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 /* N.B. This calls pcistub_put_pci_dev which does the FLR on all
97 * of the PCIe devices. */
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, 1, &vaddr);
118 xenbus_dev_fatal(pdev->xdev, err,
119 "Error mapping other domain page in ours.");
123 pdev->sh_info = vaddr;
125 err = bind_interdomain_evtchn_to_irqhandler(
126 pdev->xdev->otherend_id, remote_evtchn, xen_pcibk_handle_event,
129 xenbus_dev_fatal(pdev->xdev, err,
130 "Error binding event channel to IRQ");
133 pdev->evtchn_irq = err;
136 dev_dbg(&pdev->xdev->dev, "Attached!\n");
141 static int xen_pcibk_attach(struct xen_pcibk_device *pdev)
144 int gnt_ref, remote_evtchn;
148 mutex_lock(&pdev->dev_lock);
149 /* Make sure we only do this setup once */
150 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
151 XenbusStateInitialised)
154 /* Wait for frontend to state that it has published the configuration */
155 if (xenbus_read_driver_state(pdev->xdev->otherend) !=
156 XenbusStateInitialised)
159 dev_dbg(&pdev->xdev->dev, "Reading frontend config\n");
161 err = xenbus_gather(XBT_NIL, pdev->xdev->otherend,
162 "pci-op-ref", "%u", &gnt_ref,
163 "event-channel", "%u", &remote_evtchn,
164 "magic", NULL, &magic, NULL);
166 /* If configuration didn't get read correctly, wait longer */
167 xenbus_dev_fatal(pdev->xdev, err,
168 "Error reading configuration from frontend");
172 if (magic == NULL || strcmp(magic, XEN_PCI_MAGIC) != 0) {
173 xenbus_dev_fatal(pdev->xdev, -EFAULT,
174 "version mismatch (%s/%s) with pcifront - "
176 magic, XEN_PCI_MAGIC);
181 err = xen_pcibk_do_attach(pdev, gnt_ref, remote_evtchn);
185 dev_dbg(&pdev->xdev->dev, "Connecting...\n");
187 err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
189 xenbus_dev_fatal(pdev->xdev, err,
190 "Error switching to connected state!");
192 dev_dbg(&pdev->xdev->dev, "Connected? %d\n", err);
194 mutex_unlock(&pdev->dev_lock);
201 static int xen_pcibk_publish_pci_dev(struct xen_pcibk_device *pdev,
202 unsigned int domain, unsigned int bus,
203 unsigned int devfn, unsigned int devid)
209 len = snprintf(str, sizeof(str), "vdev-%d", devid);
210 if (unlikely(len >= (sizeof(str) - 1))) {
215 /* Note: The PV protocol uses %02x, don't change it */
216 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename, str,
217 "%04x:%02x:%02x.%02x", domain, bus,
218 PCI_SLOT(devfn), PCI_FUNC(devfn));
224 static int xen_pcibk_export_device(struct xen_pcibk_device *pdev,
225 int domain, int bus, int slot, int func,
231 dev_dbg(&pdev->xdev->dev, "exporting dom %x bus %x slot %x func %x\n",
232 domain, bus, slot, func);
234 dev = pcistub_get_pci_dev_by_slot(pdev, domain, bus, slot, func);
237 xenbus_dev_fatal(pdev->xdev, err,
238 "Couldn't locate PCI device "
239 "(%04x:%02x:%02x.%d)! "
240 "perhaps already in-use?",
241 domain, bus, slot, func);
245 err = xen_pcibk_add_pci_dev(pdev, dev, devid,
246 xen_pcibk_publish_pci_dev);
250 dev_info(&dev->dev, "registering for %d\n", pdev->xdev->otherend_id);
251 if (xen_register_device_domain_owner(dev,
252 pdev->xdev->otherend_id) != 0) {
253 dev_err(&dev->dev, "Stealing ownership from dom%d.\n",
254 xen_find_device_domain_owner(dev));
255 xen_unregister_device_domain_owner(dev);
256 xen_register_device_domain_owner(dev, pdev->xdev->otherend_id);
259 /* TODO: It'd be nice to export a bridge and have all of its children
260 * get exported with it. This may be best done in xend (which will
261 * have to calculate resource usage anyway) but we probably want to
262 * put something in here to ensure that if a bridge gets given to a
263 * driver domain, that all devices under that bridge are not given
264 * to other driver domains (as he who controls the bridge can disable
265 * it and stop the other devices from working).
