palacios/src/devices/lnx_virtio_blk.c

   1 /*
   2  * This file is part of the Palacios Virtual Machine Monitor developed
   3  * by the V3VEE Project with funding from the United States National
   4  * Science Foundation and the Department of Energy.
   5  *
   6  * The V3VEE Project is a joint project between Northwestern University
   7  * and the University of New Mexico.  You can find out more at
   8  * http://www.v3vee.org
   9  *
  10  * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
  11  * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
  12  * All rights reserved.
  13  *
  14  * Author: Jack Lange <jarusl@cs.northwestern.edu>
  15  *
  16  * This is free software.  You are permitted to use,
  17  * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
  18  */
  19
  20 #include <palacios/vmm.h>
  21 #include <palacios/vmm_dev_mgr.h>
  22 #include <devices/lnx_virtio_pci.h>
  23 #include <palacios/vm_guest_mem.h>
  24
  25 #include <devices/pci.h>
  26
  27
  28
  29 #ifndef CONFIG_DEBUG_VIRTIO_BLK
  30 #undef PrintDebug
  31 #define PrintDebug(fmt, args...)
  32 #endif
  33
  34
  35 #define SECTOR_SIZE 512
  36
  37 #define BLK_CAPACITY_PORT     20
  38 #define BLK_MAX_SIZE_PORT     28
  39 #define BLK_MAX_SEG_PORT      32
  40 #define BLK_CYLINDERS_PORT    36
  41 #define BLK_HEADS_PORT        38
  42 #define BLK_SECTS_PORT        39
  43
  44 #define BLK_IN_REQ            0
  45 #define BLK_OUT_REQ           1
  46 #define BLK_SCSI_CMD          2
  47
  48 #define BLK_BARRIER_FLAG     0x80000000
  49
  50 #define BLK_STATUS_OK             0
  51 #define BLK_STATUS_ERR            1
  52 #define BLK_STATUS_NOT_SUPPORTED  2
  53
  54
  55 struct blk_config {
  56     uint64_t capacity;
  57     uint32_t max_size;
  58     uint32_t max_seg;
  59     uint16_t cylinders;
  60     uint8_t heads;
  61     uint8_t sectors;
  62 } __attribute__((packed));
  63
  64
  65
  66 struct blk_op_hdr {
  67     uint32_t type;
  68     uint32_t prior;
  69     uint64_t sector;
  70 } __attribute__((packed));
  71
  72 #define QUEUE_SIZE 128
  73
  74 /* Host Feature flags */
  75 #define VIRTIO_BARRIER       0x01       /* Does host support barriers? */
  76 #define VIRTIO_SIZE_MAX      0x02       /* Indicates maximum segment size */
  77 #define VIRTIO_SEG_MAX       0x04       /* Indicates maximum # of segments */
  78 #define VIRTIO_LEGACY_GEOM   0x10       /* Indicates support of legacy geometry */
  79
  80
  81 struct virtio_dev_state {
  82     struct vm_device * pci_bus;
  83     struct list_head dev_list;
  84 };
  85
  86 struct virtio_blk_state {
  87
  88     struct pci_device * pci_dev;
  89     struct blk_config block_cfg;
  90     struct virtio_config virtio_cfg;
  91
  92
  93     struct virtio_queue queue;
  94
  95     struct v3_dev_blk_ops * ops;
  96
  97     void * backend_data;
  98
  99     int io_range_size;
 100
 101     struct virtio_dev_state * virtio_dev;
 102
 103     struct list_head dev_link;
 104 };
 105
 106
 107 static int virtio_free(struct vm_device * dev) {
 108     return -1;
 109 }
 110
 111 static int blk_reset(struct virtio_blk_state * virtio) {
 112
 113     virtio->queue.ring_desc_addr = 0;
 114     virtio->queue.ring_avail_addr = 0;
 115     virtio->queue.ring_used_addr = 0;
 116     virtio->queue.pfn = 0;
 117     virtio->queue.cur_avail_idx = 0;
 118
 119     virtio->virtio_cfg.status = 0;
 120     virtio->virtio_cfg.pci_isr = 0;
 121     return 0;
 122 }
 123
 124
 125
 126
 127 static int handle_read_op(struct virtio_blk_state * blk_state, uint8_t * buf, uint64_t * sector, uint64_t len) {
 128     int ret = -1;
 129
 130     PrintDebug("Reading Disk\n");
 131     ret = blk_state->ops->read(buf, (*sector) * SECTOR_SIZE, len, (void *)(blk_state->backend_data));
 132     *sector += (len / SECTOR_SIZE);
 133
 134     return ret;
 135 }
 136
 137
 138 static int handle_write_op(struct virtio_blk_state * blk_state, uint8_t * buf, uint64_t * sector, uint64_t len) {
 139     int ret = -1;
 140
 141     PrintDebug("Writing Disk\n");
 142     ret = blk_state->ops->write(buf, (*sector) * SECTOR_SIZE, len, (void *)(blk_state->backend_data));
 143     *sector += (len / SECTOR_SIZE);
 144
 145     return ret;
 146 }
 147
 148
 149
 150 // multiple block operations need to increment the sector
 151
 152 static int handle_block_op(struct guest_info * core, struct virtio_blk_state * blk_state, struct blk_op_hdr * hdr,
 153                            struct vring_desc * buf_desc, uint8_t * status) {
 154     uint8_t * buf = NULL;
 155
 156     PrintDebug("Handling Block op\n");
 157     if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
 158         PrintError("Could not translate buffer address\n");
 159         return -1;
 160     }
 161
