Commit | Line | Data |
---|---|---|
19f1537b | 1 | /*P:050 Lguest guests use a very simple method to describe devices. It's a |
a6bd8e13 | 2 | * series of device descriptors contained just above the top of normal Guest |
19f1537b RR |
3 | * memory. |
4 | * | |
5 | * We use the standard "virtio" device infrastructure, which provides us with a | |
6 | * console, a network and a block driver. Each one expects some configuration | |
a6bd8e13 | 7 | * information and a "virtqueue" or two to send and receive data. :*/ |
19f1537b RR |
8 | #include <linux/init.h> |
9 | #include <linux/bootmem.h> | |
10 | #include <linux/lguest_launcher.h> | |
11 | #include <linux/virtio.h> | |
12 | #include <linux/virtio_config.h> | |
13 | #include <linux/interrupt.h> | |
14 | #include <linux/virtio_ring.h> | |
15 | #include <linux/err.h> | |
16 | #include <asm/io.h> | |
17 | #include <asm/paravirt.h> | |
18 | #include <asm/lguest_hcall.h> | |
19 | ||
20 | /* The pointer to our (page) of device descriptions. */ | |
21 | static void *lguest_devices; | |
22 | ||
23 | /* Unique numbering for lguest devices. */ | |
24 | static unsigned int dev_index; | |
25 | ||
26 | /* For Guests, device memory can be used as normal memory, so we cast away the | |
27 | * __iomem to quieten sparse. */ | |
28 | static inline void *lguest_map(unsigned long phys_addr, unsigned long pages) | |
29 | { | |
30 | return (__force void *)ioremap(phys_addr, PAGE_SIZE*pages); | |
31 | } | |
32 | ||
33 | static inline void lguest_unmap(void *addr) | |
34 | { | |
35 | iounmap((__force void __iomem *)addr); | |
36 | } | |
37 | ||
38 | /*D:100 Each lguest device is just a virtio device plus a pointer to its entry | |
39 | * in the lguest_devices page. */ | |
40 | struct lguest_device { | |
41 | struct virtio_device vdev; | |
42 | ||
43 | /* The entry in the lguest_devices page for this device. */ | |
44 | struct lguest_device_desc *desc; | |
45 | }; | |
46 | ||
47 | /* Since the virtio infrastructure hands us a pointer to the virtio_device all | |
48 | * the time, it helps to have a curt macro to get a pointer to the struct | |
49 | * lguest_device it's enclosed in. */ | |
25478445 | 50 | #define to_lgdev(vd) container_of(vd, struct lguest_device, vdev) |
19f1537b RR |
51 | |
52 | /*D:130 | |
53 | * Device configurations | |
54 | * | |
a586d4f6 | 55 | * The configuration information for a device consists of one or more |
a6bd8e13 | 56 | * virtqueues, a feature bitmap, and some configuration bytes. The |
6e5aa7ef | 57 | * configuration bytes don't really matter to us: the Launcher sets them up, and |
a586d4f6 | 58 | * the driver will look at them during setup. |
19f1537b | 59 | * |
a586d4f6 RR |
60 | * A convenient routine to return the device's virtqueue config array: |
61 | * immediately after the descriptor. */ | |
62 | static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc) | |
63 | { | |
64 | return (void *)(desc + 1); | |
65 | } | |
19f1537b | 66 | |
a586d4f6 RR |
67 | /* The features come immediately after the virtqueues. */ |
68 | static u8 *lg_features(const struct lguest_device_desc *desc) | |
69 | { | |
70 | return (void *)(lg_vq(desc) + desc->num_vq); | |
71 | } | |
19f1537b | 72 | |
a586d4f6 RR |
73 | /* The config space comes after the two feature bitmasks. */ |
74 | static u8 *lg_config(const struct lguest_device_desc *desc) | |
19f1537b | 75 | { |
a586d4f6 RR |
76 | return lg_features(desc) + desc->feature_len * 2; |
77 | } | |
19f1537b | 78 | |
a586d4f6 RR |
79 | /* The total size of the config page used by this device (incl. desc) */ |
