]>
Commit | Line | Data |
---|---|---|
09c434b8 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
283c0972 JP |
2 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
3 | ||
4 | #define DPRINTK(fmt, ...) \ | |
5 | pr_debug("(%s:%d) " fmt "\n", \ | |
6 | __func__, __LINE__, ##__VA_ARGS__) | |
2de06cc1 IC |
7 | |
8 | #include <linux/kernel.h> | |
9 | #include <linux/err.h> | |
10 | #include <linux/string.h> | |
11 | #include <linux/ctype.h> | |
12 | #include <linux/fcntl.h> | |
13 | #include <linux/mm.h> | |
14 | #include <linux/proc_fs.h> | |
15 | #include <linux/notifier.h> | |
16 | #include <linux/kthread.h> | |
17 | #include <linux/mutex.h> | |
18 | #include <linux/io.h> | |
72ee5112 | 19 | #include <linux/module.h> |
2de06cc1 IC |
20 | |
21 | #include <asm/page.h> | |
2de06cc1 IC |
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 | ||
332f791d | 30 | #include "xenbus.h" |
2de06cc1 IC |
31 | |
32 | ||
2abb2746 | 33 | |
2de06cc1 IC |
34 | /* device/<type>/<id> => <type>-<id> */ |
35 | static int frontend_bus_id(char bus_id[XEN_BUS_ID_SIZE], const char *nodename) | |
36 | { | |
37 | nodename = strchr(nodename, '/'); | |
38 | if (!nodename || strlen(nodename + 1) >= XEN_BUS_ID_SIZE) { | |
283c0972 | 39 | pr_warn("bad frontend %s\n", nodename); |
2de06cc1 IC |
40 | return -EINVAL; |
41 | } | |
42 | ||
43 | strlcpy(bus_id, nodename + 1, XEN_BUS_ID_SIZE); | |
44 | if (!strchr(bus_id, '/')) { | |
283c0972 | 45 | pr_warn("bus_id %s no slash\n", bus_id); |
2de06cc1 IC |
46 | return -EINVAL; |
47 | } | |
48 | *strchr(bus_id, '/') = '-'; | |
49 | return 0; | |
50 | } | |
51 | ||
52 | /* device/<typename>/<name> */ | |
6bac7f9f IC |
53 | static int xenbus_probe_frontend(struct xen_bus_type *bus, const char *type, |
54 | const char *name) | |
2de06cc1 IC |
55 | { |
56 | char *nodename; | |
57 | int err; | |
58 | ||
42c46e6b SS |
59 | /* ignore console/0 */ |
60 | if (!strncmp(type, "console", 7) && !strncmp(name, "0", 1)) { | |
61 | DPRINTK("Ignoring buggy device entry console/0"); | |
62 | return 0; | |
63 | } | |
64 | ||
2de06cc1 IC |
65 | nodename = kasprintf(GFP_KERNEL, "%s/%s/%s", bus->root, type, name); |
66 | if (!nodename) | |
67 | return -ENOMEM; | |
68 | ||
69 | DPRINTK("%s", nodename); | |
70 | ||
71 | err = xenbus_probe_node(bus, type, nodename); | |
72 | kfree(nodename); | |
73 | return err; | |
74 | } | |
75 | ||
6bac7f9f IC |
76 | static int xenbus_uevent_frontend(struct device *_dev, |
77 | struct kobj_uevent_env *env) | |
df660251 IC |
78 | { |
79 | struct xenbus_device *dev = to_xenbus_device(_dev); | |
80 | ||
81 | if (add_uevent_var(env, "MODALIAS=xen:%s", dev->devicetype)) | |
82 | return -ENOMEM; | |
83 | ||
84 | return 0; | |
85 | } | |
86 | ||
87 | ||
2de06cc1 | 88 | static void backend_changed(struct xenbus_watch *watch, |
5584ea25 | 89 | const char *path, const char *token) |
2de06cc1 | 90 | { |
5584ea25 | 91 | xenbus_otherend_changed(watch, path, token, 1); |
2de06cc1 IC |
92 | } |
93 | ||
2abb2746 AC |
94 | static void xenbus_frontend_delayed_resume(struct work_struct *w) |
95 | { | |
96 | struct xenbus_device *xdev = container_of(w, struct xenbus_device, work); | |
97 | ||
98 | xenbus_dev_resume(&xdev->dev); | |
99 | } | |
