err.no Git - linux-2.6/blob - arch/i386/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/kernel.h>
  15 #include <linux/init.h>
  16 #include <linux/smp.h>
  17 #include <linux/preempt.h>
  18 #include <linux/percpu.h>
  19 #include <linux/delay.h>
  20 #include <linux/start_kernel.h>
  21 #include <linux/sched.h>
  22 #include <linux/bootmem.h>
  23 #include <linux/module.h>
  24
  25 #include <xen/interface/xen.h>
  26 #include <xen/interface/physdev.h>
  27 #include <xen/interface/vcpu.h>
  28 #include <xen/features.h>
  29 #include <xen/page.h>
  30
  31 #include <asm/paravirt.h>
  32 #include <asm/page.h>
  33 #include <asm/xen/hypercall.h>
  34 #include <asm/xen/hypervisor.h>
  35 #include <asm/fixmap.h>
  36 #include <asm/processor.h>
  37 #include <asm/setup.h>
  38 #include <asm/desc.h>
  39 #include <asm/pgtable.h>
  40
  41 #include "xen-ops.h"
  42 #include "multicalls.h"
  43
  44 EXPORT_SYMBOL_GPL(hypercall_page);
  45
  46 DEFINE_PER_CPU(enum paravirt_lazy_mode, xen_lazy_mode);
  47
  48 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
  49 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
  50 DEFINE_PER_CPU(unsigned long, xen_cr3);
  51
  52 struct start_info *xen_start_info;
  53 EXPORT_SYMBOL_GPL(xen_start_info);
  54
  55 static void xen_vcpu_setup(int cpu)
  56 {
  57         per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
  58 }
  59
  60 static void __init xen_banner(void)
  61 {
  62         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  63                paravirt_ops.name);
  64         printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic);
  65 }
  66
  67 static void xen_cpuid(unsigned int *eax, unsigned int *ebx,
  68                       unsigned int *ecx, unsigned int *edx)
  69 {
  70         unsigned maskedx = ~0;
  71
  72         /*
  73          * Mask out inconvenient features, to try and disable as many
  74          * unsupported kernel subsystems as possible.
  75          */
  76         if (*eax == 1)
  77                 maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
  78                             (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
  79                             (1 << X86_FEATURE_ACC));   /* thermal monitoring */
  80
  81         asm(XEN_EMULATE_PREFIX "cpuid"
  82                 : "=a" (*eax),
  83                   "=b" (*ebx),
  84                   "=c" (*ecx),
  85                   "=d" (*edx)
  86                 : "0" (*eax), "2" (*ecx));
  87         *edx &= maskedx;
  88 }
  89
  90 static void xen_set_debugreg(int reg, unsigned long val)
  91 {
  92         HYPERVISOR_set_debugreg(reg, val);
  93 }
  94
  95 static unsigned long xen_get_debugreg(int reg)
  96 {
  97         return HYPERVISOR_get_debugreg(reg);
  98 }
  99
 100 static unsigned long xen_save_fl(void)
 101 {
 102         struct vcpu_info *vcpu;
 103         unsigned long flags;
 104
 105         preempt_disable();
 106         vcpu = x86_read_percpu(xen_vcpu);
 107         /* flag has opposite sense of mask */
 108         flags = !vcpu->evtchn_upcall_mask;
 109         preempt_enable();
 110
 111         /* convert to IF type flag
 112            -0 -> 0x00000000
 113            -1 -> 0xffffffff
 114         */
 115         return (-flags) & X86_EFLAGS_IF;
 116 }
 117
 118 static void xen_restore_fl(unsigned long flags)
 119 {
 120         struct vcpu_info *vcpu;
 121
 122         preempt_disable();
 123
 124         /* convert from IF type flag */
 125         flags = !(flags & X86_EFLAGS_IF);
 126         vcpu = x86_read_percpu(xen_vcpu);
 127         vcpu->evtchn_upcall_mask = flags;
 128
 129         if (flags == 0) {
 130                 /* Unmask then check (avoid races).  We're only protecting
 131                    against updates by this CPU, so there's no need for
 132                    anything stronger. */
 133                 barrier();
 134
 135                 if (unlikely(vcpu->evtchn_upcall_pending))
 136                         force_evtchn_callback();
 137                 preempt_enable();
 138         } else
 139                 preempt_enable_no_resched();
 140 }
 141
 142 static void xen_irq_disable(void)
 143 {
 144         struct vcpu_info *vcpu;
 145         preempt_disable();
