blob: 1a8de57289eb97e5877b9016855cfe9e99adaac1 [file] [log] [blame]
Rusty Russell19f15372007-10-22 11:24:21 +10001/*P:050 Lguest guests use a very simple method to describe devices. It's a
Rusty Russella6bd8e12008-03-28 11:05:53 -05002 * series of device descriptors contained just above the top of normal Guest
Rusty Russell19f15372007-10-22 11:24:21 +10003 * 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
Rusty Russella6bd8e12008-03-28 11:05:53 -05007 * information and a "virtqueue" or two to send and receive data. :*/
Rusty Russell19f15372007-10-22 11:24:21 +10008#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. */
21static void *lguest_devices;
22
Rusty Russell19f15372007-10-22 11:24:21 +100023/* For Guests, device memory can be used as normal memory, so we cast away the
24 * __iomem to quieten sparse. */
25static inline void *lguest_map(unsigned long phys_addr, unsigned long pages)
26{
Rusty Russelle27810f2008-05-30 15:09:40 -050027 return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages);
Rusty Russell19f15372007-10-22 11:24:21 +100028}
29
30static inline void lguest_unmap(void *addr)
31{
32 iounmap((__force void __iomem *)addr);
33}
34
35/*D:100 Each lguest device is just a virtio device plus a pointer to its entry
36 * in the lguest_devices page. */
37struct lguest_device {
38 struct virtio_device vdev;
39
40 /* The entry in the lguest_devices page for this device. */
41 struct lguest_device_desc *desc;
42};
43
44/* Since the virtio infrastructure hands us a pointer to the virtio_device all
45 * the time, it helps to have a curt macro to get a pointer to the struct
46 * lguest_device it's enclosed in. */
Alexey Dobriyan25478442008-02-08 04:20:14 -080047#define to_lgdev(vd) container_of(vd, struct lguest_device, vdev)
Rusty Russell19f15372007-10-22 11:24:21 +100048
49/*D:130
50 * Device configurations
51 *
Rusty Russella586d4f2008-02-04 23:49:56 -050052 * The configuration information for a device consists of one or more
Rusty Russella6bd8e12008-03-28 11:05:53 -050053 * virtqueues, a feature bitmap, and some configuration bytes. The
Rusty Russell6e5aa7e2008-02-04 23:50:03 -050054 * configuration bytes don't really matter to us: the Launcher sets them up, and
Rusty Russella586d4f2008-02-04 23:49:56 -050055 * the driver will look at them during setup.
Rusty Russell19f15372007-10-22 11:24:21 +100056 *
Rusty Russella586d4f2008-02-04 23:49:56 -050057 * A convenient routine to return the device's virtqueue config array:
58 * immediately after the descriptor. */
59static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc)
60{
61 return (void *)(desc + 1);
62}
Rusty Russell19f15372007-10-22 11:24:21 +100063
Rusty Russella586d4f2008-02-04 23:49:56 -050064/* The features come immediately after the virtqueues. */
65static u8 *lg_features(const struct lguest_device_desc *desc)
66{
67 return (void *)(lg_vq(desc) + desc->num_vq);
68}
Rusty Russell19f15372007-10-22 11:24:21 +100069
Rusty Russella586d4f2008-02-04 23:49:56 -050070/* The config space comes after the two feature bitmasks. */
71static u8 *lg_config(const struct lguest_device_desc *desc)
72{
73 return lg_features(desc) + desc->feature_len * 2;
74}
75
76/* The total size of the config page used by this device (incl. desc) */
77static unsigned desc_size(const struct lguest_device_desc *desc)
78{
79 return sizeof(*desc)
80 + desc->num_vq * sizeof(struct lguest_vqconfig)
81 + desc->feature_len * 2
82 + desc->config_len;
83}
84
Rusty Russellc45a6812008-05-02 21:50:50 -050085/* This gets the device's feature bits. */
86static u32 lg_get_features(struct virtio_device *vdev)
Rusty Russell19f15372007-10-22 11:24:21 +100087{
Rusty Russellc45a6812008-05-02 21:50:50 -050088 unsigned int i;
89 u32 features = 0;
Rusty Russell19f15372007-10-22 11:24:21 +100090 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