 162     PrintDebug("Sector=%p Length=%d\n", (void *)(addr_t)(hdr->sector), buf_desc->length);
 163
 164     if (hdr->type == BLK_IN_REQ) {
 165         if (handle_read_op(blk_state, buf, &(hdr->sector), buf_desc->length) == -1) {
 166             *status = BLK_STATUS_ERR;
 167             return -1;
 168         } else {
 169             *status = BLK_STATUS_OK;
 170         }
 171     } else if (hdr->type == BLK_OUT_REQ) {
 172         if (handle_write_op(blk_state, buf, &(hdr->sector), buf_desc->length) == -1) {
 173             *status = BLK_STATUS_ERR;
 174             return -1;
 175         } else {
 176             *status = BLK_STATUS_OK;
 177         }
 178     } else if (hdr->type == BLK_SCSI_CMD) {
 179         PrintError("VIRTIO: SCSI Command Not supported!!!\n");
 180         *status = BLK_STATUS_NOT_SUPPORTED;
 181         return -1;
 182     }
 183
 184     PrintDebug("Returning Status: %d\n", *status);
 185
 186     return 0;
 187 }
 188
 189 static int get_desc_count(struct virtio_queue * q, int index) {
 190     struct vring_desc * tmp_desc = &(q->desc[index]);
 191     int cnt = 1;
 192
 193     while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
 194         tmp_desc = &(q->desc[tmp_desc->next]);
 195         cnt++;
 196     }
 197
 198     return cnt;
 199 }
 200
 201
 202
 203 static int handle_kick(struct guest_info * core, struct virtio_blk_state * blk_state) {
 204     struct virtio_queue * q = &(blk_state->queue);
 205
 206     PrintDebug("VIRTIO KICK: cur_index=%d (mod=%d), avail_index=%d\n",
 207                q->cur_avail_idx, q->cur_avail_idx % QUEUE_SIZE, q->avail->index);
 208
 209     while (q->cur_avail_idx != q->avail->index) {
 210         struct vring_desc * hdr_desc = NULL;
 211         struct vring_desc * buf_desc = NULL;
 212         struct vring_desc * status_desc = NULL;
 213         struct blk_op_hdr hdr;
 214         addr_t hdr_addr = 0;
 215         uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
 216         int desc_cnt = get_desc_count(q, desc_idx);
 217         int i = 0;
 218         uint8_t * status_ptr = NULL;
 219         uint8_t status = BLK_STATUS_OK;
 220         uint32_t req_len = 0;
 221
 222         PrintDebug("Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE);
 223
 224         if (desc_cnt < 3) {
 225             PrintError("Block operations must include at least 3 descriptors\n");
 226             return -1;
 227         }
 228
 229         hdr_desc = &(q->desc[desc_idx]);
 230
 231
 232         PrintDebug("Header Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", hdr_desc,
 233                    (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next);
 234
 235         if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
 236             PrintError("Could not translate block header address\n");
 237             return -1;
 238         }
 239
 240         // We copy the block op header out because we are going to modify its contents
 241         memcpy(&hdr, (void *)hdr_addr, sizeof(struct blk_op_hdr));
 242
 243         PrintDebug("Blk Op Hdr (ptr=%p) type=%d, sector=%p\n", (void *)hdr_addr, hdr.type, (void *)hdr.sector);
 244
 245         desc_idx = hdr_desc->next;
 246
 247         for (i = 0; i < desc_cnt - 2; i++) {
 248             uint8_t tmp_status = BLK_STATUS_OK;
 249
 250             buf_desc = &(q->desc[desc_idx]);
 251
 252             PrintDebug("Buffer Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", buf_desc,
 253                        (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next);
 254
 255             if (handle_block_op(core, blk_state, &hdr, buf_desc, &tmp_status) == -1) {
 256                 PrintError("Error handling block operation\n");
 257                 return -1;
 258             }
 259
 260             if (tmp_status != BLK_STATUS_OK) {
 261                 status = tmp_status;
 262             }
 263
 264             req_len += buf_desc->length;
 265             desc_idx = buf_desc->next;
 266         }
 267
 268         status_desc = &(q->desc[desc_idx]);
 269
 270         PrintDebug("Status Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", status_desc,
 271                    (void *)(status_desc->addr_gpa), status_desc->length, status_desc->flags, status_desc->next);
 272
 273         if (v3_gpa_to_hva(core, status_desc->addr_gpa, (addr_t *)&(status_ptr)) == -1) {
 274             PrintError("Could not translate status address\n");
 275             return -1;
 276         }
 277
 278         req_len += status_desc->length;
 279         *status_ptr = status;
 280
 281         q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
 282         q->used->ring[q->used->index % QUEUE_SIZE].length = req_len; // What do we set this to????