 146         vcpu = x86_read_percpu(xen_vcpu);
 147         vcpu->evtchn_upcall_mask = 1;
 148         preempt_enable_no_resched();
 149 }
 150
 151 static void xen_irq_enable(void)
 152 {
 153         struct vcpu_info *vcpu;
 154
 155         preempt_disable();
 156         vcpu = x86_read_percpu(xen_vcpu);
 157         vcpu->evtchn_upcall_mask = 0;
 158
 159         /* Unmask then check (avoid races).  We're only protecting
 160            against updates by this CPU, so there's no need for
 161            anything stronger. */
 162         barrier();
 163
 164         if (unlikely(vcpu->evtchn_upcall_pending))
 165                 force_evtchn_callback();
 166         preempt_enable();
 167 }
 168
 169 static void xen_safe_halt(void)
 170 {
 171         /* Blocking includes an implicit local_irq_enable(). */
 172         if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0)
 173                 BUG();
 174 }
 175
 176 static void xen_halt(void)
 177 {
 178         if (irqs_disabled())
 179                 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
 180         else
 181                 xen_safe_halt();
 182 }
 183
 184 static void xen_set_lazy_mode(enum paravirt_lazy_mode mode)
 185 {
 186         switch (mode) {
 187         case PARAVIRT_LAZY_NONE:
 188                 BUG_ON(x86_read_percpu(xen_lazy_mode) == PARAVIRT_LAZY_NONE);
 189                 break;
 190
 191         case PARAVIRT_LAZY_MMU:
 192         case PARAVIRT_LAZY_CPU:
 193                 BUG_ON(x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE);
 194                 break;
 195
 196         case PARAVIRT_LAZY_FLUSH:
 197                 /* flush if necessary, but don't change state */
 198                 if (x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE)
 199                         xen_mc_flush();
 200                 return;
 201         }
 202
 203         xen_mc_flush();
 204         x86_write_percpu(xen_lazy_mode, mode);
 205 }
 206
 207 static unsigned long xen_store_tr(void)
 208 {
 209         return 0;
 210 }
 211
 212 static void xen_set_ldt(const void *addr, unsigned entries)
 213 {
 214         unsigned long linear_addr = (unsigned long)addr;
 215         struct mmuext_op *op;
 216         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 217
 218         op = mcs.args;
 219         op->cmd = MMUEXT_SET_LDT;
 220         if (linear_addr) {
 221                 /* ldt my be vmalloced, use arbitrary_virt_to_machine */
 222                 xmaddr_t maddr;
 223                 maddr = arbitrary_virt_to_machine((unsigned long)addr);
 224                 linear_addr = (unsigned long)maddr.maddr;
 225         }
 226         op->arg1.linear_addr = linear_addr;
 227         op->arg2.nr_ents = entries;
 228
 229         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 230
 231         xen_mc_issue(PARAVIRT_LAZY_CPU);
 232 }
 233
 234 static void xen_load_gdt(const struct Xgt_desc_struct *dtr)
 235 {
 236         unsigned long *frames;
 237         unsigned long va = dtr->address;
 238         unsigned int size = dtr->size + 1;
 239         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 240         int f;
 241         struct multicall_space mcs;
 242
 243         /* A GDT can be up to 64k in size, which corresponds to 8192
 244            8-byte entries, or 16 4k pages.. */
 245
 246         BUG_ON(size > 65536);
 247         BUG_ON(va & ~PAGE_MASK);
 248
 249         mcs = xen_mc_entry(sizeof(*frames) * pages);
 250         frames = mcs.args;
 251
 252         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
 253                 frames[f] = virt_to_mfn(va);
 254                 make_lowmem_page_readonly((void *)va);
 255         }
 256
 257         MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
 258
 259         xen_mc_issue(PARAVIRT_LAZY_CPU);
 260 }
 261
 262 static void load_TLS_descriptor(struct thread_struct *t,
 263                                 unsigned int cpu, unsigned int i)
 264 {
 265         struct desc_struct *gdt = get_cpu_gdt_table(cpu);
 266         xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
 267         struct multicall_space mc = __xen_mc_entry(0);
 268
 269         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
 270 }
 271
