| 1 | /* |
| 2 | * libqos virtio driver |
| 3 | * |
| 4 | * Copyright (c) 2014 Marc Marí |
| 5 | * |
| 6 | * This work is licensed under the terms of the GNU GPL, version 2 or later. |
| 7 | * See the COPYING file in the top-level directory. |
| 8 | */ |
| 9 | |
| 10 | #include "qemu/osdep.h" |
| 11 | #include "qemu/bswap.h" |
| 12 | #include "../libqtest.h" |
| 13 | #include "virtio.h" |
| 14 | #include "standard-headers/linux/virtio_config.h" |
| 15 | #include "standard-headers/linux/virtio_ring.h" |
| 16 | |
| 17 | /* |
| 18 | * qtest_readX/writeX() functions transfer host endian from/to guest endian. |
| 19 | * This works great for Legacy VIRTIO devices where we need guest endian |
| 20 | * accesses. For VIRTIO 1.0 the vring is little-endian so the automatic guest |
| 21 | * endianness conversion is not wanted. |
| 22 | * |
| 23 | * The following qvirtio_readX/writeX() functions handle Legacy and VIRTIO 1.0 |
| 24 | * accesses seamlessly. |
| 25 | */ |
| 26 | static uint16_t qvirtio_readw(QVirtioDevice *d, QTestState *qts, uint64_t addr) |
| 27 | { |
| 28 | uint16_t val; |
| 29 | |
| 30 | if (d->features & (1ull << VIRTIO_F_VERSION_1)) { |
| 31 | qtest_memread(qts, addr, &val, sizeof(val)); |
| 32 | val = le16_to_cpu(val); |
| 33 | } else { |
| 34 | val = qtest_readw(qts, addr); |
| 35 | } |
| 36 | |
| 37 | return val; |
| 38 | } |
| 39 | |
| 40 | static uint32_t qvirtio_readl(QVirtioDevice *d, QTestState *qts, uint64_t addr) |
| 41 | { |
| 42 | uint32_t val; |
| 43 | |
| 44 | if (d->features & (1ull << VIRTIO_F_VERSION_1)) { |
| 45 | qtest_memread(qts, addr, &val, sizeof(val)); |
| 46 | val = le32_to_cpu(val); |
| 47 | } else { |
| 48 | val = qtest_readl(qts, addr); |
| 49 | } |
| 50 | |
| 51 | return val; |
| 52 | } |
| 53 | |
| 54 | static void qvirtio_writew(QVirtioDevice *d, QTestState *qts, |
| 55 | uint64_t addr, uint16_t val) |
| 56 | { |
| 57 | if (d->features & (1ull << VIRTIO_F_VERSION_1)) { |
| 58 | val = cpu_to_le16(val); |
| 59 | qtest_memwrite(qts, addr, &val, sizeof(val)); |
| 60 | } else { |
| 61 | qtest_writew(qts, addr, val); |
| 62 | } |
| 63 | } |
| 64 | |
| 65 | static void qvirtio_writel(QVirtioDevice *d, QTestState *qts, |
| 66 | uint64_t addr, uint32_t val) |
| 67 | { |
| 68 | if (d->features & (1ull << VIRTIO_F_VERSION_1)) { |
| 69 | val = cpu_to_le32(val); |
| 70 | qtest_memwrite(qts, addr, &val, sizeof(val)); |
| 71 | } else { |
| 72 | qtest_writel(qts, addr, val); |
| 73 | } |
| 74 | } |
| 75 | |
| 76 | static void qvirtio_writeq(QVirtioDevice *d, QTestState *qts, |
| 77 | uint64_t addr, uint64_t val) |
| 78 | { |
| 79 | if (d->features & (1ull << VIRTIO_F_VERSION_1)) { |
| 80 | val = cpu_to_le64(val); |
| 81 | qtest_memwrite(qts, addr, &val, sizeof(val)); |
| 82 | } else { |
| 83 | qtest_writeq(qts, addr, val); |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | uint8_t qvirtio_config_readb(QVirtioDevice *d, uint64_t addr) |
| 88 | { |
| 89 | g_assert_true(d->features_negotiated); |
| 90 | return d->bus->config_readb(d, addr); |
| 91 | } |
| 92 | |
