virtio.c revision 1.64
1 /* $NetBSD: virtio.c,v 1.64 2022/12/30 21:38:13 jakllsch Exp $ */ 2 3 /* 4 * Copyright (c) 2020 The NetBSD Foundation, Inc. 5 * Copyright (c) 2012 Stefan Fritsch, Alexander Fiveg. 6 * Copyright (c) 2010 Minoura Makoto. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __KERNEL_RCSID(0, "$NetBSD: virtio.c,v 1.64 2022/12/30 21:38:13 jakllsch Exp $"); 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/kernel.h> 36 #include <sys/atomic.h> 37 #include <sys/bus.h> 38 #include <sys/device.h> 39 #include <sys/kmem.h> 40 #include <sys/module.h> 41 42 #define VIRTIO_PRIVATE 43 44 #include <dev/pci/virtioreg.h> /* XXX: move to non-pci */ 45 #include <dev/pci/virtiovar.h> /* XXX: move to non-pci */ 46 47 #define MINSEG_INDIRECT 2 /* use indirect if nsegs >= this value */ 48 49 /* incomplete list */ 50 static const char *virtio_device_name[] = { 51 "unknown (0)", /* 0 */ 52 "network", /* 1 */ 53 "block", /* 2 */ 54 "console", /* 3 */ 55 "entropy", /* 4 */ 56 "memory balloon", /* 5 */ 57 "I/O memory", /* 6 */ 58 "remote processor messaging", /* 7 */ 59 "SCSI", /* 8 */ 60 "9P transport", /* 9 */ 61 }; 62 #define NDEVNAMES __arraycount(virtio_device_name) 63 64 static void virtio_init_vq(struct virtio_softc *, 65 struct virtqueue *, const bool); 66 67 void 68 virtio_set_status(struct virtio_softc *sc, int status) 69 { 70 sc->sc_ops->set_status(sc, status); 71 } 72 73 /* 74 * Reset the device. 75 */ 76 /* 77 * To reset the device to a known state, do following: 78 * virtio_reset(sc); // this will stop the device activity 79 * <dequeue finished requests>; // virtio_dequeue() still can be called 80 * <revoke pending requests in the vqs if any>; 81 * virtio_reinit_start(sc); // dequeue prohibitted 82 * newfeatures = virtio_negotiate_features(sc, requestedfeatures); 83 * <some other initialization>; 84 * virtio_reinit_end(sc); // device activated; enqueue allowed 85 * Once attached, feature negotiation can only be allowed after virtio_reset. 86 */ 87 void 88 virtio_reset(struct virtio_softc *sc) 89 { 90 virtio_device_reset(sc); 91 } 92 93 int 94 virtio_reinit_start(struct virtio_softc *sc) 95 { 96 int i, r; 97 98 virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_ACK); 99 virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER); 100 for (i = 0; i < sc->sc_nvqs; i++) { 101 int n; 102 struct virtqueue *vq = &sc->sc_vqs[i]; 103 n = sc->sc_ops->read_queue_size(sc, vq->vq_index); 104 if (n == 0) /* vq disappeared */ 105 continue; 106 if (n != vq->vq_num) { 107 panic("%s: virtqueue size changed, vq index %d\n", 108 device_xname(sc->sc_dev), 109 vq->vq_index); 110 } 111 virtio_init_vq(sc, vq, true); 112 sc->sc_ops->setup_queue(sc, vq->vq_index, 113 vq->vq_dmamap->dm_segs[0].ds_addr); 114 } 115 116 r = sc->sc_ops->setup_interrupts(sc, 1); 117 if (r != 0) 118 goto fail; 119 120 return 0; 121 122 fail: 123 virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED); 124 125 return 1; 126 } 127 128 void 129 virtio_reinit_end(struct virtio_softc *sc) 130 { 131 virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK); 132 } 133 134 /* 135 * Feature negotiation. 136 */ 137 void 138 virtio_negotiate_features(struct virtio_softc *sc, uint64_t guest_features) 139 { 140 if (!(device_cfdata(sc->sc_dev)->cf_flags & 1) && 141 !(device_cfdata(sc->sc_child)->cf_flags & 1)) /* XXX */ 142 guest_features |= VIRTIO_F_RING_INDIRECT_DESC; 143 sc->sc_ops->neg_features(sc, guest_features); 144 if (sc->sc_active_features & VIRTIO_F_RING_INDIRECT_DESC) 145 sc->sc_indirect = true; 146 else 147 sc->sc_indirect = false; 148 } 149 150 151 /* 152 * Device configuration registers readers/writers 153 */ 154 #if 0 155 #define DPRINTFR(n, fmt, val, index, num) \ 156 printf("\n%s (", n); \ 157 for (int i = 0; i < num; i++) \ 158 printf("%02x ", bus_space_read_1(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index+i)); \ 159 printf(") -> "); printf(fmt, val); printf("\n"); 160 #define DPRINTFR2(n, fmt, val_s, val_n) \ 161 printf("%s ", n); \ 162 printf("\n stream "); printf(fmt, val_s); printf(" norm "); printf(fmt, val_n); printf("\n"); 163 #else 164 #define DPRINTFR(n, fmt, val, index, num) 165 #define DPRINTFR2(n, fmt, val_s, val_n) 166 #endif 167 168 169 uint8_t 170 virtio_read_device_config_1(struct virtio_softc *sc, int index) 171 { 172 bus_space_tag_t iot = sc->sc_devcfg_iot; 173 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 174 uint8_t val; 175 176 val = bus_space_read_1(iot, ioh, index); 177 178 DPRINTFR("read_1", "%02x", val, index, 1); 179 return val; 180 } 181 182 uint16_t 183 virtio_read_device_config_2(struct virtio_softc *sc, int index) 184 { 185 bus_space_tag_t iot = sc->sc_devcfg_iot; 186 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 187 uint16_t val; 188 189 val = bus_space_read_2(iot, ioh, index); 190 if (BYTE_ORDER != sc->sc_bus_endian) 191 val = bswap16(val); 192 193 DPRINTFR("read_2", "%04x", val, index, 2); 194 DPRINTFR2("read_2", "%04x", 195 bus_space_read_stream_2(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 196 index), 197 bus_space_read_2(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index)); 198 return val; 199 } 200 201 uint32_t 202 virtio_read_device_config_4(struct