 283
 284         q->used->index++;
 285         q->cur_avail_idx++;
 286     }
 287
 288     if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
 289         PrintDebug("Raising IRQ %d\n",  blk_state->pci_dev->config_header.intr_line);
 290         v3_pci_raise_irq(blk_state->virtio_dev->pci_bus, 0, blk_state->pci_dev);
 291         blk_state->virtio_cfg.pci_isr = 1;
 292     }
 293
 294     return 0;
 295 }
 296
 297 static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, uint_t length, void * private_data) {
 298     struct virtio_blk_state * blk_state = (struct virtio_blk_state *)private_data;
 299     int port_idx = port % blk_state->io_range_size;
 300
 301
 302     PrintDebug("VIRTIO BLOCK Write for port %d (index=%d) len=%d, value=%x\n",
 303                port, port_idx,  length, *(uint32_t *)src);
 304
 305
 306
 307     switch (port_idx) {
 308         case GUEST_FEATURES_PORT:
 309             if (length != 4) {
 310                 PrintError("Illegal write length for guest features\n");
 311                 return -1;
 312             }
 313
 314             blk_state->virtio_cfg.guest_features = *(uint32_t *)src;
 315             PrintDebug("Setting Guest Features to %x\n", blk_state->virtio_cfg.guest_features);
 316
 317             break;
 318         case VRING_PG_NUM_PORT:
 319             if (length == 4) {
 320                 addr_t pfn = *(uint32_t *)src;
 321                 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
 322
 323
 324                 blk_state->queue.pfn = pfn;
 325
 326                 blk_state->queue.ring_desc_addr = page_addr ;
 327                 blk_state->queue.ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
 328                 blk_state->queue.ring_used_addr = ( blk_state->queue.ring_avail_addr + \
 329                                                  sizeof(struct vring_avail)    + \
 330                                                  (QUEUE_SIZE * sizeof(uint16_t)));
 331
 332                 // round up to next page boundary.
 333                 blk_state->queue.ring_used_addr = (blk_state->queue.ring_used_addr + 0xfff) & ~0xfff;
 334
 335                 if (v3_gpa_to_hva(core, blk_state->queue.ring_desc_addr, (addr_t *)&(blk_state->queue.desc)) == -1) {
 336                     PrintError("Could not translate ring descriptor address\n");
 337                     return -1;
 338                 }
 339
 340
 341                 if (v3_gpa_to_hva(core, blk_state->queue.ring_avail_addr, (addr_t *)&(blk_state->queue.avail)) == -1) {
 342                     PrintError("Could not translate ring available address\n");
 343                     return -1;
 344                 }
 345
 346
 347                 if (v3_gpa_to_hva(core, blk_state->queue.ring_used_addr, (addr_t *)&(blk_state->queue.used)) == -1) {
 348                     PrintError("Could not translate ring used address\n");
 349                     return -1;
 350                 }
 351
 352                 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
 353                            (void *)(blk_state->queue.ring_desc_addr),
 354                            (void *)(blk_state->queue.ring_avail_addr),
 355                            (void *)(blk_state->queue.ring_used_addr));
 356
 357                 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
 358                            blk_state->queue.desc, blk_state->queue.avail, blk_state->queue.used);
 359
 360             } else {
 361                 PrintError("Illegal write length for page frame number\n");
 362                 return -1;
 363             }
 364             break;
 365         case VRING_Q_SEL_PORT:
 366             blk_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
 367
 368             if (blk_state->virtio_cfg.vring_queue_selector != 0) {
 369                 PrintError("Virtio Block device only uses 1 queue, selected %d\n",
 370                            blk_state->virtio_cfg.vring_queue_selector);
 371                 return -1;
 372             }
 373
 374             break;
 375         case VRING_Q_NOTIFY_PORT:
 376             PrintDebug("Handling Kick\n");
