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 | ||
2abb2746 AC |
118 | queue_work(xenbus_frontend_wq, &xdev->work); |
119 | ||
120 | return 0; | |
121 | } | |
122 | ||
123 | return xenbus_dev_resume(dev); | |
124 | } | |
125 | ||
d7ead0c3 AC |
126 | static int xenbus_frontend_dev_probe(struct device *dev) |
127 | { | |
128 | if (xen_store_domain_type == XS_LOCAL) { | |
129 | struct xenbus_device *xdev = to_xenbus_device(dev); | |
130 | INIT_WORK(&xdev->work, xenbus_frontend_delayed_resume); | |
131 | } | |
132 | ||
133 | return xenbus_dev_probe(dev); | |
134 | } | |
135 | ||
c7853aea | 136 | static const struct dev_pm_ops xenbus_pm_ops = { |
b3e96c0c | 137 | .suspend = xenbus_dev_suspend, |
2abb2746 | 138 | .resume = xenbus_frontend_dev_resume, |
b3e96c0c SR |
139 | .freeze = xenbus_dev_suspend, |
140 | .thaw = xenbus_dev_cancel, | |
141 | .restore = xenbus_dev_resume, | |
c7853aea KS |
142 | }; |
143 | ||
2de06cc1 IC |
144 | static struct xen_bus_type xenbus_frontend = { |
145 | .root = "device", | |
6bac7f9f | 146 | .levels = 2, /* device/type/<id> */ |
2de06cc1 IC |
147 | .get_bus_id = frontend_bus_id, |
148 | .probe = xenbus_probe_frontend, | |
149 | .otherend_changed = backend_changed, | |
150 | .bus = { | |
6bac7f9f IC |
151 | .name = "xen", |
152 | .match = xenbus_match, | |
153 | .uevent = xenbus_uevent_frontend, | |
d7ead0c3 | 154 | .probe = xenbus_frontend_dev_probe, |
6bac7f9f IC |
155 | .remove = xenbus_dev_remove, |
156 | .shutdown = xenbus_dev_shutdown, | |
85dd9268 | 157 | .dev_groups = xenbus_dev_groups, |
6bac7f9f | 158 | |
c7853aea | 159 | .pm = &xenbus_pm_ops, |
2de06cc1 IC |
160 | }, |
161 | }; | |
162 | ||
163 | static void frontend_changed(struct xenbus_watch *watch, | |
164 | const char **vec, unsigned int len) | |
165 | { | |
166 | DPRINTK(""); | |
167 | ||
168 | xenbus_dev_changed(vec[XS_WATCH_PATH], &xenbus_frontend); | |
169 | } | |
170 | ||
171 | ||
172 | /* We watch for devices appearing and vanishing. */ | |
173 | static struct xenbus_watch fe_watch = { | |
174 | .node = "device", | |
175 | .callback = frontend_changed, | |
176 | }; | |
177 | ||
178 | static int read_backend_details(struct xenbus_device *xendev) | |
179 | { | |
180 | return xenbus_read_otherend_details(xendev, "backend-id", "backend"); | |
181 | } | |
182 | ||
3066616c | 183 | static int is_device_connecting(struct device *dev, void *data, bool ignore_nonessential) |
2de06cc1 IC |
184 | { |
185 | struct xenbus_device *xendev = to_xenbus_device(dev); | |
186 | struct device_driver *drv = data; | |
187 | struct xenbus_driver *xendrv; | |
188 | ||
189 | /* | |
190 | * A device with no driver will never connect. We care only about | |
191 | * devices which should currently be in the process of connecting. | |
192 | */ | |
193 | if (!dev->driver) | |
194 | return 0; | |
195 | ||
196 | /* Is this search limited to a particular driver? */ | |
197 | if (drv && (dev->driver != drv)) | |
198 | return 0; | |
199 | ||
3066616c KRW |
200 | if (ignore_nonessential) { |
201 | /* With older QEMU, for PVonHVM guests the guest config files | |
202 | * could contain: vfb = [ 'vnc=1, vnclisten=0.0.0.0'] | |
203 | * which is nonsensical as there is no PV FB (there can be | |
204 | * a PVKB) running as HVM guest. */ | |
205 | ||
206 | if ((strncmp(xendev->nodename, "device/vkbd", 11) == 0)) | |
207 | return 0; | |
208 | ||
209 | if ((strncmp(xendev->nodename, "device/vfb", 10) == 0)) | |
210 | return 0; | |
211 | } | |
2de06cc1 IC |
212 | xendrv = to_xenbus_driver(dev->driver); |
213 | return (xendev->state < XenbusStateConnected || | |
214 | (xendev->state == XenbusStateConnected && | |
215 | xendrv->is_ready && !xendrv->is_ready(xendev))); | |
216 | } | |
3066616c KRW |