virtio_softc *sc, int index) 203 { 204 bus_space_tag_t iot = sc->sc_devcfg_iot; 205 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 206 uint32_t val; 207 208 val = bus_space_read_4(iot, ioh, index); 209 if (BYTE_ORDER != sc->sc_bus_endian) 210 val = bswap32(val); 211 212 DPRINTFR("read_4", "%08x", val, index, 4); 213 DPRINTFR2("read_4", "%08x", 214 bus_space_read_stream_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 215 index), 216 bus_space_read_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index)); 217 return val; 218 } 219 220 /* 221 * The Virtio spec explicitly tells that reading and writing 8 bytes are not 222 * considered atomic and no triggers may be connected to reading or writing 223 * it. We access it using two 32 reads. See virtio spec 4.1.3.1. 224 */ 225 uint64_t 226 virtio_read_device_config_8(struct virtio_softc *sc, int index) 227 { 228 bus_space_tag_t iot = sc->sc_devcfg_iot; 229 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 230 union { 231 uint64_t u64; 232 uint32_t l[2]; 233 } v; 234 uint64_t val; 235 236 v.l[0] = bus_space_read_4(iot, ioh, index); 237 v.l[1] = bus_space_read_4(iot, ioh, index + 4); 238 if (sc->sc_bus_endian != sc->sc_struct_endian) { 239 v.l[0] = bswap32(v.l[0]); 240 v.l[1] = bswap32(v.l[1]); 241 } 242 val = v.u64; 243 244 if (BYTE_ORDER != sc->sc_struct_endian) 245 val = bswap64(val); 246 247 DPRINTFR("read_8", "%08"PRIx64, val, index, 8); 248 DPRINTFR2("read_8 low ", "%08x", 249 bus_space_read_stream_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 250 index), 251 bus_space_read_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index)); 252 DPRINTFR2("read_8 high ", "%08x", 253 bus_space_read_stream_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 254 index + 4), 255 bus_space_read_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, index + 4)); 256 return val; 257 } 258 259 /* 260 * In the older virtio spec, device config registers are host endian. On newer 261 * they are little endian. Some newer devices however explicitly specify their 262 * register to always be little endian. These functions cater for these. 263 */ 264 uint16_t 265 virtio_read_device_config_le_2(struct virtio_softc *sc, int index) 266 { 267 bus_space_tag_t iot = sc->sc_devcfg_iot; 268 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 269 uint16_t val; 270 271 val = bus_space_read_2(iot, ioh, index); 272 if (sc->sc_bus_endian != LITTLE_ENDIAN) 273 val = bswap16(val); 274 275 DPRINTFR("read_le_2", "%04x", val, index, 2); 276 DPRINTFR2("read_le_2", "%04x", 277 bus_space_read_stream_2(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 0), 278 bus_space_read_2(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 0)); 279 return val; 280 } 281 282 uint32_t 283 virtio_read_device_config_le_4(struct virtio_softc *sc, int index) 284 { 285 bus_space_tag_t iot = sc->sc_devcfg_iot; 286 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 287 uint32_t val; 288 289 val = bus_space_read_4(iot, ioh, index); 290 if (sc->sc_bus_endian != LITTLE_ENDIAN) 291 val = bswap32(val); 292 293 DPRINTFR("read_le_4", "%08x", val, index, 4); 294 DPRINTFR2("read_le_4", "%08x", 295 bus_space_read_stream_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 0), 296 bus_space_read_4(sc->sc_devcfg_iot, sc->sc_devcfg_ioh, 0)); 297 return val; 298 } 299 300 void 301 virtio_write_device_config_1(struct virtio_softc *sc, int index, uint8_t value) 302 { 303 bus_space_tag_t iot = sc->sc_devcfg_iot; 304 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 305 306 bus_space_write_1(iot, ioh, index, value); 307 } 308 309 void 310 virtio_write_device_config_2(struct virtio_softc *sc, int index, 311 uint16_t value) 312 { 313 bus_space_tag_t iot = sc->sc_devcfg_iot; 314 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 315 316 if (BYTE_ORDER != sc->sc_bus_endian) 317 value = bswap16(value); 318 bus_space_write_2(iot, ioh, index, value); 319 } 320 321 void 322 virtio_write_device_config_4(struct virtio_softc *sc, int index, 323 uint32_t value) 324 { 325 bus_space_tag_t iot = sc->sc_devcfg_iot; 326 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 327 328 if (BYTE_ORDER != sc->sc_bus_endian) 329 value = bswap32(value); 330 bus_space_write_4(iot, ioh, index, value); 331 } 332 333 /* 334 * The Virtio spec explicitly tells that reading and writing 8 bytes are not 335 * considered atomic and no triggers may be connected to reading or writing 336 * it. We access it using two 32 bit writes. For good measure it is stated to 337 * always write lsb first just in case of a hypervisor bug. See See virtio 338 * spec 4.1.3.1. 339 */ 340 void 341 virtio_write_device_config_8(struct virtio_softc *sc, int index, 342 uint64_t value) 343 { 344 bus_space_tag_t iot = sc->sc_devcfg_iot; 345 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 346 union { 347 uint64_t u64; 348 uint32_t l[2]; 349 } v; 350 351 if (BYTE_ORDER != sc->sc_struct_endian) 352 value = bswap64(value); 353 354 v.u64 = value; 355 if (sc->sc_bus_endian != sc->sc_struct_endian) { 356 v.l[0] = bswap32(v.l[0]); 357 v.l[1] = bswap32(v.l[1]); 358 } 359 360 if (sc->sc_struct_endian == LITTLE_ENDIAN) { 361 bus_space_write_4(iot, ioh, index, v.l[0]); 362 bus_space_write_4(iot, ioh, index + 4, v.l[1]); 363 } else { 364 bus_space_write_4(iot, ioh, index + 4, v.l[1]); 365 bus_space_write_4(iot, ioh, index, v.l[0]); 366 } 367 } 368 369 /* 370 * In the older virtio spec, device config registers are host endian. On newer 371 * they are little endian. Some newer devices however explicitly specify their 372 * register to always be little endian. These functions cater for these. 