vmbus_chan.c revision 307595
1/*- 2 * Copyright (c) 2009-2012,2016 Microsoft Corp. 3 * Copyright (c) 2012 NetApp Inc. 4 * Copyright (c) 2012 Citrix Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice unmodified, this list of conditions, and the following 12 * disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29#include <sys/cdefs.h> 30__FBSDID("$FreeBSD: stable/11/sys/dev/hyperv/vmbus/vmbus_chan.c 307595 2016-10-19 02:14:04Z sephe $"); 31 32#include <sys/param.h> 33#include <sys/kernel.h> 34#include <sys/lock.h> 35#include <sys/malloc.h> 36#include <sys/mutex.h> 37#include <sys/smp.h> 38#include <sys/sysctl.h> 39#include <sys/systm.h> 40 41#include <machine/atomic.h> 42 43#include <dev/hyperv/include/hyperv_busdma.h> 44#include <dev/hyperv/vmbus/hyperv_var.h> 45#include <dev/hyperv/vmbus/vmbus_reg.h> 46#include <dev/hyperv/vmbus/vmbus_var.h> 47#include <dev/hyperv/vmbus/vmbus_brvar.h> 48#include <dev/hyperv/vmbus/vmbus_chanvar.h> 49 50static void vmbus_chan_update_evtflagcnt( 51 struct vmbus_softc *, 52 const struct vmbus_channel *); 53static void vmbus_chan_close_internal( 54 struct vmbus_channel *); 55static int vmbus_chan_sysctl_mnf(SYSCTL_HANDLER_ARGS); 56static void vmbus_chan_sysctl_create( 57 struct vmbus_channel *); 58static struct vmbus_channel *vmbus_chan_alloc(struct vmbus_softc *); 59static void vmbus_chan_free(struct vmbus_channel *); 60static int vmbus_chan_add(struct vmbus_channel *); 61static void vmbus_chan_cpu_default(struct vmbus_channel *); 62 63static void vmbus_chan_task(void *, int); 64static void vmbus_chan_task_nobatch(void *, int); 65static void vmbus_chan_detach_task(void *, int); 66 67static void vmbus_chan_msgproc_choffer(struct vmbus_softc *, 68 const struct vmbus_message *); 69static void vmbus_chan_msgproc_chrescind( 70 struct vmbus_softc *, 71 const struct vmbus_message *); 72 73/* 74 * Vmbus channel message processing. 75 */ 76static const vmbus_chanmsg_proc_t 77vmbus_chan_msgprocs[VMBUS_CHANMSG_TYPE_MAX] = { 78 VMBUS_CHANMSG_PROC(CHOFFER, vmbus_chan_msgproc_choffer), 79 VMBUS_CHANMSG_PROC(CHRESCIND, vmbus_chan_msgproc_chrescind), 80 81 VMBUS_CHANMSG_PROC_WAKEUP(CHOPEN_RESP), 82 VMBUS_CHANMSG_PROC_WAKEUP(GPADL_CONNRESP), 83 VMBUS_CHANMSG_PROC_WAKEUP(GPADL_DISCONNRESP) 84}; 85 86/* 87 * Notify host that there are data pending on our TX bufring. 88 */ 89static __inline void 90vmbus_chan_signal_tx(const struct vmbus_channel *chan) 91{ 92 atomic_set_long(chan->ch_evtflag, chan->ch_evtflag_mask); 93 if (chan->ch_txflags & VMBUS_CHAN_TXF_HASMNF) 94 atomic_set_int(chan->ch_montrig, chan->ch_montrig_mask); 95 else 96 hypercall_signal_event(chan->ch_monprm_dma.hv_paddr); 97} 98 99static int 100vmbus_chan_sysctl_mnf(SYSCTL_HANDLER_ARGS) 101{ 102 struct vmbus_channel *chan = arg1; 103 int mnf = 0; 104 105 if (chan->ch_txflags & VMBUS_CHAN_TXF_HASMNF) 106 mnf = 1; 107 return sysctl_handle_int(oidp, &mnf, 0, req); 108} 109 110static void 111vmbus_chan_sysctl_create(struct vmbus_channel *chan) 112{ 113 struct sysctl_oid *ch_tree, *chid_tree, *br_tree; 114 struct sysctl_ctx_list *ctx; 115 uint32_t ch_id; 116 char name[16]; 117 118 /* 119 * Add sysctl nodes related to this channel to this 120 * channel's sysctl ctx, so that they can be destroyed 121 * independently upon close of this channel, which can 122 * happen even if the device is not detached. 123 */ 124 ctx = &chan->ch_sysctl_ctx; 125 sysctl_ctx_init(ctx); 126 127 /* 128 * Create dev.NAME.UNIT.channel tree. 129 */ 130 ch_tree = SYSCTL_ADD_NODE(ctx, 131 SYSCTL_CHILDREN(device_get_sysctl_tree(chan->ch_dev)), 132 OID_AUTO, "channel", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); 133 if (ch_tree == NULL) 134 return; 135 136 /* 137 * Create dev.NAME.UNIT.channel.CHANID tree. 138 */ 139 if (VMBUS_CHAN_ISPRIMARY(chan)) 140 ch_id = chan->ch_id; 141 else 142 ch_id = chan->ch_prichan->ch_id; 143 snprintf(name, sizeof(name), "%d", ch_id); 144 chid_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(ch_tree), 145 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); 146 if (chid_tree == NULL) 147 return; 148 149 if (!VMBUS_CHAN_ISPRIMARY(chan)) { 150 /* 151 * Create dev.NAME.UNIT.channel.CHANID.sub tree. 152 */ 153 ch_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(chid_tree), 154 OID_AUTO, "sub", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); 155 if (ch_tree == NULL) 156 return; 157 158 /* 159 * Create dev.NAME.UNIT.channel.CHANID.sub.SUBIDX tree. 160 * 161 * NOTE: 162 * chid_tree is changed to this new sysctl tree. 163 */ 164 snprintf(name, sizeof(name), "%d", chan->ch_subidx); 165 chid_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(ch_tree), 166 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); 167 if (chid_tree == NULL) 168 return; 169 170 SYSCTL_ADD_UINT(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO, 171 "chanid", CTLFLAG_RD, &chan->ch_id, 0, "channel id"); 172 } 173 174 SYSCTL_ADD_UINT(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO, 175 "cpu", CTLFLAG_RD, &chan->ch_cpuid, 0, "owner CPU id"); 176 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO, 177 "mnf", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, 178 chan, 0, vmbus_chan_sysctl_mnf, "I", 179 "has monitor notification facilities"); 180 181 br_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO, 182 "br", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, ""); 183 if (br_tree != NULL) { 184 /* 185 * Create sysctl tree for RX bufring. 