blob: e2eec38c83c2ef21248353d132e3308810e1e17d [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
2 * series of device descriptors contained just above the top of normal
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
7 * information and a "virtqueue" mechanism to send and receive data. :*/
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. */
21static void *lguest_devices;
22
23/* Unique numbering for lguest devices. */
24static 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. */
28static inline void *lguest_map(unsigned long phys_addr, unsigned long pages)
29{
30 return (__force void *)ioremap(phys_addr, PAGE_SIZE*pages);
31}
32
33static 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. */
40struct 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. */
50#define to_lgdev(vdev) container_of(vdev, struct lguest_device, vdev)
51
52/*D:130
53 * Device configurations
54 *
55 * The configuration information for a device consists of a series of fields.
Rusty Russelle1e72962007-10-25 15:02:50 +100056 * We don't really care what they are: the Launcher set them up, and the driver
57 * will look at them during setup.
Rusty Russell19f15372007-10-22 11:24:21 +100058 *
59 * For us these fields come immediately after that device's descriptor in the
60 * lguest_devices page.
61 *
62 * Each field starts with a "type" byte, a "length" byte, then that number of
63 * bytes of configuration information. The device descriptor tells us the
64 * total configuration length so we know when we've reached the last field. */
65
66/* type + length bytes */
67#define FHDR_LEN 2
68
69/* This finds the first field of a given type for a device's configuration. */
70static void *lg_find(struct virtio_device *vdev, u8 type, unsigned int *len)
71{
72 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
73 int i;
74
75 for (i = 0; i < desc->config_len; i += FHDR_LEN + desc->config[i+1]) {
76 if (desc->config[i] == type) {
77 /* Mark it used, so Host can know we looked at it, and
78 * also so we won't find the same one twice. */
79 desc->config[i] |= 0x80;
80 /* Remember, the second byte is the length. */
81 *len = desc->config[i+1];
82 /* We return a pointer to the field header. */
83 return desc->config + i;
84 }
85 }
86
87 /* Not found: return NULL for failure. */
88 return NULL;
89}
90
91/* Once they've found a field, getting a copy of it is easy. */
92static void lg_get(struct virtio_device *vdev, void *token,
93 void *buf, unsigned len)
94{
95 /* Check they didn't ask for more than the length of the field! */
96 BUG_ON(len > ((u8 *)token)[1]);
97 memcpy(buf, token + FHDR_LEN, len);
98}
99
100/* Setting the contents is also trivial. */
101static void lg_set(struct virtio_device *vdev, void *token,
102 const void *buf, unsigned len)
103{
104 BUG_ON(len > ((u8 *)token)[1]);
105 memcpy(token + FHDR_LEN, buf, len);
106}
107
108/* The operations to get and set the status word just access the status field
109 * of the device descriptor. */
110static u8 lg_get_status(struct virtio_device *vdev)
111{
112 return to_lgdev(vdev)->desc->status;
113}
114
115static void lg_set_status(struct virtio_device *vdev, u8 status)
116{
117 to_lgdev(vdev)->desc->status = status;
118}
119
120/*
121 * Virtqueues
122 *
123 * The other piece of infrastructure virtio needs is a "virtqueue": a way of
124 * the Guest device registering buffers for the other side to read from or
125 * write into (ie. send and receive buffers). Each device can have multiple
Rusty Russelle1e72962007-10-25 15:02:50 +1000126 * virtqueues: for example the console driver uses one queue for sending and
127 * another for receiving.
Rusty Russell19f15372007-10-22 11:24:21 +1000128 *
129 * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue
130 * already exists in virtio_ring.c. We just need to connect it up.
131 *
132 * We start with the information we need to keep about each virtqueue.
133 */
134
135/*D:140 This is the information we remember about each virtqueue. */
136struct lguest_vq_info
137{
138 /* A copy of the information contained in the device config. */
139 struct lguest_vqconfig config;
140
141 /* The address where we mapped the virtio ring, so we can unmap it. */
142 void *pages;
143};
144
145/* When the virtio_ring code wants to prod the Host, it calls us here and we
146 * make a hypercall. We hand the page number of the virtqueue so the Host
147 * knows which virtqueue we're talking about. */
148static void lg_notify(struct virtqueue *vq)
149{
150 /* We store our virtqueue information in the "priv" pointer of the
151 * virtqueue structure. */
152 struct lguest_vq_info *lvq = vq->priv;
153
154 hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0);
155}
156
157/* This routine finds the first virtqueue described in the configuration of
158 * this device and sets it up.
