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 /* 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)
 106 {
 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;
 111
 112         /* Give virtio_ring a chance to accept features. */
 113         vring_transport_features(vdev);
 114
 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));
 124         }
 125 }
 126
 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)
 130 {
 131         struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
 132
 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);
 136 }
 137
 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)
 141 {
 142         struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
 143
 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);
 147 }
 148
 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)
 152 {
 153         return to_lgdev(vdev)->desc->status;
 154 }
 155
 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)
 159 {
 160         unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices;
 161
 162         /* We set the status. */
 163         to_lgdev(vdev)->desc->status = status;
 164         kvm_hypercall1(LHCALL_NOTIFY, (max_pfn << PAGE_SHIFT) + offset);
 165 }
 166
 167 static void lg_set_status(struct virtio_device *vdev, u8 status)
 168 {
 169         BUG_ON(!status);
 170         set_status(vdev, status);
 171 }
 172
 173 static void lg_reset(struct virtio_device *vdev)
 174 {
 175         set_status(vdev, 0);
 176 }
 177
 178 /*
 179  * Virtqueues
 180  *
 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.
 186  *
 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.
 189  *
 190  * We start with the information we need to keep about each virtqueue.
 191  */
 192
 193 /*D:140 This is the information we remember about each virtqueue. */
 194 struct lguest_vq_info
 195 {
 196         /* A copy of the information contained in the device config. */
 197         struct lguest_vqconfig config;
 198
 199         /* The address where we mapped the virtio ring, so we can unmap it. */
 200         void *pages;
 201 };
 202
 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)
 207 {
 208         /* We store our virtqueue information in the "priv" pointer of the
 209          * virtqueue structure. */
 210         struct lguest_vq_info *lvq = vq->priv;
 211
 212         kvm_hypercall1(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT);
 213 }
 214
 215 /* An extern declaration inside a C file is bad form.  Don't do it. */
 216 extern void lguest_setup_irq(unsigned int irq);
 217
 218 /* This routine finds the first virtqueue described in the configuration of
 219  * this device and sets it up.
 220  *
 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.
 226  *
 227  * So we provide drivers with a "find the Nth virtqueue and set it up"
 228  * function. */
 229 static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
 230                                     unsigned index,
 231                                     void (*callback)(struct virtqueue *vq))
 232 {
 233         struct lguest_device *ldev = to_lgdev(vdev);
 234         struct lguest_vq_info *lvq;
 235         struct virtqueue *vq;
 236         int err;
 237
 238         /* We must have this many virtqueues. */
 239         if (index >= ldev->desc->num_vq)
 240                 return ERR_PTR(-ENOENT);
 241
 242         lvq = kmalloc(sizeof(*lvq), GFP_KERNEL);
 243         if (!lvq)
 244                 return ERR_PTR(-ENOMEM);
 245
 246         /* Make a copy of the "struct lguest_vqconfig" entry, which sits after
 247          * the descriptor.  We need a copy because the config space might not
 248          * be aligned correctly. */
 249         memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config));
 250
 251         printk("Mapping virtqueue %i addr %lx\n", index,
 252                (unsigned long)lvq->config.pfn << PAGE_SHIFT);
 253         /* Figure out how many pages the ring will take, and map that memory */
 254         lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT,
 255                                 DIV_ROUND_UP(vring_size(lvq->config.num,
 256                                                         LGUEST_VRING_ALIGN),
 257                                              PAGE_SIZE));
 258         if (!lvq->pages) {
 259                 err = -ENOMEM;
 260                 goto free_lvq;
 261         }
 262
 263         /* OK, tell virtio_ring.c to set up a virtqueue now we know its size
 264          * and we've got a pointer to its pages. */
 265         vq = vring_new_virtqueue(lvq->config.num, LGUEST_VRING_ALIGN,
 266                                  vdev, lvq->pages, lg_notify, callback);
 267         if (!vq) {
 268                 err = -ENOMEM;
 269                 goto unmap;
 270         }
 271
 272         /* Make sure the interrupt is allocated. */
 273         lguest_setup_irq(lvq->config.irq);
 274
 275         /* Tell the interrupt for this virtqueue to go to the virtio_ring
 276          * interrupt handler. */
 277         /* FIXME: We used to have a flag for the Host to tell us we could use
 278          * the interrupt as a source of randomness: it'd be nice to have that
 279          * back.. */
 280         err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
 281                           dev_name(&vdev->dev), vq);
 282         if (err)
 283                 goto destroy_vring;
 284
 285         /* Last of all we hook up our 'struct lguest_vq_info" to the
 286          * virtqueue's priv pointer. */
 287         vq->priv = lvq;
 288         return vq;
 289
 290 destroy_vring:
 291         vring_del_virtqueue(vq);
 292 unmap:
 293         lguest_unmap(lvq->pages);
 294 free_lvq:
 295         kfree(lvq);
 296         return ERR_PTR(err);
 297 }
 298 /*:*/
 299
 300 /* Cleaning up a virtqueue is easy */
 301 static void lg_del_vq(struct virtqueue *vq)
 302 {
 303         struct lguest_vq_info *lvq = vq->priv;
 304
 305         /* Release the interrupt */
 306         free_irq(lvq->config.irq, vq);
 307         /* Tell virtio_ring.c to free the virtqueue. */
 308         vring_del_virtqueue(vq);
 309         /* Unmap the pages containing the ring. */
 310         lguest_unmap(lvq->pages);
 311         /* Free our own queue information. */
 312         kfree(lvq);
 313 }
 314
 315 /* The ops structure which hooks everything together. */
 316 static struct virtio_config_ops lguest_config_ops = {
 317         .get_features = lg_get_features,
 318         .finalize_features = lg_finalize_features,
 319         .get = lg_get,
 320         .set = lg_set,
 321         .get_status = lg_get_status,
 322         .set_status = lg_set_status,
 323         .reset = lg_reset,
 324         .find_vq = lg_find_vq,
 325         .del_vq = lg_del_vq,
 326 };
 327
 328 /* The root device for the lguest virtio devices.  This makes them appear as
 329  * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */
 330 static struct device *lguest_root;
 331
 332 /*D:120 This is the core of the lguest bus: actually adding a new device.
