diff options
Diffstat (limited to 'drivers/lguest/lguest_device.c')
-rw-r--r-- | drivers/lguest/lguest_device.c | 540 |
1 files changed, 0 insertions, 540 deletions
diff --git a/drivers/lguest/lguest_device.c b/drivers/lguest/lguest_device.c deleted file mode 100644 index 89088d6538fd..000000000000 --- a/drivers/lguest/lguest_device.c +++ /dev/null | |||
@@ -1,540 +0,0 @@ | |||
1 | /*P:050 | ||
2 | * Lguest guests use a very simple method to describe devices. It's a | ||
3 | * series of device descriptors contained just above the top of normal Guest | ||
4 | * memory. | ||
5 | * | ||
6 | * We use the standard "virtio" device infrastructure, which provides us with a | ||
7 | * console, a network and a block driver. Each one expects some configuration | ||
8 | * information and a "virtqueue" or two to send and receive data. | ||
9 | :*/ | ||
10 | #include <linux/init.h> | ||
11 | #include <linux/bootmem.h> | ||
12 | #include <linux/lguest_launcher.h> | ||
13 | #include <linux/virtio.h> | ||
14 | #include <linux/virtio_config.h> | ||
15 | #include <linux/interrupt.h> | ||
16 | #include <linux/virtio_ring.h> | ||
17 | #include <linux/err.h> | ||
18 | #include <linux/export.h> | ||
19 | #include <linux/slab.h> | ||
20 | #include <asm/io.h> | ||
21 | #include <asm/paravirt.h> | ||
22 | #include <asm/lguest_hcall.h> | ||
23 | |||
24 | /* The pointer to our (page) of device descriptions. */ | ||
25 | static void *lguest_devices; | ||
26 | |||
27 | /* | ||
28 | * For Guests, device memory can be used as normal memory, so we cast away the | ||
29 | * __iomem to quieten sparse. | ||
30 | */ | ||
31 | static inline void *lguest_map(unsigned long phys_addr, unsigned long pages) | ||
32 | { | ||
33 | return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages); | ||
34 | } | ||
35 | |||
36 | static inline void lguest_unmap(void *addr) | ||
37 | { | ||
38 | iounmap((__force void __iomem *)addr); | ||
39 | } | ||
40 | |||
41 | /*D:100 | ||
42 | * Each lguest device is just a virtio device plus a pointer to its entry | ||
43 | * in the lguest_devices page. | ||
44 | */ | ||
45 | struct lguest_device { | ||
46 | struct virtio_device vdev; | ||
47 | |||
48 | /* The entry in the lguest_devices page for this device. */ | ||
49 | struct lguest_device_desc *desc; | ||
50 | }; | ||
51 | |||
52 | /* | ||
53 | * Since the virtio infrastructure hands us a pointer to the virtio_device all | ||
54 | * the time, it helps to have a curt macro to get a pointer to the struct | ||
55 | * lguest_device it's enclosed in. | ||
56 | */ | ||
57 | #define to_lgdev(vd) container_of(vd, struct lguest_device, vdev) | ||
58 | |||
59 | /*D:130 | ||
60 | * Device configurations | ||
61 | * | ||
62 | * The configuration information for a device consists of one or more | ||
63 | * virtqueues, a feature bitmap, and some configuration bytes. The | ||
64 | * configuration bytes don't really matter to us: the Launcher sets them up, and | ||
65 | * the driver will look at them during setup. | ||
66 | * | ||
67 | * A convenient routine to return the device's virtqueue config array: | ||
68 | * immediately after the descriptor. | ||
69 | */ | ||
70 | static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc) | ||
71 | { | ||
72 | return (void *)(desc + 1); | ||
73 | } | ||
74 | |||
75 | /* The features come immediately after the virtqueues. */ | ||
76 | static u8 *lg_features(const struct lguest_device_desc *desc) | ||
77 | { | ||
78 | return (void *)(lg_vq(desc) + desc->num_vq); | ||
79 | } | ||
80 | |||
81 | /* The config space comes after the two feature bitmasks. */ | ||
