1 /*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 Guest
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" or two 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>
17 #include <asm/paravirt.h>
18 #include <asm/lguest_hcall.h>
20 /* The pointer to our (page) of device descriptions. */
21 static void *lguest_devices;
23 /* For Guests, device memory can be used as normal memory, so we cast away the
24 * __iomem to quieten sparse. */
25 static inline void *lguest_map(unsigned long phys_addr, unsigned long pages)
27 return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages);
30 static inline void lguest_unmap(void *addr)
32 iounmap((__force void __iomem *)addr);
35 /*D:100 Each lguest device is just a virtio device plus a pointer to its entry
36 * in the lguest_devices page. */
37 struct lguest_device {
38 struct virtio_device vdev;
40 /* The entry in the lguest_devices page for this device. */
41 struct lguest_device_desc *desc;
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. */
47 #define to_lgdev(vd) container_of(vd, struct lguest_device, vdev)
50 * Device configurations
52 * The configuration information for a device consists of one or more
53 * virtqueues, a feature bitmap, and some configuration bytes. The
54 * configuration bytes don't really matter to us: the Launcher sets them up, and
55 * the driver will look at them during setup.
57 * A convenient routine to return the device's virtqueue config array:
58 * immediately after the descriptor. */
59 static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc)
61 return (void *)(desc + 1);
64 /* The features come immediately after the virtqueues. */
65 static u8 *lg_features(const struct lguest_device_desc *desc)
67 return (void *)(lg_vq(desc) + desc->num_vq);
70 /* The config space comes after the two feature bitmasks. */
71 static u8 *lg_config(const struct lguest_device_desc *desc)
73 return lg_features(desc) + desc->feature_len * 2;
76 /* The total size of the config page used by this device (incl. desc) */
77 static unsigned desc_size(const struct lguest_device_desc *desc)
80 + desc->num_vq * sizeof(struct lguest_vqconfig)
81 + desc->feature_len * 2
85 /* This gets the device's feature bits. */
86 static u32 lg_get_features(struct virtio_device *vdev)
90 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
91 u8 *in_features = lg_features(desc);
93 /* 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)))
101 /* The virtio core takes the features the Host offers, and copies the
102 * ones supported by the driver into the vdev->features array. Once
103 * that's all sorted out, this routine is called so we can tell the
104 * Host which features we understand and accept. */
105 static void lg_finalize_features(struct virtio_device *vdev)
107 unsigned int i, bits;
108 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
109 /* Second half of bitmap is features we accept. */
110 u8 *out_features = lg_features(desc) + desc->feature_len;
112 /* Give virtio_ring a chance to accept features. */
113 vring_transport_features(vdev);
115 /* The vdev->feature array is a Linux bitmask: this isn't the
116 * same as a the simple array of bits used by lguest devices
117 * for features. So we do this slow, manual conversion which is
118 * completely general. */
119 memset(out_features, 0, desc->feature_len);
120 bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8;
121 for (i = 0; i < bits; i++) {
122 if (test_bit(i, vdev->features))
123 out_features[i / 8] |= (1 << (i % 8));
127 /* Once they've found a field, getting a copy of it is easy. */
128 static void lg_get(struct virtio_device *vdev, unsigned int offset,
129 void *buf, unsigned len)
131 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
133 /* Check they didn't ask for more than the length of the config! */
134 BUG_ON(offset + len > desc->config_len);
135 memcpy(buf, lg_config(desc) + offset, len);
138 /* Setting the contents is also trivial. */
139 static void lg_set(struct virtio_device *vdev, unsigned int offset,
140 const void *buf, unsigned len)
142 struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
144 /* Check they didn't ask for more than the length of the config! */
145 BUG_ON(offset + len > desc->config_len);
146 memcpy(lg_config(desc) + offset, buf, len);
149 /* The operations to get and set the status word just access the status field
150 * of the device descriptor. */
151 static u8 lg_get_status(struct virtio_device *vdev)
153 return to_lgdev(vdev)->desc->status;
156 /* To notify on status updates, we (ab)use the NOTIFY hypercall, with the
157 * descriptor address of the device. A zero status means "reset". */
158 static void set_status(struct virtio_device *vdev, u8 status)
160 unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices;
162 /* We set the status. */
163 to_lgdev(vdev)->desc->status = status;
164 kvm_hypercall1(LHCALL_NOTIFY, (max_pfn << PAGE_SHIFT) + offset);
167 static void lg_set_status(struct virtio_device *vdev, u8 status)
170 set_status(vdev, status);
173 static void lg_reset(struct virtio_device *vdev)
181 * The other piece of infrastructure virtio needs is a "virtqueue": a way of
182 * the Guest device registering buffers for the other side to read from or
183 * write into (ie. send and receive buffers). Each device can have multiple
184 * virtqueues: for example the console driver uses one queue for sending and
185 * another for receiving.