80 | static unsigned desc_size(const struct lguest_device_desc *desc) | |
81 | { | |
82 | return sizeof(*desc) | |
83 | + desc->num_vq * sizeof(struct lguest_vqconfig) | |
84 | + desc->feature_len * 2 | |
85 | + desc->config_len; | |
86 | } | |
87 | ||
88 | /* This tests (and acknowleges) a feature bit. */ | |
89 | static bool lg_feature(struct virtio_device *vdev, unsigned fbit) | |
90 | { | |
91 | struct lguest_device_desc *desc = to_lgdev(vdev)->desc; | |
92 | u8 *features; | |
93 | ||
94 | /* Obviously if they ask for a feature off the end of our feature | |
95 | * bitmap, it's not set. */ | |
96 | if (fbit / 8 > desc->feature_len) | |
97 | return false; | |
98 | ||
99 | /* The feature bitmap comes after the virtqueues. */ | |
100 | features = lg_features(desc); | |
101 | if (!(features[fbit / 8] & (1 << (fbit % 8)))) | |
102 | return false; | |
103 | ||
104 | /* We set the matching bit in the other half of the bitmap to tell the | |
105 | * Host we want to use this feature. We don't use this yet, but we | |
106 | * could in future. */ | |
107 | features[desc->feature_len + fbit / 8] |= (1 << (fbit % 8)); | |
108 | return true; | |
19f1537b RR |
109 | } |
110 | ||
111 | /* Once they've found a field, getting a copy of it is easy. */ | |
a586d4f6 | 112 | static void lg_get(struct virtio_device *vdev, unsigned int offset, |
19f1537b RR |
113 | void *buf, unsigned len) |
114 | { | |
a586d4f6 RR |
115 | struct lguest_device_desc *desc = to_lgdev(vdev)->desc; |
116 | ||
117 | /* Check they didn't ask for more than the length of the config! */ | |
118 | BUG_ON(offset + len > desc->config_len); | |
119 | memcpy(buf, lg_config(desc) + offset, len); | |
19f1537b RR |
120 | } |
121 | ||
122 | /* Setting the contents is also trivial. */ | |
a586d4f6 | 123 | static void lg_set(struct virtio_device *vdev, unsigned int offset, |
19f1537b RR |
124 | const void *buf, unsigned len) |
125 | { | |
a586d4f6 RR |
126 | struct lguest_device_desc *desc = to_lgdev(vdev)->desc; |
127 | ||
128 | /* Check they didn't ask for more than the length of the config! */ | |
129 | BUG_ON(offset + len > desc->config_len); | |
130 | memcpy(lg_config(desc) + offset, buf, len); | |
19f1537b RR |
131 | } |
132 | ||
133 | /* The operations to get and set the status word just access the status field | |
134 | * of the device descriptor. */ | |
135 | static u8 lg_get_status(struct virtio_device *vdev) | |
136 | { | |
137 | return to_lgdev(vdev)->desc->status; | |
138 | } | |
139 | ||
140 | static void lg_set_status(struct virtio_device *vdev, u8 status) | |
141 | { | |
6e5aa7ef | 142 | BUG_ON(!status); |
19f1537b RR |
143 | to_lgdev(vdev)->desc->status = status; |
144 | } | |
145 | ||
6e5aa7ef RR |
146 | /* To reset the device, we (ab)use the NOTIFY hypercall, with the descriptor |
147 | * address of the device. The Host will zero the status and all the | |
148 | * features. */ | |
149 | static void lg_reset(struct virtio_device *vdev) | |
150 | { | |
151 | unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices; | |
152 | ||
153 | hcall(LHCALL_NOTIFY, (max_pfn<<PAGE_SHIFT) + offset, 0, 0); | |
154 | } | |
155 | ||
19f1537b RR |
156 | /* |
157 | * Virtqueues | |
158 | * | |
159 | * The other piece of infrastructure virtio needs is a "virtqueue": a way of | |
160 | * the Guest device registering buffers for the other side to read from or | |
161 | * write into (ie. send and receive buffers). Each device can have multiple | |
e1e72965 RR |
162 | * virtqueues: for example the console driver uses one queue for sending and |
