palacios/src/devices/lnx_virtio_vnet.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 <palacios/vm_guest_mem.h>
  23 #include <devices/lnx_virtio_pci.h>
  24
  25 #include <devices/pci.h>
  26
  27
  28 #define QUEUE_SIZE 128
  29 #define NUM_QUEUES 3
  30
  31 struct vnet_config {
  32     uint32_t num_links;
  33     uint32_t num_routes;
  34 } __attribute__((packed));
  35
  36
  37
  38 struct virtio_vnet_state {
  39     struct vnet_config vnet_cfg;
  40     struct virtio_config virtio_cfg;
  41
  42
  43     struct vm_device * pci_bus;
  44     struct pci_device * pci_dev;
  45
  46
  47 #define CTRL_QUEUE 0
  48 #define RECV_QUEUE 1
  49 #define XMIT_QUEUE 2
  50     struct virtio_queue queue[NUM_QUEUES];
  51
  52     struct virtio_queue * cur_queue;
  53
  54
  55     int io_range_size;
  56 };
  57
  58 #define VNET_GET_ROUTES 10
  59 #define VNET_ADD_ROUTE 11
  60 #define VNET_DEL_ROUTE 12
  61
  62 #define VNET_GET_LINKS 20
  63 #define VNET_ADD_LINK 21
  64 #define VNET_DEL_LINK 22
  65
  66 // structure of the symmod notifier ring structures
  67 struct vnet_ctrl_hdr {
  68     uint8_t cmd_type;
  69     uint32_t num_cmds;
  70 } __attribute__((packed));
  71
  72
  73 static int virtio_reset(struct virtio_vnet_state * virtio) {
  74
  75     memset(virtio->queue, 0, sizeof(struct virtio_queue) * 2);
  76
  77     virtio->cur_queue = &(virtio->queue[0]);
  78
  79     virtio->virtio_cfg.status = 0;
  80     virtio->virtio_cfg.pci_isr = 0;
  81
  82     virtio->queue[0].queue_size = QUEUE_SIZE;
  83
  84     memset(&(virtio->vnet_cfg), 0, sizeof(struct vnet_config));
  85
  86     return 0;
  87 }
  88
  89
  90
  91 static int get_desc_count(struct virtio_queue * q, int index) {
  92     struct vring_desc * tmp_desc = &(q->desc[index]);
  93     int cnt = 1;
  94
  95     while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
  96         tmp_desc = &(q->desc[tmp_desc->next]);
  97         cnt++;
  98     }
  99
 100     return cnt;
 101 }
 102
 103
 104
 105
 106 static int handle_cmd_kick(struct guest_info * core, struct virtio_vnet_state * vnet_state) {
 107     struct virtio_queue * q = vnet_state->cur_queue;
 108
 109     PrintDebug("VirtioVNET: Virtio Kick on command  queue\n");
 110
 111     while (q->cur_avail_idx < q->avail->index) {
 112         struct vring_desc * hdr_desc = NULL;
 113         struct vring_desc * buf_desc = NULL;
 114         struct vring_desc * status_desc = NULL;
 115         uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
 116         uint16_t desc_cnt = get_desc_count(q, desc_idx);
 117         struct vnet_ctrl_hdr * hdr = NULL;
 118         int i;
 119         int xfer_len = 0;
 120         uint8_t * status_ptr = NULL;
 121         uint8_t status = 0;
 122
 123
 124         PrintDebug("Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE);
 125
 126         if (desc_cnt < 3) {
 127             PrintError("Symmod loads must include at least 3 descriptors (cnt=%d)\n", desc_cnt);
 128             return -1;
 129         }
 130
 131         hdr_desc = &(q->desc[desc_idx]);
 132
 133         if (guest_pa_to_host_va(core, hdr_desc->addr_gpa, (addr_t *)&hdr) == -1) {
 134             PrintError("Could not translate VirtioVNET header address\n");
 135             return -1;
 136         }
 137
 138         desc_idx = hdr_desc->next;
 139
 140         if (hdr->cmd_type == VNET_ADD_ROUTE) {
 141
 142             for (i = 0; i < hdr->num_cmds; i++) {
