Commit | Line | Data |
---|---|---|
4bac07c9 JF |
1 | /****************************************************************************** |
2 | * Client-facing interface for the Xenbus driver. In other words, the | |
3 | * interface between the Xenbus and the device-specific code, be it the | |
4 | * frontend or the backend of that driver. | |
5 | * | |
6 | * Copyright (C) 2005 XenSource Ltd | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU General Public License version 2 | |
10 | * as published by the Free Software Foundation; or, when distributed | |
11 | * separately from the Linux kernel or incorporated into other | |
12 | * software packages, subject to the following license: | |
13 | * | |
14 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
15 | * of this source file (the "Software"), to deal in the Software without | |
16 | * restriction, including without limitation the rights to use, copy, modify, | |
17 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
18 | * and to permit persons to whom the Software is furnished to do so, subject to | |
19 | * the following conditions: | |
20 | * | |
21 | * The above copyright notice and this permission notice shall be included in | |
22 | * all copies or substantial portions of the Software. | |
23 | * | |
24 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
25 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
26 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
27 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
28 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
29 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
30 | * IN THE SOFTWARE. | |
31 | */ | |
32 | ||
45e27161 | 33 | #include <linux/mm.h> |
5a0e3ad6 | 34 | #include <linux/slab.h> |
4bac07c9 | 35 | #include <linux/types.h> |
2c5d37d3 | 36 | #include <linux/spinlock.h> |
4bac07c9 | 37 | #include <linux/vmalloc.h> |
63c9744b | 38 | #include <linux/export.h> |
4bac07c9 | 39 | #include <asm/xen/hypervisor.h> |
a9fd60e2 | 40 | #include <xen/page.h> |
4bac07c9 JF |
41 | #include <xen/interface/xen.h> |
42 | #include <xen/interface/event_channel.h> | |
2c5d37d3 | 43 | #include <xen/balloon.h> |
4bac07c9 JF |
44 | #include <xen/events.h> |
45 | #include <xen/grant_table.h> | |
46 | #include <xen/xenbus.h> | |
2c5d37d3 | 47 | #include <xen/xen.h> |
be3e9cf3 | 48 | #include <xen/features.h> |
2c5d37d3 DDG |
49 | |
50 | #include "xenbus_probe.h" | |
51 | ||
89bf4b4e JG |
52 | #define XENBUS_PAGES(_grants) (DIV_ROUND_UP(_grants, XEN_PFN_PER_PAGE)) |
53 | ||
54 | #define XENBUS_MAX_RING_PAGES (XENBUS_PAGES(XENBUS_MAX_RING_GRANTS)) | |
55 | ||
2c5d37d3 DDG |
56 | struct xenbus_map_node { |
57 | struct list_head next; | |
58 | union { | |
ccc9d90a WL |
59 | struct { |
60 | struct vm_struct *area; | |
61 | } pv; | |
62 | struct { | |
89bf4b4e JG |
63 | struct page *pages[XENBUS_MAX_RING_PAGES]; |
64 | unsigned long addrs[XENBUS_MAX_RING_GRANTS]; | |
ccc9d90a WL |
65 | void *addr; |
66 | } hvm; | |
2c5d37d3 | 67 | }; |
9cce2914 | 68 | grant_handle_t handles[XENBUS_MAX_RING_GRANTS]; |
ccc9d90a | 69 | unsigned int nr_handles; |
2c5d37d3 DDG |
70 | }; |
71 | ||
72 | static DEFINE_SPINLOCK(xenbus_valloc_lock); | |
73 | static LIST_HEAD(xenbus_valloc_pages); | |
74 | ||
75 | struct xenbus_ring_ops { | |
ccc9d90a WL |
76 | int (*map)(struct xenbus_device *dev, |
77 | grant_ref_t *gnt_refs, unsigned int nr_grefs, | |
78 | void **vaddr); | |
2c5d37d3 DDG |
79 | int (*unmap)(struct xenbus_device *dev, void *vaddr); |
80 | }; | |
81 | ||
82 | static const struct xenbus_ring_ops *ring_ops __read_mostly; | |
4bac07c9 JF |
83 | |
84 | const char *xenbus_strstate(enum xenbus_state state) | |
85 | { | |
86 | static const char *const name[] = { | |
87 | [ XenbusStateUnknown ] = "Unknown", | |
88 | [ XenbusStateInitialising ] = "Initialising", | |
89 | [ XenbusStateInitWait ] = "InitWait", | |
90 | [ XenbusStateInitialised ] = "Initialised", | |
91 | [ XenbusStateConnected ] = "Connected", | |
92 | [ XenbusStateClosing ] = "Closing", | |
93 | [ XenbusStateClosed ] = "Closed", | |
89afb6e4 YI |
94 | [XenbusStateReconfiguring] = "Reconfiguring", |
95 | [XenbusStateReconfigured] = "Reconfigured", | |
4bac07c9 JF |
96 | }; |
97 | return (state < ARRAY_SIZE(name)) ? name[state] : "INVALID"; | |
98 | } | |
99 | EXPORT_SYMBOL_GPL(xenbus_strstate); | |
100 | ||
101 | /** | |
102 | * xenbus_watch_path - register a watch | |
103 | * @dev: xenbus device | |