 272 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
 273 {
 274         xen_mc_batch();
 275
 276         load_TLS_descriptor(t, cpu, 0);
 277         load_TLS_descriptor(t, cpu, 1);
 278         load_TLS_descriptor(t, cpu, 2);
 279
 280         xen_mc_issue(PARAVIRT_LAZY_CPU);
 281 }
 282
 283 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
 284                                 u32 low, u32 high)
 285 {
 286         unsigned long lp = (unsigned long)&dt[entrynum];
 287         xmaddr_t mach_lp = virt_to_machine(lp);
 288         u64 entry = (u64)high << 32 | low;
 289
 290         xen_mc_flush();
 291         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
 292                 BUG();
 293 }
 294
 295 static int cvt_gate_to_trap(int vector, u32 low, u32 high,
 296                             struct trap_info *info)
 297 {
 298         u8 type, dpl;
 299
 300         type = (high >> 8) & 0x1f;
 301         dpl = (high >> 13) & 3;
 302
 303         if (type != 0xf && type != 0xe)
 304                 return 0;
 305
 306         info->vector = vector;
 307         info->address = (high & 0xffff0000) | (low & 0x0000ffff);
 308         info->cs = low >> 16;
 309         info->flags = dpl;
 310         /* interrupt gates clear IF */
 311         if (type == 0xe)
 312                 info->flags |= 4;
 313
 314         return 1;
 315 }
 316
 317 /* Locations of each CPU's IDT */
 318 static DEFINE_PER_CPU(struct Xgt_desc_struct, idt_desc);
 319
 320 /* Set an IDT entry.  If the entry is part of the current IDT, then
 321    also update Xen. */
 322 static void xen_write_idt_entry(struct desc_struct *dt, int entrynum,
 323                                 u32 low, u32 high)
 324 {
 325
 326         int cpu = smp_processor_id();
 327         unsigned long p = (unsigned long)&dt[entrynum];
 328         unsigned long start = per_cpu(idt_desc, cpu).address;
 329         unsigned long end = start + per_cpu(idt_desc, cpu).size + 1;
 330
 331         xen_mc_flush();
 332
 333         write_dt_entry(dt, entrynum, low, high);
 334
 335         if (p >= start && (p + 8) <= end) {
 336                 struct trap_info info[2];
 337
 338                 info[1].address = 0;
 339
 340                 if (cvt_gate_to_trap(entrynum, low, high, &info[0]))
 341                         if (HYPERVISOR_set_trap_table(info))
 342                                 BUG();
 343         }
 344 }
 345
 346 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
 347    hold a spinlock to protect the static traps[] array (static because
 348    it avoids allocation, and saves stack space). */
 349 static void xen_load_idt(const struct Xgt_desc_struct *desc)
 350 {
 351         static DEFINE_SPINLOCK(lock);
 352         static struct trap_info traps[257];
 353
 354         int cpu = smp_processor_id();
 355         unsigned in, out, count;
 356
 357         per_cpu(idt_desc, cpu) = *desc;
 358
 359         count = (desc->size+1) / 8;
 360         BUG_ON(count > 256);
 361
 362         spin_lock(&lock);
 363         for (in = out = 0; in < count; in++) {
 364                 const u32 *entry = (u32 *)(desc->address + in * 8);
 365
 366                 if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out]))
 367                         out++;
 368         }
 369         traps[out].address = 0;
 370
 371         xen_mc_flush();
 372         if (HYPERVISOR_set_trap_table(traps))
 373                 BUG();
 374
 375         spin_unlock(&lock);
 376 }
 377
 378 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
 379    they're handled differently. */
 380 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
 381                                 u32 low, u32 high)
 382 {
 383         switch ((high >> 8) & 0xff) {
 384         case DESCTYPE_LDT:
 385         case DESCTYPE_TSS:
 386                 /* ignore */
 387                 break;
 388
 389         default: {
 390                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
 391                 u64 desc = (u64)high << 32 | low;
 392
 393                 xen_mc_flush();
 394                 if (HYPERVISOR_update_descriptor(maddr.maddr, desc))
 395                         BUG();
 396         }
 397
 398         }
 399 }
 400
 401 static void xen_load_esp0(struct tss_struct *tss,
