sys/x86/xen/hvm.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2008, 2013 Citrix Systems, Inc.
   5  * Copyright (c) 2012 Spectra Logic Corporation
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following 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 AND CONTRIBUTORS AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  */
  29
  30 #include <sys/cdefs.h>
  31 __FBSDID("$FreeBSD$");
  32
  33 #include <sys/param.h>
  34 #include <sys/bus.h>
  35 #include <sys/kernel.h>
  36 #include <sys/malloc.h>
  37 #include <sys/proc.h>
  38 #include <sys/smp.h>
  39 #include <sys/systm.h>
  40
  41 #include <vm/vm.h>
  42 #include <vm/pmap.h>
  43 #include <vm/vm_param.h>
  44
  45 #include <dev/pci/pcivar.h>
  46
  47 #include <machine/cpufunc.h>
  48 #include <machine/cpu.h>
  49 #include <machine/smp.h>
  50
  51 #include <x86/apicreg.h>
  52
  53 #include <xen/xen-os.h>
  54 #include <xen/features.h>
  55 #include <xen/gnttab.h>
  56 #include <xen/hypervisor.h>
  57 #include <xen/hvm.h>
  58 #include <xen/xen_intr.h>
  59
  60 #include <xen/interface/arch-x86/cpuid.h>
  61 #include <xen/interface/hvm/params.h>
  62 #include <xen/interface/vcpu.h>
  63
  64 /*--------------------------- Forward Declarations ---------------------------*/
  65 static void xen_hvm_cpu_init(void);
  66
  67 /*-------------------------------- Global Data -------------------------------*/
  68 enum xen_domain_type xen_domain_type = XEN_NATIVE;
  69
  70 #ifdef SMP
  71 struct cpu_ops xen_hvm_cpu_ops = {
  72         .cpu_init       = xen_hvm_cpu_init,
  73         .cpu_resume     = xen_hvm_cpu_init
  74 };
  75 #endif
  76
  77 static MALLOC_DEFINE(M_XENHVM, "xen_hvm", "Xen HVM PV Support");
  78
  79 /**
  80  * If non-zero, the hypervisor has been configured to use a direct
  81  * IDT event callback for interrupt injection.
  82  */
  83 int xen_vector_callback_enabled;
  84
  85 /**
  86  * Start info flags. ATM this only used to store the initial domain flag for
  87  * PVHv2, and it's always empty for HVM guests.
  88  */
  89 uint32_t hvm_start_flags;
  90
  91 /*------------------------------- Per-CPU Data -------------------------------*/
  92 DPCPU_DEFINE(struct vcpu_info, vcpu_local_info);
  93 DPCPU_DEFINE(struct vcpu_info *, vcpu_info);
  94
  95 /*------------------ Hypervisor Access Shared Memory Regions -----------------*/
  96 shared_info_t *HYPERVISOR_shared_info;
  97
  98 /*------------------------------ Sysctl tunables -----------------------------*/
  99 int xen_disable_pv_disks = 0;
 100 int xen_disable_pv_nics = 0;
 101 TUNABLE_INT("hw.xen.disable_pv_disks", &xen_disable_pv_disks);
 102 TUNABLE_INT("hw.xen.disable_pv_nics", &xen_disable_pv_nics);
 103
 104 /*---------------------- XEN Hypervisor Probe and Setup ----------------------*/
 105
 106 static uint32_t cpuid_base;
 107
 108 static uint32_t
 109 xen_hvm_cpuid_base(void)
 110 {
 111         uint32_t base, regs[4];
 112
 113         for (base = 0x40000000; base < 0x40010000; base += 0x100) {
 114                 do_cpuid(base, regs);
 115                 if (!memcmp("XenVMMXenVMM", &regs[1], 12)
 116                     && (regs[0] - base) >= 2)
 117                         return (base);
 118         }
 119         return (0);
 120 }
 121
 122 static void
 123 hypervisor_quirks(unsigned int major, unsigned int minor)
 124 {
 125 #ifdef SMP
 126         if (((major < 4) || (major == 4 && minor <= 5)) &&
 127             msix_disable_migration == -1) {
 128                 /*
 129                  * Xen hypervisors prior to 4.6.0 do not properly
 130                  * handle updates to enabled MSI-X table entries,
 131                  * so disable MSI-X interrupt migration in that
 132                  * case.