82 | static u8 *lg_config(const struct lguest_device_desc *desc) | ||
83 | { | ||
84 | return lg_features(desc) + desc->feature_len * 2; | ||
85 | } | ||
86 | |||
87 | /* The total size of the config page used by this device (incl. desc) */ | ||
88 | static unsigned desc_size(const struct lguest_device_desc *desc) | ||
89 | { | ||
90 | return sizeof(*desc) | ||
91 | + desc->num_vq * sizeof(struct lguest_vqconfig) | ||
92 | + desc->feature_len * 2 | ||
93 | + desc->config_len; | ||
94 | } | ||
95 | |||
96 | /* This gets the device's feature bits. */ | ||
97 | static u64 lg_get_features(struct virtio_device *vdev) | ||
98 | { | ||
99 | unsigned int i; | ||
100 | u32 features = 0; | ||
101 | struct lguest_device_desc *desc = to_lgdev(vdev)->desc; | ||
102 | u8 *in_features = lg_features(desc); | ||
103 | |||
104 | /* We do this the slow but generic way. */ | ||
105 | for (i = 0; i < min(desc->feature_len * 8, 32); i++) | ||
106 | if (in_features[i / 8] & (1 << (i % 8))) | ||
107 | features |= (1 << i); | ||
108 | |||
109 | return features; | ||
110 | } | ||
111 | |||
112 | /* | ||
113 | * To notify on reset or feature finalization, we (ab)use the NOTIFY | ||
114 | * hypercall, with the descriptor address of the device. | ||
115 | */ | ||
116 | static void status_notify(struct virtio_device *vdev) | ||
117 | { | ||
118 | unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices; | ||
119 | |||
120 | hcall(LHCALL_NOTIFY, (max_pfn << PAGE_SHIFT) + offset, 0, 0, 0); | ||
121 | } | ||
122 | |||
123 | /* | ||
124 | * The virtio core takes the features the Host offers, and copies the ones | ||
125 | * supported by the driver into the vdev->features array. Once that's all | ||
126 | * sorted out, this routine is called so we can tell the Host which features we | ||
127 | * understand and accept. | ||
128 | */ | ||
129 | static int lg_finalize_features(struct virtio_device *vdev) | ||
130 | { | ||
131 | unsigned int i, bits; | ||
132 | struct lguest_device_desc *desc = to_lgdev(vdev)->desc; | ||
133 | /* Second half of bitmap is features we accept. */ | ||
134 | u8 *out_features = lg_features(desc) + desc->feature_len; | ||
135 | |||
136 | /* Give virtio_ring a chance to accept features. */ | ||
137 | vring_transport_features(vdev); | ||
138 | |||
139 | /* Make sure we don't have any features > 32 bits! */ | ||
140 | BUG_ON((u32)vdev->features != vdev->features); | ||
141 | |||
142 | /* | ||
143 | * Since lguest is currently x86-only, we're little-endian. That | ||
144 | * means we could just memcpy. But it's not time critical, and in | ||
145 | * case someone copies this code, we do it the slow, obvious way. | ||
146 | */ | ||
147 | memset(out_features, 0, desc->feature_len); | ||
148 | bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8; | ||
149 | for (i = 0; i < bits; i++) { | ||
150 | if (__virtio_test_bit(vdev, i)) | ||
151 | out_features[i / 8] |= (1 << (i % 8)); | ||
152 | } | ||
153 | |||
154 | /* Tell Host we've finished with this device's feature negotiation */ | ||
155 | status_notify(vdev); | ||
156 | |||
157 | return 0; | ||
158 | } | ||
159 | |||
160 | /* Once they've found a field, getting a copy of it is easy. */ | ||
161 | static void lg_get(struct virtio_device *vdev, unsigned int offset, | ||
162 | void *buf, unsigned len) | ||
163 | { | ||
164 | struct lguest_device_desc *desc = to_lgdev(vdev)->desc; | ||
165 | |||
166 | /* Check they didn't ask for more than the length of the config! */ | ||
167 | BUG_ON(offset + len > desc->config_len); | ||
168 | memcpy(buf, lg_config(desc) + offset, len); | ||
169 | } | ||