373 */ 374 void 375 virtio_write_device_config_le_2(struct virtio_softc *sc, int index, 376 uint16_t value) 377 { 378 bus_space_tag_t iot = sc->sc_devcfg_iot; 379 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 380 381 if (sc->sc_bus_endian != LITTLE_ENDIAN) 382 value = bswap16(value); 383 bus_space_write_2(iot, ioh, index, value); 384 } 385 386 void 387 virtio_write_device_config_le_4(struct virtio_softc *sc, int index, 388 uint32_t value) 389 { 390 bus_space_tag_t iot = sc->sc_devcfg_iot; 391 bus_space_handle_t ioh = sc->sc_devcfg_ioh; 392 393 if (sc->sc_bus_endian != LITTLE_ENDIAN) 394 value = bswap32(value); 395 bus_space_write_4(iot, ioh, index, value); 396 } 397 398 399 /* 400 * data structures endian helpers 401 */ 402 uint16_t 403 virtio_rw16(struct virtio_softc *sc, uint16_t val) 404 { 405 KASSERT(sc); 406 return BYTE_ORDER != sc->sc_struct_endian ? bswap16(val) : val; 407 } 408 409 uint32_t 410 virtio_rw32(struct virtio_softc *sc, uint32_t val) 411 { 412 KASSERT(sc); 413 return BYTE_ORDER != sc->sc_struct_endian ? bswap32(val) : val; 414 } 415 416 uint64_t 417 virtio_rw64(struct virtio_softc *sc, uint64_t val) 418 { 419 KASSERT(sc); 420 return BYTE_ORDER != sc->sc_struct_endian ? bswap64(val) : val; 421 } 422 423 424 /* 425 * Interrupt handler. 426 */ 427 static void 428 virtio_soft_intr(void *arg) 429 { 430 struct virtio_softc *sc = arg; 431 432 KASSERT(sc->sc_intrhand != NULL); 433 434 (*sc->sc_intrhand)(sc); 435 } 436 437 /* 438 * dmamap sync operations for a virtqueue. 439 */ 440 static inline void 441 vq_sync_descs(struct virtio_softc *sc, struct virtqueue *vq, int ops) 442 { 443 444 /* availoffset == sizeof(vring_desc) * vq_num */ 445 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 0, vq->vq_availoffset, 446 ops); 447 } 448 449 static inline void 450 vq_sync_aring_all(struct virtio_softc *sc, struct virtqueue *vq, int ops) 451 { 452 uint16_t hdrlen = offsetof(struct vring_avail, ring); 453 size_t payloadlen = vq->vq_num * sizeof(uint16_t); 454 size_t usedlen = 0; 455 456 if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) 457 usedlen = sizeof(uint16_t); 458 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 459 vq->vq_availoffset, hdrlen + payloadlen + usedlen, ops); 460 } 461 462 static inline void 463 vq_sync_aring_header(struct virtio_softc *sc, struct virtqueue *vq, int ops) 464 { 465 uint16_t hdrlen = offsetof(struct vring_avail, ring); 466 467 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 468 vq->vq_availoffset, hdrlen, ops); 469 } 470 471 static inline void 472 vq_sync_aring_payload(struct virtio_softc *sc, struct virtqueue *vq, int ops) 473 { 474 uint16_t hdrlen = offsetof(struct vring_avail, ring); 475 size_t payloadlen = vq->vq_num * sizeof(uint16_t); 476 477 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 478 vq->vq_availoffset + hdrlen, payloadlen, ops); 479 } 480 481 static inline void 482 vq_sync_aring_used(struct virtio_softc *sc, struct virtqueue *vq, int ops) 483 { 484 uint16_t hdrlen = offsetof(struct vring_avail, ring); 485 size_t payloadlen = vq->vq_num * sizeof(uint16_t); 486 size_t usedlen = sizeof(uint16_t); 487 488 if ((sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) == 0) 489 return; 490 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 491 vq->vq_availoffset + hdrlen + payloadlen, usedlen, ops); 492 } 493 494 static inline void 495 vq_sync_uring_all(struct virtio_softc *sc, struct virtqueue *vq, int ops) 496 { 497 uint16_t hdrlen = offsetof(struct vring_used, ring); 498 size_t payloadlen = vq->vq_num * sizeof(struct vring_used_elem); 499 size_t availlen = 0; 500 501 if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) 502 availlen = sizeof(uint16_t); 503 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 504 vq->vq_usedoffset, hdrlen + payloadlen + availlen, ops); 505 } 506 507 static inline void 508 vq_sync_uring_header(struct virtio_softc *sc, struct virtqueue *vq, int ops) 509 { 510 uint16_t hdrlen = offsetof(struct vring_used, ring); 511 512 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 513 vq->vq_usedoffset, hdrlen, ops); 514 } 515 516 static inline void 517 vq_sync_uring_payload(struct virtio_softc *sc, struct virtqueue *vq, int ops) 518 { 519 uint16_t hdrlen = offsetof(struct vring_used, ring); 520 size_t payloadlen = vq->vq_num * sizeof(struct vring_used_elem); 521 522 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 523 vq->vq_usedoffset + hdrlen, payloadlen, ops); 524 } 525 526 static inline void 527 vq_sync_uring_avail(struct virtio_softc *sc, struct virtqueue *vq, int ops) 528 { 529 uint16_t hdrlen = offsetof(struct vring_used, ring); 530 size_t payloadlen = vq->vq_num * sizeof(struct vring_used_elem); 531 size_t availlen = sizeof(uint16_t); 