163 | * another for receiving. | |
19f1537b RR |
164 | * |
165 | * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue | |
166 | * already exists in virtio_ring.c. We just need to connect it up. | |
167 | * | |
168 | * We start with the information we need to keep about each virtqueue. | |
169 | */ | |
170 | ||
171 | /*D:140 This is the information we remember about each virtqueue. */ | |
172 | struct lguest_vq_info | |
173 | { | |
174 | /* A copy of the information contained in the device config. */ | |
175 | struct lguest_vqconfig config; | |
176 | ||
177 | /* The address where we mapped the virtio ring, so we can unmap it. */ | |
178 | void *pages; | |
179 | }; | |
180 | ||
181 | /* When the virtio_ring code wants to prod the Host, it calls us here and we | |
a6bd8e13 | 182 | * make a hypercall. We hand the physical address of the virtqueue so the Host |
19f1537b RR |
183 | * knows which virtqueue we're talking about. */ |
184 | static void lg_notify(struct virtqueue *vq) | |
185 | { | |
186 | /* We store our virtqueue information in the "priv" pointer of the | |
187 | * virtqueue structure. */ | |
188 | struct lguest_vq_info *lvq = vq->priv; | |
189 | ||
190 | hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0); | |
191 | } | |
192 | ||
193 | /* This routine finds the first virtqueue described in the configuration of | |
194 | * this device and sets it up. | |
195 | * | |
196 | * This is kind of an ugly duckling. It'd be nicer to have a standard | |
197 | * representation of a virtqueue in the configuration space, but it seems that | |
e1e72965 | 198 | * everyone wants to do it differently. The KVM coders want the Guest to |
19f1537b RR |
199 | * allocate its own pages and tell the Host where they are, but for lguest it's |
200 | * simpler for the Host to simply tell us where the pages are. | |
201 | * | |
a6bd8e13 RR |
202 | * So we provide drivers with a "find the Nth virtqueue and set it up" |
203 | * function. */ | |
19f1537b | 204 | static struct virtqueue *lg_find_vq(struct virtio_device *vdev, |
a586d4f6 | 205 | unsigned index, |
18445c4d | 206 | void (*callback)(struct virtqueue *vq)) |
19f1537b | 207 | { |
a586d4f6 | 208 | struct lguest_device *ldev = to_lgdev(vdev); |
19f1537b RR |
209 | struct lguest_vq_info *lvq; |
210 | struct virtqueue *vq; | |
19f1537b RR |
211 | int err; |
212 | ||
a586d4f6 RR |
213 | /* We must have this many virtqueues. */ |
214 | if (index >= ldev->desc->num_vq) | |
19f1537b RR |
215 | return ERR_PTR(-ENOENT); |
216 | ||
217 | lvq = kmalloc(sizeof(*lvq), GFP_KERNEL); | |
218 | if (!lvq) | |
219 | return ERR_PTR(-ENOMEM); | |
220 | ||
a586d4f6 RR |
221 | /* Make a copy of the "struct lguest_vqconfig" entry, which sits after |
222 | * the descriptor. We need a copy because the config space might not | |
223 | * be aligned correctly. */ | |
224 | memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config)); | |
19f1537b | 225 | |
a586d4f6 RR |
226 | printk("Mapping virtqueue %i addr %lx\n", index, |
227 | (unsigned long)lvq->config.pfn << PAGE_SHIFT); | |
19f1537b RR |
228 | /* Figure out how many pages the ring will take, and map that memory */ |
229 | lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT, | |
42b36cc0 RR |
230 | DIV_ROUND_UP(vring_size(lvq->config.num, |
231 | PAGE_SIZE), | |
19f1537b RR |
232 | PAGE_SIZE)); |
233 | if (!lvq->pages) { | |
234 | err = -ENOMEM; | |
235 | goto free_lvq; | |
236 | } | |
237 | ||
238 | /* OK, tell virtio_ring.c to set up a virtqueue now we know its size | |
239 | * and we've got a pointer to its pages. */ | |
