2 * PCI Backend Xenbus Setup - handles setup with frontend and xend
4 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/list.h>
12 #include <linux/vmalloc.h>
13 #include <linux/workqueue.h>
14 #include <xen/xenbus.h>
15 #include <xen/events.h>
16 #include <asm/xen/pci.h>
19 #define INVALID_EVTCHN_IRQ (-1)
20 struct workqueue_struct
*xen_pcibk_wq
;
22 static bool __read_mostly passthrough
;
23 module_param(passthrough
, bool, S_IRUGO
);
24 MODULE_PARM_DESC(passthrough
,
25 "Option to specify how to export PCI topology to guest:\n"\
26 " 0 - (default) Hide the true PCI topology and makes the frontend\n"\
27 " there is a single PCI bus with only the exported devices on it.\n"\
28 " For example, a device at 03:05.0 will be re-assigned to 00:00.0\n"\
29 " while second device at 02:1a.1 will be re-assigned to 00:01.1.\n"\
30 " 1 - Passthrough provides a real view of the PCI topology to the\n"\
31 " frontend (for example, a device at 06:01.b will still appear at\n"\
32 " 06:01.b to the frontend). This is similar to how Xen 2.0.x\n"\
33 " exposed PCI devices to its driver domains. This may be required\n"\
34 " for drivers which depend on finding their hardward in certain\n"\
35 " bus/slot locations.");
37 static struct xen_pcibk_device
*alloc_pdev(struct xenbus_device
*xdev
)
39 struct xen_pcibk_device
*pdev
;
41 pdev
= kzalloc(sizeof(struct xen_pcibk_device
), GFP_KERNEL
);
44 dev_dbg(&xdev
->dev
, "allocated pdev @ 0x%p\n", pdev
);
47 dev_set_drvdata(&xdev
->dev
, pdev
);
49 mutex_init(&pdev
->dev_lock
);
52 pdev
->evtchn_irq
= INVALID_EVTCHN_IRQ
;
53 pdev
->be_watching
= 0;
55 INIT_WORK(&pdev
->op_work
, xen_pcibk_do_op
);
57 if (xen_pcibk_init_devices(pdev
)) {
65 static void xen_pcibk_disconnect(struct xen_pcibk_device
*pdev
)
67 mutex_lock(&pdev
->dev_lock
);
68 /* Ensure the guest can't trigger our handler before removing devices */
69 if (pdev
->evtchn_irq
!= INVALID_EVTCHN_IRQ
) {
70 unbind_from_irqhandler(pdev
->evtchn_irq
, pdev
);
71 pdev
->evtchn_irq
= INVALID_EVTCHN_IRQ
;
74 /* If the driver domain started an op, make sure we complete it
75 * before releasing the shared memory */
77 /* Note, the workqueue does not use spinlocks at all.*/
78 flush_workqueue(xen_pcibk_wq
);
80 if (pdev
->sh_info
!= NULL
) {
81 xenbus_unmap_ring_vfree(pdev
->xdev
, pdev
->sh_info
);
84 mutex_unlock(&pdev
->dev_lock
);
87 static void free_pdev(struct xen_pcibk_device
*pdev
)
89 if (pdev
->be_watching
) {
90 unregister_xenbus_watch(&pdev
->be_watch
);
91 pdev
->be_watching
= 0;
94 xen_pcibk_disconnect(pdev
);
96 xen_pcibk_release_devices(pdev
);
98 dev_set_drvdata(&pdev
->xdev
->dev
, NULL
);
104 static int xen_pcibk_do_attach(struct xen_pcibk_device
*pdev
, int gnt_ref
,
110 dev_dbg(&pdev
->xdev
->dev
,
111 "Attaching to frontend resources - gnt_ref=%d evtchn=%d\n",
112 gnt_ref
, remote_evtchn
);
114 err
= xenbus_map_ring_valloc(pdev
->xdev
, gnt_ref
, &vaddr
);
116 xenbus_dev_fatal(pdev
->xdev
, err
,
117 "Error mapping other domain page in ours.");
121 pdev
->sh_info
= vaddr
;
123 err
= bind_interdomain_evtchn_to_irqhandler(
124 pdev
->xdev
->otherend_id
, remote_evtchn
, xen_pcibk_handle_event
,
127 xenbus_dev_fatal(pdev
->xdev
, err
,
128 "Error binding event channel to IRQ");
131 pdev
->evtchn_irq
= err
;
134 dev_dbg(&pdev
->xdev
->dev