104 | * @path: path to watch | |
105 | * @watch: watch to register | |
106 | * @callback: callback to register | |
107 | * | |
108 | * Register a @watch on the given path, using the given xenbus_watch structure | |
109 | * for storage, and the given @callback function as the callback. Return 0 on | |
110 | * success, or -errno on error. On success, the given @path will be saved as | |
111 | * @watch->node, and remains the caller's to free. On error, @watch->node will | |
112 | * be NULL, the device will switch to %XenbusStateClosing, and the error will | |
113 | * be saved in the store. | |
114 | */ | |
115 | int xenbus_watch_path(struct xenbus_device *dev, const char *path, | |
116 | struct xenbus_watch *watch, | |
117 | void (*callback)(struct xenbus_watch *, | |
118 | const char **, unsigned int)) | |
119 | { | |
120 | int err; | |
121 | ||
122 | watch->node = path; | |
123 | watch->callback = callback; | |
124 | ||
125 | err = register_xenbus_watch(watch); | |
126 | ||
127 | if (err) { | |
128 | watch->node = NULL; | |
129 | watch->callback = NULL; | |
130 | xenbus_dev_fatal(dev, err, "adding watch on %s", path); | |
131 | } | |
132 | ||
133 | return err; | |
134 | } | |
135 | EXPORT_SYMBOL_GPL(xenbus_watch_path); | |
136 | ||
137 | ||
138 | /** | |
139 | * xenbus_watch_pathfmt - register a watch on a sprintf-formatted path | |
140 | * @dev: xenbus device | |
141 | * @watch: watch to register | |
142 | * @callback: callback to register | |
143 | * @pathfmt: format of path to watch | |
144 | * | |
145 | * Register a watch on the given @path, using the given xenbus_watch | |
146 | * structure for storage, and the given @callback function as the callback. | |
147 | * Return 0 on success, or -errno on error. On success, the watched path | |
148 | * (@path/@path2) will be saved as @watch->node, and becomes the caller's to | |
149 | * kfree(). On error, watch->node will be NULL, so the caller has nothing to | |
150 | * free, the device will switch to %XenbusStateClosing, and the error will be | |
151 | * saved in the store. | |
152 | */ | |
153 | int xenbus_watch_pathfmt(struct xenbus_device *dev, | |
154 | struct xenbus_watch *watch, | |
155 | void (*callback)(struct xenbus_watch *, | |
156 | const char **, unsigned int), | |
157 | const char *pathfmt, ...) | |
158 | { | |
159 | int err; | |
160 | va_list ap; | |
161 | char *path; | |
162 | ||
163 | va_start(ap, pathfmt); | |
a144ff09 | 164 | path = kvasprintf(GFP_NOIO | __GFP_HIGH, pathfmt, ap); |
4bac07c9 JF |
165 | va_end(ap); |
166 | ||
167 | if (!path) { | |
168 | xenbus_dev_fatal(dev, -ENOMEM, "allocating path for watch"); | |
169 | return -ENOMEM; | |
170 | } | |
171 | err = xenbus_watch_path(dev, path, watch, callback); | |
172 | ||
173 | if (err) | |
174 | kfree(path); | |
175 | return err; | |
176 | } | |
177 | EXPORT_SYMBOL_GPL(xenbus_watch_pathfmt); | |
178 | ||
5b61cb90 DS |
179 | static void xenbus_switch_fatal(struct xenbus_device *, int, int, |
180 | const char *, ...); | |
4bac07c9 | 181 | |
5b61cb90 DS |
182 | static int |
183 | __xenbus_switch_state(struct xenbus_device *dev, | |
184 | enum xenbus_state state, int depth) | |
4bac07c9 JF |
185 | { |
186 | /* We check whether the state is currently set to the given value, and | |
187 | if not, then the state is set. We don't want to unconditionally | |
188 | write the given state, because we don't want to fire watches | |
189 | unnecessarily. Furthermore, if the node has gone, we don't write | |
190 | to it, as the device will be tearing down, and we don't want to | |
191 | resurrect that directory. | |
192 | ||
5b61cb90 DS |
193 | Note that, because of this cached value of our state, this |
194 | function will not take a caller's Xenstore transaction | |
195 | (something it was trying to in the past) because dev->state | |
196 | would not get reset if the transaction was aborted. | |
4bac07c9 JF |
197 | */ |
198 | ||
5b61cb90 | 199 | struct xenbus_transaction xbt; |
4bac07c9 | 200 | int current_state; |
5b61cb90 | 201 | int err, abort; |
4bac07c9 JF |
202 | |
203 | if (state == dev->state) | |
204 | return 0; | |
205 | ||
5b61cb90 DS |
206 | again: |
207 | abort = 1; | |
208 | ||
209 | err = xenbus_transaction_start(&xbt); | |
210 | if (err) { | |
211 | xenbus_switch_fatal(dev, depth, err, "starting transaction"); | |
4bac07c9 | 212 | return 0; |
5b61cb90 | 213 | } |
4bac07c9 | 214 | |
5b61cb90 DS |