217 | static int essential_device_connecting(struct device *dev, void *data) |
218 | { | |
219 | return is_device_connecting(dev, data, true /* ignore PV[KBB+FB] */); | |
220 | } | |
221 | static int non_essential_device_connecting(struct device *dev, void *data) | |
222 | { | |
223 | return is_device_connecting(dev, data, false); | |
224 | } | |
2de06cc1 | 225 | |
3066616c | 226 | static int exists_essential_connecting_device(struct device_driver *drv) |
2de06cc1 IC |
227 | { |
228 | return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
3066616c KRW |
229 | essential_device_connecting); |
230 | } | |
231 | static int exists_non_essential_connecting_device(struct device_driver *drv) | |
232 | { | |
233 | return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
234 | non_essential_device_connecting); | |
2de06cc1 IC |
235 | } |
236 | ||
237 | static int print_device_status(struct device *dev, void *data) | |
238 | { | |
239 | struct xenbus_device *xendev = to_xenbus_device(dev); | |
240 | struct device_driver *drv = data; | |
241 | ||
242 | /* Is this operation limited to a particular driver? */ | |
243 | if (drv && (dev->driver != drv)) | |
244 | return 0; | |
245 | ||
246 | if (!dev->driver) { | |
247 | /* Information only: is this too noisy? */ | |
283c0972 | 248 | pr_info("Device with no driver: %s\n", xendev->nodename); |
2de06cc1 IC |
249 | } else if (xendev->state < XenbusStateConnected) { |
250 | enum xenbus_state rstate = XenbusStateUnknown; | |
251 | if (xendev->otherend) | |
252 | rstate = xenbus_read_driver_state(xendev->otherend); | |
283c0972 JP |
253 | pr_warn("Timeout connecting to device: %s (local state %d, remote state %d)\n", |
254 | xendev->nodename, xendev->state, rstate); | |
2de06cc1 IC |
255 | } |
256 | ||
257 | return 0; | |
258 | } | |
259 | ||
260 | /* We only wait for device setup after most initcalls have run. */ | |
261 | static int ready_to_wait_for_devices; | |
262 | ||
3066616c KRW |
263 | static bool wait_loop(unsigned long start, unsigned int max_delay, |
264 | unsigned int *seconds_waited) | |
265 | { | |
266 | if (time_after(jiffies, start + (*seconds_waited+5)*HZ)) { | |
267 | if (!*seconds_waited) | |
283c0972 | 268 | pr_warn("Waiting for devices to initialise: "); |
3066616c | 269 | *seconds_waited += 5; |
283c0972 JP |
270 | pr_cont("%us...", max_delay - *seconds_waited); |
271 | if (*seconds_waited == max_delay) { | |
272 | pr_cont("\n"); | |
3066616c | 273 | return true; |
283c0972 | 274 | } |
3066616c KRW |
275 | } |
276 | ||
277 | schedule_timeout_interruptible(HZ/10); | |
278 | ||
279 | return false; | |
280 | } | |
2de06cc1 IC |
281 | /* |
282 | * On a 5-minute timeout, wait for all devices currently configured. We need | |
283 | * to do this to guarantee that the filesystems and / or network devices | |
284 | * needed for boot are available, before we can allow the boot to proceed. | |
285 | * | |
286 | * This needs to be on a late_initcall, to happen after the frontend device | |
287 | * drivers have been initialised, but before the root fs is mounted. | |
288 | * | |
289 | * A possible improvement here would be to have the tools add a per-device | |
290 | * flag to the store entry, indicating whether it is needed at boot time. | |
291 | * This would allow people who knew what they were doing to accelerate their | |
292 | * boot slightly, but of course needs tools or manual intervention to set up | |
293 | * those flags correctly. | |
294 | */ | |
295 | static void wait_for_devices(struct xenbus_driver *xendrv) | |
296 | { | |
297 | unsigned long start = jiffies; | |
298 | struct device_driver *drv = xendrv ? &xendrv->driver : NULL; | |
299 | unsigned int seconds_waited = 0; | |
300 | ||
301 | if (!ready_to_wait_for_devices || !xen_domain()) | |
