1/*P:050 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 4 * memory. 5 * 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. 9:*/ 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/slab.h> 19#include <asm/io.h> 20#include <asm/paravirt.h> 21#include <asm/lguest_hcall.h> 22 23/* The pointer to our (page) of device descriptions. */ 24static void *lguest_devices; 25 26/* 27 * For Guests, device memory can be used as normal memory, so we cast away the 28 * __iomem to quieten sparse. 29 */ 30static inline void *lguest_map(unsigned long phys_addr, unsigned long pages) 31{ 32 return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages); 33} 34 35static inline void lguest_unmap(void *addr) 36{ 37 iounmap((__force void __iomem *)addr); 38} 39 40/*D:100 41 * Each lguest device is just a virtio device plus a pointer to its entry 42 * in the lguest_devices page. 43 */ 44struct lguest_device { 45 struct virtio_device vdev; 46 47 /* The entry in the lguest_devices page for this device. */ 48 struct lguest_device_desc *desc; 49}; 50 51/* 52 * Since the virtio infrastructure hands us a pointer to the virtio_device all 53 * the time, it helps to have a curt macro to get a pointer to the struct 54 * lguest_device it's enclosed in. 55 */ 56#define to_lgdev(vd) container_of(vd, struct lguest_device, vdev) 57 58/*D:130 59 * Device configurations 60 * 61 * The configuration information for a device consists of one or more 62 * virtqueues, a feature bitmap, and some configuration bytes. The 63 * configuration bytes don't really matter to us: the Launcher sets them up, and 64 * the driver will look at them during setup. 65 * 66 * A convenient routine to return the device's virtqueue config array: 67 * immediately after the descriptor. 68 */ 69static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc) 70{ 71 return (void *)(desc + 1); 72} 73 74/* The features come immediately after the virtqueues. */ 75static u8 *lg_features(const struct lguest_device_desc *desc) 76{ 77 return (void *)(lg_vq(desc) + desc->num_vq); 78} 79 80/* The config space comes after the two feature bitmasks. */ 81static u8 *lg_config(const struct lguest_device_desc *desc) 82{ 83 return lg_features(desc) + desc->feature_len * 2; 84} 85 86/* The total size of the config page used by this device (incl. desc) */ 87static unsigned desc_size(const struct lguest_device_desc *desc) 88{ 89 return sizeof(*desc) 90 + desc->num_vq * sizeof(struct lguest_vqconfig) 91 + desc->feature_len * 2 92 + desc->config_len; 93} 94 95/* This gets the device's feature bits. */ 96static u32 lg_get_features(struct virtio_device *vdev) 97{ 98 unsigned int i; 99 u32 features = 0; 100 struct lguest_device_desc *desc = to_lgdev(vdev)->desc; 101 u8 *in_features = lg_features(desc); 102 103 /* We do this the slow but generic way. */ 104 for (i = 0; i < min(desc->feature_len * 8, 32); i++) 105 if (in_features[i / 8] & (1 << (i % 8))) 106 features |= (1 << i); 107 108 return features; 109} 110 111/* 112 * The virtio core takes the features the Host offers, and copies the ones 113 * supported by the driver into the vdev->features array. Once that's all 114 * sorted out, this routine is called so we can tell the Host which features we 115 * understand and accept. 116 */ 117static void lg_finalize_features(struct virtio_device *vdev) 118{ 119 unsigned int i, bits; 120 struct lguest_device_desc *desc = to_lgdev(vdev)->desc; 121 /* Second half of bitmap is features we accept. */ 122 u8 *out_features = lg_features(desc) + desc->feature_len; 123 124 /* Give virtio_ring a chance to accept features. */ 125 vring_transport_features(vdev); 126 127 /* 128 * The vdev->feature array is a Linux bitmask: this isn't the same as a 129 * the simple array of bits used by lguest devices for features. So we 130 * do this slow, manual conversion which is completely general. 