100 | ||
101 | static int xenbus_frontend_dev_resume(struct device *dev) | |
102 | { | |
103 | /* | |
104 | * If xenstored is running in this domain, we cannot access the backend | |
105 | * state at the moment, so we need to defer xenbus_dev_resume | |
106 | */ | |
107 | if (xen_store_domain_type == XS_LOCAL) { | |
108 | struct xenbus_device *xdev = to_xenbus_device(dev); | |
109 | ||
5ee405d9 | 110 | schedule_work(&xdev->work); |
2abb2746 AC |
111 | |
112 | return 0; | |
113 | } | |
114 | ||
115 | return xenbus_dev_resume(dev); | |
116 | } | |
117 | ||
d7ead0c3 AC |
118 | static int xenbus_frontend_dev_probe(struct device *dev) |
119 | { | |
120 | if (xen_store_domain_type == XS_LOCAL) { | |
121 | struct xenbus_device *xdev = to_xenbus_device(dev); | |
122 | INIT_WORK(&xdev->work, xenbus_frontend_delayed_resume); | |
123 | } | |
124 | ||
125 | return xenbus_dev_probe(dev); | |
126 | } | |
127 | ||
c534374e PD |
128 | static void xenbus_frontend_dev_shutdown(struct device *_dev) |
129 | { | |
130 | struct xenbus_device *dev = to_xenbus_device(_dev); | |
131 | unsigned long timeout = 5*HZ; | |
132 | ||
133 | DPRINTK("%s", dev->nodename); | |
134 | ||
135 | get_device(&dev->dev); | |
136 | if (dev->state != XenbusStateConnected) { | |
137 | pr_info("%s: %s: %s != Connected, skipping\n", | |
138 | __func__, dev->nodename, xenbus_strstate(dev->state)); | |
139 | goto out; | |
140 | } | |
141 | xenbus_switch_state(dev, XenbusStateClosing); | |
142 | timeout = wait_for_completion_timeout(&dev->down, timeout); | |
143 | if (!timeout) | |
144 | pr_info("%s: %s timeout closing device\n", | |
145 | __func__, dev->nodename); | |
146 | out: | |
147 | put_device(&dev->dev); | |
148 | } | |
149 | ||
c7853aea | 150 | static const struct dev_pm_ops xenbus_pm_ops = { |
b3e96c0c | 151 | .suspend = xenbus_dev_suspend, |
2abb2746 | 152 | .resume = xenbus_frontend_dev_resume, |
b3e96c0c SR |
153 | .freeze = xenbus_dev_suspend, |
154 | .thaw = xenbus_dev_cancel, | |
155 | .restore = xenbus_dev_resume, | |
c7853aea KS |
156 | }; |
157 | ||
2de06cc1 IC |
158 | static struct xen_bus_type xenbus_frontend = { |
159 | .root = "device", | |
6bac7f9f | 160 | .levels = 2, /* device/type/<id> */ |
2de06cc1 IC |
161 | .get_bus_id = frontend_bus_id, |
162 | .probe = xenbus_probe_frontend, | |
163 | .otherend_changed = backend_changed, | |
164 | .bus = { | |
6bac7f9f IC |
165 | .name = "xen", |
166 | .match = xenbus_match, | |
167 | .uevent = xenbus_uevent_frontend, | |
d7ead0c3 | 168 | .probe = xenbus_frontend_dev_probe, |
6bac7f9f | 169 | .remove = xenbus_dev_remove, |
c534374e | 170 | .shutdown = xenbus_frontend_dev_shutdown, |
85dd9268 | 171 | .dev_groups = xenbus_dev_groups, |
6bac7f9f | 172 | |
c7853aea | 173 | .pm = &xenbus_pm_ops, |
2de06cc1 IC |
174 | }, |
175 | }; | |
176 | ||
177 | static void frontend_changed(struct xenbus_watch *watch, | |
5584ea25 | 178 | const char *path, const char *token) |
2de06cc1 IC |
179 | { |
180 | DPRINTK(""); | |
181 | ||
5584ea25 | 182 | xenbus_dev_changed(path, &xenbus_frontend); |
2de06cc1 IC |
183 | } |
184 | ||
185 | ||
186 | /* We watch for devices appearing and vanishing. */ | |
187 | static struct xenbus_watch fe_watch = { | |
