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 DRV_NAME "xen-pciback"
17 #define INVALID_EVTCHN_IRQ (-1)
18 struct workqueue_struct
*xen_pcibk_wq
;
20 static int __read_mostly passthrough
;
21 module_param(passthrough
, bool, S_IRUGO
);
22 MODULE_PARM_DESC(passthrough
,
23 "Option to specify how to export PCI topology to guest:\n"\
24 " 0 - (default) Hide the true PCI topology and makes the frontend\n"\
25 " there is a single PCI bus with only the exported devices on it.\n"\
26 " For example, a device at 03:05.0 will be re-assigned to 00:00.0\n"\
27 " while second device at 02:1a.1 will be re-assigned to 00:01.1.\n"\
28 " 1 - Passthrough provides a real view of the PCI topology to the\n"\
29 " frontend (for example, a device at 06:01.b will still appear at\n"\
30 " 06:01.b to the frontend). This is similar to how Xen 2.0.x\n"\
31 " exposed PCI devices to its driver domains. This may be required\n"\
32 " for drivers which depend on finding their hardward in certain\n"\
33 " bus/slot locations.");
35 static struct xen_pcibk_device
*alloc_pdev(struct xenbus_device
*xdev
)
37 struct xen_pcibk_device
*pdev
;
39 pdev
= kzalloc(sizeof(struct xen_pcibk_device
), GFP_KERNEL
);
42 dev_dbg(&xdev
->dev
, "allocated pdev @ 0x%p\n", pdev
);
45 dev_set_drvdata(&xdev
->dev
, pdev
);
47 spin_lock_init(&pdev
->dev_lock
);
50 pdev
->evtchn_irq
= INVALID_EVTCHN_IRQ
;
51 pdev
->be_watching
= 0;
53 INIT_WORK(&pdev
->op_work
, xen_pcibk_do_op
);
55 if (xen_pcibk_init_devices(pdev
)) {
63 static void xen_pcibk_disconnect(struct xen_pcibk_device
*pdev
)
65 spin_lock(&pdev
->dev_lock
);
67 /* Ensure the guest can't trigger our handler before removing devices */
68 if (pdev
->evtchn_irq
!= INVALID_EVTCHN_IRQ
) {
69 unbind_from_irqhandler(pdev
->evtchn_irq
, pdev
);
70 pdev
->evtchn_irq
= INVALID_EVTCHN_IRQ
;
72 spin_unlock(&pdev
->dev_lock
);
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 spin_lock(&pdev
->dev_lock
);
81 if (pdev
->sh_info
!= NULL
) {
82 xenbus_unmap_ring_vfree(pdev
->xdev
, pdev
->sh_info
);
85 spin_unlock(&pdev
->dev_lock
);
89 static void free_pdev(struct xen_pcibk_device
*pdev
)
91 if (pdev
->be_watching
) {
92 unregister_xenbus_watch(&pdev
->be_watch
);
93 pdev
->be_watching
= 0;
96 xen_pcibk_disconnect(pdev
);
98 xen_pcibk_release_devices(pdev
);
100 dev_set_drvdata(&pdev
->xdev
->dev
, NULL
);
106 static int xen_pcibk_do_attach(struct xen_pcibk_device
*pdev
, int gnt_ref
,
112 dev_dbg(&pdev
->xdev
->dev
,
113 "Attaching to frontend resources - gnt_ref=%d evtchn=%d\n",
114 gnt_ref
, remote_evtchn
);
116 err
= xenbus_map_ring_valloc(pdev
->xdev
, gnt_ref
, &vaddr
);
118 xenbus_dev_fatal(pdev
->xdev
, err
,
119 "Error mapping other domain page in ours.");
123 spin_lock(&pdev
->dev_lock
);
124 pdev
->sh_info
= vaddr
;
125 spin_unlock(&pdev
->dev_lock
);
127 err
= bind_interdomain_evtchn_to_irqhandler(
128 pdev
->xdev
->otherend_id
, remote_evtchn
, xen_pcibk_handle_event
,
131 xenbus_dev_fatal(pdev
->xdev
, err
,
132 "Error binding event channel to IRQ");
136 spin_lock(&pdev
->dev_lock
);
137 pdev
->evtchn_irq
= err
;
138 spin_unlock(&pdev
->dev_lock
);
141 dev_dbg(&pdev
->xdev
->dev
, "Attached!\n");
