vmbus.c revision 307164
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/* 30 * VM Bus Driver Implementation 31 */ 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: stable/10/sys/dev/hyperv/vmbus/vmbus.c 307164 2016-10-13 02:28:40Z sephe $"); 34 35#include <sys/param.h> 36#include <sys/bus.h> 37#include <sys/kernel.h> 38#include <sys/lock.h> 39#include <sys/malloc.h> 40#include <sys/module.h> 41#include <sys/mutex.h> 42#include <sys/proc.h> 43#include <sys/smp.h> 44#include <sys/sysctl.h> 45#include <sys/systm.h> 46#include <sys/taskqueue.h> 47 48#include <machine/intr_machdep.h> 49#include <machine/apicvar.h> 50#include <machine/md_var.h> 51 52#include <contrib/dev/acpica/include/acpi.h> 53 54#include <dev/hyperv/include/hyperv.h> 55#include <dev/hyperv/include/vmbus_xact.h> 56#include <dev/hyperv/vmbus/hyperv_reg.h> 57#include <dev/hyperv/vmbus/hyperv_var.h> 58#include <dev/hyperv/vmbus/vmbus_reg.h> 59#include <dev/hyperv/vmbus/vmbus_var.h> 60#include <dev/hyperv/vmbus/vmbus_chanvar.h> 61 62#include "acpi_if.h" 63#include "vmbus_if.h" 64 65#define VMBUS_GPADL_START 0xe1e10 66 67struct vmbus_msghc { 68 struct vmbus_xact *mh_xact; 69 struct hypercall_postmsg_in mh_inprm_save; 70}; 71 72static int vmbus_probe(device_t); 73static int vmbus_attach(device_t); 74static int vmbus_detach(device_t); 75static int vmbus_read_ivar(device_t, device_t, int, 76 uintptr_t *); 77static int vmbus_child_pnpinfo_str(device_t, device_t, 78 char *, size_t); 79static uint32_t vmbus_get_version_method(device_t, device_t); 80static int vmbus_probe_guid_method(device_t, device_t, 81 const struct hyperv_guid *); 82 83static int vmbus_init(struct vmbus_softc *); 84static int vmbus_connect(struct vmbus_softc *, uint32_t); 85static int vmbus_req_channels(struct vmbus_softc *sc); 86static void vmbus_disconnect(struct vmbus_softc *); 87static int vmbus_scan(struct vmbus_softc *); 88static void vmbus_scan_wait(struct vmbus_softc *); 89static void vmbus_scan_newchan(struct vmbus_softc *); 90static void vmbus_scan_newdev(struct vmbus_softc *); 91static void vmbus_scan_done(struct vmbus_softc *, 92 const struct vmbus_message *); 93static void vmbus_chanmsg_handle(struct vmbus_softc *, 94 const struct vmbus_message *); 95static void vmbus_msg_task(void *, int); 96static void vmbus_synic_setup(void *); 97static void vmbus_synic_teardown(void *); 98static int vmbus_sysctl_version(SYSCTL_HANDLER_ARGS); 99static int vmbus_dma_alloc(struct vmbus_softc *); 100static void vmbus_dma_free(struct vmbus_softc *); 101static int vmbus_intr_setup(struct vmbus_softc *); 102static void vmbus_intr_teardown(struct vmbus_softc *); 103static int vmbus_doattach(struct vmbus_softc *); 104static void vmbus_event_proc_dummy(struct vmbus_softc *, 105 int); 106 107static struct vmbus_softc *vmbus_sc; 108 109extern inthand_t IDTVEC(rsvd), IDTVEC(vmbus_isr); 110 111static const uint32_t vmbus_version[] = { 112 VMBUS_VERSION_WIN8_1, 113 VMBUS_VERSION_WIN8, 114 VMBUS_VERSION_WIN7, 115 VMBUS_VERSION_WS2008 116}; 117 118static const vmbus_chanmsg_proc_t 119vmbus_chanmsg_handlers[VMBUS_CHANMSG_TYPE_MAX] = { 120 VMBUS_CHANMSG_PROC(CHOFFER_DONE, vmbus_scan_done), 121 VMBUS_CHANMSG_PROC_WAKEUP(CONNECT_RESP) 122}; 123 124static device_method_t vmbus_methods[] = { 125 /* Device interface */ 126 DEVMETHOD(device_probe, vmbus_probe), 127 DEVMETHOD(device_attach, vmbus_attach), 128 DEVMETHOD(device_detach, vmbus_detach), 129 DEVMETHOD(device_shutdown, bus_generic_shutdown), 130 DEVMETHOD(device_suspend, bus_generic_suspend), 131 DEVMETHOD(device_resume, bus_generic_resume), 132 133 /* Bus interface */ 134 DEVMETHOD(bus_add_child, bus_generic_add_child), 135 DEVMETHOD(bus_print_child, bus_generic_print_child), 136 DEVMETHOD(bus_read_ivar, vmbus_read_ivar), 137 DEVMETHOD(bus_child_pnpinfo_str, vmbus_child_pnpinfo_str), 138 139 /* Vmbus interface */ 140 DEVMETHOD(vmbus_get_version, vmbus_get_version_method), 141 DEVMETHOD(vmbus_probe_guid, vmbus_probe_guid_method), 142 143 DEVMETHOD_END 144}; 145 146static driver_t vmbus_driver = { 147 "vmbus", 148 vmbus_methods, 149 sizeof(struct vmbus_softc) 150}; 151 152static devclass_t vmbus_devclass; 153 