170 | |||
171 | /* Setting the contents is also trivial. */ | ||
172 | static void lg_set(struct virtio_device *vdev, unsigned int offset, | ||
173 | const void *buf, unsigned len) | ||
174 | { | ||
175 | struct lguest_device_desc *desc = to_lgdev(vdev)->desc; | ||
176 | |||
177 | /* Check they didn't ask for more than the length of the config! */ | ||
178 | BUG_ON(offset + len > desc->config_len); | ||
179 | memcpy(lg_config(desc) + offset, buf, len); | ||
180 | } | ||
181 | |||
182 | /* | ||
183 | * The operations to get and set the status word just access the status field | ||
184 | * of the device descriptor. | ||
185 | */ | ||
186 | static u8 lg_get_status(struct virtio_device *vdev) | ||
187 | { | ||
188 | return to_lgdev(vdev)->desc->status; | ||
189 | } | ||
190 | |||
191 | static void lg_set_status(struct virtio_device *vdev, u8 status) | ||
192 | { | ||
193 | BUG_ON(!status); | ||
194 | to_lgdev(vdev)->desc->status = status; | ||
195 | |||
196 | /* Tell Host immediately if we failed. */ | ||
197 | if (status & VIRTIO_CONFIG_S_FAILED) | ||
198 | status_notify(vdev); | ||
199 | } | ||
200 | |||
201 | static void lg_reset(struct virtio_device *vdev) | ||
202 | { | ||
203 | /* 0 status means "reset" */ | ||
204 | to_lgdev(vdev)->desc->status = 0; | ||
205 | status_notify(vdev); | ||
206 | } | ||
207 | |||
208 | /* | ||
209 | * Virtqueues | ||
210 | * | ||
211 | * The other piece of infrastructure virtio needs is a "virtqueue": a way of | ||
212 | * the Guest device registering buffers for the other side to read from or | ||
213 | * write into (ie. send and receive buffers). Each device can have multiple | ||
214 | * virtqueues: for example the console driver uses one queue for sending and | ||
215 | * another for receiving. | ||
216 | * | ||
217 | * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue | ||
218 | * already exists in virtio_ring.c. We just need to connect it up. | ||
219 | * | ||
220 | * We start with the information we need to keep about each virtqueue. | ||
221 | */ | ||
222 | |||
223 | /*D:140 This is the information we remember about each virtqueue. */ | ||
224 | struct lguest_vq_info { | ||
225 | /* A copy of the information contained in the device config. */ | ||
226 | struct lguest_vqconfig config; | ||
227 | |||
228 | /* The address where we mapped the virtio ring, so we can unmap it. */ | ||
229 | void *pages; | ||
230 | }; | ||
231 | |||
232 | /* | ||
233 | * When the virtio_ring code wants to prod the Host, it calls us here and we | ||
234 | * make a hypercall. We hand the physical address of the virtqueue so the Host | ||
235 | * knows which virtqueue we're talking about. | ||
236 | */ | ||
237 | static bool lg_notify(struct virtqueue *vq) | ||
238 | { | ||
239 | /* | ||
240 | * We store our virtqueue information in the "priv" pointer of the | ||
241 | * virtqueue structure. | ||
242 | */ | ||
243 | struct lguest_vq_info *lvq = vq->priv; | ||
244 | |||
245 | hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0, 0); | ||
246 | return true; | ||
247 | } | ||
248 | |||
249 | /* An extern declaration inside a C file is bad form. Don't do it. */ | ||
250 | extern int lguest_setup_irq(unsigned int irq); | ||
251 | |||
252 | /* | ||
253 | * This routine finds the Nth virtqueue described in the configuration of | ||
254 | * this device and sets it up. | ||
255 | * | ||
256 | * This is kind of an ugly duckling. It'd be nicer to have a standard | ||
257 | * representation of a virtqueue in the configuration space, but it seems that | ||
258 | * everyone wants to do it differently. The KVM coders want the Guest to | ||
