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
  44 #include <dev/pci/pcivar.h>
  45
  46 #include <machine/cpufunc.h>
  47 #include <machine/cpu.h>
  48 #include <machine/smp.h>
  49
  50 #include <x86/apicreg.h>
  51
  52 #include <xen/xen-os.h>
  53 #include <xen/features.h>
  54 #include <xen/gnttab.h>
  55 #include <xen/hypervisor.h>
  56 #include <xen/hvm.h>
  57 #include <xen/xen_intr.h>
  58
  59 #include <xen/interface/hvm/params.h>
  60 #include <xen/interface/vcpu.h>
  61
  62 /*--------------------------- Forward Declarations ---------------------------*/
  63 static void xen_hvm_cpu_init(void);
  64
  65 /*-------------------------------- Local Types -------------------------------*/
  66 enum xen_hvm_init_type {
  67         XEN_HVM_INIT_COLD,
  68         XEN_HVM_INIT_CANCELLED_SUSPEND,
  69         XEN_HVM_INIT_RESUME
  70 };
  71
  72 /*-------------------------------- Global Data -------------------------------*/
  73 enum xen_domain_type xen_domain_type = XEN_NATIVE;
  74
  75 #ifdef SMP
  76 struct cpu_ops xen_hvm_cpu_ops = {
  77         .cpu_init       = xen_hvm_cpu_init,
  78         .cpu_resume     = xen_hvm_cpu_init
  79 };
  80 #endif
  81
  82 static MALLOC_DEFINE(M_XENHVM, "xen_hvm", "Xen HVM PV Support");
  83
  84 /**
  85  * If non-zero, the hypervisor has been configured to use a direct
  86  * IDT event callback for interrupt injection.
  87  */
  88 int xen_vector_callback_enabled;
  89
  90 /*------------------------------- Per-CPU Data -------------------------------*/
  91 DPCPU_DEFINE(struct vcpu_info, vcpu_local_info);
  92 DPCPU_DEFINE(struct vcpu_info *, vcpu_info);
  93
  94 /*------------------ Hypervisor Access Shared Memory Regions -----------------*/
  95 shared_info_t *HYPERVISOR_shared_info;
  96 start_info_t *HYPERVISOR_start_info;
  97
  98
  99 /*------------------------------ Sysctl tunables -----------------------------*/
 100 int xen_disable_pv_disks = 0;
 101 int xen_disable_pv_nics = 0;
 102 TUNABLE_INT("hw.xen.disable_pv_disks", &xen_disable_pv_disks);
 103 TUNABLE_INT("hw.xen.disable_pv_nics", &xen_disable_pv_nics);
 104
 105 /*---------------------- XEN Hypervisor Probe and Setup ----------------------*/
 106 static uint32_t
 107 xen_hvm_cpuid_base(void)
 108 {
 109         uint32_t base, regs[4];
 110
 111         for (base = 0x40000000; base < 0x40010000; base += 0x100) {
 112                 do_cpuid(base, regs);
 113                 if (!memcmp("XenVMMXenVMM", &regs[1], 12)
 114                     && (regs[0] - base) >= 2)
 115                         return (base);
 116         }
 117         return (0);
 118 }
 119
 120 /*
 121  * Allocate and fill in the hypcall page.
 122  */
 123 static int
 124 xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type init_type)
 125 {
 126         uint32_t base, regs[4];
 127         int i;
 128
 129         if (xen_pv_domain()) {
 130                 /* hypercall page is already set in the PV case */
 131                 return (0);
 132         }
 133
 134         base = xen_hvm_cpuid_base();
 135         if (base == 0)
 136                 return (ENXIO);
 137
 138         if (init_type == XEN_HVM_INIT_COLD) {
 139                 int major, minor;
 140
 141                 do_cpuid(base + 1, regs);
 142
 143                 major = regs[0] >> 16;
 144                 minor = regs[0] & 0xffff;
 145                 printf("XEN: Hypervisor version %d.%d detected.\n", major,
 146                         minor);
 147
 148 #ifdef SMP
 149                 if (((major < 4) || (major == 4 && minor <= 5)) &&
 150                     msix_disable_migration == -1) {
 151                         /*
 152                          * Xen hypervisors prior to 4.6.0 do not properly
 153                          * handle updates to enabled MSI-X table entries,
 154                          * so disable MSI-X interrupt migration in that
 155                          * case.
 156                          */
 157                         if (bootverbose)
 158                                 printf(
 159 "Disabling MSI-X interrupt migration due to Xen hypervisor bug.\n"
 160 "Set machdep.msix_disable_migration=0 to forcefully enable it.\n");
 161                         msix_disable_migration = 1;
 162                 }
 163 #endif
 164         }
 165
 166         /*
 167          * Find the hypercall pages.