 377             if (handle_kick(core, blk_state) == -1) {
 378                 PrintError("Could not handle Block Notification\n");
 379                 return -1;
 380             }
 381             break;
 382         case VIRTIO_STATUS_PORT:
 383             blk_state->virtio_cfg.status = *(uint8_t *)src;
 384
 385             if (blk_state->virtio_cfg.status == 0) {
 386                 PrintDebug("Resetting device\n");
 387                 blk_reset(blk_state);
 388             }
 389
 390             break;
 391
 392         case VIRTIO_ISR_PORT:
 393             blk_state->virtio_cfg.pci_isr = *(uint8_t *)src;
 394             break;
 395         default:
 396             return -1;
 397             break;
 398     }
 399
 400     return length;
 401 }
 402
 403
 404 static int virtio_io_read(struct guest_info * core, uint16_t port, void * dst, uint_t length, void * private_data) {
 405     struct virtio_blk_state * blk_state = (struct virtio_blk_state *)private_data;
 406     int port_idx = port % blk_state->io_range_size;
 407
 408
 409     PrintDebug("VIRTIO BLOCK Read  for port %d (index =%d), length=%d\n",
 410                port, port_idx, length);
 411
 412     switch (port_idx) {
 413         case HOST_FEATURES_PORT:
 414             if (length != 4) {
 415                 PrintError("Illegal read length for host features\n");
 416                 return -1;
 417             }
 418
 419             *(uint32_t *)dst = blk_state->virtio_cfg.host_features;
 420
 421             break;
 422         case VRING_PG_NUM_PORT:
 423             if (length != 4) {
 424                 PrintError("Illegal read length for page frame number\n");
 425                 return -1;
 426             }
 427
 428             *(uint32_t *)dst = blk_state->queue.pfn;
 429
 430             break;
 431         case VRING_SIZE_PORT:
 432             if (length != 2) {
 433                 PrintError("Illegal read length for vring size\n");
 434                 return -1;
 435             }
 436
 437             *(uint16_t *)dst = blk_state->queue.queue_size;
 438
 439             break;
 440
 441         case VIRTIO_STATUS_PORT:
 442             if (length != 1) {
 443                 PrintError("Illegal read length for status\n");
 444                 return -1;
 445             }
 446
 447             *(uint8_t *)dst = blk_state->virtio_cfg.status;
 448             break;
 449
 450         case VIRTIO_ISR_PORT:
 451             *(uint8_t *)dst = blk_state->virtio_cfg.pci_isr;
 452             blk_state->virtio_cfg.pci_isr = 0;
 453             v3_pci_lower_irq(blk_state->virtio_dev->pci_bus, 0, blk_state->pci_dev);
 454             break;
 455
 456         default:
 457             if ( (port_idx >= sizeof(struct virtio_config)) &&
 458                  (port_idx < (sizeof(struct virtio_config) + sizeof(struct blk_config))) ) {
 459                 int cfg_offset = port_idx - sizeof(struct virtio_config);
 460                 uint8_t * cfg_ptr = (uint8_t *)&(blk_state->block_cfg);
 461
 462                 memcpy(dst, cfg_ptr + cfg_offset, length);
 463
 464             } else {
 465                 PrintError("Read of Unhandled Virtio Read\n");
 466                 return -1;
 467             }
 468
 469             break;
 470     }
 471
 472     return length;
 473 }
 474
 475
 476
 477
 478 static struct v3_device_ops dev_ops = {
 479     .free = virtio_free,
 480
 481 };
 482
 483
 484
 485
 486
 487 static int register_dev(struct virtio_dev_state * virtio, struct virtio_blk_state * blk_state) {
 488     // initialize PCI
 489     struct pci_device * pci_dev = NULL;
 490     struct v3_pci_bar bars[6];
 491     int num_ports = sizeof(struct virtio_config) + sizeof(struct blk_config);
 492     int tmp_ports = num_ports;
 493     int i;
 494
 495
 496
 497     // This gets the number of ports, rounded up to a power of 2
 498     blk_state->io_range_size = 1; // must be a power of 2
 499
 500     while (tmp_ports > 0) {
 501         tmp_ports >>= 1;
 502         blk_state->io_range_size <<= 1;
 503     }
 504
 505     // this is to account for any low order bits being set in num_ports
 506     // if there are none, then num_ports was already a power of 2 so we shift right to reset it
