git.oblomov.eu Git - linux-2.6/blob - 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/kernel.h>
  15 #include <linux/init.h>
  16 #include <linux/smp.h>
  17 #include <linux/preempt.h>
  18 #include <linux/hardirq.h>
  19 #include <linux/percpu.h>
  20 #include <linux/delay.h>
  21 #include <linux/start_kernel.h>
  22 #include <linux/sched.h>
  23 #include <linux/bootmem.h>
  24 #include <linux/module.h>
  25 #include <linux/mm.h>
  26 #include <linux/page-flags.h>
  27 #include <linux/highmem.h>
  28 #include <linux/console.h>
  29
  30 #include <xen/interface/xen.h>
  31 #include <xen/interface/version.h>
  32 #include <xen/interface/physdev.h>
  33 #include <xen/interface/vcpu.h>
  34 #include <xen/features.h>
  35 #include <xen/page.h>
  36 #include <xen/hvc-console.h>
  37
  38 #include <asm/paravirt.h>
  39 #include <asm/apic.h>
  40 #include <asm/page.h>
  41 #include <asm/xen/hypercall.h>
  42 #include <asm/xen/hypervisor.h>
  43 #include <asm/fixmap.h>
  44 #include <asm/processor.h>
  45 #include <asm/msr-index.h>
  46 #include <asm/setup.h>
  47 #include <asm/desc.h>
  48 #include <asm/pgtable.h>
  49 #include <asm/tlbflush.h>
  50 #include <asm/reboot.h>
  51
  52 #include "xen-ops.h"
  53 #include "mmu.h"
  54 #include "multicalls.h"
  55
  56 EXPORT_SYMBOL_GPL(hypercall_page);
  57
  58 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
  59 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
  60
  61 enum xen_domain_type xen_domain_type = XEN_NATIVE;
  62 EXPORT_SYMBOL_GPL(xen_domain_type);
  63
  64 struct start_info *xen_start_info;
  65 EXPORT_SYMBOL_GPL(xen_start_info);
  66
  67 struct shared_info xen_dummy_shared_info;
  68
  69 void *xen_initial_gdt;
  70
  71 /*
  72  * Point at some empty memory to start with. We map the real shared_info
  73  * page as soon as fixmap is up and running.
  74  */
  75 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
  76
  77 /*
  78  * Flag to determine whether vcpu info placement is available on all
  79  * VCPUs.  We assume it is to start with, and then set it to zero on
  80  * the first failure.  This is because it can succeed on some VCPUs
  81  * and not others, since it can involve hypervisor memory allocation,
  82  * or because the guest failed to guarantee all the appropriate
  83  * constraints on all VCPUs (ie buffer can't cross a page boundary).
  84  *
  85  * Note that any particular CPU may be using a placed vcpu structure,
  86  * but we can only optimise if the all are.
  87  *
  88  * 0: not available, 1: available
  89  */
  90 static int have_vcpu_info_placement = 1;
  91
  92 static void xen_vcpu_setup(int cpu)
  93 {
  94         struct vcpu_register_vcpu_info info;
  95         int err;
  96         struct vcpu_info *vcpup;
  97
  98         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
  99         per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
 100
 101         if (!have_vcpu_info_placement)
 102                 return;         /* already tested, not available */
 103
 104         vcpup = &per_cpu(xen_vcpu_info, cpu);
 105
 106         info.mfn = virt_to_mfn(vcpup);
 107         info.offset = offset_in_page(vcpup);
 108
 109         printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
 110                cpu, vcpup, info.mfn, info.offset);
 111
 112         /* Check to see if the hypervisor will put the vcpu_info
 113            structure where we want it, which allows direct access via
 114            a percpu-variable. */
 115         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
 116
 117         if (err) {
 118                 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
 119                 have_vcpu_info_placement = 0;
 120         } else {
 121                 /* This cpu is using the registered vcpu info, even if
 122                    later ones fail to. */
 123                 per_cpu(xen_vcpu, cpu) = vcpup;
 124
 125                 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
 126                        cpu, vcpup);
 127         }
 128 }
 129
 130 /*
 131  * On restore, set the vcpu placement up again.