131 */ 132 memset(out_features, 0, desc->feature_len); 133 bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8; 134 for (i = 0; i < bits; i++) { 135 if (test_bit(i, vdev->features)) 136 out_features[i / 8] |= (1 << (i % 8)); 137 } 138} 139 140/* Once they've found a field, getting a copy of it is easy. */ 141static void lg_get(struct virtio_device *vdev, unsigned int offset, 142 void *buf, unsigned len) 143{ 144 struct lguest_device_desc *desc = to_lgdev(vdev)->desc; 145 146 /* Check they didn't ask for more than the length of the config! */ 147 BUG_ON(offset + len > desc->config_len); 148 memcpy(buf, lg_config(desc) + offset, len); 149} 150 151/* Setting the contents is also trivial. */ 152static void lg_set(struct virtio_device *vdev, unsigned int offset, 153 const void *buf, unsigned len) 154{ 155 struct lguest_device_desc *desc = to_lgdev(vdev)->desc; 156 157 /* Check they didn't ask for more than the length of the config! */ 158 BUG_ON(offset + len > desc->config_len); 159 memcpy(lg_config(desc) + offset, buf, len); 160} 161 162/* 163 * The operations to get and set the status word just access the status field 164 * of the device descriptor. 165 */ 166static u8 lg_get_status(struct virtio_device *vdev) 167{ 168 return to_lgdev(vdev)->desc->status; 169} 170 171/* 172 * To notify on status updates, we (ab)use the NOTIFY hypercall, with the 173 * descriptor address of the device. A zero status means "reset". 174 */ 175static void set_status(struct virtio_device *vdev, u8 status) 176{ 177 unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices; 178 179 /* We set the status. */ 180 to_lgdev(vdev)->desc->status = status; 181 hcall(LHCALL_NOTIFY, (max_pfn << PAGE_SHIFT) + offset, 0, 0, 0); 182} 183 184static void lg_set_status(struct virtio_device *vdev, u8 status) 185{ 186 BUG_ON(!status); 187 set_status(vdev, status); 188} 189 190static void lg_reset(struct virtio_device *vdev) 191{ 192 set_status(vdev, 0); 193} 194 195/* 196 * Virtqueues 197 * 198 * The other piece of infrastructure virtio needs is a "virtqueue": a way of 199 * the Guest device registering buffers for the other side to read from or 200 * write into (ie. send and receive buffers). Each device can have multiple 201 * virtqueues: for example the console driver uses one queue for sending and 202 * another for receiving. 203 * 204 * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue 205 * already exists in virtio_ring.c. We just need to connect it up. 206 * 207 * We start with the information we need to keep about each virtqueue. 208 */ 209 210/*D:140 This is the information we remember about each virtqueue. */ 211struct lguest_vq_info { 212 /* A copy of the information contained in the device config. */ 213 struct lguest_vqconfig config; 214 215 /* The address where we mapped the virtio ring, so we can unmap it. */ 216 void *pages; 217}; 218 219/* 220 * When the virtio_ring code wants to prod the Host, it calls us here and we 221 * make a hypercall. We hand the physical address of the virtqueue so the Host 222 * knows which virtqueue we're talking about. 223 */ 224static void lg_notify(struct virtqueue *vq) 225{ 226 /* 227 * We store our virtqueue information in the "priv" pointer of the 228 * virtqueue structure. 229 */ 230 struct lguest_vq_info *lvq = vq->priv; 231 232 hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0, 0); 233} 234 235/* An extern declaration inside a C file is bad form. Don't do it. */ 236extern void lguest_setup_irq(unsigned int irq); 237 238/* 239 * This routine finds the Nth virtqueue described in the configuration of 240 * this device and sets it up. 241 * 242 * This is kind of an ugly duckling. It'd be nicer to have a standard 243 * representation of a virtqueue in the configuration space, but it seems that 244 * everyone wants to do it differently. The KVM coders want the Guest to 245 * allocate its own pages and tell the Host where they are, but for lguest it's 246 * simpler for the Host to simply tell us where the pages are. 247 */ 248static struct virtqueue *lg_find_vq(struct virtio_device *vdev, 249 unsigned index, 250 void (*callback)(struct virtqueue *vq), 251 const char *name) 252{ 253 struct lguest_device *ldev = to_lgdev(vdev); 254 struct lguest_vq_info *lvq; 255 struct virtqueue *vq; 256 int err; 257 258 /* We must have this many virtqueues. */ 259 if (index >= ldev->desc->num_vq) 260 return ERR_PTR(-ENOENT); 261 262 lvq = kmalloc(sizeof(*lvq), GFP_KERNEL); 263 if (!lvq) 264 return ERR_PTR(-ENOMEM); 265 266 /* 267 * Make a copy of the "struct lguest_vqconfig" entry, which sits after 268 * the descriptor. We need a copy because the config space might not 269 * be aligned correctly. 270 */ 271 memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config)); 272 273 printk("Mapping virtqueue %i addr %lx\n", index, 274 (unsigned long)lvq->config.pfn << PAGE_SHIFT); 275 /* Figure out how many pages the ring will take, and map that memory */ 276 lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT, 277 DIV_ROUND_UP(vring_size(lvq->config.num, 278 LGUEST_VRING_ALIGN), 279 PAGE_SIZE)); 280 if (!lvq->pages) { 281 err = -ENOMEM; 282 goto free_lvq; 283 } 284 285 /* 286 * OK, tell virtio_ring.c to set up a virtqueue now we know its size 287 * and we've got a pointer to its pages. 288 */ 289 vq = vring_new_virtqueue(lvq->config.num, LGUEST_VRING_ALIGN, 290 vdev, lvq->pages, lg_notify, callback, name); 291 if (!vq) { 292 err = -ENOMEM; 293 goto unmap; 294 } 295 296 /* Make sure the interrupt is allocated. */ 297 lguest_setup_irq(lvq->config.irq); 298 299 err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED, 300 dev_name(&vdev->dev), vq); 301 if (err) 302 goto destroy_vring; 303 304 /* 305 * Last of all we hook up our 'struct lguest_vq_info" to the 306 * virtqueue's priv pointer. 307 */ 308 vq->priv = lvq; 309 return vq; 310 311destroy_vring: 312 vring_del_virtqueue(vq); 313unmap: 314 lguest_unmap(lvq->pages); 315free_lvq: 316 kfree(lvq); 317 return ERR_PTR(err); 318} 319/*:*/ 320 321/* Cleaning up a virtqueue is easy */ 322static void lg_del_vq(struct virtqueue *vq) 323{ 324 struct lguest_vq_info *lvq = vq->priv; 325 326 /* Release the interrupt */ 327 free_irq(lvq->config.irq, vq); 328 /* Tell virtio_ring.c to free the virtqueue. */ 329 vring_del_virtqueue(vq); 330 /* Unmap the pages containing the ring. */ 331 lguest_unmap(lvq->pages); 332 /* Free our own queue information. */ 333 kfree(lvq); 334} 335 336static void lg_del_vqs(struct virtio_device *vdev) 337{ 338 struct virtqueue *vq, *n; 339 340 list_for_each_entry_safe(vq, n, &vdev->vqs, list) 341 lg_del_vq(vq); 342} 343 344static int lg_find_vqs(struct virtio_device *vdev, unsigned nvqs, 345 struct virtqueue *vqs[], 346 vq_callback_t *callbacks[], 347 const char *names[]) 348{ 349 struct lguest_device *ldev = to_lgdev(vdev); 350 int i; 351 352 /* We must have this many virtqueues. */ 353 if (nvqs > ldev->desc->num_vq) 354 return -ENOENT; 355 356 for (i = 0; i < nvqs; ++i) { 357 vqs[i] = lg_find_vq(vdev, i, callbacks[i], names[i]); 358 if (IS_ERR(vqs[i])) 359 goto error; 360 } 361 return 0; 362 363error: 364 lg_del_vqs(vdev); 365 return PTR_ERR(vqs[i]); 366} 367 368/* The ops structure which hooks everything together. */ 369static struct virtio_config_ops lguest_config_ops = { 370 .get_features = lg_get_features, 371 .finalize_features = lg_finalize_features, 372 .get = lg_get, 373 .set = lg_set, 374 .get_status = lg_get_status, 375 .set_status = lg_set_status, 376 .reset = lg_reset, 377 .find_vqs = lg_find_vqs, 378 .del_vqs = lg_del_vqs, 379}; 380 381/* 382 * The root device for the lguest virtio devices. This makes them appear as 383 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. 384 */ 385static struct device *lguest_root; 386 387/*D:120 388 * This is the core of the lguest bus: actually adding a new device. 389 * It's a separate function because it's neater that way, and because an 390 * earlier version of the code supported hotplug and unplug. They were removed 391 * early on because they were never used. 392 * 393 * As Andrew Tridgell says, "Untested code is buggy code". 