186 */ 187 vmbus_br_sysctl_create(ctx, br_tree, &chan->ch_rxbr.rxbr, "rx"); 188 /* 189 * Create sysctl tree for TX bufring. 190 */ 191 vmbus_br_sysctl_create(ctx, br_tree, &chan->ch_txbr.txbr, "tx"); 192 } 193} 194 195int 196vmbus_chan_open(struct vmbus_channel *chan, int txbr_size, int rxbr_size, 197 const void *udata, int udlen, vmbus_chan_callback_t cb, void *cbarg) 198{ 199 struct vmbus_chan_br cbr; 200 int error; 201 202 /* 203 * Allocate the TX+RX bufrings. 204 */ 205 KASSERT(chan->ch_bufring == NULL, ("bufrings are allocated")); 206 chan->ch_bufring = hyperv_dmamem_alloc(bus_get_dma_tag(chan->ch_dev), 207 PAGE_SIZE, 0, txbr_size + rxbr_size, &chan->ch_bufring_dma, 208 BUS_DMA_WAITOK); 209 if (chan->ch_bufring == NULL) { 210 device_printf(chan->ch_dev, "bufring allocation failed\n"); 211 return (ENOMEM); 212 } 213 214 cbr.cbr = chan->ch_bufring; 215 cbr.cbr_paddr = chan->ch_bufring_dma.hv_paddr; 216 cbr.cbr_txsz = txbr_size; 217 cbr.cbr_rxsz = rxbr_size; 218 219 error = vmbus_chan_open_br(chan, &cbr, udata, udlen, cb, cbarg); 220 if (error) { 221 hyperv_dmamem_free(&chan->ch_bufring_dma, chan->ch_bufring); 222 chan->ch_bufring = NULL; 223 } 224 return (error); 225} 226 227int 228vmbus_chan_open_br(struct vmbus_channel *chan, const struct vmbus_chan_br *cbr, 229 const void *udata, int udlen, vmbus_chan_callback_t cb, void *cbarg) 230{ 231 struct vmbus_softc *sc = chan->ch_vmbus; 232 const struct vmbus_chanmsg_chopen_resp *resp; 233 const struct vmbus_message *msg; 234 struct vmbus_chanmsg_chopen *req; 235 struct vmbus_msghc *mh; 236 uint32_t status; 237 int error, txbr_size, rxbr_size; 238 uint8_t *br; 239 240 if (udlen > VMBUS_CHANMSG_CHOPEN_UDATA_SIZE) { 241 device_printf(sc->vmbus_dev, 242 "invalid udata len %d for chan%u\n", udlen, chan->ch_id); 243 return EINVAL; 244 } 245 246 br = cbr->cbr; 247 txbr_size = cbr->cbr_txsz; 248 rxbr_size = cbr->cbr_rxsz; 249 KASSERT((txbr_size & PAGE_MASK) == 0, 250 ("send bufring size is not multiple page")); 251 KASSERT((rxbr_size & PAGE_MASK) == 0, 252 ("recv bufring size is not multiple page")); 253 KASSERT((cbr->cbr_paddr & PAGE_MASK) == 0, 254 ("bufring is not page aligned")); 255 256 /* 257 * Zero out the TX/RX bufrings, in case that they were used before. 258 */ 259 memset(br, 0, txbr_size + rxbr_size); 260 261 if (atomic_testandset_int(&chan->ch_stflags, 262 VMBUS_CHAN_ST_OPENED_SHIFT)) 263 panic("double-open chan%u", chan->ch_id); 264 265 chan->ch_cb = cb; 266 chan->ch_cbarg = cbarg; 267 268 vmbus_chan_update_evtflagcnt(sc, chan); 269 270 chan->ch_tq = VMBUS_PCPU_GET(chan->ch_vmbus, event_tq, chan->ch_cpuid); 271 if (chan->ch_flags & VMBUS_CHAN_FLAG_BATCHREAD) 272 TASK_INIT(&chan->ch_task, 0, vmbus_chan_task, chan); 273 else 274 TASK_INIT(&chan->ch_task, 0, vmbus_chan_task_nobatch, chan); 275 276 /* TX bufring comes first */ 277 vmbus_txbr_setup(&chan->ch_txbr, br, txbr_size); 278 /* RX bufring immediately follows TX bufring */ 279 vmbus_rxbr_setup(&chan->ch_rxbr, br + txbr_size, rxbr_size); 280 281 /* Create sysctl tree for this channel */ 282 vmbus_chan_sysctl_create(chan); 283 284 /* 285 * Connect the bufrings, both RX and TX, to this channel. 286 */ 287 error = vmbus_chan_gpadl_connect(chan, cbr->cbr_paddr, 288 txbr_size + rxbr_size, &chan->ch_bufring_gpadl); 289 if (error) { 290 device_printf(sc->vmbus_dev, 291 "failed to connect bufring GPADL to chan%u\n", chan->ch_id); 292 goto failed; 293 } 294 295 /* 296 * Open channel w/ the bufring GPADL on the target CPU. 297 */ 298 mh = vmbus_msghc_get(sc, sizeof(*req)); 299 if (mh == NULL) { 300 device_printf(sc->vmbus_dev, 301 "can not get msg hypercall for chopen(chan%u)\n", 302 chan->ch_id); 303 error = ENXIO; 304 goto failed; 305 } 306 307 req = vmbus_msghc_dataptr(mh); 308 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHOPEN; 309 req->chm_chanid = chan->ch_id; 310 req->chm_openid = chan->ch_id; 311 req->chm_gpadl = chan->ch_bufring_gpadl; 312 req->chm_vcpuid = chan->ch_vcpuid; 313 req->chm_txbr_pgcnt = txbr_size >> PAGE_SHIFT; 314 if (udlen > 0) 315 memcpy(req->chm_udata, udata, udlen); 316 317 error = vmbus_msghc_exec(sc, mh); 318 if (error) { 319 device_printf(sc->vmbus_dev, 320 "chopen(chan%u) msg hypercall exec failed: %d\n", 321 chan->ch_id, error); 322 vmbus_msghc_put(sc, mh); 323 goto failed; 324 } 325 326 msg = vmbus_msghc_wait_result(sc, mh); 327 resp = (const struct vmbus_chanmsg_chopen_resp *)msg->msg_data; 328 status = resp->chm_status; 329 330 vmbus_msghc_put(sc, mh); 331 332 if (status == 0) { 333 if (bootverbose) { 334 device_printf(sc->vmbus_dev, "chan%u opened\n", 335 chan->ch_id); 336 } 337 return 0; 338 } 339 340 device_printf(sc->vmbus_dev, "failed to open chan%u\n", chan->ch_id); 341 error = ENXIO; 342 343failed: 344 if (chan->ch_bufring_gpadl) { 345 vmbus_chan_gpadl_disconnect(chan, chan->ch_bufring_gpadl); 346 chan->ch_bufring_gpadl = 0; 347 } 348 atomic_clear_int(&chan->ch_stflags, VMBUS_CHAN_ST_OPENED); 349 return error; 350} 351 352int 353vmbus_chan_gpadl_connect(struct vmbus_channel *chan, bus_addr_t paddr, 354 int size, uint32_t *gpadl0) 355{ 356 struct vmbus_softc *sc = chan->ch_vmbus; 357 struct vmbus_msghc *mh; 358 struct vmbus_chanmsg_gpadl_conn *req; 359 const struct vmbus_message *msg; 360 size_t reqsz; 361 uint32_t gpadl, status; 362 int page_count, range_len, i, cnt, error; 363 uint64_t page_id; 364 365 /* 366 * Preliminary checks. 