 132  * If it fails, then we're in a bad state, since
 133  * we can't back out from using it...
 134  */
 135 void xen_vcpu_restore(void)
 136 {
 137         if (have_vcpu_info_placement) {
 138                 int cpu;
 139
 140                 for_each_online_cpu(cpu) {
 141                         bool other_cpu = (cpu != smp_processor_id());
 142
 143                         if (other_cpu &&
 144                             HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
 145                                 BUG();
 146
 147                         xen_vcpu_setup(cpu);
 148
 149                         if (other_cpu &&
 150                             HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
 151                                 BUG();
 152                 }
 153
 154                 BUG_ON(!have_vcpu_info_placement);
 155         }
 156 }
 157
 158 static void __init xen_banner(void)
 159 {
 160         unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
 161         struct xen_extraversion extra;
 162         HYPERVISOR_xen_version(XENVER_extraversion, &extra);
 163
 164         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
 165                pv_info.name);
 166         printk(KERN_INFO "Xen version: %d.%d%s%s\n",
 167                version >> 16, version & 0xffff, extra.extraversion,
 168                xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
 169 }
 170
 171 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
 172                       unsigned int *cx, unsigned int *dx)
 173 {
 174         unsigned maskedx = ~0;
 175
 176         /*
 177          * Mask out inconvenient features, to try and disable as many
 178          * unsupported kernel subsystems as possible.
 179          */
 180         if (*ax == 1)
 181                 maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
 182                             (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
 183                             (1 << X86_FEATURE_MCE)  |  /* disable MCE */
 184                             (1 << X86_FEATURE_MCA)  |  /* disable MCA */
 185                             (1 << X86_FEATURE_ACC));   /* thermal monitoring */
 186
 187         asm(XEN_EMULATE_PREFIX "cpuid"
 188                 : "=a" (*ax),
 189                   "=b" (*bx),
 190                   "=c" (*cx),
 191                   "=d" (*dx)
 192                 : "0" (*ax), "2" (*cx));
 193         *dx &= maskedx;
 194 }
 195
 196 static void xen_set_debugreg(int reg, unsigned long val)
 197 {
 198         HYPERVISOR_set_debugreg(reg, val);
 199 }
 200
 201 static unsigned long xen_get_debugreg(int reg)
 202 {
 203         return HYPERVISOR_get_debugreg(reg);
 204 }
 205
 206 void xen_leave_lazy(void)
 207 {
 208         paravirt_leave_lazy(paravirt_get_lazy_mode());
 209         xen_mc_flush();
 210 }
 211
 212 static unsigned long xen_store_tr(void)
 213 {
 214         return 0;
 215 }
 216
 217 /*
 218  * Set the page permissions for a particular virtual address.  If the
 219  * address is a vmalloc mapping (or other non-linear mapping), then
 220  * find the linear mapping of the page and also set its protections to
 221  * match.