188 | .node = "device", | |
189 | .callback = frontend_changed, | |
190 | }; | |
191 | ||
192 | static int read_backend_details(struct xenbus_device *xendev) | |
193 | { | |
194 | return xenbus_read_otherend_details(xendev, "backend-id", "backend"); | |
195 | } | |
196 | ||
3066616c | 197 | static int is_device_connecting(struct device *dev, void *data, bool ignore_nonessential) |
2de06cc1 IC |
198 | { |
199 | struct xenbus_device *xendev = to_xenbus_device(dev); | |
200 | struct device_driver *drv = data; | |
201 | struct xenbus_driver *xendrv; | |
202 | ||
203 | /* | |
204 | * A device with no driver will never connect. We care only about | |
205 | * devices which should currently be in the process of connecting. | |
206 | */ | |
207 | if (!dev->driver) | |
208 | return 0; | |
209 | ||
210 | /* Is this search limited to a particular driver? */ | |
211 | if (drv && (dev->driver != drv)) | |
212 | return 0; | |
213 | ||
3066616c KRW |
214 | if (ignore_nonessential) { |
215 | /* With older QEMU, for PVonHVM guests the guest config files | |
216 | * could contain: vfb = [ 'vnc=1, vnclisten=0.0.0.0'] | |
217 | * which is nonsensical as there is no PV FB (there can be | |
218 | * a PVKB) running as HVM guest. */ | |
219 | ||
220 | if ((strncmp(xendev->nodename, "device/vkbd", 11) == 0)) | |
221 | return 0; | |
222 | ||
223 | if ((strncmp(xendev->nodename, "device/vfb", 10) == 0)) | |
224 | return 0; | |
225 | } | |
2de06cc1 IC |
226 | xendrv = to_xenbus_driver(dev->driver); |
227 | return (xendev->state < XenbusStateConnected || | |
228 | (xendev->state == XenbusStateConnected && | |
229 | xendrv->is_ready && !xendrv->is_ready(xendev))); | |
230 | } | |
3066616c KRW |
231 | static int essential_device_connecting(struct device *dev, void *data) |
232 | { | |
233 | return is_device_connecting(dev, data, true /* ignore PV[KBB+FB] */); | |
234 | } | |
235 | static int non_essential_device_connecting(struct device *dev, void *data) | |
236 | { | |
237 | return is_device_connecting(dev, data, false); | |
238 | } | |
2de06cc1 | 239 | |
3066616c | 240 | static int exists_essential_connecting_device(struct device_driver *drv) |
2de06cc1 IC |
241 | { |
242 | return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
3066616c KRW |
243 | essential_device_connecting); |
244 | } | |
245 | static int exists_non_essential_connecting_device(struct device_driver *drv) | |
246 | { | |
247 | return bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
248 | non_essential_device_connecting); | |
2de06cc1 IC |
249 | } |
250 | ||
251 | static int print_device_status(struct device *dev, void *data) | |
252 | { | |
253 | struct xenbus_device *xendev = to_xenbus_device(dev); | |
254 | struct device_driver *drv = data; | |
255 | ||
256 | /* Is this operation limited to a particular driver? */ | |
257 | if (drv && (dev->driver != drv)) | |
258 | return 0; | |
259 | ||
260 | if (!dev->driver) { | |
261 | /* Information only: is this too noisy? */ | |
283c0972 | 262 | pr_info("Device with no driver: %s\n", xendev->nodename); |
2de06cc1 IC |
263 | } else if (xendev->state < XenbusStateConnected) { |
264 | enum xenbus_state rstate = XenbusStateUnknown; | |
265 | if (xendev->otherend) | |