367 */ 368 369 KASSERT((size & PAGE_MASK) == 0, 370 ("invalid GPA size %d, not multiple page size", size)); 371 page_count = size >> PAGE_SHIFT; 372 373 KASSERT((paddr & PAGE_MASK) == 0, 374 ("GPA is not page aligned %jx", (uintmax_t)paddr)); 375 page_id = paddr >> PAGE_SHIFT; 376 377 range_len = __offsetof(struct vmbus_gpa_range, gpa_page[page_count]); 378 /* 379 * We don't support multiple GPA ranges. 380 */ 381 if (range_len > UINT16_MAX) { 382 device_printf(sc->vmbus_dev, "GPA too large, %d pages\n", 383 page_count); 384 return EOPNOTSUPP; 385 } 386 387 /* 388 * Allocate GPADL id. 389 */ 390 gpadl = vmbus_gpadl_alloc(sc); 391 *gpadl0 = gpadl; 392 393 /* 394 * Connect this GPADL to the target channel. 395 * 396 * NOTE: 397 * Since each message can only hold small set of page 398 * addresses, several messages may be required to 399 * complete the connection. 400 */ 401 if (page_count > VMBUS_CHANMSG_GPADL_CONN_PGMAX) 402 cnt = VMBUS_CHANMSG_GPADL_CONN_PGMAX; 403 else 404 cnt = page_count; 405 page_count -= cnt; 406 407 reqsz = __offsetof(struct vmbus_chanmsg_gpadl_conn, 408 chm_range.gpa_page[cnt]); 409 mh = vmbus_msghc_get(sc, reqsz); 410 if (mh == NULL) { 411 device_printf(sc->vmbus_dev, 412 "can not get msg hypercall for gpadl->chan%u\n", 413 chan->ch_id); 414 return EIO; 415 } 416 417 req = vmbus_msghc_dataptr(mh); 418 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_GPADL_CONN; 419 req->chm_chanid = chan->ch_id; 420 req->chm_gpadl = gpadl; 421 req->chm_range_len = range_len; 422 req->chm_range_cnt = 1; 423 req->chm_range.gpa_len = size; 424 req->chm_range.gpa_ofs = 0; 425 for (i = 0; i < cnt; ++i) 426 req->chm_range.gpa_page[i] = page_id++; 427 428 error = vmbus_msghc_exec(sc, mh); 429 if (error) { 430 device_printf(sc->vmbus_dev, 431 "gpadl->chan%u msg hypercall exec failed: %d\n", 432 chan->ch_id, error); 433 vmbus_msghc_put(sc, mh); 434 return error; 435 } 436 437 while (page_count > 0) { 438 struct vmbus_chanmsg_gpadl_subconn *subreq; 439 440 if (page_count > VMBUS_CHANMSG_GPADL_SUBCONN_PGMAX) 441 cnt = VMBUS_CHANMSG_GPADL_SUBCONN_PGMAX; 442 else 443 cnt = page_count; 444 page_count -= cnt; 445 446 reqsz = __offsetof(struct vmbus_chanmsg_gpadl_subconn, 447 chm_gpa_page[cnt]); 448 vmbus_msghc_reset(mh, reqsz); 449 450 subreq = vmbus_msghc_dataptr(mh); 451 subreq->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_GPADL_SUBCONN; 452 subreq->chm_gpadl = gpadl; 453 for (i = 0; i < cnt; ++i) 454 subreq->chm_gpa_page[i] = page_id++; 455 456 vmbus_msghc_exec_noresult(mh); 457 } 458 KASSERT(page_count == 0, ("invalid page count %d", page_count)); 459 460 msg = vmbus_msghc_wait_result(sc, mh); 461 status = ((const struct vmbus_chanmsg_gpadl_connresp *) 462 msg->msg_data)->chm_status; 463 464 vmbus_msghc_put(sc, mh); 465 466 if (status != 0) { 467 device_printf(sc->vmbus_dev, "gpadl->chan%u failed: " 468 "status %u\n", chan->ch_id, status); 469 return EIO; 470 } else { 471 if (bootverbose) { 472 device_printf(sc->vmbus_dev, "gpadl->chan%u " 473 "succeeded\n", chan->ch_id); 474 } 475 } 476 return 0; 477} 478 479/* 480 * Disconnect the GPA from the target channel 481 */ 482int 483vmbus_chan_gpadl_disconnect(struct vmbus_channel *chan, uint32_t gpadl) 484{ 485 struct vmbus_softc *sc = chan->ch_vmbus; 486 struct vmbus_msghc *mh; 487 struct vmbus_chanmsg_gpadl_disconn *req; 488 int error; 489 490 mh = vmbus_msghc_get(sc, sizeof(*req)); 491 if (mh == NULL) { 492 device_printf(sc->vmbus_dev, 493 "can not get msg hypercall for gpa x->chan%u\n", 494 chan->ch_id); 495 return EBUSY; 496 } 497 498 req = vmbus_msghc_dataptr(mh); 499 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_GPADL_DISCONN; 500 req->chm_chanid = chan->ch_id; 501 req->chm_gpadl = gpadl; 502 503 error = vmbus_msghc_exec(sc, mh); 504 if (error) { 505 device_printf(sc->vmbus_dev, 506 "gpa x->chan%u msg hypercall exec failed: %d\n", 507 chan->ch_id, error); 508 vmbus_msghc_put(sc, mh); 509 return error; 510 } 511 512 vmbus_msghc_wait_result(sc, mh); 513 /* Discard result; no useful information */ 514 vmbus_msghc_put(sc, mh); 515 516 return 0; 517} 518 519static void 520vmbus_chan_close_internal(struct vmbus_channel *chan) 521{ 522 struct vmbus_softc *sc = chan->ch_vmbus; 523 struct vmbus_msghc *mh; 524 struct vmbus_chanmsg_chclose *req; 525 struct taskqueue *tq = chan->ch_tq; 526 int error; 527 528 /* TODO: stringent check */ 529 atomic_clear_int(&chan->ch_stflags, VMBUS_CHAN_ST_OPENED); 530 531 /* 532 * Free this channel's sysctl tree attached to its device's 533 * sysctl tree. 534 */ 535 sysctl_ctx_free(&chan->ch_sysctl_ctx); 536 537 /* 538 * Set ch_tq to NULL to avoid more requests be scheduled. 539 * XXX pretty broken; need rework. 540 */ 541 chan->ch_tq = NULL; 542 taskqueue_drain(tq, &chan->ch_task); 543 chan->ch_cb = NULL; 544 545 /* 546 * Close this channel. 