146 static int xen_pcibk_attach(struct xen_pcibk_device
*pdev
)
149 int gnt_ref
, remote_evtchn
;
153 /* Make sure we only do this setup once */
154 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
155 XenbusStateInitialised
)
158 /* Wait for frontend to state that it has published the configuration */
159 if (xenbus_read_driver_state(pdev
->xdev
->otherend
) !=
160 XenbusStateInitialised
)
163 dev_dbg(&pdev
->xdev
->dev
, "Reading frontend config\n");
165 err
= xenbus_gather(XBT_NIL
, pdev
->xdev
->otherend
,
166 "pci-op-ref", "%u", &gnt_ref
,
167 "event-channel", "%u", &remote_evtchn
,
168 "magic", NULL
, &magic
, NULL
);
170 /* If configuration didn't get read correctly, wait longer */
171 xenbus_dev_fatal(pdev
->xdev
, err
,
172 "Error reading configuration from frontend");
176 if (magic
== NULL
|| strcmp(magic
, XEN_PCI_MAGIC
) != 0) {
177 xenbus_dev_fatal(pdev
->xdev
, -EFAULT
,
178 "version mismatch (%s/%s) with pcifront - "
180 magic
, XEN_PCI_MAGIC
);
184 err
= xen_pcibk_do_attach(pdev
, gnt_ref
, remote_evtchn
);
188 dev_dbg(&pdev
->xdev
->dev
, "Connecting...\n");
190 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateConnected
);
192 xenbus_dev_fatal(pdev
->xdev
, err
,
193 "Error switching to connected state!");
195 dev_dbg(&pdev
->xdev
->dev
, "Connected? %d\n", err
);
203 static int xen_pcibk_publish_pci_dev(struct xen_pcibk_device
*pdev
,
204 unsigned int domain
, unsigned int bus
,
205 unsigned int devfn
, unsigned int devid
)
211 len
= snprintf(str
, sizeof(str
), "vdev-%d", devid
);
212 if (unlikely(len
>= (sizeof(str
) - 1))) {
217 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, str
,
218 "%04x:%02x:%02x.%02x", domain
, bus
,
219 PCI_SLOT(devfn
), PCI_FUNC(devfn
));
225 static int xen_pcibk_export_device(struct xen_pcibk_device
*pdev
,
226 int domain
, int bus
, int slot
, int func
,
232 dev_dbg(&pdev
->xdev
->dev
, "exporting dom %x bus %x slot %x func %x\n",
233 domain
, bus
, slot
, func
);
235 dev
= pcistub_get_pci_dev_by_slot(pdev
, domain
, bus
, slot
, func
);
238 xenbus_dev_fatal(pdev
->xdev
, err
,
239 "Couldn't locate PCI device "
240 "(%04x:%02x:%02x.%01x)! "
241 "perhaps already in-use?",
242 domain
, bus
, slot
, func
);
246 err
= xen_pcibk_add_pci_dev(pdev
, dev
, devid
,
247 xen_pcibk_publish_pci_dev
);
251 dev_dbg(&dev
->dev
, "registering for %d\n", pdev
->xdev
->otherend_id
);
252 if (xen_register_device_domain_owner(dev
,
253 pdev
->xdev
->otherend_id
) != 0) {
254 dev_err(&dev
->dev
, "device has been assigned to another " \
255 "domain! Over-writting the ownership, but beware.\n");
256 xen_unregister_device_domain_owner(dev
);
257 xen_register_device_domain_owner(dev
, pdev
->xdev
->otherend_id
);
260 /* TODO: It'd be nice to export a bridge and have all of its children
261 * get exported with it. This may be best done in xend (which will
262 * have to calculate resource usage anyway) but we probably want to
263 * put something in here to ensure that if a bridge gets given to a
264 * driver domain, that all devices under that bridge are not given
265 * to other driver domains (as he who controls the bridge can disable
266 * it and stop the other devices from working).
