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 int __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 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, str
,
210 "%04x:%02x:%02x.%02x", domain
, bus
,
211 PCI_SLOT(devfn
), PCI_FUNC(devfn
));
217 static int xen_pcibk_export_device(struct xen_pcibk_device
*pdev
,
218 int domain
, int bus
, int slot
, int func
,
224 dev_dbg(&pdev
->xdev
->dev
, "exporting dom %x bus %x slot %x func %x\n",
225 domain
, bus
, slot
, func
);
227 dev
= pcistub_get_pci_dev_by_slot(pdev
, domain
, bus
, slot
, func
);
230 xenbus_dev_fatal(pdev
->xdev
, err
,
231 "Couldn't locate PCI device "
232 "(%04x:%02x:%02x.%01x)! "
233 "perhaps already in-use?",
234 domain
, bus
, slot
, func
);
238 err
= xen_pcibk_add_pci_dev(pdev
, dev
, devid
,
239 xen_pcibk_publish_pci_dev
);
243 dev_dbg(&dev
->dev
, "registering for %d\n", pdev
->xdev
->otherend_id
);
244 dev
->dev_flags
|= PCI_DEV_FLAGS_ASSIGNED
;
245 if (xen_register_device_domain_owner(dev
,
246 pdev
->xdev
->otherend_id
) != 0) {
247 dev_err(&dev
->dev
, "device has been assigned to another " \
248 "domain! Over-writting the ownership, but beware.\n");
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.%01x)! 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 dev
->dev_flags
&= ~PCI_DEV_FLAGS_ASSIGNED
;
285 xen_unregister_device_domain_owner(dev
);
287 xen_pcibk_release_pci_dev(pdev
, dev
);
293 static int xen_pcibk_publish_pci_root(struct xen_pcibk_device
*pdev
,
294 unsigned int domain
, unsigned int bus
)
297 int i
, root_num
, len
, err
;
300 dev_dbg(&pdev
->xdev
->dev
, "Publishing pci roots\n");
302 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
303 "root_num", "%d", &root_num
);
304 if (err
== 0 || err
== -ENOENT
)
309 /* Verify that we haven't already published this pci root */
310 for (i
= 0; i
< root_num
; i
++) {
311 len
= snprintf(str
, sizeof(str
), "root-%d", i
);
312 if (unlikely(len
>= (sizeof(str
) - 1))) {
317 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
318 str
, "%x:%x", &d
, &b
);
326 if (d
== domain
&& b
== bus
) {
332 len
= snprintf(str
, sizeof(str
), "root-%d", root_num
);
333 if (unlikely(len
>= (sizeof(str
) - 1))) {
338 dev_dbg(&pdev
->xdev
->dev
, "writing root %d at %04x:%02x\n",
339 root_num
, domain
, bus
);
341 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, str
,
342 "%04x:%02x", domain
, bus
);
346 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
,
347 "root_num", "%d", (root_num
+ 1));
353 static int xen_pcibk_reconfigure(struct xen_pcibk_device
*pdev
)
357 int domain
, bus
, slot
, func
;
364 dev_dbg(&pdev
->xdev
->dev
, "Reconfiguring device ...\n");
366 mutex_lock(&pdev
->dev_lock
);
367 /* Make sure we only reconfigure once */
368 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
369 XenbusStateReconfiguring
)
372 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, "num_devs", "%d",
377 xenbus_dev_fatal(pdev
->xdev
, err
,
378 "Error reading number of devices");
382 for (i
= 0; i
< num_devs
; i
++) {
383 len
= snprintf(state_str
, sizeof(state_str
), "state-%d", i
);
384 if (unlikely(len
>= (sizeof(state_str
) - 1))) {
386 xenbus_dev_fatal(pdev
->xdev
, err
,
387 "String overflow while reading "
391 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, state_str
,
394 substate
= XenbusStateUnknown
;