266 | rstate = xenbus_read_driver_state(xendev->otherend); | |
283c0972 JP |
267 | pr_warn("Timeout connecting to device: %s (local state %d, remote state %d)\n", |
268 | xendev->nodename, xendev->state, rstate); | |
2de06cc1 IC |
269 | } |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
274 | /* We only wait for device setup after most initcalls have run. */ | |
275 | static int ready_to_wait_for_devices; | |
276 | ||
3066616c KRW |
277 | static bool wait_loop(unsigned long start, unsigned int max_delay, |
278 | unsigned int *seconds_waited) | |
279 | { | |
280 | if (time_after(jiffies, start + (*seconds_waited+5)*HZ)) { | |
281 | if (!*seconds_waited) | |
283c0972 | 282 | pr_warn("Waiting for devices to initialise: "); |
3066616c | 283 | *seconds_waited += 5; |
283c0972 JP |
284 | pr_cont("%us...", max_delay - *seconds_waited); |
285 | if (*seconds_waited == max_delay) { | |
286 | pr_cont("\n"); | |
3066616c | 287 | return true; |
283c0972 | 288 | } |
3066616c KRW |
289 | } |
290 | ||
291 | schedule_timeout_interruptible(HZ/10); | |
292 | ||
293 | return false; | |
294 | } | |
2de06cc1 IC |
295 | /* |
296 | * On a 5-minute timeout, wait for all devices currently configured. We need | |
297 | * to do this to guarantee that the filesystems and / or network devices | |
298 | * needed for boot are available, before we can allow the boot to proceed. | |
299 | * | |
300 | * This needs to be on a late_initcall, to happen after the frontend device | |
301 | * drivers have been initialised, but before the root fs is mounted. | |
302 | * | |
303 | * A possible improvement here would be to have the tools add a per-device | |
304 | * flag to the store entry, indicating whether it is needed at boot time. | |
305 | * This would allow people who knew what they were doing to accelerate their | |
306 | * boot slightly, but of course needs tools or manual intervention to set up | |
307 | * those flags correctly. | |
308 | */ | |
309 | static void wait_for_devices(struct xenbus_driver *xendrv) | |
310 | { | |
311 | unsigned long start = jiffies; | |
312 | struct device_driver *drv = xendrv ? &xendrv->driver : NULL; | |
313 | unsigned int seconds_waited = 0; | |
314 | ||
315 | if (!ready_to_wait_for_devices || !xen_domain()) | |
316 | return; | |
317 | ||
3066616c KRW |
318 | while (exists_non_essential_connecting_device(drv)) |
319 | if (wait_loop(start, 30, &seconds_waited)) | |
320 | break; | |
321 | ||
322 | /* Skips PVKB and PVFB check.*/ | |
323 | while (exists_essential_connecting_device(drv)) | |
324 | if (wait_loop(start, 270, &seconds_waited)) | |
325 | break; | |
2de06cc1 IC |
326 | |
327 | if (seconds_waited) | |
328 | printk("\n"); | |
329 | ||
330 | bus_for_each_dev(&xenbus_frontend.bus, NULL, drv, | |
331 | print_device_status); | |
332 | } | |
333 | ||
95afae48 DV |
334 | int __xenbus_register_frontend(struct xenbus_driver *drv, struct module *owner, |
335 | const char *mod_name) | |
2de06cc1 IC |
336 | { |
337 | int ret; | |
338 | ||
339 | drv->read_otherend_details = read_backend_details; | |
340 | ||
95afae48 DV |
341 | ret = xenbus_register_driver_common(drv, &xenbus_frontend, |
342 | owner, mod_name); | |
2de06cc1 IC |
343 | if (ret) |
344 | return ret; | |