 222  */
 223 static void set_aliased_prot(void *v, pgprot_t prot)
 224 {
 225         int level;
 226         pte_t *ptep;
 227         pte_t pte;
 228         unsigned long pfn;
 229         struct page *page;
 230
 231         ptep = lookup_address((unsigned long)v, &level);
 232         BUG_ON(ptep == NULL);
 233
 234         pfn = pte_pfn(*ptep);
 235         page = pfn_to_page(pfn);
 236
 237         pte = pfn_pte(pfn, prot);
 238
 239         if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
 240                 BUG();
 241
 242         if (!PageHighMem(page)) {
 243                 void *av = __va(PFN_PHYS(pfn));
 244
 245                 if (av != v)
 246                         if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
 247                                 BUG();
 248         } else
 249                 kmap_flush_unused();
 250 }
 251
 252 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
 253 {
 254         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
 255         int i;
 256
 257         for(i = 0; i < entries; i += entries_per_page)
 258                 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
 259 }
 260
 261 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
 262 {
 263         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
 264         int i;
 265
 266         for(i = 0; i < entries; i += entries_per_page)
 267                 set_aliased_prot(ldt + i, PAGE_KERNEL);
 268 }
 269
 270 static void xen_set_ldt(const void *addr, unsigned entries)
 271 {
 272         struct mmuext_op *op;
 273         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 274
 275         op = mcs.args;
 276         op->cmd = MMUEXT_SET_LDT;
 277         op->arg1.linear_addr = (unsigned long)addr;
 278         op->arg2.nr_ents = entries;
 279
 280         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 281
 282         xen_mc_issue(PARAVIRT_LAZY_CPU);
 283 }
 284
 285 static void xen_load_gdt(const struct desc_ptr *dtr)
 286 {
 287         unsigned long *frames;
 288         unsigned long va = dtr->address;
 289         unsigned int size = dtr->size + 1;
 290         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 291         int f;
 292         struct multicall_space mcs;
 293
 294         /* A GDT can be up to 64k in size, which corresponds to 8192
 295            8-byte entries, or 16 4k pages.. */
 296
 297         BUG_ON(size > 65536);
 298         BUG_ON(va & ~PAGE_MASK);
 299
 300         mcs = xen_mc_entry(sizeof(*frames) * pages);
 301         frames = mcs.args;
 302
 303         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
 304                 frames[f] = virt_to_mfn(va);
 305                 make_lowmem_page_readonly((void *)va);
 306         }
 307
 308         MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
 309
 310         xen_mc_issue(PARAVIRT_LAZY_CPU);
 311 }
 312
 313 static void load_TLS_descriptor(struct thread_struct *t,
 314                                 unsigned int cpu, unsigned int i)
 315 {
 316         struct desc_struct *gdt = get_cpu_gdt_table(cpu);
 317         xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
 318         struct multicall_space mc = __xen_mc_entry(0);
 319
 320         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
 321 }
 322
 323 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
 324 {
 325         /*
 326          * XXX sleazy hack: If we're being called in a lazy-cpu zone,
 327          * it means we're in a context switch, and %gs has just been
 328          * saved.  This means we can zero it out to prevent faults on
 329          * exit from the hypervisor if the next process has no %gs.
 330          * Either way, it has been saved, and the new value will get
 331          * loaded properly.  This will go away as soon as Xen has been
 332          * modified to not save/restore %gs for normal hypercalls.
 333          *
 334          * On x86_64, this hack is not used for %gs, because gs points
 335          * to KERNEL_GS_BASE (and uses it for PDA references), so we
 336          * must not zero %gs on x86_64
 337          *
 338          * For x86_64, we need to zero %fs, otherwise we may get an
 339          * exception between the new %fs descriptor being loaded and
 340          * %fs being effectively cleared at __switch_to().
