Commit | Line | Data |
---|---|---|
283c0972 JP |
1 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
2 | ||
3 | #define DPRINTK(fmt, ...) \ | |
4 | pr_debug("(%s:%d) " fmt "\n", \ | |
5 | __func__, __LINE__, ##__VA_ARGS__) | |
2de06cc1 IC |
6 | |
7 | #include <linux/kernel.h> | |
8 | #include <linux/err.h> | |
9 | #include <linux/string.h> | |
10 | #include <linux/ctype.h> | |
11 | #include <linux/fcntl.h> | |
12 | #include <linux/mm.h> | |
13 | #include <linux/proc_fs.h> | |
14 | #include <linux/notifier.h> | |
15 | #include <linux/kthread.h> | |
16 | #include <linux/mutex.h> | |
17 | #include <linux/io.h> | |
72ee5112 | 18 | #include <linux/module.h> |
2de06cc1 IC |
19 | |
20 | #include <asm/page.h> | |
21 | #include <asm/pgtable.h> | |
22 | #include <asm/xen/hypervisor.h> | |
23 | #include <xen/xenbus.h> | |
24 | #include <xen/events.h> | |
25 | #include <xen/page.h> | |
4d9310e3 | 26 | #include <xen/xen.h> |
2de06cc1 IC |
27 | |
28 | #include <xen/platform_pci.h> | |
29 | ||
30 | #include "xenbus_comms.h" | |
31 | #include "xenbus_probe.h" | |
32 | ||
33 | ||
2abb2746 AC |
34 | static struct workqueue_struct *xenbus_frontend_wq; |
35 | ||
2de06cc1 IC |
36 | /* device/<type>/<id> => <type>-<id> */ |
37 | static int frontend_bus_id(char bus_id[XEN_BUS_ID_SIZE], const char *nodename) | |
38 | { | |
39 | nodename = strchr(nodename, '/'); | |
40 | if (!nodename || strlen(nodename + 1) >= XEN_BUS_ID_SIZE) { | |
283c0972 | 41 | pr_warn("bad frontend %s\n", nodename); |
2de06cc1 IC |
42 | return -EINVAL; |
43 | } | |
44 | ||
45 | strlcpy(bus_id, nodename + 1, XEN_BUS_ID_SIZE); | |
46 | if (!strchr(bus_id, '/')) { | |
283c0972 | 47 | pr_warn("bus_id %s no slash\n", bus_id); |
2de06cc1 IC |
48 | return -EINVAL; |
49 | } | |
50 | *strchr(bus_id, '/') = '-'; | |
51 | return 0; | |
52 | } | |
53 | ||
54 | /* device/<typename>/<name> */ | |
6bac7f9f IC |
55 | static int xenbus_probe_frontend(struct xen_bus_type *bus, const char *type, |
56 | const char *name) | |
2de06cc1 IC |
57 | { |
58 | char *nodename; | |
59 | int err; | |
60 | ||
42c46e6b SS |
61 | /* ignore console/0 */ |
62 | if (!strncmp(type, "console", 7) && !strncmp(name, "0", 1)) { | |
63 | DPRINTK("Ignoring buggy device entry console/0"); | |
64 | return 0; | |
65 | } | |
66 | ||
2de06cc1 IC |
67 | nodename = kasprintf(GFP_KERNEL, "%s/%s/%s", bus->root, type, name); |
68 | if (!nodename) | |
69 | return -ENOMEM; | |
70 | ||
71 | DPRINTK("%s", nodename); | |
72 | ||
73 | err = xenbus_probe_node(bus, type, nodename); | |
74 | kfree(nodename); | |
75 | return err; | |
76 | } | |
77 | ||
6bac7f9f IC |
78 | static int xenbus_uevent_frontend(struct device *_dev, |
79 | struct kobj_uevent_env *env) | |
df660251 IC |
80 | { |
81 | struct xenbus_device *dev = to_xenbus_device(_dev); | |
82 | ||
83 | if (add_uevent_var(env, "MODALIAS=xen:%s", dev->devicetype)) | |
84 | return -ENOMEM; | |
85 | ||
86 | return 0; | |
87 | } | |
88 | ||
89 | ||
2de06cc1 IC |
90 | static void backend_changed(struct xenbus_watch *watch, |
91 | const char **vec, unsigned int len) | |
92 | { | |
93 | xenbus_otherend_changed(watch, vec, len, 1); | |
94 | } | |
95 | ||
2abb2746 AC |
96 | static void xenbus_frontend_delayed_resume(struct work_struct *w) |
97 | { | |
98 | struct xenbus_device *xdev = container_of(w, struct xenbus_device, work); | |
99 | ||
100 | xenbus_dev_resume(&xdev->dev); | |
101 | } | |
102 | ||
103 | static int xenbus_frontend_dev_resume(struct device *dev) | |
104 | { | |
105 | /* | |
106 | * If xenstored is running in this domain, we cannot access the backend | |
107 | * state at the moment, so we need to defer xenbus_dev_resume | |
108 | */ | |
