hvm.c revision 291687
1/* 2 * Copyright (c) 2008, 2013 Citrix Systems, Inc. 3 * Copyright (c) 2012 Spectra Logic Corporation 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28#include <sys/cdefs.h> 29__FBSDID("$FreeBSD: stable/10/sys/x86/xen/hvm.c 291687 2015-12-03 11:05:35Z royger $"); 30 31#include <sys/param.h> 32#include <sys/bus.h> 33#include <sys/kernel.h> 34#include <sys/malloc.h> 35#include <sys/proc.h> 36#include <sys/smp.h> 37#include <sys/systm.h> 38 39#include <vm/vm.h> 40#include <vm/pmap.h> 41 42#include <dev/pci/pcivar.h> 43 44#include <machine/cpufunc.h> 45#include <machine/cpu.h> 46#include <machine/smp.h> 47 48#include <x86/apicreg.h> 49 50#include <xen/xen-os.h> 51#include <xen/features.h> 52#include <xen/gnttab.h> 53#include <xen/hypervisor.h> 54#include <xen/hvm.h> 55#include <xen/xen_intr.h> 56 57#include <xen/interface/hvm/params.h> 58#include <xen/interface/vcpu.h> 59 60/*--------------------------- Forward Declarations ---------------------------*/ 61#ifdef SMP 62static driver_filter_t xen_smp_rendezvous_action; 63static driver_filter_t xen_invltlb; 64static driver_filter_t xen_invlpg; 65static driver_filter_t xen_invlrng; 66static driver_filter_t xen_invlcache; 67#ifdef __i386__ 68static driver_filter_t xen_lazypmap; 69#endif 70static driver_filter_t xen_ipi_bitmap_handler; 71static driver_filter_t xen_cpustop_handler; 72static driver_filter_t xen_cpususpend_handler; 73static driver_filter_t xen_cpustophard_handler; 74static void xen_ipi_vectored(u_int vector, int dest); 75#endif 76static void xen_hvm_cpu_init(void); 77 78/*---------------------------- Extern Declarations ---------------------------*/ 79#ifdef __i386__ 80extern void pmap_lazyfix_action(void); 81#endif 82#ifdef __amd64__ 83extern int pmap_pcid_enabled; 84#endif 85 86/*---------------------------------- Macros ----------------------------------*/ 87#define IPI_TO_IDX(ipi) ((ipi) - APIC_IPI_INTS) 88 89/*-------------------------------- Local Types -------------------------------*/ 90enum xen_hvm_init_type { 91 XEN_HVM_INIT_COLD, 92 XEN_HVM_INIT_CANCELLED_SUSPEND, 93 XEN_HVM_INIT_RESUME 94}; 95 96struct xen_ipi_handler 97{ 98 driver_filter_t *filter; 99 const char *description; 100}; 101 102/*-------------------------------- Global Data -------------------------------*/ 103enum xen_domain_type xen_domain_type = XEN_NATIVE; 104 105#ifdef SMP 106struct cpu_ops xen_hvm_cpu_ops = { 107 .ipi_vectored = lapic_ipi_vectored, 108 .cpu_init = xen_hvm_cpu_init, 109 .cpu_resume = xen_hvm_cpu_init 110}; 111#endif 112 113static MALLOC_DEFINE(M_XENHVM, "xen_hvm", "Xen HVM PV Support"); 114 115#ifdef SMP 116static struct xen_ipi_handler xen_ipis[] = 117{ 118 [IPI_TO_IDX(IPI_RENDEZVOUS)] = { xen_smp_rendezvous_action, "r" }, 119 [IPI_TO_IDX(IPI_INVLTLB)] = { xen_invltlb, "itlb"}, 120 [IPI_TO_IDX(IPI_INVLPG)] = { xen_invlpg, "ipg" }, 121 [IPI_TO_IDX(IPI_INVLRNG)] = { xen_invlrng, "irg" }, 122 [IPI_TO_IDX(IPI_INVLCACHE)] = { xen_invlcache, "ic" }, 123#ifdef __i386__ 124 [IPI_TO_IDX(IPI_LAZYPMAP)] = { xen_lazypmap, "lp" }, 125#endif 126 [IPI_TO_IDX(IPI_BITMAP_VECTOR)] = { xen_ipi_bitmap_handler, "b" }, 127 [IPI_TO_IDX(IPI_STOP)] = { xen_cpustop_handler, "st" }, 128 [IPI_TO_IDX(IPI_SUSPEND)] = { xen_cpususpend_handler, "sp" }, 129 [IPI_TO_IDX(IPI_STOP_HARD)] = { xen_cpustophard_handler, "sth" }, 130}; 131#endif 132 133/** 134 * If non-zero, the hypervisor has been configured to use a direct 135 * IDT event callback for interrupt injection. 136 */ 137int xen_vector_callback_enabled; 138 139/*------------------------------- Per-CPU Data -------------------------------*/ 140DPCPU_DEFINE(struct vcpu_info, vcpu_local_info); 141DPCPU_DEFINE(struct vcpu_info *, vcpu_info); 142#ifdef SMP 143DPCPU_DEFINE(xen_intr_handle_t, ipi_handle[nitems(xen_ipis)]); 144#endif 145 146/*------------------ Hypervisor Access Shared Memory Regions -----------------*/ 147/** Hypercall table accessed via HYPERVISOR_*_op() methods. */ 148char *hypercall_stubs; 149shared_info_t *HYPERVISOR_shared_info; 150 151 152/*------------------------------ Sysctl tunables -----------------------------*/ 153int xen_disable_pv_disks = 0; 154int xen_disable_pv_nics = 0; 155TUNABLE_INT("hw.xen.disable_pv_disks", &xen_disable_pv_disks); 156TUNABLE_INT("hw.xen.disable_pv_nics", &xen_disable_pv_nics); 157 158#ifdef SMP 159/*---------------------------- XEN PV IPI Handlers ---------------------------*/ 160/* 161 * This are C clones of the ASM functions found in apic_vector.s 162 */ 163static int 164xen_ipi_bitmap_handler(void *arg) 165{ 166 struct trapframe *frame; 167 168 frame = arg; 169 ipi_bitmap_handler(*frame); 170 return (FILTER_HANDLED); 171} 172 173static int 174xen_smp_rendezvous_action(void *arg) 175{ 176#ifdef COUNT_IPIS 177 (*ipi_rendezvous_counts[PCPU_GET(cpuid)])++; 178#endif /* COUNT_IPIS */ 179 180 smp_rendezvous_action(); 181 return (FILTER_HANDLED); 182} 183 184static int 185xen_invltlb(void *arg) 186{ 187 188 invltlb_handler(); 189 return (FILTER_HANDLED); 190} 191 192#ifdef __amd64__ 193static int 194xen_invltlb_pcid(void *arg) 195{ 196 197 invltlb_pcid_handler(); 198 return (FILTER_HANDLED); 199} 200#endif 201 202static int 203xen_invlpg(void *arg) 204{ 205 206 invlpg_handler(); 207 return (FILTER_HANDLED); 208} 209 210#ifdef __amd64__ 211static int 212xen_invlpg_pcid(void *arg) 213{ 214 215 invlpg_pcid_handler(); 216 return (FILTER_HANDLED); 217} 218#endif 219 220static int 221xen_invlrng(void *arg) 222{ 223 224 invlrng_handler(); 225 return (FILTER_HANDLED); 226} 227 228static int 229xen_invlcache(void *arg) 230{ 231 232 invlcache_handler(); 233 return (FILTER_HANDLED); 234} 235 236#ifdef __i386__ 237static int 238xen_lazypmap(void *arg) 239{ 240 241 pmap_lazyfix_action(); 242 return (FILTER_HANDLED); 243} 244#endif 245 246static int 247xen_cpustop_handler(void *arg) 248{ 249 250 cpustop_handler(); 251 return (FILTER_HANDLED); 252} 253 254static int 255xen_cpususpend_handler(void *arg) 256{ 257 258 