| 93 | uint16_t qvirtio_config_readw(QVirtioDevice *d, uint64_t addr) |
| 94 | { |
| 95 | g_assert_true(d->features_negotiated); |
| 96 | return d->bus->config_readw(d, addr); |
| 97 | } |
| 98 | |
| 99 | uint32_t qvirtio_config_readl(QVirtioDevice *d, uint64_t addr) |
| 100 | { |
| 101 | g_assert_true(d->features_negotiated); |
| 102 | return d->bus->config_readl(d, addr); |
| 103 | } |
| 104 | |
| 105 | uint64_t qvirtio_config_readq(QVirtioDevice *d, uint64_t addr) |
| 106 | { |
| 107 | g_assert_true(d->features_negotiated); |
| 108 | return d->bus->config_readq(d, addr); |
| 109 | } |
| 110 | |
| 111 | uint64_t qvirtio_get_features(QVirtioDevice *d) |
| 112 | { |
| 113 | return d->bus->get_features(d); |
| 114 | } |
| 115 | |
| 116 | void qvirtio_set_features(QVirtioDevice *d, uint64_t features) |
| 117 | { |
| 118 | g_assert(!(features & QVIRTIO_F_BAD_FEATURE)); |
| 119 | |
| 120 | d->features = features; |
| 121 | d->bus->set_features(d, features); |
| 122 | |
| 123 | /* |
| 124 | * This could be a separate function for drivers that want to access |
| 125 | * configuration space before setting FEATURES_OK, but no existing users |
| 126 | * need that and it's less code for callers if this is done implicitly. |
| 127 | */ |
| 128 | if (features & (1ull << VIRTIO_F_VERSION_1)) { |
| 129 | uint8_t status = d->bus->get_status(d) | |
| 130 | VIRTIO_CONFIG_S_FEATURES_OK; |
| 131 | |
| 132 | d->bus->set_status(d, status); |
| 133 | g_assert_cmphex(d->bus->get_status(d), ==, status); |
| 134 | } |
| 135 | |
| 136 | d->features_negotiated = true; |
| 137 | } |
| 138 | |
| 139 | QVirtQueue *qvirtqueue_setup(QVirtioDevice *d, |
| 140 | QGuestAllocator *alloc, uint16_t index) |
| 141 | { |
| 142 | g_assert_true(d->features_negotiated); |
| 143 | return d->bus->virtqueue_setup(d, alloc, index); |
| 144 | } |
| 145 | |
| 146 | void qvirtqueue_cleanup(const QVirtioBus *bus, QVirtQueue *vq, |
| 147 | QGuestAllocator *alloc) |
| 148 | { |
| 149 | return bus->virtqueue_cleanup(vq, alloc); |
| 150 | } |
| 151 | |
| 152 | void qvirtio_reset(QVirtioDevice *d) |
| 153 | { |
| 154 | d->bus->set_status(d, 0); |
| 155 | g_assert_cmphex(d->bus->get_status(d), ==, 0); |
| 156 | d->features_negotiated = false; |
| 157 | } |
| 158 | |
| 159 | void qvirtio_set_acknowledge(QVirtioDevice *d) |
| 160 | { |
| 161 | d->bus->set_status(d, d->bus->get_status(d) | VIRTIO_CONFIG_S_ACKNOWLEDGE); |
| 162 | g_assert_cmphex(d->bus->get_status(d), ==, VIRTIO_CONFIG_S_ACKNOWLEDGE); |
| 163 | } |
| 164 | |
| 165 | void qvirtio_set_driver(QVirtioDevice *d) |
| 166 | { |
| 167 | d->bus->set_status(d, d->bus->get_status(d) | VIRTIO_CONFIG_S_DRIVER); |
| 168 | g_assert_cmphex(d->bus->get_status(d), ==, |
| 169 | VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_ACKNOWLEDGE); |
| 170 | } |
| 171 | |
| 172 | void qvirtio_set_driver_ok(QVirtioDevice *d) |
| 173 | { |
| 174 | d->bus->set_status(d, d->bus->get_status(d) | VIRTIO_CONFIG_S_DRIVER_OK); |
| 175 | g_assert_cmphex(d->bus->get_status(d), ==, VIRTIO_CONFIG_S_DRIVER_OK | |