532 533 if ((sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) == 0) 534 return; 535 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 536 vq->vq_usedoffset + hdrlen + payloadlen, availlen, ops); 537 } 538 539 static inline void 540 vq_sync_indirect(struct virtio_softc *sc, struct virtqueue *vq, int slot, 541 int ops) 542 { 543 int offset = vq->vq_indirectoffset + 544 sizeof(struct vring_desc) * vq->vq_maxnsegs * slot; 545 546 bus_dmamap_sync(sc->sc_dmat, vq->vq_dmamap, 547 offset, sizeof(struct vring_desc) * vq->vq_maxnsegs, ops); 548 } 549 550 bool 551 virtio_vq_is_enqueued(struct virtio_softc *sc, struct virtqueue *vq) 552 { 553 554 if (vq->vq_queued) { 555 vq->vq_queued = 0; 556 vq_sync_aring_all(sc, vq, BUS_DMASYNC_POSTWRITE); 557 } 558 559 vq_sync_uring_header(sc, vq, BUS_DMASYNC_POSTREAD); 560 if (vq->vq_used_idx == virtio_rw16(sc, vq->vq_used->idx)) 561 return 0; 562 vq_sync_uring_payload(sc, vq, BUS_DMASYNC_POSTREAD); 563 return 1; 564 } 565 566 /* 567 * Scan vq, bus_dmamap_sync for the vqs (not for the payload), 568 * and calls (*vq_done)() if some entries are consumed. 569 * 570 * Can be used as sc_intrhand. 571 */ 572 int 573 virtio_vq_intr(struct virtio_softc *sc) 574 { 575 struct virtqueue *vq; 576 int i, r = 0; 577 578 for (i = 0; i < sc->sc_nvqs; i++) { 579 vq = &sc->sc_vqs[i]; 580 if (virtio_vq_is_enqueued(sc, vq) == 1) { 581 if (vq->vq_done) 582 r |= (*vq->vq_done)(vq); 583 } 584 } 585 586 return r; 587 } 588 589 int 590 virtio_vq_intrhand(struct virtio_softc *sc) 591 { 592 struct virtqueue *vq; 593 int i, r = 0; 594 595 for (i = 0; i < sc->sc_nvqs; i++) { 596 vq = &sc->sc_vqs[i]; 597 r |= (*vq->vq_intrhand)(vq->vq_intrhand_arg); 598 } 599 600 return r; 601 } 602 603 604 /* 605 * Increase the event index in order to delay interrupts. 606 */ 607 int 608 virtio_postpone_intr(struct virtio_softc *sc, struct virtqueue *vq, 609 uint16_t nslots) 610 { 611 uint16_t idx, nused; 612 613 idx = vq->vq_used_idx + nslots; 614 615 /* set the new event index: avail_ring->used_event = idx */ 616 *vq->vq_used_event = virtio_rw16(sc, idx); 617 vq_sync_aring_used(vq->vq_owner, vq, BUS_DMASYNC_PREWRITE); 618 vq->vq_queued++; 619 620 nused = (uint16_t) 621 (virtio_rw16(sc, vq->vq_used->idx) - vq->vq_used_idx); 622 KASSERT(nused <= vq->vq_num); 623 624 return nslots < nused; 625 } 626 627 /* 628 * Postpone interrupt until 3/4 of the available descriptors have been 629 * consumed. 630 */ 631 int 632 virtio_postpone_intr_smart(struct virtio_softc *sc, struct virtqueue *vq) 633 { 634 uint16_t nslots; 635 636 nslots = (uint16_t) 637 (virtio_rw16(sc, vq->vq_avail->idx) - vq->vq_used_idx) * 3 / 4; 638 639 return virtio_postpone_intr(sc, vq, nslots); 640 } 641 642 /* 643 * Postpone interrupt until all of the available descriptors have been 644 * consumed. 645 */ 646 int 647 virtio_postpone_intr_far(struct virtio_softc *sc, struct virtqueue *vq) 648 { 649 uint16_t nslots; 650 651 nslots = (uint16_t) 652 (virtio_rw16(sc, vq->vq_avail->idx) - vq->vq_used_idx); 653 654 return virtio_postpone_intr(sc, vq, nslots); 655 } 656 657 /* 658 * Start/stop vq interrupt. No guarantee. 659 */ 660 void 661 virtio_stop_vq_intr(struct virtio_softc *sc, struct virtqueue *vq) 662 { 663 664 if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) { 665 /* 666 * No way to disable the interrupt completely with 667 * RingEventIdx. Instead advance used_event by half the 668 * possible value. This won't happen soon and is far enough in 669 * the past to not trigger a spurios interrupt. 670 */ 671 *vq->vq_used_event = virtio_rw16(sc, vq->vq_used_idx + 0x8000); 672 vq_sync_aring_used(sc, vq, BUS_DMASYNC_PREWRITE); 673 } else { 674 vq->vq_avail->flags |= 675 virtio_rw16(sc, VRING_AVAIL_F_NO_INTERRUPT); 676 vq_sync_aring_header(sc, vq, BUS_DMASYNC_PREWRITE); 677 } 678 vq->vq_queued++; 679 } 680 681 int 682 virtio_start_vq_intr(struct virtio_softc *sc, struct virtqueue *vq) 683 { 684 685 if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) { 686 /* 687 * If event index feature is negotiated, enabling interrupts 688 * is done through setting the latest consumed index in the 689 * used_event field 690 */ 691 *vq->vq_used_event = virtio_rw16(sc, vq->vq_used_idx); 692 vq_sync_aring_used(sc, vq, BUS_DMASYNC_PREWRITE); 693 } else { 694 vq->vq_avail->flags &= 695 ~virtio_rw16(sc, VRING_AVAIL_F_NO_INTERRUPT); 696 vq_sync_aring_header(sc, vq, BUS_DMASYNC_PREWRITE); 697 } 698 vq->vq_queued++; 699 700 vq_sync_uring_header(sc, vq, BUS_DMASYNC_POSTREAD); 701 if (vq->vq_used_idx == virtio_rw16(sc, vq->vq_used->idx)) 702 return 0; 703 vq_sync_uring_payload(sc, vq, BUS_DMASYNC_POSTREAD); 704 return 1; 705 } 706 707 /* 708 * Initialize vq structure. 