 402                                    struct thread_struct *thread)
 403 {
 404         struct multicall_space mcs = xen_mc_entry(0);
 405         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->esp0);
 406         xen_mc_issue(PARAVIRT_LAZY_CPU);
 407 }
 408
 409 static void xen_set_iopl_mask(unsigned mask)
 410 {
 411         struct physdev_set_iopl set_iopl;
 412
 413         /* Force the change at ring 0. */
 414         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
 415         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 416 }
 417
 418 static void xen_io_delay(void)
 419 {
 420 }
 421
 422 #ifdef CONFIG_X86_LOCAL_APIC
 423 static unsigned long xen_apic_read(unsigned long reg)
 424 {
 425         return 0;
 426 }
 427 #endif
 428
 429 static void xen_flush_tlb(void)
 430 {
 431         struct mmuext_op op;
 432
 433         op.cmd = MMUEXT_TLB_FLUSH_LOCAL;
 434         if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
 435                 BUG();
 436 }
 437
 438 static void xen_flush_tlb_single(unsigned long addr)
 439 {
 440         struct mmuext_op op;
 441
 442         op.cmd = MMUEXT_INVLPG_LOCAL;
 443         op.arg1.linear_addr = addr & PAGE_MASK;
 444         if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
 445                 BUG();
 446 }
 447
 448 static unsigned long xen_read_cr2(void)
 449 {
 450         return x86_read_percpu(xen_vcpu)->arch.cr2;
 451 }
 452
 453 static void xen_write_cr4(unsigned long cr4)
 454 {
 455         /* never allow TSC to be disabled */
 456         native_write_cr4(cr4 & ~X86_CR4_TSD);
 457 }
 458
 459 /*
 460  * Page-directory addresses above 4GB do not fit into architectural %cr3.
 461  * When accessing %cr3, or equivalent field in vcpu_guest_context, guests
 462  * must use the following accessor macros to pack/unpack valid MFNs.
 463  *
 464  * Note that Xen is using the fact that the pagetable base is always
 465  * page-aligned, and putting the 12 MSB of the address into the 12 LSB
 466  * of cr3.
 467  */
 468 #define xen_pfn_to_cr3(pfn) (((unsigned)(pfn) << 12) | ((unsigned)(pfn) >> 20))
 469 #define xen_cr3_to_pfn(cr3) (((unsigned)(cr3) >> 12) | ((unsigned)(cr3) << 20))
 470
 471 static unsigned long xen_read_cr3(void)
 472 {
 473         return x86_read_percpu(xen_cr3);
 474 }
 475
 476 static void xen_write_cr3(unsigned long cr3)
 477 {
 478         if (cr3 == x86_read_percpu(xen_cr3)) {
 479                 /* just a simple tlb flush */
 480                 xen_flush_tlb();
 481                 return;
 482         }
 483
 484         x86_write_percpu(xen_cr3, cr3);
 485
 486
 487         {
 488                 struct mmuext_op *op;
 489                 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 490                 unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3));
 491
 492                 op = mcs.args;
 493                 op->cmd = MMUEXT_NEW_BASEPTR;
 494                 op->arg1.mfn = mfn;
 495
 496                 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 497
 498                 xen_mc_issue(PARAVIRT_LAZY_CPU);
 499         }
 500 }
 501
 502 static void xen_alloc_pt(struct mm_struct *mm, u32 pfn)
 503 {
 504         /* XXX pfn isn't necessarily a lowmem page */
 505         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
 506 }
 507
 508 static void xen_alloc_pd(u32 pfn)
 509 {
 510         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
 511 }
 512
 513 static void xen_release_pd(u32 pfn)
 514 {
 515         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
 516 }
 517
 518 static void xen_release_pt(u32 pfn)
 519 {
 520         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
 521 }
 522
 523 static void xen_alloc_pd_clone(u32 pfn, u32 clonepfn,
 524                                         u32 start, u32 count)
 525 {
 526         xen_alloc_pd(pfn);
 527 }
 528
 529 static __init void xen_pagetable_setup_start(pgd_t *base)
 530 {
 531         pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base;
 532
 533         init_mm.pgd = base;
 534         /*
 535          * copy top-level of Xen-supplied pagetable into place.  For
 536          * !PAE we can use this as-is, but for PAE it is a stand-in
 537          * while we copy the pmd pages.