215 | err = xenbus_scanf(xbt, dev->nodename, "state", "%d", ¤t_state); |
216 | if (err != 1) | |
217 | goto abort; | |
218 | ||
219 | err = xenbus_printf(xbt, dev->nodename, "state", "%d", state); | |
4bac07c9 | 220 | if (err) { |
5b61cb90 DS |
221 | xenbus_switch_fatal(dev, depth, err, "writing new state"); |
222 | goto abort; | |
4bac07c9 JF |
223 | } |
224 | ||
5b61cb90 DS |
225 | abort = 0; |
226 | abort: | |
227 | err = xenbus_transaction_end(xbt, abort); | |
228 | if (err) { | |
229 | if (err == -EAGAIN && !abort) | |
230 | goto again; | |
231 | xenbus_switch_fatal(dev, depth, err, "ending transaction"); | |
232 | } else | |
233 | dev->state = state; | |
4bac07c9 JF |
234 | |
235 | return 0; | |
236 | } | |
5b61cb90 DS |
237 | |
238 | /** | |
239 | * xenbus_switch_state | |
240 | * @dev: xenbus device | |
241 | * @state: new state | |
242 | * | |
243 | * Advertise in the store a change of the given driver to the given new_state. | |
244 | * Return 0 on success, or -errno on error. On error, the device will switch | |
245 | * to XenbusStateClosing, and the error will be saved in the store. | |
246 | */ | |
247 | int xenbus_switch_state(struct xenbus_device *dev, enum xenbus_state state) | |
248 | { | |
249 | return __xenbus_switch_state(dev, state, 0); | |
250 | } | |
251 | ||
4bac07c9 JF |
252 | EXPORT_SYMBOL_GPL(xenbus_switch_state); |
253 | ||
254 | int xenbus_frontend_closed(struct xenbus_device *dev) | |
255 | { | |
256 | xenbus_switch_state(dev, XenbusStateClosed); | |
257 | complete(&dev->down); | |
258 | return 0; | |
259 | } | |
260 | EXPORT_SYMBOL_GPL(xenbus_frontend_closed); | |
261 | ||
262 | /** | |
263 | * Return the path to the error node for the given device, or NULL on failure. | |
264 | * If the value returned is non-NULL, then it is the caller's to kfree. | |
265 | */ | |
266 | static char *error_path(struct xenbus_device *dev) | |
267 | { | |
268 | return kasprintf(GFP_KERNEL, "error/%s", dev->nodename); | |
269 | } | |
270 | ||
271 | ||
272 | static void xenbus_va_dev_error(struct xenbus_device *dev, int err, | |
273 | const char *fmt, va_list ap) | |
274 | { | |
4bac07c9 JF |
275 | unsigned int len; |
276 | char *printf_buffer = NULL; | |
277 | char *path_buffer = NULL; | |
278 | ||
279 | #define PRINTF_BUFFER_SIZE 4096 | |
280 | printf_buffer = kmalloc(PRINTF_BUFFER_SIZE, GFP_KERNEL); | |
281 | if (printf_buffer == NULL) | |
282 | goto fail; | |
283 | ||
284 | len = sprintf(printf_buffer, "%i ", -err); | |
305559f1 | 285 | vsnprintf(printf_buffer+len, PRINTF_BUFFER_SIZE-len, fmt, ap); |
4bac07c9 JF |
286 | |
287 | dev_err(&dev->dev, "%s\n", printf_buffer); | |
288 | ||
289 | path_buffer = error_path(dev); | |
290 | ||
291 | if (path_buffer == NULL) { | |
292 | dev_err(&dev->dev, "failed to write error node for %s (%s)\n", | |
293 | dev->nodename, printf_buffer); | |
294 | goto fail; | |
295 | } | |
296 | ||
297 | if (xenbus_write(XBT_NIL, path_buffer, "error", printf_buffer) != 0) { | |
298 | dev_err(&dev->dev, "failed to write error node for %s (%s)\n", | |
299 | dev->nodename, printf_buffer); | |
300 | goto fail; | |
301 | } | |
302 | ||
303 | fail: | |
304 | kfree(printf_buffer); | |
305 | kfree(path_buffer); | |
306 | } | |
307 | ||
308 | ||
309 | /** | |
310 | * xenbus_dev_error | |
311 | * @dev: xenbus device | |
312 | * @err: error to report | |
313 | * @fmt: error message format | |
314 | * | |
315 | * Report the given negative errno into the store, along with the given | |
316 | * formatted message. | |
317 | */ | |
318 | void xenbus_dev_error(struct xenbus_device *dev, int err, const char *fmt, ...) | |
319 | { | |
320 | va_list ap; | |
321 | ||
322 | va_start(ap, fmt); | |
323 | xenbus_va_dev_error(dev, err, fmt, ap); | |
324 | va_end(ap); | |
325 | } | |
326 | EXPORT_SYMBOL_GPL(xenbus_dev_error); | |
327 | ||
328 | /** | |
329 | * xenbus_dev_fatal | |
330 | * @dev: xenbus device | |
331 | * @err: error to report | |
332 | * @fmt: error message format | |
333 | * | |
334 | * Equivalent to xenbus_dev_error(dev, err, fmt, args), followed by | |
d8220347 | 335 | * xenbus_switch_state(dev, XenbusStateClosing) to schedule an orderly |
4bac07c9 JF |
336 | * closedown of this driver and its peer. |
337 | */ | |
338 | ||
339 | void xenbus_dev_fatal(struct xenbus_device *dev, int err, const char *fmt, ...) | |
340 | { | |
341 | va_list ap; | |
342 | ||
343 | va_start(ap, fmt); | |
344 | xenbus_va_dev_error(dev, err, fmt, ap); | |
345 | va_end(ap); | |