709 */ 710 static void 711 virtio_init_vq(struct virtio_softc *sc, struct virtqueue *vq, 712 const bool reinit) 713 { 714 int i, j; 715 int vq_size = vq->vq_num; 716 717 memset(vq->vq_vaddr, 0, vq->vq_bytesize); 718 719 /* build the indirect descriptor chain */ 720 if (vq->vq_indirect != NULL) { 721 struct vring_desc *vd; 722 723 for (i = 0; i < vq_size; i++) { 724 vd = vq->vq_indirect; 725 vd += vq->vq_maxnsegs * i; 726 for (j = 0; j < vq->vq_maxnsegs - 1; j++) { 727 vd[j].next = virtio_rw16(sc, j + 1); 728 } 729 } 730 } 731 732 /* free slot management */ 733 SIMPLEQ_INIT(&vq->vq_freelist); 734 for (i = 0; i < vq_size; i++) { 735 SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, &vq->vq_entries[i], 736 qe_list); 737 vq->vq_entries[i].qe_index = i; 738 } 739 if (!reinit) 740 mutex_init(&vq->vq_freelist_lock, MUTEX_SPIN, sc->sc_ipl); 741 742 /* enqueue/dequeue status */ 743 vq->vq_avail_idx = 0; 744 vq->vq_used_idx = 0; 745 vq->vq_queued = 0; 746 if (!reinit) { 747 mutex_init(&vq->vq_aring_lock, MUTEX_SPIN, sc->sc_ipl); 748 mutex_init(&vq->vq_uring_lock, MUTEX_SPIN, sc->sc_ipl); 749 } 750 vq_sync_uring_all(sc, vq, BUS_DMASYNC_PREREAD); 751 vq->vq_queued++; 752 } 753 754 /* 755 * Allocate/free a vq. 756 */ 757 int 758 virtio_alloc_vq(struct virtio_softc *sc, struct virtqueue *vq, int index, 759 int maxsegsize, int maxnsegs, const char *name) 760 { 761 int vq_size, allocsize1, allocsize2, allocsize3, allocsize = 0; 762 int rsegs, r, hdrlen; 763 #define VIRTQUEUE_ALIGN(n) roundup(n, VIRTIO_PAGE_SIZE) 764 765 /* Make sure callers allocate vqs in order */ 766 KASSERT(sc->sc_nvqs == index); 767 768 memset(vq, 0, sizeof(*vq)); 769 770 vq_size = sc->sc_ops->read_queue_size(sc, index); 771 if (vq_size == 0) { 772 aprint_error_dev(sc->sc_dev, 773 "virtqueue not exist, index %d for %s\n", 774 index, name); 775 goto err; 776 } 777 778 hdrlen = sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX ? 3 : 2; 779 780 /* allocsize1: descriptor table + avail ring + pad */ 781 allocsize1 = VIRTQUEUE_ALIGN(sizeof(struct vring_desc) * vq_size 782 + sizeof(uint16_t) * (hdrlen + vq_size)); 783 /* allocsize2: used ring + pad */ 784 allocsize2 = VIRTQUEUE_ALIGN(sizeof(uint16_t) * hdrlen 785 + sizeof(struct vring_used_elem) * vq_size); 786 /* allocsize3: indirect table */ 787 if (sc->sc_indirect && maxnsegs >= MINSEG_INDIRECT) 788 allocsize3 = sizeof(struct vring_desc) * maxnsegs * vq_size; 789 else 790 allocsize3 = 0; 791 allocsize = allocsize1 + allocsize2 + allocsize3; 792 793 /* alloc and map the memory */ 794 r = bus_dmamem_alloc(sc->sc_dmat, allocsize, VIRTIO_PAGE_SIZE, 0, 795 &vq->vq_segs[0], 1, &rsegs, BUS_DMA_WAITOK); 796 if (r != 0) { 797 aprint_error_dev(sc->sc_dev, 798 "virtqueue %d for %s allocation failed, " 799 "error code %d\n", index, name, r); 800 goto err; 801 } 802 r = bus_dmamem_map(sc->sc_dmat, &vq->vq_segs[0], rsegs, allocsize, 803 &vq->vq_vaddr, BUS_DMA_WAITOK); 804 if (r != 0) { 805 aprint_error_dev(sc->sc_dev, 806 "virtqueue %d for %s map failed, " 807 "error code %d\n", index, name, r); 808 goto err; 809 } 810 r = bus_dmamap_create(sc->sc_dmat, allocsize, 1, allocsize, 0, 811 BUS_DMA_WAITOK, &vq->vq_dmamap); 812 if (r != 0) { 813 aprint_error_dev(sc->sc_dev, 814 "virtqueue %d for %s dmamap creation failed, " 815 "error code %d\n", index, name, r); 816 goto err; 817 } 818 r = bus_dmamap_load(sc->sc_dmat, vq->vq_dmamap, 819 vq->vq_vaddr, allocsize, NULL, BUS_DMA_WAITOK); 820 if (r != 0) { 821 aprint_error_dev(sc->sc_dev, 822 "virtqueue %d for %s dmamap load failed, " 823 "error code %d\n", index, name, r); 824 goto err; 825 } 826 827 /* remember addresses and offsets for later use */ 828 vq->vq_owner = sc; 829 vq->vq_num = vq_size; 830 vq->vq_index = index; 831 vq->vq_desc = vq->vq_vaddr; 832 vq->vq_availoffset = sizeof(struct vring_desc) * vq_size; 833 vq->vq_avail = (void *)(((char *)vq->vq_desc) + vq->vq_availoffset); 834 vq->vq_used_event = (uint16_t *)((char *)vq->vq_avail + 835 offsetof(struct vring_avail, ring[vq->vq_num])); 836 vq->vq_usedoffset = allocsize1; 837 vq->vq_used = (void *)(((char *)vq->vq_desc) + vq->vq_usedoffset); 838 vq->vq_avail_event = (uint16_t *)((char *)vq->vq_used + 839 offsetof(struct vring_used, ring[vq->vq_num])); 840 841 if (allocsize3 > 0) { 842 vq->vq_indirectoffset = allocsize1 + allocsize2; 843 vq->vq_indirect = (void *)(((char *)vq->vq_desc) 844 + vq->vq_indirectoffset); 845 } 846 vq->vq_bytesize = allocsize; 847 vq->vq_maxsegsize = maxsegsize; 848 vq->vq_maxnsegs = maxnsegs; 849 850 /* free slot management */ 851 vq->vq_entries = kmem_zalloc(sizeof(struct vq_entry) * vq_size, 852 KM_SLEEP); 853 virtio_init_vq(sc, vq, false); 854 855 /* set the vq address */ 856 sc->sc_ops->setup_queue(sc, index, 857 vq->vq_dmamap->dm_segs[0].ds_addr); 858 859 aprint_verbose_dev(sc->sc_dev, 860 "allocated %u byte for virtqueue %d for %s, size %d\n", 861 allocsize, index, name, vq_size); 862 if (allocsize3 > 0) 863 aprint_verbose_dev(sc->sc_dev, 864 "using %d byte (%d entries) indirect descriptors\n", 865 allocsize3, maxnsegs * vq_size); 866 867 sc->sc_nvqs++; 868 869 return 0; 870 871 err: 872 sc->sc_ops->setup_queue(sc, index, 0); 873 if (vq->vq_dmamap) 874 bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap); 875 if (vq->vq_vaddr) 876 bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, allocsize); 877 if (vq->vq_segs[0].ds_addr) 878 bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1); 879 memset(vq, 0, sizeof(*vq)); 880 881 return -1; 882 } 883 884 int 885 virtio_free_vq(struct virtio_softc *sc, struct virtqueue *vq) 886 { 887 struct vq_entry *qe; 888 int i = 0; 889 890 /* device must be already deactivated */ 891 /* confirm the vq is empty */ 892 