Rusty Russellc45a6812008-05-02 21:50:50 -050091 u8 *in_features = lg_features(desc);
Rusty Russell19f15372007-10-22 11:24:21 +100092
Rusty Russellc45a6812008-05-02 21:50:50 -050093 /* We do this the slow but generic way. */
94 for (i = 0; i < min(desc->feature_len * 8, 32); i++)
95 if (in_features[i / 8] & (1 << (i % 8)))
96 features |= (1 << i);
Rusty Russell19f15372007-10-22 11:24:21 +100097
Rusty Russellc45a6812008-05-02 21:50:50 -050098 return features;
99}
Rusty Russella586d4f2008-02-04 23:49:56 -0500100
Rusty Russellc45a6812008-05-02 21:50:50 -0500101static void lg_set_features(struct virtio_device *vdev, u32 features)
102{
103 unsigned int i;
104 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
105 /* Second half of bitmap is features we accept. */
106 u8 *out_features = lg_features(desc) + desc->feature_len;
107
108 memset(out_features, 0, desc->feature_len);
109 for (i = 0; i < min(desc->feature_len * 8, 32); i++) {
110 if (features & (1 << i))
111 out_features[i / 8] |= (1 << (i % 8));
112 }
Rusty Russell19f15372007-10-22 11:24:21 +1000113}
114
115/* Once they've found a field, getting a copy of it is easy. */
Rusty Russella586d4f2008-02-04 23:49:56 -0500116static void lg_get(struct virtio_device *vdev, unsigned int offset,
Rusty Russell19f15372007-10-22 11:24:21 +1000117 void *buf, unsigned len)
118{
Rusty Russella586d4f2008-02-04 23:49:56 -0500119 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
120
121 /* Check they didn't ask for more than the length of the config! */
122 BUG_ON(offset + len > desc->config_len);
123 memcpy(buf, lg_config(desc) + offset, len);
Rusty Russell19f15372007-10-22 11:24:21 +1000124}
125
126/* Setting the contents is also trivial. */
Rusty Russella586d4f2008-02-04 23:49:56 -0500127static void lg_set(struct virtio_device *vdev, unsigned int offset,
Rusty Russell19f15372007-10-22 11:24:21 +1000128 const void *buf, unsigned len)
129{
Rusty Russella586d4f2008-02-04 23:49:56 -0500130 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
131
132 /* Check they didn't ask for more than the length of the config! */
133 BUG_ON(offset + len > desc->config_len);
134 memcpy(lg_config(desc) + offset, buf, len);
Rusty Russell19f15372007-10-22 11:24:21 +1000135}
136
137/* The operations to get and set the status word just access the status field
138 * of the device descriptor. */
139static u8 lg_get_status(struct virtio_device *vdev)
140{
141 return to_lgdev(vdev)->desc->status;
142}
143
Rusty Russella007a752008-05-02 21:50:53 -0500144/* To notify on status updates, we (ab)use the NOTIFY hypercall, with the
145 * descriptor address of the device. A zero status means "reset". */
146static void set_status(struct virtio_device *vdev, u8 status)
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500147{
148 unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices;
149
Rusty Russella007a752008-05-02 21:50:53 -0500150 /* We set the status. */
151 to_lgdev(vdev)->desc->status = status;
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500152 hcall(LHCALL_NOTIFY, (max_pfn<<PAGE_SHIFT) + offset, 0, 0);
153}
154
Rusty Russella007a752008-05-02 21:50:53 -0500155static void lg_set_status(struct virtio_device *vdev, u8 status)
156{
157 BUG_ON(!status);
158 set_status(vdev, status);
159}
160
161static void lg_reset(struct virtio_device *vdev)
162{
163 set_status(vdev, 0);
164}
165
Rusty Russell19f15372007-10-22 11:24:21 +1000166/*
167 * Virtqueues
168 *
169 * The other piece of infrastructure virtio needs is a "virtqueue": a way of
170 * the Guest device registering buffers for the other side to read from or
171 * write into (ie. send and receive buffers). Each device can have multiple
Rusty Russelle1e72962007-10-25 15:02:50 +1000172 * virtqueues: for example the console driver uses one queue for sending and
173 * another for receiving.
