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);
48 mutex_init(&pdev->dev_lock);
51 pdev->evtchn_irq = INVALID_EVTCHN_IRQ;
52 pdev->be_watching = 0;
54 INIT_WORK(&pdev->op_work, xen_pcibk_do_op);
56 if (xen_pcibk_init_devices(pdev)) {
61 dev_set_drvdata(&xdev->dev, pdev);
67 static void xen_pcibk_disconnect(struct xen_pcibk_device *pdev)
69 mutex_lock(&pdev->dev_lock);
70 /* Ensure the guest can't trigger our handler before removing devices */
71 if (pdev->evtchn_irq != INVALID_EVTCHN_IRQ) {
72 unbind_from_irqhandler(pdev->evtchn_irq, pdev);
73 pdev->evtchn_irq = INVALID_EVTCHN_IRQ;
76 /* If the driver domain started an op, make sure we complete it
77 * before releasing the shared memory */
79 /* Note, the workqueue does not use spinlocks at all.*/
80 flush_workqueue(xen_pcibk_wq);
82 if (pdev->sh_info != NULL) {
83 xenbus_unmap_ring_vfree(pdev->xdev, pdev->sh_info);
86 mutex_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 /* N.B. This calls pcistub_put_pci_dev which does the FLR on all
99 * of the PCIe devices. */
100 xen_pcibk_release_devices(pdev);
102 dev_set_drvdata(&pdev->xdev->dev, NULL);
108 static int xen_pcibk_do_attach(struct xen_pcibk_device *pdev, int gnt_ref,
114 dev_dbg(&pdev->xdev->dev,
115 "Attaching to frontend resources - gnt_ref=%d evtchn=%d\n",
116 gnt_ref, remote_evtchn);
118 err = xenbus_map_ring_valloc(pdev->xdev, &gnt_ref, 1, &vaddr);
120 xenbus_dev_fatal(pdev->xdev, err,
121 "Error mapping other domain page in ours.");
125 pdev->sh_info = vaddr;
127 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
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");
135 pdev->evtchn_irq = err;
138 dev_dbg(&pdev->xdev->dev, "Attached!\n");
143 static int xen_pcibk_attach(struct xen_pcibk_device *pdev)
146 int gnt_ref, remote_evtchn;
150 mutex_lock(&pdev->dev_lock);
151 /* Make sure we only do this setup once */
152 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
153 XenbusStateInitialised)
156 /* Wait for frontend to state that it has published the configuration */
157 if (xenbus_read_driver_state(pdev->xdev->otherend) !=
158 XenbusStateInitialised)
161 dev_dbg(&pdev->xdev->dev, "Reading frontend config\n");
163 err = xenbus_gather(XBT_NIL, pdev->xdev->otherend,
164 "pci-op-ref", "%u", &gnt_ref,
165 "event-channel", "%u", &remote_evtchn,
166 "magic", NULL, &magic, NULL);
168 /* If configuration didn't get read correctly, wait longer */
169 xenbus_dev_fatal(pdev->xdev, err,
170 "Error reading configuration from frontend");
174 if (magic == NULL || strcmp(magic, XEN_PCI_MAGIC) != 0) {
175 xenbus_dev_fatal(pdev->xdev, -EFAULT,
176 "version mismatch (%s/%s) with pcifront - "
178 magic, XEN_PCI_MAGIC);
183 err = xen_pcibk_do_attach(pdev, gnt_ref, remote_evtchn);
187 dev_dbg(&pdev->xdev->dev, "Connecting...\n");
189 err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
191 xenbus_dev_fatal(pdev->xdev, err,
192 "Error switching to connected state!");
194 dev_dbg(&pdev->xdev->dev, "Connected? %d\n", err);
196 mutex_unlock(&pdev->dev_lock);
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 /* Note: The PV protocol uses %02x, don't change it */
218 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename, str,
219 "%04x:%02x:%02x.%02x", domain, bus,
220 PCI_SLOT(devfn), PCI_FUNC(devfn));
226 static int xen_pcibk_export_device(struct xen_pcibk_device *pdev,
227 int domain, int bus, int slot, int func,
233 dev_dbg(&pdev->xdev->dev, "exporting dom %x bus %x slot %x func %x\n",
234 domain, bus, slot, func);
236 dev = pcistub_get_pci_dev_by_slot(pdev, domain, bus, slot, func);
239 xenbus_dev_fatal(pdev->xdev, err,
240 "Couldn't locate PCI device "
241 "(%04x:%02x:%02x.%d)! "
242 "perhaps already in-use?",
243 domain, bus, slot, func);
247 err = xen_pcibk_add_pci_dev(pdev, dev, devid,
248 xen_pcibk_publish_pci_dev);
252 dev_info(&dev->dev, "registering for %d\n", pdev->xdev->otherend_id);
253 if (xen_register_device_domain_owner(dev,
254 pdev->xdev->otherend_id) != 0) {
255 dev_err(&dev->dev, "Stealing ownership from dom%d.\n",
256 xen_find_device_domain_owner(dev));
257 xen_unregister_device_domain_owner(dev);
258 xen_register_device_domain_owner(dev, pdev->xdev->otherend_id);
261 /* TODO: It'd be nice to export a bridge and have all of its children
262 * get exported with it. This may be best done in xend (which will
263 * have to calculate resource usage anyway) but we probably want to
264 * put something in here to ensure that if a bridge gets given to a
265 * driver domain, that all devices under that bridge are not given
266 * to other driver domains (as he who controls the bridge can disable
267 * it and stop the other devices from working).