 341          */
 342         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
 343 #ifdef CONFIG_X86_32
 344                 loadsegment(gs, 0);
 345 #else
 346                 loadsegment(fs, 0);
 347 #endif
 348         }
 349
 350         xen_mc_batch();
 351
 352         load_TLS_descriptor(t, cpu, 0);
 353         load_TLS_descriptor(t, cpu, 1);
 354         load_TLS_descriptor(t, cpu, 2);
 355
 356         xen_mc_issue(PARAVIRT_LAZY_CPU);
 357 }
 358
 359 #ifdef CONFIG_X86_64
 360 static void xen_load_gs_index(unsigned int idx)
 361 {
 362         if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
 363                 BUG();
 364 }
 365 #endif
 366
 367 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
 368                                 const void *ptr)
 369 {
 370         xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
 371         u64 entry = *(u64 *)ptr;
 372
 373         preempt_disable();
 374
 375         xen_mc_flush();
 376         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
 377                 BUG();
 378
 379         preempt_enable();
 380 }
 381
 382 static int cvt_gate_to_trap(int vector, const gate_desc *val,
 383                             struct trap_info *info)
 384 {
 385         if (val->type != 0xf && val->type != 0xe)
 386                 return 0;
 387
 388         info->vector = vector;
 389         info->address = gate_offset(*val);
 390         info->cs = gate_segment(*val);
 391         info->flags = val->dpl;
 392         /* interrupt gates clear IF */
 393         if (val->type == 0xe)
 394                 info->flags |= 4;
 395
 396         return 1;
 397 }
 398
 399 /* Locations of each CPU's IDT */
 400 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
 401
 402 /* Set an IDT entry.  If the entry is part of the current IDT, then
 403    also update Xen. */
 404 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
 405 {
 406         unsigned long p = (unsigned long)&dt[entrynum];
 407         unsigned long start, end;
 408
 409         preempt_disable();
 410
 411         start = __get_cpu_var(idt_desc).address;
 412         end = start + __get_cpu_var(idt_desc).size + 1;
 413
 414         xen_mc_flush();
 415
 416         native_write_idt_entry(dt, entrynum, g);
 417
 418         if (p >= start && (p + 8) <= end) {
 419                 struct trap_info info[2];
 420
 421                 info[1].address = 0;
 422
 423                 if (cvt_gate_to_trap(entrynum, g, &info[0]))
 424                         if (HYPERVISOR_set_trap_table(info))
 425                                 BUG();
 426         }
 427
 428         preempt_enable();
 429 }
 430
 431 static void xen_convert_trap_info(const struct desc_ptr *desc,
 432                                   struct trap_info *traps)
 433 {
 434         unsigned in, out, count;
 435
 436         count = (desc->size+1) / sizeof(gate_desc);
 437         BUG_ON(count > 256);
 438
 439         for (in = out = 0; in < count; in++) {
 440                 gate_desc *entry = (gate_desc*)(desc->address) + in;
 441
 442                 if (cvt_gate_to_trap(in, entry, &traps[out]))
 443                         out++;
 444         }
 445         traps[out].address = 0;
 446 }
 447
 448 void xen_copy_trap_info(struct trap_info *traps)
 449 {
 450         const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
 451
 452         xen_convert_trap_info(desc, traps);
 453 }
 454
 455 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
 456    hold a spinlock to protect the static traps[] array (static because
 457    it avoids allocation, and saves stack space). */
 458 static void xen_load_idt(const struct desc_ptr *desc)
 459 {
 460         static DEFINE_SPINLOCK(lock);
 461         static struct trap_info traps[257];
 462
 463         spin_lock(&lock);
 464
 465         __get_cpu_var(idt_desc) = *desc;
 466
 467         xen_convert_trap_info(desc, traps);
 468
 469         xen_mc_flush();
 470         if (HYPERVISOR_set_trap_table(traps))
 471                 BUG();
 472
 473         spin_unlock(&lock);
 474 }
 475
 476 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
 477    they're handled differently. */
 478 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
 479                                 const void *desc, int type)
 480 {
 481         preempt_disable();
 482
 483         switch (type) {
 484         case DESC_LDT:
 485         case DESC_TSS:
 486                 /* ignore */
 487                 break;