109 | if (xen_store_domain_type == XS_LOCAL) { | |
110 | struct xenbus_device *xdev = to_xenbus_device(dev); | |
111 | ||
112 | if (!xenbus_frontend_wq) { | |
113 | pr_err("%s: no workqueue to process delayed resume\n", | |
114 | xdev->nodename); | |
115 | return -EFAULT; | |
116 | } | |
117 | ||
118 | INIT_WORK(&xdev->work, xenbus_frontend_delayed_resume); | |
119 | queue_work(xenbus_frontend_wq, &xdev->work); | |
120 | ||
121 | return 0; | |
122 | } | |
123 | ||
124 | return xenbus_dev_resume(dev); | |
125 | } | |
126 | ||
c7853aea | 127 | static const struct dev_pm_ops xenbus_pm_ops = { |
b3e96c0c | 128 | .suspend = xenbus_dev_suspend, |
2abb2746 | 129 | .resume = xenbus_frontend_dev_resume, |
b3e96c0c SR |
130 | .freeze = xenbus_dev_suspend, |
131 | .thaw = xenbus_dev_cancel, | |
132 | .restore = xenbus_dev_resume, | |
c7853aea KS |
133 | }; |
134 | ||
2de06cc1 IC |
135 | static struct xen_bus_type xenbus_frontend = { |
136 | .root = "device", | |
6bac7f9f | 137 | .levels = 2, /* device/type/<id> */ |
2de06cc1 IC |
138 | .get_bus_id = frontend_bus_id, |
139 | .probe = xenbus_probe_frontend, | |
140 | .otherend_changed = backend_changed, | |
141 | .bus = { | |
6bac7f9f IC |
142 | .name = "xen", |
143 | .match = xenbus_match, | |
144 | .uevent = xenbus_uevent_frontend, | |
145 | .probe = xenbus_dev_probe, | |
146 | .remove = xenbus_dev_remove, | |
147 | .shutdown = xenbus_dev_shutdown, | |
cc85e933 | 148 | .dev_attrs = xenbus_dev_attrs, |
6bac7f9f | 149 | |
c7853aea | 150 | .pm = &xenbus_pm_ops, |
2de06cc1 IC |
151 | }, |
152 | }; | |
153 | ||
154 | static void frontend_changed(struct xenbus_watch *watch, | |
155 | const char **vec, unsigned int len) | |
156 | { | |
157 | DPRINTK(""); | |
158 | ||
159 | xenbus_dev_changed(vec[XS_WATCH_PATH], &xenbus_frontend); | |
160 | } | |
161 | ||
162 | ||
163 | /* We watch for devices appearing and vanishing. */ | |
164 | static struct xenbus_watch fe_watch = { | |
165 | .node = "device", | |
166 | .callback = frontend_changed, | |
167 | }; | |
168 | ||
169 | static int read_backend_details(struct xenbus_device *xendev) | |
170 | { | |
171 | return xenbus_read_otherend_details(xendev, "backend-id", "backend"); | |
172 | } | |
173 | ||
3066616c | 174 | static int is_device_connecting(struct device *dev, void *data, bool ignore_nonessential) |
2de06cc1 IC |
175 | { |
176 | struct xenbus_device *xendev = to_xenbus_device(dev); | |
177 | struct device_driver *drv = data; | |
178 | struct xenbus_driver *xendrv; | |
179 | ||
180 | /* | |
181 | * A device with no driver will never connect. We care only about | |
182 | * devices which should currently be in the process of connecting. | |
183 | */ | |
184 | if (!dev->driver) | |
185 | return 0; | |
186 | ||
187 | /* Is this search limited to a particular driver? */ | |
188 | if (drv && (dev->driver != drv)) | |
189 | return 0; | |
190 | ||
3066616c KRW |
191 | if (ignore_nonessential) { |
192 | /* With older QEMU, for PVonHVM guests the guest config files | |
193 | * could contain: vfb = [ 'vnc=1, vnclisten=0.0.0.0'] | |
194 | * which is nonsensical as there is no PV FB (there can be | |
195 | * a PVKB) running as HVM guest. */ | |
196 | ||
197 | if ((strncmp(xendev->nodename, "device/vkbd", 11) == 0)) | |
198 | return 0; | |
199 | ||
200 | if ((strncmp(xendev->nodename, "device/vfb", 10) == 0)) | |
201 | return 0; | |
202 | } | |
2de06cc1 IC |
203 | xendrv = to_xenbus_driver(dev->driver); |
204 | return (xendev->state < XenbusStateConnected || | |
205 | (xendev->state == XenbusStateConnected && | |
206 | xendrv->is_ready && !xendrv->is_ready(xendev))); | |
207 | } | |
3066616c KRW |