259 | * allocate its own pages and tell the Host where they are, but for lguest it's | ||
260 | * simpler for the Host to simply tell us where the pages are. | ||
261 | */ | ||
262 | static struct virtqueue *lg_find_vq(struct virtio_device *vdev, | ||
263 | unsigned index, | ||
264 | void (*callback)(struct virtqueue *vq), | ||
265 | const char *name) | ||
266 | { | ||
267 | struct lguest_device *ldev = to_lgdev(vdev); | ||
268 | struct lguest_vq_info *lvq; | ||
269 | struct virtqueue *vq; | ||
270 | int err; | ||
271 | |||
272 | if (!name) | ||
273 | return NULL; | ||
274 | |||
275 | /* We must have this many virtqueues. */ | ||
276 | if (index >= ldev->desc->num_vq) | ||
277 | return ERR_PTR(-ENOENT); | ||
278 | |||
279 | lvq = kmalloc(sizeof(*lvq), GFP_KERNEL); | ||
280 | if (!lvq) | ||
281 | return ERR_PTR(-ENOMEM); | ||
282 | |||
283 | /* | ||
284 | * Make a copy of the "struct lguest_vqconfig" entry, which sits after | ||
285 | * the descriptor. We need a copy because the config space might not | ||
286 | * be aligned correctly. | ||
287 | */ | ||
288 | memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config)); | ||
289 | |||
290 | printk("Mapping virtqueue %i addr %lx\n", index, | ||
291 | (unsigned long)lvq->config.pfn << PAGE_SHIFT); | ||
292 | /* Figure out how many pages the ring will take, and map that memory */ | ||
293 | lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT, | ||
294 | DIV_ROUND_UP(vring_size(lvq->config.num, | ||
295 | LGUEST_VRING_ALIGN), | ||
296 | PAGE_SIZE)); | ||
297 | if (!lvq->pages) { | ||
298 | err = -ENOMEM; | ||
299 | goto free_lvq; | ||
300 | } | ||
301 | |||
302 | /* | ||
303 | * OK, tell virtio_ring.c to set up a virtqueue now we know its size | ||
304 | * and we've got a pointer to its pages. Note that we set weak_barriers | ||
305 | * to 'true': the host just a(nother) SMP CPU, so we only need inter-cpu | ||
306 | * barriers. | ||
307 | */ | ||
308 | vq = vring_new_virtqueue(index, lvq->config.num, LGUEST_VRING_ALIGN, vdev, | ||
309 | true, lvq->pages, lg_notify, callback, name); | ||
310 | if (!vq) { | ||
311 | err = -ENOMEM; | ||
312 | goto unmap; | ||
313 | } | ||
314 | |||
315 | /* Make sure the interrupt is allocated. */ | ||
316 | err = lguest_setup_irq(lvq->config.irq); | ||
317 | if (err) | ||
318 | goto destroy_vring; | ||
319 | |||
320 | /* | ||
321 | * Tell the interrupt for this virtqueue to go to the virtio_ring | ||
322 | * interrupt handler. | ||
323 | * | ||
324 | * FIXME: We used to have a flag for the Host to tell us we could use | ||
325 | * the interrupt as a source of randomness: it'd be nice to have that | ||
326 | * back. | ||
327 | */ | ||
328 | err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED, | ||
329 | dev_name(&vdev->dev), vq); | ||
330 | if (err) | ||
331 | goto free_desc; | ||
332 | |||
333 | /* | ||
334 | * Last of all we hook up our 'struct lguest_vq_info" to the | ||
335 | * virtqueue's priv pointer. | ||
336 | */ | ||
337 | vq->priv = lvq; | ||
338 | return vq; | ||
339 | |||
340 | free_desc: | ||
341 | irq_free_desc(lvq->config.irq); | ||
342 | destroy_vring: | ||
343 | vring_del_virtqueue(vq); | ||
344 | unmap: | ||
345 | lguest_unmap(lvq->pages); | ||
346 | free_lvq: | ||
347 | kfree(lvq); | ||
348 | return ERR_PTR(err); | ||
349 | } | ||
350 | /*:*/ | ||
351 | |||
352 | /* Cleaning up a virtqueue is easy */ | ||
353 | static void lg_del_vq(struct virtqueue *vq) | ||
354 | { | ||
355 | struct lguest_vq_info *lvq = vq->priv; | ||
356 | |||
357 | /* Release the interrupt */ | ||
358 | free_irq(lvq->config.irq, vq); | ||