547 */ 548 mh = vmbus_msghc_get(sc, sizeof(*req)); 549 if (mh == NULL) { 550 device_printf(sc->vmbus_dev, 551 "can not get msg hypercall for chclose(chan%u)\n", 552 chan->ch_id); 553 return; 554 } 555 556 req = vmbus_msghc_dataptr(mh); 557 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHCLOSE; 558 req->chm_chanid = chan->ch_id; 559 560 error = vmbus_msghc_exec_noresult(mh); 561 vmbus_msghc_put(sc, mh); 562 563 if (error) { 564 device_printf(sc->vmbus_dev, 565 "chclose(chan%u) msg hypercall exec failed: %d\n", 566 chan->ch_id, error); 567 return; 568 } else if (bootverbose) { 569 device_printf(sc->vmbus_dev, "close chan%u\n", chan->ch_id); 570 } 571 572 /* 573 * Disconnect the TX+RX bufrings from this channel. 574 */ 575 if (chan->ch_bufring_gpadl) { 576 vmbus_chan_gpadl_disconnect(chan, chan->ch_bufring_gpadl); 577 chan->ch_bufring_gpadl = 0; 578 } 579 580 /* 581 * Destroy the TX+RX bufrings. 582 */ 583 if (chan->ch_bufring != NULL) { 584 hyperv_dmamem_free(&chan->ch_bufring_dma, chan->ch_bufring); 585 chan->ch_bufring = NULL; 586 } 587} 588 589/* 590 * Caller should make sure that all sub-channels have 591 * been added to 'chan' and all to-be-closed channels 592 * are not being opened. 593 */ 594void 595vmbus_chan_close(struct vmbus_channel *chan) 596{ 597 int subchan_cnt; 598 599 if (!VMBUS_CHAN_ISPRIMARY(chan)) { 600 /* 601 * Sub-channel is closed when its primary channel 602 * is closed; done. 603 */ 604 return; 605 } 606 607 /* 608 * Close all sub-channels, if any. 609 */ 610 subchan_cnt = chan->ch_subchan_cnt; 611 if (subchan_cnt > 0) { 612 struct vmbus_channel **subchan; 613 int i; 614 615 subchan = vmbus_subchan_get(chan, subchan_cnt); 616 for (i = 0; i < subchan_cnt; ++i) 617 vmbus_chan_close_internal(subchan[i]); 618 vmbus_subchan_rel(subchan, subchan_cnt); 619 } 620 621 /* Then close the primary channel. */ 622 vmbus_chan_close_internal(chan); 623} 624 625int 626vmbus_chan_send(struct vmbus_channel *chan, uint16_t type, uint16_t flags, 627 void *data, int dlen, uint64_t xactid) 628{ 629 struct vmbus_chanpkt pkt; 630 int pktlen, pad_pktlen, hlen, error; 631 uint64_t pad = 0; 632 struct iovec iov[3]; 633 boolean_t send_evt; 634 635 hlen = sizeof(pkt); 636 pktlen = hlen + dlen; 637 pad_pktlen = VMBUS_CHANPKT_TOTLEN(pktlen); 638 KASSERT(pad_pktlen <= vmbus_txbr_maxpktsz(&chan->ch_txbr), 639 ("invalid packet size %d", pad_pktlen)); 640 641 pkt.cp_hdr.cph_type = type; 642 pkt.cp_hdr.cph_flags = flags; 643 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_hlen, hlen); 644 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_tlen, pad_pktlen); 645 pkt.cp_hdr.cph_xactid = xactid; 646 647 iov[0].iov_base = &pkt; 648 iov[0].iov_len = hlen; 649 iov[1].iov_base = data; 650 iov[1].iov_len = dlen; 651 iov[2].iov_base = &pad; 652 iov[2].iov_len = pad_pktlen - pktlen; 653 654 error = vmbus_txbr_write(&chan->ch_txbr, iov, 3, &send_evt); 655 if (!error && send_evt) 656 vmbus_chan_signal_tx(chan); 657 return error; 658} 659 660int 661vmbus_chan_send_sglist(struct vmbus_channel *chan, 662 struct vmbus_gpa sg[], int sglen, void *data, int dlen, uint64_t xactid) 663{ 664 struct vmbus_chanpkt_sglist pkt; 665 int pktlen, pad_pktlen, hlen, error; 666 struct iovec iov[4]; 667 boolean_t send_evt; 668 uint64_t pad = 0; 669 670 hlen = __offsetof(struct vmbus_chanpkt_sglist, cp_gpa[sglen]); 671 pktlen = hlen + dlen; 672 pad_pktlen = VMBUS_CHANPKT_TOTLEN(pktlen); 673 KASSERT(pad_pktlen <= vmbus_txbr_maxpktsz(&chan->ch_txbr), 674 ("invalid packet size %d", pad_pktlen)); 675 676 pkt.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA; 677 pkt.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC; 678 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_hlen, hlen); 679 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_tlen, pad_pktlen); 680 pkt.cp_hdr.cph_xactid = xactid; 681 pkt.cp_rsvd = 0; 682 pkt.cp_gpa_cnt = sglen; 683 684 iov[0].iov_base = &pkt; 685 iov[0].iov_len = sizeof(pkt); 686 iov[1].iov_base = sg; 687 iov[1].iov_len = sizeof(struct vmbus_gpa) * sglen; 688 iov[2].iov_base = data; 689 iov[2].iov_len = dlen; 690 iov[3].iov_base = &pad; 691 iov[3].iov_len = pad_pktlen - pktlen; 692 693 error = vmbus_txbr_write(&chan->ch_txbr, iov, 4, &send_evt); 694 if (!error && send_evt) 695 vmbus_chan_signal_tx(chan); 696 return error; 697} 698 699int 700vmbus_chan_send_prplist(struct vmbus_channel *chan, 701 struct vmbus_gpa_range *prp, int prp_cnt, void *data, int dlen, 702 uint64_t xactid) 703{ 704 struct vmbus_chanpkt_prplist pkt; 705 int pktlen, pad_pktlen, hlen, error; 706 struct iovec iov[4]; 707 boolean_t send_evt; 708 uint64_t pad = 0; 709 710 hlen = __offsetof(struct vmbus_chanpkt_prplist, 711 cp_range[0].gpa_page[prp_cnt]); 712 pktlen = hlen + dlen; 713 pad_pktlen = VMBUS_CHANPKT_TOTLEN(pktlen); 714 KASSERT(pad_pktlen <= vmbus_txbr_maxpktsz(&chan->ch_txbr), 715 ("invalid packet size %d", pad_pktlen)); 716 717 pkt.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA; 718 pkt.