| 176 | VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_ACKNOWLEDGE | |
| 177 | (d->features & (1ull << VIRTIO_F_VERSION_1) ? |
| 178 | VIRTIO_CONFIG_S_FEATURES_OK : 0)); |
| 179 | } |
| 180 | |
| 181 | void qvirtio_wait_queue_isr(QTestState *qts, QVirtioDevice *d, |
| 182 | QVirtQueue *vq, gint64 timeout_us) |
| 183 | { |
| 184 | gint64 start_time = g_get_monotonic_time(); |
| 185 | |
| 186 | for (;;) { |
| 187 | if (d->bus->get_queue_isr_status(d, vq)) { |
| 188 | return; |
| 189 | } |
| 190 | g_assert(g_get_monotonic_time() - start_time <= timeout_us); |
| 191 | } |
| 192 | } |
| 193 | |
| 194 | /* Wait for the status byte at given guest memory address to be set |
| 195 | * |
| 196 | * The virtqueue interrupt must not be raised, making this useful for testing |
| 197 | * event_index functionality. |
| 198 | */ |
| 199 | uint8_t qvirtio_wait_status_byte_no_isr(QTestState *qts, QVirtioDevice *d, |
| 200 | QVirtQueue *vq, |
| 201 | uint64_t addr, |
| 202 | gint64 timeout_us) |
| 203 | { |
| 204 | gint64 start_time = g_get_monotonic_time(); |
| 205 | uint8_t val; |
| 206 | |
| 207 | while ((val = qtest_readb(qts, addr)) == 0xff) { |
| 208 | g_assert(!d->bus->get_queue_isr_status(d, vq)); |
| 209 | g_assert(g_get_monotonic_time() - start_time <= timeout_us); |
| 210 | } |
| 211 | return val; |
| 212 | } |
| 213 | |
| 214 | /* |
| 215 | * qvirtio_wait_used_elem: |
| 216 | * @desc_idx: The next expected vq->desc[] index in the used ring |
| 217 | * @len: A pointer that is filled with the length written into the buffer, may |
| 218 | * be NULL |
| 219 | * @timeout_us: How many microseconds to wait before failing |
| 220 | * |
| 221 | * This function waits for the next completed request on the used ring. |
| 222 | */ |
| 223 | void qvirtio_wait_used_elem(QTestState *qts, QVirtioDevice *d, |
| 224 | QVirtQueue *vq, |
| 225 | uint32_t desc_idx, |
| 226 | uint32_t *len, |
| 227 | gint64 timeout_us) |
| 228 | { |
| 229 | gint64 start_time = g_get_monotonic_time(); |
| 230 | |
| 231 | for (;;) { |
| 232 | uint32_t got_desc_idx; |
| 233 | |
| 234 | |
| 235 | if (d->bus->get_queue_isr_status(d, vq) && |
| 236 | qvirtqueue_get_buf(qts, vq, &got_desc_idx, len)) { |
| 237 | g_assert_cmpint(got_desc_idx, ==, desc_idx); |
| 238 | return; |
| 239 | } |
| 240 | g_assert(g_get_monotonic_time() - start_time <= timeout_us); |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | void qvirtio_wait_config_isr(QVirtioDevice *d, gint64 timeout_us) |
| 245 | { |
| 246 | d->bus->wait_config_isr_status(d, timeout_us); |
| 247 | } |
| 248 | |
| 249 | void qvring_init(QTestState *qts, const QGuestAllocator *alloc, QVirtQueue *vq, |
| 250 | uint64_t addr) |
| 251 | { |
| 252 | int i; |
| 253 | |
| 254 | vq->desc = addr; |
| 255 | vq->avail = vq->desc + vq->size * sizeof(struct vring_desc); |
| 256 | vq->used = (uint64_t)((vq->avail + sizeof(uint16_t) * (3 + vq->size) |
| 257 | + vq->align - 1) & ~(vq->align - 1)); |
| 258 | |
| 259 | for (i = 0; i < vq->size - 1; i++) { |
| 260 | /* vq->desc[i].addr */ |
| 261 | qvirtio_writeq(vq->vdev, qts, vq->desc + (16 * i), 0); |