154DRIVER_MODULE(vmbus, acpi, vmbus_driver, vmbus_devclass, NULL, NULL); 155MODULE_DEPEND(vmbus, acpi, 1, 1, 1); 156MODULE_VERSION(vmbus, 1); 157 158static __inline struct vmbus_softc * 159vmbus_get_softc(void) 160{ 161 return vmbus_sc; 162} 163 164void 165vmbus_msghc_reset(struct vmbus_msghc *mh, size_t dsize) 166{ 167 struct hypercall_postmsg_in *inprm; 168 169 if (dsize > HYPERCALL_POSTMSGIN_DSIZE_MAX) 170 panic("invalid data size %zu", dsize); 171 172 inprm = vmbus_xact_req_data(mh->mh_xact); 173 memset(inprm, 0, HYPERCALL_POSTMSGIN_SIZE); 174 inprm->hc_connid = VMBUS_CONNID_MESSAGE; 175 inprm->hc_msgtype = HYPERV_MSGTYPE_CHANNEL; 176 inprm->hc_dsize = dsize; 177} 178 179struct vmbus_msghc * 180vmbus_msghc_get(struct vmbus_softc *sc, size_t dsize) 181{ 182 struct vmbus_msghc *mh; 183 struct vmbus_xact *xact; 184 185 if (dsize > HYPERCALL_POSTMSGIN_DSIZE_MAX) 186 panic("invalid data size %zu", dsize); 187 188 xact = vmbus_xact_get(sc->vmbus_xc, 189 dsize + __offsetof(struct hypercall_postmsg_in, hc_data[0])); 190 if (xact == NULL) 191 return (NULL); 192 193 mh = vmbus_xact_priv(xact, sizeof(*mh)); 194 mh->mh_xact = xact; 195 196 vmbus_msghc_reset(mh, dsize); 197 return (mh); 198} 199 200void 201vmbus_msghc_put(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh) 202{ 203 204 vmbus_xact_put(mh->mh_xact); 205} 206 207void * 208vmbus_msghc_dataptr(struct vmbus_msghc *mh) 209{ 210 struct hypercall_postmsg_in *inprm; 211 212 inprm = vmbus_xact_req_data(mh->mh_xact); 213 return (inprm->hc_data); 214} 215 216int 217vmbus_msghc_exec_noresult(struct vmbus_msghc *mh) 218{ 219 sbintime_t time = SBT_1MS; 220 struct hypercall_postmsg_in *inprm; 221 bus_addr_t inprm_paddr; 222 int i; 223 224 inprm = vmbus_xact_req_data(mh->mh_xact); 225 inprm_paddr = vmbus_xact_req_paddr(mh->mh_xact); 226 227 /* 228 * Save the input parameter so that we could restore the input 229 * parameter if the Hypercall failed. 230 * 231 * XXX 232 * Is this really necessary?! i.e. Will the Hypercall ever 233 * overwrite the input parameter? 234 */ 235 memcpy(&mh->mh_inprm_save, inprm, HYPERCALL_POSTMSGIN_SIZE); 236 237 /* 238 * In order to cope with transient failures, e.g. insufficient 239 * resources on host side, we retry the post message Hypercall 240 * several times. 20 retries seem sufficient. 241 */ 242#define HC_RETRY_MAX 20 243 244 for (i = 0; i < HC_RETRY_MAX; ++i) { 245 uint64_t status; 246 247 status = hypercall_post_message(inprm_paddr); 248 if (status == HYPERCALL_STATUS_SUCCESS) 249 return 0; 250 251 pause_sbt("hcpmsg", time, 0, C_HARDCLOCK); 252 if (time < SBT_1S * 2) 253 time *= 2; 254 255 /* Restore input parameter and try again */ 256 memcpy(inprm, &mh->mh_inprm_save, HYPERCALL_POSTMSGIN_SIZE); 257 } 258 259#undef HC_RETRY_MAX 260 261 return EIO; 262} 263 264int 265vmbus_msghc_exec(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh) 266{ 267 int error; 268 269 vmbus_xact_activate(mh->mh_xact); 270 error = vmbus_msghc_exec_noresult(mh); 271 if (error) 272 vmbus_xact_deactivate(mh->mh_xact); 273 return error; 274} 275 276const struct vmbus_message * 277vmbus_msghc_wait_result(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh) 278{ 279 size_t resp_len; 280 281 return (vmbus_xact_wait(mh->mh_xact, &resp_len)); 282} 283 284void 285vmbus_msghc_wakeup(struct vmbus_softc *sc, const struct vmbus_message *msg) 286{ 287 288 vmbus_xact_ctx_wakeup(sc->vmbus_xc, msg, sizeof(*msg)); 289} 290 291uint32_t 292vmbus_gpadl_alloc(struct vmbus_softc *sc) 293{ 294 return atomic_fetchadd_int(&sc->vmbus_gpadl, 1); 295} 296 297static int 298vmbus_connect(struct vmbus_softc *sc, uint32_t version) 299{ 300 struct vmbus_chanmsg_connect *req; 301 const struct vmbus_message *msg; 302 struct vmbus_msghc *mh; 303 int error, done = 0; 304 305 mh = vmbus_msghc_get(sc, sizeof(*req)); 306 if (mh == NULL) 307 return ENXIO; 308 309 req = vmbus_msghc_dataptr(mh); 310 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CONNECT; 311 req->chm_ver = version; 312 req->chm_evtflags = sc->vmbus_evtflags_dma.hv_paddr; 313 req->chm_mnf1 = sc->vmbus_mnf1_dma.hv_paddr; 314 req->chm_mnf2 = sc->vmbus_mnf2_dma.hv_paddr; 315 316 error = vmbus_msghc_exec(sc, mh); 317 if (error) { 318 vmbus_msghc_put(sc, mh); 319 return error; 320 } 321 322 msg = vmbus_msghc_wait_result(sc, mh); 323 done = ((const struct vmbus_chanmsg_connect_resp *) 324 msg->msg_data)->chm_done; 325 326 vmbus_msghc_put(sc, mh); 327 328 return (done ? 