 143                 uint8_t tmp_status = 0;
 144                 uint8_t * buf = NULL;
 145
 146                 buf_desc = &(q->desc[desc_idx]);
 147
 148                 if (guest_pa_to_host_va(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
 149                     PrintError("Could not translate buffer address\n");
 150                     return -1;
 151                 }
 152
 153                 // add route
 154                 PrintDebug("Adding VNET Route\n");
 155
 156                 if (tmp_status != 0) {
 157                     PrintError("Error loading module segment\n");
 158                     status = tmp_status;
 159                 }
 160
 161                 xfer_len += buf_desc->length;
 162                 desc_idx = buf_desc->next;
 163             }
 164
 165         } else if (hdr->cmd_type == VNET_ADD_LINK) {
 166
 167
 168         }
 169
 170
 171
 172         status_desc = &(q->desc[desc_idx]);
 173
 174         if (guest_pa_to_host_va(core, status_desc->addr_gpa, (addr_t *)&status_ptr) == -1) {
 175             PrintError("SYMMOD Error could not translate status address\n");
 176             return -1;
 177         }
 178
 179         xfer_len += status_desc->length;
 180         *status_ptr = status;
 181
 182         PrintDebug("Transferred %d bytes (xfer_len)\n", xfer_len);
 183         q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
 184         q->used->ring[q->used->index % QUEUE_SIZE].length = xfer_len; // set to total inbound xfer length
 185
 186         q->used->index++;
 187         q->cur_avail_idx++;
 188     }
 189
 190
 191     if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
 192         PrintDebug("Raising IRQ %d\n",  vnet_state->pci_dev->config_header.intr_line);
 193         v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
 194         vnet_state->virtio_cfg.pci_isr = 1;
 195     }
 196
 197
 198     return 0;
 199 }
 200
 201
 202 static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, uint_t length, void * private_data) {
 203     struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
 204     int port_idx = port % vnet_state->io_range_size;
 205
 206
 207     PrintDebug("VirtioVNET: VIRTIO VNET Write for port %d len=%d, value=%x\n",
 208                port, length, *(uint32_t *)src);
 209     PrintDebug("VirtioVNET: port idx=%d\n", port_idx);
 210
 211
 212     switch (port_idx) {
 213         case GUEST_FEATURES_PORT:
 214             if (length != 4) {
 215                 PrintError("Illegal write length for guest features\n");
 216                 return -1;
 217             }
 218
 219             vnet_state->virtio_cfg.guest_features = *(uint32_t *)src;
 220
 221             break;
 222         case VRING_PG_NUM_PORT:
 223             if (length == 4) {
 224                 addr_t pfn = *(uint32_t *)src;
 225                 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
 226
 227                 vnet_state->cur_queue->pfn = pfn;
 228
 229                 vnet_state->cur_queue->ring_desc_addr = page_addr ;
 230                 vnet_state->cur_queue->ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
 231                 vnet_state->cur_queue->ring_used_addr = ( vnet_state->cur_queue->ring_avail_addr + \
 232                                                  sizeof(struct vring_avail)    + \
 233                                                  (QUEUE_SIZE * sizeof(uint16_t)));
 234
 235                 // round up to next page boundary.