273 static int xen_pcibk_remove_device(struct xen_pcibk_device *pdev,
274 int domain, int bus, int slot, int func)
279 dev_dbg(&pdev->xdev->dev, "removing dom %x bus %x slot %x func %x\n",
280 domain, bus, slot, func);
282 dev = xen_pcibk_get_pci_dev(pdev, domain, bus, PCI_DEVFN(slot, func));
285 dev_dbg(&pdev->xdev->dev, "Couldn't locate PCI device "
286 "(%04x:%02x:%02x.%d)! not owned by this domain\n",
287 domain, bus, slot, func);
291 dev_dbg(&dev->dev, "unregistering for %d\n", pdev->xdev->otherend_id);
292 xen_unregister_device_domain_owner(dev);
294 /* N.B. This ends up calling pcistub_put_pci_dev which ends up
296 xen_pcibk_release_pci_dev(pdev, dev, true /* use the lock. */);
302 static int xen_pcibk_publish_pci_root(struct xen_pcibk_device *pdev,
303 unsigned int domain, unsigned int bus)
306 int i, root_num, len, err;
309 dev_dbg(&pdev->xdev->dev, "Publishing pci roots\n");
311 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename,
312 "root_num", "%d", &root_num);
313 if (err == 0 || err == -ENOENT)
318 /* Verify that we haven't already published this pci root */
319 for (i = 0; i < root_num; i++) {
320 len = snprintf(str, sizeof(str), "root-%d", i);
321 if (unlikely(len >= (sizeof(str) - 1))) {
326 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename,
327 str, "%x:%x", &d, &b);
335 if (d == domain && b == bus) {
341 len = snprintf(str, sizeof(str), "root-%d", root_num);
342 if (unlikely(len >= (sizeof(str) - 1))) {
347 dev_dbg(&pdev->xdev->dev, "writing root %d at %04x:%02x\n",
348 root_num, domain, bus);
350 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename, str,
351 "%04x:%02x", domain, bus);
355 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename,
356 "root_num", "%d", (root_num + 1));
362 static int xen_pcibk_reconfigure(struct xen_pcibk_device *pdev,
363 enum xenbus_state state)
367 int domain, bus, slot, func;
374 dev_dbg(&pdev->xdev->dev, "Reconfiguring device ...\n");
376 mutex_lock(&pdev->dev_lock);
377 if (xenbus_read_driver_state(pdev->xdev->nodename) != state)
380 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename, "num_devs", "%d",
385 xenbus_dev_fatal(pdev->xdev, err,
386 "Error reading number of devices");
390 for (i = 0; i < num_devs; i++) {
391 len = snprintf(state_str, sizeof(state_str), "state-%d", i);
392 if (unlikely(len >= (sizeof(state_str) - 1))) {
394 xenbus_dev_fatal(pdev->xdev, err,
395 "String overflow while reading "
399 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename, state_str,
402 substate = XenbusStateUnknown;
405 case XenbusStateInitialising:
406 dev_dbg(&pdev->xdev->dev, "Attaching dev-%d ...\n", i);
408 len = snprintf(dev_str, sizeof(dev_str), "dev-%d", i);
409 if (unlikely(len >= (sizeof(dev_str) - 1))) {
411 xenbus_dev_fatal(pdev->xdev, err,
412 "String overflow while "
413 "reading configuration");
416 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename,
417 dev_str, "%x:%x:%x.%x",