 488
 489         default: {
 490                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
 491
 492                 xen_mc_flush();
 493                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
 494                         BUG();
 495         }
 496
 497         }
 498
 499         preempt_enable();
 500 }
 501
 502 static void xen_load_sp0(struct tss_struct *tss,
 503                          struct thread_struct *thread)
 504 {
 505         struct multicall_space mcs = xen_mc_entry(0);
 506         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
 507         xen_mc_issue(PARAVIRT_LAZY_CPU);
 508 }
 509
 510 static void xen_set_iopl_mask(unsigned mask)
 511 {
 512         struct physdev_set_iopl set_iopl;
 513
 514         /* Force the change at ring 0. */
 515         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
 516         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 517 }
 518
 519 static void xen_io_delay(void)
 520 {
 521 }
 522
 523 #ifdef CONFIG_X86_LOCAL_APIC
 524 static u32 xen_apic_read(u32 reg)
 525 {
 526         return 0;
 527 }
 528
 529 static void xen_apic_write(u32 reg, u32 val)
 530 {
 531         /* Warn to see if there's any stray references */
 532         WARN_ON(1);
 533 }
 534
 535 static u64 xen_apic_icr_read(void)
 536 {
 537         return 0;
 538 }
 539
 540 static void xen_apic_icr_write(u32 low, u32 id)
 541 {
 542         /* Warn to see if there's any stray references */
 543         WARN_ON(1);
 544 }
 545
 546 static void xen_apic_wait_icr_idle(void)
 547 {
 548         return;
 549 }
 550
 551 static u32 xen_safe_apic_wait_icr_idle(void)
 552 {
 553         return 0;
 554 }
 555
 556 static struct apic_ops xen_basic_apic_ops = {
 557         .read = xen_apic_read,
 558         .write = xen_apic_write,
 559         .icr_read = xen_apic_icr_read,
 560         .icr_write = xen_apic_icr_write,
 561         .wait_icr_idle = xen_apic_wait_icr_idle,
 562         .safe_wait_icr_idle = xen_safe_apic_wait_icr_idle,
 563 };
 564
 565 #endif
 566
 567
 568 static void xen_clts(void)
 569 {
 570         struct multicall_space mcs;
 571
 572         mcs = xen_mc_entry(0);
 573
 574         MULTI_fpu_taskswitch(mcs.mc, 0);
 575
 576         xen_mc_issue(PARAVIRT_LAZY_CPU);
 577 }
 578
 579 static void xen_write_cr0(unsigned long cr0)
 580 {
 581         struct multicall_space mcs;
 582
 583         /* Only pay attention to cr0.TS; everything else is
 584            ignored. */
 585         mcs = xen_mc_entry(0);
 586
 587         MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
 588
 589         xen_mc_issue(PARAVIRT_LAZY_CPU);
 590 }
 591
 592 static void xen_write_cr4(unsigned long cr4)
 593 {
 594         cr4 &= ~X86_CR4_PGE;
 595         cr4 &= ~X86_CR4_PSE;
 596
 597         native_write_cr4(cr4);
 598 }
 599
 600 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
 601 {
 602         int ret;
 603
 604         ret = 0;
 605
 606         switch (msr) {
 607 #ifdef CONFIG_X86_64
 608                 unsigned which;
 609                 u64 base;
 610
 611         case MSR_FS_BASE:               which = SEGBASE_FS; goto set;
 612         case MSR_KERNEL_GS_BASE:        which = SEGBASE_GS_USER; goto set;
 613         case MSR_GS_BASE:               which = SEGBASE_GS_KERNEL; goto set;
 614
 615         set:
 616                 base = ((u64)high << 32) | low;
 617                 if (HYPERVISOR_set_segment_base(which, base) != 0)
 618                         ret = -EFAULT;
 619                 break;
 620 #endif
 621
 622         case MSR_STAR:
 623         case MSR_CSTAR:
 624         case MSR_LSTAR:
 625         case MSR_SYSCALL_MASK:
 626         case MSR_IA32_SYSENTER_CS:
 627         case MSR_IA32_SYSENTER_ESP:
 628         case MSR_IA32_SYSENTER_EIP:
 629                 /* Fast syscall setup is all done in hypercalls, so
 630                    these are all ignored.  Stub them out here to stop
 631                    Xen console noise. */
 632                 break;
 633
 634         default:
 635                 ret = native_write_msr_safe(msr, low, high);
 636         }
 637
 638         return ret;
 639 }
 640
 641 void xen_setup_shared_info(void)
 642 {
 643         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 644                 set_fixmap(FIX_PARAVIRT_BOOTMAP,
 645                            xen_start_info->shared_info);
 646
 647                 HYPERVISOR_shared_info =
 648                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
