drivers/lguest/lguest_device.c

   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
   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" 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>
  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. */
  21 static void *lguest_devices;
  22
  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)
  26 {
  27         return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages);
  28 }
  29
  30 static 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. */
  37 struct 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.  */
  47 #define to_lgdev(vd) container_of(vd, struct lguest_device, vdev)
  48
  49 /*D:130
  50  * Device configurations
  51  *
  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.
  56  *
  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)
  60 {
  61         return (void *)(desc + 1);
  62 }
  63
  64 /* The features come immediately after the virtqueues. */
  65 static u8 *lg_features(const struct lguest_device_desc *desc)
  66 {
  67         return (void *)(lg_vq(desc) + desc->num_vq);
  68 }
  69
  70 /* The config space comes after the two feature bitmasks. */
  71 static 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) */
  77 static 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
  85 /* This gets the device's feature bits. */
  86 static u32 lg_get_features(struct virtio_device *vdev)
  87 {
  88         unsigned int i;
  89         u32 features = 0;
  90         struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
  91         u8 *in_features = lg_features(desc);
  92
  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)))
  96                         features |= (1 << i);
  97
  98         return features;
  99 }
 100
 101 static void lg_finalize_features(struct virtio_device *vdev)
 102 {
 103         unsigned int i, bits;
 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         /* Give virtio_ring a chance to accept features. */
 109         vring_transport_features(vdev);
 110
 111         memset(out_features, 0, desc->feature_len);
 112         bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8;
 113         for (i = 0; i < bits; i++) {
 114                 if (test_bit(i, vdev->features))
 115                         out_features[i / 8] |= (1 << (i % 8));
 116         }
 117 }
 118
 119 /* Once they've found a field, getting a copy of it is easy. */
 120 static void lg_get(struct virtio_device *vdev, unsigned int offset,
 121                    void *buf, unsigned len)
 122 {
 123         struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
 124
 125         /* Check they didn't ask for more than the length of the config! */
 126         BUG_ON(offset + len > desc->config_len);
 127         memcpy(buf, lg_config(desc) + offset, len);
 128 }
 129
 130 /* Setting the contents is also trivial. */
 131 static void lg_set(struct virtio_device *vdev, unsigned int offset,
 132                    const void *buf, unsigned len)
 133 {
 134         struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
 135
 136         /* Check they didn't ask for more than the length of the config! */
 137         BUG_ON(offset + len > desc->config_len);
 138         memcpy(lg_config(desc) + offset, buf, len);
 139 }
 140
 141 /* The operations to get and set the status word just access the status field
 142  * of the device descriptor. */
 143 static u8 lg_get_status(struct virtio_device *vdev)
 144 {
 145         return to_lgdev(vdev)->desc->status;
 146 }
 147
 148 /* To notify on status updates, we (ab)use the NOTIFY hypercall, with the
 149  * descriptor address of the device.  A zero status means "reset". */
 150 static void set_status(struct virtio_device *vdev, u8 status)
 151 {
 152         unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices;
 153
 154         /* We set the status. */
 155         to_lgdev(vdev)->desc->status = status;
 156         hcall(LHCALL_NOTIFY, (max_pfn<<PAGE_SHIFT) + offset, 0, 0);
 157 }
 158
 159 static void lg_set_status(struct virtio_device *vdev, u8 status)
 160 {
 161         BUG_ON(!status);
 162         set_status(vdev, status);
 163 }
 164
 165 static void lg_reset(struct virtio_device *vdev)
 166 {
 167         set_status(vdev, 0);
 168 }
 169
 170 /*
 171  * Virtqueues
 172  *
 173  * The other piece of infrastructure virtio needs is a "virtqueue": a way of
 174  * the Guest device registering buffers for the other side to read from or
 175  * write into (ie. send and receive buffers).  Each device can have multiple
 176  * virtqueues: for example the console driver uses one queue for sending and
 177  * another for receiving.
 178  *
 179  * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue
 180  * already exists in virtio_ring.c.  We just need to connect it up.
 181  *
 182  * We start with the information we need to keep about each virtqueue.