 236                 vnet_state->cur_queue->ring_used_addr = (vnet_state->cur_queue->ring_used_addr + 0xfff) & ~0xfff;
 237
 238                 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_desc_addr, (addr_t *)&(vnet_state->cur_queue->desc)) == -1) {
 239                     PrintError("Could not translate ring descriptor address\n");
 240                     return -1;
 241                 }
 242
 243
 244                 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_avail_addr, (addr_t *)&(vnet_state->cur_queue->avail)) == -1) {
 245                     PrintError("Could not translate ring available address\n");
 246                     return -1;
 247                 }
 248
 249
 250                 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_used_addr, (addr_t *)&(vnet_state->cur_queue->used)) == -1) {
 251                     PrintError("Could not translate ring used address\n");
 252                     return -1;
 253                 }
 254
 255                 PrintDebug("VirtioVNET: RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
 256                            (void *)(vnet_state->cur_queue->ring_desc_addr),
 257                            (void *)(vnet_state->cur_queue->ring_avail_addr),
 258                            (void *)(vnet_state->cur_queue->ring_used_addr));
 259
 260                 PrintDebug("VirtioVNET: RingDesc=%p, Avail=%p, Used=%p\n",
 261                            vnet_state->cur_queue->desc, vnet_state->cur_queue->avail, vnet_state->cur_queue->used);
 262
 263             } else {
 264                 PrintError("Illegal write length for page frame number\n");
 265                 return -1;
 266             }
 267             break;
 268         case VRING_Q_SEL_PORT:
 269             vnet_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
 270
 271             if (vnet_state->virtio_cfg.vring_queue_selector > NUM_QUEUES) {
 272                 PrintError("Virtio Symbiotic device has no qeueues. Selected %d\n",
 273                            vnet_state->virtio_cfg.vring_queue_selector);
 274                 return -1;
 275             }
 276
 277             vnet_state->cur_queue = &(vnet_state->queue[vnet_state->virtio_cfg.vring_queue_selector]);
 278
 279             break;
 280         case VRING_Q_NOTIFY_PORT: {
 281             uint16_t queue_idx = *(uint16_t *)src;
 282
 283             PrintDebug("VirtioVNET: Handling Kick\n");
 284
 285             if (queue_idx == 0) {
 286                 if (handle_cmd_kick(core, vnet_state) == -1) {
 287                     PrintError("Could not handle VNET Control command\n");
 288                     return -1;
 289                 }
 290             } else {
 291                 PrintError("Kick on invalid queue (%d)\n", queue_idx);
 292                 return -1;
 293             }
 294
 295             break;
 296         }
 297         case VIRTIO_STATUS_PORT:
 298             vnet_state->virtio_cfg.status = *(uint8_t *)src;
 299
 300             if (vnet_state->virtio_cfg.status == 0) {
 301                 PrintDebug("VirtioVNET: Resetting device\n");
 302                 virtio_reset(vnet_state);
 303             }
 304
 305             break;
 306
 307         case VIRTIO_ISR_PORT:
 308             vnet_state->virtio_cfg.pci_isr = *(uint8_t *)src;
 309             break;
 310         default:
 311             return -1;
 312             break;
 313     }
 314
 315     return length;
 316 }
 317
 318
 319 static int virtio_io_read(struct guest_info * core, uint16_t port, void * dst, uint_t length, void * private_data) {
 320
 321     struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
 322     int port_idx = port % vnet_state->io_range_size;
 323
 324 /*
 325     PrintDebug("VirtioVNET: VIRTIO SYMBIOTIC Read  for port %d (index =%d), length=%d\n",
 326                port, port_idx, length);
 327 */
 328     switch (port_idx) {
 329         case HOST_FEATURES_PORT:
 330             if (length != 4) {
 331                 PrintError("Illegal read length for host features\n");
 332                 return -1;
 333             }
 334
 335             *(uint32_t *)dst = vnet_state->virtio_cfg.host_features;
 336
 337             break;
 338         case VRING_PG_NUM_PORT:
 339             if (length != 4) {
 340                 PrintError("Illegal read length for page frame number\n");
 341                 return -1;
 342             }
 343
 344             *(uint32_t *)dst = vnet_state->cur_queue->pfn;
 345
 346             break;
 347         case VRING_SIZE_PORT:
 348             if (length != 2) {
 349                 PrintError("Illegal read length for vring size\n");
 350                 return -1;
 351             }
 352
 353             *(uint16_t *)dst = vnet_state->cur_queue->queue_size;
 354
 355             break;