359 | /* Tell virtio_ring.c to free the virtqueue. */ | ||
360 | vring_del_virtqueue(vq); | ||
361 | /* Unmap the pages containing the ring. */ | ||
362 | lguest_unmap(lvq->pages); | ||
363 | /* Free our own queue information. */ | ||
364 | kfree(lvq); | ||
365 | } | ||
366 | |||
367 | static void lg_del_vqs(struct virtio_device *vdev) | ||
368 | { | ||
369 | struct virtqueue *vq, *n; | ||
370 | |||
371 | list_for_each_entry_safe(vq, n, &vdev->vqs, list) | ||
372 | lg_del_vq(vq); | ||
373 | } | ||
374 | |||
375 | static int lg_find_vqs(struct virtio_device *vdev, unsigned nvqs, | ||
376 | struct virtqueue *vqs[], | ||
377 | vq_callback_t *callbacks[], | ||
378 | const char *names[]) | ||
379 | { | ||
380 | struct lguest_device *ldev = to_lgdev(vdev); | ||
381 | int i; | ||
382 | |||
383 | /* We must have this many virtqueues. */ | ||
384 | if (nvqs > ldev->desc->num_vq) | ||
385 | return -ENOENT; | ||
386 | |||
387 | for (i = 0; i < nvqs; ++i) { | ||
388 | vqs[i] = lg_find_vq(vdev, i, callbacks[i], names[i]); | ||
389 | if (IS_ERR(vqs[i])) | ||
390 | goto error; | ||
391 | } | ||
392 | return 0; | ||
393 | |||
394 | error: | ||
395 | lg_del_vqs(vdev); | ||
396 | return PTR_ERR(vqs[i]); | ||
397 | } | ||
398 | |||
399 | static const char *lg_bus_name(struct virtio_device *vdev) | ||
400 | { | ||
401 | return ""; | ||
402 | } | ||
403 | |||
404 | /* The ops structure which hooks everything together. */ | ||
405 | static const struct virtio_config_ops lguest_config_ops = { | ||
406 | .get_features = lg_get_features, | ||
407 | .finalize_features = lg_finalize_features, | ||
408 | .get = lg_get, | ||
409 | .set = lg_set, | ||
410 | .get_status = lg_get_status, | ||
411 | .set_status = lg_set_status, | ||
412 | .reset = lg_reset, | ||
413 | .find_vqs = lg_find_vqs, | ||
414 | .del_vqs = lg_del_vqs, | ||
415 | .bus_name = lg_bus_name, | ||
416 | }; | ||
417 | |||
418 | /* | ||
419 | * The root device for the lguest virtio devices. This makes them appear as | ||
420 | * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. | ||
421 | */ | ||
422 | static struct device *lguest_root; | ||
423 | |||
424 | /*D:120 | ||
425 | * This is the core of the lguest bus: actually adding a new device. | ||
426 | * It's a separate function because it's neater that way, and because an | ||
427 | * earlier version of the code supported hotplug and unplug. They were removed | ||
428 | * early on because they were never used. | ||
429 | * | ||
430 | * As Andrew Tridgell says, "Untested code is buggy code". | ||
431 | * | ||
432 | * It's worth reading this carefully: we start with a pointer to the new device | ||
433 | * descriptor in the "lguest_devices" page, and the offset into the device | ||
434 | * descriptor page so we can uniquely identify it if things go badly wrong. | ||
435 | */ | ||
436 | static void add_lguest_device(struct lguest_device_desc *d, | ||
437 | unsigned int offset) | ||
438 | { | ||
439 | struct lguest_device *ldev; | ||
440 | |||
441 | /* Start with zeroed memory; Linux's device layer counts on it. */ | ||
442 | ldev = kzalloc(sizeof(*ldev), GFP_KERNEL); | ||
443 | if (!ldev) { | ||
444 | printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n", | ||
445 | offset, d->type); | ||
446 | return; | ||
447 | } | ||
448 | |||
449 | /* This devices' parent is the lguest/ dir. */ | ||
450 | ldev->vdev.dev.parent = lguest_root; | ||
451 | /* | ||
452 | * The device type comes straight from the descriptor. There's also a | ||
453 | * device vendor field in the virtio_device struct, which we leave as | ||
454 | * 0. | ||
455 | */ | ||