0 : EOPNOTSUPP); 329} 330 331static int 332vmbus_init(struct vmbus_softc *sc) 333{ 334 int i; 335 336 for (i = 0; i < nitems(vmbus_version); ++i) { 337 int error; 338 339 error = vmbus_connect(sc, vmbus_version[i]); 340 if (!error) { 341 sc->vmbus_version = vmbus_version[i]; 342 device_printf(sc->vmbus_dev, "version %u.%u\n", 343 VMBUS_VERSION_MAJOR(sc->vmbus_version), 344 VMBUS_VERSION_MINOR(sc->vmbus_version)); 345 return 0; 346 } 347 } 348 return ENXIO; 349} 350 351static void 352vmbus_disconnect(struct vmbus_softc *sc) 353{ 354 struct vmbus_chanmsg_disconnect *req; 355 struct vmbus_msghc *mh; 356 int error; 357 358 mh = vmbus_msghc_get(sc, sizeof(*req)); 359 if (mh == NULL) { 360 device_printf(sc->vmbus_dev, 361 "can not get msg hypercall for disconnect\n"); 362 return; 363 } 364 365 req = vmbus_msghc_dataptr(mh); 366 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_DISCONNECT; 367 368 error = vmbus_msghc_exec_noresult(mh); 369 vmbus_msghc_put(sc, mh); 370 371 if (error) { 372 device_printf(sc->vmbus_dev, 373 "disconnect msg hypercall failed\n"); 374 } 375} 376 377static int 378vmbus_req_channels(struct vmbus_softc *sc) 379{ 380 struct vmbus_chanmsg_chrequest *req; 381 struct vmbus_msghc *mh; 382 int error; 383 384 mh = vmbus_msghc_get(sc, sizeof(*req)); 385 if (mh == NULL) 386 return ENXIO; 387 388 req = vmbus_msghc_dataptr(mh); 389 req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHREQUEST; 390 391 error = vmbus_msghc_exec_noresult(mh); 392 vmbus_msghc_put(sc, mh); 393 394 return error; 395} 396 397static void 398vmbus_scan_newchan(struct vmbus_softc *sc) 399{ 400 mtx_lock(&sc->vmbus_scan_lock); 401 if ((sc->vmbus_scan_chcnt & VMBUS_SCAN_CHCNT_DONE) == 0) 402 sc->vmbus_scan_chcnt++; 403 mtx_unlock(&sc->vmbus_scan_lock); 404} 405 406static void 407vmbus_scan_done(struct vmbus_softc *sc, 408 const struct vmbus_message *msg __unused) 409{ 410 mtx_lock(&sc->vmbus_scan_lock); 411 sc->vmbus_scan_chcnt |= VMBUS_SCAN_CHCNT_DONE; 412 mtx_unlock(&sc->vmbus_scan_lock); 413 wakeup(&sc->vmbus_scan_chcnt); 414} 415 416static void 417vmbus_scan_newdev(struct vmbus_softc *sc) 418{ 419 mtx_lock(&sc->vmbus_scan_lock); 420 sc->vmbus_scan_devcnt++; 421 mtx_unlock(&sc->vmbus_scan_lock); 422 wakeup(&sc->vmbus_scan_devcnt); 423} 424 425static void 426vmbus_scan_wait(struct vmbus_softc *sc) 427{ 428 uint32_t chancnt; 429 430 mtx_lock(&sc->vmbus_scan_lock); 431 while ((sc->vmbus_scan_chcnt & VMBUS_SCAN_CHCNT_DONE) == 0) { 432 mtx_sleep(&sc->vmbus_scan_chcnt, &sc->vmbus_scan_lock, 0, 433 "waitch", 0); 434 } 435 chancnt = sc->vmbus_scan_chcnt & ~VMBUS_SCAN_CHCNT_DONE; 436 437 while (sc->vmbus_scan_devcnt != chancnt) { 438 mtx_sleep(&sc->vmbus_scan_devcnt, &sc->vmbus_scan_lock, 0, 439 "waitdev", 0); 440 } 441 mtx_unlock(&sc->vmbus_scan_lock); 442} 443 444static int 445vmbus_scan(struct vmbus_softc *sc) 446{ 447 int error; 448 449 /* 450 * Start vmbus scanning. 451 */ 452 error = vmbus_req_channels(sc); 453 if (error) { 454 device_printf(sc->vmbus_dev, "channel request failed: %d\n", 455 error); 456 return error; 457 } 458 459 /* 460 * Wait for all devices are added to vmbus. 461 */ 462 vmbus_scan_wait(sc); 463 464 /* 465 * Identify, probe and attach. 466 */ 467 bus_generic_probe(sc->vmbus_dev); 468 bus_generic_attach(sc->vmbus_dev); 469 470 if (bootverbose) { 471 device_printf(sc->vmbus_dev, "device scan, probe and attach " 472 "done\n"); 473 } 474 return 0; 475} 476 477static void 478vmbus_chanmsg_handle(struct vmbus_softc *sc, const struct vmbus_message *msg) 479{ 480 vmbus_chanmsg_proc_t msg_proc; 481 uint32_t msg_type; 482 483 msg_type = ((const struct vmbus_chanmsg_hdr *)msg->msg_data)->chm_type; 484 if (msg_type >= VMBUS_CHANMSG_TYPE_MAX) { 485 device_printf(sc->vmbus_dev, "unknown message type 0x%x\n", 486 msg_type); 487 return; 488 } 489 490 msg_proc = vmbus_chanmsg_handlers[msg_type]; 491 if (msg_proc != NULL) 492 msg_proc(sc, msg); 493 494 /* Channel specific processing */ 495 vmbus_chan_msgproc(sc, msg); 496} 497 498static void 499vmbus_msg_task(void *xsc, int pending __unused) 500{ 501 struct vmbus_softc *sc = xsc; 502 volatile struct vmbus_message *msg; 503 504 msg = VMBUS_PCPU_GET(sc, message, curcpu) + VMBUS_SINT_MESSAGE; 505 for (;;) { 506 if (msg->msg_type == HYPERV_MSGTYPE_NONE) { 507 /* No message */ 508 break; 509 } else if (msg->msg_type == HYPERV_MSGTYPE_CHANNEL) { 510 /* Channel message */ 511 vmbus_chanmsg_handle(sc, 512 __DEVOLATILE(const struct vmbus_message *, msg)); 513 } 514 515 msg->msg_type = HYPERV_MSGTYPE_NONE; 516 /* 517 * Make sure the write to msg_type (i.e. set to 518 * HYPERV_MSGTYPE_NONE) happens before we read the 519 * msg_flags and EOMing. Otherwise, the EOMing will 520 * not deliver any more messages since there is no 521 * empty slot 522 * 523 * NOTE: 524 * mb() is used here, since atomic_thread_fence_seq_cst() 525 * will become compiler fence on UP kernel. 526 */ 527 mb(); 528 if (msg->msg_flags & VMBUS_MSGFLAG_PENDING) { 529 /* 530 * This will cause message queue rescan to possibly 531 * deliver another msg from the hypervisor 532 */ 533 wrmsr(MSR_HV_EOM, 0); 534 } 535 } 536} 537 538static __inline int 539vmbus_handle_intr1(struct vmbus_softc *sc, struct trapframe *frame, int cpu) 540{ 541 volatile struct vmbus_message *msg; 542 struct vmbus_message *msg_base; 543 544 msg_base = VMBUS_PCPU_GET(sc, message, cpu); 545 546 /* 547 * Check event timer. 548 * 549 * TODO: move this to independent IDT vector. 550 */ 551 msg = msg_base + VMBUS_SINT_TIMER; 552 if (msg->msg_type == HYPERV_MSGTYPE_TIMER_EXPIRED) { 553 msg->msg_type = HYPERV_MSGTYPE_NONE; 554 555 vmbus_et_intr(frame); 556 557 /* 558 * Make sure the write to msg_type (i.e. set to 559 * HYPERV_MSGTYPE_NONE) happens before we read the 560 * msg_flags and EOMing. Otherwise, the EOMing will 561 * not deliver any more messages since there is no 562 * empty slot 563 * 564 * NOTE: 565 * mb() is used here, since atomic_thread_fence_seq_cst() 566 * will become compiler fence on UP kernel. 567 */ 568 mb(); 569 if (msg->msg_flags & VMBUS_MSGFLAG_PENDING) { 570 /* 571 * This will cause message queue rescan to possibly 572 * deliver another msg from the hypervisor 573 */ 574 wrmsr(MSR_HV_EOM, 0); 575 } 576 } 577 578 /* 579 * Check events. Hot path for network and storage I/O data; high rate. 580 * 581 * NOTE: 582 * As recommended by the Windows guest fellows, we check events before 583 * checking messages. 584 */ 585 sc->vmbus_event_proc(sc, cpu); 586 587 /* 588 * Check messages. Mainly management stuffs; ultra low rate. 589 */ 590 msg = msg_base + VMBUS_SINT_MESSAGE; 591 if (__predict_false(msg->msg_type != HYPERV_MSGTYPE_NONE)) { 592 taskqueue_enqueue(VMBUS_PCPU_GET(sc, message_tq, cpu), 593 VMBUS_PCPU_PTR(sc, message_task, cpu)); 594 } 595 596 return (FILTER_HANDLED); 597} 598 599void 600vmbus_handle_intr(struct trapframe *trap_frame) 601{ 602 struct vmbus_softc *sc = vmbus_get_softc(); 603 int cpu = curcpu; 604 605 /* 606 * Disable preemption. 607 */ 608 critical_enter(); 609 610 /* 611 * Do a little interrupt counting. 612 */ 613 (*VMBUS_PCPU_GET(sc, intr_cnt, cpu))++; 614 615 vmbus_handle_intr1(sc, trap_frame, cpu); 616 617 /* 618 * Enable preemption. 619 */ 620 critical_exit(); 621} 622 623static void 624vmbus_synic_setup(void *xsc) 625{ 626 struct vmbus_softc *sc = xsc; 627 int cpu = curcpu; 628 uint64_t val, orig; 629 uint32_t sint; 630 631 if (hyperv_features & CPUID_HV_MSR_VP_INDEX) { 632 /* Save virtual processor id. */ 633 VMBUS_PCPU_GET(sc, vcpuid, cpu) = rdmsr(MSR_HV_VP_INDEX); 634 } else { 635 /* Set virtual processor id to 0 for compatibility. */ 636 VMBUS_PCPU_GET(sc, vcpuid, cpu) = 0; 637 } 638 639 /* 640 * Setup the SynIC message. 641 */ 642 orig = rdmsr(MSR_HV_SIMP); 643 val = MSR_HV_SIMP_ENABLE | (orig & MSR_HV_SIMP_RSVD_MASK) | 644 ((VMBUS_PCPU_GET(sc, message_dma.hv_paddr, cpu) >> PAGE_SHIFT) << 645 MSR_HV_SIMP_PGSHIFT); 646 wrmsr(MSR_HV_SIMP, val); 647 648 /* 649 * Setup the SynIC event flags. 