345 | ||
346 | /* If this driver is loaded as a module wait for devices to attach. */ | |
347 | wait_for_devices(drv); | |
348 | ||
349 | return 0; | |
350 | } | |
95afae48 | 351 | EXPORT_SYMBOL_GPL(__xenbus_register_frontend); |
2de06cc1 | 352 | |
116df6f0 OH |
353 | static DECLARE_WAIT_QUEUE_HEAD(backend_state_wq); |
354 | static int backend_state; | |
355 | ||
356 | static void xenbus_reset_backend_state_changed(struct xenbus_watch *w, | |
5584ea25 | 357 | const char *path, const char *token) |
116df6f0 | 358 | { |
5584ea25 | 359 | if (xenbus_scanf(XBT_NIL, path, "", "%i", |
c251f15c JB |
360 | &backend_state) != 1) |
361 | backend_state = XenbusStateUnknown; | |
116df6f0 | 362 | printk(KERN_DEBUG "XENBUS: backend %s %s\n", |
5584ea25 | 363 | path, xenbus_strstate(backend_state)); |
116df6f0 OH |
364 | wake_up(&backend_state_wq); |
365 | } | |
366 | ||
367 | static void xenbus_reset_wait_for_backend(char *be, int expected) | |
368 | { | |
369 | long timeout; | |
370 | timeout = wait_event_interruptible_timeout(backend_state_wq, | |
371 | backend_state == expected, 5 * HZ); | |
372 | if (timeout <= 0) | |
283c0972 | 373 | pr_info("backend %s timed out\n", be); |
116df6f0 OH |
374 | } |
375 | ||
376 | /* | |
377 | * Reset frontend if it is in Connected or Closed state. | |
378 | * Wait for backend to catch up. | |
379 | * State Connected happens during kdump, Closed after kexec. | |
380 | */ | |
381 | static void xenbus_reset_frontend(char *fe, char *be, int be_state) | |
382 | { | |
383 | struct xenbus_watch be_watch; | |
384 | ||
385 | printk(KERN_DEBUG "XENBUS: backend %s %s\n", | |
386 | be, xenbus_strstate(be_state)); | |
387 | ||
388 | memset(&be_watch, 0, sizeof(be_watch)); | |
389 | be_watch.node = kasprintf(GFP_NOIO | __GFP_HIGH, "%s/state", be); | |
390 | if (!be_watch.node) | |
391 | return; | |
392 | ||
393 | be_watch.callback = xenbus_reset_backend_state_changed; | |
394 | backend_state = XenbusStateUnknown; | |
395 | ||
283c0972 | 396 | pr_info("triggering reconnect on %s\n", be); |
116df6f0 OH |
397 | register_xenbus_watch(&be_watch); |
398 | ||
399 | /* fall through to forward backend to state XenbusStateInitialising */ | |
400 | switch (be_state) { | |
401 | case XenbusStateConnected: | |
402 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosing); | |
403 | xenbus_reset_wait_for_backend(be, XenbusStateClosing); | |
df561f66 | 404 | fallthrough; |
116df6f0 OH |
405 | |
406 | case XenbusStateClosing: | |
407 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateClosed); | |
408 | xenbus_reset_wait_for_backend(be, XenbusStateClosed); | |
df561f66 | 409 | fallthrough; |
116df6f0 OH |
410 | |
411 | case XenbusStateClosed: | |
412 | xenbus_printf(XBT_NIL, fe, "state", "%d", XenbusStateInitialising); | |
413 | xenbus_reset_wait_for_backend(be, XenbusStateInitWait); | |
414 | } | |
415 | ||
416 | unregister_xenbus_watch(&be_watch); | |
283c0972 | 417 | pr_info("reconnect done on %s\n", be); |
116df6f0 OH |
418 | kfree(be_watch.node); |
419 | } | |
420 | ||
421 | static void xenbus_check_frontend(char *class, char *dev) | |
422 | { | |
423 | int be_state, fe_state, err; | |
424 | char *backend, *frontend; | |
425 | ||