| 262 | /* vq->desc[i].next */ |
| 263 | qvirtio_writew(vq->vdev, qts, vq->desc + (16 * i) + 14, i + 1); |
| 264 | } |
| 265 | |
| 266 | /* vq->avail->flags */ |
| 267 | qvirtio_writew(vq->vdev, qts, vq->avail, 0); |
| 268 | |
| 269 | qvirtqueue_set_avail_idx(qts, vq->vdev, vq, 0); |
| 270 | |
| 271 | /* vq->avail->used_event */ |
| 272 | qvirtio_writew(vq->vdev, qts, vq->avail + 4 + (2 * vq->size), 0); |
| 273 | |
| 274 | /* vq->used->flags */ |
| 275 | qvirtio_writew(vq->vdev, qts, vq->used, 0); |
| 276 | /* vq->used->idx */ |
| 277 | qvirtio_writew(vq->vdev, qts, vq->used + 2, 0); |
| 278 | /* vq->used->avail_event */ |
| 279 | qvirtio_writew(vq->vdev, qts, vq->used + 4 + |
| 280 | sizeof(struct vring_used_elem) * vq->size, 0); |
| 281 | } |
| 282 | |
| 283 | QVRingIndirectDesc *qvring_indirect_desc_setup(QTestState *qs, QVirtioDevice *d, |
| 284 | QGuestAllocator *alloc, |
| 285 | uint16_t elem) |
| 286 | { |
| 287 | int i; |
| 288 | QVRingIndirectDesc *indirect = g_malloc(sizeof(*indirect)); |
| 289 | |
| 290 | indirect->index = 0; |
| 291 | indirect->elem = elem; |
| 292 | indirect->desc = guest_alloc(alloc, sizeof(struct vring_desc) * elem); |
| 293 | |
| 294 | for (i = 0; i < elem; ++i) { |
| 295 | /* indirect->desc[i].addr */ |
| 296 | qvirtio_writeq(d, qs, indirect->desc + (16 * i), 0); |
| 297 | |
| 298 | /* |
| 299 | * If it's not the last element of the ring, set |
| 300 | * the chain (VRING_DESC_F_NEXT) flag and |
| 301 | * desc->next. Clear the last element - there's |
| 302 | * no guarantee that guest_alloc() will do it. |
| 303 | */ |
| 304 | if (i != elem - 1) { |
| 305 | /* indirect->desc[i].flags */ |
| 306 | qvirtio_writew(d, qs, indirect->desc + (16 * i) + 12, |
| 307 | VRING_DESC_F_NEXT); |
| 308 | |
| 309 | /* indirect->desc[i].next */ |
| 310 | qvirtio_writew(d, qs, indirect->desc + (16 * i) + 14, i + 1); |
| 311 | } else { |
| 312 | qvirtio_writew(d, qs, indirect->desc + (16 * i) + 12, 0); |
| 313 | qvirtio_writew(d, qs, indirect->desc + (16 * i) + 14, 0); |
| 314 | } |
| 315 | } |
| 316 | |
| 317 | return indirect; |
| 318 | } |
| 319 | |
| 320 | void qvring_indirect_desc_add(QVirtioDevice *d, QTestState *qts, |
| 321 | QVRingIndirectDesc *indirect, |
| 322 | uint64_t data, uint32_t len, bool write) |
| 323 | { |
| 324 | uint16_t flags; |
| 325 | |
| 326 | g_assert_cmpint(indirect->index, <, indirect->elem); |
| 327 | |
| 328 | flags = qvirtio_readw(d, qts, indirect->desc + |
| 329 | (16 * indirect->index) + 12); |
| 330 | |
| 331 | if (write) { |
| 332 | flags |= VRING_DESC_F_WRITE; |
| 333 | } |
| 334 | |
| 335 | /* indirect->desc[indirect->index].addr */ |
| 336 | qvirtio_writeq(d, qts, indirect->desc + (16 * indirect->index), data); |
| 337 | /* indirect->desc[indirect->index].len */ |
| 338 | qvirtio_writel(d, qts, indirect->desc + (16 * indirect->index) + 8, len); |
| 339 | /* indirect->desc[indirect->index].flags */ |
| 340 | qvirtio_writew(d, qts, indirect->desc + (16 * indirect->index) + 12, |
| 341 | flags); |
| 342 | |
| 343 | indirect->index++; |
| 344 | } |