187 * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue
188 * already exists in virtio_ring.c. We just need to connect it up.
190 * We start with the information we need to keep about each virtqueue.
193 /*D:140 This is the information we remember about each virtqueue. */
194 struct lguest_vq_info
196 /* A copy of the information contained in the device config. */
197 struct lguest_vqconfig config;
199 /* The address where we mapped the virtio ring, so we can unmap it. */
203 /* When the virtio_ring code wants to prod the Host, it calls us here and we
204 * make a hypercall. We hand the physical address of the virtqueue so the Host
205 * knows which virtqueue we're talking about. */
206 static void lg_notify(struct virtqueue *vq)
208 /* We store our virtqueue information in the "priv" pointer of the
209 * virtqueue structure. */
210 struct lguest_vq_info *lvq = vq->priv;
212 kvm_hypercall1(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT);
215 /* An extern declaration inside a C file is bad form. Don't do it. */
216 extern void lguest_setup_irq(unsigned int irq);
218 /* This routine finds the first virtqueue described in the configuration of
219 * this device and sets it up.
221 * This is kind of an ugly duckling. It'd be nicer to have a standard
222 * representation of a virtqueue in the configuration space, but it seems that
223 * everyone wants to do it differently. The KVM coders want the Guest to
224 * allocate its own pages and tell the Host where they are, but for lguest it's
225 * simpler for the Host to simply tell us where the pages are.
227 * So we provide drivers with a "find the Nth virtqueue and set it up"
229 static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
231 void (*callback)(struct virtqueue *vq),
234 struct lguest_device *ldev = to_lgdev(vdev);
235 struct lguest_vq_info *lvq;
236 struct virtqueue *vq;
239 /* We must have this many virtqueues. */
240 if (index >= ldev->desc->num_vq)
241 return ERR_PTR(-ENOENT);
243 lvq = kmalloc(sizeof(*lvq), GFP_KERNEL);
245 return ERR_PTR(-ENOMEM);
247 /* Make a copy of the "struct lguest_vqconfig" entry, which sits after
248 * the descriptor. We need a copy because the config space might not
249 * be aligned correctly. */
250 memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config));
252 printk("Mapping virtqueue %i addr %lx\n", index,
253 (unsigned long)lvq->config.pfn << PAGE_SHIFT);
254 /* Figure out how many pages the ring will take, and map that memory */
255 lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT,
256 DIV_ROUND_UP(vring_size(lvq->config.num,
264 /* OK, tell virtio_ring.c to set up a virtqueue now we know its size
265 * and we've got a pointer to its pages. */
266 vq = vring_new_virtqueue(lvq->config.num, LGUEST_VRING_ALIGN,
267 vdev, lvq->pages, lg_notify, callback, name);
273 /* Make sure the interrupt is allocated. */
274 lguest_setup_irq(lvq->config.irq);
276 /* Tell the interrupt for this virtqueue to go to the virtio_ring
277 * interrupt handler. */
278 /* FIXME: We used to have a flag for the Host to tell us we could use
279 * the interrupt as a source of randomness: it'd be nice to have that
281 err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
282 dev_name(&vdev->dev), vq);
286 /* Last of all we hook up our 'struct lguest_vq_info" to the
287 * virtqueue's priv pointer. */
292 vring_del_virtqueue(vq);
294 lguest_unmap(lvq->pages);
301 /* Cleaning up a virtqueue is easy */
302 static void lg_del_vq(struct virtqueue *vq)
304 struct lguest_vq_info *lvq = vq->priv;
306 /* Release the interrupt */
307 free_irq(lvq->config.irq, vq);
308 /* Tell virtio_ring.c to free the virtqueue. */
309 vring_del_virtqueue(vq);
310 /* Unmap the pages containing the ring. */
311 lguest_unmap(lvq->pages);
312 /* Free our own queue information. */
316 static void lg_del_vqs(struct virtio_device *vdev)
318 struct virtqueue *vq, *n;
320 list_for_each_entry_safe(vq, n, &vdev->vqs, list)
324 static int lg_find_vqs(struct virtio_device *vdev, unsigned nvqs,
325 struct virtqueue *vqs[],
326 vq_callback_t *callbacks[],
329 struct lguest_device *ldev = to_lgdev(vdev);
332 /* We must have this many virtqueues. */
333 if (nvqs > ldev->desc->num_vq)
336 for (i = 0; i < nvqs; ++i) {
337 vqs[i] = lg_find_vq(vdev, i, callbacks[i], names[i]);
345 return PTR_ERR(vqs[i]);
348 /* The ops structure which hooks everything together. */
349 static struct virtio_config_ops lguest_config_ops = {
350 .get_features = lg_get_features,
351 .finalize_features = lg_finalize_features,
354 .get_status = lg_get_status,
355 .set_status = lg_set_status,
357 .find_vqs = lg_find_vqs,
358 .del_vqs = lg_del_vqs,
361 /* The root device for the lguest virtio devices. This makes them appear as
362 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */
363 static struct device *lguest_root;
365 /*D:120 This is the core of the lguest bus: actually adding a new device.