cpususpend_handler(); 259 return (FILTER_HANDLED); 260} 261 262static int 263xen_cpustophard_handler(void *arg) 264{ 265 266 ipi_nmi_handler(); 267 return (FILTER_HANDLED); 268} 269 270/* Xen PV IPI sender */ 271static void 272xen_ipi_vectored(u_int vector, int dest) 273{ 274 xen_intr_handle_t *ipi_handle; 275 int ipi_idx, to_cpu, self; 276 277 ipi_idx = IPI_TO_IDX(vector); 278 if (ipi_idx > nitems(xen_ipis)) 279 panic("IPI out of range"); 280 281 switch(dest) { 282 case APIC_IPI_DEST_SELF: 283 ipi_handle = DPCPU_GET(ipi_handle); 284 xen_intr_signal(ipi_handle[ipi_idx]); 285 break; 286 case APIC_IPI_DEST_ALL: 287 CPU_FOREACH(to_cpu) { 288 ipi_handle = DPCPU_ID_GET(to_cpu, ipi_handle); 289 xen_intr_signal(ipi_handle[ipi_idx]); 290 } 291 break; 292 case APIC_IPI_DEST_OTHERS: 293 self = PCPU_GET(cpuid); 294 CPU_FOREACH(to_cpu) { 295 if (to_cpu != self) { 296 ipi_handle = DPCPU_ID_GET(to_cpu, ipi_handle); 297 xen_intr_signal(ipi_handle[ipi_idx]); 298 } 299 } 300 break; 301 default: 302 to_cpu = apic_cpuid(dest); 303 ipi_handle = DPCPU_ID_GET(to_cpu, ipi_handle); 304 xen_intr_signal(ipi_handle[ipi_idx]); 305 break; 306 } 307} 308 309/*---------------------- XEN diverged cpu operations -------------------------*/ 310static void 311xen_cpu_ipi_init(int cpu) 312{ 313 xen_intr_handle_t *ipi_handle; 314 const struct xen_ipi_handler *ipi; 315 device_t dev; 316 int idx, rc; 317 318 ipi_handle = DPCPU_ID_GET(cpu, ipi_handle); 319 dev = pcpu_find(cpu)->pc_device; 320 KASSERT((dev != NULL), ("NULL pcpu device_t")); 321 322 for (ipi = xen_ipis, idx = 0; idx < nitems(xen_ipis); ipi++, idx++) { 323 324 if (ipi->filter == NULL) { 325 ipi_handle[idx] = NULL; 326 continue; 327 } 328 329 rc = xen_intr_alloc_and_bind_ipi(dev, cpu, ipi->filter, 330 INTR_TYPE_TTY, &ipi_handle[idx]); 331 if (rc != 0) 332 panic("Unable to allocate a XEN IPI port"); 333 xen_intr_describe(ipi_handle[idx], "%s", ipi->description); 334 } 335} 336 337static void 338xen_setup_cpus(void) 339{ 340 int i; 341 342 if (!xen_hvm_domain() || !xen_vector_callback_enabled) 343 return; 344 345#ifdef __amd64__ 346 if (pmap_pcid_enabled) { 347 xen_ipis[IPI_TO_IDX(IPI_INVLTLB)].filter = xen_invltlb_pcid; 348 xen_ipis[IPI_TO_IDX(IPI_INVLPG)].filter = xen_invlpg_pcid; 349 } 350#endif 351 CPU_FOREACH(i) 352 xen_cpu_ipi_init(i); 353 354 /* Set the xen pv ipi ops to replace the native ones */ 355 cpu_ops.ipi_vectored = xen_ipi_vectored; 356} 357#endif 358 359/*---------------------- XEN Hypervisor Probe and Setup ----------------------*/ 360static uint32_t 361xen_hvm_cpuid_base(void) 362{ 363 uint32_t base, regs[4]; 364 365 for (base = 0x40000000; base < 0x40010000; base += 0x100) { 366 do_cpuid(base, regs); 367 if (!memcmp("XenVMMXenVMM", ®s[1], 12) 368 && (regs[0] - base) >= 2) 369 return (base); 370 } 371 return (0); 372} 373 374/* 375 * Allocate and fill in the hypcall page. 376 */ 377static int 378xen_hvm_init_hypercall_stubs(void) 379{ 380 uint32_t base, regs[4]; 381 int i; 382 383 base = xen_hvm_cpuid_base(); 384 if (base == 0) 385 return (ENXIO); 386 387 if (hypercall_stubs == NULL) { 388 do_cpuid(base + 1, regs); 389 printf("XEN: Hypervisor version %d.