397 case XenbusStateInitialising
:
398 dev_dbg(&pdev
->xdev
->dev
, "Attaching dev-%d ...\n", i
);
400 len
= snprintf(dev_str
, sizeof(dev_str
), "dev-%d", i
);
401 if (unlikely(len
>= (sizeof(dev_str
) - 1))) {
403 xenbus_dev_fatal(pdev
->xdev
, err
,
404 "String overflow while "
405 "reading configuration");
408 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
409 dev_str
, "%x:%x:%x.%x",
410 &domain
, &bus
, &slot
, &func
);
412 xenbus_dev_fatal(pdev
->xdev
, err
,
413 "Error reading device "
419 xenbus_dev_fatal(pdev
->xdev
, err
,
420 "Error parsing pci device "
425 err
= xen_pcibk_export_device(pdev
, domain
, bus
, slot
,
430 /* Publish pci roots. */
431 err
= xen_pcibk_publish_pci_roots(pdev
,
432 xen_pcibk_publish_pci_root
);
434 xenbus_dev_fatal(pdev
->xdev
, err
,
435 "Error while publish PCI root"
436 "buses for frontend");
440 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
,
442 XenbusStateInitialised
);
444 xenbus_dev_fatal(pdev
->xdev
, err
,
445 "Error switching substate of "
451 case XenbusStateClosing
:
452 dev_dbg(&pdev
->xdev
->dev
, "Detaching dev-%d ...\n", i
);
454 len
= snprintf(dev_str
, sizeof(dev_str
), "vdev-%d", i
);
455 if (unlikely(len
>= (sizeof(dev_str
) - 1))) {
457 xenbus_dev_fatal(pdev
->xdev
, err
,
458 "String overflow while "
459 "reading configuration");
462 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
463 dev_str
, "%x:%x:%x.%x",
464 &domain
, &bus
, &slot
, &func
);
466 xenbus_dev_fatal(pdev
->xdev
, err
,
467 "Error reading device "
473 xenbus_dev_fatal(pdev
->xdev
, err
,
474 "Error parsing pci device "
479 err
= xen_pcibk_remove_device(pdev
, domain
, bus
, slot
,
484 /* TODO: If at some point we implement support for pci
485 * root hot-remove on pcifront side, we'll need to
486 * remove unnecessary xenstore nodes of pci roots here.
496 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateReconfigured
);
498 xenbus_dev_fatal(pdev
->xdev
, err
,
499 "Error switching to reconfigured state!");
504 mutex_unlock(&pdev
->dev_lock
);
508 static void xen_pcibk_frontend_changed(struct xenbus_device
*xdev
,
509 enum xenbus_state fe_state
)
511 struct xen_pcibk_device
*pdev
= dev_get_drvdata(&xdev
->dev
);
513 dev_dbg(&xdev
->dev
, "fe state changed %d\n", fe_state
);
516 case XenbusStateInitialised
:
517 xen_pcibk_attach(pdev
);
520 case XenbusStateReconfiguring
:
521 xen_pcibk_reconfigure(pdev
);
524 case XenbusStateConnected
:
525 /* pcifront switched its state from reconfiguring to connected.
526 * Then switch to connected state.
528 xenbus_switch_state(xdev
, XenbusStateConnected
);
531 case XenbusStateClosing
:
532 xen_pcibk_disconnect(pdev
);
533 xenbus_switch_state(xdev
, XenbusStateClosing
);
536 case XenbusStateClosed
:
537 xen_pcibk_disconnect(pdev
);
538 xenbus_switch_state(xdev
, XenbusStateClosed
);
539 if (xenbus_dev_is_online(xdev
))
541 /* fall through if not online */
542 case XenbusStateUnknown
:
543 dev_dbg(&xdev
->dev
, "frontend is gone! unregister device\n");
544 device_unregister(&xdev
->dev
);
552 static int xen_pcibk_setup_backend(struct xen_pcibk_device
*pdev
)
554 /* Get configuration from xend (if available now) */
555 int domain
, bus
, slot
, func
;
561 mutex_lock(&pdev
->dev_lock
);
562 /* It's possible we could get the call to setup twice, so make sure
563 * we're not already connected.