272 static int xen_pcibk_remove_device(struct xen_pcibk_device
*pdev
,
273 int domain
, int bus
, int slot
, int func
)
278 dev_dbg(&pdev
->xdev
->dev
, "removing dom %x bus %x slot %x func %x\n",
279 domain
, bus
, slot
, func
);
281 dev
= xen_pcibk_get_pci_dev(pdev
, domain
, bus
, PCI_DEVFN(slot
, func
));
284 dev_dbg(&pdev
->xdev
->dev
, "Couldn't locate PCI device "
285 "(%04x:%02x:%02x.%01x)! not owned by this domain\n",
286 domain
, bus
, slot
, func
);
290 dev_dbg(&dev
->dev
, "unregistering for %d\n", pdev
->xdev
->otherend_id
);
291 xen_unregister_device_domain_owner(dev
);
293 xen_pcibk_release_pci_dev(pdev
, dev
);
299 static int xen_pcibk_publish_pci_root(struct xen_pcibk_device
*pdev
,
300 unsigned int domain
, unsigned int bus
)
303 int i
, root_num
, len
, err
;
306 dev_dbg(&pdev
->xdev
->dev
, "Publishing pci roots\n");
308 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
309 "root_num", "%d", &root_num
);
310 if (err
== 0 || err
== -ENOENT
)
315 /* Verify that we haven't already published this pci root */
316 for (i
= 0; i
< root_num
; i
++) {
317 len
= snprintf(str
, sizeof(str
), "root-%d", i
);
318 if (unlikely(len
>= (sizeof(str
) - 1))) {
323 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
324 str
, "%x:%x", &d
, &b
);
332 if (d
== domain
&& b
== bus
) {
338 len
= snprintf(str
, sizeof(str
), "root-%d", root_num
);
339 if (unlikely(len
>= (sizeof(str
) - 1))) {
344 dev_dbg(&pdev
->xdev
->dev
, "writing root %d at %04x:%02x\n",
345 root_num
, domain
, bus
);
347 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, str
,
348 "%04x:%02x", domain
, bus
);
352 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
,
353 "root_num", "%d", (root_num
+ 1));
359 static int xen_pcibk_reconfigure(struct xen_pcibk_device
*pdev
)
363 int domain
, bus
, slot
, func
;
370 dev_dbg(&pdev
->xdev
->dev
, "Reconfiguring device ...\n");
372 /* Make sure we only reconfigure once */
373 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
374 XenbusStateReconfiguring
)
377 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, "num_devs", "%d",
382 xenbus_dev_fatal(pdev
->xdev
, err
,
383 "Error reading number of devices");
387 for (i
= 0; i
< num_devs
; i
++) {
388 len
= snprintf(state_str
, sizeof(state_str
), "state-%d", i
);
389 if (unlikely(len
>= (sizeof(state_str
) - 1))) {
391 xenbus_dev_fatal(pdev
->xdev
, err
,
392 "String overflow while reading "
396 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, state_str
,
399 substate
= XenbusStateUnknown
;
402 case XenbusStateInitialising
:
403 dev_dbg(&pdev
->xdev
->dev
, "Attaching dev-%d ...\n", i
);
405 len
= snprintf(dev_str
, sizeof(dev_str
), "dev-%d", i
);
406 if (unlikely(len
>= (sizeof(dev_str
) - 1))) {
408 xenbus_dev_fatal(pdev
->xdev
, err
,
409 "String overflow while "
410 "reading configuration");
413 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