346 | ||
347 | xenbus_switch_state(dev, XenbusStateClosing); | |
348 | } | |
349 | EXPORT_SYMBOL_GPL(xenbus_dev_fatal); | |
350 | ||
5b61cb90 DS |
351 | /** |
352 | * Equivalent to xenbus_dev_fatal(dev, err, fmt, args), but helps | |
353 | * avoiding recursion within xenbus_switch_state. | |
354 | */ | |
355 | static void xenbus_switch_fatal(struct xenbus_device *dev, int depth, int err, | |
356 | const char *fmt, ...) | |
357 | { | |
358 | va_list ap; | |
359 | ||
360 | va_start(ap, fmt); | |
361 | xenbus_va_dev_error(dev, err, fmt, ap); | |
362 | va_end(ap); | |
363 | ||
364 | if (!depth) | |
365 | __xenbus_switch_state(dev, XenbusStateClosing, 1); | |
366 | } | |
367 | ||
4bac07c9 JF |
368 | /** |
369 | * xenbus_grant_ring | |
370 | * @dev: xenbus device | |
ccc9d90a WL |
371 | * @vaddr: starting virtual address of the ring |
372 | * @nr_pages: number of pages to be granted | |
373 | * @grefs: grant reference array to be filled in | |
374 | * | |
375 | * Grant access to the given @vaddr to the peer of the given device. | |
376 | * Then fill in @grefs with grant references. Return 0 on success, or | |
377 | * -errno on error. On error, the device will switch to | |
378 | * XenbusStateClosing, and the error will be saved in the store. | |
4bac07c9 | 379 | */ |
ccc9d90a WL |
380 | int xenbus_grant_ring(struct xenbus_device *dev, void *vaddr, |
381 | unsigned int nr_pages, grant_ref_t *grefs) | |
4bac07c9 | 382 | { |
ccc9d90a WL |
383 | int err; |
384 | int i, j; | |
385 | ||
386 | for (i = 0; i < nr_pages; i++) { | |
ccc9d90a | 387 | err = gnttab_grant_foreign_access(dev->otherend_id, |
0df4f266 | 388 | virt_to_gfn(vaddr), 0); |
ccc9d90a WL |
389 | if (err < 0) { |
390 | xenbus_dev_fatal(dev, err, | |
391 | "granting access to ring page"); | |
392 | goto fail; | |
393 | } | |
394 | grefs[i] = err; | |
c9fd55eb | 395 | |
7d567928 | 396 | vaddr = vaddr + XEN_PAGE_SIZE; |
ccc9d90a WL |
397 | } |
398 | ||
399 | return 0; | |
400 | ||
401 | fail: | |
402 | for (j = 0; j < i; j++) | |
403 | gnttab_end_foreign_access_ref(grefs[j], 0); | |
4bac07c9 JF |
404 | return err; |
405 | } | |
406 | EXPORT_SYMBOL_GPL(xenbus_grant_ring); | |
407 | ||
408 | ||
409 | /** | |
410 | * Allocate an event channel for the given xenbus_device, assigning the newly | |
411 | * created local port to *port. Return 0 on success, or -errno on error. On | |
412 | * error, the device will switch to XenbusStateClosing, and the error will be | |
413 | * saved in the store. | |
414 | */ | |
415 | int xenbus_alloc_evtchn(struct xenbus_device *dev, int *port) | |
416 | { | |
417 | struct evtchn_alloc_unbound alloc_unbound; | |
418 | int err; | |
419 | ||
420 | alloc_unbound.dom = DOMID_SELF; | |
421 | alloc_unbound.remote_dom = dev->otherend_id; | |
422 | ||
423 | err = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound, | |
424 | &alloc_unbound); | |
425 | if (err) | |
426 | xenbus_dev_fatal(dev, err, "allocating event channel"); | |
427 | else | |
428 | *port = alloc_unbound.port; | |
429 | ||
430 | return err; | |
431 | } | |
432 | EXPORT_SYMBOL_GPL(xenbus_alloc_evtchn); | |
433 | ||
434 | ||
4bac07c9 JF |
435 | /** |
436 | * Free an existing event channel. Returns 0 on success or -errno on error. | |
437 | */ | |
438 | int xenbus_free_evtchn(struct xenbus_device *dev, int port) | |
439 | { | |
440 | struct evtchn_close close; | |
441 | int err; | |
442 | ||
443 | close.port = port; | |
444 | ||
445 | err = HYPERVISOR_event_channel_op(EVTCHNOP_close, &close); | |
446 | if (err) | |
447 | xenbus_dev_error(dev, err, "freeing event channel %d", port); | |
448 | ||
449 | return err; | |
450 | } | |
451 | EXPORT_SYMBOL_GPL(xenbus_free_evtchn); | |
452 | ||
453 | ||
454 | /** | |
455 | * xenbus_map_ring_valloc | |
456 | * @dev: xenbus device | |
ccc9d90a WL |
457 | * @gnt_refs: grant reference array |
458 | * @nr_grefs: number of grant references | |
4bac07c9 JF |
459 | * @vaddr: pointer to address to be filled out by mapping |
460 | * | |
ccc9d90a WL |
461 | * Map @nr_grefs pages of memory into this domain from another |
462 | * domain's grant table. xenbus_map_ring_valloc allocates @nr_grefs | |
463 | * pages of virtual address space, maps the pages to that address, and | |
464 | * sets *vaddr to that address. Returns 0 on success, and GNTST_* | |
465 | * (see xen/include/interface/grant_table.h) or -ENOMEM / -EINVAL on | |
466 | * error. If an error is returned, device will switch to | |