 538          */
 539         memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t));
 540
 541         if (PTRS_PER_PMD > 1) {
 542                 int i;
 543                 /*
 544                  * For PAE, need to allocate new pmds, rather than
 545                  * share Xen's, since Xen doesn't like pmd's being
 546                  * shared between address spaces.
 547                  */
 548                 for (i = 0; i < PTRS_PER_PGD; i++) {
 549                         if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
 550                                 pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 551
 552                                 memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
 553                                        PAGE_SIZE);
 554
 555                                 xen_alloc_pd(PFN_DOWN(__pa(pmd)));
 556
 557                                 set_pgd(&base[i], __pgd(1 + __pa(pmd)));
 558                         } else
 559                                 pgd_clear(&base[i]);
 560                 }
 561         }
 562
 563         /* make sure zero_page is mapped RO so we can use it in pagetables */
 564         make_lowmem_page_readonly(empty_zero_page);
 565         make_lowmem_page_readonly(base);
 566         /*
 567          * Switch to new pagetable.  This is done before
 568          * pagetable_init has done anything so that the new pages
 569          * added to the table can be prepared properly for Xen.
 570          */
 571         xen_write_cr3(__pa(base));
 572 }
 573
 574 static __init void xen_pagetable_setup_done(pgd_t *base)
 575 {
 576         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 577                 /*
 578                  * Create a mapping for the shared info page.
 579                  * Should be set_fixmap(), but shared_info is a machine
 580                  * address with no corresponding pseudo-phys address.
 581                  */
 582 #if 0
 583                 set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP),
 584                             PFN_DOWN(xen_start_info->shared_info),
 585                             PAGE_KERNEL);
 586 #endif
 587
 588                 HYPERVISOR_shared_info =
 589                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
 590
 591         } else
 592                 HYPERVISOR_shared_info =
 593                         (struct shared_info *)__va(xen_start_info->shared_info);
 594
 595 #if 0
 596         xen_pgd_pin(base);
 597 #endif
 598
 599         xen_vcpu_setup(smp_processor_id());
 600 }
 601
 602 static const struct paravirt_ops xen_paravirt_ops __initdata = {
 603         .paravirt_enabled = 1,
 604         .shared_kernel_pmd = 0,
 605
 606         .name = "Xen",
 607         .banner = xen_banner,
 608
 609         .patch = paravirt_patch_default,
 610
 611         .memory_setup = xen_memory_setup,
 612         .arch_setup = xen_arch_setup,
 613
 614         .cpuid = xen_cpuid,
 615
 616         .set_debugreg = xen_set_debugreg,
 617         .get_debugreg = xen_get_debugreg,
 618
 619         .clts = native_clts,
 620
 621         .read_cr0 = native_read_cr0,
 622         .write_cr0 = native_write_cr0,
 623
 624         .read_cr2 = xen_read_cr2,
 625         .write_cr2 = native_write_cr2,
 626
 627         .read_cr3 = xen_read_cr3,
 628         .write_cr3 = xen_write_cr3,
 629
 630         .read_cr4 = native_read_cr4,
 631         .read_cr4_safe = native_read_cr4_safe,
 632         .write_cr4 = xen_write_cr4,
 633
 634         .save_fl = xen_save_fl,
 635         .restore_fl = xen_restore_fl,
 636         .irq_disable = xen_irq_disable,
 637         .irq_enable = xen_irq_enable,
 638         .safe_halt = xen_safe_halt,
 639         .halt = xen_halt,