 133                  */
 134                 if (bootverbose)
 135                         printf(
 136 "Disabling MSI-X interrupt migration due to Xen hypervisor bug.\n"
 137 "Set machdep.msix_disable_migration=0 to forcefully enable it.\n");
 138                 msix_disable_migration = 1;
 139         }
 140 #endif
 141 }
 142
 143 static void
 144 hypervisor_version(void)
 145 {
 146         uint32_t regs[4];
 147         int major, minor;
 148
 149         do_cpuid(cpuid_base + 1, regs);
 150
 151         major = regs[0] >> 16;
 152         minor = regs[0] & 0xffff;
 153         printf("XEN: Hypervisor version %d.%d detected.\n", major, minor);
 154
 155         hypervisor_quirks(major, minor);
 156 }
 157
 158 /*
 159  * Allocate and fill in the hypcall page.
 160  */
 161 int
 162 xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type init_type)
 163 {
 164         uint32_t regs[4];
 165
 166         if (xen_domain() && init_type == XEN_HVM_INIT_LATE) {
 167                 /*
 168                  * If the domain type is already set we can assume that the
 169                  * hypercall page has been populated too, so just print the
 170                  * version (and apply any quirks) and exit.
 171                  */
 172                 hypervisor_version();
 173                 return 0;
 174         }
 175
 176         cpuid_base = xen_hvm_cpuid_base();
 177         if (cpuid_base == 0)
 178                 return (ENXIO);
 179
 180         if (init_type == XEN_HVM_INIT_LATE)
 181                 hypervisor_version();
 182
 183         /*
 184          * Find the hypercall pages.
 185          */
 186         do_cpuid(cpuid_base + 2, regs);
 187         if (regs[0] != 1)
 188                 return (EINVAL);
 189
 190         wrmsr(regs[1], (init_type == XEN_HVM_INIT_EARLY)
 191             ? ((vm_paddr_t)&hypercall_page - KERNBASE)
 192             : vtophys(&hypercall_page));
 193
 194         return (0);
 195 }
 196
 197 static void
 198 xen_hvm_init_shared_info_page(void)
 199 {
 200         struct xen_add_to_physmap xatp;
 201
 202         if (xen_pv_domain()) {
 203                 /*
 204                  * Already setup in the PV case, shared_info is passed inside
 205                  * of the start_info struct at start of day.
 206                  */
 207                 return;
 208         }
 209
 210         if (HYPERVISOR_shared_info == NULL) {
 211                 HYPERVISOR_shared_info = malloc(PAGE_SIZE, M_XENHVM, M_NOWAIT);
 212                 if (HYPERVISOR_shared_info == NULL)
 213                         panic("Unable to allocate Xen shared info page");
 214         }
 215
 216         xatp.domid = DOMID_SELF;
 217         xatp.idx = 0;
 218         xatp.space = XENMAPSPACE_shared_info;
 219         xatp.gpfn = vtophys(HYPERVISOR_shared_info) >> PAGE_SHIFT;
 220         if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
 221                 panic("HYPERVISOR_memory_op failed");
 222 }
 223
 224 /*
 225  * Tell the hypervisor how to contact us for event channel callbacks.
 226  */
 227 void
 228 xen_hvm_set_callback(device_t dev)
 229 {
 230         struct xen_hvm_param xhp;
 231         int irq;
 232
 233         if (xen_vector_callback_enabled)
 234                 return;
 235
 236         xhp.domid = DOMID_SELF;
 237         xhp.index = HVM_PARAM_CALLBACK_IRQ;
 238         if (xen_feature(XENFEAT_hvm_callback_vector) != 0) {
 239                 int error;
 240
 241                 xhp.value = HVM_CALLBACK_VECTOR(IDT_EVTCHN);
 242                 error = HYPERVISOR_hvm_op(HVMOP_set_param, &xhp);
 243                 if (error == 0) {
 244                         xen_vector_callback_enabled = 1;
 245                         return;
 246                 }
 247                 printf("Xen HVM callback vector registration failed (%d). "
 248                     "Falling back to emulated device interrupt\n", error);
 249         }
 250         xen_vector_callback_enabled = 0;
 251         if (dev == NULL) {
 252                 /*
 253                  * Called from early boot or resume.
 254                  * xenpci will invoke us again later.