302 | return; | |
303 | ||
3066616c KRW |
304 | while (exists_non_essential_connecting_device(drv)) |
305 | if (wait_loop(start, 30, &seconds_waited)) | |
306 | break; | |
307 | ||
308 | /* Skips PVKB and PVFB check.*/ | |
309 | while (exists_essential_connecting_device(drv)) | |
310 | if (wait_loop(start, 270, &seconds_waited)) | |
311 | break; | |
2de06cc1 IC |
312 | |
313 | if (seconds_waited) | |
314 | printk("\n"); | |
315 | ||
316 | bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
317 | print_device_status); | |
318 | } | |
319 | ||
95afae48 DV |
320 | int __xenbus_register_frontend(struct xenbus_driver *drv, struct module *owner, |
321 | const char *mod_name) | |
2de06cc1 IC |
322 | { |
323 | int ret; | |
324 | ||
325 | drv->read_otherend_details = read_backend_details; | |
326 | ||
95afae48 DV |
327 | ret = xenbus_register_driver_common(drv, &xenbus_frontend, |
328 | owner, mod_name); | |
2de06cc1 IC |
329 | if (ret) |
330 | return ret; | |
331 | ||
332 | /* If this driver is loaded as a module wait for devices to attach. */ | |
333 | wait_for_devices(drv); | |
334 | ||
335 | return 0; | |
336 | } | |
95afae48 | 337 | EXPORT_SYMBOL_GPL(__xenbus_register_frontend); |
2de06cc1 | 338 | |
116df6f0 OH |
339 | static DECLARE_WAIT_QUEUE_HEAD(backend_state_wq); |
340 | static int backend_state; | |
341 | ||
342 | static void xenbus_reset_backend_state_changed(struct xenbus_watch *w, | |
343 | const char **v, unsigned int l) | |
344 | { | |
345 | xenbus_scanf(XBT_NIL, v[XS_WATCH_PATH], "", "%i", &backend_state); | |
346 | printk(KERN_DEBUG "XENBUS: backend %s %s\n", | |
347 | v[XS_WATCH_PATH], xenbus_strstate(backend_state)); | |
348 | wake_up(&backend_state_wq); | |
349 | } | |
350 | ||
351 | static void xenbus_reset_wait_for_backend(char *be, int expected) | |
352 | { | |
353 | long timeout; | |
354 | timeout = wait_event_interruptible_timeout(backend_state_wq, | |
355 | backend_state == expected, 5 * HZ); | |
356 | if (timeout <= 0) | |
283c0972 | 357 | pr_info("backend %s timed out\n", be); |
116df6f0 OH |
358 | } |
359 | ||
360 | /* | |
361 | * Reset frontend if it is in Connected or Closed state. | |
362 | * Wait for backend to catch up. | |
363 | * State Connected happens during kdump, Closed after kexec. | |
364 | */ | |
365 | static void xenbus_reset_frontend(char *fe, char *be, int be_state) | |
366 | { | |
367 | struct xenbus_watch be_watch; | |
368 | ||
369 | printk(KERN_DEBUG "XENBUS: backend %s %s\n", | |
370 | be, xenbus_strstate(be_state)); | |
371 | ||
372 | memset(&be_watch, 0, sizeof(be_watch)); | |
373 | be_watch.node = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/state", be); | |
374 | if (!be_watch.node) | |
375 | return; | |
376 | ||
377 | be_watch.callback = xenbus_reset_backend_state_changed; | |
378 | backend_state = XenbusStateUnknown; | |
379 | ||
283c0972 | 380 | pr_info("triggering reconnect on %s\n", be); |
116df6f0 OH |
381 | register_xenbus_watch(&be_watch); |
382 | ||
383 | /* fall through to forward backend to state XenbusStateInitialising */ | |
384 | switch (be_state) { | |
385 | case XenbusStateConnected: | |
386 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosing); | |
387 | xenbus_reset_wait_for_backend(be, XenbusStateClosing); | |
388 | ||
389 | case XenbusStateClosing: | |
390 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosed); | |
391 | xenbus_reset_wait_for_backend(be, XenbusStateClosed); | |
392 | ||
393 | case XenbusStateClosed: | |
394 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateInitialising); | |
395 | xenbus_reset_wait_for_backend(be, XenbusStateInitWait); | |
396 | } | |
397 | ||
398 | unregister_xenbus_watch(&be_watch); | |
283c0972 | 399 | pr_info("reconnect done on %s\n", be); |