271 static int xen_pcibk_remove_device(struct xen_pcibk_device *pdev,
272 int domain, int bus, int slot, int func)
277 dev_dbg(&pdev->xdev->dev, "removing dom %x bus %x slot %x func %x\n",
278 domain, bus, slot, func);
280 dev = xen_pcibk_get_pci_dev(pdev, domain, bus, PCI_DEVFN(slot, func));
283 dev_dbg(&pdev->xdev->dev, "Couldn't locate PCI device "
284 "(%04x:%02x:%02x.%d)! not owned by this domain\n",
285 domain, bus, slot, func);
289 dev_dbg(&dev->dev, "unregistering for %d\n", pdev->xdev->otherend_id);
290 xen_unregister_device_domain_owner(dev);
292 /* N.B. This ends up calling pcistub_put_pci_dev which ends up
294 xen_pcibk_release_pci_dev(pdev, dev, true /* use the lock. */);
300 static int xen_pcibk_publish_pci_root(struct xen_pcibk_device *pdev,
301 unsigned int domain, unsigned int bus)
304 int i, root_num, len, err;
307 dev_dbg(&pdev->xdev->dev, "Publishing pci roots\n");
309 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename,
310 "root_num", "%d", &root_num);
311 if (err == 0 || err == -ENOENT)
316 /* Verify that we haven't already published this pci root */
317 for (i = 0; i < root_num; i++) {
318 len = snprintf(str, sizeof(str), "root-%d", i);
319 if (unlikely(len >= (sizeof(str) - 1))) {
324 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename,
325 str, "%x:%x", &d, &b);
333 if (d == domain && b == bus) {
339 len = snprintf(str, sizeof(str), "root-%d", root_num);
340 if (unlikely(len >= (sizeof(str) - 1))) {
345 dev_dbg(&pdev->xdev->dev, "writing root %d at %04x:%02x\n",
346 root_num, domain, bus);
348 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename, str,
349 "%04x:%02x", domain, bus);
353 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename,
354 "root_num", "%d", (root_num + 1));
360 static int xen_pcibk_reconfigure(struct xen_pcibk_device *pdev)
364 int domain, bus, slot, func;
371 dev_dbg(&pdev->xdev->dev, "Reconfiguring device ...\n");
373 mutex_lock(&pdev->dev_lock);
374 /* Make sure we only reconfigure once */
375 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
376 XenbusStateReconfiguring)
379 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename, "num_devs", "%d",
384 xenbus_dev_fatal(pdev->xdev, err,
385 "Error reading number of devices");
389 for (i = 0; i < num_devs; i++) {
390 len = snprintf(state_str, sizeof(state_str), "state-%d", i);
391 if (unlikely(len >= (sizeof(state_str) - 1))) {
393 xenbus_dev_fatal(pdev->xdev, err,
394 "String overflow while reading "
398 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename, state_str,
401 substate = XenbusStateUnknown;
404 case XenbusStateInitialising:
405 dev_dbg(&pdev->xdev->dev, "Attaching dev-%d ...\n", i);
407 len = snprintf(dev_str, sizeof(dev_str), "dev-%d", i);
408 if (unlikely(len >= (sizeof(dev_str) - 1))) {
410 xenbus_dev_fatal(pdev->xdev, err,
411 "String overflow while "
412 "reading configuration");