 183  */
 184
 185 /*D:140 This is the information we remember about each virtqueue. */
 186 struct lguest_vq_info
 187 {
 188         /* A copy of the information contained in the device config. */
 189         struct lguest_vqconfig config;
 190
 191         /* The address where we mapped the virtio ring, so we can unmap it. */
 192         void *pages;
 193 };
 194
 195 /* When the virtio_ring code wants to prod the Host, it calls us here and we
 196  * make a hypercall.  We hand the physical address of the virtqueue so the Host
 197  * knows which virtqueue we're talking about. */
 198 static void lg_notify(struct virtqueue *vq)
 199 {
 200         /* We store our virtqueue information in the "priv" pointer of the
 201          * virtqueue structure. */
 202         struct lguest_vq_info *lvq = vq->priv;
 203
 204         hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0);
 205 }
 206
 207 /* This routine finds the first virtqueue described in the configuration of
 208  * this device and sets it up.
 209  *
 210  * This is kind of an ugly duckling.  It'd be nicer to have a standard
 211  * representation of a virtqueue in the configuration space, but it seems that
 212  * everyone wants to do it differently.  The KVM coders want the Guest to
 213  * allocate its own pages and tell the Host where they are, but for lguest it's
 214  * simpler for the Host to simply tell us where the pages are.
 215  *
 216  * So we provide drivers with a "find the Nth virtqueue and set it up"
 217  * function. */
 218 static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
 219                                     unsigned index,
 220                                     void (*callback)(struct virtqueue *vq))
 221 {
 222         struct lguest_device *ldev = to_lgdev(vdev);
 223         struct lguest_vq_info *lvq;
 224         struct virtqueue *vq;
 225         int err;
 226
 227         /* We must have this many virtqueues. */
 228         if (index >= ldev->desc->num_vq)
 229                 return ERR_PTR(-ENOENT);
 230
 231         lvq = kmalloc(sizeof(*lvq), GFP_KERNEL);
 232         if (!lvq)
 233                 return ERR_PTR(-ENOMEM);
 234
 235         /* Make a copy of the "struct lguest_vqconfig" entry, which sits after
 236          * the descriptor.  We need a copy because the config space might not
 237          * be aligned correctly. */
 238         memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config));
 239
 240         printk("Mapping virtqueue %i addr %lx\n", index,
 241                (unsigned long)lvq->config.pfn << PAGE_SHIFT);
 242         /* Figure out how many pages the ring will take, and map that memory */
 243         lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT,
 244                                 DIV_ROUND_UP(vring_size(lvq->config.num,
 245                                                         PAGE_SIZE),
 246                                              PAGE_SIZE));
 247         if (!lvq->pages) {
 248                 err = -ENOMEM;
 249                 goto free_lvq;
 250         }
 251
 252         /* OK, tell virtio_ring.c to set up a virtqueue now we know its size
 253          * and we've got a pointer to its pages. */
 254         vq = vring_new_virtqueue(lvq->config.num, vdev, lvq->pages,
 255                                  lg_notify, callback);
 256         if (!vq) {
 257                 err = -ENOMEM;
 258                 goto unmap;
 259         }
 260
 261         /* Tell the interrupt for this virtqueue to go to the virtio_ring
 262          * interrupt handler. */
 263         /* FIXME: We used to have a flag for the Host to tell us we could use
 264          * the interrupt as a source of randomness: it'd be nice to have that
 265          * back.. */
 266         err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
 267                           vdev->dev.bus_id, vq);
 268         if (err)
 269                 goto destroy_vring;
 270
 271         /* Last of all we hook up our 'struct lguest_vq_info" to the
 272          * virtqueue's priv pointer. */
 273         vq->priv = lvq;
 274         return vq;
 275
 276 destroy_vring:
 277         vring_del_virtqueue(vq);
 278 unmap:
 279         lguest_unmap(lvq->pages);
 280 free_lvq:
 281         kfree(lvq);
 282         return ERR_PTR(err);
 283 }
 284 /*:*/
 285
 286 /* Cleaning up a virtqueue is easy */
 287 static void lg_del_vq(struct virtqueue *vq)
 288 {
 289         struct lguest_vq_info *lvq = vq->priv;
 290
 291         /* Release the interrupt */
 292         free_irq(lvq->config.irq, vq);
 293         /* Tell virtio_ring.c to free the virtqueue. */
 294         vring_del_virtqueue(vq);
 295         /* Unmap the pages containing the ring. */
 296         lguest_unmap(lvq->pages);
 297         /* Free our own queue information. */
 298         kfree(lvq);
 299 }
 300
 301 /* The ops structure which hooks everything together. */
 302 static struct virtio_config_ops lguest_config_ops = {
 303         .get_features = lg_get_features,
 304         .finalize_features = lg_finalize_features,
 305         .get = lg_get,
 306         .set = lg_set,
 307         .get_status = lg_get_status,
 308         .set_status = lg_set_status,
 309         .reset = lg_reset,
 310         .find_vq = lg_find_vq,
 311         .del_vq = lg_del_vq,
 312 };
 313
 314 /* The root device for the lguest virtio devices.  This makes them appear as
 315  * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */
 316 static struct device lguest_root = {
 317         .parent = NULL,
 318         .bus_id = "lguest",
 319 };
 320
 321 /*D:120 This is the core of the lguest bus: actually adding a new device.