 255                  */
 256                 return;
 257         }
 258
 259         irq = pci_get_irq(dev);
 260         if (irq < 16) {
 261                 xhp.value = HVM_CALLBACK_GSI(irq);
 262         } else {
 263                 u_int slot;
 264                 u_int pin;
 265
 266                 slot = pci_get_slot(dev);
 267                 pin = pci_get_intpin(dev) - 1;
 268                 xhp.value = HVM_CALLBACK_PCI_INTX(slot, pin);
 269         }
 270
 271         if (HYPERVISOR_hvm_op(HVMOP_set_param, &xhp) != 0)
 272                 panic("Can't set evtchn callback");
 273 }
 274
 275 #define XEN_MAGIC_IOPORT 0x10
 276 enum {
 277         XMI_MAGIC                        = 0x49d2,
 278         XMI_UNPLUG_IDE_DISKS             = 0x01,
 279         XMI_UNPLUG_NICS                  = 0x02,
 280         XMI_UNPLUG_IDE_EXCEPT_PRI_MASTER = 0x04
 281 };
 282
 283 static void
 284 xen_hvm_disable_emulated_devices(void)
 285 {
 286         u_short disable_devs = 0;
 287
 288         if (xen_pv_domain()) {
 289                 /*
 290                  * No emulated devices in the PV case, so no need to unplug
 291                  * anything.
 292                  */
 293                 if (xen_disable_pv_disks != 0 || xen_disable_pv_nics != 0)
 294                         printf("PV devices cannot be disabled in PV guests\n");
 295                 return;
 296         }
 297
 298         if (inw(XEN_MAGIC_IOPORT) != XMI_MAGIC)
 299                 return;
 300
 301         if (xen_disable_pv_disks == 0) {
 302                 if (bootverbose)
 303                         printf("XEN: disabling emulated disks\n");
 304                 disable_devs |= XMI_UNPLUG_IDE_DISKS;
 305         }
 306         if (xen_disable_pv_nics == 0) {
 307                 if (bootverbose)
 308                         printf("XEN: disabling emulated nics\n");
 309                 disable_devs |= XMI_UNPLUG_NICS;
 310         }
 311
 312         if (disable_devs != 0)
 313                 outw(XEN_MAGIC_IOPORT, disable_devs);
 314 }
 315
 316 static void
 317 xen_hvm_init(enum xen_hvm_init_type init_type)
 318 {
 319         int error;
 320         int i;
 321
 322         if (init_type == XEN_HVM_INIT_CANCELLED_SUSPEND)
 323                 return;
 324
 325         error = xen_hvm_init_hypercall_stubs(init_type);
 326
 327         switch (init_type) {
 328         case XEN_HVM_INIT_LATE:
 329                 if (error != 0)
 330                         return;
 331
 332                 /*
 333                  * If xen_domain_type is not set at this point
 334                  * it means we are inside a (PV)HVM guest, because
 335                  * for PVH the guest type is set much earlier
 336                  * (see hammer_time_xen).
 337                  */
 338                 if (!xen_domain()) {
 339                         xen_domain_type = XEN_HVM_DOMAIN;
 340                         vm_guest = VM_GUEST_XEN;
 341                 }
 342
 343                 setup_xen_features();
 344 #ifdef SMP
 345                 cpu_ops = xen_hvm_cpu_ops;
 346 #endif
 347                 break;
 348         case XEN_HVM_INIT_RESUME:
 349                 if (error != 0)
 350                         panic("Unable to init Xen hypercall stubs on resume");
 351
 352                 /* Clear stale vcpu_info. */
 353                 CPU_FOREACH(i)
 354                         DPCPU_ID_SET(i, vcpu_info, NULL);
 355                 break;
 356         default:
 357                 panic("Unsupported HVM initialization type");
 358         }
 359
 360         xen_vector_callback_enabled = 0;
 361         xen_hvm_set_callback(NULL);
 362
 363         /*
 364          * On (PV)HVM domains we need to request the hypervisor to
 365          * fill the shared info page, for PVH guest the shared_info page
 366          * is passed inside the start_info struct and is already set, so this
 367          * functions are no-ops.
 368          */
 369         xen_hvm_init_shared_info_page();
 370         xen_hvm_disable_emulated_devices();
 371 }
 372
 373 void
 374 xen_hvm_suspend(void)
 375 {
 376 }
 377
 378 void
 379 xen_hvm_resume(bool suspend_cancelled)
 380 {
 381
 382         xen_hvm_init(suspend_cancelled ?