426 | frontend = kasprintf(GFP_NOIO | __GFP_HIGH, "device/%s/%s", class, dev); | |
427 | if (!frontend) | |
428 | return; | |
429 | ||
430 | err = xenbus_scanf(XBT_NIL, frontend, "state", "%i", &fe_state); | |
431 | if (err != 1) | |
432 | goto out; | |
433 | ||
434 | switch (fe_state) { | |
435 | case XenbusStateConnected: | |
436 | case XenbusStateClosed: | |
437 | printk(KERN_DEBUG "XENBUS: frontend %s %s\n", | |
438 | frontend, xenbus_strstate(fe_state)); | |
439 | backend = xenbus_read(XBT_NIL, frontend, "backend", NULL); | |
440 | if (!backend || IS_ERR(backend)) | |
441 | goto out; | |
442 | err = xenbus_scanf(XBT_NIL, backend, "state", "%i", &be_state); | |
443 | if (err == 1) | |
444 | xenbus_reset_frontend(frontend, backend, be_state); | |
445 | kfree(backend); | |
446 | break; | |
447 | default: | |
448 | break; | |
449 | } | |
450 | out: | |
451 | kfree(frontend); | |
452 | } | |
453 | ||
454 | static void xenbus_reset_state(void) | |
455 | { | |
456 | char **devclass, **dev; | |
457 | int devclass_n, dev_n; | |
458 | int i, j; | |
459 | ||
460 | devclass = xenbus_directory(XBT_NIL, "device", "", &devclass_n); | |
461 | if (IS_ERR(devclass)) | |
462 | return; | |
463 | ||
464 | for (i = 0; i < devclass_n; i++) { | |
465 | dev = xenbus_directory(XBT_NIL, "device", devclass[i], &dev_n); | |
466 | if (IS_ERR(dev)) | |
467 | continue; | |
468 | for (j = 0; j < dev_n; j++) | |
469 | xenbus_check_frontend(devclass[i], dev[j]); | |
470 | kfree(dev); | |
471 | } | |
472 | kfree(devclass); | |
473 | } | |
474 | ||
df660251 IC |
475 | static int frontend_probe_and_watch(struct notifier_block *notifier, |
476 | unsigned long event, | |
477 | void *data) | |
478 | { | |
116df6f0 OH |
479 | /* reset devices in Connected or Closed state */ |
480 | if (xen_hvm_domain()) | |
481 | xenbus_reset_state(); | |
df660251 IC |
482 | /* Enumerate devices in xenstore and watch for changes. */ |
483 | xenbus_probe_devices(&xenbus_frontend); | |
df660251 | 484 | register_xenbus_watch(&fe_watch); |
0ff4fdf0 | 485 | |
df660251 IC |
486 | return NOTIFY_DONE; |
487 | } | |
488 | ||
489 | ||
2de06cc1 IC |
490 | static int __init xenbus_probe_frontend_init(void) |
491 | { | |
df660251 IC |
492 | static struct notifier_block xenstore_notifier = { |
493 | .notifier_call = frontend_probe_and_watch | |
494 | }; | |
2de06cc1 IC |
495 | int err; |
496 | ||
497 | DPRINTK(""); | |
498 | ||
499 | /* Register ourselves with the kernel bus subsystem */ | |
500 | err = bus_register(&xenbus_frontend.bus); | |
0ff4fdf0 | 501 | if (err) |
2de06cc1 | 502 | return err; |
2de06cc1 | 503 | |
df660251 | 504 | register_xenstore_notifier(&xenstore_notifier); |
2de06cc1 IC |
505 | |
506 | return 0; | |
507 | } | |
806f5463 | 508 | subsys_initcall(xenbus_probe_frontend_init); |
2de06cc1 IC |
509 | |
510 | #ifndef MODULE | |
511 | static int __init boot_wait_for_devices(void) | |
512 | { | |
51c71a3b | 513 | if (!xen_has_pv_devices()) |
2de06cc1 IC |
514 | return -ENODEV; |
515 | ||
516 | ready_to_wait_for_devices = 1; | |
517 | wait_for_devices(NULL); | |
518 | return 0; | |
519 | } | |
520 | ||
521 | late_initcall(boot_wait_for_devices); | |
522 | #endif | |
1b31a143 JF |
523 | |
524 | MODULE_LICENSE("GPL"); |