Rusty Russell19f15372007-10-22 11:24:21 +1000174 *
175 * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue
176 * already exists in virtio_ring.c. We just need to connect it up.
177 *
178 * We start with the information we need to keep about each virtqueue.
179 */
180
181/*D:140 This is the information we remember about each virtqueue. */
182struct lguest_vq_info
183{
184 /* A copy of the information contained in the device config. */
185 struct lguest_vqconfig config;
186
187 /* The address where we mapped the virtio ring, so we can unmap it. */
188 void *pages;
189};
190
191/* When the virtio_ring code wants to prod the Host, it calls us here and we
Rusty Russella6bd8e12008-03-28 11:05:53 -0500192 * make a hypercall. We hand the physical address of the virtqueue so the Host
Rusty Russell19f15372007-10-22 11:24:21 +1000193 * knows which virtqueue we're talking about. */
194static void lg_notify(struct virtqueue *vq)
195{
196 /* We store our virtqueue information in the "priv" pointer of the
197 * virtqueue structure. */
198 struct lguest_vq_info *lvq = vq->priv;
199
200 hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0);
201}
202
203/* This routine finds the first virtqueue described in the configuration of
204 * this device and sets it up.
205 *
206 * This is kind of an ugly duckling. It'd be nicer to have a standard
207 * representation of a virtqueue in the configuration space, but it seems that
Rusty Russelle1e72962007-10-25 15:02:50 +1000208 * everyone wants to do it differently. The KVM coders want the Guest to
Rusty Russell19f15372007-10-22 11:24:21 +1000209 * allocate its own pages and tell the Host where they are, but for lguest it's
210 * simpler for the Host to simply tell us where the pages are.
211 *
Rusty Russella6bd8e12008-03-28 11:05:53 -0500212 * So we provide drivers with a "find the Nth virtqueue and set it up"
213 * function. */
Rusty Russell19f15372007-10-22 11:24:21 +1000214static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
Rusty Russella586d4f2008-02-04 23:49:56 -0500215 unsigned index,
Rusty Russell18445c42008-02-04 23:49:57 -0500216 void (*callback)(struct virtqueue *vq))
Rusty Russell19f15372007-10-22 11:24:21 +1000217{
Rusty Russella586d4f2008-02-04 23:49:56 -0500218 struct lguest_device *ldev = to_lgdev(vdev);
Rusty Russell19f15372007-10-22 11:24:21 +1000219 struct lguest_vq_info *lvq;
220 struct virtqueue *vq;
Rusty Russell19f15372007-10-22 11:24:21 +1000221 int err;
222
Rusty Russella586d4f2008-02-04 23:49:56 -0500223 /* We must have this many virtqueues. */
224 if (index >= ldev->desc->num_vq)
Rusty Russell19f15372007-10-22 11:24:21 +1000225 return ERR_PTR(-ENOENT);
226
227 lvq = kmalloc(sizeof(*lvq), GFP_KERNEL);
228 if (!lvq)
229 return ERR_PTR(-ENOMEM);
230
Rusty Russella586d4f2008-02-04 23:49:56 -0500231 /* Make a copy of the "struct lguest_vqconfig" entry, which sits after
232 * the descriptor. We need a copy because the config space might not
233 * be aligned correctly. */
234 memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config));
Rusty Russell19f15372007-10-22 11:24:21 +1000235