208 | static int essential_device_connecting(struct device *dev, void *data) |
209 | { | |
210 | return is_device_connecting(dev, data, true /* ignore PV[KBB+FB] */); | |
211 | } | |
212 | static int non_essential_device_connecting(struct device *dev, void *data) | |
213 | { | |
214 | return is_device_connecting(dev, data, false); | |
215 | } | |
2de06cc1 | 216 | |
3066616c | 217 | static int exists_essential_connecting_device(struct device_driver *drv) |
2de06cc1 IC |
218 | { |
219 | return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
3066616c KRW |
220 | essential_device_connecting); |
221 | } | |
222 | static int exists_non_essential_connecting_device(struct device_driver *drv) | |
223 | { | |
224 | return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
225 | non_essential_device_connecting); | |
2de06cc1 IC |
226 | } |
227 | ||
228 | static int print_device_status(struct device *dev, void *data) | |
229 | { | |
230 | struct xenbus_device *xendev = to_xenbus_device(dev); | |
231 | struct device_driver *drv = data; | |
232 | ||
233 | /* Is this operation limited to a particular driver? */ | |
234 | if (drv && (dev->driver != drv)) | |
235 | return 0; | |
236 | ||
237 | if (!dev->driver) { | |
238 | /* Information only: is this too noisy? */ | |
283c0972 | 239 | pr_info("Device with no driver: %s\n", xendev->nodename); |
2de06cc1 IC |
240 | } else if (xendev->state < XenbusStateConnected) { |
241 | enum xenbus_state rstate = XenbusStateUnknown; | |
242 | if (xendev->otherend) | |
243 | rstate = xenbus_read_driver_state(xendev->otherend); | |
283c0972 JP |
244 | pr_warn("Timeout connecting to device: %s (local state %d, remote state %d)\n", |
245 | xendev->nodename, xendev->state, rstate); | |
2de06cc1 IC |
246 | } |
247 | ||
248 | return 0; | |
249 | } | |
250 | ||
251 | /* We only wait for device setup after most initcalls have run. */ | |
252 | static int ready_to_wait_for_devices; | |
253 | ||
3066616c KRW |
254 | static bool wait_loop(unsigned long start, unsigned int max_delay, |
255 | unsigned int *seconds_waited) | |
256 | { | |
257 | if (time_after(jiffies, start + (*seconds_waited+5)*HZ)) { | |
258 | if (!*seconds_waited) | |
283c0972 | 259 | pr_warn("Waiting for devices to initialise: "); |
3066616c | 260 | *seconds_waited += 5; |
283c0972 JP |
261 | pr_cont("%us...", max_delay - *seconds_waited); |
262 | if (*seconds_waited == max_delay) { | |
263 | pr_cont("\n"); | |
3066616c | 264 | return true; |
283c0972 | 265 | } |
3066616c KRW |
266 | } |
267 | ||
268 | schedule_timeout_interruptible(HZ/10); | |
269 | ||
270 | return false; | |
271 | } | |
2de06cc1 IC |
272 | /* |
273 | * On a 5-minute timeout, wait for all devices currently configured. We need | |
274 | * to do this to guarantee that the filesystems and / or network devices | |
275 | * needed for boot are available, before we can allow the boot to proceed. | |
276 | * | |
277 | * This needs to be on a late_initcall, to happen after the frontend device | |
278 | * drivers have been initialised, but before the root fs is mounted. | |
279 | * | |
280 | * A possible improvement here would be to have the tools add a per-device | |
281 | * flag to the store entry, indicating whether it is needed at boot time. | |
282 | * This would allow people who knew what they were doing to accelerate their | |
283 | * boot slightly, but of course needs tools or manual intervention to set up | |
284 | * those flags correctly. | |
285 | */ | |
286 | static void wait_for_devices(struct xenbus_driver *xendrv) | |
287 | { | |
288 | unsigned long start = jiffies; | |
289 | struct device_driver *drv = xendrv ? &xendrv->driver : NULL; | |