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC; 719 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_hlen, hlen); 720 VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_tlen, pad_pktlen); 721 pkt.cp_hdr.cph_xactid = xactid; 722 pkt.cp_rsvd = 0; 723 pkt.cp_range_cnt = 1; 724 725 iov[0].iov_base = &pkt; 726 iov[0].iov_len = sizeof(pkt); 727 iov[1].iov_base = prp; 728 iov[1].iov_len = __offsetof(struct vmbus_gpa_range, gpa_page[prp_cnt]); 729 iov[2].iov_base = data; 730 iov[2].iov_len = dlen; 731 iov[3].iov_base = &pad; 732 iov[3].iov_len = pad_pktlen - pktlen; 733 734 error = vmbus_txbr_write(&chan->ch_txbr, iov, 4, &send_evt); 735 if (!error && send_evt) 736 vmbus_chan_signal_tx(chan); 737 return error; 738} 739 740int 741vmbus_chan_recv(struct vmbus_channel *chan, void *data, int *dlen0, 742 uint64_t *xactid) 743{ 744 struct vmbus_chanpkt_hdr pkt; 745 int error, dlen, hlen; 746 747 error = vmbus_rxbr_peek(&chan->ch_rxbr, &pkt, sizeof(pkt)); 748 if (error) 749 return (error); 750 751 if (__predict_false(pkt.cph_hlen < VMBUS_CHANPKT_HLEN_MIN)) { 752 device_printf(chan->ch_dev, "invalid hlen %u\n", 753 pkt.cph_hlen); 754 /* XXX this channel is dead actually. */ 755 return (EIO); 756 } 757 if (__predict_false(pkt.cph_hlen > pkt.cph_tlen)) { 758 device_printf(chan->ch_dev, "invalid hlen %u and tlen %u\n", 759 pkt.cph_hlen, pkt.cph_tlen); 760 /* XXX this channel is dead actually. */ 761 return (EIO); 762 } 763 764 hlen = VMBUS_CHANPKT_GETLEN(pkt.cph_hlen); 765 dlen = VMBUS_CHANPKT_GETLEN(pkt.cph_tlen) - hlen; 766 767 if (*dlen0 < dlen) { 768 /* Return the size of this packet's data. */ 769 *dlen0 = dlen; 770 return (ENOBUFS); 771 } 772 773 *xactid = pkt.cph_xactid; 774 *dlen0 = dlen; 775 776 /* Skip packet header */ 777 error = vmbus_rxbr_read(&chan->ch_rxbr, data, dlen, hlen); 778 KASSERT(!error, ("vmbus_rxbr_read failed")); 779 780 return (0); 781} 782 783int 784vmbus_chan_recv_pkt(struct vmbus_channel *chan, 785 struct vmbus_chanpkt_hdr *pkt0, int *pktlen0) 786{ 787 struct vmbus_chanpkt_hdr pkt; 788 int error, pktlen; 789 790 error = vmbus_rxbr_peek(&chan->ch_rxbr, &pkt, sizeof(pkt)); 791 if (error) 792 return (error); 793 794 if (__predict_false(pkt.cph_hlen < VMBUS_CHANPKT_HLEN_MIN)) { 795 device_printf(chan->ch_dev, "invalid hlen %u\n", 796 pkt.cph_hlen); 797 /* XXX this channel is dead actually. */ 798 return (EIO); 799 } 800 if (__predict_false(pkt.cph_hlen > pkt.cph_tlen)) { 801 device_printf(chan->ch_dev, "invalid hlen %u and tlen %u\n", 802 pkt.cph_hlen, pkt.cph_tlen); 803 /* XXX this channel is dead actually. */ 804 return (EIO); 805 } 806 807 pktlen = VMBUS_CHANPKT_GETLEN(pkt.cph_tlen); 808 if (*pktlen0 < pktlen) { 809 /* Return the size of this packet. */ 810 *pktlen0 = pktlen; 811 return (ENOBUFS); 812 } 813 *pktlen0 = pktlen; 814 815 /* Include packet header */ 816 error = vmbus_rxbr_read(&chan->ch_rxbr, pkt0, pktlen, 0); 817 KASSERT(!error, ("vmbus_rxbr_read failed")); 818 819 return (0); 820} 821 822static void 823vmbus_chan_task(void *xchan, int pending __unused) 824{ 825 struct vmbus_channel *chan = xchan; 826 vmbus_chan_callback_t cb = chan->ch_cb; 827 void *cbarg = chan->ch_cbarg; 828 829 /* 830 * Optimize host to guest signaling by ensuring: 831 * 1. While reading the channel, we disable interrupts from 832 * host. 833 * 2. Ensure that we process all posted messages from the host 834 * before returning from this callback. 835 * 3. Once we return, enable signaling from the host. Once this 836 * state is set we check to see if additional packets are 837 * available to read. In this case we repeat the process. 838 * 839 * NOTE: Interrupt has been disabled in the ISR. 840 */ 841 for (;;) { 842 uint32_t left; 843 844 cb(chan, cbarg); 845 846 left = vmbus_rxbr_intr_unmask(&chan->ch_rxbr); 847 if (left == 0) { 848 /* No more data in RX bufring; done */ 849 break; 850 } 851 vmbus_rxbr_intr_mask(&chan->ch_rxbr); 852 } 853} 854 855static void 856vmbus_chan_task_nobatch(void *xchan, int pending __unused) 857{ 858 struct vmbus_channel *chan = xchan; 859 860 chan->ch_cb(chan, chan->ch_cbarg); 861} 862 863static __inline void 864vmbus_event_flags_proc(struct vmbus_softc *sc, volatile u_long *event_flags, 865 int flag_cnt) 866{ 867 int f; 868 869 for (f = 0; f < flag_cnt; ++f) { 870 uint32_t chid_base; 871 u_long flags; 872 int chid_ofs; 873 874 if (event_flags[f] == 0) 875 continue; 876 877 flags = atomic_swap_long(&event_flags[f], 0); 878 chid_base = f << VMBUS_EVTFLAG_SHIFT; 879 880 while ((chid_ofs = ffsl(flags)) != 0) { 881 struct vmbus_channel *chan; 882 883 --chid_ofs; /* NOTE: ffsl is 1-based */ 884 flags &= ~(1UL << chid_ofs); 885 886 chan = sc->vmbus_chmap[chid_base + chid_ofs]; 887 888 /* if channel is closed or closing */ 889 if (chan == NULL || chan->ch_tq == NULL) 890 continue; 891 892 if (chan->ch_flags & VMBUS_CHAN_FLAG_BATCHREAD) 893 vmbus_rxbr_intr_mask(&chan->ch_rxbr); 894 taskqueue_enqueue(chan->ch_tq, &chan->ch_task); 895 } 896 } 897} 898 899void 900vmbus_event_proc(struct vmbus_softc *sc, int cpu) 901{ 902 struct vmbus_evtflags *eventf; 903 904 /* 905 * On Host with Win8 or above, the event page can be checked directly 906 * to get the id of the channel that has the pending interrupt. 