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 | 34 | |
2de06cc1 IC |
35 | /* device/<type>/<id> => <type>-<id> */ |
36 | static int frontend_bus_id(char bus_id[XEN_BUS_ID_SIZE], const char *nodename) | |
37 | { | |
38 | nodename = strchr(nodename, '/'); | |
39 | if (!nodename || strlen(nodename + 1) >= XEN_BUS_ID_SIZE) { | |
283c0972 | 40 | pr_warn("bad frontend %s\n", nodename); |
2de06cc1 IC |
41 | return -EINVAL; |
42 | } | |
43 | ||
44 | strlcpy(bus_id, nodename + 1, XEN_BUS_ID_SIZE); | |
45 | if (!strchr(bus_id, '/')) { | |
283c0972 | 46 | pr_warn("bus_id %s no slash\n", bus_id); |
2de06cc1 IC |
47 | return -EINVAL; |
48 | } | |
49 | *strchr(bus_id, '/') = '-'; | |
50 | return 0; | |
51 | } | |
52 | ||
53 | /* device/<typename>/<name> */ | |
6bac7f9f IC |
54 | static int xenbus_probe_frontend(struct xen_bus_type *bus, const char *type, |
55 | const char *name) | |
2de06cc1 IC |
56 | { |
57 | char *nodename; | |
58 | int err; | |
59 | ||
42c46e6b SS |
60 | /* ignore console/0 */ |
61 | if (!strncmp(type, "console", 7) && !strncmp(name, "0", 1)) { | |
62 | DPRINTK("Ignoring buggy device entry console/0"); | |
63 | return 0; | |
64 | } | |
65 | ||
2de06cc1 IC |
66 | nodename = kasprintf(GFP_KERNEL, "%s/%s/%s", bus->root, type, name); |
67 | if (!nodename) | |
68 | return -ENOMEM; | |
69 | ||
70 | DPRINTK("%s", nodename); | |
71 | ||
72 | err = xenbus_probe_node(bus, type, nodename); | |
73 | kfree(nodename); | |
74 | return err; | |
75 | } | |
76 | ||
6bac7f9f IC |
77 | static int xenbus_uevent_frontend(struct device *_dev, |
78 | struct kobj_uevent_env *env) | |
df660251 IC |
79 | { |
80 | struct xenbus_device *dev = to_xenbus_device(_dev); | |
81 | ||
82 | if (add_uevent_var(env, "MODALIAS=xen:%s", dev->devicetype)) | |
83 | return -ENOMEM; | |
84 | ||
85 | return 0; | |
86 | } | |
87 | ||
88 | ||
2de06cc1 IC |
89 | static void backend_changed(struct xenbus_watch *watch, |
90 | const char **vec, unsigned int len) | |
91 | { | |
92 | xenbus_otherend_changed(watch, vec, len, 1); | |
93 | } | |
94 | ||
2abb2746 AC |
95 | static void xenbus_frontend_delayed_resume(struct work_struct *w) |
96 | { | |
97 | struct xenbus_device *xdev = container_of(w, struct xenbus_device, work); | |
98 | ||
99 | xenbus_dev_resume(&xdev->dev); | |
100 | } | |
101 | ||
102 | static int xenbus_frontend_dev_resume(struct device *dev) | |
103 | { | |
104 | /* | |
105 | * If xenstored is running in this domain, we cannot access the backend | |
106 | * state at the moment, so we need to defer xenbus_dev_resume | |
107 | */ | |
108 | if (xen_store_domain_type == XS_LOCAL) { | |
109 | struct xenbus_device *xdev = to_xenbus_device(dev); | |
110 | ||
5ee405d9 | 111 | schedule_work(&xdev->work); |
2abb2746 AC |
112 | |
113 | return 0; | |
114 | } | |
115 | ||
116 | return xenbus_dev_resume(dev); | |
117 | } | |
118 | ||
d7ead0c3 AC |
119 | static int xenbus_frontend_dev_probe(struct device *dev) |
120 | { | |
121 | if (xen_store_domain_type == XS_LOCAL) { | |
122 | struct xenbus_device *xdev = to_xenbus_device(dev); | |
123 | INIT_WORK(&xdev->work, xenbus_frontend_delayed_resume); | |
124 | } | |
125 | ||
126 | return xenbus_dev_probe(dev); | |
127 | } | |
128 | ||
c7853aea | 129 | static const struct dev_pm_ops xenbus_pm_ops = { |
b3e96c0c | 130 | .suspend = xenbus_dev_suspend, |
2abb2746 | 131 | .resume = xenbus_frontend_dev_resume, |
b3e96c0c SR |
132 | .freeze = xenbus_dev_suspend, |
133 | .thaw = xenbus_dev_cancel, | |