Rusty Russella586d4f2008-02-04 23:49:56 -0500236 printk("Mapping virtqueue %i addr %lx\n", index,
237 (unsigned long)lvq->config.pfn << PAGE_SHIFT);
Rusty Russell19f15372007-10-22 11:24:21 +1000238 /* Figure out how many pages the ring will take, and map that memory */
239 lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT,
Rusty Russell42b36cc2007-11-12 13:39:18 +1100240 DIV_ROUND_UP(vring_size(lvq->config.num,
241 PAGE_SIZE),
Rusty Russell19f15372007-10-22 11:24:21 +1000242 PAGE_SIZE));
243 if (!lvq->pages) {
244 err = -ENOMEM;
245 goto free_lvq;
246 }
247
248 /* OK, tell virtio_ring.c to set up a virtqueue now we know its size
249 * and we've got a pointer to its pages. */
250 vq = vring_new_virtqueue(lvq->config.num, vdev, lvq->pages,
251 lg_notify, callback);
252 if (!vq) {
253 err = -ENOMEM;
254 goto unmap;
255 }
256
257 /* Tell the interrupt for this virtqueue to go to the virtio_ring
258 * interrupt handler. */
259 /* FIXME: We used to have a flag for the Host to tell us we could use
260 * the interrupt as a source of randomness: it'd be nice to have that
261 * back.. */
262 err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
263 vdev->dev.bus_id, vq);
264 if (err)
265 goto destroy_vring;
266
267 /* Last of all we hook up our 'struct lguest_vq_info" to the
268 * virtqueue's priv pointer. */
269 vq->priv = lvq;
270 return vq;
271
272destroy_vring:
273 vring_del_virtqueue(vq);
274unmap:
275 lguest_unmap(lvq->pages);
276free_lvq:
277 kfree(lvq);
278 return ERR_PTR(err);
279}
280/*:*/
281
282/* Cleaning up a virtqueue is easy */
283static void lg_del_vq(struct virtqueue *vq)
284{
285 struct lguest_vq_info *lvq = vq->priv;
286
Rusty Russell74b25532007-11-19 11:20:42 -0500287 /* Release the interrupt */
288 free_irq(lvq->config.irq, vq);
Rusty Russell19f15372007-10-22 11:24:21 +1000289 /* Tell virtio_ring.c to free the virtqueue. */
290 vring_del_virtqueue(vq);
291 /* Unmap the pages containing the ring. */
292 lguest_unmap(lvq->pages);
293 /* Free our own queue information. */
294 kfree(lvq);
295}
296
297/* The ops structure which hooks everything together. */
298static struct virtio_config_ops lguest_config_ops = {
Rusty Russellc45a6812008-05-02 21:50:50 -0500299 .get_features = lg_get_features,
300 .set_features = lg_set_features,
Rusty Russell19f15372007-10-22 11:24:21 +1000301 .get = lg_get,
302 .set = lg_set,
303 .get_status = lg_get_status,
304 .set_status = lg_set_status,
Rusty Russell6e5aa7e2008-02-04 23:50:03 -0500305 .reset = lg_reset,
Rusty Russell19f15372007-10-22 11:24:21 +1000306 .find_vq = lg_find_vq,
307 .del_vq = lg_del_vq,
308};
309
310/* The root device for the lguest virtio devices. This makes them appear as
311 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */
312static struct device lguest_root = {
313 .parent = NULL,
314 .bus_id = "lguest",
315};
316
317/*D:120 This is the core of the lguest bus: actually adding a new device.
318 * It's a separate function because it's neater that way, and because an
319 * earlier version of the code supported hotplug and unplug. They were removed
320 * early on because they were never used.
321 *
322 * As Andrew Tridgell says, "Untested code is buggy code".
