linux/drivers/lguest/lguest_device.c
<<
>>
Prefs
   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. */
  21static 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. */
  25static 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
  30static 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. */
  37struct 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. */
  59static 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. */
  65static 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. */
  71static 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) */
  77static 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. */
  86static 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. */
 105static 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. */
 128static 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. */
 139static 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. */
 151static 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". */
 158static 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        hcall(LHCALL_NOTIFY, (max_pfn<<PAGE_SHIFT) + offset, 0, 0);
 165}
 166
 167static void lg_set_status(struct virtio_device *vdev, u8 status)
 168{
 169        BUG_ON(!status);
 170        set_status(vdev, status);
 171}
 172
 173static 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. */
 194struct 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. */
 206static 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        hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0);
 213}
 214
 215/* This routine finds the first virtqueue described in the configuration of
 216 * this device and sets it up.
 217 *
 218 * This is kind of an ugly duckling.  It'd be nicer to have a standard
 219 * representation of a virtqueue in the configuration space, but it seems that
 220 * everyone wants to do it differently.  The KVM coders want the Guest to
 221 * allocate its own pages and tell the Host where they are, but for lguest it's
 222 * simpler for the Host to simply tell us where the pages are.
 223 *
 224 * So we provide drivers with a "find the Nth virtqueue and set it up"
 225 * function. */
 226static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
 227                                    unsigned index,
 228                                    void (*callback)(struct virtqueue *vq))
 229{
 230        struct lguest_device *ldev = to_lgdev(vdev);
 231        struct lguest_vq_info *lvq;
 232        struct virtqueue *vq;
 233        int err;
 234
 235        /* We must have this many virtqueues. */
 236        if (index >= ldev->desc->num_vq)
 237                return ERR_PTR(-ENOENT);
 238
 239        lvq = kmalloc(sizeof(*lvq), GFP_KERNEL);
 240        if (!lvq)
 241                return ERR_PTR(-ENOMEM);
 242
 243        /* Make a copy of the "struct lguest_vqconfig" entry, which sits after
 244         * the descriptor.  We need a copy because the config space might not
 245         * be aligned correctly. */
 246        memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config));
 247
 248        printk("Mapping virtqueue %i addr %lx\n", index,
 249               (unsigned long)lvq->config.pfn << PAGE_SHIFT);
 250        /* Figure out how many pages the ring will take, and map that memory */
 251        lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT,
 252                                DIV_ROUND_UP(vring_size(lvq->config.num,
 253                                                        PAGE_SIZE),
 254                                             PAGE_SIZE));
 255        if (!lvq->pages) {
 256                err = -ENOMEM;
 257                goto free_lvq;
 258        }
 259
 260        /* OK, tell virtio_ring.c to set up a virtqueue now we know its size
 261         * and we've got a pointer to its pages. */
 262        vq = vring_new_virtqueue(lvq->config.num, vdev, lvq->pages,
 263                                 lg_notify, callback);
 264        if (!vq) {
 265                err = -ENOMEM;
 266                goto unmap;
 267        }
 268
 269        /* Tell the interrupt for this virtqueue to go to the virtio_ring
 270         * interrupt handler. */
 271        /* FIXME: We used to have a flag for the Host to tell us we could use
 272         * the interrupt as a source of randomness: it'd be nice to have that
 273         * back.. */
 274        err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
 275                          vdev->dev.bus_id, vq);
 276        if (err)
 277                goto destroy_vring;
 278
 279        /* Last of all we hook up our 'struct lguest_vq_info" to the
 280         * virtqueue's priv pointer. */
 281        vq->priv = lvq;
 282        return vq;
 283
 284destroy_vring:
 285        vring_del_virtqueue(vq);
 286unmap:
 287        lguest_unmap(lvq->pages);
 288free_lvq:
 289        kfree(lvq);
 290        return ERR_PTR(err);
 291}
 292/*:*/
 293
 294/* Cleaning up a virtqueue is easy */
 295static void lg_del_vq(struct virtqueue *vq)
 296{
 297        struct lguest_vq_info *lvq = vq->priv;
 298
 299        /* Release the interrupt */
 300        free_irq(lvq->config.irq, vq);
 301        /* Tell virtio_ring.c to free the virtqueue. */
 302        vring_del_virtqueue(vq);
 303        /* Unmap the pages containing the ring. */
 304        lguest_unmap(lvq->pages);
 305        /* Free our own queue information. */
 306        kfree(lvq);
 307}
 308
 309/* The ops structure which hooks everything together. */
 310static struct virtio_config_ops lguest_config_ops = {
 311        .get_features = lg_get_features,
 312        .finalize_features = lg_finalize_features,
 313        .get = lg_get,
 314        .set = lg_set,
 315        .get_status = lg_get_status,
 316        .set_status = lg_set_status,
 317        .reset = lg_reset,
 318        .find_vq = lg_find_vq,
 319        .del_vq = lg_del_vq,
 320};
 321
 322/* The root device for the lguest virtio devices.  This makes them appear as
 323 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */
 324static struct device lguest_root = {
 325        .parent = NULL,
 326        .bus_id = "lguest",
 327};
 328
 329/*D:120 This is the core of the lguest bus: actually adding a new device.