 322  * It's a separate function because it's neater that way, and because an
 323  * earlier version of the code supported hotplug and unplug.  They were removed
 324  * early on because they were never used.
 325  *
 326  * As Andrew Tridgell says, "Untested code is buggy code".
 327  *
 328  * It's worth reading this carefully: we start with a pointer to the new device
 329  * descriptor in the "lguest_devices" page, and the offset into the device
 330  * descriptor page so we can uniquely identify it if things go badly wrong. */
 331 static void add_lguest_device(struct lguest_device_desc *d,
 332                               unsigned int offset)
 333 {
 334         struct lguest_device *ldev;
 335
 336         /* Start with zeroed memory; Linux's device layer seems to count on
 337          * it. */
 338         ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
 339         if (!ldev) {
 340                 printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n",
 341                        offset, d->type);
 342                 return;
 343         }
 344
 345         /* This devices' parent is the lguest/ dir. */
 346         ldev->vdev.dev.parent = &lguest_root;
 347         /* We have a unique device index thanks to the dev_index counter. */
 348         ldev->vdev.id.device = d->type;
 349         /* We have a simple set of routines for querying the device's
 350          * configuration information and setting its status. */
 351         ldev->vdev.config = &lguest_config_ops;
 352         /* And we remember the device's descriptor for lguest_config_ops. */
 353         ldev->desc = d;
 354
 355         /* register_virtio_device() sets up the generic fields for the struct
 356          * virtio_device and calls device_register().  This makes the bus
 357          * infrastructure look for a matching driver. */
 358         if (register_virtio_device(&ldev->vdev) != 0) {
 359                 printk(KERN_ERR "Failed to register lguest dev %u type %u\n",
 360                        offset, d->type);
 361                 kfree(ldev);
 362         }
 363 }
 364
 365 /*D:110 scan_devices() simply iterates through the device page.  The type 0 is
 366  * reserved to mean "end of devices". */
 367 static void scan_devices(void)
 368 {
 369         unsigned int i;
 370         struct lguest_device_desc *d;
 371
 372         /* We start at the page beginning, and skip over each entry. */
 373         for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
 374                 d = lguest_devices + i;
 375
 376                 /* Once we hit a zero, stop. */
 377                 if (d->type == 0)
 378                         break;
 379
 380                 printk("Device at %i has size %u\n", i, desc_size(d));
 381                 add_lguest_device(d, i);
 382         }
 383 }
 384
 385 /*D:105 Fairly early in boot, lguest_devices_init() is called to set up the
 386  * lguest device infrastructure.  We check that we are a Guest by checking
 387  * pv_info.name: there are other ways of checking, but this seems most
 388  * obvious to me.
 389  *
 390  * So we can access the "struct lguest_device_desc"s easily, we map that memory
 391  * and store the pointer in the global "lguest_devices".  Then we register a
 392  * root device from which all our devices will hang (this seems to be the
 393  * correct sysfs incantation).
 394  *
 395  * Finally we call scan_devices() which adds all the devices found in the
 396  * lguest_devices page. */
 397 static int __init lguest_devices_init(void)
 398 {
 399         if (strcmp(pv_info.name, "lguest") != 0)
 400                 return 0;
 401
 402         if (device_register(&lguest_root) != 0)
 403                 panic("Could not register lguest root");
 404
 405         /* Devices are in a single page above top of "normal" mem */
 406         lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
 407
 408         scan_devices();
 409         return 0;
 410 }
 411 /* We do this after core stuff, but before the drivers. */
 412 postcore_initcall(lguest_devices_init);
 413
 414 /*D:150 At this point in the journey we used to now wade through the lguest
 415  * devices themselves: net, block and console.  Since they're all now virtio
 416  * devices rather than lguest-specific, I've decided to ignore them.  Mostly,
 417  * they're kind of boring.  But this does mean you'll never experience the
 418  * thrill of reading the forbidden love scene buried deep in the block driver.
 419  *
 420  * "make Launcher" beckons, where we answer questions like "Where do Guests
 421  * come from?", and "What do you do when someone asks for optimization?". */