323 *
324 * It's worth reading this carefully: we start with a pointer to the new device
Rusty Russellb769f572008-05-30 15:09:42 -0500325 * descriptor in the "lguest_devices" page, and the offset into the device
326 * descriptor page so we can uniquely identify it if things go badly wrong. */
327static void add_lguest_device(struct lguest_device_desc *d,
328 unsigned int offset)
Rusty Russell19f15372007-10-22 11:24:21 +1000329{
330 struct lguest_device *ldev;
331
Rusty Russelle1e72962007-10-25 15:02:50 +1000332 /* Start with zeroed memory; Linux's device layer seems to count on
333 * it. */
Rusty Russell19f15372007-10-22 11:24:21 +1000334 ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
335 if (!ldev) {
Rusty Russellb769f572008-05-30 15:09:42 -0500336 printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n",
337 offset, d->type);
Rusty Russell19f15372007-10-22 11:24:21 +1000338 return;
339 }
340
341 /* This devices' parent is the lguest/ dir. */
342 ldev->vdev.dev.parent = &lguest_root;
343 /* We have a unique device index thanks to the dev_index counter. */
Rusty Russell19f15372007-10-22 11:24:21 +1000344 ldev->vdev.id.device = d->type;
345 /* We have a simple set of routines for querying the device's
346 * configuration information and setting its status. */
347 ldev->vdev.config = &lguest_config_ops;
348 /* And we remember the device's descriptor for lguest_config_ops. */
349 ldev->desc = d;
350
351 /* register_virtio_device() sets up the generic fields for the struct
352 * virtio_device and calls device_register(). This makes the bus
353 * infrastructure look for a matching driver. */
354 if (register_virtio_device(&ldev->vdev) != 0) {
Rusty Russellb769f572008-05-30 15:09:42 -0500355 printk(KERN_ERR "Failed to register lguest dev %u type %u\n",
356 offset, d->type);
Rusty Russell19f15372007-10-22 11:24:21 +1000357 kfree(ldev);
358 }
359}
360
361/*D:110 scan_devices() simply iterates through the device page. The type 0 is
362 * reserved to mean "end of devices". */
363static void scan_devices(void)
364{
365 unsigned int i;
366 struct lguest_device_desc *d;
367
368 /* We start at the page beginning, and skip over each entry. */
Rusty Russella586d4f2008-02-04 23:49:56 -0500369 for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
Rusty Russell19f15372007-10-22 11:24:21 +1000370 d = lguest_devices + i;
371
372 /* Once we hit a zero, stop. */
373 if (d->type == 0)
374 break;
375
Rusty Russella586d4f2008-02-04 23:49:56 -0500376 printk("Device at %i has size %u\n", i, desc_size(d));
Rusty Russellb769f572008-05-30 15:09:42 -0500377 add_lguest_device(d, i);
Rusty Russell19f15372007-10-22 11:24:21 +1000378 }
379}
380
381/*D:105 Fairly early in boot, lguest_devices_init() is called to set up the
382 * lguest device infrastructure. We check that we are a Guest by checking
383 * pv_info.name: there are other ways of checking, but this seems most
384 * obvious to me.
385 *
386 * So we can access the "struct lguest_device_desc"s easily, we map that memory
387 * and store the pointer in the global "lguest_devices". Then we register a
388 * root device from which all our devices will hang (this seems to be the
389 * correct sysfs incantation).
390 *
391 * Finally we call scan_devices() which adds all the devices found in the
392 * lguest_devices page. */
393static int __init lguest_devices_init(void)
394{
395 if (strcmp(pv_info.name, "lguest") != 0)
396 return 0;
397
398 if (device_register(&lguest_root) != 0)
399 panic("Could not register lguest root");
400
401 /* Devices are in a single page above top of "normal" mem */
402 lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
403
404 scan_devices();
405 return 0;
406}
407/* We do this after core stuff, but before the drivers. */
408postcore_initcall(lguest_devices_init);
409
410/*D:150 At this point in the journey we used to now wade through the lguest
411 * devices themselves: net, block and console. Since they're all now virtio
412 * devices rather than lguest-specific, I've decided to ignore them. Mostly,
413 * they're kind of boring. But this does mean you'll never experience the
414 * thrill of reading the forbidden love scene buried deep in the block driver.
415 *
416 * "make Launcher" beckons, where we answer questions like "Where do Guests
417 * come from?", and "What do you do when someone asks for optimization?". */