4bac07c9 JF |
467 | * XenbusStateClosing and the error message will be saved in XenStore. |
468 | */ | |
ccc9d90a WL |
469 | int xenbus_map_ring_valloc(struct xenbus_device *dev, grant_ref_t *gnt_refs, |
470 | unsigned int nr_grefs, void **vaddr) | |
2c5d37d3 | 471 | { |
ccc9d90a | 472 | return ring_ops->map(dev, gnt_refs, nr_grefs, vaddr); |
2c5d37d3 DDG |
473 | } |
474 | EXPORT_SYMBOL_GPL(xenbus_map_ring_valloc); | |
475 | ||
ccc9d90a WL |
476 | /* N.B. sizeof(phys_addr_t) doesn't always equal to sizeof(unsigned |
477 | * long), e.g. 32-on-64. Caller is responsible for preparing the | |
478 | * right array to feed into this function */ | |
479 | static int __xenbus_map_ring(struct xenbus_device *dev, | |
480 | grant_ref_t *gnt_refs, | |
481 | unsigned int nr_grefs, | |
482 | grant_handle_t *handles, | |
483 | phys_addr_t *addrs, | |
484 | unsigned int flags, | |
485 | bool *leaked) | |
486 | { | |
9cce2914 JG |
487 | struct gnttab_map_grant_ref map[XENBUS_MAX_RING_GRANTS]; |
488 | struct gnttab_unmap_grant_ref unmap[XENBUS_MAX_RING_GRANTS]; | |
ccc9d90a WL |
489 | int i, j; |
490 | int err = GNTST_okay; | |
491 | ||
9cce2914 | 492 | if (nr_grefs > XENBUS_MAX_RING_GRANTS) |
ccc9d90a WL |
493 | return -EINVAL; |
494 | ||
495 | for (i = 0; i < nr_grefs; i++) { | |
496 | memset(&map[i], 0, sizeof(map[i])); | |
497 | gnttab_set_map_op(&map[i], addrs[i], flags, gnt_refs[i], | |
498 | dev->otherend_id); | |
499 | handles[i] = INVALID_GRANT_HANDLE; | |
500 | } | |
501 | ||
502 | gnttab_batch_map(map, i); | |
503 | ||
504 | for (i = 0; i < nr_grefs; i++) { | |
505 | if (map[i].status != GNTST_okay) { | |
506 | err = map[i].status; | |
507 | xenbus_dev_fatal(dev, map[i].status, | |
508 | "mapping in shared page %d from domain %d", | |
509 | gnt_refs[i], dev->otherend_id); | |
510 | goto fail; | |
511 | } else | |
512 | handles[i] = map[i].handle; | |
513 | } | |
514 | ||
515 | return GNTST_okay; | |
516 | ||
517 | fail: | |
518 | for (i = j = 0; i < nr_grefs; i++) { | |
519 | if (handles[i] != INVALID_GRANT_HANDLE) { | |
520 | memset(&unmap[j], 0, sizeof(unmap[j])); | |
521 | gnttab_set_unmap_op(&unmap[j], (phys_addr_t)addrs[i], | |
522 | GNTMAP_host_map, handles[i]); | |
523 | j++; | |
524 | } | |
525 | } | |
526 | ||
527 | if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, unmap, j)) | |
528 | BUG(); | |
529 | ||
530 | *leaked = false; | |
531 | for (i = 0; i < j; i++) { | |
532 | if (unmap[i].status != GNTST_okay) { | |
533 | *leaked = true; | |
534 | break; | |
535 | } | |
536 | } | |
537 | ||
538 | return err; | |
539 | } | |
540 | ||
2c5d37d3 | 541 | static int xenbus_map_ring_valloc_pv(struct xenbus_device *dev, |
ccc9d90a WL |
542 | grant_ref_t *gnt_refs, |
543 | unsigned int nr_grefs, | |
544 | void **vaddr) | |
4bac07c9 | 545 | { |
2c5d37d3 | 546 | struct xenbus_map_node *node; |
4bac07c9 | 547 | struct vm_struct *area; |
9cce2914 JG |
548 | pte_t *ptes[XENBUS_MAX_RING_GRANTS]; |
549 | phys_addr_t phys_addrs[XENBUS_MAX_RING_GRANTS]; | |
ccc9d90a WL |
550 | int err = GNTST_okay; |
551 | int i; | |
552 | bool leaked; | |
4bac07c9 JF |
553 | |
554 | *vaddr = NULL; | |
555 | ||
9cce2914 | 556 | if (nr_grefs > XENBUS_MAX_RING_GRANTS) |
ccc9d90a WL |
557 | return -EINVAL; |
558 | ||
2c5d37d3 DDG |
559 | node = kzalloc(sizeof(*node), GFP_KERNEL); |
560 | if (!node) | |
561 | return -ENOMEM; | |
562 | ||
7d567928 | 563 | area = alloc_vm_area(XEN_PAGE_SIZE * nr_grefs, ptes); |
2c5d37d3 DDG |
564 | if (!area) { |
565 | kfree(node); | |
4bac07c9 | 566 | return -ENOMEM; |
2c5d37d3 | 567 | } |
4bac07c9 | 568 | |
ccc9d90a WL |
569 | for (i = 0; i < nr_grefs; i++) |
570 | phys_addrs[i] = arbitrary_virt_to_machine(ptes[i]).maddr; | |
4bac07c9 | 571 | |
ccc9d90a WL |
572 | err = __xenbus_map_ring(dev, gnt_refs, nr_grefs, node->handles, |
573 | phys_addrs, | |
574 | GNTMAP_host_map | GNTMAP_contains_pte, | |
575 | &leaked); | |
576 | if (err) | |
577 | goto failed; | |
4bac07c9 | 578 | |
ccc9d90a WL |
579 | node->nr_handles = nr_grefs; |
580 | node->pv.area = area; | |
2c5d37d3 DDG |
581 | |
582 | spin_lock(&xenbus_valloc_lock); | |
583 | list_add(&node->next, &xenbus_valloc_pages); | |
584 | spin_unlock(&xenbus_valloc_lock); | |
4bac07c9 JF |
585 | |
586 | *vaddr = area->addr; | |
587 | return 0; | |
ccc9d90a WL |
588 | |
589 | failed: | |
590 | if (!leaked) | |
591 | free_vm_area(area); | |
592 | else | |
593 | pr_alert("leaking VM area %p size %u page(s)", area, nr_grefs); | |
