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
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.
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>
21 #include <asm/paravirt.h>
22 #include <asm/lguest_hcall.h>
24 /* The pointer to our (page) of device descriptions. */
25 static void *lguest_devices;
28 * For Guests, device memory can be used as normal memory, so we cast away the
29 * __iomem to quieten sparse.
31 static inline void *lguest_map(unsigned long phys_addr, unsigned long pages)
33 return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages);
36 static inline void lguest_unmap(void *addr)
38 iounmap((__force void __iomem *)addr);
42 * Each lguest device is just a virtio device plus a pointer to its entry
43 * in the lguest_devices page.
45 struct lguest_device {
46 struct virtio_device vdev;
48 /* The entry in the lguest_devices page for this device. */
49 struct lguest_device_desc *desc;
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.
57 #define to_lgdev(vd) container_of(vd, struct lguest_device, vdev)
60 * Device configurations
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.
67 * A convenient routine to return the device's virtqueue config array:
68 * immediately after the descriptor.
70 static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc)
72 return (void *)(desc + 1);
75 /* The features come immediately after the virtqueues. */
76 static u8 *lg_features(const struct lguest_device_desc *desc)
78 return (void *)(lg_vq(desc) + desc->num_vq);
81 /* The config space comes after the two feature bitmasks. */
82 static u8 *lg_config(const struct lguest_device_desc *desc)
84 return lg_features(desc) + desc->feature_len * 2;
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)
91 + desc->num_vq * sizeof(struct lguest_vqconfig)
92 + desc->feature_len * 2
96 /* This gets the device's feature bits. */
97 static u32 lg_get_features(struct virtio_device *vdev)
101 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
102 u8 *in_features = lg_features(desc);
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);
113 * To notify on reset or feature finalization, we (ab)use the NOTIFY
114 * hypercall, with the descriptor address of the device.
116 static void status_notify(struct virtio_device *vdev)
118 unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices;
120 hcall(LHCALL_NOTIFY, (max_pfn << PAGE_SHIFT) + offset, 0, 0, 0);
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.
129 static void lg_finalize_features(struct virtio_device *vdev)
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;
136 /* Give virtio_ring a chance to accept features. */
137 vring_transport_features(vdev);
140 * The vdev->feature array is a Linux bitmask: this isn't the same as a
141 * the simple array of bits used by lguest devices for features. So we
142 * do this slow, manual conversion which is completely general.
144 memset(out_features, 0, desc->feature_len);
145 bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8;
146 for (i = 0; i < bits; i++) {
147 if (test_bit(i, vdev->features))
148 out_features[i / 8] |= (1 << (i % 8));
151 /* Tell Host we've finished with this device's feature negotiation */
155 /* Once they've found a field, getting a copy of it is easy. */
156 static void lg_get(struct virtio_device *vdev, unsigned int offset,
157 void *buf, unsigned len)
159 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
161 /* Check they didn't ask for more than the length of the config! */
162 BUG_ON(offset + len > desc->config_len);
163 memcpy(buf, lg_config(desc) + offset, len);
166 /* Setting the contents is also trivial. */
167 static void lg_set(struct virtio_device *vdev, unsigned int offset,
168 const void *buf, unsigned len)
170 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
172 /* Check they didn't ask for more than the length of the config! */
173 BUG_ON(offset + len > desc->config_len);
174 memcpy(lg_config(desc) + offset, buf, len);
178 * The operations to get and set the status word just access the status field
179 * of the device descriptor.
181 static u8 lg_get_status(struct virtio_device *vdev)
183 return to_lgdev(vdev)->desc->status;
186 static void lg_set_status(struct virtio_device *vdev, u8 status)
189 to_lgdev(vdev)->desc->status = status;
191 /* Tell Host immediately if we failed. */
192 if (status & VIRTIO_CONFIG_S_FAILED)
196 static void lg_reset(struct virtio_device *vdev)
198 /* 0 status means "reset" */
199 to_lgdev(vdev)->desc->status = 0;
206 * The other piece of infrastructure virtio needs is a "virtqueue": a way of
207 * the Guest device registering buffers for the other side to read from or
208 * write into (ie. send and receive buffers). Each device can have multiple
209 * virtqueues: for example the console driver uses one queue for sending and
210 * another for receiving.
