163 "Error reading configuration from frontend");
167 if (magic
== NULL
|| strcmp(magic
, XEN_PCI_MAGIC
) != 0) {
168 xenbus_dev_fatal(pdev
->xdev
, -EFAULT
,
169 "version mismatch (%s/%s) with pcifront - "
171 magic
, XEN_PCI_MAGIC
);
175 err
= xen_pcibk_do_attach(pdev
, gnt_ref
, remote_evtchn
);
179 dev_dbg(&pdev
->xdev
->dev
, "Connecting...\n");
181 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateConnected
);
183 xenbus_dev_fatal(pdev
->xdev
, err
,
184 "Error switching to connected state!");
186 dev_dbg(&pdev
->xdev
->dev
, "Connected? %d\n", err
);
188 mutex_unlock(&pdev
->dev_lock
);
195 static int xen_pcibk_publish_pci_dev(struct xen_pcibk_device
*pdev
,
196 unsigned int domain
, unsigned int bus
,
197 unsigned int devfn
, unsigned int devid
)
203 len
= snprintf(str
, sizeof(str
), "vdev-%d", devid
);
204 if (unlikely(len
>= (sizeof(str
) - 1))) {
209 /* Note: The PV protocol uses %02x, don't change it */
210 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, str
,
211 "%04x:%02x:%02x.%02x", domain
, bus
,
212 PCI_SLOT(devfn
), PCI_FUNC(devfn
));
218 static int xen_pcibk_export_device(struct xen_pcibk_device
*pdev
,
219 int domain
, int bus
, int slot
, int func
,
225 dev_dbg(&pdev
->xdev
->dev
, "exporting dom %x bus %x slot %x func %x\n",
226 domain
, bus
, slot
, func
);
228 dev
= pcistub_get_pci_dev_by_slot(pdev
, domain
, bus
, slot
, func
);
231 xenbus_dev_fatal(pdev
->xdev
, err
,
232 "Couldn't locate PCI device "
233 "(%04x:%02x:%02x.%d)! "
234 "perhaps already in-use?",
235 domain
, bus
, slot
, func
);
239 err
= xen_pcibk_add_pci_dev(pdev
, dev
, devid
,
240 xen_pcibk_publish_pci_dev
);
244 dev_dbg(&dev
->dev
, "registering for %d\n", pdev
->xdev
->otherend_id
);
245 if (xen_register_device_domain_owner(dev
,
246 pdev
->xdev
->otherend_id
) != 0) {
247 dev_err(&dev
->dev
, "Stealing ownership from dom%d.\n",
248 xen_find_device_domain_owner(dev
));
249 xen_unregister_device_domain_owner(dev
);
250 xen_register_device_domain_owner(dev
, pdev
->xdev
->otherend_id
);
253 /* TODO: It'd be nice to export a bridge and have all of its children
254 * get exported with it. This may be best done in xend (which will
255 * have to calculate resource usage anyway) but we probably want to
256 * put something in here to ensure that if a bridge gets given to a
257 * driver domain, that all devices under that bridge are not given
258 * to other driver domains (as he who controls the bridge can disable
259 * it and stop the other devices from working).
265 static int xen_pcibk_remove_device(struct xen_pcibk_device
*pdev
,
266 int domain
, int bus
, int slot
, int func
)
271 dev_dbg(&pdev
->xdev
->dev
, "removing dom %x bus %x slot %x func %x\n",
272 domain
, bus
, slot
, func
);
274 dev
= xen_pcibk_get_pci_dev(pdev
, domain
, bus
, PCI_DEVFN(slot
, func
));
277 dev_dbg(&pdev
->xdev
->dev
, "Couldn't locate PCI device "
278 "(%04x:%02x:%02x.%d)! not owned by this domain\n",
279 domain
, bus
, slot
, func
);
283 dev_dbg(&dev
->dev
, "unregistering for %d\n", pdev
->xdev
->otherend_id
);
284 xen_unregister_device_domain_owner(dev
);
286 xen_pcibk_release_pci_dev(pdev
, dev
);
292 static int xen_pcibk_publish_pci_root(struct xen_pcibk_device
*pdev
,
293 unsigned int domain
, unsigned int bus
)
296 int i
, root_num
, len
, err
;
299 dev_dbg(&pdev
->xdev
->dev
, "Publishing pci roots\n");
301 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
302 "root_num", "%d", &root_num
);
303 if (err
== 0 || err
== -ENOENT
)
308 /* Verify that we haven't already published this pci root */
309 for (i
= 0; i
< root_num
; i
++) {
310 len
= snprintf(str
, sizeof(str
), "root-%d", i
);
311 if (unlikely(len
>= (sizeof(str
) - 1))) {
316 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
317 str
, "%x:%x", &d
, &b
);
325 if (d
== domain
&& b
== bus
) {
331 len
= snprintf(str
, sizeof(str
), "root-%d", root_num
);
332 if (unlikely(len
>= (sizeof(str
) - 1))) {
337 dev_dbg(&pdev
->xdev
->dev
, "writing root %d at %04x:%02x\n",
338 root_num
, domain
, bus
);
340 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, str
,
341 "%04x:%02x", domain
, bus
);
345 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
,
346 "root_num", "%d", (root_num