415 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename,
416 dev_str, "%x:%x:%x.%x",
417 &domain, &bus, &slot, &func);
419 xenbus_dev_fatal(pdev->xdev, err,
420 "Error reading device "
426 xenbus_dev_fatal(pdev->xdev, err,
427 "Error parsing pci device "
432 err = xen_pcibk_export_device(pdev, domain, bus, slot,
437 /* Publish pci roots. */
438 err = xen_pcibk_publish_pci_roots(pdev,
439 xen_pcibk_publish_pci_root);
441 xenbus_dev_fatal(pdev->xdev, err,
442 "Error while publish PCI root"
443 "buses for frontend");
447 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename,
449 XenbusStateInitialised);
451 xenbus_dev_fatal(pdev->xdev, err,
452 "Error switching substate of "
458 case XenbusStateClosing:
459 dev_dbg(&pdev->xdev->dev, "Detaching dev-%d ...\n", i);
461 len = snprintf(dev_str, sizeof(dev_str), "vdev-%d", i);
462 if (unlikely(len >= (sizeof(dev_str) - 1))) {
464 xenbus_dev_fatal(pdev->xdev, err,
465 "String overflow while "
466 "reading configuration");
469 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename,
470 dev_str, "%x:%x:%x.%x",
471 &domain, &bus, &slot, &func);
473 xenbus_dev_fatal(pdev->xdev, err,
474 "Error reading device "
480 xenbus_dev_fatal(pdev->xdev, err,
481 "Error parsing pci device "
486 err = xen_pcibk_remove_device(pdev, domain, bus, slot,
491 /* TODO: If at some point we implement support for pci
492 * root hot-remove on pcifront side, we'll need to
493 * remove unnecessary xenstore nodes of pci roots here.
503 err = xenbus_switch_state(pdev->xdev, XenbusStateReconfigured);
505 xenbus_dev_fatal(pdev->xdev, err,
506 "Error switching to reconfigured state!");
511 mutex_unlock(&pdev->dev_lock);
515 static void xen_pcibk_frontend_changed(struct xenbus_device *xdev,
516 enum xenbus_state fe_state)
518 struct xen_pcibk_device *pdev = dev_get_drvdata(&xdev->dev);
520 dev_dbg(&xdev->dev, "fe state changed %d\n", fe_state);
523 case XenbusStateInitialised:
524 xen_pcibk_attach(pdev);
527 case XenbusStateReconfiguring:
528 xen_pcibk_reconfigure(pdev);
531 case XenbusStateConnected:
532 /* pcifront switched its state from reconfiguring to connected.
533 * Then switch to connected state.
535 xenbus_switch_state(xdev, XenbusStateConnected);
538 case XenbusStateClosing:
539 xen_pcibk_disconnect(pdev);
540 xenbus_switch_state(xdev, XenbusStateClosing);
543 case XenbusStateClosed:
544 xen_pcibk_disconnect(pdev);
545 xenbus_switch_state(xdev, XenbusStateClosed);
546 if (xenbus_dev_is_online(xdev))
548 /* fall through if not online */
549 case XenbusStateUnknown:
550 dev_dbg(&xdev->dev, "frontend is gone! unregister device\n");
551 device_unregister(&xdev->dev);
559 static int xen_pcibk_setup_backend(struct xen_pcibk_device *pdev)
561 /* Get configuration from xend (if available now) */
562 int domain, bus, slot, func;
568 mutex_lock(&pdev->dev_lock);
569 /* It's possible we could get the call to setup twice, so make sure
570 * we're not already connected.