 333  * It's a separate function because it's neater that way, and because an
 334  * earlier version of the code supported hotplug and unplug.  They were removed
 335  * early on because they were never used.
 336  *
 337  * As Andrew Tridgell says, "Untested code is buggy code".
 338  *
 339  * It's worth reading this carefully: we start with a pointer to the new device
 340  * descriptor in the "lguest_devices" page, and the offset into the device
 341  * descriptor page so we can uniquely identify it if things go badly wrong. */
 342 static void add_lguest_device(struct lguest_device_desc *d,
 343                               unsigned int offset)
 344 {
 345         struct lguest_device *ldev;
 346
 347         /* Start with zeroed memory; Linux's device layer seems to count on
 348          * it. */
 349         ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
 350         if (!ldev) {
 351                 printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n",
 352                        offset, d->type);
 353                 return;
 354         }
 355
 356         /* This devices' parent is the lguest/ dir. */
 357         ldev->vdev.dev.parent = lguest_root;
 358         /* We have a unique device index thanks to the dev_index counter. */
 359         ldev->vdev.id.device = d->type;
 360         /* We have a simple set of routines for querying the device's
 361          * configuration information and setting its status. */
 362         ldev->vdev.config = &lguest_config_ops;
 363         /* And we remember the device's descriptor for lguest_config_ops. */
 364         ldev->desc = d;
 365
 366         /* register_virtio_device() sets up the generic fields for the struct
 367          * virtio_device and calls device_register().  This makes the bus
 368          * infrastructure look for a matching driver. */
 369         if (register_virtio_device(&ldev->vdev) != 0) {
 370                 printk(KERN_ERR "Failed to register lguest dev %u type %u\n",
 371                        offset, d->type);
 372                 kfree(ldev);
 373         }
 374 }
 375
 376 /*D:110 scan_devices() simply iterates through the device page.  The type 0 is
 377  * reserved to mean "end of devices". */
 378 static void scan_devices(void)
 379 {
 380         unsigned int i;
 381         struct lguest_device_desc *d;
 382
 383         /* We start at the page beginning, and skip over each entry. */
 384         for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
 385                 d = lguest_devices + i;
 386
 387                 /* Once we hit a zero, stop. */
 388                 if (d->type == 0)
 389                         break;
 390
 391                 printk("Device at %i has size %u\n", i, desc_size(d));
 392                 add_lguest_device(d, i);
 393         }
 394 }
 395
 396 /*D:105 Fairly early in boot, lguest_devices_init() is called to set up the
 397  * lguest device infrastructure.  We check that we are a Guest by checking
 398  * pv_info.name: there are other ways of checking, but this seems most
 399  * obvious to me.
 400  *
 401  * So we can access the "struct lguest_device_desc"s easily, we map that memory
 402  * and store the pointer in the global "lguest_devices".  Then we register a
 403  * root device from which all our devices will hang (this seems to be the
 404  * correct sysfs incantation).
 405  *
 406  * Finally we call scan_devices() which adds all the devices found in the
 407  * lguest_devices page. */
 408 static int __init lguest_devices_init(void)
 409 {
 410         if (strcmp(pv_info.name, "lguest") != 0)
 411                 return 0;
 412
 413         lguest_root = root_device_register("lguest");
 414         if (IS_ERR(lguest_root))
 415                 panic("Could not register lguest root");
 416
 417         /* Devices are in a single page above top of "normal" mem */
 418         lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
 419
 420         scan_devices();
 421         return 0;
 422 }
 423 /* We do this after core stuff, but before the drivers. */
 424 postcore_initcall(lguest_devices_init);
 425
 426 /*D:150 At this point in the journey we used to now wade through the lguest
 427  * devices themselves: net, block and console.  Since they're all now virtio
 428  * devices rather than lguest-specific, I've decided to ignore them.  Mostly,
 429  * they're kind of boring.  But this does mean you'll never experience the
 430  * thrill of reading the forbidden love scene buried deep in the block driver.
 431  *
 432  * "make Launcher" beckons, where we answer questions like "Where do Guests
 433  * come from?", and "What do you do when someone asks for optimization?". */