159 *
160 * This is kind of an ugly duckling. It'd be nicer to have a standard
161 * representation of a virtqueue in the configuration space, but it seems that
Rusty Russelle1e72962007-10-25 15:02:50 +1000162 * everyone wants to do it differently. The KVM coders want the Guest to
Rusty Russell19f15372007-10-22 11:24:21 +1000163 * allocate its own pages and tell the Host where they are, but for lguest it's
164 * simpler for the Host to simply tell us where the pages are.
165 *
166 * So we provide devices with a "find virtqueue and set it up" function. */
167static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
168 bool (*callback)(struct virtqueue *vq))
169{
170 struct lguest_vq_info *lvq;
171 struct virtqueue *vq;
172 unsigned int len;
173 void *token;
174 int err;
175
176 /* Look for a field of the correct type to mark a virtqueue. Note that
177 * if this succeeds, then the type will be changed so it won't be found
178 * again, and future lg_find_vq() calls will find the next
179 * virtqueue (if any). */
180 token = vdev->config->find(vdev, VIRTIO_CONFIG_F_VIRTQUEUE, &len);
181 if (!token)
182 return ERR_PTR(-ENOENT);
183
184 lvq = kmalloc(sizeof(*lvq), GFP_KERNEL);
185 if (!lvq)
186 return ERR_PTR(-ENOMEM);
187
188 /* Note: we could use a configuration space inside here, just like we
189 * do for the device. This would allow expansion in future, because
190 * our configuration system is designed to be expansible. But this is
191 * way easier. */
192 if (len != sizeof(lvq->config)) {
193 dev_err(&vdev->dev, "Unexpected virtio config len %u\n", len);
194 err = -EIO;
195 goto free_lvq;
196 }
197 /* Make a copy of the "struct lguest_vqconfig" field. We need a copy
198 * because the config space might not be aligned correctly. */
199 vdev->config->get(vdev, token, &lvq->config, sizeof(lvq->config));
200
201 /* Figure out how many pages the ring will take, and map that memory */
202 lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT,
Rusty Russell42b36cc2007-11-12 13:39:18 +1100203 DIV_ROUND_UP(vring_size(lvq->config.num,
204 PAGE_SIZE),
Rusty Russell19f15372007-10-22 11:24:21 +1000205 PAGE_SIZE));
206 if (!lvq->pages) {
207 err = -ENOMEM;
208 goto free_lvq;
209 }
210
211 /* OK, tell virtio_ring.c to set up a virtqueue now we know its size
212 * and we've got a pointer to its pages. */
213 vq = vring_new_virtqueue(lvq->config.num, vdev, lvq->pages,
214 lg_notify, callback);
215 if (!vq) {
216 err = -ENOMEM;
217 goto unmap;
218 }
219
220 /* Tell the interrupt for this virtqueue to go to the virtio_ring
221 * interrupt handler. */
222 /* FIXME: We used to have a flag for the Host to tell us we could use
223 * the interrupt as a source of randomness: it'd be nice to have that
224 * back.. */
225 err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
226 vdev->dev.bus_id, vq);
227 if (err)
228 goto destroy_vring;
229
230 /* Last of all we hook up our 'struct lguest_vq_info" to the
231 * virtqueue's priv pointer. */
232 vq->priv = lvq;
233 return vq;
234
235destroy_vring:
236 vring_del_virtqueue(vq);
237unmap:
238 lguest_unmap(lvq->pages);
239free_lvq:
240 kfree(lvq);
241 return ERR_PTR(err);
242}
243/*:*/
244
245/* Cleaning up a virtqueue is easy */
246static void lg_del_vq(struct virtqueue *vq)
247{
248 struct lguest_vq_info *lvq = vq->priv;
249
Rusty Russell74b25532007-11-19 11:20:42 -0500250 /* Release the interrupt */
251 free_irq(lvq->config.irq, vq);
Rusty Russell19f15372007-10-22 11:24:21 +1000252 /* Tell virtio_ring.c to free the virtqueue. */
253 vring_del_virtqueue(vq);
254 /* Unmap the pages containing the ring. */
255 lguest_unmap(lvq->pages);
256 /* Free our own queue information. */
257 kfree(lvq);
258}
259
260/* The ops structure which hooks everything together. */
261static struct virtio_config_ops lguest_config_ops = {
262 .find = lg_find,
263 .get = lg_get,
264 .set = lg_set,
265 .get_status = lg_get_status,
266 .set_status = lg_set_status,
267 .find_vq = lg_find_vq,
268 .del_vq = lg_del_vq,
269};
270
271/* The root device for the lguest virtio devices. This makes them appear as
272 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */
273static struct device lguest_root = {
274 .parent = NULL,
275 .bus_id = "lguest",
276};
277
278/*D:120 This is the core of the lguest bus: actually adding a new device.