907 */ 908 eventf = VMBUS_PCPU_GET(sc, event_flags, cpu) + VMBUS_SINT_MESSAGE; 909 vmbus_event_flags_proc(sc, eventf->evt_flags, 910 VMBUS_PCPU_GET(sc, event_flags_cnt, cpu)); 911} 912 913void 914vmbus_event_proc_compat(struct vmbus_softc *sc, int cpu) 915{ 916 struct vmbus_evtflags *eventf; 917 918 eventf = VMBUS_PCPU_GET(sc, event_flags, cpu) + VMBUS_SINT_MESSAGE; 919 if (atomic_testandclear_long(&eventf->evt_flags[0], 0)) { 920 vmbus_event_flags_proc(sc, sc->vmbus_rx_evtflags, 921 VMBUS_CHAN_MAX_COMPAT >> VMBUS_EVTFLAG_SHIFT); 922 } 923} 924 925static void 926vmbus_chan_update_evtflagcnt(struct vmbus_softc *sc, 927 const struct vmbus_channel *chan) 928{ 929 volatile int *flag_cnt_ptr; 930 int flag_cnt; 931 932 flag_cnt = (chan->ch_id / VMBUS_EVTFLAG_LEN) + 1; 933 flag_cnt_ptr = VMBUS_PCPU_PTR(sc, event_flags_cnt, chan->ch_cpuid); 934 935 for (;;) { 936 int old_flag_cnt; 937 938 old_flag_cnt = *flag_cnt_ptr; 939 if (old_flag_cnt >= flag_cnt) 940 break; 941 if (atomic_cmpset_int(flag_cnt_ptr, old_flag_cnt, flag_cnt)) { 942 if (bootverbose) { 943 device_printf(sc->vmbus_dev, 944 "channel%u update cpu%d flag_cnt to %d\n", 945 chan->ch_id, chan->ch_cpuid, flag_cnt); 946 } 947 break; 948 } 949 } 950} 951 952static struct vmbus_channel * 953vmbus_chan_alloc(struct vmbus_softc *sc) 954{ 955 struct vmbus_channel *chan; 956 957 chan = malloc(sizeof(*chan), M_DEVBUF, M_WAITOK | M_ZERO); 958 959 chan->ch_monprm = hyperv_dmamem_alloc(bus_get_dma_tag(sc->vmbus_dev), 960 HYPERCALL_PARAM_ALIGN, 0, sizeof(struct hyperv_mon_param), 961 &chan->ch_monprm_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO); 962 if (chan->ch_monprm == NULL) { 963 device_printf(sc->vmbus_dev, "monprm alloc failed\n"); 964 free(chan, M_DEVBUF); 965 return NULL; 966 } 967 968 chan->ch_vmbus = sc; 969 mtx_init(&chan->ch_subchan_lock, "vmbus subchan", NULL, MTX_DEF); 970 TAILQ_INIT(&chan->ch_subchans); 971 TASK_INIT(&chan->ch_detach_task, 0, vmbus_chan_detach_task, chan); 972 vmbus_rxbr_init(&chan->ch_rxbr); 973 vmbus_txbr_init(&chan->ch_txbr); 974 975 return chan; 976} 977 978static void 979vmbus_chan_free(struct vmbus_channel *chan) 980{ 981 /* TODO: assert sub-channel list is empty */ 982 /* TODO: asset no longer on the primary channel's sub-channel list */ 983 /* TODO: asset no longer on the vmbus channel list */ 984 hyperv_dmamem_free(&chan->ch_monprm_dma, chan->ch_monprm); 985 mtx_destroy(&chan->ch_subchan_lock); 986 vmbus_rxbr_deinit(&chan->ch_rxbr); 987 vmbus_txbr_deinit(&chan->ch_txbr); 988 free(chan, M_DEVBUF); 989} 990 991static int 992vmbus_chan_add(struct vmbus_channel *newchan) 993{ 994 struct vmbus_softc *sc = newchan->ch_vmbus; 995 struct vmbus_channel *prichan; 996 997 if (newchan->ch_id == 0) { 998 /* 999 * XXX 1000 * Chan0 will neither be processed nor should be offered; 1001 * skip it. 1002 */ 1003 device_printf(sc->vmbus_dev, "got chan0 offer, discard\n"); 1004 return EINVAL; 1005 } else if (newchan->ch_id >= VMBUS_CHAN_MAX) { 1006 device_printf(sc->vmbus_dev, "invalid chan%u offer\n", 1007 newchan->ch_id); 1008 return EINVAL; 1009 } 1010 sc->vmbus_chmap[newchan->ch_id] = newchan; 1011 1012 if (bootverbose) { 1013 device_printf(sc->vmbus_dev, "chan%u subidx%u offer\n", 1014 newchan->ch_id, newchan->ch_subidx); 1015 } 1016 1017 mtx_lock(&sc->vmbus_prichan_lock); 1018 TAILQ_FOREACH(prichan, &sc->vmbus_prichans, ch_prilink) { 1019 /* 1020 * Sub-channel will have the same type GUID and instance 1021 * GUID as its primary channel. 1022 */ 1023 if (memcmp(&prichan->ch_guid_type, &newchan->ch_guid_type, 1024 sizeof(struct hyperv_guid)) == 0 && 1025 memcmp(&prichan->ch_guid_inst, &newchan->ch_guid_inst, 1026 sizeof(struct hyperv_guid)) == 0) 1027 break; 1028 } 1029 if (VMBUS_CHAN_ISPRIMARY(newchan)) { 1030 if (prichan == NULL) { 1031 /* Install the new primary channel */ 1032 TAILQ_INSERT_TAIL(&sc->vmbus_prichans, newchan, 1033 ch_prilink); 1034 mtx_unlock(&sc->vmbus_prichan_lock); 1035 return 0; 1036 } else { 1037 mtx_unlock(&sc->vmbus_prichan_lock); 1038 device_printf(sc->vmbus_dev, "duplicated primary " 1039 "chan%u\n", newchan->ch_id); 1040 return EINVAL; 1041 } 1042 } else { /* Sub-channel */ 1043 if (prichan == NULL) { 1044 mtx_unlock(&sc->vmbus_prichan_lock); 1045 device_printf(sc->vmbus_dev, "no primary chan for " 1046 "chan%u\n", newchan->ch_id); 1047 return EINVAL; 1048 } 1049 /* 1050 * Found the primary channel for this sub-channel and 1051 * move on. 1052 * 1053 * XXX refcnt prichan 1054 */ 1055 } 1056 mtx_unlock(&sc->vmbus_prichan_lock); 1057 1058 /* 1059 * This is a sub-channel; link it with the primary channel. 1060 */ 1061 KASSERT(!VMBUS_CHAN_ISPRIMARY(newchan), 1062 ("new channel is not sub-channel")); 1063 KASSERT(prichan != NULL, ("no primary channel")); 1064 1065 newchan->ch_prichan = prichan; 1066 newchan->ch_dev = prichan->ch_dev; 1067 1068 mtx_lock(&prichan->ch_subchan_lock); 1069 TAILQ_INSERT_TAIL(&prichan->ch_subchans, newchan, ch_sublink); 1070 /* 1071 * Bump up sub-channel count and notify anyone that is 1072 * interested in this sub-channel, after this sub-channel 1073 * is setup. 