418 &domain, &bus, &slot, &func);
420 xenbus_dev_fatal(pdev->xdev, err,
421 "Error reading device "
427 xenbus_dev_fatal(pdev->xdev, err,
428 "Error parsing pci device "
433 err = xen_pcibk_export_device(pdev, domain, bus, slot,
438 /* Publish pci roots. */
439 err = xen_pcibk_publish_pci_roots(pdev,
440 xen_pcibk_publish_pci_root);
442 xenbus_dev_fatal(pdev->xdev, err,
443 "Error while publish PCI root"
444 "buses for frontend");
448 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename,
450 XenbusStateInitialised);
452 xenbus_dev_fatal(pdev->xdev, err,
453 "Error switching substate of "
459 case XenbusStateClosing:
460 dev_dbg(&pdev->xdev->dev, "Detaching dev-%d ...\n", i);
462 len = snprintf(dev_str, sizeof(dev_str), "vdev-%d", i);
463 if (unlikely(len >= (sizeof(dev_str) - 1))) {
465 xenbus_dev_fatal(pdev->xdev, err,
466 "String overflow while "
467 "reading configuration");
470 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename,
471 dev_str, "%x:%x:%x.%x",
472 &domain, &bus, &slot, &func);
474 xenbus_dev_fatal(pdev->xdev, err,
475 "Error reading device "
481 xenbus_dev_fatal(pdev->xdev, err,
482 "Error parsing pci device "
487 err = xen_pcibk_remove_device(pdev, domain, bus, slot,
492 /* TODO: If at some point we implement support for pci
493 * root hot-remove on pcifront side, we'll need to
494 * remove unnecessary xenstore nodes of pci roots here.
504 if (state != XenbusStateReconfiguring)
505 /* Make sure we only reconfigure once. */
508 err = xenbus_switch_state(pdev->xdev, XenbusStateReconfigured);
510 xenbus_dev_fatal(pdev->xdev, err,
511 "Error switching to reconfigured state!");
516 mutex_unlock(&pdev->dev_lock);
520 static void xen_pcibk_frontend_changed(struct xenbus_device *xdev,
521 enum xenbus_state fe_state)
523 struct xen_pcibk_device *pdev = dev_get_drvdata(&xdev->dev);
525 dev_dbg(&xdev->dev, "fe state changed %d\n", fe_state);
528 case XenbusStateInitialised:
529 xen_pcibk_attach(pdev);
532 case XenbusStateReconfiguring:
533 xen_pcibk_reconfigure(pdev, XenbusStateReconfiguring);
536 case XenbusStateConnected:
537 /* pcifront switched its state from reconfiguring to connected.
538 * Then switch to connected state.
540 xenbus_switch_state(xdev, XenbusStateConnected);
543 case XenbusStateClosing:
544 xen_pcibk_disconnect(pdev);
545 xenbus_switch_state(xdev, XenbusStateClosing);
548 case XenbusStateClosed:
549 xen_pcibk_disconnect(pdev);
550 xenbus_switch_state(xdev, XenbusStateClosed);
551 if (xenbus_dev_is_online(xdev))
553 /* fall through if not online */
554 case XenbusStateUnknown:
555 dev_dbg(&xdev->dev, "frontend is gone! unregister device\n");
556 device_unregister(&xdev->dev);
564 static int xen_pcibk_setup_backend(struct xen_pcibk_device *pdev)
566 /* Get configuration from xend (if available now) */
567 int domain, bus, slot, func;
573 mutex_lock(&pdev->dev_lock);
574 /* It's possible we could get the call to setup twice, so make sure
575 * we're not already connected.