414 dev_str
, "%x:%x:%x.%x",
415 &domain
, &bus
, &slot
, &func
);
417 xenbus_dev_fatal(pdev
->xdev
, err
,
418 "Error reading device "
424 xenbus_dev_fatal(pdev
->xdev
, err
,
425 "Error parsing pci device "
430 err
= xen_pcibk_export_device(pdev
, domain
, bus
, slot
,
435 /* Publish pci roots. */
436 err
= xen_pcibk_publish_pci_roots(pdev
,
437 xen_pcibk_publish_pci_root
);
439 xenbus_dev_fatal(pdev
->xdev
, err
,
440 "Error while publish PCI root"
441 "buses for frontend");
445 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
,
447 XenbusStateInitialised
);
449 xenbus_dev_fatal(pdev
->xdev
, err
,
450 "Error switching substate of "
456 case XenbusStateClosing
:
457 dev_dbg(&pdev
->xdev
->dev
, "Detaching dev-%d ...\n", i
);
459 len
= snprintf(dev_str
, sizeof(dev_str
), "vdev-%d", i
);
460 if (unlikely(len
>= (sizeof(dev_str
) - 1))) {
462 xenbus_dev_fatal(pdev
->xdev
, err
,
463 "String overflow while "
464 "reading configuration");
467 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
,
468 dev_str
, "%x:%x:%x.%x",
469 &domain
, &bus
, &slot
, &func
);
471 xenbus_dev_fatal(pdev
->xdev
, err
,
472 "Error reading device "
478 xenbus_dev_fatal(pdev
->xdev
, err
,
479 "Error parsing pci device "
484 err
= xen_pcibk_remove_device(pdev
, domain
, bus
, slot
,
489 /* TODO: If at some point we implement support for pci
490 * root hot-remove on pcifront side, we'll need to
491 * remove unnecessary xenstore nodes of pci roots here.
501 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateReconfigured
);
503 xenbus_dev_fatal(pdev
->xdev
, err
,
504 "Error switching to reconfigured state!");
512 static void xen_pcibk_frontend_changed(struct xenbus_device
*xdev
,
513 enum xenbus_state fe_state
)
515 struct xen_pcibk_device
*pdev
= dev_get_drvdata(&xdev
->dev
);
517 dev_dbg(&xdev
->dev
, "fe state changed %d\n", fe_state
);
520 case XenbusStateInitialised
:
521 xen_pcibk_attach(pdev
);
524 case XenbusStateReconfiguring
:
525 xen_pcibk_reconfigure(pdev
);
528 case XenbusStateConnected
:
529 /* pcifront switched its state from reconfiguring to connected.
530 * Then switch to connected state.
532 xenbus_switch_state(xdev
, XenbusStateConnected
);
535 case XenbusStateClosing
:
536 xen_pcibk_disconnect(pdev
);
537 xenbus_switch_state(xdev
, XenbusStateClosing
);
540 case XenbusStateClosed
:
541 xen_pcibk_disconnect(pdev
);
542 xenbus_switch_state(xdev
, XenbusStateClosed
);
543 if (xenbus_dev_is_online(xdev
))
545 /* fall through if not online */
546 case XenbusStateUnknown
:
547 dev_dbg(&xdev
->dev
, "frontend is gone! unregister device\n");
548 device_unregister(&xdev
->dev
);
556 static int xen_pcibk_setup_backend(struct xen_pcibk_device
*pdev
)
558 /* Get configuration from xend (if available now) */
559 int domain
, bus
, slot
, func
;
565 /* It's possible we could get the call to setup twice, so make sure
566 * we're not already connected.