1074 */ 1075 prichan->ch_subchan_cnt++; 1076 mtx_unlock(&prichan->ch_subchan_lock); 1077 wakeup(prichan); 1078 1079 return 0; 1080} 1081 1082void 1083vmbus_chan_cpu_set(struct vmbus_channel *chan, int cpu) 1084{ 1085 KASSERT(cpu >= 0 && cpu < mp_ncpus, ("invalid cpu %d", cpu)); 1086 1087 if (chan->ch_vmbus->vmbus_version == VMBUS_VERSION_WS2008 || 1088 chan->ch_vmbus->vmbus_version == VMBUS_VERSION_WIN7) { 1089 /* Only cpu0 is supported */ 1090 cpu = 0; 1091 } 1092 1093 chan->ch_cpuid = cpu; 1094 chan->ch_vcpuid = VMBUS_PCPU_GET(chan->ch_vmbus, vcpuid, cpu); 1095 1096 if (bootverbose) { 1097 printf("vmbus_chan%u: assigned to cpu%u [vcpu%u]\n", 1098 chan->ch_id, chan->ch_cpuid, chan->ch_vcpuid); 1099 } 1100} 1101 1102void 1103vmbus_chan_cpu_rr(struct vmbus_channel *chan) 1104{ 1105 static uint32_t vmbus_chan_nextcpu; 1106 int cpu; 1107 1108 cpu = atomic_fetchadd_int(&vmbus_chan_nextcpu, 1) % mp_ncpus; 1109 vmbus_chan_cpu_set(chan, cpu); 1110} 1111 1112static void 1113vmbus_chan_cpu_default(struct vmbus_channel *chan) 1114{ 1115 /* 1116 * By default, pin the channel to cpu0. Devices having 1117 * special channel-cpu mapping requirement should call 1118 * vmbus_chan_cpu_{set,rr}(). 1119 */ 1120 vmbus_chan_cpu_set(chan, 0); 1121} 1122 1123static void 1124vmbus_chan_msgproc_choffer(struct vmbus_softc *sc, 1125 const struct vmbus_message *msg) 1126{ 1127 const struct vmbus_chanmsg_choffer *offer; 1128 struct vmbus_channel *chan; 1129 int error; 1130 1131 offer = (const struct vmbus_chanmsg_choffer *)msg->msg_data; 1132 1133 chan = vmbus_chan_alloc(sc); 1134 if (chan == NULL) { 1135 device_printf(sc->vmbus_dev, "allocate chan%u failed\n", 1136 offer->chm_chanid); 1137 return; 1138 } 1139 1140 chan->ch_id = offer->chm_chanid; 1141 chan->ch_subidx = offer->chm_subidx; 1142 chan->ch_guid_type = offer->chm_chtype; 1143 chan->ch_guid_inst = offer->chm_chinst; 1144 1145 /* Batch reading is on by default */ 1146 chan->ch_flags |= VMBUS_CHAN_FLAG_BATCHREAD; 1147 1148 chan->ch_monprm->mp_connid = VMBUS_CONNID_EVENT; 1149 if (sc->vmbus_version != VMBUS_VERSION_WS2008) 1150 chan->ch_monprm->mp_connid = offer->chm_connid; 1151 1152 if (offer->chm_flags1 & VMBUS_CHOFFER_FLAG1_HASMNF) { 1153 int trig_idx; 1154 1155 /* 1156 * Setup MNF stuffs. 1157 */ 1158 chan->ch_txflags |= VMBUS_CHAN_TXF_HASMNF; 1159 1160 trig_idx = offer->chm_montrig / VMBUS_MONTRIG_LEN; 1161 if (trig_idx >= VMBUS_MONTRIGS_MAX) 1162 panic("invalid monitor trigger %u", offer->chm_montrig); 1163 chan->ch_montrig = 1164 &sc->vmbus_mnf2->mnf_trigs[trig_idx].mt_pending; 1165 1166 chan->ch_montrig_mask = 1167 1 << (offer->chm_montrig % VMBUS_MONTRIG_LEN); 1168 } 1169 1170 /* 1171 * Setup event flag. 1172 */ 1173 chan->ch_evtflag = 1174 &sc->vmbus_tx_evtflags[chan->ch_id >> VMBUS_EVTFLAG_SHIFT]; 1175 chan->ch_evtflag_mask = 1UL << (chan->ch_id & VMBUS_EVTFLAG_MASK); 1176 1177 /* Select default cpu for this channel. */ 1178 vmbus_chan_cpu_default(chan); 1179 1180 error = vmbus_chan_add(chan); 1181 if (error) { 1182 device_printf(sc->vmbus_dev, "add chan%u failed: %d\n", 1183 chan->ch_id, error); 1184 vmbus_chan_free(chan); 1185 return; 1186 } 1187 1188 if (VMBUS_CHAN_ISPRIMARY(chan)) { 1189 /* 1190 * Add device for this primary channel. 1191 * 1192 * NOTE: 1193 * Error is ignored here; don't have much to do if error 1194 * really happens. 1195 */ 1196 vmbus_add_child(chan); 1197 } 1198} 1199 1200/* 1201 * XXX pretty broken; need rework. 1202 */ 1203static void 1204vmbus_chan_msgproc_chrescind(struct vmbus_softc *sc, 1205 const struct vmbus_message *msg) 1206{ 1207 const struct vmbus_chanmsg_chrescind *note; 1208 struct vmbus_channel *chan; 1209 1210 note = (const struct vmbus_chanmsg_chrescind *)msg->msg_data; 1211 if (note->chm_chanid > VMBUS_CHAN_MAX) { 1212 device_printf(sc->vmbus_dev, "invalid rescinded chan%u\n", 1213 note->chm_chanid); 1214 return; 1215 } 1216 1217 if (bootverbose) { 1218 device_printf(sc->vmbus_dev, "chan%u rescinded\n", 1219 note->chm_chanid); 1220 } 1221 1222 chan = sc->vmbus_chmap[note->chm_chanid]; 1223 if (chan == NULL) 1224 return; 1225 sc->vmbus_chmap[note->chm_chanid] = NULL; 1226 1227 taskqueue_enqueue(taskqueue_thread, &chan->ch_detach_task); 1228} 1229 1230static void 1231vmbus_chan_detach_task(void *xchan, int pending __unused) 1232{ 1233 struct vmbus_channel *chan = xchan; 1234 1235 if (VMBUS_CHAN_ISPRIMARY(chan)) { 1236 /* Only primary channel owns the device */ 1237 vmbus_delete_child(chan); 1238 /* NOTE: DO NOT free primary channel for now */ 1239 } else { 1240 struct vmbus_softc *sc = chan->ch_vmbus; 1241 struct vmbus_channel *pri_chan = chan->ch_prichan; 1242 struct vmbus_chanmsg_chfree *req; 1243 struct vmbus_msghc *mh; 1244 int error; 1245 1246 mh = vmbus_msghc_get(sc, sizeof(*req)); 1247 if (mh == NULL) { 1248 device_printf(sc->vmbus_dev, 1249 "can not get msg hypercall for chfree(chan%u)\n", 1250 chan->ch_id); 1251 goto remove; 1252 } 1253 1254 req = vmbus_msghc_dataptr(mh); 1255 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHFREE; 1256 req->chm_chanid = chan->ch_id; 1257 1258 error = vmbus_msghc_exec_noresult(mh); 1259 vmbus_msghc_put(sc, mh); 1260 1261 if (error) { 1262 device_printf(sc->vmbus_dev, 1263 "chfree(chan%u) failed: %d", 1264 chan->ch_id, error); 1265 /* NOTE: Move on! */ 1266 } else { 1267 if (bootverbose) { 1268 device_printf(sc->vmbus_dev, "chan%u freed\n", 1269 chan->ch_id); 1270 } 1271 } 1272remove: 1273 mtx_lock(&pri_chan->ch_subchan_lock); 1274 TAILQ_REMOVE(&pri_chan->ch_subchans, chan, ch_sublink); 1275 KASSERT(pri_chan->ch_subchan_cnt > 0, 1276 ("invalid subchan_cnt %d", pri_chan->ch_subchan_cnt)); 1277 pri_chan->ch_subchan_cnt--; 1278 mtx_unlock(&pri_chan->ch_subchan_lock); 1279 wakeup(pri_chan); 1280 1281 vmbus_chan_free(chan); 1282 } 1283} 1284 1285/* 1286 * Detach all devices and destroy the corresponding primary channels. 