arch/x86/xen/enlighten.c

   1 /*
   2  * Core of Xen paravirt_ops implementation.
   3  *
   4  * This file contains the xen_paravirt_ops structure itself, and the
   5  * implementations for:
   6  * - privileged instructions
   7  * - interrupt flags
   8  * - segment operations
   9  * - booting and setup
  10  *
  11  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  12  */
  13
  14 #include <linux/cpu.h>
  15 #include <linux/kernel.h>
  16 #include <linux/init.h>
  17 #include <linux/smp.h>
  18 #include <linux/preempt.h>
  19 #include <linux/hardirq.h>
  20 #include <linux/percpu.h>
  21 #include <linux/delay.h>
  22 #include <linux/start_kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/kprobes.h>
  25 #include <linux/bootmem.h>
  26 #include <linux/module.h>
  27 #include <linux/mm.h>
  28 #include <linux/page-flags.h>
  29 #include <linux/highmem.h>
  30 #include <linux/console.h>
  31 #include <linux/pci.h>
  32 #include <linux/gfp.h>
  33 #include <linux/memblock.h>
  34
  35 #include <xen/xen.h>
  36 #include <xen/interface/xen.h>
  37 #include <xen/interface/version.h>
  38 #include <xen/interface/physdev.h>
  39 #include <xen/interface/vcpu.h>
  40 #include <xen/interface/memory.h>
  41 #include <xen/features.h>
  42 #include <xen/page.h>
  43 #include <xen/hvm.h>
  44 #include <xen/hvc-console.h>
  45
  46 #include <asm/paravirt.h>
  47 #include <asm/apic.h>
  48 #include <asm/page.h>
  49 #include <asm/xen/pci.h>
  50 #include <asm/xen/hypercall.h>
  51 #include <asm/xen/hypervisor.h>
  52 #include <asm/fixmap.h>
  53 #include <asm/processor.h>
  54 #include <asm/proto.h>
  55 #include <asm/msr-index.h>
  56 #include <asm/traps.h>
  57 #include <asm/setup.h>
  58 #include <asm/desc.h>
  59 #include <asm/pgalloc.h>
  60 #include <asm/pgtable.h>
  61 #include <asm/tlbflush.h>
  62 #include <asm/reboot.h>
  63 #include <asm/stackprotector.h>
  64 #include <asm/hypervisor.h>
  65
  66 #include "xen-ops.h"
  67 #include "mmu.h"
  68 #include "multicalls.h"
  69
  70 EXPORT_SYMBOL_GPL(hypercall_page);
  71
  72 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
  73 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
  74
  75 enum xen_domain_type xen_domain_type = XEN_NATIVE;
  76 EXPORT_SYMBOL_GPL(xen_domain_type);
  77
  78 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
  79 EXPORT_SYMBOL(machine_to_phys_mapping);
  80 unsigned int   machine_to_phys_order;
  81 EXPORT_SYMBOL(machine_to_phys_order);
  82
  83 struct start_info *xen_start_info;
  84 EXPORT_SYMBOL_GPL(xen_start_info);
  85
  86 struct shared_info xen_dummy_shared_info;
  87
  88 void *xen_initial_gdt;
  89
  90 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
  91 __read_mostly int xen_have_vector_callback;
  92 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
  93
  94 /*
  95  * Point at some empty memory to start with. We map the real shared_info
  96  * page as soon as fixmap is up and running.
  97  */
  98 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
  99
 100 /*
 101  * Flag to determine whether vcpu info placement is available on all
 102  * VCPUs.  We assume it is to start with, and then set it to zero on
 103  * the first failure.  This is because it can succeed on some VCPUs
 104  * and not others, since it can involve hypervisor memory allocation,
 105  * or because the guest failed to guarantee all the appropriate
 106  * constraints on all VCPUs (ie buffer can't cross a page boundary).
 107  *
 108  * Note that any particular CPU may be using a placed vcpu structure,
 109  * but we can only optimise if the all are.
 110  *
 111  * 0: not available, 1: available
 112  */
 113 static int have_vcpu_info_placement = 1;
 114
 115 static void clamp_max_cpus(void)
 116 {
 117 #ifdef CONFIG_SMP
 118         if (setup_max_cpus > MAX_VIRT_CPUS)
 119                 setup_max_cpus = MAX_VIRT_CPUS;
 120 #endif
 121 }
 122
 123 static void xen_vcpu_setup(int cpu)
 124 {
 125         struct vcpu_register_vcpu_info info;
 126         int err;
 127         struct vcpu_info *vcpup;
 128
 129         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
 130
 131         if (cpu < MAX_VIRT_CPUS)
 132                 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
 133
 134         if (!have_vcpu_info_placement) {
 135                 if (cpu >= MAX_VIRT_CPUS)
 136                         clamp_max_cpus();
 137                 return;
 138         }
 139
 140         vcpup = &per_cpu(xen_vcpu_info, cpu);
 141         info.mfn = arbitrary_virt_to_mfn(vcpup);
 142         info.offset = offset_in_page(vcpup);
 143
 144         /* Check to see if the hypervisor will put the vcpu_info
 145            structure where we want it, which allows direct access via
 146            a percpu-variable. */
 147         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
 148
 149         if (err) {
 150                 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
 151                 have_vcpu_info_placement = 0;
 152                 clamp_max_cpus();
 153         } else {
 154                 /* This cpu is using the registered vcpu info, even if
 155                    later ones fail to. */
 156                 per_cpu(xen_vcpu, cpu) = vcpup;
 157         }
 158 }
 159
 160 /*
 161  * On restore, set the vcpu placement up again.