577 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
581 dev_dbg(&pdev->xdev->dev, "getting be setup\n");
583 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename, "num_devs", "%d",
588 xenbus_dev_fatal(pdev->xdev, err,
589 "Error reading number of devices");
593 for (i = 0; i < num_devs; i++) {
594 int l = snprintf(dev_str, sizeof(dev_str), "dev-%d", i);
595 if (unlikely(l >= (sizeof(dev_str) - 1))) {
597 xenbus_dev_fatal(pdev->xdev, err,
598 "String overflow while reading "
603 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename, dev_str,
604 "%x:%x:%x.%x", &domain, &bus, &slot, &func);
606 xenbus_dev_fatal(pdev->xdev, err,
607 "Error reading device configuration");
612 xenbus_dev_fatal(pdev->xdev, err,
613 "Error parsing pci device "
618 err = xen_pcibk_export_device(pdev, domain, bus, slot, func, i);
622 /* Switch substate of this device. */
623 l = snprintf(state_str, sizeof(state_str), "state-%d", i);
624 if (unlikely(l >= (sizeof(state_str) - 1))) {
626 xenbus_dev_fatal(pdev->xdev, err,
627 "String overflow while reading "
631 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename, state_str,
632 "%d", XenbusStateInitialised);
634 xenbus_dev_fatal(pdev->xdev, err, "Error switching "
635 "substate of dev-%d\n", i);
640 err = xen_pcibk_publish_pci_roots(pdev, xen_pcibk_publish_pci_root);
642 xenbus_dev_fatal(pdev->xdev, err,
643 "Error while publish PCI root buses "
648 err = xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
650 xenbus_dev_fatal(pdev->xdev, err,
651 "Error switching to initialised state!");
654 mutex_unlock(&pdev->dev_lock);
656 /* see if pcifront is already configured (if not, we'll wait) */
657 xen_pcibk_attach(pdev);
661 static void xen_pcibk_be_watch(struct xenbus_watch *watch,
662 const char **vec, unsigned int len)
664 struct xen_pcibk_device *pdev =
665 container_of(watch, struct xen_pcibk_device, be_watch);
667 switch (xenbus_read_driver_state(pdev->xdev->nodename)) {
668 case XenbusStateInitWait:
669 xen_pcibk_setup_backend(pdev);
672 case XenbusStateInitialised:
674 * We typically move to Initialised when the first device was
675 * added. Hence subsequent devices getting added may need
678 xen_pcibk_reconfigure(pdev, XenbusStateInitialised);
686 static int xen_pcibk_xenbus_probe(struct xenbus_device *dev,
687 const struct xenbus_device_id *id)
690 struct xen_pcibk_device *pdev = alloc_pdev(dev);
694 xenbus_dev_fatal(dev, err,
695 "Error allocating xen_pcibk_device struct");
699 /* wait for xend to configure us */
700 err = xenbus_switch_state(dev, XenbusStateInitWait);
704 /* watch the backend node for backend configuration information */
705 err = xenbus_watch_path(dev, dev->nodename, &pdev->be_watch,
706 NULL, xen_pcibk_be_watch);
710 pdev->be_watching = 1;
712 /* We need to force a call to our callback here in case
713 * xend already configured us!
715 xen_pcibk_be_watch(&pdev->be_watch, NULL, 0);
721 static int xen_pcibk_xenbus_remove(struct xenbus_device *dev)
723 struct xen_pcibk_device *pdev = dev_get_drvdata(&dev->dev);
731 static const struct xenbus_device_id xen_pcibk_ids[] = {
736 static struct xenbus_driver xen_pcibk_driver = {
738 .ids = xen_pcibk_ids,
739 .probe = xen_pcibk_xenbus_probe,
740 .remove = xen_pcibk_xenbus_remove,
741 .otherend_changed = xen_pcibk_frontend_changed,
744 const struct xen_pcibk_backend *__read_mostly xen_pcibk_backend;
746 int __init xen_pcibk_xenbus_register(void)
748 xen_pcibk_wq = create_workqueue("xen_pciback_workqueue");
750 pr_err("%s: create xen_pciback_workqueue failed\n", __func__);
753 xen_pcibk_backend = &xen_pcibk_vpci_backend;
755 xen_pcibk_backend = &xen_pcibk_passthrough_backend;
756 pr_info("backend is %s\n", xen_pcibk_backend->name);
757 return xenbus_register_backend(&xen_pcibk_driver);
760 void __exit xen_pcibk_xenbus_unregister(void)
762 destroy_workqueue(xen_pcibk_wq);
763 xenbus_unregister_driver(&xen_pcibk_driver);