1287 */ 1288void 1289vmbus_chan_destroy_all(struct vmbus_softc *sc) 1290{ 1291 struct vmbus_channel *chan; 1292 1293 mtx_lock(&sc->vmbus_prichan_lock); 1294 while ((chan = TAILQ_FIRST(&sc->vmbus_prichans)) != NULL) { 1295 KASSERT(VMBUS_CHAN_ISPRIMARY(chan), ("not primary channel")); 1296 TAILQ_REMOVE(&sc->vmbus_prichans, chan, ch_prilink); 1297 mtx_unlock(&sc->vmbus_prichan_lock); 1298 1299 vmbus_delete_child(chan); 1300 vmbus_chan_free(chan); 1301 1302 mtx_lock(&sc->vmbus_prichan_lock); 1303 } 1304 bzero(sc->vmbus_chmap, 1305 sizeof(struct vmbus_channel *) * VMBUS_CHAN_MAX); 1306 mtx_unlock(&sc->vmbus_prichan_lock); 1307} 1308 1309/* 1310 * The channel whose vcpu binding is closest to the currect vcpu will 1311 * be selected. 1312 * If no multi-channel, always select primary channel. 1313 */ 1314struct vmbus_channel * 1315vmbus_chan_cpu2chan(struct vmbus_channel *prichan, int cpu) 1316{ 1317 struct vmbus_channel *sel, *chan; 1318 uint32_t vcpu, sel_dist; 1319 1320 KASSERT(cpu >= 0 && cpu < mp_ncpus, ("invalid cpuid %d", cpu)); 1321 if (TAILQ_EMPTY(&prichan->ch_subchans)) 1322 return prichan; 1323 1324 vcpu = VMBUS_PCPU_GET(prichan->ch_vmbus, vcpuid, cpu); 1325 1326#define CHAN_VCPU_DIST(ch, vcpu) \ 1327 (((ch)->ch_vcpuid > (vcpu)) ? \ 1328 ((ch)->ch_vcpuid - (vcpu)) : ((vcpu) - (ch)->ch_vcpuid)) 1329 1330#define CHAN_SELECT(ch) \ 1331do { \ 1332 sel = ch; \ 1333 sel_dist = CHAN_VCPU_DIST(ch, vcpu); \ 1334} while (0) 1335 1336 CHAN_SELECT(prichan); 1337 1338 mtx_lock(&prichan->ch_subchan_lock); 1339 TAILQ_FOREACH(chan, &prichan->ch_subchans, ch_sublink) { 1340 uint32_t dist; 1341 1342 KASSERT(chan->ch_stflags & VMBUS_CHAN_ST_OPENED, 1343 ("chan%u is not opened", chan->ch_id)); 1344 1345 if (chan->ch_vcpuid == vcpu) { 1346 /* Exact match; done */ 1347 CHAN_SELECT(chan); 1348 break; 1349 } 1350 1351 dist = CHAN_VCPU_DIST(chan, vcpu); 1352 if (sel_dist <= dist) { 1353 /* Far or same distance; skip */ 1354 continue; 1355 } 1356 1357 /* Select the closer channel. */ 1358 CHAN_SELECT(chan); 1359 } 1360 mtx_unlock(&prichan->ch_subchan_lock); 1361 1362#undef CHAN_SELECT 1363#undef CHAN_VCPU_DIST 1364 1365 return sel; 1366} 1367 1368struct vmbus_channel ** 1369vmbus_subchan_get(struct vmbus_channel *pri_chan, int subchan_cnt) 1370{ 1371 struct vmbus_channel **ret, *chan; 1372 int i; 1373 1374 KASSERT(subchan_cnt > 0, ("invalid sub-channel count %d", subchan_cnt)); 1375 1376 ret = malloc(subchan_cnt * sizeof(struct vmbus_channel *), M_TEMP, 1377 M_WAITOK); 1378 1379 mtx_lock(&pri_chan->ch_subchan_lock); 1380 1381 while (pri_chan->ch_subchan_cnt < subchan_cnt) 1382 mtx_sleep(pri_chan, &pri_chan->ch_subchan_lock, 0, "subch", 0); 1383 1384 i = 0; 1385 TAILQ_FOREACH(chan, &pri_chan->ch_subchans, ch_sublink) { 1386 /* TODO: refcnt chan */ 1387 ret[i] = chan; 1388 1389 ++i; 1390 if (i == subchan_cnt) 1391 break; 1392 } 1393 KASSERT(i == subchan_cnt, ("invalid subchan count %d, should be %d", 1394 pri_chan->ch_subchan_cnt, subchan_cnt)); 1395 1396 mtx_unlock(&pri_chan->ch_subchan_lock); 1397 1398 return ret; 1399} 1400 1401void 1402vmbus_subchan_rel(struct vmbus_channel **subchan, int subchan_cnt __unused) 1403{ 1404 1405 free(subchan, M_TEMP); 1406} 1407 1408void 1409vmbus_subchan_drain(struct vmbus_channel *pri_chan) 1410{ 1411 mtx_lock(&pri_chan->ch_subchan_lock); 1412 while (pri_chan->ch_subchan_cnt > 0) 1413 mtx_sleep(pri_chan, &pri_chan->ch_subchan_lock, 0, "dsubch", 0); 1414 mtx_unlock(&pri_chan->ch_subchan_lock); 1415} 1416 1417void 1418vmbus_chan_msgproc(struct vmbus_softc *sc, const struct vmbus_message *msg) 1419{ 1420 vmbus_chanmsg_proc_t msg_proc; 1421 uint32_t msg_type; 1422 1423 msg_type = ((const struct vmbus_chanmsg_hdr *)msg->msg_data)->chm_type; 1424 KASSERT(msg_type < VMBUS_CHANMSG_TYPE_MAX, 1425 ("invalid message type %u", msg_type)); 1426 1427 msg_proc = vmbus_chan_msgprocs[msg_type]; 1428 if (msg_proc != NULL) 1429 msg_proc(sc, msg); 1430} 1431 1432void 1433vmbus_chan_set_readbatch(struct vmbus_channel *chan, bool on) 1434{ 1435 if (!on) 1436 chan->ch_flags &= ~VMBUS_CHAN_FLAG_BATCHREAD; 1437 else 1438 chan->ch_flags |= VMBUS_CHAN_FLAG_BATCHREAD; 1439} 1440 1441uint32_t 1442vmbus_chan_id(const struct vmbus_channel *chan) 1443{ 1444 return chan->ch_id; 1445} 1446 1447uint32_t 1448vmbus_chan_subidx(const struct vmbus_channel *chan) 1449{ 1450 return chan->ch_subidx; 1451} 1452 1453bool 1454vmbus_chan_is_primary(const struct vmbus_channel *chan) 1455{ 1456 if (VMBUS_CHAN_ISPRIMARY(chan)) 1457 return true; 1458 else 1459 return false; 1460} 1461 1462const struct hyperv_guid * 1463vmbus_chan_guid_inst(const struct vmbus_channel *chan) 1464{ 1465 return &chan->ch_guid_inst; 1466} 1467 1468int 1469vmbus_chan_prplist_nelem(int br_size, int prpcnt_max, int dlen_max) 1470{ 1471 int elem_size; 1472 1473 elem_size = __offsetof(struct vmbus_chanpkt_prplist, 1474 cp_range[0].gpa_page[prpcnt_max]); 1475 elem_size += dlen_max; 1476 elem_size = VMBUS_CHANPKT_TOTLEN(elem_size); 1477 1478 return (vmbus_br_nelem(br_size, elem_size)); 1479} 1480