 162  * If it fails, then we're in a bad state, since
 163  * we can't back out from using it...
 164  */
 165 void xen_vcpu_restore(void)
 166 {
 167         int cpu;
 168
 169         for_each_online_cpu(cpu) {
 170                 bool other_cpu = (cpu != smp_processor_id());
 171
 172                 if (other_cpu &&
 173                     HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
 174                         BUG();
 175
 176                 xen_setup_runstate_info(cpu);
 177
 178                 if (have_vcpu_info_placement)
 179                         xen_vcpu_setup(cpu);
 180
 181                 if (other_cpu &&
 182                     HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
 183                         BUG();
 184         }
 185 }
 186
 187 static void __init xen_banner(void)
 188 {
 189         unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
 190         struct xen_extraversion extra;
 191         HYPERVISOR_xen_version(XENVER_extraversion, &extra);
 192
 193         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
 194                pv_info.name);
 195         printk(KERN_INFO "Xen version: %d.%d%s%s\n",
 196                version >> 16, version & 0xffff, extra.extraversion,
 197                xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
 198 }
 199
 200 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
 201 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
 202
 203 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
 204                       unsigned int *cx, unsigned int *dx)
 205 {
 206         unsigned maskebx = ~0;
 207         unsigned maskecx = ~0;
 208         unsigned maskedx = ~0;
 209
 210         /*
 211          * Mask out inconvenient features, to try and disable as many
 212          * unsupported kernel subsystems as possible.
 213          */
 214         switch (*ax) {
 215         case 1:
 216                 maskecx = cpuid_leaf1_ecx_mask;
 217                 maskedx = cpuid_leaf1_edx_mask;
 218                 break;
 219
 220         case 0xb:
 221                 /* Suppress extended topology stuff */
 222                 maskebx = 0;
 223                 break;
 224         }
 225
 226         asm(XEN_EMULATE_PREFIX "cpuid"
 227                 : "=a" (*ax),
 228                   "=b" (*bx),
 229                   "=c" (*cx),
 230                   "=d" (*dx)
 231                 : "0" (*ax), "2" (*cx));
 232
 233         *bx &= maskebx;
 234         *cx &= maskecx;
 235         *dx &= maskedx;
 236 }
 237
 238 static void __init xen_init_cpuid_mask(void)
 239 {
 240         unsigned int ax, bx, cx, dx;
 241         unsigned int xsave_mask;
 242
 243         cpuid_leaf1_edx_mask =
 244                 ~((1 << X86_FEATURE_MCE)  |  /* disable MCE */
 245                   (1 << X86_FEATURE_MCA)  |  /* disable MCA */
 246                   (1 << X86_FEATURE_MTRR) |  /* disable MTRR */
 247                   (1 << X86_FEATURE_ACC));   /* thermal monitoring */
 248
 249         if (!xen_initial_domain())
 250                 cpuid_leaf1_edx_mask &=
 251                         ~((1 << X86_FEATURE_APIC) |  /* disable local APIC */
 252                           (1 << X86_FEATURE_ACPI));  /* disable ACPI */
 253         ax = 1;
 254         xen_cpuid(&ax, &bx, &cx, &dx);
 255
 256         xsave_mask =
 257                 (1 << (X86_FEATURE_XSAVE % 32)) |
 258                 (1 << (X86_FEATURE_OSXSAVE % 32));
 259
 260         /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
 261         if ((cx & xsave_mask) != xsave_mask)
 262                 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
 263 }
 264
 265 static void xen_set_debugreg(int reg, unsigned long val)
 266 {
 267         HYPERVISOR_set_debugreg(reg, val);
 268 }
 269
 270 static unsigned long xen_get_debugreg(int reg)
 271 {
 272         return HYPERVISOR_get_debugreg(reg);
 273 }
 274
 275 static void xen_end_context_switch(struct task_struct *next)
 276 {
 277         xen_mc_flush();
 278         paravirt_end_context_switch(next);
 279 }
 280
 281 static unsigned long xen_store_tr(void)
 282 {
 283         return 0;
 284 }
 285
 286 /*
 287  * Set the page permissions for a particular virtual address.  If the
 288  * address is a vmalloc mapping (or other non-linear mapping), then
 289  * find the linear mapping of the page and also set its protections to
 290  * match.
 291  */
 292 static void set_aliased_prot(void *v, pgprot_t prot)
 293 {
 294         int level;
 295         pte_t *ptep;
 296         pte_t pte;
 297         unsigned long pfn;
 298         struct page *page;
 299
 300         ptep = lookup_address((unsigned long)v, &level);
 301         BUG_ON(ptep == NULL);
 302
 303         pfn = pte_pfn(*ptep);
 304         page = pfn_to_page(pfn);
 305
 306         pte = pfn_pte(pfn, prot);
 307
 308         if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
 309                 BUG();
 310
 311         if (!PageHighMem(page)) {
 312                 void *av = __va(PFN_PHYS(pfn));
 313
 314                 if (av != v)
 315                         if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
 316                                 BUG();
 317         } else
 318                 kmap_flush_unused();
 319 }
 320
 321 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
 322 {
 323         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
 324         int i;
 325
 326         for(i = 0; i < entries; i += entries_per_page)
 327                 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
 328 }
 329
 330 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
 331 {
 332         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
 333         int i;
 334
 335         for(i = 0; i < entries; i += entries_per_page)
 336                 set_aliased_prot(ldt + i, PAGE_KERNEL);
 337 }
 338
 339 static void xen_set_ldt(const void *addr, unsigned entries)
 340 {
 341         struct mmuext_op *op;
 342         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 343
 344         trace_xen_cpu_set_ldt(addr, entries);
 345
 346         op = mcs.args;
 347         op->cmd = MMUEXT_SET_LDT;
 348         op->arg1.linear_addr = (unsigned long)addr;
 349         op->arg2.nr_ents = entries;
 350
 351         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 352
 353         xen_mc_issue(PARAVIRT_LAZY_CPU);
 354 }
 355
 356 static void xen_load_gdt(const struct desc_ptr *dtr)
 357 {
 358         unsigned long va = dtr->address;
 359         unsigned int size = dtr->size + 1;
 360         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 361         unsigned long frames[pages];
 362         int f;
 363
 364         /*
 365          * A GDT can be up to 64k in size, which corresponds to 8192
 366          * 8-byte entries, or 16 4k pages..