, "Attached!\n");
139 static int xen_pcibk_attach(struct xen_pcibk_device
*pdev
)
142 int gnt_ref
, remote_evtchn
;
146 mutex_lock(&pdev
->dev_lock
);
147 /* Make sure we only do this setup once */
148 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
149 XenbusStateInitialised
)
152 /* Wait for frontend to state that it has published the configuration */
153 if (xenbus_read_driver_state(pdev
->xdev
->otherend
) !=
154 XenbusStateInitialised
)
157 dev_dbg(&pdev
->xdev
->dev
, "Reading frontend config\n");
159 err
= xenbus_gather(XBT_NIL
, pdev
->xdev
->otherend
,
160 "pci-op-ref", "%u", &gnt_ref
,
161 "event-channel", "%u", &remote_evtchn
,
162 "magic", NULL
, &magic
, NULL
);
164 /* If configuration didn't get read correctly, wait longer */
165 xenbus_dev_fatal(pdev
->xdev
, err
,
166 "Error reading configuration from frontend");
170 if (magic
== NULL
|| strcmp(magic
, XEN_PCI_MAGIC
) != 0) {
171 xenbus_dev_fatal(pdev
->xdev
, -EFAULT
,
172 "version mismatch (%s/%s) with pcifront - "
174 magic
, XEN_PCI_MAGIC
);
178 err
= xen_pcibk_do_attach(pdev
, gnt_ref
, remote_evtchn
);
182 dev_dbg(&pdev
->xdev
->dev
, "Connecting...\n");
184 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateConnected
);
186 xenbus_dev_fatal(pdev
->xdev
, err
,
187 "Error switching to connected state!");
189 dev_dbg(&pdev
->xdev
->dev
, "Connected? %d\n", err
);
191 mutex_unlock(&pdev
->dev_lock
);
198 static int xen_pcibk_publish_pci_dev(struct xen_pcibk_device
*pdev
,
199 unsigned int domain
, unsigned int bus
,
200 unsigned int devfn
, unsigned int devid
)
206 len
= snprintf(str
, sizeof(str
), "vdev-%d", devid
);
207 if (unlikely(len
>= (sizeof(str
) - 1))) {
212 /* Note: The PV protocol uses %02x, don't change it */
213 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, str
,
214 "%04x:%02x:%02x.%02x", domain
, bus
,
215 PCI_SLOT(devfn
), PCI_FUNC(devfn
));
221 static int xen_pcibk_export_device(struct xen_pcibk_device
*pdev
,
222 int domain
, int bus
, int slot
, int func
,
228 dev_dbg(&pdev
->xdev
->dev
, "exporting dom %x bus %x slot %x func %x\n",
229 domain
, bus
, slot
, func
);
231 dev
= pcistub_get_pci_dev_by_slot(pdev
, domain
, bus
, slot
, func
);
234 xenbus_dev_fatal(pdev
->xdev
, err
,
235 "Couldn't locate PCI device "
236 "(%04x:%02x:%02x.%d)! "
237 "perhaps already in-use?",
238 domain
, bus
, slot
, func
);
242 err
= xen_pcibk_add_pci_dev(pdev
, dev
, devid
,
243 xen_pcibk_publish_pci_dev
);
247 dev_dbg(&dev
->dev
, "registering for %d\n", pdev
->xdev
->otherend_id
);
248 if (xen_register_device_domain_owner(dev
,
249 pdev
->xdev
->otherend_id
) != 0) {
250 dev_err(&dev
->dev
, "Stealing ownership from dom%d.\n",
251 xen_find_device_domain_owner(dev
));
252 xen_unregister_device_domain_owner(dev
);
253 xen_register_device_domain_owner(dev
, pdev
->xdev
->otherend_id
);
256 /* TODO: It'd be nice to export a bridge and have all of its children
257 * get exported with it. This may be best done in xend (which will
258 * have to calculate resource usage anyway) but we probably want to
259 * put something in here to ensure that if a bridge gets given to a
260 * driver domain, that all devices under that bridge are not given
261 * to other driver domains (as he who controls the bridge can disable
262 * it and stop the other devices from working).