134 | .restore = xenbus_dev_resume, | |
c7853aea KS |
135 | }; |
136 | ||
2de06cc1 IC |
137 | static struct xen_bus_type xenbus_frontend = { |
138 | .root = "device", | |
6bac7f9f | 139 | .levels = 2, /* device/type/<id> */ |
2de06cc1 IC |
140 | .get_bus_id = frontend_bus_id, |
141 | .probe = xenbus_probe_frontend, | |
142 | .otherend_changed = backend_changed, | |
143 | .bus = { | |
6bac7f9f IC |
144 | .name = "xen", |
145 | .match = xenbus_match, | |
146 | .uevent = xenbus_uevent_frontend, | |
d7ead0c3 | 147 | .probe = xenbus_frontend_dev_probe, |
6bac7f9f IC |
148 | .remove = xenbus_dev_remove, |
149 | .shutdown = xenbus_dev_shutdown, | |
85dd9268 | 150 | .dev_groups = xenbus_dev_groups, |
6bac7f9f | 151 | |
c7853aea | 152 | .pm = &xenbus_pm_ops, |
2de06cc1 IC |
153 | }, |
154 | }; | |
155 | ||
156 | static void frontend_changed(struct xenbus_watch *watch, | |
157 | const char **vec, unsigned int len) | |
158 | { | |
159 | DPRINTK(""); | |
160 | ||
161 | xenbus_dev_changed(vec[XS_WATCH_PATH], &xenbus_frontend); | |
162 | } | |
163 | ||
164 | ||
165 | /* We watch for devices appearing and vanishing. */ | |
166 | static struct xenbus_watch fe_watch = { | |
167 | .node = "device", | |
168 | .callback = frontend_changed, | |
169 | }; | |
170 | ||
171 | static int read_backend_details(struct xenbus_device *xendev) | |
172 | { | |
173 | return xenbus_read_otherend_details(xendev, "backend-id", "backend"); | |
174 | } | |
175 | ||
3066616c | 176 | static int is_device_connecting(struct device *dev, void *data, bool ignore_nonessential) |
2de06cc1 IC |
177 | { |
178 | struct xenbus_device *xendev = to_xenbus_device(dev); | |
179 | struct device_driver *drv = data; | |
180 | struct xenbus_driver *xendrv; | |
181 | ||
182 | /* | |
183 | * A device with no driver will never connect. We care only about | |
184 | * devices which should currently be in the process of connecting. | |
185 | */ | |
186 | if (!dev->driver) | |
187 | return 0; | |
188 | ||
189 | /* Is this search limited to a particular driver? */ | |
190 | if (drv && (dev->driver != drv)) | |
191 | return 0; | |
192 | ||
3066616c KRW |
193 | if (ignore_nonessential) { |
194 | /* With older QEMU, for PVonHVM guests the guest config files | |
195 | * could contain: vfb = [ 'vnc=1, vnclisten=0.0.0.0'] | |
196 | * which is nonsensical as there is no PV FB (there can be | |
197 | * a PVKB) running as HVM guest. */ | |
198 | ||
199 | if ((strncmp(xendev->nodename, "device/vkbd", 11) == 0)) | |
200 | return 0; | |
201 | ||
202 | if ((strncmp(xendev->nodename, "device/vfb", 10) == 0)) | |
203 | return 0; | |
204 | } | |
2de06cc1 IC |
205 | xendrv = to_xenbus_driver(dev->driver); |
206 | return (xendev->state < XenbusStateConnected || | |
207 | (xendev->state == XenbusStateConnected && | |
208 | xendrv->is_ready && !xendrv->is_ready(xendev))); | |
209 | } | |
3066616c KRW |
210 | static int essential_device_connecting(struct device *dev, void *data) |
211 | { | |
212 | return is_device_connecting(dev, data, true /* ignore PV[KBB+FB] */); | |
213 | } | |
214 | static int non_essential_device_connecting(struct device *dev, void *data) | |
215 | { | |
216 | return is_device_connecting(dev, data, false); | |
217 | } | |
2de06cc1 | 218 | |
3066616c | 219 | static int exists_essential_connecting_device(struct device_driver *drv) |
2de06cc1 IC |
220 | { |
221 | return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
3066616c KRW |
222 | essential_device_connecting); |
223 | } | |