SIMPLEQ_FOREACH(qe, &vq->vq_freelist, qe_list) { 893 i++; 894 } 895 if (i != vq->vq_num) { 896 printf("%s: freeing non-empty vq, index %d\n", 897 device_xname(sc->sc_dev), vq->vq_index); 898 return EBUSY; 899 } 900 901 /* tell device that there's no virtqueue any longer */ 902 sc->sc_ops->setup_queue(sc, vq->vq_index, 0); 903 904 vq_sync_aring_all(sc, vq, BUS_DMASYNC_POSTWRITE); 905 906 kmem_free(vq->vq_entries, sizeof(*vq->vq_entries) * vq->vq_num); 907 bus_dmamap_unload(sc->sc_dmat, vq->vq_dmamap); 908 bus_dmamap_destroy(sc->sc_dmat, vq->vq_dmamap); 909 bus_dmamem_unmap(sc->sc_dmat, vq->vq_vaddr, vq->vq_bytesize); 910 bus_dmamem_free(sc->sc_dmat, &vq->vq_segs[0], 1); 911 mutex_destroy(&vq->vq_freelist_lock); 912 mutex_destroy(&vq->vq_uring_lock); 913 mutex_destroy(&vq->vq_aring_lock); 914 memset(vq, 0, sizeof(*vq)); 915 916 sc->sc_nvqs--; 917 918 return 0; 919 } 920 921 /* 922 * Free descriptor management. 923 */ 924 static struct vq_entry * 925 vq_alloc_entry(struct virtqueue *vq) 926 { 927 struct vq_entry *qe; 928 929 mutex_enter(&vq->vq_freelist_lock); 930 if (SIMPLEQ_EMPTY(&vq->vq_freelist)) { 931 mutex_exit(&vq->vq_freelist_lock); 932 return NULL; 933 } 934 qe = SIMPLEQ_FIRST(&vq->vq_freelist); 935 SIMPLEQ_REMOVE_HEAD(&vq->vq_freelist, qe_list); 936 mutex_exit(&vq->vq_freelist_lock); 937 938 return qe; 939 } 940 941 static void 942 vq_free_entry(struct virtqueue *vq, struct vq_entry *qe) 943 { 944 mutex_enter(&vq->vq_freelist_lock); 945 SIMPLEQ_INSERT_TAIL(&vq->vq_freelist, qe, qe_list); 946 mutex_exit(&vq->vq_freelist_lock); 947 948 return; 949 } 950 951 /* 952 * Enqueue several dmamaps as a single request. 953 */ 954 /* 955 * Typical usage: 956 * <queue size> number of followings are stored in arrays 957 * - command blocks (in dmamem) should be pre-allocated and mapped 958 * - dmamaps for command blocks should be pre-allocated and loaded 959 * - dmamaps for payload should be pre-allocated 960 * r = virtio_enqueue_prep(sc, vq, &slot); // allocate a slot 961 * if (r) // currently 0 or EAGAIN 962 * return r; 963 * r = bus_dmamap_load(dmat, dmamap_payload[slot], data, count, ..); 964 * if (r) { 965 * virtio_enqueue_abort(sc, vq, slot); 966 * return r; 967 * } 968 * r = virtio_enqueue_reserve(sc, vq, slot, 969 * dmamap_payload[slot]->dm_nsegs + 1); 970 * // ^ +1 for command 971 * if (r) { // currently 0 or EAGAIN 972 * bus_dmamap_unload(dmat, dmamap_payload[slot]); 973 * return r; // do not call abort() 974 * } 975 * <setup and prepare commands> 976 * bus_dmamap_sync(dmat, dmamap_cmd[slot],... BUS_DMASYNC_PREWRITE); 977 * bus_dmamap_sync(dmat, dmamap_payload[slot],...); 978 * virtio_enqueue(sc, vq, slot, dmamap_cmd[slot], false); 979 * virtio_enqueue(sc, vq, slot, dmamap_payload[slot], iswrite); 980 * virtio_enqueue_commit(sc, vq, slot, true); 981 */ 982 983 /* 984 * enqueue_prep: allocate a slot number 985 */ 986 int 987 virtio_enqueue_prep(struct virtio_softc *sc, struct virtqueue *vq, int *slotp) 988 { 989 struct vq_entry *qe1; 990 991 KASSERT(slotp != NULL); 992 993 qe1 = vq_alloc_entry(vq); 994 if (qe1 == NULL) 995 return EAGAIN; 996 /* next slot is not allocated yet */ 997 qe1->qe_next = -1; 998 *slotp = qe1->qe_index; 999 1000 return 0; 1001 } 1002 1003 /* 1004 * enqueue_reserve: allocate remaining slots and build the descriptor chain. 1005 */ 1006 int 1007 virtio_enqueue_reserve(struct virtio_softc *sc, struct virtqueue *vq, 1008 int slot, int nsegs) 1009 { 1010 int indirect; 1011 struct vq_entry *qe1 = &vq->vq_entries[slot]; 1012 1013 KASSERT(qe1->qe_next == -1); 1014 KASSERT(1 <= nsegs && nsegs <= vq->vq_num); 1015 1016 if ((vq->vq_indirect != NULL) && 1017 (nsegs >= MINSEG_INDIRECT) && 1018 (nsegs <= vq->vq_maxnsegs)) 1019 indirect = 1; 1020 else 1021 indirect = 0; 1022 qe1->qe_indirect = indirect; 1023 1024 if (indirect) { 1025 struct vring_desc *vd; 1026 uint64_t addr; 1027 int i; 1028 1029 vd = &vq->vq_desc[qe1->qe_index]; 1030 addr = vq->vq_dmamap->dm_segs[0].ds_addr 1031 + vq->vq_indirectoffset; 1032 addr += sizeof(struct vring_desc) 1033 * vq->vq_maxnsegs * qe1->qe_index; 1034 vd->addr = virtio_rw64(sc, addr); 1035 vd->len = virtio_rw32(sc, sizeof(struct vring_desc) * nsegs); 1036 vd->flags = virtio_rw16(sc, VRING_DESC_F_INDIRECT); 1037 1038 vd = vq->vq_indirect; 1039 vd += vq->vq_maxnsegs * qe1->qe_index; 1040 qe1->qe_desc_base = vd; 1041 1042 for (i = 0; i < nsegs - 1; i++) { 1043 vd[i].flags = virtio_rw16(sc, VRING_DESC_F_NEXT); 1044 } 1045 vd[i].flags = virtio_rw16(sc, 0); 1046 qe1->qe_next = 0; 1047 1048 return 0; 1049 } else { 1050 struct vring_desc *vd; 1051 struct vq_entry *qe; 1052 int i, s; 1053 1054 vd = &vq->vq_desc[0]; 1055 qe1->qe_desc_base = vd; 1056 qe1->qe_next = qe1->qe_index; 1057 s = slot; 1058 for (i = 0; i < nsegs - 1; i++) { 1059 qe = vq_alloc_entry(vq); 1060 if (qe == NULL) { 1061 vd[s].flags = virtio_rw16(sc, 0); 1062 virtio_enqueue_abort(sc, vq, slot); 1063 return EAGAIN; 1064 } 1065 vd[s].flags = virtio_rw16(sc, VRING_DESC_F_NEXT); 1066 vd[s].next = virtio_rw16(sc, qe->qe_index); 1067 s = qe->qe_index; 1068 } 1069 vd[s].flags = virtio_rw16(sc, 0); 1070 1071 return 0; 1072 } 1073 } 1074 1075 /* 1076 * enqueue: enqueue a single dmamap. 