%d detected.\n", 390 regs[0] >> 16, regs[0] & 0xffff); 391 } 392 393 /* 394 * Find the hypercall pages. 395 */ 396 do_cpuid(base + 2, regs); 397 398 if (hypercall_stubs == NULL) { 399 size_t call_region_size; 400 401 call_region_size = regs[0] * PAGE_SIZE; 402 hypercall_stubs = malloc(call_region_size, M_XENHVM, M_NOWAIT); 403 if (hypercall_stubs == NULL) 404 panic("Unable to allocate Xen hypercall region"); 405 } 406 407 for (i = 0; i < regs[0]; i++) 408 wrmsr(regs[1], vtophys(hypercall_stubs + i * PAGE_SIZE) + i); 409 410 return (0); 411} 412 413static void 414xen_hvm_init_shared_info_page(void) 415{ 416 struct xen_add_to_physmap xatp; 417 418 if (HYPERVISOR_shared_info == NULL) { 419 HYPERVISOR_shared_info = malloc(PAGE_SIZE, M_XENHVM, M_NOWAIT); 420 if (HYPERVISOR_shared_info == NULL) 421 panic("Unable to allocate Xen shared info page"); 422 } 423 424 xatp.domid = DOMID_SELF; 425 xatp.idx = 0; 426 xatp.space = XENMAPSPACE_shared_info; 427 xatp.gpfn = vtophys(HYPERVISOR_shared_info) >> PAGE_SHIFT; 428 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 429 panic("HYPERVISOR_memory_op failed"); 430} 431 432/* 433 * Tell the hypervisor how to contact us for event channel callbacks. 434 */ 435void 436xen_hvm_set_callback(device_t dev) 437{ 438 struct xen_hvm_param xhp; 439 int irq; 440 441 if (xen_vector_callback_enabled) 442 return; 443 444 xhp.domid = DOMID_SELF; 445 xhp.index = HVM_PARAM_CALLBACK_IRQ; 446 if (xen_feature(XENFEAT_hvm_callback_vector) != 0) { 447 int error; 448 449 xhp.value = HVM_CALLBACK_VECTOR(IDT_EVTCHN); 450 error = HYPERVISOR_hvm_op(HVMOP_set_param, &xhp); 451 if (error == 0) { 452 xen_vector_callback_enabled = 1; 453 return; 454 } 455 printf("Xen HVM callback vector registration failed (%d). " 456 "Falling back to emulated device interrupt\n", error); 457 } 458 xen_vector_callback_enabled = 0; 459 if (dev == NULL) { 460 /* 461 * Called from early boot or resume. 462 * xenpci will invoke us again later. 463 */ 464 return; 465 } 466 467 irq = pci_get_irq(dev); 468 if (irq < 16) { 469 xhp.value = HVM_CALLBACK_GSI(irq); 470 } else { 471 u_int slot; 472 u_int pin; 473 474 slot = pci_get_slot(dev); 475 pin = pci_get_intpin(dev) - 1; 476 xhp.value = HVM_CALLBACK_PCI_INTX(slot, pin); 477 } 478 479 if (HYPERVISOR_hvm_op(HVMOP_set_param, &xhp) != 0) 480 panic("Can't set evtchn callback"); 481} 482 483#define XEN_MAGIC_IOPORT 0x10 484enum { 485 XMI_MAGIC = 0x49d2, 486 XMI_UNPLUG_IDE_DISKS = 0x01, 487 XMI_UNPLUG_NICS = 0x02, 488 XMI_UNPLUG_IDE_EXCEPT_PRI_MASTER = 0x04 489}; 490 491static void 492xen_hvm_disable_emulated_devices(void) 493{ 494 u_short disable_devs = 0; 495 496 if (inw(XEN_MAGIC_IOPORT) != XMI_MAGIC) 497 return; 498 499 if (xen_disable_pv_disks == 0) { 500 if (bootverbose) 501 printf("XEN: disabling emulated disks\n"); 502 disable_devs |= XMI_UNPLUG_IDE_DISKS; 503 } 504 if (xen_disable_pv_nics == 0) { 505 if (bootverbose) 506 printf("XEN: disabling emulated nics\n"); 507 disable_devs |= XMI_UNPLUG_NICS; 508 } 509 510 if (disable_devs != 0) 511 outw(XEN_MAGIC_IOPORT, disable_devs); 512} 513 514static void 515xen_hvm_init(enum xen_hvm_init_type init_type) 516{ 517 int error; 518 int i; 519 520 if (init_type == XEN_HVM_INIT_CANCELLED_SUSPEND) 521 return; 522 523 error = xen_hvm_init_hypercall_stubs(); 524 525 switch (init_type) { 526 case XEN_HVM_INIT_COLD: 527 if (error != 0) 528 return; 529 530 setup_xen_features(); 531#ifdef SMP 532 cpu_ops = xen_hvm_cpu_ops; 533#endif 534 vm_guest = VM_GUEST_XEN; 535 break; 536 case XEN_HVM_INIT_RESUME: 537 if (error != 0) 538 panic("Unable to init Xen hypercall stubs on resume"); 539 540 /* Clear stale vcpu_info. */ 541 CPU_FOREACH(i) 542 DPCPU_ID_SET(i, vcpu_info, NULL); 543 break; 544 default: 545 panic("Unsupported HVM initialization type"); 546 } 547 548 xen_vector_callback_enabled = 0; 549 xen_domain_type = XEN_HVM_DOMAIN; 550 xen_hvm_init_shared_info_page(); 551 xen_hvm_set_callback(NULL); 552 xen_hvm_disable_emulated_devices(); 553} 554 555void 556xen_hvm_suspend(void) 557{ 558} 559 560void 561xen_hvm_resume(bool suspend_cancelled) 562{ 563 564 xen_hvm_init(suspend_cancelled ? 565 XEN_HVM_INIT_CANCELLED_SUSPEND : XEN_HVM_INIT_RESUME); 566 567 /* Register vcpu_info area for CPU#0. */ 568 xen_hvm_cpu_init(); 569} 570 571static void 572xen_hvm_sysinit(void *arg __unused) 573{ 574 xen_hvm_init(XEN_HVM_INIT_COLD); 575} 576 577static void 578xen_set_vcpu_id(void) 579{ 580 struct pcpu *pc; 581 int i; 582 583 /* Set vcpu_id to acpi_id */ 584 CPU_FOREACH(i) { 585 pc = pcpu_find(i); 586 pc->pc_vcpu_id = pc->pc_acpi_id; 587 if (bootverbose) 588 printf("XEN: CPU %u has VCPU ID %u\n", 589 i, pc->pc_vcpu_id); 590 } 591} 592 593static void 594xen_hvm_cpu_init(void) 595{ 596 struct vcpu_register_vcpu_info info; 597 struct vcpu_info *vcpu_info; 598 int cpu, rc; 599 600 if (!xen_domain()) 601 return; 602 603 if (DPCPU_GET(vcpu_info) != NULL) { 604 /* 605 * vcpu_info is already set. We're resuming 606 * from a failed migration and our pre-suspend 607 * configuration is still valid. 608 */ 609 return; 610 } 611 612 vcpu_info = DPCPU_PTR(vcpu_local_info); 613 cpu = PCPU_GET(vcpu_id); 614 info.mfn = vtophys(vcpu_info) >> PAGE_SHIFT; 615 info.offset = vtophys(vcpu_info) - trunc_page(vtophys(vcpu_info)); 616 617 rc = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); 618 if (rc != 0) 619 DPCPU_SET(vcpu_info, &HYPERVISOR_shared_info->vcpu_info[cpu]); 620 else 621 DPCPU_SET(vcpu_info, vcpu_info); 622} 623 624SYSINIT(xen_hvm_init, SI_SUB_HYPERVISOR, SI_ORDER_FIRST, xen_hvm_sysinit, NULL); 625#ifdef SMP 626SYSINIT(xen_setup_cpus, SI_SUB_SMP, SI_ORDER_FIRST, xen_setup_cpus, NULL); 627#endif 628SYSINIT(xen_hvm_cpu_init, SI_SUB_INTR, SI_ORDER_FIRST, xen_hvm_cpu_init, NULL); 629SYSINIT(xen_set_vcpu_id, SI_SUB_CPU, SI_ORDER_ANY, xen_set_vcpu_id, NULL); 630