 367          */
 368
 369         BUG_ON(size > 65536);
 370         BUG_ON(va & ~PAGE_MASK);
 371
 372         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
 373                 int level;
 374                 pte_t *ptep;
 375                 unsigned long pfn, mfn;
 376                 void *virt;
 377
 378                 /*
 379                  * The GDT is per-cpu and is in the percpu data area.
 380                  * That can be virtually mapped, so we need to do a
 381                  * page-walk to get the underlying MFN for the
 382                  * hypercall.  The page can also be in the kernel's
 383                  * linear range, so we need to RO that mapping too.
 384                  */
 385                 ptep = lookup_address(va, &level);
 386                 BUG_ON(ptep == NULL);
 387
 388                 pfn = pte_pfn(*ptep);
 389                 mfn = pfn_to_mfn(pfn);
 390                 virt = __va(PFN_PHYS(pfn));
 391
 392                 frames[f] = mfn;
 393
 394                 make_lowmem_page_readonly((void *)va);
 395                 make_lowmem_page_readonly(virt);
 396         }
 397
 398         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
 399                 BUG();
 400 }
 401
 402 /*
 403  * load_gdt for early boot, when the gdt is only mapped once
 404  */
 405 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
 406 {
 407         unsigned long va = dtr->address;
 408         unsigned int size = dtr->size + 1;
 409         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 410         unsigned long frames[pages];
 411         int f;
 412
 413         /*
 414          * A GDT can be up to 64k in size, which corresponds to 8192
 415          * 8-byte entries, or 16 4k pages..
 416          */
 417
 418         BUG_ON(size > 65536);
 419         BUG_ON(va & ~PAGE_MASK);
 420
 421         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
 422                 pte_t pte;
 423                 unsigned long pfn, mfn;
 424
 425                 pfn = virt_to_pfn(va);
 426                 mfn = pfn_to_mfn(pfn);
 427
 428                 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
 429
 430                 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
 431                         BUG();
 432
 433                 frames[f] = mfn;
 434         }
 435
 436         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
 437                 BUG();
 438 }
 439
 440 static void load_TLS_descriptor(struct thread_struct *t,
 441                                 unsigned int cpu, unsigned int i)
 442 {
 443         struct desc_struct *gdt = get_cpu_gdt_table(cpu);
 444         xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
 445         struct multicall_space mc = __xen_mc_entry(0);
 446
 447         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
 448 }
 449
 450 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
 451 {
 452         /*
 453          * XXX sleazy hack: If we're being called in a lazy-cpu zone
 454          * and lazy gs handling is enabled, it means we're in a
 455          * context switch, and %gs has just been saved.  This means we
 456          * can zero it out to prevent faults on exit from the
 457          * hypervisor if the next process has no %gs.  Either way, it
 458          * has been saved, and the new value will get loaded properly.
 459          * This will go away as soon as Xen has been modified to not
 460          * save/restore %gs for normal hypercalls.
 461          *
 462          * On x86_64, this hack is not used for %gs, because gs points
 463          * to KERNEL_GS_BASE (and uses it for PDA references), so we
 464          * must not zero %gs on x86_64
 465          *
 466          * For x86_64, we need to zero %fs, otherwise we may get an
 467          * exception between the new %fs descriptor being loaded and
 468          * %fs being effectively cleared at __switch_to().
 469          */
 470         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
 471 #ifdef CONFIG_X86_32
 472                 lazy_load_gs(0);
 473 #else
 474                 loadsegment(fs, 0);
 475 #endif
 476         }
 477
 478         xen_mc_batch();
 479
 480         load_TLS_descriptor(t, cpu, 0);
 481         load_TLS_descriptor(t, cpu, 1);
 482         load_TLS_descriptor(t, cpu, 2);
 483
 484         xen_mc_issue(PARAVIRT_LAZY_CPU);
 485 }
 486
 487 #ifdef CONFIG_X86_64
 488 static void xen_load_gs_index(unsigned int idx)
 489 {
 490         if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
 491                 BUG();
 492 }
 493 #endif
 494
 495 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
 496                                 const void *ptr)
 497 {
 498         xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
 499         u64 entry = *(u64 *)ptr;
 500
 501         trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
 502
 503         preempt_disable();
 504
 505         xen_mc_flush();
 506         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
 507                 BUG();
 508
 509         preempt_enable();
 510 }
 511
 512 static int cvt_gate_to_trap(int vector, const gate_desc *val,
 513                             struct trap_info *info)
 514 {
 515         unsigned long addr;
 516
 517         if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
 518                 return 0;
 519
 520         info->vector = vector;
 521
 522         addr = gate_offset(*val);
 523 #ifdef CONFIG_X86_64
 524         /*
 525          * Look for known traps using IST, and substitute them
 526          * appropriately.  The debugger ones are the only ones we care
 527          * about.  Xen will handle faults like double_fault and
 528          * machine_check, so we should never see them.  Warn if
 529          * there's an unexpected IST-using fault handler.