650 */ 651 orig = rdmsr(MSR_HV_SIEFP); 652 val = MSR_HV_SIEFP_ENABLE | (orig & MSR_HV_SIEFP_RSVD_MASK) | 653 ((VMBUS_PCPU_GET(sc, event_flags_dma.hv_paddr, cpu) 654 >> PAGE_SHIFT) << MSR_HV_SIEFP_PGSHIFT); 655 wrmsr(MSR_HV_SIEFP, val); 656 657 658 /* 659 * Configure and unmask SINT for message and event flags. 660 */ 661 sint = MSR_HV_SINT0 + VMBUS_SINT_MESSAGE; 662 orig = rdmsr(sint); 663 val = sc->vmbus_idtvec | MSR_HV_SINT_AUTOEOI | 664 (orig & MSR_HV_SINT_RSVD_MASK); 665 wrmsr(sint, val); 666 667 /* 668 * Configure and unmask SINT for timer. 669 */ 670 sint = MSR_HV_SINT0 + VMBUS_SINT_TIMER; 671 orig = rdmsr(sint); 672 val = sc->vmbus_idtvec | MSR_HV_SINT_AUTOEOI | 673 (orig & MSR_HV_SINT_RSVD_MASK); 674 wrmsr(sint, val); 675 676 /* 677 * All done; enable SynIC. 678 */ 679 orig = rdmsr(MSR_HV_SCONTROL); 680 val = MSR_HV_SCTRL_ENABLE | (orig & MSR_HV_SCTRL_RSVD_MASK); 681 wrmsr(MSR_HV_SCONTROL, val); 682} 683 684static void 685vmbus_synic_teardown(void *arg) 686{ 687 uint64_t orig; 688 uint32_t sint; 689 690 /* 691 * Disable SynIC. 692 */ 693 orig = rdmsr(MSR_HV_SCONTROL); 694 wrmsr(MSR_HV_SCONTROL, (orig & MSR_HV_SCTRL_RSVD_MASK)); 695 696 /* 697 * Mask message and event flags SINT. 698 */ 699 sint = MSR_HV_SINT0 + VMBUS_SINT_MESSAGE; 700 orig = rdmsr(sint); 701 wrmsr(sint, orig | MSR_HV_SINT_MASKED); 702 703 /* 704 * Mask timer SINT. 705 */ 706 sint = MSR_HV_SINT0 + VMBUS_SINT_TIMER; 707 orig = rdmsr(sint); 708 wrmsr(sint, orig | MSR_HV_SINT_MASKED); 709 710 /* 711 * Teardown SynIC message. 712 */ 713 orig = rdmsr(MSR_HV_SIMP); 714 wrmsr(MSR_HV_SIMP, (orig & MSR_HV_SIMP_RSVD_MASK)); 715 716 /* 717 * Teardown SynIC event flags. 718 */ 719 orig = rdmsr(MSR_HV_SIEFP); 720 wrmsr(MSR_HV_SIEFP, (orig & MSR_HV_SIEFP_RSVD_MASK)); 721} 722 723static int 724vmbus_dma_alloc(struct vmbus_softc *sc) 725{ 726 bus_dma_tag_t parent_dtag; 727 uint8_t *evtflags; 728 int cpu; 729 730 parent_dtag = bus_get_dma_tag(sc->vmbus_dev); 731 CPU_FOREACH(cpu) { 732 void *ptr; 733 734 /* 735 * Per-cpu messages and event flags. 736 */ 737 ptr = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0, 738 PAGE_SIZE, VMBUS_PCPU_PTR(sc, message_dma, cpu), 739 BUS_DMA_WAITOK | BUS_DMA_ZERO); 740 if (ptr == NULL) 741 return ENOMEM; 742 VMBUS_PCPU_GET(sc, message, cpu) = ptr; 743 744 ptr = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0, 745 PAGE_SIZE, VMBUS_PCPU_PTR(sc, event_flags_dma, cpu), 746 BUS_DMA_WAITOK | BUS_DMA_ZERO); 747 if (ptr == NULL) 748 return ENOMEM; 749 VMBUS_PCPU_GET(sc, event_flags, cpu) = ptr; 750 } 751 752 evtflags = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0, 753 PAGE_SIZE, &sc->vmbus_evtflags_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO); 754 if (evtflags == NULL) 755 return ENOMEM; 756 sc->vmbus_rx_evtflags = (u_long *)evtflags; 757 sc->vmbus_tx_evtflags = (u_long *)(evtflags + (PAGE_SIZE / 2)); 758 sc->vmbus_evtflags = evtflags; 759 760 sc->vmbus_mnf1 = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0, 761 PAGE_SIZE, &sc->vmbus_mnf1_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO); 762 if (sc->vmbus_mnf1 == NULL) 763 return ENOMEM; 764 765 sc->vmbus_mnf2 = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0, 766 sizeof(struct vmbus_mnf), &sc->vmbus_mnf2_dma, 767 BUS_DMA_WAITOK | BUS_DMA_ZERO); 768 if (sc->vmbus_mnf2 == NULL) 769 return ENOMEM; 770 771 return 0; 772} 773 774static void 775vmbus_dma_free(struct vmbus_softc *sc) 776{ 777 int cpu; 778 779 if (sc->vmbus_evtflags != NULL) { 780 hyperv_dmamem_free(&sc->vmbus_evtflags_dma, sc->vmbus_evtflags); 781 sc->vmbus_evtflags = NULL; 782 sc->vmbus_rx_evtflags = NULL; 783 sc->vmbus_tx_evtflags = NULL; 784 } 785 if (sc->vmbus_mnf1 != NULL) { 786 hyperv_dmamem_free(&sc->vmbus_mnf1_dma, sc->vmbus_mnf1); 787 sc->vmbus_mnf1 = NULL; 788 } 789 if (sc->vmbus_mnf2 != NULL) { 790 hyperv_dmamem_free(&sc->vmbus_mnf2_dma, sc->vmbus_mnf2); 791 sc->vmbus_mnf2 = NULL; 792 } 793 794 CPU_FOREACH(cpu) { 795 if (VMBUS_PCPU_GET(sc, message, cpu) != NULL) { 796 hyperv_dmamem_free( 797 VMBUS_PCPU_PTR(sc, message_dma, cpu), 798 VMBUS_PCPU_GET(sc, message, cpu)); 799 VMBUS_PCPU_GET(sc, message, cpu) = NULL; 800 } 801 if (VMBUS_PCPU_GET(sc, event_flags, cpu) != NULL) { 802 hyperv_dmamem_free( 803 VMBUS_PCPU_PTR(sc, event_flags_dma, cpu), 804 VMBUS_PCPU_GET(sc, event_flags, cpu)); 805 VMBUS_PCPU_GET(sc, event_flags, cpu) = NULL; 806 } 807 } 808} 809 810/** 811 * @brief Find a free IDT slot and setup the interrupt handler. 