 640         .wbinvd = native_wbinvd,
 641
 642         .read_msr = native_read_msr_safe,
 643         .write_msr = native_write_msr_safe,
 644         .read_tsc = native_read_tsc,
 645         .read_pmc = native_read_pmc,
 646
 647         .iret = (void *)&hypercall_page[__HYPERVISOR_iret],
 648         .irq_enable_sysexit = NULL,  /* never called */
 649
 650         .load_tr_desc = paravirt_nop,
 651         .set_ldt = xen_set_ldt,
 652         .load_gdt = xen_load_gdt,
 653         .load_idt = xen_load_idt,
 654         .load_tls = xen_load_tls,
 655
 656         .store_gdt = native_store_gdt,
 657         .store_idt = native_store_idt,
 658         .store_tr = xen_store_tr,
 659
 660         .write_ldt_entry = xen_write_ldt_entry,
 661         .write_gdt_entry = xen_write_gdt_entry,
 662         .write_idt_entry = xen_write_idt_entry,
 663         .load_esp0 = xen_load_esp0,
 664
 665         .set_iopl_mask = xen_set_iopl_mask,
 666         .io_delay = xen_io_delay,
 667
 668 #ifdef CONFIG_X86_LOCAL_APIC
 669         .apic_write = paravirt_nop,
 670         .apic_write_atomic = paravirt_nop,
 671         .apic_read = xen_apic_read,
 672         .setup_boot_clock = paravirt_nop,
 673         .setup_secondary_clock = paravirt_nop,
 674         .startup_ipi_hook = paravirt_nop,
 675 #endif
 676
 677         .flush_tlb_user = xen_flush_tlb,
 678         .flush_tlb_kernel = xen_flush_tlb,
 679         .flush_tlb_single = xen_flush_tlb_single,
 680
 681         .pte_update = paravirt_nop,
 682         .pte_update_defer = paravirt_nop,
 683
 684         .pagetable_setup_start = xen_pagetable_setup_start,
 685         .pagetable_setup_done = xen_pagetable_setup_done,
 686
 687         .alloc_pt = xen_alloc_pt,
 688         .alloc_pd = xen_alloc_pd,
 689         .alloc_pd_clone = xen_alloc_pd_clone,
 690         .release_pd = xen_release_pd,
 691         .release_pt = xen_release_pt,
 692
 693         .set_lazy_mode = xen_set_lazy_mode,
 694 };
 695
 696 /* First C function to be called on Xen boot */
 697 asmlinkage void __init xen_start_kernel(void)
 698 {
 699         pgd_t *pgd;
 700
 701         if (!xen_start_info)
 702                 return;
 703
 704         BUG_ON(memcmp(xen_start_info->magic, "xen-3.0", 7) != 0);
 705
 706         /* Install Xen paravirt ops */
 707         paravirt_ops = xen_paravirt_ops;
 708
 709         xen_setup_features();
 710
 711         /* Get mfn list */
 712         if (!xen_feature(XENFEAT_auto_translated_physmap))
 713                 phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list;
 714
 715         pgd = (pgd_t *)xen_start_info->pt_base;
 716
 717         init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
 718
 719         init_mm.pgd = pgd; /* use the Xen pagetables to start */
 720
 721         /* keep using Xen gdt for now; no urgent need to change it */
 722
 723         x86_write_percpu(xen_cr3, __pa(pgd));
 724         xen_vcpu_setup(0);
 725
 726         paravirt_ops.kernel_rpl = 1;
 727         if (xen_feature(XENFEAT_supervisor_mode_kernel))
 728                 paravirt_ops.kernel_rpl = 0;
 729
 730         /* set the limit of our address space */
 731         reserve_top_address(-HYPERVISOR_VIRT_START + 2 * PAGE_SIZE);
 732
 733         /* set up basic CPUID stuff */
 734         cpu_detect(&new_cpu_data);
 735         new_cpu_data.hard_math = 1;
 736         new_cpu_data.x86_capability[0] = cpuid_edx(1);
 737
 738         /* Poke various useful things into boot_params */
 739         LOADER_TYPE = (9 << 4) | 0;
 740         INITRD_START = xen_start_info->mod_start ? __pa(xen_start_info->mod_start) : 0;
 741         INITRD_SIZE = xen_start_info->mod_len;
 742
 743         /* Start the world */
 744         start_kernel();
 745 }