 530          */
 531         if (addr == (unsigned long)debug)
 532                 addr = (unsigned long)xen_debug;
 533         else if (addr == (unsigned long)int3)
 534                 addr = (unsigned long)xen_int3;
 535         else if (addr == (unsigned long)stack_segment)
 536                 addr = (unsigned long)xen_stack_segment;
 537         else if (addr == (unsigned long)double_fault ||
 538                  addr == (unsigned long)nmi) {
 539                 /* Don't need to handle these */
 540                 return 0;
 541 #ifdef CONFIG_X86_MCE
 542         } else if (addr == (unsigned long)machine_check) {
 543                 return 0;
 544 #endif
 545         } else {
 546                 /* Some other trap using IST? */
 547                 if (WARN_ON(val->ist != 0))
 548                         return 0;
 549         }
 550 #endif  /* CONFIG_X86_64 */
 551         info->address = addr;
 552
 553         info->cs = gate_segment(*val);
 554         info->flags = val->dpl;
 555         /* interrupt gates clear IF */
 556         if (val->type == GATE_INTERRUPT)
 557                 info->flags |= 1 << 2;
 558
 559         return 1;
 560 }
 561
 562 /* Locations of each CPU's IDT */
 563 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
 564
 565 /* Set an IDT entry.  If the entry is part of the current IDT, then
 566    also update Xen. */
 567 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
 568 {
 569         unsigned long p = (unsigned long)&dt[entrynum];
 570         unsigned long start, end;
 571
 572         trace_xen_cpu_write_idt_entry(dt, entrynum, g);
 573
 574         preempt_disable();
 575
 576         start = __this_cpu_read(idt_desc.address);
 577         end = start + __this_cpu_read(idt_desc.size) + 1;
 578
 579         xen_mc_flush();
 580
 581         native_write_idt_entry(dt, entrynum, g);
 582
 583         if (p >= start && (p + 8) <= end) {
 584                 struct trap_info info[2];
 585
 586                 info[1].address = 0;
 587
 588                 if (cvt_gate_to_trap(entrynum, g, &info[0]))
 589                         if (HYPERVISOR_set_trap_table(info))
 590                                 BUG();
 591         }
 592
 593         preempt_enable();
 594 }
 595
 596 static void xen_convert_trap_info(const struct desc_ptr *desc,
 597                                   struct trap_info *traps)
 598 {
 599         unsigned in, out, count;
 600
 601         count = (desc->size+1) / sizeof(gate_desc);
 602         BUG_ON(count > 256);
 603
 604         for (in = out = 0; in < count; in++) {
 605                 gate_desc *entry = (gate_desc*)(desc->address) + in;
 606
 607                 if (cvt_gate_to_trap(in, entry, &traps[out]))
 608                         out++;
 609         }
 610         traps[out].address = 0;
 611 }
 612
 613 void xen_copy_trap_info(struct trap_info *traps)
 614 {
 615         const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
 616
 617         xen_convert_trap_info(desc, traps);
 618 }
 619
 620 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
 621    hold a spinlock to protect the static traps[] array (static because
 622    it avoids allocation, and saves stack space). */
 623 static void xen_load_idt(const struct desc_ptr *desc)
 624 {
 625         static DEFINE_SPINLOCK(lock);
 626         static struct trap_info traps[257];
 627
 628         trace_xen_cpu_load_idt(desc);
 629
 630         spin_lock(&lock);
 631
 632         __get_cpu_var(idt_desc) = *desc;
 633
 634         xen_convert_trap_info(desc, traps);
 635
 636         xen_mc_flush();
 637         if (HYPERVISOR_set_trap_table(traps))
 638                 BUG();
 639
 640         spin_unlock(&lock);
 641 }
 642
 643 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
 644    they're handled differently. */
 645 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
 646                                 const void *desc, int type)
 647 {
 648         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
 649
 650         preempt_disable();
 651
 652         switch (type) {
 653         case DESC_LDT:
 654         case DESC_TSS:
 655                 /* ignore */
 656                 break;
 657
 658         default: {
 659                 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
 660
 661                 xen_mc_flush();
 662                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
 663                         BUG();
 664         }
 665
 666         }
 667
 668         preempt_enable();
 669 }
 670
 671 /*
 672  * Version of write_gdt_entry for use at early boot-time needed to
 673  * update an entry as simply as possible.