| 345 | |
| 346 | uint32_t qvirtqueue_add(QTestState *qts, QVirtQueue *vq, uint64_t data, |
| 347 | uint32_t len, bool write, bool next) |
| 348 | { |
| 349 | uint16_t flags = 0; |
| 350 | vq->num_free--; |
| 351 | |
| 352 | if (write) { |
| 353 | flags |= VRING_DESC_F_WRITE; |
| 354 | } |
| 355 | |
| 356 | if (next) { |
| 357 | flags |= VRING_DESC_F_NEXT; |
| 358 | } |
| 359 | |
| 360 | /* vq->desc[vq->free_head].addr */ |
| 361 | qvirtio_writeq(vq->vdev, qts, vq->desc + (16 * vq->free_head), data); |
| 362 | /* vq->desc[vq->free_head].len */ |
| 363 | qvirtio_writel(vq->vdev, qts, vq->desc + (16 * vq->free_head) + 8, len); |
| 364 | /* vq->desc[vq->free_head].flags */ |
| 365 | qvirtio_writew(vq->vdev, qts, vq->desc + (16 * vq->free_head) + 12, flags); |
| 366 | |
| 367 | return vq->free_head++; /* Return and increase, in this order */ |
| 368 | } |
| 369 | |
| 370 | uint32_t qvirtqueue_add_indirect(QTestState *qts, QVirtQueue *vq, |
| 371 | QVRingIndirectDesc *indirect) |
| 372 | { |
| 373 | g_assert(vq->indirect); |
| 374 | g_assert_cmpint(vq->size, >=, indirect->elem); |
| 375 | g_assert_cmpint(indirect->index, ==, indirect->elem); |
| 376 | |
| 377 | vq->num_free--; |
| 378 | |
| 379 | /* vq->desc[vq->free_head].addr */ |
| 380 | qvirtio_writeq(vq->vdev, qts, vq->desc + (16 * vq->free_head), |
| 381 | indirect->desc); |
| 382 | /* vq->desc[vq->free_head].len */ |
| 383 | qvirtio_writel(vq->vdev, qts, vq->desc + (16 * vq->free_head) + 8, |
| 384 | sizeof(struct vring_desc) * indirect->elem); |
| 385 | /* vq->desc[vq->free_head].flags */ |
| 386 | qvirtio_writew(vq->vdev, qts, vq->desc + (16 * vq->free_head) + 12, |
| 387 | VRING_DESC_F_INDIRECT); |
| 388 | |
| 389 | return vq->free_head++; /* Return and increase, in this order */ |
| 390 | } |
| 391 | |
| 392 | void qvirtqueue_set_avail_idx(QTestState *qts, QVirtioDevice *d, |
| 393 | QVirtQueue *vq, uint16_t idx) |
| 394 | { |
| 395 | /* vq->avail->idx */ |
| 396 | qvirtio_writew(d, qts, vq->avail + 2, idx); |
| 397 | } |
| 398 | |
| 399 | void qvirtqueue_kick(QTestState *qts, QVirtioDevice *d, QVirtQueue *vq, |
| 400 | uint32_t free_head) |
| 401 | { |
| 402 | /* vq->avail->idx */ |
| 403 | uint16_t idx = qvirtio_readw(d, qts, vq->avail + 2); |
| 404 | /* vq->used->flags */ |
| 405 | uint16_t flags; |
| 406 | /* vq->used->avail_event */ |
| 407 | uint16_t avail_event; |
| 408 | |
| 409 | /* vq->avail->ring[idx % vq->size] */ |
| 410 | qvirtio_writew(d, qts, vq->avail + 4 + (2 * (idx % vq->size)), free_head); |
| 411 | |
| 412 | qvirtqueue_set_avail_idx(qts, d, vq, idx + 1); |
| 413 | |
| 414 | /* Must read after idx is updated */ |
| 415 | flags = qvirtio_readw(d, qts, vq->used); |
| 416 | avail_event = qvirtio_readw(d, qts, vq->used + 4 + |
| 417 | sizeof(struct vring_used_elem) * vq->size); |
| 418 | |
| 419 | /* < 1 because we add elements to avail queue one by one */ |
| 420 | if ((flags & VRING_USED_F_NO_NOTIFY) == 0 && |
| 421 | (!vq->event || (uint16_t)(idx-avail_event) < 1)) { |
| 422 | d->bus->virtqueue_kick(d, vq); |
| 423 | } |