 383             XEN_HVM_INIT_CANCELLED_SUSPEND : XEN_HVM_INIT_RESUME);
 384
 385         /* Register vcpu_info area for CPU#0. */
 386         xen_hvm_cpu_init();
 387 }
 388
 389 static void
 390 xen_hvm_sysinit(void *arg __unused)
 391 {
 392         xen_hvm_init(XEN_HVM_INIT_LATE);
 393 }
 394 SYSINIT(xen_hvm_init, SI_SUB_HYPERVISOR, SI_ORDER_FIRST, xen_hvm_sysinit, NULL);
 395
 396 static void
 397 xen_hvm_cpu_init(void)
 398 {
 399         struct vcpu_register_vcpu_info info;
 400         struct vcpu_info *vcpu_info;
 401         uint32_t regs[4];
 402         int cpu, rc;
 403
 404         if (!xen_domain())
 405                 return;
 406
 407         if (DPCPU_GET(vcpu_info) != NULL) {
 408                 /*
 409                  * vcpu_info is already set.  We're resuming
 410                  * from a failed migration and our pre-suspend
 411                  * configuration is still valid.
 412                  */
 413                 return;
 414         }
 415
 416         /*
 417          * Set vCPU ID. If available fetch the ID from CPUID, if not just use
 418          * the ACPI ID.
 419          */
 420         KASSERT(cpuid_base != 0, ("Invalid base Xen CPUID leaf"));
 421         cpuid_count(cpuid_base + 4, 0, regs);
 422         PCPU_SET(vcpu_id, (regs[0] & XEN_HVM_CPUID_VCPU_ID_PRESENT) ?
 423             regs[1] : PCPU_GET(acpi_id));
 424
 425         /*
 426          * Set the vCPU info.
 427          *
 428          * NB: the vCPU info for vCPUs < 32 can be fetched from the shared info
 429          * page, but in order to make sure the mapping code is correct always
 430          * attempt to map the vCPU info at a custom place.
 431          */
 432         vcpu_info = DPCPU_PTR(vcpu_local_info);
 433         cpu = PCPU_GET(vcpu_id);
 434         info.mfn = vtophys(vcpu_info) >> PAGE_SHIFT;
 435         info.offset = vtophys(vcpu_info) - trunc_page(vtophys(vcpu_info));
 436
 437         rc = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
 438         if (rc != 0)
 439                 DPCPU_SET(vcpu_info, &HYPERVISOR_shared_info->vcpu_info[cpu]);
 440         else
 441                 DPCPU_SET(vcpu_info, vcpu_info);
 442 }
 443 SYSINIT(xen_hvm_cpu_init, SI_SUB_INTR, SI_ORDER_FIRST, xen_hvm_cpu_init, NULL);
 444
 445 /* HVM/PVH start_info accessors */
 446 static vm_paddr_t
 447 hvm_get_xenstore_mfn(void)
 448 {
 449
 450         return (hvm_get_parameter(HVM_PARAM_STORE_PFN));
 451 }
 452
 453 static evtchn_port_t
 454 hvm_get_xenstore_evtchn(void)
 455 {
 456
 457         return (hvm_get_parameter(HVM_PARAM_STORE_EVTCHN));
 458 }
 459
 460 static vm_paddr_t
 461 hvm_get_console_mfn(void)
 462 {
 463
 464         return (hvm_get_parameter(HVM_PARAM_CONSOLE_PFN));
 465 }
 466
 467 static evtchn_port_t
 468 hvm_get_console_evtchn(void)
 469 {
 470
 471         return (hvm_get_parameter(HVM_PARAM_CONSOLE_EVTCHN));
 472 }
 473
 474 static uint32_t
 475 hvm_get_start_flags(void)
 476 {
 477
 478         return (hvm_start_flags);
 479 }
 480
 481 struct hypervisor_info hypervisor_info = {
 482         .get_xenstore_mfn               = hvm_get_xenstore_mfn,
 483         .get_xenstore_evtchn            = hvm_get_xenstore_evtchn,
 484         .get_console_mfn                = hvm_get_console_mfn,
 485         .get_console_evtchn             = hvm_get_console_evtchn,
 486         .get_start_flags                = hvm_get_start_flags,
 487 };