 356
 357         case VIRTIO_STATUS_PORT:
 358             if (length != 1) {
 359                 PrintError("Illegal read length for status\n");
 360                 return -1;
 361             }
 362
 363             *(uint8_t *)dst = vnet_state->virtio_cfg.status;
 364             break;
 365
 366         case VIRTIO_ISR_PORT:
 367             *(uint8_t *)dst = vnet_state->virtio_cfg.pci_isr;
 368             vnet_state->virtio_cfg.pci_isr = 0;
 369             v3_pci_lower_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
 370             break;
 371
 372         default:
 373             if ( (port_idx >= sizeof(struct virtio_config)) &&
 374                  (port_idx < (sizeof(struct virtio_config) + sizeof(struct vnet_config))) ) {
 375                 int cfg_offset = port_idx - sizeof(struct virtio_config);
 376                 uint8_t * cfg_ptr = (uint8_t *)&(vnet_state->vnet_cfg);
 377
 378                 memcpy(dst, cfg_ptr + cfg_offset, length);
 379
 380             } else {
 381                 PrintError("Read of Unhandled Virtio Read\n");
 382                 return -1;
 383             }
 384
 385             break;
 386     }
 387
 388     return length;
 389 }
 390
 391
 392
 393 static struct v3_device_ops dev_ops = {
 394     .free = NULL,
 395     .reset = NULL,
 396     .start = NULL,
 397     .stop = NULL,
 398 };
 399
 400
 401 static int vnet_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
 402     struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
 403     struct virtio_vnet_state * virtio_state = NULL;
 404     struct pci_device * pci_dev = NULL;
 405     char * name = v3_cfg_val(cfg, "name");
 406
 407     PrintDebug("VirtioVNET: Initializing VIRTIO VNET Control device\n");
 408
 409     if (pci_bus == NULL) {
 410         PrintError("VirtIO devices require a PCI Bus");
 411         return -1;
 412     }
 413
 414     virtio_state  = (struct virtio_vnet_state *)V3_Malloc(sizeof(struct virtio_vnet_state));
 415     memset(virtio_state, 0, sizeof(struct virtio_vnet_state));
 416
 417     struct vm_device * dev = v3_allocate_device(name, &dev_ops, virtio_state);
 418
 419     if (v3_attach_device(vm, dev) == -1) {
 420         PrintError("Could not attach device %s\n", name);
 421         return -1;
 422     }
 423
 424
 425     // PCI initialization
 426     {
 427         struct v3_pci_bar bars[6];
 428         int num_ports = sizeof(struct virtio_config) + sizeof(struct vnet_config);
 429         int tmp_ports = num_ports;
 430         int i;
 431
 432
 433         // This gets the number of ports, rounded up to a power of 2
 434         virtio_state->io_range_size = 1; // must be a power of 2
 435
 436         while (tmp_ports > 0) {
 437             tmp_ports >>= 1;
 438             virtio_state->io_range_size <<= 1;
 439         }
 440
 441         // this is to account for any low order bits being set in num_ports
 442         // if there are none, then num_ports was already a power of 2 so we shift right to reset it
 443         if ((num_ports & ((virtio_state->io_range_size >> 1) - 1)) == 0) {
 444             virtio_state->io_range_size >>= 1;
 445         }
 446
 447
 448         for (i = 0; i < 6; i++) {
 449             bars[i].type = PCI_BAR_NONE;
 450         }
 451
 452         bars[0].type = PCI_BAR_IO;
 453         bars[0].default_base_port = -1;
 454         bars[0].num_ports = virtio_state->io_range_size;
 455
 456         bars[0].io_read = virtio_io_read;
 457         bars[0].io_write = virtio_io_write;
 458         bars[0].private_data = virtio_state;
 459
 460         pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
 461                                          0, PCI_AUTO_DEV_NUM, 0,
 462                                          "LNX_VIRTIO_VNET", bars,
 463                                          NULL, NULL, NULL, virtio_state);
 464
 465         if (!pci_dev) {
 466             PrintError("Could not register PCI Device\n");
 467             return -1;
 468         }
 469
 470         pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
 471         pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
 472
 473
 474         pci_dev->config_header.device_id = VIRTIO_VNET_DEV_ID;
 475         pci_dev->config_header.class = PCI_CLASS_MEMORY;
 476         pci_dev->config_header.subclass = PCI_MEM_SUBCLASS_RAM;
 477
 478         pci_dev->config_header.subsystem_id = VIRTIO_VNET_SUBDEVICE_ID;
 479
 480
 481         pci_dev->config_header.intr_pin = 1;
 482
 483         pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
 484
 485
 486         virtio_state->pci_dev = pci_dev;
 487         virtio_state->pci_bus = pci_bus;
 488     }
 489
 490     virtio_reset(virtio_state);
 491
 492
 493     return 0;
 494 }
 495
 496
 497 device_register("LNX_VIRTIO_VNET", vnet_init)