240 | vq = vring_new_virtqueue(lvq->config.num, vdev, lvq->pages, | |
241 | lg_notify, callback); | |
242 | if (!vq) { | |
243 | err = -ENOMEM; | |
244 | goto unmap; | |
245 | } | |
246 | ||
247 | /* Tell the interrupt for this virtqueue to go to the virtio_ring | |
248 | * interrupt handler. */ | |
249 | /* FIXME: We used to have a flag for the Host to tell us we could use | |
250 | * the interrupt as a source of randomness: it'd be nice to have that | |
251 | * back.. */ | |
252 | err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED, | |
253 | vdev->dev.bus_id, vq); | |
254 | if (err) | |
255 | goto destroy_vring; | |
256 | ||
257 | /* Last of all we hook up our 'struct lguest_vq_info" to the | |
258 | * virtqueue's priv pointer. */ | |
259 | vq->priv = lvq; | |
260 | return vq; | |
261 | ||
262 | destroy_vring: | |
263 | vring_del_virtqueue(vq); | |
264 | unmap: | |
265 | lguest_unmap(lvq->pages); | |
266 | free_lvq: | |
267 | kfree(lvq); | |
268 | return ERR_PTR(err); | |
269 | } | |
270 | /*:*/ | |
271 | ||
272 | /* Cleaning up a virtqueue is easy */ | |
273 | static void lg_del_vq(struct virtqueue *vq) | |
274 | { | |
275 | struct lguest_vq_info *lvq = vq->priv; | |
276 | ||
74b2553f RR |
277 | /* Release the interrupt */ |
278 | free_irq(lvq->config.irq, vq); | |
19f1537b RR |
279 | /* Tell virtio_ring.c to free the virtqueue. */ |
280 | vring_del_virtqueue(vq); | |
281 | /* Unmap the pages containing the ring. */ | |
282 | lguest_unmap(lvq->pages); | |
283 | /* Free our own queue information. */ | |
284 | kfree(lvq); | |
285 | } | |
286 | ||
287 | /* The ops structure which hooks everything together. */ | |
288 | static struct virtio_config_ops lguest_config_ops = { | |
a586d4f6 | 289 | .feature = lg_feature, |
19f1537b RR |
290 | .get = lg_get, |
291 | .set = lg_set, | |
292 | .get_status = lg_get_status, | |
293 | .set_status = lg_set_status, | |
6e5aa7ef | 294 | .reset = lg_reset, |
19f1537b RR |
295 | .find_vq = lg_find_vq, |
296 | .del_vq = lg_del_vq, | |
297 | }; | |
298 | ||
299 | /* The root device for the lguest virtio devices. This makes them appear as | |
300 | * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */ | |
301 | static struct device lguest_root = { | |
302 | .parent = NULL, | |
303 | .bus_id = "lguest", | |
304 | }; | |
305 | ||
306 | /*D:120 This is the core of the lguest bus: actually adding a new device. | |
307 | * It's a separate function because it's neater that way, and because an | |
308 | * earlier version of the code supported hotplug and unplug. They were removed | |
309 | * early on because they were never used. | |
310 | * | |
311 | * As Andrew Tridgell says, "Untested code is buggy code". | |
312 | * | |
313 | * It's worth reading this carefully: we start with a pointer to the new device | |
314 | * descriptor in the "lguest_devices" page. */ | |
315 | static void add_lguest_device(struct lguest_device_desc *d) | |
316 | { | |
317 | struct lguest_device *ldev; | |
318 | ||
e1e72965 RR |
319 | /* Start with zeroed memory; Linux's device layer seems to count on |
320 | * it. */ | |
19f1537b RR |
321 | ldev = kzalloc(sizeof(*ldev), GFP_KERNEL); |
322 | if (!ldev) { | |
323 | printk(KERN_EMERG "Cannot allocate lguest dev %u\n", | |
324 | dev_index++); | |
325 | return; | |
326 | } | |
327 | ||
328 | /* This devices' parent is the lguest/ dir. */ | |
329 | ldev->vdev.dev.parent = &lguest_root; | |
330 | /* We have a unique device index thanks to the dev_index counter. */ | |
331 | ldev->vdev.index = dev_index++; | |