212 * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue
213 * already exists in virtio_ring.c. We just need to connect it up.
215 * We start with the information we need to keep about each virtqueue.
218 /*D:140 This is the information we remember about each virtqueue. */
219 struct lguest_vq_info {
220 /* A copy of the information contained in the device config. */
221 struct lguest_vqconfig config;
223 /* The address where we mapped the virtio ring, so we can unmap it. */
228 * When the virtio_ring code wants to prod the Host, it calls us here and we
229 * make a hypercall. We hand the physical address of the virtqueue so the Host
230 * knows which virtqueue we're talking about.
232 static bool lg_notify(struct virtqueue *vq)
235 * We store our virtqueue information in the "priv" pointer of the
236 * virtqueue structure.
238 struct lguest_vq_info *lvq = vq->priv;
240 hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0, 0);
244 /* An extern declaration inside a C file is bad form. Don't do it. */
245 extern int lguest_setup_irq(unsigned int irq);
248 * This routine finds the Nth virtqueue described in the configuration of
249 * this device and sets it up.
251 * This is kind of an ugly duckling. It'd be nicer to have a standard
252 * representation of a virtqueue in the configuration space, but it seems that
253 * everyone wants to do it differently. The KVM coders want the Guest to
254 * allocate its own pages and tell the Host where they are, but for lguest it's
255 * simpler for the Host to simply tell us where the pages are.
257 static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
259 void (*callback)(struct virtqueue *vq),
262 struct lguest_device *ldev = to_lgdev(vdev);
263 struct lguest_vq_info *lvq;
264 struct virtqueue *vq;
270 /* We must have this many virtqueues. */
271 if (index >= ldev->desc->num_vq)
272 return ERR_PTR(-ENOENT);
274 lvq = kmalloc(sizeof(*lvq), GFP_KERNEL);
276 return ERR_PTR(-ENOMEM);
279 * Make a copy of the "struct lguest_vqconfig" entry, which sits after
280 * the descriptor. We need a copy because the config space might not
281 * be aligned correctly.
283 memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config));
285 printk("Mapping virtqueue %i addr %lx\n", index,
286 (unsigned long)lvq->config.pfn << PAGE_SHIFT);
287 /* Figure out how many pages the ring will take, and map that memory */
288 lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT,
289 DIV_ROUND_UP(vring_size(lvq->config.num,
298 * OK, tell virtio_ring.c to set up a virtqueue now we know its size
299 * and we've got a pointer to its pages. Note that we set weak_barriers
300 * to 'true': the host just a(nother) SMP CPU, so we only need inter-cpu
303 vq = vring_new_virtqueue(index, lvq->config.num, LGUEST_VRING_ALIGN, vdev,
304 true, lvq->pages, lg_notify, callback, name);
310 /* Make sure the interrupt is allocated. */
311 err = lguest_setup_irq(lvq->config.irq);
316 * Tell the interrupt for this virtqueue to go to the virtio_ring
319 * FIXME: We used to have a flag for the Host to tell us we could use
320 * the interrupt as a source of randomness: it'd be nice to have that
323 err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
324 dev_name(&vdev->dev), vq);
329 * Last of all we hook up our 'struct lguest_vq_info" to the
330 * virtqueue's priv pointer.
336 irq_free_desc(lvq->config.irq);
338 vring_del_virtqueue(vq);
340 lguest_unmap(lvq->pages);
347 /* Cleaning up a virtqueue is easy */
348 static void lg_del_vq(struct virtqueue *vq)
350 struct lguest_vq_info *lvq = vq->priv;
352 /* Release the interrupt */
353 free_irq(lvq->config.irq, vq);
354 /* Tell virtio_ring.c to free the virtqueue. */
355 vring_del_virtqueue(vq);
356 /* Unmap the pages containing the ring. */
357 lguest_unmap(lvq->pages);
358 /* Free our own queue information. */
362 static void lg_del_vqs(struct virtio_device *vdev)
364 struct virtqueue *vq, *n;
366 list_for_each_entry_safe(vq, n, &vdev->vqs, list)
370 static int lg_find_vqs(struct virtio_device *vdev, unsigned nvqs,
371 struct virtqueue *vqs[],
372 vq_callback_t *callbacks[],
375 struct lguest_device *ldev = to_lgdev(vdev);
378 /* We must have this many virtqueues. */
379 if (nvqs > ldev->desc->num_vq)
382 for (i = 0; i < nvqs; ++i) {
383 vqs[i] = lg_find_vq(vdev, i, callbacks[i], names[i]);
391 return PTR_ERR(vqs[i]);
394 static const char *lg_bus_name(struct virtio_device *vdev)
399 /* The ops structure which hooks everything together. */
400 static const struct virtio_config_ops lguest_config_ops = {
401 .get_features = lg_get_features,
402 .finalize_features = lg_finalize_features,
405 .get_status = lg_get_status,
406 .set_status = lg_set_status,
408 .find_vqs = lg_find_vqs,
409 .del_vqs = lg_del_vqs,
410 .bus_name = lg_bus_name,
414 * The root device for the lguest virtio devices. This makes them appear as
415 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2.