568 if (xenbus_read_driver_state(pdev
->xdev
->nodename
) !=
572 dev_dbg(&pdev
->xdev
->dev
, "getting be setup\n");
574 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, "num_devs", "%d",
579 xenbus_dev_fatal(pdev
->xdev
, err
,
580 "Error reading number of devices");
584 for (i
= 0; i
< num_devs
; i
++) {
585 int l
= snprintf(dev_str
, sizeof(dev_str
), "dev-%d", i
);
586 if (unlikely(l
>= (sizeof(dev_str
) - 1))) {
588 xenbus_dev_fatal(pdev
->xdev
, err
,
589 "String overflow while reading "
594 err
= xenbus_scanf(XBT_NIL
, pdev
->xdev
->nodename
, dev_str
,
595 "%x:%x:%x.%x", &domain
, &bus
, &slot
, &func
);
597 xenbus_dev_fatal(pdev
->xdev
, err
,
598 "Error reading device configuration");
603 xenbus_dev_fatal(pdev
->xdev
, err
,
604 "Error parsing pci device "
609 err
= xen_pcibk_export_device(pdev
, domain
, bus
, slot
, func
, i
);
613 /* Switch substate of this device. */
614 l
= snprintf(state_str
, sizeof(state_str
), "state-%d", i
);
615 if (unlikely(l
>= (sizeof(state_str
) - 1))) {
617 xenbus_dev_fatal(pdev
->xdev
, err
,
618 "String overflow while reading "
622 err
= xenbus_printf(XBT_NIL
, pdev
->xdev
->nodename
, state_str
,
623 "%d", XenbusStateInitialised
);
625 xenbus_dev_fatal(pdev
->xdev
, err
, "Error switching "
626 "substate of dev-%d\n", i
);
631 err
= xen_pcibk_publish_pci_roots(pdev
, xen_pcibk_publish_pci_root
);
633 xenbus_dev_fatal(pdev
->xdev
, err
,
634 "Error while publish PCI root buses "
639 err
= xenbus_switch_state(pdev
->xdev
, XenbusStateInitialised
);
641 xenbus_dev_fatal(pdev
->xdev
, err
,
642 "Error switching to initialised state!");
646 /* see if pcifront is already configured (if not, we'll wait) */
647 xen_pcibk_attach(pdev
);
652 static void xen_pcibk_be_watch(struct xenbus_watch
*watch
,
653 const char **vec
, unsigned int len
)
655 struct xen_pcibk_device
*pdev
=
656 container_of(watch
, struct xen_pcibk_device
, be_watch
);
658 switch (xenbus_read_driver_state(pdev
->xdev
->nodename
)) {
659 case XenbusStateInitWait
:
660 xen_pcibk_setup_backend(pdev
);
668 static int xen_pcibk_xenbus_probe(struct xenbus_device
*dev
,
669 const struct xenbus_device_id
*id
)
672 struct xen_pcibk_device
*pdev
= alloc_pdev(dev
);
676 xenbus_dev_fatal(dev
, err
,
677 "Error allocating xen_pcibk_device struct");
681 /* wait for xend to configure us */
682 err
= xenbus_switch_state(dev
, XenbusStateInitWait
);
686 /* watch the backend node for backend configuration information */
687 err
= xenbus_watch_path(dev
, dev
->nodename
, &pdev
->be_watch
,
692 pdev
->be_watching
= 1;
694 /* We need to force a call to our callback here in case
695 * xend already configured us!
697 xen_pcibk_be_watch(&pdev
->be_watch
, NULL
, 0);
703 static int xen_pcibk_xenbus_remove(struct xenbus_device
*dev
)
705 struct xen_pcibk_device
*pdev
= dev_get_drvdata(&dev
->dev
);
713 static const struct xenbus_device_id xenpci_ids
[] = {
718 static struct xenbus_driver xenbus_xen_pcibk_driver
= {
720 .owner
= THIS_MODULE
,
722 .probe
= xen_pcibk_xenbus_probe
,
723 .remove
= xen_pcibk_xenbus_remove
,
724 .otherend_changed
= xen_pcibk_frontend_changed
,
727 struct xen_pcibk_backend
*xen_pcibk_backend
;
729 int __init
xen_pcibk_xenbus_register(void)
731 xen_pcibk_wq
= create_workqueue("xen_pciback_workqueue");
733 printk(KERN_ERR
"%s: create"
734 "xen_pciback_workqueue failed\n", __func__
);
737 xen_pcibk_backend
= &xen_pcibk_vpci_backend
;
739 xen_pcibk_backend
= &xen_pcibk_passthrough_backend
;
740 pr_info(DRV_NAME
": backend is %s\n", xen_pcibk_backend
->name
);
741 return xenbus_register_backend(&xenbus_xen_pcibk_driver
);
744 void __exit
xen_pcibk_xenbus_unregister(void)
746 destroy_workqueue(xen_pcibk_wq
);
747 xenbus_unregister_driver(&xenbus_xen_pcibk_driver
);