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