 168          */
 169         do_cpuid(base + 2, regs);
 170
 171         for (i = 0; i < regs[0]; i++)
 172                 wrmsr(regs[1], vtophys(&hypercall_page + i * PAGE_SIZE) + i);
 173
 174         return (0);
 175 }
 176
 177 static void
 178 xen_hvm_init_shared_info_page(void)
 179 {
 180         struct xen_add_to_physmap xatp;
 181
 182         if (xen_pv_domain()) {
 183                 /*
 184                  * Already setup in the PV case, shared_info is passed inside
 185                  * of the start_info struct at start of day.
 186                  */
 187                 return;
 188         }
 189
 190         if (HYPERVISOR_shared_info == NULL) {
 191                 HYPERVISOR_shared_info = malloc(PAGE_SIZE, M_XENHVM, M_NOWAIT);
 192                 if (HYPERVISOR_shared_info == NULL)
 193                         panic("Unable to allocate Xen shared info page");
 194         }
 195
 196         xatp.domid = DOMID_SELF;
 197         xatp.idx = 0;
 198         xatp.space = XENMAPSPACE_shared_info;
 199         xatp.gpfn = vtophys(HYPERVISOR_shared_info) >> PAGE_SHIFT;
 200         if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
 201                 panic("HYPERVISOR_memory_op failed");
 202 }
 203
 204 /*
 205  * Tell the hypervisor how to contact us for event channel callbacks.
 206  */
 207 void
 208 xen_hvm_set_callback(device_t dev)
 209 {
 210         struct xen_hvm_param xhp;
 211         int irq;
 212
 213         if (xen_vector_callback_enabled)
 214                 return;
 215
 216         xhp.domid = DOMID_SELF;
 217         xhp.index = HVM_PARAM_CALLBACK_IRQ;
 218         if (xen_feature(XENFEAT_hvm_callback_vector) != 0) {
 219                 int error;
 220
 221                 xhp.value = HVM_CALLBACK_VECTOR(IDT_EVTCHN);
 222                 error = HYPERVISOR_hvm_op(HVMOP_set_param, &xhp);
 223                 if (error == 0) {
 224                         xen_vector_callback_enabled = 1;
 225                         return;
 226                 }
 227                 printf("Xen HVM callback vector registration failed (%d). "
 228                     "Falling back to emulated device interrupt\n", error);
 229         }
 230         xen_vector_callback_enabled = 0;
 231         if (dev == NULL) {
 232                 /*
 233                  * Called from early boot or resume.
 234                  * xenpci will invoke us again later.
 235                  */
 236                 return;
 237         }
 238
 239         irq = pci_get_irq(dev);
 240         if (irq < 16) {
 241                 xhp.value = HVM_CALLBACK_GSI(irq);
 242         } else {
 243                 u_int slot;
 244                 u_int pin;
 245
 246                 slot = pci_get_slot(dev);
 247                 pin = pci_get_intpin(dev) - 1;
 248                 xhp.value = HVM_CALLBACK_PCI_INTX(slot, pin);
 249         }
 250
 251         if (HYPERVISOR_hvm_op(HVMOP_set_param, &xhp) != 0)
 252                 panic("Can't set evtchn callback");
 253 }
 254
 255 #define XEN_MAGIC_IOPORT 0x10
 256 enum {
 257         XMI_MAGIC                        = 0x49d2,
 258         XMI_UNPLUG_IDE_DISKS             = 0x01,
 259         XMI_UNPLUG_NICS                  = 0x02,
 260         XMI_UNPLUG_IDE_EXCEPT_PRI_MASTER = 0x04
 261 };
 262
 263 static void
 264 xen_hvm_disable_emulated_devices(void)
 265 {
 266         u_short disable_devs = 0;
 267
 268         if (xen_pv_domain()) {
 269                 /*
 270                  * No emulated devices in the PV case, so no need to unplug
 271                  * anything.
 272                  */
 273                 if (xen_disable_pv_disks != 0 || xen_disable_pv_nics != 0)
 274                         printf("PV devices cannot be disabled in PV guests\n");
 275                 return;
 276         }
 277
 278         if (inw(XEN_MAGIC_IOPORT) != XMI_MAGIC)
 279                 return;
 280
 281         if (xen_disable_pv_disks == 0) {
 282                 if (bootverbose)
 283                         printf("XEN: disabling emulated disks\n");
 284                 disable_devs |= XMI_UNPLUG_IDE_DISKS;
 285         }
 286         if (xen_disable_pv_nics == 0) {
 287                 if (bootverbose)
 288                         printf("XEN: disabling emulated nics\n");
 289                 disable_devs |= XMI_UNPLUG_NICS;
 290         }
 291
 292         if (disable_devs != 0)
 293                 outw(XEN_MAGIC_IOPORT, disable_devs);
 294 }
 295
 296 static void
 297 xen_hvm_init(enum xen_hvm_init_type init_type)
 298 {
 299         int error;
 300         int i;
 301
 302         if (init_type == XEN_HVM_INIT_CANCELLED_SUSPEND)
 303                 return;
 304
 305         error = xen_hvm_init_hypercall_stubs(init_type);
 306
 307         switch (init_type) {
 308         case XEN_HVM_INIT_COLD:
 309                 if (error != 0)
 310                         return;
 311
 312                 /*
 313                  * If xen_domain_type is not set at this point
 314                  * it means we are inside a (PV)HVM guest, because
 315                  * for PVH the guest type is set much earlier
 316                  * (see hammer_time_xen).