332 | /* The device type comes straight from the descriptor. There's also a | |
333 | * device vendor field in the virtio_device struct, which we leave as | |
334 | * 0. */ | |
335 | ldev->vdev.id.device = d->type; | |
336 | /* We have a simple set of routines for querying the device's | |
337 | * configuration information and setting its status. */ | |
338 | ldev->vdev.config = &lguest_config_ops; | |
339 | /* And we remember the device's descriptor for lguest_config_ops. */ | |
340 | ldev->desc = d; | |
341 | ||
342 | /* register_virtio_device() sets up the generic fields for the struct | |
343 | * virtio_device and calls device_register(). This makes the bus | |
344 | * infrastructure look for a matching driver. */ | |
345 | if (register_virtio_device(&ldev->vdev) != 0) { | |
346 | printk(KERN_ERR "Failed to register lguest device %u\n", | |
347 | ldev->vdev.index); | |
348 | kfree(ldev); | |
349 | } | |
350 | } | |
351 | ||
352 | /*D:110 scan_devices() simply iterates through the device page. The type 0 is | |
353 | * reserved to mean "end of devices". */ | |
354 | static void scan_devices(void) | |
355 | { | |
356 | unsigned int i; | |
357 | struct lguest_device_desc *d; | |
358 | ||
359 | /* We start at the page beginning, and skip over each entry. */ | |
a586d4f6 | 360 | for (i = 0; i < PAGE_SIZE; i += desc_size(d)) { |
19f1537b RR |
361 | d = lguest_devices + i; |
362 | ||
363 | /* Once we hit a zero, stop. */ | |
364 | if (d->type == 0) | |
365 | break; | |
366 | ||
a586d4f6 | 367 | printk("Device at %i has size %u\n", i, desc_size(d)); |
19f1537b RR |
368 | add_lguest_device(d); |
369 | } | |
370 | } | |
371 | ||
372 | /*D:105 Fairly early in boot, lguest_devices_init() is called to set up the | |
373 | * lguest device infrastructure. We check that we are a Guest by checking | |
374 | * pv_info.name: there are other ways of checking, but this seems most | |
375 | * obvious to me. | |
376 | * | |
377 | * So we can access the "struct lguest_device_desc"s easily, we map that memory | |
378 | * and store the pointer in the global "lguest_devices". Then we register a | |
379 | * root device from which all our devices will hang (this seems to be the | |
380 | * correct sysfs incantation). | |
381 | * | |
382 | * Finally we call scan_devices() which adds all the devices found in the | |
383 | * lguest_devices page. */ | |
384 | static int __init lguest_devices_init(void) | |
385 | { | |
386 | if (strcmp(pv_info.name, "lguest") != 0) | |
387 | return 0; | |
388 | ||
389 | if (device_register(&lguest_root) != 0) | |
390 | panic("Could not register lguest root"); | |
391 | ||
392 | /* Devices are in a single page above top of "normal" mem */ | |
393 | lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1); | |
394 | ||
395 | scan_devices(); | |
396 | return 0; | |
397 | } | |
398 | /* We do this after core stuff, but before the drivers. */ | |
399 | postcore_initcall(lguest_devices_init); | |
400 | ||
401 | /*D:150 At this point in the journey we used to now wade through the lguest | |
402 | * devices themselves: net, block and console. Since they're all now virtio | |
403 | * devices rather than lguest-specific, I've decided to ignore them. Mostly, | |
404 | * they're kind of boring. But this does mean you'll never experience the | |
405 | * thrill of reading the forbidden love scene buried deep in the block driver. | |
406 | * | |
407 | * "make Launcher" beckons, where we answer questions like "Where do Guests | |
408 | * come from?", and "What do you do when someone asks for optimization?". */ |