 330 * It's a separate function because it's neater that way, and because an
 331 * earlier version of the code supported hotplug and unplug.  They were removed
 332 * early on because they were never used.
 333 *
 334 * As Andrew Tridgell says, "Untested code is buggy code".
 335 *
 336 * It's worth reading this carefully: we start with a pointer to the new device
 337 * descriptor in the "lguest_devices" page, and the offset into the device
 338 * descriptor page so we can uniquely identify it if things go badly wrong. */
 339static void add_lguest_device(struct lguest_device_desc *d,
 340                              unsigned int offset)
 341{
 342        struct lguest_device *ldev;
 343
 344        /* Start with zeroed memory; Linux's device layer seems to count on
 345         * it. */
 346        ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
 347        if (!ldev) {
 348                printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n",
 349                       offset, d->type);
 350                return;
 351        }
 352
 353        /* This devices' parent is the lguest/ dir. */
 354        ldev->vdev.dev.parent = &lguest_root;
 355        /* We have a unique device index thanks to the dev_index counter. */
 356        ldev->vdev.id.device = d->type;
 357        /* We have a simple set of routines for querying the device's
 358         * configuration information and setting its status. */
 359        ldev->vdev.config = &lguest_config_ops;
 360        /* And we remember the device's descriptor for lguest_config_ops. */
 361        ldev->desc = d;
 362
 363        /* register_virtio_device() sets up the generic fields for the struct
 364         * virtio_device and calls device_register().  This makes the bus
 365         * infrastructure look for a matching driver. */
 366        if (register_virtio_device(&ldev->vdev) != 0) {
 367                printk(KERN_ERR "Failed to register lguest dev %u type %u\n",
 368                       offset, d->type);
 369                kfree(ldev);
 370        }
 371}
 372
 373/*D:110 scan_devices() simply iterates through the device page.  The type 0 is
 374 * reserved to mean "end of devices". */
 375static void scan_devices(void)
 376{
 377        unsigned int i;
 378        struct lguest_device_desc *d;
 379
 380        /* We start at the page beginning, and skip over each entry. */
 381        for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
 382                d = lguest_devices + i;
 383
 384                /* Once we hit a zero, stop. */
 385                if (d->type == 0)
 386                        break;
 387
 388                printk("Device at %i has size %u\n", i, desc_size(d));
 389                add_lguest_device(d, i);
 390        }
 391}
 392
 393/*D:105 Fairly early in boot, lguest_devices_init() is called to set up the
 394 * lguest device infrastructure.  We check that we are a Guest by checking
 395 * pv_info.name: there are other ways of checking, but this seems most
 396 * obvious to me.
 397 *
 398 * So we can access the "struct lguest_device_desc"s easily, we map that memory
 399 * and store the pointer in the global "lguest_devices".  Then we register a
 400 * root device from which all our devices will hang (this seems to be the
 401 * correct sysfs incantation).
 402 *
 403 * Finally we call scan_devices() which adds all the devices found in the
 404 * lguest_devices page. */
 405static int __init lguest_devices_init(void)
 406{
 407        if (strcmp(pv_info.name, "lguest") != 0)
 408                return 0;
 409
 410        if (device_register(&lguest_root) != 0)
 411                panic("Could not register lguest root");
 412
 413        /* Devices are in a single page above top of "normal" mem */
 414        lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
 415
 416        scan_devices();
 417        return 0;
 418}
 419/* We do this after core stuff, but before the drivers. */
 420postcore_initcall(lguest_devices_init);
 421
 422/*D:150 At this point in the journey we used to now wade through the lguest
 423 * devices themselves: net, block and console.  Since they're all now virtio
 424 * devices rather than lguest-specific, I've decided to ignore them.  Mostly,
 425 * they're kind of boring.  But this does mean you'll never experience the
 426 * thrill of reading the forbidden love scene buried deep in the block driver.
 427 *
 428 * "make Launcher" beckons, where we answer questions like "Where do Guests
 429 * come from?", and "What do you do when someone asks for optimization?". */
 430
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.