594 | ||
595 | kfree(node); | |
596 | return err; | |
4bac07c9 | 597 | } |
2c5d37d3 | 598 | |
89bf4b4e JG |
599 | struct map_ring_valloc_hvm |
600 | { | |
601 | unsigned int idx; | |
602 | ||
603 | /* Why do we need two arrays? See comment of __xenbus_map_ring */ | |
604 | phys_addr_t phys_addrs[XENBUS_MAX_RING_GRANTS]; | |
605 | unsigned long addrs[XENBUS_MAX_RING_GRANTS]; | |
606 | }; | |
607 | ||
608 | static void xenbus_map_ring_setup_grant_hvm(unsigned long gfn, | |
609 | unsigned int goffset, | |
610 | unsigned int len, | |
611 | void *data) | |
612 | { | |
613 | struct map_ring_valloc_hvm *info = data; | |
614 | unsigned long vaddr = (unsigned long)gfn_to_virt(gfn); | |
615 | ||
616 | info->phys_addrs[info->idx] = vaddr; | |
617 | info->addrs[info->idx] = vaddr; | |
618 | ||
619 | info->idx++; | |
620 | } | |
621 | ||
2c5d37d3 | 622 | static int xenbus_map_ring_valloc_hvm(struct xenbus_device *dev, |
ccc9d90a WL |
623 | grant_ref_t *gnt_ref, |
624 | unsigned int nr_grefs, | |
625 | void **vaddr) | |
2c5d37d3 DDG |
626 | { |
627 | struct xenbus_map_node *node; | |
628 | int err; | |
629 | void *addr; | |
ccc9d90a | 630 | bool leaked = false; |
89bf4b4e JG |
631 | struct map_ring_valloc_hvm info = { |
632 | .idx = 0, | |
633 | }; | |
634 | unsigned int nr_pages = XENBUS_PAGES(nr_grefs); | |
ccc9d90a | 635 | |
9cce2914 | 636 | if (nr_grefs > XENBUS_MAX_RING_GRANTS) |
ccc9d90a | 637 | return -EINVAL; |
2c5d37d3 DDG |
638 | |
639 | *vaddr = NULL; | |
640 | ||
641 | node = kzalloc(sizeof(*node), GFP_KERNEL); | |
642 | if (!node) | |
643 | return -ENOMEM; | |
644 | ||
89bf4b4e | 645 | err = alloc_xenballooned_pages(nr_pages, node->hvm.pages); |
2c5d37d3 DDG |
646 | if (err) |
647 | goto out_err; | |
648 | ||
89bf4b4e JG |
649 | gnttab_foreach_grant(node->hvm.pages, nr_grefs, |
650 | xenbus_map_ring_setup_grant_hvm, | |
651 | &info); | |
ccc9d90a WL |
652 | |
653 | err = __xenbus_map_ring(dev, gnt_ref, nr_grefs, node->handles, | |
89bf4b4e | 654 | info.phys_addrs, GNTMAP_host_map, &leaked); |
ccc9d90a | 655 | node->nr_handles = nr_grefs; |
2c5d37d3 | 656 | |
2c5d37d3 | 657 | if (err) |
ccc9d90a WL |
658 | goto out_free_ballooned_pages; |
659 | ||
89bf4b4e | 660 | addr = vmap(node->hvm.pages, nr_pages, VM_MAP | VM_IOREMAP, |
ccc9d90a WL |
661 | PAGE_KERNEL); |
662 | if (!addr) { | |
663 | err = -ENOMEM; | |
664 | goto out_xenbus_unmap_ring; | |
665 | } | |
666 | ||
667 | node->hvm.addr = addr; | |
2c5d37d3 DDG |
668 | |
669 | spin_lock(&xenbus_valloc_lock); | |
670 | list_add(&node->next, &xenbus_valloc_pages); | |
671 | spin_unlock(&xenbus_valloc_lock); | |
672 | ||
673 | *vaddr = addr; | |
674 | return 0; | |
675 | ||
ccc9d90a WL |
676 | out_xenbus_unmap_ring: |
677 | if (!leaked) | |
89bf4b4e | 678 | xenbus_unmap_ring(dev, node->handles, nr_grefs, info.addrs); |
ccc9d90a WL |
679 | else |
680 | pr_alert("leaking %p size %u page(s)", | |
89bf4b4e | 681 | addr, nr_pages); |
ccc9d90a WL |
682 | out_free_ballooned_pages: |
683 | if (!leaked) | |
89bf4b4e | 684 | free_xenballooned_pages(nr_pages, node->hvm.pages); |
8d0b8801 | 685 | out_err: |
2c5d37d3 DDG |
686 | kfree(node); |
687 | return err; | |
688 | } | |
4bac07c9 JF |
689 | |
690 | ||
691 | /** | |
692 | * xenbus_map_ring | |
693 | * @dev: xenbus device | |
ccc9d90a WL |
694 | * @gnt_refs: grant reference array |
695 | * @nr_grefs: number of grant reference | |
696 | * @handles: pointer to grant handle to be filled | |
697 | * @vaddrs: addresses to be mapped to | |
698 | * @leaked: fail to clean up a failed map, caller should not free vaddr | |
4bac07c9 | 699 | * |
ccc9d90a | 700 | * Map pages of memory into this domain from another domain's grant table. |
4bac07c9 | 701 | * xenbus_map_ring does not allocate the virtual address space (you must do |
ccc9d90a | 702 | * this yourself!). It only maps in the pages to the specified address. |
4bac07c9 | 703 | * Returns 0 on success, and GNTST_* (see xen/include/interface/grant_table.h) |
ccc9d90a WL |
704 | * or -ENOMEM / -EINVAL on error. If an error is returned, device will switch to |
705 | * XenbusStateClosing and the first error message will be saved in XenStore. | |
706 | * Further more if we fail to map the ring, caller should check @leaked. | |
707 | * If @leaked is not zero it means xenbus_map_ring fails to clean up, caller | |
708 | * should not free the address space of @vaddr. | |
4bac07c9 | 709 | */ |