 674  */
 675 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
 676                                             const void *desc, int type)
 677 {
 678         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
 679
 680         switch (type) {
 681         case DESC_LDT:
 682         case DESC_TSS:
 683                 /* ignore */
 684                 break;
 685
 686         default: {
 687                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
 688
 689                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
 690                         dt[entry] = *(struct desc_struct *)desc;
 691         }
 692
 693         }
 694 }
 695
 696 static void xen_load_sp0(struct tss_struct *tss,
 697                          struct thread_struct *thread)
 698 {
 699         struct multicall_space mcs;
 700
 701         mcs = xen_mc_entry(0);
 702         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
 703         xen_mc_issue(PARAVIRT_LAZY_CPU);
 704 }
 705
 706 static void xen_set_iopl_mask(unsigned mask)
 707 {
 708         struct physdev_set_iopl set_iopl;
 709
 710         /* Force the change at ring 0. */
 711         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
 712         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 713 }
 714
 715 static void xen_io_delay(void)
 716 {
 717 }
 718
 719 #ifdef CONFIG_X86_LOCAL_APIC
 720 static u32 xen_apic_read(u32 reg)
 721 {
 722         return 0;
 723 }
 724
 725 static void xen_apic_write(u32 reg, u32 val)
 726 {
 727         /* Warn to see if there's any stray references */
 728         WARN_ON(1);
 729 }
 730
 731 static u64 xen_apic_icr_read(void)
 732 {
 733         return 0;
 734 }
 735
 736 static void xen_apic_icr_write(u32 low, u32 id)
 737 {
 738         /* Warn to see if there's any stray references */
 739         WARN_ON(1);
 740 }
 741
 742 static void xen_apic_wait_icr_idle(void)
 743 {
 744         return;
 745 }
 746
 747 static u32 xen_safe_apic_wait_icr_idle(void)
 748 {
 749         return 0;
 750 }
 751
 752 static void set_xen_basic_apic_ops(void)
 753 {
 754         apic->read = xen_apic_read;
 755         apic->write = xen_apic_write;
 756         apic->icr_read = xen_apic_icr_read;
 757         apic->icr_write = xen_apic_icr_write;
 758         apic->wait_icr_idle = xen_apic_wait_icr_idle;
 759         apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
 760 }
 761
 762 #endif
 763
 764 static void xen_clts(void)
 765 {
 766         struct multicall_space mcs;
 767
 768         mcs = xen_mc_entry(0);
 769
 770         MULTI_fpu_taskswitch(mcs.mc, 0);
 771
 772         xen_mc_issue(PARAVIRT_LAZY_CPU);
 773 }
 774
 775 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
 776
 777 static unsigned long xen_read_cr0(void)
 778 {
 779         unsigned long cr0 = percpu_read(xen_cr0_value);
 780
 781         if (unlikely(cr0 == 0)) {
 782                 cr0 = native_read_cr0();
 783                 percpu_write(xen_cr0_value, cr0);
 784         }
 785
 786         return cr0;
 787 }
 788
 789 static void xen_write_cr0(unsigned long cr0)
 790 {
 791         struct multicall_space mcs;
 792
 793         percpu_write(xen_cr0_value, cr0);
 794
 795         /* Only pay attention to cr0.TS; everything else is
 796            ignored. */
 797         mcs = xen_mc_entry(0);
 798
 799         MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
 800
 801         xen_mc_issue(PARAVIRT_LAZY_CPU);
 802 }
 803
 804 static void xen_write_cr4(unsigned long cr4)
 805 {
 806         cr4 &= ~X86_CR4_PGE;
 807         cr4 &= ~X86_CR4_PSE;
 808
 809         native_write_cr4(cr4);
 810 }
 811
 812 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
 813 {
 814         int ret;
 815
 816         ret = 0;
 817
 818         switch (msr) {
 819 #ifdef CONFIG_X86_64
 820                 unsigned which;
 821                 u64 base;
 822
 823         case MSR_FS_BASE:               which = SEGBASE_FS; goto set;
 824         case MSR_KERNEL_GS_BASE:        which = SEGBASE_GS_USER; goto set;
 825         case MSR_GS_BASE:               which = SEGBASE_GS_KERNEL; goto set;
 826
 827         set:
 828                 base = ((u64)high << 32) | low;
 829                 if (HYPERVISOR_set_segment_base(which, base) != 0)
 830                         ret = -EIO;
 831                 break;
 832 #endif
 833
 834         case MSR_STAR:
 835         case MSR_CSTAR:
 836         case MSR_LSTAR:
 837         case MSR_SYSCALL_MASK:
 838         case MSR_IA32_SYSENTER_CS:
 839         case MSR_IA32_SYSENTER_ESP:
 840         case MSR_IA32_SYSENTER_EIP:
 841                 /* Fast syscall setup is all done in hypercalls, so
 842                    these are all ignored.  Stub them out here to stop
 843                    Xen console noise. */
 844                 break;
 845
 846         case MSR_IA32_CR_PAT:
 847                 if (smp_processor_id() == 0)
 848                         xen_set_pat(((u64)high << 32) | low);
 849                 break;
 850
 851         default:
 852                 ret = native_write_msr_safe(msr, low, high);
 853         }
 854
 855         return ret;
 856 }
 857
 858 void xen_setup_shared_info(void)
 859 {
 860         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 861                 set_fixmap(FIX_PARAVIRT_BOOTMAP,
 862                            xen_start_info->shared_info);
 863
 864                 HYPERVISOR_shared_info =
 865                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
 866         } else
 867                 HYPERVISOR_shared_info =
 868                         (struct shared_info *)__va(xen_start_info->shared_info);
 869
 870 #ifndef CONFIG_SMP
 871         /* In UP this is as good a place as any to set up shared info */
 872         xen_setup_vcpu_info_placement();
 873 #endif
 874
 875         xen_setup_mfn_list_list();
 876 }
 877
 878 /* This is called once we have the cpu_possible_map */
 879 void xen_setup_vcpu_info_placement(void)
 880 {
 881         int cpu;
 882
 883         for_each_possible_cpu(cpu)
 884                 xen_vcpu_setup(cpu);
 885
 886         /* xen_vcpu_setup managed to place the vcpu_info within the
 887            percpu area for all cpus, so make use of it */
 888         if (have_vcpu_info_placement) {
 889                 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
 890                 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
 891                 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
 892                 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