290 | unsigned int seconds_waited = 0; | |
291 | ||
292 | if (!ready_to_wait_for_devices || !xen_domain()) | |
293 | return; | |
294 | ||
3066616c KRW |
295 | while (exists_non_essential_connecting_device(drv)) |
296 | if (wait_loop(start, 30, &seconds_waited)) | |
297 | break; | |
298 | ||
299 | /* Skips PVKB and PVFB check.*/ | |
300 | while (exists_essential_connecting_device(drv)) | |
301 | if (wait_loop(start, 270, &seconds_waited)) | |
302 | break; | |
2de06cc1 IC |
303 | |
304 | if (seconds_waited) | |
305 | printk("\n"); | |
306 | ||
307 | bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
308 | print_device_status); | |
309 | } | |
310 | ||
73db144b | 311 | int xenbus_register_frontend(struct xenbus_driver *drv) |
2de06cc1 IC |
312 | { |
313 | int ret; | |
314 | ||
315 | drv->read_otherend_details = read_backend_details; | |
316 | ||
73db144b | 317 | ret = xenbus_register_driver_common(drv, &xenbus_frontend); |
2de06cc1 IC |
318 | if (ret) |
319 | return ret; | |
320 | ||
321 | /* If this driver is loaded as a module wait for devices to attach. */ | |
322 | wait_for_devices(drv); | |
323 | ||
324 | return 0; | |
325 | } | |
73db144b | 326 | EXPORT_SYMBOL_GPL(xenbus_register_frontend); |
2de06cc1 | 327 | |
116df6f0 OH |
328 | static DECLARE_WAIT_QUEUE_HEAD(backend_state_wq); |
329 | static int backend_state; | |
330 | ||
331 | static void xenbus_reset_backend_state_changed(struct xenbus_watch *w, | |
332 | const char **v, unsigned int l) | |
333 | { | |
334 | xenbus_scanf(XBT_NIL, v[XS_WATCH_PATH], "", "%i", &backend_state); | |
335 | printk(KERN_DEBUG "XENBUS: backend %s %s\n", | |
336 | v[XS_WATCH_PATH], xenbus_strstate(backend_state)); | |
337 | wake_up(&backend_state_wq); | |
338 | } | |
339 | ||
340 | static void xenbus_reset_wait_for_backend(char *be, int expected) | |
341 | { | |
342 | long timeout; | |
343 | timeout = wait_event_interruptible_timeout(backend_state_wq, | |
344 | backend_state == expected, 5 * HZ); | |
345 | if (timeout <= 0) | |
283c0972 | 346 | pr_info("backend %s timed out\n", be); |
116df6f0 OH |
347 | } |
348 | ||
349 | /* | |
350 | * Reset frontend if it is in Connected or Closed state. | |
351 | * Wait for backend to catch up. | |
352 | * State Connected happens during kdump, Closed after kexec. | |
353 | */ | |
354 | static void xenbus_reset_frontend(char *fe, char *be, int be_state) | |
355 | { | |
356 | struct xenbus_watch be_watch; | |
357 | ||
358 | printk(KERN_DEBUG "XENBUS: backend %s %s\n", | |
359 | be, xenbus_strstate(be_state)); | |
360 | ||
361 | memset(&be_watch, 0, sizeof(be_watch)); | |
362 | be_watch.node = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/state", be); | |
363 | if (!be_watch.node) | |
364 | return; | |
365 | ||
366 | be_watch.callback = xenbus_reset_backend_state_changed; | |
367 | backend_state = XenbusStateUnknown; | |
368 | ||
283c0972 | 369 | pr_info("triggering reconnect on %s\n", be); |
116df6f0 OH |
370 | register_xenbus_watch(&be_watch); |
371 | ||
372 | /* fall through to forward backend to state XenbusStateInitialising */ | |
373 | switch (be_state) { | |
374 | case XenbusStateConnected: | |
375 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosing); | |
376 | xenbus_reset_wait_for_backend(be, XenbusStateClosing); | |
377 | ||
378 | case XenbusStateClosing: | |
379 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosed); | |
380 | xenbus_reset_wait_for_backend(be, XenbusStateClosed); | |
381 | ||
382 | case XenbusStateClosed: | |
383 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateInitialising); | |
384 | xenbus_reset_wait_for_backend(be, XenbusStateInitWait); | |
385 | } | |
386 | ||