572 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
576 dev_dbg(&pdev->xdev->dev, "getting be setup\n");
578 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename, "num_devs", "%d",
583 xenbus_dev_fatal(pdev->xdev, err,
584 "Error reading number of devices");
588 for (i = 0; i < num_devs; i++) {
589 int l = snprintf(dev_str, sizeof(dev_str), "dev-%d", i);
590 if (unlikely(l >= (sizeof(dev_str) - 1))) {
592 xenbus_dev_fatal(pdev->xdev, err,
593 "String overflow while reading "
598 err = xenbus_scanf(XBT_NIL, pdev->xdev->nodename, dev_str,
599 "%x:%x:%x.%x", &domain, &bus, &slot, &func);
601 xenbus_dev_fatal(pdev->xdev, err,
602 "Error reading device configuration");
607 xenbus_dev_fatal(pdev->xdev, err,
608 "Error parsing pci device "
613 err = xen_pcibk_export_device(pdev, domain, bus, slot, func, i);
617 /* Switch substate of this device. */
618 l = snprintf(state_str, sizeof(state_str), "state-%d", i);
619 if (unlikely(l >= (sizeof(state_str) - 1))) {
621 xenbus_dev_fatal(pdev->xdev, err,
622 "String overflow while reading "
626 err = xenbus_printf(XBT_NIL, pdev->xdev->nodename, state_str,
627 "%d", XenbusStateInitialised);
629 xenbus_dev_fatal(pdev->xdev, err, "Error switching "
630 "substate of dev-%d\n", i);
635 err = xen_pcibk_publish_pci_roots(pdev, xen_pcibk_publish_pci_root);
637 xenbus_dev_fatal(pdev->xdev, err,
638 "Error while publish PCI root buses "
643 err = xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
645 xenbus_dev_fatal(pdev->xdev, err,
646 "Error switching to initialised state!");
649 mutex_unlock(&pdev->dev_lock);
651 /* see if pcifront is already configured (if not, we'll wait) */
652 xen_pcibk_attach(pdev);
656 static void xen_pcibk_be_watch(struct xenbus_watch *watch,
657 const char **vec, unsigned int len)
659 struct xen_pcibk_device *pdev =
660 container_of(watch, struct xen_pcibk_device, be_watch);
662 switch (xenbus_read_driver_state(pdev->xdev->nodename)) {
663 case XenbusStateInitWait:
664 xen_pcibk_setup_backend(pdev);
672 static int xen_pcibk_xenbus_probe(struct xenbus_device *dev,
673 const struct xenbus_device_id *id)
676 struct xen_pcibk_device *pdev = alloc_pdev(dev);
680 xenbus_dev_fatal(dev, err,
681 "Error allocating xen_pcibk_device struct");
685 /* wait for xend to configure us */
686 err = xenbus_switch_state(dev, XenbusStateInitWait);
690 /* watch the backend node for backend configuration information */
691 err = xenbus_watch_path(dev, dev->nodename, &pdev->be_watch,
696 pdev->be_watching = 1;
698 /* We need to force a call to our callback here in case
699 * xend already configured us!
701 xen_pcibk_be_watch(&pdev->be_watch, NULL, 0);
707 static int xen_pcibk_xenbus_remove(struct xenbus_device *dev)
709 struct xen_pcibk_device *pdev = dev_get_drvdata(&dev->dev);
717 static const struct xenbus_device_id xen_pcibk_ids[] = {
722 static struct xenbus_driver xen_pcibk_driver = {
724 .ids = xen_pcibk_ids,
725 .probe = xen_pcibk_xenbus_probe,
726 .remove = xen_pcibk_xenbus_remove,
727 .otherend_changed = xen_pcibk_frontend_changed,
730 const struct xen_pcibk_backend *__read_mostly xen_pcibk_backend;
732 int __init xen_pcibk_xenbus_register(void)
734 xen_pcibk_wq = create_workqueue("xen_pciback_workqueue");
736 pr_err("%s: create xen_pciback_workqueue failed\n", __func__);
739 xen_pcibk_backend = &xen_pcibk_vpci_backend;
741 xen_pcibk_backend = &xen_pcibk_passthrough_backend;
742 pr_info("backend is %s\n", xen_pcibk_backend->name);
743 return xenbus_register_backend(&xen_pcibk_driver);
746 void __exit xen_pcibk_xenbus_unregister(void)
748 destroy_workqueue(xen_pcibk_wq);
749 xenbus_unregister_driver(&xen_pcibk_driver);