 507     if ((num_ports & ((blk_state->io_range_size >> 1) - 1)) == 0) {
 508         blk_state->io_range_size >>= 1;
 509     }
 510
 511
 512     for (i = 0; i < 6; i++) {
 513         bars[i].type = PCI_BAR_NONE;
 514     }
 515
 516     PrintDebug("Virtio-BLK io_range_size = %d\n", blk_state->io_range_size);
 517
 518     bars[0].type = PCI_BAR_IO;
 519     bars[0].default_base_port = -1;
 520     bars[0].num_ports = blk_state->io_range_size;
 521
 522     bars[0].io_read = virtio_io_read;
 523     bars[0].io_write = virtio_io_write;
 524     bars[0].private_data = blk_state;
 525
 526     pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
 527                                      0, PCI_AUTO_DEV_NUM, 0,
 528                                      "LNX_VIRTIO_BLK", bars,
 529                                      NULL, NULL, NULL, blk_state);
 530
 531     if (!pci_dev) {
 532         PrintError("Could not register PCI Device\n");
 533         return -1;
 534     }
 535
 536     pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
 537     pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
 538
 539
 540     pci_dev->config_header.device_id = VIRTIO_BLOCK_DEV_ID;
 541     pci_dev->config_header.class = PCI_CLASS_STORAGE;
 542     pci_dev->config_header.subclass = PCI_STORAGE_SUBCLASS_OTHER;
 543
 544     pci_dev->config_header.subsystem_id = VIRTIO_BLOCK_SUBDEVICE_ID;
 545
 546
 547     pci_dev->config_header.intr_pin = 1;
 548
 549     pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
 550
 551
 552     blk_state->pci_dev = pci_dev;
 553
 554     /* Block configuration */
 555     blk_state->virtio_cfg.host_features = VIRTIO_SEG_MAX;
 556     blk_state->block_cfg.max_seg = QUEUE_SIZE - 2;
 557
 558     // Virtio Block only uses one queue
 559     blk_state->queue.queue_size = QUEUE_SIZE;
 560
 561     blk_state->virtio_dev = virtio;
 562
 563     blk_reset(blk_state);
 564
 565
 566     return 0;
 567 }
 568
 569
 570 static int connect_fn(struct v3_vm_info * vm,
 571                       void * frontend_data,
 572                       struct v3_dev_blk_ops * ops,
 573                       v3_cfg_tree_t * cfg,
 574                       void * private_data) {
 575
 576     struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
 577
 578     struct virtio_blk_state * blk_state  = (struct virtio_blk_state *)V3_Malloc(sizeof(struct virtio_blk_state));
 579     memset(blk_state, 0, sizeof(struct virtio_blk_state));
 580
 581     register_dev(virtio, blk_state);
 582
 583     blk_state->ops = ops;
 584     blk_state->backend_data = private_data;
 585
 586     blk_state->block_cfg.capacity = ops->get_capacity(private_data) / SECTOR_SIZE;
 587
 588     PrintDebug("Virtio Capacity = %d -- 0x%p\n", (int)(blk_state->block_cfg.capacity),
 589                (void *)(addr_t)(blk_state->block_cfg.capacity));
 590
 591     return 0;
 592 }
 593
 594
 595 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
 596     struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
 597     struct virtio_dev_state * virtio_state = NULL;
 598     char * dev_id = v3_cfg_val(cfg, "ID");
 599
 600     PrintDebug("Initializing VIRTIO Block device\n");
 601
 602     if (pci_bus == NULL) {
 603         PrintError("VirtIO devices require a PCI Bus");
 604         return -1;
 605     }
 606
 607
 608     virtio_state  = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
 609     memset(virtio_state, 0, sizeof(struct virtio_dev_state));
 610
 611     INIT_LIST_HEAD(&(virtio_state->dev_list));
 612     virtio_state->pci_bus = pci_bus;
 613
 614
 615     struct vm_device * dev = v3_allocate_device(dev_id, &dev_ops, virtio_state);
 616     if (v3_attach_device(vm, dev) == -1) {
 617         PrintError("Could not attach device %s\n", dev_id);
 618         return -1;
 619     }
 620
 621     if (v3_dev_add_blk_frontend(vm, dev_id, connect_fn, (void *)virtio_state) == -1) {
 622         PrintError("Could not register %s as block frontend\n", dev_id);
 623         return -1;
 624     }
 625
 626     return 0;
 627 }
 628
 629
 630 device_register("LNX_VIRTIO_BLK", virtio_init)