268 static int xen_pcibk_remove_device(struct xen_pcibk_device
*pdev
,
269 int domain
, int bus
, int slot
, int func
)
274 dev_dbg(&pdev
->xdev
->dev
, "removing dom %x bus %x slot %x func %x\n",
275 domain
, bus
, slot
, func
);
277 dev
= xen_pcibk_get_pci_dev(pdev
, domain
, bus
, PCI_DEVFN(slot
, func
));
280 dev_dbg(&pdev
->xdev
->dev
, "Couldn't locate PCI device "
281 "(%04x:%02x:%02x.%d)! not owned by this domain\n",
282 domain
, bus
, slot
, func
);
286 dev_dbg(&dev
->dev
, "unregistering for %d\n", pdev
->xdev
->otherend_id
);
287 xen_unregister_device_domain_owner(dev
);
289 xen_pcibk_release_pci_dev(pdev
, dev
);
295 static int xen_pcibk_publish_pci_root(struct xen_pcibk_device
*pdev
,
296 unsigned int domain
, unsigned int bus
)
299 int i
, root_num
, len
, err
;
302 dev_dbg(&pdev
->xdev
->dev
, "Publishing pci roots\n");
304 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
305 "root_num", "%d", &root_num
);
306 if (err
== 0 || err
== -ENOENT
)
311 /* Verify that we haven't already published this pci root */
312 for (i
= 0; i
< root_num
; i
++) {
313 len
= snprintf(str
, sizeof(str
), "root-%d", i
);
314 if (unlikely(len
>= (sizeof(str
) - 1))) {
319 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
320 str
, "%x:%x", &d
, &b
);
328 if (d
== domain
&& b
== bus
) {
334 len
= snprintf(str
, sizeof(str
), "root-%d", root_num
);
335 if (unlikely(len
>= (sizeof(str
) - 1))) {
340 dev_dbg(&pdev
->xdev
->dev
, "writing root %d at %04x:%02x\n",
341 root_num
, domain
, bus
);
343 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, str
,
344 "%04x:%02x", domain
, bus
);
348 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
,
349 "root_num", "%d", (root_num
+ 1));
355 static int xen_pcibk_reconfigure(struct xen_pcibk_device
*pdev
)
359 int domain
, bus
, slot
, func
;
366 dev_dbg(&pdev
->xdev
->dev
, "Reconfiguring device ...\n");
368 mutex_lock(&pdev
->dev_lock
);
369 /* Make sure we only reconfigure once */
370 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
371 XenbusStateReconfiguring
)
374 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, "num_devs", "%d",
379 xenbus_dev_fatal(pdev
->xdev
, err
,
380 "Error reading number of devices");
384 for (i
= 0; i
< num_devs
; i
++) {
385 len
= snprintf(state_str
, sizeof(state_str
), "state-%d", i
);
386 if (unlikely(len
>= (sizeof(state_str
) - 1))) {
388 xenbus_dev_fatal(pdev
->xdev
, err
,
389 "String overflow while reading "
393 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, state_str
,
396 substate
= XenbusStateUnknown
;
399 case XenbusStateInitialising
:
400 dev_dbg(&pdev
->xdev
->dev
, "Attaching dev-%d ...\n", i
);
402 len
= snprintf(dev_str
, sizeof(dev_str
), "dev-%d", i
);
403 if (unlikely(len
>= (sizeof(dev_str
) - 1))) {
405 xenbus_dev_fatal(pdev
->xdev
, err
,
406 "String overflow while "
407 "reading configuration");
410 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
411 dev_str
, "%x:%x:%x.%x",