366 * It's a separate function because it's neater that way, and because an
367 * earlier version of the code supported hotplug and unplug. They were removed
368 * early on because they were never used.
370 * As Andrew Tridgell says, "Untested code is buggy code".
372 * It's worth reading this carefully: we start with a pointer to the new device
373 * descriptor in the "lguest_devices" page, and the offset into the device
374 * descriptor page so we can uniquely identify it if things go badly wrong. */
375 static void add_lguest_device(struct lguest_device_desc *d,
378 struct lguest_device *ldev;
380 /* Start with zeroed memory; Linux's device layer seems to count on
382 ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
384 printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n",
389 /* This devices' parent is the lguest/ dir. */
390 ldev->vdev.dev.parent = lguest_root;
391 /* We have a unique device index thanks to the dev_index counter. */
392 ldev->vdev.id.device = d->type;
393 /* We have a simple set of routines for querying the device's
394 * configuration information and setting its status. */
395 ldev->vdev.config = &lguest_config_ops;
396 /* And we remember the device's descriptor for lguest_config_ops. */
399 /* register_virtio_device() sets up the generic fields for the struct
400 * virtio_device and calls device_register(). This makes the bus
401 * infrastructure look for a matching driver. */
402 if (register_virtio_device(&ldev->vdev) != 0) {
403 printk(KERN_ERR "Failed to register lguest dev %u type %u\n",
409 /*D:110 scan_devices() simply iterates through the device page. The type 0 is
410 * reserved to mean "end of devices". */
411 static void scan_devices(void)
414 struct lguest_device_desc *d;
416 /* We start at the page beginning, and skip over each entry. */
417 for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
418 d = lguest_devices + i;
420 /* Once we hit a zero, stop. */
424 printk("Device at %i has size %u\n", i, desc_size(d));
425 add_lguest_device(d, i);
429 /*D:105 Fairly early in boot, lguest_devices_init() is called to set up the
430 * lguest device infrastructure. We check that we are a Guest by checking
431 * pv_info.name: there are other ways of checking, but this seems most
434 * So we can access the "struct lguest_device_desc"s easily, we map that memory
435 * and store the pointer in the global "lguest_devices". Then we register a
436 * root device from which all our devices will hang (this seems to be the
437 * correct sysfs incantation).
439 * Finally we call scan_devices() which adds all the devices found in the
440 * lguest_devices page. */
441 static int __init lguest_devices_init(void)
443 if (strcmp(pv_info.name, "lguest") != 0)
446 lguest_root = root_device_register("lguest");
447 if (IS_ERR(lguest_root))
448 panic("Could not register lguest root");
450 /* Devices are in a single page above top of "normal" mem */
451 lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
456 /* We do this after core stuff, but before the drivers. */
457 postcore_initcall(lguest_devices_init);
459 /*D:150 At this point in the journey we used to now wade through the lguest
460 * devices themselves: net, block and console. Since they're all now virtio
461 * devices rather than lguest-specific, I've decided to ignore them. Mostly,
462 * they're kind of boring. But this does mean you'll never experience the
463 * thrill of reading the forbidden love scene buried deep in the block driver.
465 * "make Launcher" beckons, where we answer questions like "Where do Guests
466 * come from?", and "What do you do when someone asks for optimization?". */