394 * 395 * It's worth reading this carefully: we start with a pointer to the new device 396 * descriptor in the "lguest_devices" page, and the offset into the device 397 * descriptor page so we can uniquely identify it if things go badly wrong. 398 */ 399static void add_lguest_device(struct lguest_device_desc *d, 400 unsigned int offset) 401{ 402 struct lguest_device *ldev; 403 404 /* Start with zeroed memory; Linux's device layer counts on it. */ 405 ldev = kzalloc(sizeof(*ldev), GFP_KERNEL); 406 if (!ldev) { 407 printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n", 408 offset, d->type); 409 return; 410 } 411 412 /* This devices' parent is the lguest/ dir. */ 413 ldev->vdev.dev.parent = lguest_root; 414 /* 415 * The device type comes straight from the descriptor. There's also a 416 * device vendor field in the virtio_device struct, which we leave as 417 * 0. 418 */ 419 ldev->vdev.id.device = d->type; 420 /* 421 * We have a simple set of routines for querying the device's 422 * configuration information and setting its status. 423 */ 424 ldev->vdev.config = &lguest_config_ops; 425 /* And we remember the device's descriptor for lguest_config_ops. */ 426 ldev->desc = d; 427 428 /* 429 * register_virtio_device() sets up the generic fields for the struct 430 * virtio_device and calls device_register(). This makes the bus 431 * infrastructure look for a matching driver. 432 */ 433 if (register_virtio_device(&ldev->vdev) != 0) { 434 printk(KERN_ERR "Failed to register lguest dev %u type %u\n", 435 offset, d->type); 436 kfree(ldev); 437 } 438} 439 440/*D:110 441 * scan_devices() simply iterates through the device page. The type 0 is 442 * reserved to mean "end of devices". 443 */ 444static void scan_devices(void) 445{ 446 unsigned int i; 447 struct lguest_device_desc *d; 448 449 /* We start at the page beginning, and skip over each entry. */ 450 for (i = 0; i < PAGE_SIZE; i += desc_size(d)) { 451 d = lguest_devices + i; 452 453 /* Once we hit a zero, stop. */ 454 if (d->type == 0) 455 break; 456 457 printk("Device at %i has size %u\n", i, desc_size(d)); 458 add_lguest_device(d, i); 459 } 460} 461 462/*D:105 463 * Fairly early in boot, lguest_devices_init() is called to set up the 464 * lguest device infrastructure. We check that we are a Guest by checking 465 * pv_info.name: there are other ways of checking, but this seems most 466 * obvious to me. 467 * 468 * So we can access the "struct lguest_device_desc"s easily, we map that memory 469 * and store the pointer in the global "lguest_devices". Then we register a 470 * root device from which all our devices will hang (this seems to be the 471 * correct sysfs incantation). 472 * 473 * Finally we call scan_devices() which adds all the devices found in the 474 * lguest_devices page. 475 */ 476static int __init lguest_devices_init(void) 477{ 478 if (strcmp(pv_info.name, "lguest") != 0) 479 return 0; 480 481 lguest_root = root_device_register("lguest"); 482 if (IS_ERR(lguest_root)) 483 panic("Could not register lguest root"); 484 485 /* Devices are in a single page above top of "normal" mem */ 486 lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1); 487 488 scan_devices(); 489 return 0; 490} 491/* We do this after core stuff, but before the drivers. */ 492postcore_initcall(lguest_devices_init); 493 494/*D:150 495 * At this point in the journey we used to now wade through the lguest 496 * devices themselves: net, block and console. Since they're all now virtio 497 * devices rather than lguest-specific, I've decided to ignore them. Mostly, 498 * they're kind of boring. But this does mean you'll never experience the 499 * thrill of reading the forbidden love scene buried deep in the block driver. 500 * 501 * "make Launcher" beckons, where we answer questions like "Where do Guests 502 * come from?", and "What do you do when someone asks for optimization?". 503 */ 504