 893                 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
 894         }
 895 }
 896
 897 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
 898                           unsigned long addr, unsigned len)
 899 {
 900         char *start, *end, *reloc;
 901         unsigned ret;
 902
 903         start = end = reloc = NULL;
 904
 905 #define SITE(op, x)                                                     \
 906         case PARAVIRT_PATCH(op.x):                                      \
 907         if (have_vcpu_info_placement) {                                 \
 908                 start = (char *)xen_##x##_direct;                       \
 909                 end = xen_##x##_direct_end;                             \
 910                 reloc = xen_##x##_direct_reloc;                         \
 911         }                                                               \
 912         goto patch_site
 913
 914         switch (type) {
 915                 SITE(pv_irq_ops, irq_enable);
 916                 SITE(pv_irq_ops, irq_disable);
 917                 SITE(pv_irq_ops, save_fl);
 918                 SITE(pv_irq_ops, restore_fl);
 919 #undef SITE
 920
 921         patch_site:
 922                 if (start == NULL || (end-start) > len)
 923                         goto default_patch;
 924
 925                 ret = paravirt_patch_insns(insnbuf, len, start, end);
 926
 927                 /* Note: because reloc is assigned from something that
 928                    appears to be an array, gcc assumes it's non-null,
 929                    but doesn't know its relationship with start and
 930                    end. */
 931                 if (reloc > start && reloc < end) {
 932                         int reloc_off = reloc - start;
 933                         long *relocp = (long *)(insnbuf + reloc_off);
 934                         long delta = start - (char *)addr;
 935
 936                         *relocp += delta;
 937                 }
 938                 break;
 939
 940         default_patch:
 941         default:
 942                 ret = paravirt_patch_default(type, clobbers, insnbuf,
 943                                              addr, len);
 944                 break;
 945         }
 946
 947         return ret;
 948 }
 949
 950 static const struct pv_info xen_info __initconst = {
 951         .paravirt_enabled = 1,
 952         .shared_kernel_pmd = 0,
 953
 954         .name = "Xen",
 955 };
 956
 957 static const struct pv_init_ops xen_init_ops __initconst = {
 958         .patch = xen_patch,
 959 };
 960
 961 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
 962         .cpuid = xen_cpuid,
 963
 964         .set_debugreg = xen_set_debugreg,
 965         .get_debugreg = xen_get_debugreg,
 966
 967         .clts = xen_clts,
 968
 969         .read_cr0 = xen_read_cr0,
 970         .write_cr0 = xen_write_cr0,
 971
 972         .read_cr4 = native_read_cr4,
 973         .read_cr4_safe = native_read_cr4_safe,
 974         .write_cr4 = xen_write_cr4,
 975
 976         .wbinvd = native_wbinvd,
 977
 978         .read_msr = native_read_msr_safe,
 979         .write_msr = xen_write_msr_safe,
 980         .read_tsc = native_read_tsc,
 981         .read_pmc = native_read_pmc,
 982
 983         .iret = xen_iret,
 984         .irq_enable_sysexit = xen_sysexit,
 985 #ifdef CONFIG_X86_64
 986         .usergs_sysret32 = xen_sysret32,
 987         .usergs_sysret64 = xen_sysret64,
 988 #endif
 989
 990         .load_tr_desc = paravirt_nop,
 991         .set_ldt = xen_set_ldt,
 992         .load_gdt = xen_load_gdt,
 993         .load_idt = xen_load_idt,
 994         .load_tls = xen_load_tls,
 995 #ifdef CONFIG_X86_64
 996         .load_gs_index = xen_load_gs_index,
 997 #endif
 998
 999         .alloc_ldt = xen_alloc_ldt,
1000         .free_ldt = xen_free_ldt,
1001
1002         .store_gdt = native_store_gdt,
1003         .store_idt = native_store_idt,
1004         .store_tr = xen_store_tr,
1005
1006         .write_ldt_entry = xen_write_ldt_entry,
1007         .write_gdt_entry = xen_write_gdt_entry,
1008         .write_idt_entry = xen_write_idt_entry,
1009         .load_sp0 = xen_load_sp0,
1010
1011         .set_iopl_mask = xen_set_iopl_mask,
1012         .io_delay = xen_io_delay,
1013
1014         /* Xen takes care of %gs when switching to usermode for us */
1015         .swapgs = paravirt_nop,
1016
1017         .start_context_switch = paravirt_start_context_switch,
1018         .end_context_switch = xen_end_context_switch,
1019 };
1020
1021 static const struct pv_apic_ops xen_apic_ops __initconst = {
1022 #ifdef CONFIG_X86_LOCAL_APIC
1023         .startup_ipi_hook = paravirt_nop,
1024 #endif
1025 };
1026
1027 static void xen_reboot(int reason)
1028 {
1029         struct sched_shutdown r = { .reason = reason };
1030
1031         if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1032                 BUG();
1033 }
1034
1035 static void xen_restart(char *msg)
1036 {
1037         xen_reboot(SHUTDOWN_reboot);
1038 }
1039
1040 static void xen_emergency_restart(void)
1041 {
1042         xen_reboot(SHUTDOWN_reboot);
1043 }
1044
1045 static void xen_machine_halt(void)
1046 {
1047         xen_reboot(SHUTDOWN_poweroff);
1048 }
1049
1050 static void xen_machine_power_off(void)
1051 {
1052         if (pm_power_off)
1053                 pm_power_off();
1054         xen_reboot(SHUTDOWN_poweroff);
1055 }
1056
1057 static void xen_crash_shutdown(struct pt_regs *regs)
1058 {
1059         xen_reboot(SHUTDOWN_crash);
1060 }
1061
1062 static int
1063 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1064 {
1065         xen_reboot(SHUTDOWN_crash);
1066         return NOTIFY_DONE;
1067 }
1068
1069 static struct notifier_block xen_panic_block = {
1070         .notifier_call= xen_panic_event,
1071 };
1072
1073 int xen_panic_handler_init(void)
1074 {
1075         atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1076         return 0;
1077 }
1078
1079 static const struct machine_ops xen_machine_ops __initconst = {
1080         .restart = xen_restart,
1081         .halt = xen_machine_halt,
1082         .power_off = xen_machine_power_off,
1083         .shutdown = xen_machine_halt,
1084         .crash_shutdown = xen_crash_shutdown,
1085         .emergency_restart = xen_emergency_restart,
1086 };
1087
1088 /*
1089  * Set up the GDT and segment registers for -fstack-protector.  Until
1090  * we do this, we have to be careful not to call any stack-protected
1091  * function, which is most of the kernel.