224 | static int exists_non_essential_connecting_device(struct device_driver *drv) | |
225 | { | |
226 | return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
227 | non_essential_device_connecting); | |
2de06cc1 IC |
228 | } |
229 | ||
230 | static int print_device_status(struct device *dev, void *data) | |
231 | { | |
232 | struct xenbus_device *xendev = to_xenbus_device(dev); | |
233 | struct device_driver *drv = data; | |
234 | ||
235 | /* Is this operation limited to a particular driver? */ | |
236 | if (drv && (dev->driver != drv)) | |
237 | return 0; | |
238 | ||
239 | if (!dev->driver) { | |
240 | /* Information only: is this too noisy? */ | |
283c0972 | 241 | pr_info("Device with no driver: %s\n", xendev->nodename); |
2de06cc1 IC |
242 | } else if (xendev->state < XenbusStateConnected) { |
243 | enum xenbus_state rstate = XenbusStateUnknown; | |
244 | if (xendev->otherend) | |
245 | rstate = xenbus_read_driver_state(xendev->otherend); | |
283c0972 JP |
246 | pr_warn("Timeout connecting to device: %s (local state %d, remote state %d)\n", |
247 | xendev->nodename, xendev->state, rstate); | |
2de06cc1 IC |
248 | } |
249 | ||
250 | return 0; | |
251 | } | |
252 | ||
253 | /* We only wait for device setup after most initcalls have run. */ | |
254 | static int ready_to_wait_for_devices; | |
255 | ||
3066616c KRW |
256 | static bool wait_loop(unsigned long start, unsigned int max_delay, |
257 | unsigned int *seconds_waited) | |
258 | { | |
259 | if (time_after(jiffies, start + (*seconds_waited+5)*HZ)) { | |
260 | if (!*seconds_waited) | |
283c0972 | 261 | pr_warn("Waiting for devices to initialise: "); |
3066616c | 262 | *seconds_waited += 5; |
283c0972 JP |
263 | pr_cont("%us...", max_delay - *seconds_waited); |
264 | if (*seconds_waited == max_delay) { | |
265 | pr_cont("\n"); | |
3066616c | 266 | return true; |
283c0972 | 267 | } |
3066616c KRW |
268 | } |
269 | ||
270 | schedule_timeout_interruptible(HZ/10); | |
271 | ||
272 | return false; | |
273 | } | |
2de06cc1 IC |
274 | /* |
275 | * On a 5-minute timeout, wait for all devices currently configured. We need | |
276 | * to do this to guarantee that the filesystems and / or network devices | |
277 | * needed for boot are available, before we can allow the boot to proceed. | |
278 | * | |
279 | * This needs to be on a late_initcall, to happen after the frontend device | |
280 | * drivers have been initialised, but before the root fs is mounted. | |
281 | * | |
282 | * A possible improvement here would be to have the tools add a per-device | |
283 | * flag to the store entry, indicating whether it is needed at boot time. | |
284 | * This would allow people who knew what they were doing to accelerate their | |
285 | * boot slightly, but of course needs tools or manual intervention to set up | |
286 | * those flags correctly. | |
287 | */ | |
288 | static void wait_for_devices(struct xenbus_driver *xendrv) | |
289 | { | |
290 | unsigned long start = jiffies; | |
291 | struct device_driver *drv = xendrv ? &xendrv->driver : NULL; | |
292 | unsigned int seconds_waited = 0; | |
293 | ||
294 | if (!ready_to_wait_for_devices || !xen_domain()) | |
295 | return; | |
296 | ||
3066616c KRW |
297 | while (exists_non_essential_connecting_device(drv)) |
298 | if (wait_loop(start, 30, &seconds_waited)) | |
299 | break; | |
300 | ||
301 | /* Skips PVKB and PVFB check.*/ | |
302 | while (exists_essential_connecting_device(drv)) | |
303 | if (wait_loop(start, 270, &seconds_waited)) | |