412 &domain
, &bus
, &slot
, &func
);
414 xenbus_dev_fatal(pdev
->xdev
, err
,
415 "Error reading device "
421 xenbus_dev_fatal(pdev
->xdev
, err
,
422 "Error parsing pci device "
427 err
= xen_pcibk_export_device(pdev
, domain
, bus
, slot
,
432 /* Publish pci roots. */
433 err
= xen_pcibk_publish_pci_roots(pdev
,
434 xen_pcibk_publish_pci_root
);
436 xenbus_dev_fatal(pdev
->xdev
, err
,
437 "Error while publish PCI root"
438 "buses for frontend");
442 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
,
444 XenbusStateInitialised
);
446 xenbus_dev_fatal(pdev
->xdev
, err
,
447 "Error switching substate of "
453 case XenbusStateClosing
:
454 dev_dbg(&pdev
->xdev
->dev
, "Detaching dev-%d ...\n", i
);
456 len
= snprintf(dev_str
, sizeof(dev_str
), "vdev-%d", i
);
457 if (unlikely(len
>= (sizeof(dev_str
) - 1))) {
459 xenbus_dev_fatal(pdev
->xdev
, err
,
460 "String overflow while "
461 "reading configuration");
464 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
465 dev_str
, "%x:%x:%x.%x",
466 &domain
, &bus
, &slot
, &func
);
468 xenbus_dev_fatal(pdev
->xdev
, err
,
469 "Error reading device "
475 xenbus_dev_fatal(pdev
->xdev
, err
,
476 "Error parsing pci device "
481 err
= xen_pcibk_remove_device(pdev
, domain
, bus
, slot
,
486 /* TODO: If at some point we implement support for pci
487 * root hot-remove on pcifront side, we'll need to
488 * remove unnecessary xenstore nodes of pci roots here.
498 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateReconfigured
);
500 xenbus_dev_fatal(pdev
->xdev
, err
,
501 "Error switching to reconfigured state!");
506 mutex_unlock(&pdev
->dev_lock
);
510 static void xen_pcibk_frontend_changed(struct xenbus_device
*xdev
,
511 enum xenbus_state fe_state
)
513 struct xen_pcibk_device
*pdev
= dev_get_drvdata(&xdev
->dev
);
515 dev_dbg(&xdev
->dev
, "fe state changed %d\n", fe_state
);
518 case XenbusStateInitialised
:
519 xen_pcibk_attach(pdev
);
522 case XenbusStateReconfiguring
:
523 xen_pcibk_reconfigure(pdev
);
526 case XenbusStateConnected
:
527 /* pcifront switched its state from reconfiguring to connected.
528 * Then switch to connected state.
530 xenbus_switch_state(xdev
, XenbusStateConnected
);
533 case XenbusStateClosing
:
534 xen_pcibk_disconnect(pdev
);
535 xenbus_switch_state(xdev
, XenbusStateClosing
);
538 case XenbusStateClosed
:
539 xen_pcibk_disconnect(pdev
);
540 xenbus_switch_state(xdev
, XenbusStateClosed
);
541 if (xenbus_dev_is_online(xdev
))
543 /* fall through if not online */
544 case XenbusStateUnknown
:
545 dev_dbg(&xdev
->dev
, "frontend is gone! unregister device\n");
546 device_unregister(&xdev
->dev
);
554 static int xen_pcibk_setup_backend(struct xen_pcibk_device
*pdev
)
556 /* Get configuration from xend (if available now) */
557 int domain
, bus
, slot
, func
;
563 mutex_lock(&pdev
->dev_lock
);
564 /* It's possible we could get the call to setup twice, so make sure
565 * we're not already connected.