456 | ldev->vdev.id.device = d->type; | ||
457 | /* | ||
458 | * We have a simple set of routines for querying the device's | ||
459 | * configuration information and setting its status. | ||
460 | */ | ||
461 | ldev->vdev.config = &lguest_config_ops; | ||
462 | /* And we remember the device's descriptor for lguest_config_ops. */ | ||
463 | ldev->desc = d; | ||
464 | |||
465 | /* | ||
466 | * register_virtio_device() sets up the generic fields for the struct | ||
467 | * virtio_device and calls device_register(). This makes the bus | ||
468 | * infrastructure look for a matching driver. | ||
469 | */ | ||
470 | if (register_virtio_device(&ldev->vdev) != 0) { | ||
471 | printk(KERN_ERR "Failed to register lguest dev %u type %u\n", | ||
472 | offset, d->type); | ||
473 | kfree(ldev); | ||
474 | } | ||
475 | } | ||
476 | |||
477 | /*D:110 | ||
478 | * scan_devices() simply iterates through the device page. The type 0 is | ||
479 | * reserved to mean "end of devices". | ||
480 | */ | ||
481 | static void scan_devices(void) | ||
482 | { | ||
483 | unsigned int i; | ||
484 | struct lguest_device_desc *d; | ||
485 | |||
486 | /* We start at the page beginning, and skip over each entry. */ | ||
487 | for (i = 0; i < PAGE_SIZE; i += desc_size(d)) { | ||
488 | d = lguest_devices + i; | ||
489 | |||
490 | /* Once we hit a zero, stop. */ | ||
491 | if (d->type == 0) | ||
492 | break; | ||
493 | |||
494 | printk("Device at %i has size %u\n", i, desc_size(d)); | ||
495 | add_lguest_device(d, i); | ||
496 | } | ||
497 | } | ||
498 | |||
499 | /*D:105 | ||
500 | * Fairly early in boot, lguest_devices_init() is called to set up the | ||
501 | * lguest device infrastructure. We check that we are a Guest by checking | ||
502 | * pv_info.name: there are other ways of checking, but this seems most | ||
503 | * obvious to me. | ||
504 | * | ||
505 | * So we can access the "struct lguest_device_desc"s easily, we map that memory | ||
506 | * and store the pointer in the global "lguest_devices". Then we register a | ||
507 | * root device from which all our devices will hang (this seems to be the | ||
508 | * correct sysfs incantation). | ||
509 | * | ||
510 | * Finally we call scan_devices() which adds all the devices found in the | ||
511 | * lguest_devices page. | ||
512 | */ | ||
513 | static int __init lguest_devices_init(void) | ||
514 | { | ||
515 | if (strcmp(pv_info.name, "lguest") != 0) | ||
516 | return 0; | ||
517 | |||
518 | lguest_root = root_device_register("lguest"); | ||
519 | if (IS_ERR(lguest_root)) | ||
520 | panic("Could not register lguest root"); | ||
521 | |||
522 | /* Devices are in a single page above top of "normal" mem */ | ||
523 | lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1); | ||
524 | |||
525 | scan_devices(); | ||
526 | return 0; | ||
527 | } | ||
528 | /* We do this after core stuff, but before the drivers. */ | ||
529 | postcore_initcall(lguest_devices_init); | ||
530 | |||
531 | /*D:150 | ||
532 | * At this point in the journey we used to now wade through the lguest | ||
533 | * devices themselves: net, block and console. Since they're all now virtio | ||
534 | * devices rather than lguest-specific, I've decided to ignore them. Mostly, | ||
535 | * they're kind of boring. But this does mean you'll never experience the | ||
536 | * thrill of reading the forbidden love scene buried deep in the block driver. | ||
537 | * | ||
538 | * "make Launcher" beckons, where we answer questions like "Where do Guests | ||
539 | * come from?", and "What do you do when someone asks for optimization?". | ||
540 | */ | ||