565 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
569 dev_dbg(&pdev
->xdev
->dev
, "getting be setup\n");
571 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, "num_devs", "%d",
576 xenbus_dev_fatal(pdev
->xdev
, err
,
577 "Error reading number of devices");
581 for (i
= 0; i
< num_devs
; i
++) {
582 int l
= snprintf(dev_str
, sizeof(dev_str
), "dev-%d", i
);
583 if (unlikely(l
>= (sizeof(dev_str
) - 1))) {
585 xenbus_dev_fatal(pdev
->xdev
, err
,
586 "String overflow while reading "
591 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, dev_str
,
592 "%x:%x:%x.%x", &domain
, &bus
, &slot
, &func
);
594 xenbus_dev_fatal(pdev
->xdev
, err
,
595 "Error reading device configuration");
600 xenbus_dev_fatal(pdev
->xdev
, err
,
601 "Error parsing pci device "
606 err
= xen_pcibk_export_device(pdev
, domain
, bus
, slot
, func
, i
);
610 /* Switch substate of this device. */
611 l
= snprintf(state_str
, sizeof(state_str
), "state-%d", i
);
612 if (unlikely(l
>= (sizeof(state_str
) - 1))) {
614 xenbus_dev_fatal(pdev
->xdev
, err
,
615 "String overflow while reading "
619 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, state_str
,
620 "%d", XenbusStateInitialised
);
622 xenbus_dev_fatal(pdev
->xdev
, err
, "Error switching "
623 "substate of dev-%d\n", i
);
628 err
= xen_pcibk_publish_pci_roots(pdev
, xen_pcibk_publish_pci_root
);
630 xenbus_dev_fatal(pdev
->xdev
, err
,
631 "Error while publish PCI root buses "
636 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateInitialised
);
638 xenbus_dev_fatal(pdev
->xdev
, err
,
639 "Error switching to initialised state!");
642 mutex_unlock(&pdev
->dev_lock
);
644 /* see if pcifront is already configured (if not, we'll wait) */
645 xen_pcibk_attach(pdev
);
649 static void xen_pcibk_be_watch(struct xenbus_watch
*watch
,
650 const char **vec
, unsigned int len
)
652 struct xen_pcibk_device
*pdev
=
653 container_of(watch
, struct xen_pcibk_device
, be_watch
);
655 switch (xenbus_read_driver_state(pdev
->xdev
->nodename
)) {
656 case XenbusStateInitWait
:
657 xen_pcibk_setup_backend(pdev
);
665 static int xen_pcibk_xenbus_probe(struct xenbus_device
*dev
,
666 const struct xenbus_device_id
*id
)
669 struct xen_pcibk_device
*pdev
= alloc_pdev(dev
);
673 xenbus_dev_fatal(dev
, err
,
674 "Error allocating xen_pcibk_device struct");
678 /* wait for xend to configure us */
679 err
= xenbus_switch_state(dev
, XenbusStateInitWait
);
683 /* watch the backend node for backend configuration information */
684 err
= xenbus_watch_path(dev
, dev
->nodename
, &pdev
->be_watch
,
689 pdev
->be_watching
= 1;
691 /* We need to force a call to our callback here in case
692 * xend already configured us!
694 xen_pcibk_be_watch(&pdev
->be_watch
, NULL
, 0);
700 static int xen_pcibk_xenbus_remove(struct xenbus_device
*dev
)
702 struct xen_pcibk_device
*pdev
= dev_get_drvdata(&dev
->dev
);
710 static const struct xenbus_device_id xenpci_ids
[] = {
715 static struct xenbus_driver xenbus_xen_pcibk_driver
= {
717 .owner
= THIS_MODULE
,
719 .probe
= xen_pcibk_xenbus_probe
,
720 .remove
= xen_pcibk_xenbus_remove
,
721 .otherend_changed
= xen_pcibk_frontend_changed
,
724 const struct xen_pcibk_backend
*__read_mostly xen_pcibk_backend
;
726 int __init
xen_pcibk_xenbus_register(void)
728 xen_pcibk_wq
= create_workqueue("xen_pciback_workqueue");
730 printk(KERN_ERR
"%s: create"
731 "xen_pciback_workqueue failed\n", __func__
);
734 xen_pcibk_backend
= &xen_pcibk_vpci_backend
;
736 xen_pcibk_backend
= &xen_pcibk_passthrough_backend
;
737 pr_info(DRV_NAME
": backend is %s\n", xen_pcibk_backend
->name
);
738 return xenbus_register_backend(&xenbus_xen_pcibk_driver
);
741 void __exit
xen_pcibk_xenbus_unregister(void)
743 destroy_workqueue(xen_pcibk_wq
);
744 xenbus_unregister_driver(&xenbus_xen_pcibk_driver
);