ccc9d90a WL |
710 | int xenbus_map_ring(struct xenbus_device *dev, grant_ref_t *gnt_refs, |
711 | unsigned int nr_grefs, grant_handle_t *handles, | |
712 | unsigned long *vaddrs, bool *leaked) | |
4bac07c9 | 713 | { |
9cce2914 | 714 | phys_addr_t phys_addrs[XENBUS_MAX_RING_GRANTS]; |
ccc9d90a | 715 | int i; |
4bac07c9 | 716 | |
9cce2914 | 717 | if (nr_grefs > XENBUS_MAX_RING_GRANTS) |
ccc9d90a | 718 | return -EINVAL; |
4bac07c9 | 719 | |
ccc9d90a WL |
720 | for (i = 0; i < nr_grefs; i++) |
721 | phys_addrs[i] = (unsigned long)vaddrs[i]; | |
4bac07c9 | 722 | |
ccc9d90a WL |
723 | return __xenbus_map_ring(dev, gnt_refs, nr_grefs, handles, |
724 | phys_addrs, GNTMAP_host_map, leaked); | |
4bac07c9 JF |
725 | } |
726 | EXPORT_SYMBOL_GPL(xenbus_map_ring); | |
727 | ||
728 | ||
729 | /** | |
730 | * xenbus_unmap_ring_vfree | |
731 | * @dev: xenbus device | |
732 | * @vaddr: addr to unmap | |
733 | * | |
734 | * Based on Rusty Russell's skeleton driver's unmap_page. | |
735 | * Unmap a page of memory in this domain that was imported from another domain. | |
736 | * Use xenbus_unmap_ring_vfree if you mapped in your memory with | |
737 | * xenbus_map_ring_valloc (it will free the virtual address space). | |
738 | * Returns 0 on success and returns GNTST_* on error | |
739 | * (see xen/include/interface/grant_table.h). | |
740 | */ | |
741 | int xenbus_unmap_ring_vfree(struct xenbus_device *dev, void *vaddr) | |
742 | { | |
2c5d37d3 DDG |
743 | return ring_ops->unmap(dev, vaddr); |
744 | } | |
745 | EXPORT_SYMBOL_GPL(xenbus_unmap_ring_vfree); | |
746 | ||
747 | static int xenbus_unmap_ring_vfree_pv(struct xenbus_device *dev, void *vaddr) | |
748 | { | |
749 | struct xenbus_map_node *node; | |
9cce2914 | 750 | struct gnttab_unmap_grant_ref unmap[XENBUS_MAX_RING_GRANTS]; |
cd12909c | 751 | unsigned int level; |
ccc9d90a WL |
752 | int i; |
753 | bool leaked = false; | |
754 | int err; | |
4bac07c9 | 755 | |
2c5d37d3 DDG |
756 | spin_lock(&xenbus_valloc_lock); |
757 | list_for_each_entry(node, &xenbus_valloc_pages, next) { | |
ccc9d90a | 758 | if (node->pv.area->addr == vaddr) { |
2c5d37d3 DDG |
759 | list_del(&node->next); |
760 | goto found; | |
761 | } | |
4bac07c9 | 762 | } |
2c5d37d3 DDG |
763 | node = NULL; |
764 | found: | |
765 | spin_unlock(&xenbus_valloc_lock); | |
4bac07c9 | 766 | |
2c5d37d3 | 767 | if (!node) { |
4bac07c9 JF |
768 | xenbus_dev_error(dev, -ENOENT, |
769 | "can't find mapped virtual address %p", vaddr); | |
770 | return GNTST_bad_virt_addr; | |
771 | } | |
772 | ||
ccc9d90a WL |
773 | for (i = 0; i < node->nr_handles; i++) { |
774 | unsigned long addr; | |
775 | ||
776 | memset(&unmap[i], 0, sizeof(unmap[i])); | |
7d567928 | 777 | addr = (unsigned long)vaddr + (XEN_PAGE_SIZE * i); |
ccc9d90a WL |
778 | unmap[i].host_addr = arbitrary_virt_to_machine( |
779 | lookup_address(addr, &level)).maddr; | |
780 | unmap[i].dev_bus_addr = 0; | |
781 | unmap[i].handle = node->handles[i]; | |
782 | } | |
4bac07c9 | 783 | |
ccc9d90a | 784 | if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, unmap, i)) |
4bac07c9 JF |
785 | BUG(); |
786 | ||
ccc9d90a WL |
787 | err = GNTST_okay; |
788 | leaked = false; | |
789 | for (i = 0; i < node->nr_handles; i++) { | |
790 | if (unmap[i].status != GNTST_okay) { | |
791 | leaked = true; | |
792 | xenbus_dev_error(dev, unmap[i].status, | |
793 | "unmapping page at handle %d error %d", | |
794 | node->handles[i], unmap[i].status); | |
795 | err = unmap[i].status; | |
796 | break; | |
797 | } | |
798 | } | |
799 | ||
800 | if (!leaked) | |
801 | free_vm_area(node->pv.area); | |
4bac07c9 | 802 | else |
ccc9d90a WL |
803 | pr_alert("leaking VM area %p size %u page(s)", |
804 | node->pv.area, node->nr_handles); | |
4bac07c9 | 805 | |
2c5d37d3 | 806 | kfree(node); |
ccc9d90a | 807 | return err; |
4bac07c9 | 808 | } |
4bac07c9 | 809 | |
89bf4b4e JG |
810 | struct unmap_ring_vfree_hvm |
811 | { | |
812 | unsigned int idx; | |
813 | unsigned long addrs[XENBUS_MAX_RING_GRANTS]; | |
814 | }; | |
815 | ||
816 | static void xenbus_unmap_ring_setup_grant_hvm(unsigned long gfn, | |
817 | unsigned int goffset, | |
818 | unsigned int len, | |
819 | void *data) | |
820 | { | |
821 | struct unmap_ring_vfree_hvm *info = data; | |
822 | ||
823 | info->addrs[info->idx] = (unsigned long)gfn_to_virt(gfn); | |
824 | ||
825 | info->idx++; | |
826 | } | |
827 | ||
2c5d37d3 DDG |
828 | static int xenbus_unmap_ring_vfree_hvm(struct xenbus_device *dev, void *vaddr) |