| 424 | } |
| 425 | |
| 426 | /* |
| 427 | * qvirtqueue_get_buf: |
| 428 | * @desc_idx: A pointer that is filled with the vq->desc[] index, may be NULL |
| 429 | * @len: A pointer that is filled with the length written into the buffer, may |
| 430 | * be NULL |
| 431 | * |
| 432 | * This function gets the next used element if there is one ready. |
| 433 | * |
| 434 | * Returns: true if an element was ready, false otherwise |
| 435 | */ |
| 436 | bool qvirtqueue_get_buf(QTestState *qts, QVirtQueue *vq, uint32_t *desc_idx, |
| 437 | uint32_t *len) |
| 438 | { |
| 439 | uint16_t idx; |
| 440 | uint64_t elem_addr, addr; |
| 441 | |
| 442 | idx = qvirtio_readw(vq->vdev, qts, |
| 443 | vq->used + offsetof(struct vring_used, idx)); |
| 444 | if (idx == vq->last_used_idx) { |
| 445 | return false; |
| 446 | } |
| 447 | |
| 448 | elem_addr = vq->used + |
| 449 | offsetof(struct vring_used, ring) + |
| 450 | (vq->last_used_idx % vq->size) * |
| 451 | sizeof(struct vring_used_elem); |
| 452 | |
| 453 | if (desc_idx) { |
| 454 | addr = elem_addr + offsetof(struct vring_used_elem, id); |
| 455 | *desc_idx = qvirtio_readl(vq->vdev, qts, addr); |
| 456 | } |
| 457 | |
| 458 | if (len) { |
| 459 | addr = elem_addr + offsetof(struct vring_used_elem, len); |
| 460 | *len = qvirtio_readw(vq->vdev, qts, addr); |
| 461 | } |
| 462 | |
| 463 | vq->last_used_idx++; |
| 464 | return true; |
| 465 | } |
| 466 | |
| 467 | /* |
| 468 | * qvirtqueue_reset_pool: |
| 469 | * @vq: The virtqueue to reset |
| 470 | * |
| 471 | * Reset the descriptor pool state without reinitializing the device. |
| 472 | * This is useful for tests that issue a high number of requests and |
| 473 | * are limited by the simplified virtio test driver's descriptor tracking, |
| 474 | * which decrements num_free but never increments it back. |
| 475 | * |
| 476 | * This is only safe for synchronous test code where requests are |
| 477 | * sent and completed before the next request is issued. Do not use |
| 478 | * with asynchronous code where multiple requests may be in-flight. |
| 479 | * |
| 480 | * Note: This only resets the available descriptor pool (free_head, |
| 481 | * num_free). The used ring position (last_used_idx) is NOT reset |
| 482 | * and should continue to track consumed responses across iterations. |
| 483 | */ |
| 484 | void qvirtqueue_reset_pool(QVirtQueue *vq) |
| 485 | { |
| 486 | vq->free_head = 0; |
| 487 | vq->num_free = vq->size; |
| 488 | } |
| 489 | |
| 490 | void qvirtqueue_set_used_event(QTestState *qts, QVirtQueue *vq, uint16_t idx) |
| 491 | { |
| 492 | g_assert(vq->event); |
| 493 | |
| 494 | /* vq->avail->used_event */ |
| 495 | qvirtio_writew(vq->vdev, qts, vq->avail + 4 + (2 * vq->size), idx); |
| 496 | } |
| 497 | |
| 498 | void qvirtio_start_device(QVirtioDevice *vdev) |
| 499 | { |
| 500 | qvirtio_reset(vdev); |
| 501 | qvirtio_set_acknowledge(vdev); |
| 502 | qvirtio_set_driver(vdev); |
| 503 | } |
| 504 | |
| 505 | bool qvirtio_is_big_endian(QVirtioDevice *d) |
| 506 | { |
| 507 | return d->big_endian; |
| 508 | } |