1092  */
1093 static void __init xen_setup_stackprotector(void)
1094 {
1095         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1096         pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1097
1098         setup_stack_canary_segment(0);
1099         switch_to_new_gdt(0);
1100
1101         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1102         pv_cpu_ops.load_gdt = xen_load_gdt;
1103 }
1104
1105 /* First C function to be called on Xen boot */
1106 asmlinkage void __init xen_start_kernel(void)
1107 {
1108         struct physdev_set_iopl set_iopl;
1109         int rc;
1110         pgd_t *pgd;
1111
1112         if (!xen_start_info)
1113                 return;
1114
1115         xen_domain_type = XEN_PV_DOMAIN;
1116
1117         xen_setup_machphys_mapping();
1118
1119         /* Install Xen paravirt ops */
1120         pv_info = xen_info;
1121         pv_init_ops = xen_init_ops;
1122         pv_cpu_ops = xen_cpu_ops;
1123         pv_apic_ops = xen_apic_ops;
1124
1125         x86_init.resources.memory_setup = xen_memory_setup;
1126         x86_init.oem.arch_setup = xen_arch_setup;
1127         x86_init.oem.banner = xen_banner;
1128
1129         xen_init_time_ops();
1130
1131         /*
1132          * Set up some pagetable state before starting to set any ptes.
1133          */
1134
1135         xen_init_mmu_ops();
1136
1137         /* Prevent unwanted bits from being set in PTEs. */
1138         __supported_pte_mask &= ~_PAGE_GLOBAL;
1139         if (!xen_initial_domain())
1140                 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1141
1142         __supported_pte_mask |= _PAGE_IOMAP;
1143
1144         /*
1145          * Prevent page tables from being allocated in highmem, even
1146          * if CONFIG_HIGHPTE is enabled.
1147          */
1148         __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1149
1150         /* Work out if we support NX */
1151         x86_configure_nx();
1152
1153         xen_setup_features();
1154
1155         /* Get mfn list */
1156         if (!xen_feature(XENFEAT_auto_translated_physmap))
1157                 xen_build_dynamic_phys_to_machine();
1158
1159         /*
1160          * Set up kernel GDT and segment registers, mainly so that
1161          * -fstack-protector code can be executed.
1162          */
1163         xen_setup_stackprotector();
1164
1165         xen_init_irq_ops();
1166         xen_init_cpuid_mask();
1167
1168 #ifdef CONFIG_X86_LOCAL_APIC
1169         /*
1170          * set up the basic apic ops.
1171          */
1172         set_xen_basic_apic_ops();
1173 #endif
1174
1175         if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1176                 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1177                 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1178         }
1179
1180         machine_ops = xen_machine_ops;
1181
1182         /*
1183          * The only reliable way to retain the initial address of the
1184          * percpu gdt_page is to remember it here, so we can go and
1185          * mark it RW later, when the initial percpu area is freed.
1186          */
1187         xen_initial_gdt = &per_cpu(gdt_page, 0);
1188
1189         xen_smp_init();
1190
1191 #ifdef CONFIG_ACPI_NUMA
1192         /*
1193          * The pages we from Xen are not related to machine pages, so
1194          * any NUMA information the kernel tries to get from ACPI will
1195          * be meaningless.  Prevent it from trying.
1196          */
1197         acpi_numa = -1;
1198 #endif
1199
1200         pgd = (pgd_t *)xen_start_info->pt_base;
1201
1202         if (!xen_initial_domain())
1203                 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1204
1205         __supported_pte_mask |= _PAGE_IOMAP;
1206         /* Don't do the full vcpu_info placement stuff until we have a
1207            possible map and a non-dummy shared_info. */
1208         per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1209
1210         local_irq_disable();
1211         early_boot_irqs_disabled = true;
1212
1213         memblock_init();
1214
1215         xen_raw_console_write("mapping kernel into physical memory\n");
1216         pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1217         xen_ident_map_ISA();
1218
1219         /* Allocate and initialize top and mid mfn levels for p2m structure */
1220         xen_build_mfn_list_list();
1221
1222         /* keep using Xen gdt for now; no urgent need to change it */
1223
1224 #ifdef CONFIG_X86_32
1225         pv_info.kernel_rpl = 1;
1226         if (xen_feature(XENFEAT_supervisor_mode_kernel))
1227                 pv_info.kernel_rpl = 0;
1228 #else
1229         pv_info.kernel_rpl = 0;
1230 #endif
1231         /* set the limit of our address space */
1232         xen_reserve_top();
1233
1234         /* We used to do this in xen_arch_setup, but that is too late on AMD
1235          * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1236          * which pokes 0xcf8 port.