116df6f0 OH |
400 | kfree(be_watch.node); |
401 | } | |
402 | ||
403 | static void xenbus_check_frontend(char *class, char *dev) | |
404 | { | |
405 | int be_state, fe_state, err; | |
406 | char *backend, *frontend; | |
407 | ||
408 | frontend = kasprintf(GFP_NOIO | __GFP_HIGH, "device/%s/%s", class, dev); | |
409 | if (!frontend) | |
410 | return; | |
411 | ||
412 | err = xenbus_scanf(XBT_NIL, frontend, "state", "%i", &fe_state); | |
413 | if (err != 1) | |
414 | goto out; | |
415 | ||
416 | switch (fe_state) { | |
417 | case XenbusStateConnected: | |
418 | case XenbusStateClosed: | |
419 | printk(KERN_DEBUG "XENBUS: frontend %s %s\n", | |
420 | frontend, xenbus_strstate(fe_state)); | |
421 | backend = xenbus_read(XBT_NIL, frontend, "backend", NULL); | |
422 | if (!backend || IS_ERR(backend)) | |
423 | goto out; | |
424 | err = xenbus_scanf(XBT_NIL, backend, "state", "%i", &be_state); | |
425 | if (err == 1) | |
426 | xenbus_reset_frontend(frontend, backend, be_state); | |
427 | kfree(backend); | |
428 | break; | |
429 | default: | |
430 | break; | |
431 | } | |
432 | out: | |
433 | kfree(frontend); | |
434 | } | |
435 | ||
436 | static void xenbus_reset_state(void) | |
437 | { | |
438 | char **devclass, **dev; | |
439 | int devclass_n, dev_n; | |
440 | int i, j; | |
441 | ||
442 | devclass = xenbus_directory(XBT_NIL, "device", "", &devclass_n); | |
443 | if (IS_ERR(devclass)) | |
444 | return; | |
445 | ||
446 | for (i = 0; i < devclass_n; i++) { | |
447 | dev = xenbus_directory(XBT_NIL, "device", devclass[i], &dev_n); | |
448 | if (IS_ERR(dev)) | |
449 | continue; | |
450 | for (j = 0; j < dev_n; j++) | |
451 | xenbus_check_frontend(devclass[i], dev[j]); | |
452 | kfree(dev); | |
453 | } | |
454 | kfree(devclass); | |
455 | } | |
456 | ||
df660251 IC |
457 | static int frontend_probe_and_watch(struct notifier_block *notifier, |
458 | unsigned long event, | |
459 | void *data) | |
460 | { | |
116df6f0 OH |
461 | /* reset devices in Connected or Closed state */ |
462 | if (xen_hvm_domain()) | |
463 | xenbus_reset_state(); | |
df660251 IC |
464 | /* Enumerate devices in xenstore and watch for changes. */ |
465 | xenbus_probe_devices(&xenbus_frontend); | |
df660251 | 466 | register_xenbus_watch(&fe_watch); |
0ff4fdf0 | 467 | |
df660251 IC |
468 | return NOTIFY_DONE; |
469 | } | |
470 | ||
471 | ||
2de06cc1 IC |
472 | static int __init xenbus_probe_frontend_init(void) |
473 | { | |
df660251 IC |
474 | static struct notifier_block xenstore_notifier = { |
475 | .notifier_call = frontend_probe_and_watch | |
476 | }; | |
2de06cc1 IC |
477 | int err; |
478 | ||
479 | DPRINTK(""); | |
480 | ||
481 | /* Register ourselves with the kernel bus subsystem */ | |
482 | err = bus_register(&xenbus_frontend.bus); | |
0ff4fdf0 | 483 | if (err) |
2de06cc1 | 484 | return err; |
2de06cc1 | 485 | |
df660251 | 486 | register_xenstore_notifier(&xenstore_notifier); |
2de06cc1 | 487 | |
d7ead0c3 AC |
488 | if (xen_store_domain_type == XS_LOCAL) { |
489 | xenbus_frontend_wq = create_workqueue("xenbus_frontend"); | |
490 | if (!xenbus_frontend_wq) | |
491 | pr_warn("create xenbus frontend workqueue failed, S3 resume is likely to fail\n"); | |
492 | } | |
2abb2746 | 493 | |
2de06cc1 IC |
494 | return 0; |
495 | } | |
806f5463 | 496 | subsys_initcall(xenbus_probe_frontend_init); |
2de06cc1 IC |
497 | |
498 | #ifndef MODULE | |
499 | static int __init boot_wait_for_devices(void) | |
500 | { | |
51c71a3b | 501 | if (!xen_has_pv_devices()) |
2de06cc1 IC |
502 | return -ENODEV; |
503 | ||
504 | ready_to_wait_for_devices = 1; | |
505 | wait_for_devices(NULL); | |
506 | return 0; | |
507 | } | |
508 | ||
509 | late_initcall(boot_wait_for_devices); | |
510 | #endif | |
1b31a143 JF |
511 | |
512 | MODULE_LICENSE("GPL"); |