+ 1));
352 static int xen_pcibk_reconfigure(struct xen_pcibk_device
*pdev
)
356 int domain
, bus
, slot
, func
;
363 dev_dbg(&pdev
->xdev
->dev
, "Reconfiguring device ...\n");
365 mutex_lock(&pdev
->dev_lock
);
366 /* Make sure we only reconfigure once */
367 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
368 XenbusStateReconfiguring
)
371 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, "num_devs", "%d",
376 xenbus_dev_fatal(pdev
->xdev
, err
,
377 "Error reading number of devices");
381 for (i
= 0; i
< num_devs
; i
++) {
382 len
= snprintf(state_str
, sizeof(state_str
), "state-%d", i
);
383 if (unlikely(len
>= (sizeof(state_str
) - 1))) {
385 xenbus_dev_fatal(pdev
->xdev
, err
,
386 "String overflow while reading "
390 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, state_str
,
393 substate
= XenbusStateUnknown
;
396 case XenbusStateInitialising
:
397 dev_dbg(&pdev
->xdev
->dev
, "Attaching dev-%d ...\n", i
);
399 len
= snprintf(dev_str
, sizeof(dev_str
), "dev-%d", i
);
400 if (unlikely(len
>= (sizeof(dev_str
) - 1))) {
402 xenbus_dev_fatal(pdev
->xdev
, err
,
403 "String overflow while "
404 "reading configuration");
407 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
408 dev_str
, "%x:%x:%x.%x",
409 &domain
, &bus
, &slot
, &func
);
411 xenbus_dev_fatal(pdev
->xdev
, err
,
412 "Error reading device "
418 xenbus_dev_fatal(pdev
->xdev
, err
,
419 "Error parsing pci device "
424 err
= xen_pcibk_export_device(pdev
, domain
, bus
, slot
,
429 /* Publish pci roots. */
430 err
= xen_pcibk_publish_pci_roots(pdev
,
431 xen_pcibk_publish_pci_root
);
433 xenbus_dev_fatal(pdev
->xdev
, err
,
434 "Error while publish PCI root"
435 "buses for frontend");
439 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
,
441 XenbusStateInitialised
);
443 xenbus_dev_fatal(pdev
->xdev
, err
,
444 "Error switching substate of "
450 case XenbusStateClosing
:
451 dev_dbg(&pdev
->xdev
->dev
, "Detaching dev-%d ...\n", i
);
453 len
= snprintf(dev_str
, sizeof(dev_str
), "vdev-%d", i
);
454 if (unlikely(len
>= (sizeof(dev_str
) - 1))) {
456 xenbus_dev_fatal(pdev
->xdev
, err
,
457 "String overflow while "
458 "reading configuration");
461 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
462 dev_str
, "%x:%x:%x.%x",
463 &domain
, &bus
, &slot
, &func
);
465 xenbus_dev_fatal(pdev
->xdev
, err
,
466 "Error reading device "
472 xenbus_dev_fatal(pdev
->xdev
, err
,
473 "Error parsing pci device "
478 err
= xen_pcibk_remove_device(pdev
, domain
, bus
, slot
,
483 /* TODO: If at some point we implement support for pci
484 * root hot-remove on pcifront side, we'll need to
485 * remove unnecessary xenstore nodes of pci roots here.
495 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateReconfigured
);
497 xenbus_dev_fatal(pdev
->xdev
, err
,
498 "Error switching to reconfigured state!");
503 mutex_unlock(&pdev
->dev_lock
);
507 static void xen_pcibk_frontend_changed(struct xenbus_device
*xdev
,
508 enum xenbus_state fe_state
)
510 struct xen_pcibk_device
*pdev
= dev_get_drvdata(&xdev
->dev
);
512 dev_dbg(&xdev
->dev
, "fe state changed %d\n", fe_state
);
515 case XenbusStateInitialised
:
516 xen_pcibk_attach(pdev
);
519 case XenbusStateReconfiguring
:
520 xen_pcibk_reconfigure(pdev
);
523 case XenbusStateConnected
:
524 /* pcifront switched its state from reconfiguring to connected.
525 * Then switch to connected state.
527 xenbus_switch_state(xdev
, XenbusStateConnected
);
530 case XenbusStateClosing
:
531 xen_pcibk_disconnect(pdev
);
532 xenbus_switch_state(xdev
, XenbusStateClosing
);
535 case XenbusStateClosed
:
536 xen_pcibk_disconnect(pdev
);
537 xenbus_switch_state(xdev
, XenbusStateClosed
);
538 if (xenbus_dev_is_online(xdev
))
540 /* fall through if not online */
541 case XenbusStateUnknown
:
542 dev_dbg(&xdev
->dev
, "frontend is gone! unregister device\n");
543 device_unregister(&xdev
->dev
);
551 static int xen_pcibk_setup_backend(struct xen_pcibk_device
*pdev
)
553 /* Get configuration from xend (if available now) */
554 int domain
, bus
, slot
, func
;
560 mutex_lock(&pdev
->dev_lock
);
561 /* It's possible we could get the call to setup twice, so make sure
562 * we're not already connected.