812 */ 813static int 814vmbus_vector_alloc(void) 815{ 816 int vector; 817 uintptr_t func; 818 struct gate_descriptor *ip; 819 820 /* 821 * Search backwards form the highest IDT vector available for use 822 * as vmbus channel callback vector. We install 'vmbus_isr' 823 * handler at that vector and use it to interrupt vcpus. 824 */ 825 vector = APIC_SPURIOUS_INT; 826 while (--vector >= APIC_IPI_INTS) { 827 ip = &idt[vector]; 828 func = ((long)ip->gd_hioffset << 16 | ip->gd_looffset); 829 if (func == (uintptr_t)&IDTVEC(rsvd)) { 830#ifdef __i386__ 831 setidt(vector , IDTVEC(vmbus_isr), SDT_SYS386IGT, 832 SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); 833#else 834 setidt(vector , IDTVEC(vmbus_isr), SDT_SYSIGT, 835 SEL_KPL, 0); 836#endif 837 838 return (vector); 839 } 840 } 841 return (0); 842} 843 844/** 845 * @brief Restore the IDT slot to rsvd. 846 */ 847static void 848vmbus_vector_free(int vector) 849{ 850 uintptr_t func; 851 struct gate_descriptor *ip; 852 853 if (vector == 0) 854 return; 855 856 KASSERT(vector >= APIC_IPI_INTS && vector < APIC_SPURIOUS_INT, 857 ("invalid vector %d", vector)); 858 859 ip = &idt[vector]; 860 func = ((long)ip->gd_hioffset << 16 | ip->gd_looffset); 861 KASSERT(func == (uintptr_t)&IDTVEC(vmbus_isr), 862 ("invalid vector %d", vector)); 863 864 setidt(vector, IDTVEC(rsvd), SDT_SYSIGT, SEL_KPL, 0); 865} 866 867static void 868vmbus_cpuset_setthread_task(void *xmask, int pending __unused) 869{ 870 cpuset_t *mask = xmask; 871 int error; 872 873 error = cpuset_setthread(curthread->td_tid, mask); 874 if (error) { 875 panic("curthread=%ju: can't pin; error=%d", 876 (uintmax_t)curthread->td_tid, error); 877 } 878} 879 880static int 881vmbus_intr_setup(struct vmbus_softc *sc) 882{ 883 int cpu; 884 885 CPU_FOREACH(cpu) { 886 struct task cpuset_task; 887 char buf[MAXCOMLEN + 1]; 888 cpuset_t cpu_mask; 889 890 /* Allocate an interrupt counter for Hyper-V interrupt */ 891 snprintf(buf, sizeof(buf), "cpu%d:hyperv", cpu); 892 intrcnt_add(buf, VMBUS_PCPU_PTR(sc, intr_cnt, cpu)); 893 894 /* 895 * Setup taskqueue to handle events. Task will be per- 896 * channel. 897 */ 898 VMBUS_PCPU_GET(sc, event_tq, cpu) = taskqueue_create_fast( 899 "hyperv event", M_WAITOK, taskqueue_thread_enqueue, 900 VMBUS_PCPU_PTR(sc, event_tq, cpu)); 901 taskqueue_start_threads(VMBUS_PCPU_PTR(sc, event_tq, cpu), 902 1, PI_NET, "hvevent%d", cpu); 903 904 CPU_SETOF(cpu, &cpu_mask); 905 TASK_INIT(&cpuset_task, 0, vmbus_cpuset_setthread_task, 906 &cpu_mask); 907 taskqueue_enqueue(VMBUS_PCPU_GET(sc, event_tq, cpu), 908 &cpuset_task); 909 taskqueue_drain(VMBUS_PCPU_GET(sc, event_tq, cpu), 910 &cpuset_task); 911 912 /* 913 * Setup tasks and taskqueues to handle messages. 914 */ 915 VMBUS_PCPU_GET(sc, message_tq, cpu) = taskqueue_create_fast( 916 "hyperv msg", M_WAITOK, taskqueue_thread_enqueue, 917 VMBUS_PCPU_PTR(sc, message_tq, cpu)); 918 taskqueue_start_threads(VMBUS_PCPU_PTR(sc, message_tq, cpu), 1, 919 PI_NET, "hvmsg%d", cpu); 920 TASK_INIT(VMBUS_PCPU_PTR(sc, message_task, cpu), 0, 921 vmbus_msg_task, sc); 922 923 CPU_SETOF(cpu, &cpu_mask); 924 TASK_INIT(&cpuset_task, 0, vmbus_cpuset_setthread_task, 925 &cpu_mask); 926 taskqueue_enqueue(VMBUS_PCPU_GET(sc, message_tq, cpu), 927 &cpuset_task); 928 taskqueue_drain(VMBUS_PCPU_GET(sc, message_tq, cpu), 929 &cpuset_task); 930 } 931 932 /* 933 * All Hyper-V ISR required resources are setup, now let's find a 934 * free IDT vector for Hyper-V ISR and set it up. 935 */ 936 sc->vmbus_idtvec = vmbus_vector_alloc(); 937 if (sc->vmbus_idtvec == 0) { 938 device_printf(sc->vmbus_dev, "cannot find free IDT vector\n"); 939 return ENXIO; 940 } 941 if(bootverbose) { 942 device_printf(sc->vmbus_dev, "vmbus IDT vector %d\n", 943 sc->vmbus_idtvec); 944 } 945 return 0; 946} 947 948static void 949vmbus_intr_teardown(struct vmbus_softc *sc) 950{ 951 int cpu; 952 953 vmbus_vector_free(sc->vmbus_idtvec); 954 955 CPU_FOREACH(cpu) { 956 if (VMBUS_PCPU_GET(sc, event_tq, cpu) != NULL) { 957 taskqueue_free(VMBUS_PCPU_GET(sc, event_tq, cpu)); 