279 * It's a separate function because it's neater that way, and because an
280 * earlier version of the code supported hotplug and unplug. They were removed
281 * early on because they were never used.
282 *
283 * As Andrew Tridgell says, "Untested code is buggy code".
284 *
285 * It's worth reading this carefully: we start with a pointer to the new device
286 * descriptor in the "lguest_devices" page. */
287static void add_lguest_device(struct lguest_device_desc *d)
288{
289 struct lguest_device *ldev;
290
Rusty Russelle1e72962007-10-25 15:02:50 +1000291 /* Start with zeroed memory; Linux's device layer seems to count on
292 * it. */
Rusty Russell19f15372007-10-22 11:24:21 +1000293 ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
294 if (!ldev) {
295 printk(KERN_EMERG "Cannot allocate lguest dev %u\n",
296 dev_index++);
297 return;
298 }
299
300 /* This devices' parent is the lguest/ dir. */
301 ldev->vdev.dev.parent = &lguest_root;
302 /* We have a unique device index thanks to the dev_index counter. */
303 ldev->vdev.index = dev_index++;
304 /* The device type comes straight from the descriptor. There's also a
305 * device vendor field in the virtio_device struct, which we leave as
306 * 0. */
307 ldev->vdev.id.device = d->type;
308 /* We have a simple set of routines for querying the device's
309 * configuration information and setting its status. */
310 ldev->vdev.config = &lguest_config_ops;
311 /* And we remember the device's descriptor for lguest_config_ops. */
312 ldev->desc = d;
313
314 /* register_virtio_device() sets up the generic fields for the struct
315 * virtio_device and calls device_register(). This makes the bus
316 * infrastructure look for a matching driver. */
317 if (register_virtio_device(&ldev->vdev) != 0) {
318 printk(KERN_ERR "Failed to register lguest device %u\n",
319 ldev->vdev.index);
320 kfree(ldev);
321 }
322}
323
324/*D:110 scan_devices() simply iterates through the device page. The type 0 is
325 * reserved to mean "end of devices". */
326static void scan_devices(void)
327{
328 unsigned int i;
329 struct lguest_device_desc *d;
330
331 /* We start at the page beginning, and skip over each entry. */
332 for (i = 0; i < PAGE_SIZE; i += sizeof(*d) + d->config_len) {
333 d = lguest_devices + i;
334
335 /* Once we hit a zero, stop. */
336 if (d->type == 0)
337 break;
338
339 add_lguest_device(d);
340 }
341}
342
343/*D:105 Fairly early in boot, lguest_devices_init() is called to set up the
344 * lguest device infrastructure. We check that we are a Guest by checking
345 * pv_info.name: there are other ways of checking, but this seems most
346 * obvious to me.
347 *
348 * So we can access the "struct lguest_device_desc"s easily, we map that memory
349 * and store the pointer in the global "lguest_devices". Then we register a
350 * root device from which all our devices will hang (this seems to be the
351 * correct sysfs incantation).
352 *
353 * Finally we call scan_devices() which adds all the devices found in the
354 * lguest_devices page. */
355static int __init lguest_devices_init(void)
356{
357 if (strcmp(pv_info.name, "lguest") != 0)
358 return 0;
359
360 if (device_register(&lguest_root) != 0)
361 panic("Could not register lguest root");
362
363 /* Devices are in a single page above top of "normal" mem */
364 lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
365
366 scan_devices();
367 return 0;
368}
369/* We do this after core stuff, but before the drivers. */
370postcore_initcall(lguest_devices_init);
371
372/*D:150 At this point in the journey we used to now wade through the lguest
373 * devices themselves: net, block and console. Since they're all now virtio
374 * devices rather than lguest-specific, I've decided to ignore them. Mostly,
375 * they're kind of boring. But this does mean you'll never experience the
376 * thrill of reading the forbidden love scene buried deep in the block driver.
377 *
378 * "make Launcher" beckons, where we answer questions like "Where do Guests
379 * come from?", and "What do you do when someone asks for optimization?". */