sys/x86/xen/hvm.c

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