1077 */ 1078 int 1079 virtio_enqueue(struct virtio_softc *sc, struct virtqueue *vq, int slot, 1080 bus_dmamap_t dmamap, bool write) 1081 { 1082 struct vq_entry *qe1 = &vq->vq_entries[slot]; 1083 struct vring_desc *vd = qe1->qe_desc_base; 1084 int i; 1085 int s = qe1->qe_next; 1086 1087 KASSERT(s >= 0); 1088 KASSERT(dmamap->dm_nsegs > 0); 1089 1090 for (i = 0; i < dmamap->dm_nsegs; i++) { 1091 vd[s].addr = virtio_rw64(sc, dmamap->dm_segs[i].ds_addr); 1092 vd[s].len = virtio_rw32(sc, dmamap->dm_segs[i].ds_len); 1093 if (!write) 1094 vd[s].flags |= virtio_rw16(sc, VRING_DESC_F_WRITE); 1095 s = virtio_rw16(sc, vd[s].next); 1096 } 1097 qe1->qe_next = s; 1098 1099 return 0; 1100 } 1101 1102 int 1103 virtio_enqueue_p(struct virtio_softc *sc, struct virtqueue *vq, int slot, 1104 bus_dmamap_t dmamap, bus_addr_t start, bus_size_t len, 1105 bool write) 1106 { 1107 struct vq_entry *qe1 = &vq->vq_entries[slot]; 1108 struct vring_desc *vd = qe1->qe_desc_base; 1109 int s = qe1->qe_next; 1110 1111 KASSERT(s >= 0); 1112 KASSERT(dmamap->dm_nsegs == 1); /* XXX */ 1113 KASSERT(dmamap->dm_segs[0].ds_len > start); 1114 KASSERT(dmamap->dm_segs[0].ds_len >= start + len); 1115 1116 vd[s].addr = virtio_rw64(sc, dmamap->dm_segs[0].ds_addr + start); 1117 vd[s].len = virtio_rw32(sc, len); 1118 if (!write) 1119 vd[s].flags |= virtio_rw16(sc, VRING_DESC_F_WRITE); 1120 qe1->qe_next = virtio_rw16(sc, vd[s].next); 1121 1122 return 0; 1123 } 1124 1125 /* 1126 * enqueue_commit: add it to the aring. 1127 */ 1128 int 1129 virtio_enqueue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot, 1130 bool notifynow) 1131 { 1132 struct vq_entry *qe1; 1133 1134 if (slot < 0) { 1135 mutex_enter(&vq->vq_aring_lock); 1136 goto notify; 1137 } 1138 vq_sync_descs(sc, vq, BUS_DMASYNC_PREWRITE); 1139 qe1 = &vq->vq_entries[slot]; 1140 if (qe1->qe_indirect) 1141 vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_PREWRITE); 1142 mutex_enter(&vq->vq_aring_lock); 1143 vq->vq_avail->ring[(vq->vq_avail_idx++) % vq->vq_num] = 1144 virtio_rw16(sc, slot); 1145 1146 notify: 1147 if (notifynow) { 1148 uint16_t o, n, t; 1149 uint16_t flags; 1150 1151 o = virtio_rw16(sc, vq->vq_avail->idx); 1152 n = vq->vq_avail_idx; 1153 1154 /* 1155 * Prepare for `device->CPU' (host->guest) transfer 1156 * into the buffer. This must happen before we commit 1157 * the vq->vq_avail->idx update to ensure we're not 1158 * still using the buffer in case program-prior loads 1159 * or stores in it get delayed past the store to 1160 * vq->vq_avail->idx. 1161 */ 1162 vq_sync_uring_all(sc, vq, BUS_DMASYNC_PREREAD); 1163 1164 /* ensure payload is published, then avail idx */ 1165 vq_sync_aring_payload(sc, vq, BUS_DMASYNC_PREWRITE); 1166 vq->vq_avail->idx = virtio_rw16(sc, vq->vq_avail_idx); 1167 vq_sync_aring_header(sc, vq, BUS_DMASYNC_PREWRITE); 1168 vq->vq_queued++; 1169 1170 if (sc->sc_active_features & VIRTIO_F_RING_EVENT_IDX) { 1171 vq_sync_uring_avail(sc, vq, BUS_DMASYNC_POSTREAD); 1172 t = virtio_rw16(sc, *vq->vq_avail_event) + 1; 1173 if ((uint16_t) (n - t) < (uint16_t) (n - o)) 1174 sc->sc_ops->kick(sc, vq->vq_index); 1175 } else { 1176 vq_sync_uring_header(sc, vq, BUS_DMASYNC_POSTREAD); 1177 flags = virtio_rw16(sc, vq->vq_used->flags); 1178 if (!(flags & VRING_USED_F_NO_NOTIFY)) 1179 sc->sc_ops->kick(sc, vq->vq_index); 1180 } 1181 } 1182 mutex_exit(&vq->vq_aring_lock); 1183 1184 return 0; 1185 } 1186 1187 /* 1188 * enqueue_abort: rollback. 1189 */ 1190 int 1191 virtio_enqueue_abort(struct virtio_softc *sc, struct virtqueue *vq, int slot) 1192 { 1193 struct vq_entry *qe = &vq->vq_entries[slot]; 1194 struct vring_desc *vd; 1195 int s; 1196 1197 if (qe->qe_next < 0) { 1198 vq_free_entry(vq, qe); 1199 return 0; 1200 } 1201 1202 s = slot; 1203 vd = &vq->vq_desc[0]; 1204 while (virtio_rw16(sc, vd[s].flags) & VRING_DESC_F_NEXT) { 1205 s = virtio_rw16(sc, vd[s].next); 1206 vq_free_entry(vq, qe); 1207 qe = &vq->vq_entries[s]; 1208 } 1209 vq_free_entry(vq, qe); 1210 return 0; 1211 } 1212 1213 /* 1214 * Dequeue a request. 1215 */ 1216 /* 1217 * dequeue: dequeue a request from uring; dmamap_sync for uring is 1218 * already done in the interrupt handler. 1219 */ 1220 int 1221 virtio_dequeue(struct virtio_softc *sc, struct virtqueue *vq, 1222 int *slotp, int *lenp) 1223 { 1224 uint16_t slot, usedidx; 1225 struct vq_entry *qe; 1226 1227 if (vq->vq_used_idx == virtio_rw16(sc, vq->vq_used->idx)) 1228 return ENOENT; 1229 mutex_enter(&vq->vq_uring_lock); 1230 usedidx = vq->vq_used_idx++; 1231 mutex_exit(&vq->vq_uring_lock); 1232 usedidx %= vq->vq_num; 1233 slot = virtio_rw32(sc, vq->vq_used->ring[usedidx].id); 1234 qe = &vq->vq_entries[slot]; 1235 1236 if (qe->qe_indirect) 1237 vq_sync_indirect(sc, vq, slot, BUS_DMASYNC_POSTWRITE); 1238 1239 if (slotp) 1240 *slotp = slot; 1241 if (lenp) 1242 *lenp = virtio_rw32(sc, vq->vq_used->ring[usedidx].len); 1243 1244 return 0; 1245 } 1246 1247 /* 1248 * dequeue_commit: complete dequeue; the slot is recycled for future use. 1249 * if you forget to call this the slot will be leaked. 