829 | { | |
830 | int rv; | |
831 | struct xenbus_map_node *node; | |
832 | void *addr; | |
89bf4b4e JG |
833 | struct unmap_ring_vfree_hvm info = { |
834 | .idx = 0, | |
835 | }; | |
836 | unsigned int nr_pages; | |
2c5d37d3 DDG |
837 | |
838 | spin_lock(&xenbus_valloc_lock); | |
839 | list_for_each_entry(node, &xenbus_valloc_pages, next) { | |
ccc9d90a | 840 | addr = node->hvm.addr; |
2c5d37d3 DDG |
841 | if (addr == vaddr) { |
842 | list_del(&node->next); | |
843 | goto found; | |
844 | } | |
845 | } | |
5ac08001 | 846 | node = addr = NULL; |
2c5d37d3 DDG |
847 | found: |
848 | spin_unlock(&xenbus_valloc_lock); | |
849 | ||
850 | if (!node) { | |
851 | xenbus_dev_error(dev, -ENOENT, | |
852 | "can't find mapped virtual address %p", vaddr); | |
853 | return GNTST_bad_virt_addr; | |
854 | } | |
855 | ||
89bf4b4e JG |
856 | nr_pages = XENBUS_PAGES(node->nr_handles); |
857 | ||
858 | gnttab_foreach_grant(node->hvm.pages, node->nr_handles, | |
859 | xenbus_unmap_ring_setup_grant_hvm, | |
860 | &info); | |
2c5d37d3 | 861 | |
ccc9d90a | 862 | rv = xenbus_unmap_ring(dev, node->handles, node->nr_handles, |
89bf4b4e | 863 | info.addrs); |
c22fe519 | 864 | if (!rv) { |
ccc9d90a | 865 | vunmap(vaddr); |
89bf4b4e | 866 | free_xenballooned_pages(nr_pages, node->hvm.pages); |
c22fe519 | 867 | } |
2c5d37d3 | 868 | else |
89bf4b4e | 869 | WARN(1, "Leaking %p, size %u page(s)\n", vaddr, nr_pages); |
2c5d37d3 DDG |
870 | |
871 | kfree(node); | |
872 | return rv; | |
873 | } | |
4bac07c9 JF |
874 | |
875 | /** | |
876 | * xenbus_unmap_ring | |
877 | * @dev: xenbus device | |
ccc9d90a WL |
878 | * @handles: grant handle array |
879 | * @nr_handles: number of handles in the array | |
880 | * @vaddrs: addresses to unmap | |
4bac07c9 | 881 | * |
ccc9d90a | 882 | * Unmap memory in this domain that was imported from another domain. |
4bac07c9 JF |
883 | * Returns 0 on success and returns GNTST_* on error |
884 | * (see xen/include/interface/grant_table.h). | |
885 | */ | |
886 | int xenbus_unmap_ring(struct xenbus_device *dev, | |
ccc9d90a WL |
887 | grant_handle_t *handles, unsigned int nr_handles, |
888 | unsigned long *vaddrs) | |
4bac07c9 | 889 | { |
9cce2914 | 890 | struct gnttab_unmap_grant_ref unmap[XENBUS_MAX_RING_GRANTS]; |
ccc9d90a WL |
891 | int i; |
892 | int err; | |
2946a52a | 893 | |
9cce2914 | 894 | if (nr_handles > XENBUS_MAX_RING_GRANTS) |
ccc9d90a | 895 | return -EINVAL; |
4bac07c9 | 896 | |
ccc9d90a WL |
897 | for (i = 0; i < nr_handles; i++) |
898 | gnttab_set_unmap_op(&unmap[i], vaddrs[i], | |
899 | GNTMAP_host_map, handles[i]); | |
900 | ||
901 | if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, unmap, i)) | |
4bac07c9 JF |
902 | BUG(); |
903 | ||
ccc9d90a WL |
904 | err = GNTST_okay; |
905 | for (i = 0; i < nr_handles; i++) { | |
906 | if (unmap[i].status != GNTST_okay) { | |
907 | xenbus_dev_error(dev, unmap[i].status, | |
908 | "unmapping page at handle %d error %d", | |
909 | handles[i], unmap[i].status); | |
910 | err = unmap[i].status; | |
911 | break; | |
912 | } | |
913 | } | |
4bac07c9 | 914 | |
ccc9d90a | 915 | return err; |
4bac07c9 JF |
916 | } |
917 | EXPORT_SYMBOL_GPL(xenbus_unmap_ring); | |
918 | ||
919 | ||
920 | /** | |
921 | * xenbus_read_driver_state | |
922 | * @path: path for driver | |
923 | * | |
924 | * Return the state of the driver rooted at the given store path, or | |
925 | * XenbusStateUnknown if no state can be read. | |
926 | */ | |
927 | enum xenbus_state xenbus_read_driver_state(const char *path) | |
928 | { | |
929 | enum xenbus_state result; | |
930 | int err = xenbus_gather(XBT_NIL, path, "state", "%d", &result, NULL); | |
931 | if (err) | |
932 | result = XenbusStateUnknown; | |
933 | ||
934 | return result; | |
935 | } | |
936 | EXPORT_SYMBOL_GPL(xenbus_read_driver_state); | |
2c5d37d3 DDG |
937 | |
938 | static const struct xenbus_ring_ops ring_ops_pv = { | |
939 | .map = xenbus_map_ring_valloc_pv, | |
940 | .unmap = xenbus_unmap_ring_vfree_pv, | |
941 | }; | |
942 | ||
943 | static const struct xenbus_ring_ops ring_ops_hvm = { | |
944 | .map = xenbus_map_ring_valloc_hvm, | |
945 | .unmap = xenbus_unmap_ring_vfree_hvm, | |
946 | }; | |
947 | ||
948 | void __init xenbus_ring_ops_init(void) | |
949 | { | |
be3e9cf3 | 950 | if (!xen_feature(XENFEAT_auto_translated_physmap)) |
2c5d37d3 DDG |
951 | ring_ops = &ring_ops_pv; |
952 | else | |
953 | ring_ops = &ring_ops_hvm; | |
954 | } |