304 | break; | |
2de06cc1 IC |
305 | |
306 | if (seconds_waited) | |
307 | printk("\n"); | |
308 | ||
309 | bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
310 | print_device_status); | |
311 | } | |
312 | ||
95afae48 DV |
313 | int __xenbus_register_frontend(struct xenbus_driver *drv, struct module *owner, |
314 | const char *mod_name) | |
2de06cc1 IC |
315 | { |
316 | int ret; | |
317 | ||
318 | drv->read_otherend_details = read_backend_details; | |
319 | ||
95afae48 DV |
320 | ret = xenbus_register_driver_common(drv, &xenbus_frontend, |
321 | owner, mod_name); | |
2de06cc1 IC |
322 | if (ret) |
323 | return ret; | |
324 | ||
325 | /* If this driver is loaded as a module wait for devices to attach. */ | |
326 | wait_for_devices(drv); | |
327 | ||
328 | return 0; | |
329 | } | |
95afae48 | 330 | EXPORT_SYMBOL_GPL(__xenbus_register_frontend); |
2de06cc1 | 331 | |
116df6f0 OH |
332 | static DECLARE_WAIT_QUEUE_HEAD(backend_state_wq); |
333 | static int backend_state; | |
334 | ||
335 | static void xenbus_reset_backend_state_changed(struct xenbus_watch *w, | |
336 | const char **v, unsigned int l) | |
337 | { | |
c251f15c JB |
338 | if (xenbus_scanf(XBT_NIL, v[XS_WATCH_PATH], "", "%i", |
339 | &backend_state) != 1) | |
340 | backend_state = XenbusStateUnknown; | |
116df6f0 OH |
341 | printk(KERN_DEBUG "XENBUS: backend %s %s\n", |
342 | v[XS_WATCH_PATH], xenbus_strstate(backend_state)); | |
343 | wake_up(&backend_state_wq); | |
344 | } | |
345 | ||
346 | static void xenbus_reset_wait_for_backend(char *be, int expected) | |
347 | { | |
348 | long timeout; | |
349 | timeout = wait_event_interruptible_timeout(backend_state_wq, | |
350 | backend_state == expected, 5 * HZ); | |
351 | if (timeout <= 0) | |
283c0972 | 352 | pr_info("backend %s timed out\n", be); |
116df6f0 OH |
353 | } |
354 | ||
355 | /* | |
356 | * Reset frontend if it is in Connected or Closed state. | |
357 | * Wait for backend to catch up. | |
358 | * State Connected happens during kdump, Closed after kexec. | |
359 | */ | |
360 | static void xenbus_reset_frontend(char *fe, char *be, int be_state) | |
361 | { | |
362 | struct xenbus_watch be_watch; | |
363 | ||
364 | printk(KERN_DEBUG "XENBUS: backend %s %s\n", | |
365 | be, xenbus_strstate(be_state)); | |
366 | ||
367 | memset(&be_watch, 0, sizeof(be_watch)); | |
368 | be_watch.node = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/state", be); | |
369 | if (!be_watch.node) | |
370 | return; | |
371 | ||
372 | be_watch.callback = xenbus_reset_backend_state_changed; | |
373 | backend_state = XenbusStateUnknown; | |
374 | ||
283c0972 | 375 | pr_info("triggering reconnect on %s\n", be); |
116df6f0 OH |
376 | register_xenbus_watch(&be_watch); |
377 | ||
378 | /* fall through to forward backend to state XenbusStateInitialising */ | |
379 | switch (be_state) { | |
380 | case XenbusStateConnected: | |
381 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosing); | |
382 | xenbus_reset_wait_for_backend(be, XenbusStateClosing); | |
383 | ||
384 | case XenbusStateClosing: | |
385 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosed); | |
386 | xenbus_reset_wait_for_backend(be, XenbusStateClosed); | |
387 | ||
388 | case XenbusStateClosed: | |
389 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateInitialising); | |
390 | xenbus_reset_wait_for_backend(be, XenbusStateInitWait); | |
391 | } | |
392 | ||
393 | unregister_xenbus_watch(&be_watch); | |
283c0972 | 394 | pr_info("reconnect done on %s\n", be); |