1250 */ 1251 int 1252 virtio_dequeue_commit(struct virtio_softc *sc, struct virtqueue *vq, int slot) 1253 { 1254 struct vq_entry *qe = &vq->vq_entries[slot]; 1255 struct vring_desc *vd = &vq->vq_desc[0]; 1256 int s = slot; 1257 1258 while (virtio_rw16(sc, vd[s].flags) & VRING_DESC_F_NEXT) { 1259 s = virtio_rw16(sc, vd[s].next); 1260 vq_free_entry(vq, qe); 1261 qe = &vq->vq_entries[s]; 1262 } 1263 vq_free_entry(vq, qe); 1264 1265 return 0; 1266 } 1267 1268 /* 1269 * Attach a child, fill all the members. 1270 */ 1271 void 1272 virtio_child_attach_start(struct virtio_softc *sc, device_t child, int ipl, 1273 struct virtqueue *vqs, 1274 virtio_callback config_change, 1275 virtio_callback intr_hand, 1276 int req_flags, int req_features, const char *feat_bits) 1277 { 1278 char buf[1024]; 1279 1280 sc->sc_child = child; 1281 sc->sc_ipl = ipl; 1282 sc->sc_vqs = vqs; 1283 sc->sc_config_change = config_change; 1284 sc->sc_intrhand = intr_hand; 1285 sc->sc_flags = req_flags; 1286 1287 virtio_negotiate_features(sc, req_features); 1288 snprintb(buf, sizeof(buf), feat_bits, sc->sc_active_features); 1289 aprint_normal(": features: %s\n", buf); 1290 aprint_naive("\n"); 1291 } 1292 1293 void 1294 virtio_child_attach_set_vqs(struct virtio_softc *sc, 1295 struct virtqueue *vqs, int nvq_pairs) 1296 { 1297 1298 KASSERT(nvq_pairs == 1 || 1299 (sc->sc_flags & VIRTIO_F_INTR_SOFTINT) == 0); 1300 if (nvq_pairs > 1) 1301 sc->sc_child_mq = true; 1302 1303 sc->sc_vqs = vqs; 1304 } 1305 1306 int 1307 virtio_child_attach_finish(struct virtio_softc *sc) 1308 { 1309 int r; 1310 1311 sc->sc_finished_called = true; 1312 r = sc->sc_ops->alloc_interrupts(sc); 1313 if (r != 0) { 1314 aprint_error_dev(sc->sc_dev, 1315 "failed to allocate interrupts\n"); 1316 goto fail; 1317 } 1318 1319 r = sc->sc_ops->setup_interrupts(sc, 0); 1320 if (r != 0) { 1321 aprint_error_dev(sc->sc_dev, "failed to setup interrupts\n"); 1322 goto fail; 1323 } 1324 1325 KASSERT(sc->sc_soft_ih == NULL); 1326 if (sc->sc_flags & VIRTIO_F_INTR_SOFTINT) { 1327 u_int flags = SOFTINT_NET; 1328 if (sc->sc_flags & VIRTIO_F_INTR_MPSAFE) 1329 flags |= SOFTINT_MPSAFE; 1330 1331 sc->sc_soft_ih = softint_establish(flags, virtio_soft_intr, 1332 sc); 1333 if (sc->sc_soft_ih == NULL) { 1334 sc->sc_ops->free_interrupts(sc); 1335 aprint_error_dev(sc->sc_dev, 1336 "failed to establish soft interrupt\n"); 1337 goto fail; 1338 } 1339 } 1340 1341 virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK); 1342 return 0; 1343 1344 fail: 1345 if (sc->sc_soft_ih) { 1346 softint_disestablish(sc->sc_soft_ih); 1347 sc->sc_soft_ih = NULL; 1348 } 1349 1350 sc->sc_ops->free_interrupts(sc); 1351 1352 virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED); 1353 return 1; 1354 } 1355 1356 void 1357 virtio_child_detach(struct virtio_softc *sc) 1358 { 1359 sc->sc_child = NULL; 1360 sc->sc_vqs = NULL; 1361 1362 virtio_device_reset(sc); 1363 1364 sc->sc_ops->free_interrupts(sc); 1365 1366 if (sc->sc_soft_ih) { 1367 softint_disestablish(sc->sc_soft_ih); 1368 sc->sc_soft_ih = NULL; 1369 } 1370 } 1371 1372 void 1373 virtio_child_attach_failed(struct virtio_softc *sc) 1374 { 1375 virtio_child_detach(sc); 1376 1377 virtio_set_status(sc, VIRTIO_CONFIG_DEVICE_STATUS_FAILED); 1378 1379 sc->sc_child = VIRTIO_CHILD_FAILED; 1380 } 1381 1382 bus_dma_tag_t 1383 virtio_dmat(struct virtio_softc *sc) 1384 { 1385 return sc->sc_dmat; 1386 } 1387 1388 device_t 1389 virtio_child(struct virtio_softc *sc) 1390 { 1391 return sc->sc_child; 1392 } 1393 1394 int 1395 virtio_intrhand(struct virtio_softc *sc) 1396 { 1397 return (*sc->sc_intrhand)(sc); 1398 } 1399 1400 uint64_t 1401 virtio_features(struct virtio_softc *sc) 1402 { 1403 return sc->sc_active_features; 1404 } 1405 1406 int 1407 virtio_attach_failed(struct virtio_softc *sc) 1408 { 1409 device_t self = sc->sc_dev; 1410 1411 /* no error if its not connected, but its failed */ 1412 if (sc->sc_childdevid == 0) 1413 return 1; 1414 1415 if (sc->sc_child == NULL) { 1416 aprint_error_dev(self, 1417 "no matching child driver; not configured\n"); 1418 return 1; 1419 } 1420 1421 if (sc->sc_child == VIRTIO_CHILD_FAILED) { 1422 aprint_error_dev(self, "virtio configuration failed\n"); 1423 return 1; 1424 } 1425 1426 /* sanity check */ 1427 if (!sc->sc_finished_called) { 1428 aprint_error_dev(self, "virtio internal error, child driver " 1429 "signaled OK but didn't initialize interrupts\n"); 1430 return 1; 1431 } 1432 1433 return 0; 1434 } 1435 1436 void 1437 virtio_print_device_type(device_t self, int id, int revision) 1438 { 1439 aprint_normal_dev(self, "%s device (id %d, rev. 0x%02x)\n", 1440 (id < NDEVNAMES ? virtio_device_name[id] : "Unknown"), 1441 id, 1442 revision); 1443 } 1444 1445 1446 MODULE(MODULE_CLASS_DRIVER, virtio, NULL); 1447 1448 #ifdef _MODULE 1449 #include "ioconf.c" 1450 #endif 1451 1452 static int 1453 virtio_modcmd(modcmd_t cmd, void *opaque) 1454 { 1455 int error = 0; 1456 1457 #ifdef _MODULE 1458 switch (cmd) { 1459 case MODULE_CMD_INIT: 1460 error = config_init_component(cfdriver_ioconf_virtio, 1461 cfattach_ioconf_virtio, cfdata_ioconf_virtio); 1462 break; 1463 case MODULE_CMD_FINI: 1464 error = config_fini_component(cfdriver_ioconf_virtio, 1465 cfattach_ioconf_virtio, cfdata_ioconf_virtio); 1466 break; 1467 default: 1468 error = ENOTTY; 1469 break; 1470 } 1471 #endif 1472 1473 return error; 1474 } 1475
Indexes created Mon Sep 22 05:09:51 GMT 2025