 317                  */
 318                 if (!xen_domain()) {
 319                         xen_domain_type = XEN_HVM_DOMAIN;
 320                         vm_guest = VM_GUEST_XEN;
 321                 }
 322
 323                 setup_xen_features();
 324 #ifdef SMP
 325                 cpu_ops = xen_hvm_cpu_ops;
 326 #endif
 327                 break;
 328         case XEN_HVM_INIT_RESUME:
 329                 if (error != 0)
 330                         panic("Unable to init Xen hypercall stubs on resume");
 331
 332                 /* Clear stale vcpu_info. */
 333                 CPU_FOREACH(i)
 334                         DPCPU_ID_SET(i, vcpu_info, NULL);
 335                 break;
 336         default:
 337                 panic("Unsupported HVM initialization type");
 338         }
 339
 340         xen_vector_callback_enabled = 0;
 341         xen_hvm_set_callback(NULL);
 342
 343         /*
 344          * On (PV)HVM domains we need to request the hypervisor to
 345          * fill the shared info page, for PVH guest the shared_info page
 346          * is passed inside the start_info struct and is already set, so this
 347          * functions are no-ops.
 348          */
 349         xen_hvm_init_shared_info_page();
 350         xen_hvm_disable_emulated_devices();
 351 }
 352
 353 void
 354 xen_hvm_suspend(void)
 355 {
 356 }
 357
 358 void
 359 xen_hvm_resume(bool suspend_cancelled)
 360 {
 361
 362         xen_hvm_init(suspend_cancelled ?
 363             XEN_HVM_INIT_CANCELLED_SUSPEND : XEN_HVM_INIT_RESUME);
 364
 365         /* Register vcpu_info area for CPU#0. */
 366         xen_hvm_cpu_init();
 367 }
 368
 369 static void
 370 xen_hvm_sysinit(void *arg __unused)
 371 {
 372         xen_hvm_init(XEN_HVM_INIT_COLD);
 373 }
 374
 375 static void
 376 xen_set_vcpu_id(void)
 377 {
 378         struct pcpu *pc;
 379         int i;
 380
 381         if (!xen_hvm_domain())
 382                 return;
 383
 384         /* Set vcpu_id to acpi_id */
 385         CPU_FOREACH(i) {
 386                 pc = pcpu_find(i);
 387                 pc->pc_vcpu_id = pc->pc_acpi_id;
 388                 if (bootverbose)
 389                         printf("XEN: CPU %u has VCPU ID %u\n",
 390                                i, pc->pc_vcpu_id);
 391         }
 392 }
 393
 394 static void
 395 xen_hvm_cpu_init(void)
 396 {
 397         struct vcpu_register_vcpu_info info;
 398         struct vcpu_info *vcpu_info;
 399         int cpu, rc;
 400
 401         if (!xen_domain())
 402                 return;
 403
 404         if (DPCPU_GET(vcpu_info) != NULL) {
 405                 /*
 406                  * vcpu_info is already set.  We're resuming
 407                  * from a failed migration and our pre-suspend
 408                  * configuration is still valid.
 409                  */
 410                 return;
 411         }
 412
 413         vcpu_info = DPCPU_PTR(vcpu_local_info);
 414         cpu = PCPU_GET(vcpu_id);
 415         info.mfn = vtophys(vcpu_info) >> PAGE_SHIFT;
 416         info.offset = vtophys(vcpu_info) - trunc_page(vtophys(vcpu_info));
 417
 418         rc = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
 419         if (rc != 0)
 420                 DPCPU_SET(vcpu_info, &HYPERVISOR_shared_info->vcpu_info[cpu]);
 421         else
 422                 DPCPU_SET(vcpu_info, vcpu_info);
 423 }
 424
 425 SYSINIT(xen_hvm_init, SI_SUB_HYPERVISOR, SI_ORDER_FIRST, xen_hvm_sysinit, NULL);
 426 SYSINIT(xen_hvm_cpu_init, SI_SUB_INTR, SI_ORDER_FIRST, xen_hvm_cpu_init, NULL);
 427 SYSINIT(xen_set_vcpu_id, SI_SUB_CPU, SI_ORDER_ANY, xen_set_vcpu_id, NULL);