116df6f0 OH |
395 | kfree(be_watch.node); |
396 | } | |
397 | ||
398 | static void xenbus_check_frontend(char *class, char *dev) | |
399 | { | |
400 | int be_state, fe_state, err; | |
401 | char *backend, *frontend; | |
402 | ||
403 | frontend = kasprintf(GFP_NOIO | __GFP_HIGH, "device/%s/%s", class, dev); | |
404 | if (!frontend) | |
405 | return; | |
406 | ||
407 | err = xenbus_scanf(XBT_NIL, frontend, "state", "%i", &fe_state); | |
408 | if (err != 1) | |
409 | goto out; | |
410 | ||
411 | switch (fe_state) { | |
412 | case XenbusStateConnected: | |
413 | case XenbusStateClosed: | |
414 | printk(KERN_DEBUG "XENBUS: frontend %s %s\n", | |
415 | frontend, xenbus_strstate(fe_state)); | |
416 | backend = xenbus_read(XBT_NIL, frontend, "backend", NULL); | |
417 | if (!backend || IS_ERR(backend)) | |
418 | goto out; | |
419 | err = xenbus_scanf(XBT_NIL, backend, "state", "%i", &be_state); | |
420 | if (err == 1) | |
421 | xenbus_reset_frontend(frontend, backend, be_state); | |
422 | kfree(backend); | |
423 | break; | |
424 | default: | |
425 | break; | |
426 | } | |
427 | out: | |
428 | kfree(frontend); | |
429 | } | |
430 | ||
431 | static void xenbus_reset_state(void) | |
432 | { | |
433 | char **devclass, **dev; | |
434 | int devclass_n, dev_n; | |
435 | int i, j; | |
436 | ||
437 | devclass = xenbus_directory(XBT_NIL, "device", "", &devclass_n); | |
438 | if (IS_ERR(devclass)) | |
439 | return; | |
440 | ||
441 | for (i = 0; i < devclass_n; i++) { | |
442 | dev = xenbus_directory(XBT_NIL, "device", devclass[i], &dev_n); | |
443 | if (IS_ERR(dev)) | |
444 | continue; | |
445 | for (j = 0; j < dev_n; j++) | |
446 | xenbus_check_frontend(devclass[i], dev[j]); | |
447 | kfree(dev); | |
448 | } | |
449 | kfree(devclass); | |
450 | } | |
451 | ||
df660251 IC |
452 | static int frontend_probe_and_watch(struct notifier_block *notifier, |
453 | unsigned long event, | |
454 | void *data) | |
455 | { | |
116df6f0 OH |
456 | /* reset devices in Connected or Closed state */ |
457 | if (xen_hvm_domain()) | |
458 | xenbus_reset_state(); | |
df660251 IC |
459 | /* Enumerate devices in xenstore and watch for changes. */ |
460 | xenbus_probe_devices(&xenbus_frontend); | |
df660251 | 461 | register_xenbus_watch(&fe_watch); |
0ff4fdf0 | 462 | |
df660251 IC |
463 | return NOTIFY_DONE; |
464 | } | |
465 | ||
466 | ||
2de06cc1 IC |
467 | static int __init xenbus_probe_frontend_init(void) |
468 | { | |
df660251 IC |
469 | static struct notifier_block xenstore_notifier = { |
470 | .notifier_call = frontend_probe_and_watch | |
471 | }; | |
2de06cc1 IC |
472 | int err; |
473 | ||
474 | DPRINTK(""); | |
475 | ||
476 | /* Register ourselves with the kernel bus subsystem */ | |
477 | err = bus_register(&xenbus_frontend.bus); | |
0ff4fdf0 | 478 | if (err) |
2de06cc1 | 479 | return err; |
2de06cc1 | 480 | |
df660251 | 481 | register_xenstore_notifier(&xenstore_notifier); |
2de06cc1 IC |
482 | |
483 | return 0; | |
484 | } | |
806f5463 | 485 | subsys_initcall(xenbus_probe_frontend_init); |
2de06cc1 IC |
486 | |
487 | #ifndef MODULE | |
488 | static int __init boot_wait_for_devices(void) | |
489 | { | |
51c71a3b | 490 | if (!xen_has_pv_devices()) |
2de06cc1 IC |
491 | return -ENODEV; |
492 | ||
493 | ready_to_wait_for_devices = 1; | |
494 | wait_for_devices(NULL); | |
495 | return 0; | |
496 | } | |
497 | ||
498 | late_initcall(boot_wait_for_devices); | |
499 | #endif | |
1b31a143 JF |
500 | |
501 | MODULE_LICENSE("GPL"); |