567 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
571 dev_dbg(&pdev
->xdev
->dev
, "getting be setup\n");
573 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, "num_devs", "%d",
578 xenbus_dev_fatal(pdev
->xdev
, err
,
579 "Error reading number of devices");
583 for (i
= 0; i
< num_devs
; i
++) {
584 int l
= snprintf(dev_str
, sizeof(dev_str
), "dev-%d", i
);
585 if (unlikely(l
>= (sizeof(dev_str
) - 1))) {
587 xenbus_dev_fatal(pdev
->xdev
, err
,
588 "String overflow while reading "
593 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, dev_str
,
594 "%x:%x:%x.%x", &domain
, &bus
, &slot
, &func
);
596 xenbus_dev_fatal(pdev
->xdev
, err
,
597 "Error reading device configuration");
602 xenbus_dev_fatal(pdev
->xdev
, err
,
603 "Error parsing pci device "
608 err
= xen_pcibk_export_device(pdev
, domain
, bus
, slot
, func
, i
);
612 /* Switch substate of this device. */
613 l
= snprintf(state_str
, sizeof(state_str
), "state-%d", i
);
614 if (unlikely(l
>= (sizeof(state_str
) - 1))) {
616 xenbus_dev_fatal(pdev
->xdev
, err
,
617 "String overflow while reading "
621 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, state_str
,
622 "%d", XenbusStateInitialised
);
624 xenbus_dev_fatal(pdev
->xdev
, err
, "Error switching "
625 "substate of dev-%d\n", i
);
630 err
= xen_pcibk_publish_pci_roots(pdev
, xen_pcibk_publish_pci_root
);
632 xenbus_dev_fatal(pdev
->xdev
, err
,
633 "Error while publish PCI root buses "
638 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateInitialised
);
640 xenbus_dev_fatal(pdev
->xdev
, err
,
641 "Error switching to initialised state!");
644 mutex_unlock(&pdev
->dev_lock
);
646 /* see if pcifront is already configured (if not, we'll wait) */
647 xen_pcibk_attach(pdev
);
651 static void xen_pcibk_be_watch(struct xenbus_watch
*watch
,
652 const char **vec
, unsigned int len
)
654 struct xen_pcibk_device
*pdev
=
655 container_of(watch
, struct xen_pcibk_device
, be_watch
);
657 switch (xenbus_read_driver_state(pdev
->xdev
->nodename
)) {
658 case XenbusStateInitWait
:
659 xen_pcibk_setup_backend(pdev
);
667 static int xen_pcibk_xenbus_probe(struct xenbus_device
*dev
,
668 const struct xenbus_device_id
*id
)
671 struct xen_pcibk_device
*pdev
= alloc_pdev(dev
);
675 xenbus_dev_fatal(dev
, err
,
676 "Error allocating xen_pcibk_device struct");
680 /* wait for xend to configure us */
681 err
= xenbus_switch_state(dev
, XenbusStateInitWait
);
685 /* watch the backend node for backend configuration information */
686 err
= xenbus_watch_path(dev
, dev
->nodename
, &pdev
->be_watch
,
691 pdev
->be_watching
= 1;
693 /* We need to force a call to our callback here in case
694 * xend already configured us!
696 xen_pcibk_be_watch(&pdev
->be_watch
, NULL
, 0);
702 static int xen_pcibk_xenbus_remove(struct xenbus_device
*dev
)
704 struct xen_pcibk_device
*pdev
= dev_get_drvdata(&dev
->dev
);
712 static const struct xenbus_device_id xen_pcibk_ids
[] = {
717 static DEFINE_XENBUS_DRIVER(xen_pcibk
, DRV_NAME
,
718 .probe
= xen_pcibk_xenbus_probe
,
719 .remove
= xen_pcibk_xenbus_remove
,
720 .otherend_changed
= xen_pcibk_frontend_changed
,
723 const struct xen_pcibk_backend
*__read_mostly xen_pcibk_backend
;
725 int __init
xen_pcibk_xenbus_register(void)
727 xen_pcibk_wq
= create_workqueue("xen_pciback_workqueue");
729 pr_err("%s: create xen_pciback_workqueue failed\n", __func__
);
732 xen_pcibk_backend
= &xen_pcibk_vpci_backend
;
734 xen_pcibk_backend
= &xen_pcibk_passthrough_backend
;
735 pr_info("backend is %s\n", xen_pcibk_backend
->name
);
736 return xenbus_register_backend(&xen_pcibk_driver
);
739 void __exit
xen_pcibk_xenbus_unregister(void)
741 destroy_workqueue(xen_pcibk_wq
);
742 xenbus_unregister_driver(&xen_pcibk_driver
);