387 | unregister_xenbus_watch(&be_watch); | |
283c0972 | 388 | pr_info("reconnect done on %s\n", be); |
116df6f0 OH |
389 | kfree(be_watch.node); |
390 | } | |
391 | ||
392 | static void xenbus_check_frontend(char *class, char *dev) | |
393 | { | |
394 | int be_state, fe_state, err; | |
395 | char *backend, *frontend; | |
396 | ||
397 | frontend = kasprintf(GFP_NOIO | __GFP_HIGH, "device/%s/%s", class, dev); | |
398 | if (!frontend) | |
399 | return; | |
400 | ||
401 | err = xenbus_scanf(XBT_NIL, frontend, "state", "%i", &fe_state); | |
402 | if (err != 1) | |
403 | goto out; | |
404 | ||
405 | switch (fe_state) { | |
406 | case XenbusStateConnected: | |
407 | case XenbusStateClosed: | |
408 | printk(KERN_DEBUG "XENBUS: frontend %s %s\n", | |
409 | frontend, xenbus_strstate(fe_state)); | |
410 | backend = xenbus_read(XBT_NIL, frontend, "backend", NULL); | |
411 | if (!backend || IS_ERR(backend)) | |
412 | goto out; | |
413 | err = xenbus_scanf(XBT_NIL, backend, "state", "%i", &be_state); | |
414 | if (err == 1) | |
415 | xenbus_reset_frontend(frontend, backend, be_state); | |
416 | kfree(backend); | |
417 | break; | |
418 | default: | |
419 | break; | |
420 | } | |
421 | out: | |
422 | kfree(frontend); | |
423 | } | |
424 | ||
425 | static void xenbus_reset_state(void) | |
426 | { | |
427 | char **devclass, **dev; | |
428 | int devclass_n, dev_n; | |
429 | int i, j; | |
430 | ||
431 | devclass = xenbus_directory(XBT_NIL, "device", "", &devclass_n); | |
432 | if (IS_ERR(devclass)) | |
433 | return; | |
434 | ||
435 | for (i = 0; i < devclass_n; i++) { | |
436 | dev = xenbus_directory(XBT_NIL, "device", devclass[i], &dev_n); | |
437 | if (IS_ERR(dev)) | |
438 | continue; | |
439 | for (j = 0; j < dev_n; j++) | |
440 | xenbus_check_frontend(devclass[i], dev[j]); | |
441 | kfree(dev); | |
442 | } | |
443 | kfree(devclass); | |
444 | } | |
445 | ||
df660251 IC |
446 | static int frontend_probe_and_watch(struct notifier_block *notifier, |
447 | unsigned long event, | |
448 | void *data) | |
449 | { | |
116df6f0 OH |
450 | /* reset devices in Connected or Closed state */ |
451 | if (xen_hvm_domain()) | |
452 | xenbus_reset_state(); | |
df660251 IC |
453 | /* Enumerate devices in xenstore and watch for changes. */ |
454 | xenbus_probe_devices(&xenbus_frontend); | |
df660251 | 455 | register_xenbus_watch(&fe_watch); |
0ff4fdf0 | 456 | |
df660251 IC |
457 | return NOTIFY_DONE; |
458 | } | |
459 | ||
460 | ||
2de06cc1 IC |
461 | static int __init xenbus_probe_frontend_init(void) |
462 | { | |
df660251 IC |
463 | static struct notifier_block xenstore_notifier = { |
464 | .notifier_call = frontend_probe_and_watch | |
465 | }; | |
2de06cc1 IC |
466 | int err; |
467 | ||
468 | DPRINTK(""); | |
469 | ||
470 | /* Register ourselves with the kernel bus subsystem */ | |
471 | err = bus_register(&xenbus_frontend.bus); | |
0ff4fdf0 | 472 | if (err) |
2de06cc1 | 473 | return err; |
2de06cc1 | 474 | |
df660251 | 475 | register_xenstore_notifier(&xenstore_notifier); |
2de06cc1 | 476 | |
2abb2746 AC |
477 | xenbus_frontend_wq = create_workqueue("xenbus_frontend"); |
478 | ||
2de06cc1 IC |
479 | return 0; |
480 | } | |
806f5463 | 481 | subsys_initcall(xenbus_probe_frontend_init); |
2de06cc1 IC |
482 | |
483 | #ifndef MODULE | |
484 | static int __init boot_wait_for_devices(void) | |
485 | { | |
486 | if (xen_hvm_domain() && !xen_platform_pci_unplug) | |
487 | return -ENODEV; | |
488 | ||
489 | ready_to_wait_for_devices = 1; | |
490 | wait_for_devices(NULL); | |
491 | return 0; | |
492 | } | |
493 | ||
494 | late_initcall(boot_wait_for_devices); | |
495 | #endif | |
1b31a143 JF |
496 | |
497 | MODULE_LICENSE("GPL"); |