958 VMBUS_PCPU_GET(sc, event_tq, cpu) = NULL; 959 } 960 if (VMBUS_PCPU_GET(sc, message_tq, cpu) != NULL) { 961 taskqueue_drain(VMBUS_PCPU_GET(sc, message_tq, cpu), 962 VMBUS_PCPU_PTR(sc, message_task, cpu)); 963 taskqueue_free(VMBUS_PCPU_GET(sc, message_tq, cpu)); 964 VMBUS_PCPU_GET(sc, message_tq, cpu) = NULL; 965 } 966 } 967} 968 969static int 970vmbus_read_ivar(device_t dev, device_t child, int index, uintptr_t *result) 971{ 972 return (ENOENT); 973} 974 975static int 976vmbus_child_pnpinfo_str(device_t dev, device_t child, char *buf, size_t buflen) 977{ 978 const struct vmbus_channel *chan; 979 char guidbuf[HYPERV_GUID_STRLEN]; 980 981 chan = vmbus_get_channel(child); 982 if (chan == NULL) { 983 /* Event timer device, which does not belong to a channel */ 984 return (0); 985 } 986 987 strlcat(buf, "classid=", buflen); 988 hyperv_guid2str(&chan->ch_guid_type, guidbuf, sizeof(guidbuf)); 989 strlcat(buf, guidbuf, buflen); 990 991 strlcat(buf, " deviceid=", buflen); 992 hyperv_guid2str(&chan->ch_guid_inst, guidbuf, sizeof(guidbuf)); 993 strlcat(buf, guidbuf, buflen); 994 995 return (0); 996} 997 998int 999vmbus_add_child(struct vmbus_channel *chan) 1000{ 1001 struct vmbus_softc *sc = chan->ch_vmbus; 1002 device_t parent = sc->vmbus_dev; 1003 int error = 0; 1004 1005 /* New channel has been offered */ 1006 vmbus_scan_newchan(sc); 1007 1008 chan->ch_dev = device_add_child(parent, NULL, -1); 1009 if (chan->ch_dev == NULL) { 1010 device_printf(parent, "device_add_child for chan%u failed\n", 1011 chan->ch_id); 1012 error = ENXIO; 1013 goto done; 1014 } 1015 device_set_ivars(chan->ch_dev, chan); 1016 1017done: 1018 /* New device has been/should be added to vmbus. */ 1019 vmbus_scan_newdev(sc); 1020 return error; 1021} 1022 1023int 1024vmbus_delete_child(struct vmbus_channel *chan) 1025{ 1026 int error; 1027 1028 if (chan->ch_dev == NULL) { 1029 /* Failed to add a device. */ 1030 return 0; 1031 } 1032 1033 /* 1034 * XXXKYS: Ensure that this is the opposite of 1035 * device_add_child() 1036 */ 1037 mtx_lock(&Giant); 1038 error = device_delete_child(chan->ch_vmbus->vmbus_dev, chan->ch_dev); 1039 mtx_unlock(&Giant); 1040 1041 return error; 1042} 1043 1044static int 1045vmbus_sysctl_version(SYSCTL_HANDLER_ARGS) 1046{ 1047 struct vmbus_softc *sc = arg1; 1048 char verstr[16]; 1049 1050 snprintf(verstr, sizeof(verstr), "%u.%u", 1051 VMBUS_VERSION_MAJOR(sc->vmbus_version), 1052 VMBUS_VERSION_MINOR(sc->vmbus_version)); 1053 return sysctl_handle_string(oidp, verstr, sizeof(verstr), req); 1054} 1055 1056static uint32_t 1057vmbus_get_version_method(device_t bus, device_t dev) 1058{ 1059 struct vmbus_softc *sc = device_get_softc(bus); 1060 1061 return sc->vmbus_version; 1062} 1063 1064static int 1065vmbus_probe_guid_method(device_t bus, device_t dev, 1066 const struct hyperv_guid *guid) 1067{ 1068 const struct vmbus_channel *chan = vmbus_get_channel(dev); 1069 1070 if (memcmp(&chan->ch_guid_type, guid, sizeof(struct hyperv_guid)) == 0) 1071 return 0; 1072 return ENXIO; 1073} 1074 1075static int 1076vmbus_probe(device_t dev) 1077{ 1078 char *id[] = { "VMBUS", NULL }; 1079 1080 if (ACPI_ID_PROBE(device_get_parent(dev), dev, id) == NULL || 1081 device_get_unit(dev) != 0 || vm_guest != VM_GUEST_HV || 1082 (hyperv_features & CPUID_HV_MSR_SYNIC) == 0) 1083 return (ENXIO); 1084 1085 device_set_desc(dev, "Hyper-V Vmbus"); 1086 1087 return (BUS_PROBE_DEFAULT); 1088} 1089 1090/** 1091 * @brief Main vmbus driver initialization routine. 1092 * 1093 * Here, we 1094 * - initialize the vmbus driver context 1095 * - setup various driver entry points 1096 * - invoke the vmbus hv main init routine 1097 * - get the irq resource 1098 * - invoke the vmbus to add the vmbus root device 1099 * - setup the vmbus root device 1100 * - retrieve the channel offers 1101 */ 1102static int 1103vmbus_doattach(struct vmbus_softc *sc) 1104{ 1105 struct sysctl_oid_list *child; 1106 struct sysctl_ctx_list *ctx; 1107 int ret; 1108 1109 if (sc->vmbus_flags & VMBUS_FLAG_ATTACHED) 1110 return (0); 1111 sc->vmbus_flags |= VMBUS_FLAG_ATTACHED; 1112 1113 mtx_init(&sc->vmbus_scan_lock, "vmbus scan", NULL, MTX_DEF); 1114 sc->vmbus_gpadl = VMBUS_GPADL_START; 1115 mtx_init(&sc->vmbus_prichan_lock, "vmbus prichan", NULL, MTX_DEF); 1116 TAILQ_INIT(&sc->vmbus_prichans); 1117 sc->vmbus_chmap = malloc( 1118 sizeof(struct vmbus_channel *) * VMBUS_CHAN_MAX, M_DEVBUF, 1119 M_WAITOK | M_ZERO); 1120 1121 /* 1122 * Create context for "post message" Hypercalls 1123 */ 1124 sc->vmbus_xc = vmbus_xact_ctx_create(bus_get_dma_tag(sc->vmbus_dev), 1125 HYPERCALL_POSTMSGIN_SIZE, VMBUS_MSG_SIZE, 1126 sizeof(struct vmbus_msghc)); 1127 if (sc->vmbus_xc == NULL) { 1128 ret = ENXIO; 1129 goto cleanup; 1130 } 1131 1132 /* 1133 * Allocate DMA stuffs. 