1237          */
1238         set_iopl.iopl = 1;
1239         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1240         if (rc != 0)
1241                 xen_raw_printk("physdev_op failed %d\n", rc);
1242
1243 #ifdef CONFIG_X86_32
1244         /* set up basic CPUID stuff */
1245         cpu_detect(&new_cpu_data);
1246         new_cpu_data.hard_math = 1;
1247         new_cpu_data.wp_works_ok = 1;
1248         new_cpu_data.x86_capability[0] = cpuid_edx(1);
1249 #endif
1250
1251         /* Poke various useful things into boot_params */
1252         boot_params.hdr.type_of_loader = (9 << 4) | 0;
1253         boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1254                 ? __pa(xen_start_info->mod_start) : 0;
1255         boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1256         boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1257
1258         if (!xen_initial_domain()) {
1259                 add_preferred_console("xenboot", 0, NULL);
1260                 add_preferred_console("tty", 0, NULL);
1261                 add_preferred_console("hvc", 0, NULL);
1262                 if (pci_xen)
1263                         x86_init.pci.arch_init = pci_xen_init;
1264         } else {
1265                 const struct dom0_vga_console_info *info =
1266                         (void *)((char *)xen_start_info +
1267                                  xen_start_info->console.dom0.info_off);
1268
1269                 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1270                 xen_start_info->console.domU.mfn = 0;
1271                 xen_start_info->console.domU.evtchn = 0;
1272
1273                 /* Make sure ACS will be enabled */
1274                 pci_request_acs();
1275         }
1276
1277
1278         xen_raw_console_write("about to get started...\n");
1279
1280         xen_setup_runstate_info(0);
1281
1282         /* Start the world */
1283 #ifdef CONFIG_X86_32
1284         i386_start_kernel();
1285 #else
1286         x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1287 #endif
1288 }
1289
1290 static int init_hvm_pv_info(int *major, int *minor)
1291 {
1292         uint32_t eax, ebx, ecx, edx, pages, msr, base;
1293         u64 pfn;
1294
1295         base = xen_cpuid_base();
1296         cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1297
1298         *major = eax >> 16;
1299         *minor = eax & 0xffff;
1300         printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1301
1302         cpuid(base + 2, &pages, &msr, &ecx, &edx);
1303
1304         pfn = __pa(hypercall_page);
1305         wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1306
1307         xen_setup_features();
1308
1309         pv_info.name = "Xen HVM";
1310
1311         xen_domain_type = XEN_HVM_DOMAIN;
1312
1313         return 0;
1314 }
1315
1316 void __ref xen_hvm_init_shared_info(void)
1317 {
1318         int cpu;
1319         struct xen_add_to_physmap xatp;
1320         static struct shared_info *shared_info_page = 0;
1321
1322         if (!shared_info_page)
1323                 shared_info_page = (struct shared_info *)
1324                         extend_brk(PAGE_SIZE, PAGE_SIZE);
1325         xatp.domid = DOMID_SELF;
1326         xatp.idx = 0;
1327         xatp.space = XENMAPSPACE_shared_info;
1328         xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1329         if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1330                 BUG();
1331
1332         HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1333
1334         /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1335          * page, we use it in the event channel upcall and in some pvclock
1336          * related functions. We don't need the vcpu_info placement
1337          * optimizations because we don't use any pv_mmu or pv_irq op on
1338          * HVM.
1339          * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1340          * online but xen_hvm_init_shared_info is run at resume time too and
1341          * in that case multiple vcpus might be online. */
1342         for_each_online_cpu(cpu) {
1343                 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1344         }
1345 }
1346
1347 #ifdef CONFIG_XEN_PVHVM
1348 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1349                                     unsigned long action, void *hcpu)
1350 {
1351         int cpu = (long)hcpu;
1352         switch (action) {
1353         case CPU_UP_PREPARE:
1354                 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1355                 if (xen_have_vector_callback)
1356                         xen_init_lock_cpu(cpu);
1357                 break;
1358         default:
1359                 break;
1360         }
1361         return NOTIFY_OK;
1362 }
1363
1364 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1365         .notifier_call  = xen_hvm_cpu_notify,
1366 };
1367
1368 static void __init xen_hvm_guest_init(void)
1369 {
1370         int r;
1371         int major, minor;
1372
1373         r = init_hvm_pv_info(&major, &minor);
1374         if (r < 0)
1375                 return;
1376
1377         xen_hvm_init_shared_info();
1378
1379         if (xen_feature(XENFEAT_hvm_callback_vector))
1380                 xen_have_vector_callback = 1;
1381         xen_hvm_smp_init();
1382         register_cpu_notifier(&xen_hvm_cpu_notifier);
1383         xen_unplug_emulated_devices();
1384         have_vcpu_info_placement = 0;
1385         x86_init.irqs.intr_init = xen_init_IRQ;
1386         xen_hvm_init_time_ops();
1387         xen_hvm_init_mmu_ops();
1388 }
1389
1390 static bool __init xen_hvm_platform(void)
1391 {
1392         if (xen_pv_domain())
1393                 return false;
1394
1395         if (!xen_cpuid_base())
1396                 return false;
1397
1398         return true;
1399 }
1400
1401 bool xen_hvm_need_lapic(void)
1402 {
1403         if (xen_pv_domain())
1404                 return false;
1405         if (!xen_hvm_domain())
1406                 return false;
1407         if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1408                 return false;
1409         return true;
1410 }
1411 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1412
1413 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1414         .name                   = "Xen HVM",
1415         .detect                 = xen_hvm_platform,
1416         .init_platform          = xen_hvm_guest_init,
1417 };
1418 EXPORT_SYMBOL(x86_hyper_xen_hvm);
1419 #endif