564 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
568 dev_dbg(&pdev
->xdev
->dev
, "getting be setup\n");
570 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, "num_devs", "%d",
575 xenbus_dev_fatal(pdev
->xdev
, err
,
576 "Error reading number of devices");
580 for (i
= 0; i
< num_devs
; i
++) {
581 int l
= snprintf(dev_str
, sizeof(dev_str
), "dev-%d", i
);
582 if (unlikely(l
>= (sizeof(dev_str
) - 1))) {
584 xenbus_dev_fatal(pdev
->xdev
, err
,
585 "String overflow while reading "
590 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, dev_str
,
591 "%x:%x:%x.%x", &domain
, &bus
, &slot
, &func
);
593 xenbus_dev_fatal(pdev
->xdev
, err
,
594 "Error reading device configuration");
599 xenbus_dev_fatal(pdev
->xdev
, err
,
600 "Error parsing pci device "
605 err
= xen_pcibk_export_device(pdev
, domain
, bus
, slot
, func
, i
);
609 /* Switch substate of this device. */
610 l
= snprintf(state_str
, sizeof(state_str
), "state-%d", i
);
611 if (unlikely(l
>= (sizeof(state_str
) - 1))) {
613 xenbus_dev_fatal(pdev
->xdev
, err
,
614 "String overflow while reading "
618 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, state_str
,
619 "%d", XenbusStateInitialised
);
621 xenbus_dev_fatal(pdev
->xdev
, err
, "Error switching "
622 "substate of dev-%d\n", i
);
627 err
= xen_pcibk_publish_pci_roots(pdev
, xen_pcibk_publish_pci_root
);
629 xenbus_dev_fatal(pdev
->xdev
, err
,
630 "Error while publish PCI root buses "
635 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateInitialised
);
637 xenbus_dev_fatal(pdev
->xdev
, err
,
638 "Error switching to initialised state!");
641 mutex_unlock(&pdev
->dev_lock
);
643 /* see if pcifront is already configured (if not, we'll wait) */
644 xen_pcibk_attach(pdev
);
648 static void xen_pcibk_be_watch(struct xenbus_watch
*watch
,
649 const char **vec
, unsigned int len
)
651 struct xen_pcibk_device
*pdev
=
652 container_of(watch
, struct xen_pcibk_device
, be_watch
);
654 switch (xenbus_read_driver_state(pdev
->xdev
->nodename
)) {
655 case XenbusStateInitWait
:
656 xen_pcibk_setup_backend(pdev
);
664 static int xen_pcibk_xenbus_probe(struct xenbus_device
*dev
,
665 const struct xenbus_device_id
*id
)
668 struct xen_pcibk_device
*pdev
= alloc_pdev(dev
);
672 xenbus_dev_fatal(dev
, err
,
673 "Error allocating xen_pcibk_device struct");
677 /* wait for xend to configure us */
678 err
= xenbus_switch_state(dev
, XenbusStateInitWait
);
682 /* watch the backend node for backend configuration information */
683 err
= xenbus_watch_path(dev
, dev
->nodename
, &pdev
->be_watch
,
688 pdev
->be_watching
= 1;
690 /* We need to force a call to our callback here in case
691 * xend already configured us!
693 xen_pcibk_be_watch(&pdev
->be_watch
, NULL
, 0);
699 static int xen_pcibk_xenbus_remove(struct xenbus_device
*dev
)
701 struct xen_pcibk_device
*pdev
= dev_get_drvdata(&dev
->dev
);
709 static const struct xenbus_device_id xen_pcibk_ids
[] = {
714 static DEFINE_XENBUS_DRIVER(xen_pcibk
, DRV_NAME
,
715 .probe
= xen_pcibk_xenbus_probe
,
716 .remove
= xen_pcibk_xenbus_remove
,
717 .otherend_changed
= xen_pcibk_frontend_changed
,
720 const struct xen_pcibk_backend
*__read_mostly xen_pcibk_backend
;
722 int __init
xen_pcibk_xenbus_register(void)
724 xen_pcibk_wq
= create_workqueue("xen_pciback_workqueue");
726 printk(KERN_ERR
"%s: create"
727 "xen_pciback_workqueue failed\n", __func__
);
730 xen_pcibk_backend
= &xen_pcibk_vpci_backend
;
732 xen_pcibk_backend
= &xen_pcibk_passthrough_backend
;
733 pr_info(DRV_NAME
": backend is %s\n", xen_pcibk_backend
->name
);
734 return xenbus_register_backend(&xen_pcibk_driver
);
737 void __exit
xen_pcibk_xenbus_unregister(void)
739 destroy_workqueue(xen_pcibk_wq
);
740 xenbus_unregister_driver(&xen_pcibk_driver
);