1134 */ 1135 ret = vmbus_dma_alloc(sc); 1136 if (ret != 0) 1137 goto cleanup; 1138 1139 /* 1140 * Setup interrupt. 1141 */ 1142 ret = vmbus_intr_setup(sc); 1143 if (ret != 0) 1144 goto cleanup; 1145 1146 /* 1147 * Setup SynIC. 1148 */ 1149 if (bootverbose) 1150 device_printf(sc->vmbus_dev, "smp_started = %d\n", smp_started); 1151 smp_rendezvous(NULL, vmbus_synic_setup, NULL, sc); 1152 sc->vmbus_flags |= VMBUS_FLAG_SYNIC; 1153 1154 /* 1155 * Initialize vmbus, e.g. connect to Hypervisor. 1156 */ 1157 ret = vmbus_init(sc); 1158 if (ret != 0) 1159 goto cleanup; 1160 1161 if (sc->vmbus_version == VMBUS_VERSION_WS2008 || 1162 sc->vmbus_version == VMBUS_VERSION_WIN7) 1163 sc->vmbus_event_proc = vmbus_event_proc_compat; 1164 else 1165 sc->vmbus_event_proc = vmbus_event_proc; 1166 1167 ret = vmbus_scan(sc); 1168 if (ret != 0) 1169 goto cleanup; 1170 1171 ctx = device_get_sysctl_ctx(sc->vmbus_dev); 1172 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->vmbus_dev)); 1173 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "version", 1174 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0, 1175 vmbus_sysctl_version, "A", "vmbus version"); 1176 1177 return (ret); 1178 1179cleanup: 1180 vmbus_intr_teardown(sc); 1181 vmbus_dma_free(sc); 1182 if (sc->vmbus_xc != NULL) { 1183 vmbus_xact_ctx_destroy(sc->vmbus_xc); 1184 sc->vmbus_xc = NULL; 1185 } 1186 free(sc->vmbus_chmap, M_DEVBUF); 1187 mtx_destroy(&sc->vmbus_scan_lock); 1188 mtx_destroy(&sc->vmbus_prichan_lock); 1189 1190 return (ret); 1191} 1192 1193static void 1194vmbus_event_proc_dummy(struct vmbus_softc *sc __unused, int cpu __unused) 1195{ 1196} 1197 1198static int 1199vmbus_attach(device_t dev) 1200{ 1201 vmbus_sc = device_get_softc(dev); 1202 vmbus_sc->vmbus_dev = dev; 1203 1204 /* 1205 * Event processing logic will be configured: 1206 * - After the vmbus protocol version negotiation. 1207 * - Before we request channel offers. 1208 */ 1209 vmbus_sc->vmbus_event_proc = vmbus_event_proc_dummy; 1210 1211 /* 1212 * If the system has already booted and thread 1213 * scheduling is possible indicated by the global 1214 * cold set to zero, we just call the driver 1215 * initialization directly. 1216 */ 1217 if (!cold) 1218 vmbus_doattach(vmbus_sc); 1219 1220 return (0); 1221} 1222 1223static int 1224vmbus_detach(device_t dev) 1225{ 1226 struct vmbus_softc *sc = device_get_softc(dev); 1227 1228 vmbus_chan_destroy_all(sc); 1229 1230 vmbus_disconnect(sc); 1231 1232 if (sc->vmbus_flags & VMBUS_FLAG_SYNIC) { 1233 sc->vmbus_flags &= ~VMBUS_FLAG_SYNIC; 1234 smp_rendezvous(NULL, vmbus_synic_teardown, NULL, NULL); 1235 } 1236 1237 vmbus_intr_teardown(sc); 1238 vmbus_dma_free(sc); 1239 1240 if (sc->vmbus_xc != NULL) { 1241 vmbus_xact_ctx_destroy(sc->vmbus_xc); 1242 sc->vmbus_xc = NULL; 1243 } 1244 1245 free(sc->vmbus_chmap, M_DEVBUF); 1246 mtx_destroy(&sc->vmbus_scan_lock); 1247 mtx_destroy(&sc->vmbus_prichan_lock); 1248 1249 return (0); 1250} 1251 1252static void 1253vmbus_sysinit(void *arg __unused) 1254{ 1255 struct vmbus_softc *sc = vmbus_get_softc(); 1256 1257 if (vm_guest != VM_GUEST_HV || sc == NULL) 1258 return; 1259 1260 /* 1261 * If the system has already booted and thread 1262 * scheduling is possible, as indicated by the 1263 * global cold set to zero, we just call the driver 1264 * initialization directly. 1265 */ 1266 if (!cold) 1267 vmbus_doattach(sc); 1268} 1269/* 1270 * NOTE: 1271 * We have to start as the last step of SI_SUB_SMP, i.e. after SMP is 1272 * initialized. 1273 */ 1274SYSINIT(vmbus_initialize, SI_SUB_SMP, SI_ORDER_ANY, vmbus_sysinit, NULL); 1275