417 static struct device *lguest_root;
420 * This is the core of the lguest bus: actually adding a new device.
421 * It's a separate function because it's neater that way, and because an
422 * earlier version of the code supported hotplug and unplug. They were removed
423 * early on because they were never used.
425 * As Andrew Tridgell says, "Untested code is buggy code".
427 * It's worth reading this carefully: we start with a pointer to the new device
428 * descriptor in the "lguest_devices" page, and the offset into the device
429 * descriptor page so we can uniquely identify it if things go badly wrong.
431 static void add_lguest_device(struct lguest_device_desc *d,
434 struct lguest_device *ldev;
436 /* Start with zeroed memory; Linux's device layer counts on it. */
437 ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
439 printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n",
444 /* This devices' parent is the lguest/ dir. */
445 ldev->vdev.dev.parent = lguest_root;
447 * The device type comes straight from the descriptor. There's also a
448 * device vendor field in the virtio_device struct, which we leave as
451 ldev->vdev.id.device = d->type;
453 * We have a simple set of routines for querying the device's
454 * configuration information and setting its status.
456 ldev->vdev.config = &lguest_config_ops;
457 /* And we remember the device's descriptor for lguest_config_ops. */
461 * register_virtio_device() sets up the generic fields for the struct
462 * virtio_device and calls device_register(). This makes the bus
463 * infrastructure look for a matching driver.
465 if (register_virtio_device(&ldev->vdev) != 0) {
466 printk(KERN_ERR "Failed to register lguest dev %u type %u\n",
473 * scan_devices() simply iterates through the device page. The type 0 is
474 * reserved to mean "end of devices".
476 static void scan_devices(void)
479 struct lguest_device_desc *d;
481 /* We start at the page beginning, and skip over each entry. */
482 for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
483 d = lguest_devices + i;
485 /* Once we hit a zero, stop. */
489 printk("Device at %i has size %u\n", i, desc_size(d));
490 add_lguest_device(d, i);
495 * Fairly early in boot, lguest_devices_init() is called to set up the
496 * lguest device infrastructure. We check that we are a Guest by checking
497 * pv_info.name: there are other ways of checking, but this seems most
500 * So we can access the "struct lguest_device_desc"s easily, we map that memory
501 * and store the pointer in the global "lguest_devices". Then we register a
502 * root device from which all our devices will hang (this seems to be the
503 * correct sysfs incantation).
505 * Finally we call scan_devices() which adds all the devices found in the
506 * lguest_devices page.
508 static int __init lguest_devices_init(void)
510 if (strcmp(pv_info.name, "lguest") != 0)
513 lguest_root = root_device_register("lguest");
514 if (IS_ERR(lguest_root))
515 panic("Could not register lguest root");
517 /* Devices are in a single page above top of "normal" mem */
518 lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
523 /* We do this after core stuff, but before the drivers. */
524 postcore_initcall(lguest_devices_init);
527 * At this point in the journey we used to now wade through the lguest
528 * devices themselves: net, block and console. Since they're all now virtio
529 * devices rather than lguest-specific, I've decided to ignore them. Mostly,
530 * they're kind of boring. But this does mean you'll never experience the
531 * thrill of reading the forbidden love scene buried deep in the block driver.
533 * "make Launcher" beckons, where we answer questions like "Where do Guests
534 * come from?", and "What do you do when someone asks for optimization?".