 649         } else
 650                 HYPERVISOR_shared_info =
 651                         (struct shared_info *)__va(xen_start_info->shared_info);
 652
 653 #ifndef CONFIG_SMP
 654         /* In UP this is as good a place as any to set up shared info */
 655         xen_setup_vcpu_info_placement();
 656 #endif
 657
 658         xen_setup_mfn_list_list();
 659 }
 660
 661 /* This is called once we have the cpu_possible_map */
 662 void xen_setup_vcpu_info_placement(void)
 663 {
 664         int cpu;
 665
 666         for_each_possible_cpu(cpu)
 667                 xen_vcpu_setup(cpu);
 668
 669         /* xen_vcpu_setup managed to place the vcpu_info within the
 670            percpu area for all cpus, so make use of it */
 671         if (have_vcpu_info_placement) {
 672                 printk(KERN_INFO "Xen: using vcpu_info placement\n");
 673
 674                 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
 675                 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
 676                 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
 677                 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
 678                 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
 679         }
 680 }
 681
 682 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
 683                           unsigned long addr, unsigned len)
 684 {
 685         char *start, *end, *reloc;
 686         unsigned ret;
 687
 688         start = end = reloc = NULL;
 689
 690 #define SITE(op, x)                                                     \
 691         case PARAVIRT_PATCH(op.x):                                      \
 692         if (have_vcpu_info_placement) {                                 \
 693                 start = (char *)xen_##x##_direct;                       \
 694                 end = xen_##x##_direct_end;                             \
 695                 reloc = xen_##x##_direct_reloc;                         \
 696         }                                                               \
 697         goto patch_site
 698
 699         switch (type) {
 700                 SITE(pv_irq_ops, irq_enable);
 701                 SITE(pv_irq_ops, irq_disable);
 702                 SITE(pv_irq_ops, save_fl);
 703                 SITE(pv_irq_ops, restore_fl);
 704 #undef SITE
 705
 706         patch_site:
 707                 if (start == NULL || (end-start) > len)
 708                         goto default_patch;
 709
 710                 ret = paravirt_patch_insns(insnbuf, len, start, end);
 711
 712                 /* Note: because reloc is assigned from something that
 713                    appears to be an array, gcc assumes it's non-null,
 714                    but doesn't know its relationship with start and
 715                    end. */
 716                 if (reloc > start && reloc < end) {
 717                         int reloc_off = reloc - start;
 718                         long *relocp = (long *)(insnbuf + reloc_off);
 719                         long delta = start - (char *)addr;
 720
 721                         *relocp += delta;
 722                 }
 723                 break;
 724
 725         default_patch:
 726         default:
 727                 ret = paravirt_patch_default(type, clobbers, insnbuf,
 728                                              addr, len);
 729                 break;
 730         }
 731
 732         return ret;
 733 }
 734
 735 static const struct pv_info xen_info __initdata = {
 736         .paravirt_enabled = 1,
 737         .shared_kernel_pmd = 0,
 738
 739         .name = "Xen",
 740 };
 741
 742 static const struct pv_init_ops xen_init_ops __initdata = {
 743         .patch = xen_patch,
 744
 745         .banner = xen_banner,
 746         .memory_setup = xen_memory_setup,
 747         .arch_setup = xen_arch_setup,
 748         .post_allocator_init = xen_post_allocator_init,
 749 };
 750
 751 static const struct pv_time_ops xen_time_ops __initdata = {
 752         .time_init = xen_time_init,
 753
 754         .set_wallclock = xen_set_wallclock,
 755         .get_wallclock = xen_get_wallclock,
 756         .get_tsc_khz = xen_tsc_khz,
 757         .sched_clock = xen_sched_clock,
 758 };
 759
 760 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
 761         .cpuid = xen_cpuid,
 762
 763         .set_debugreg = xen_set_debugreg,
 764         .get_debugreg = xen_get_debugreg,
 765
 766         .clts = xen_clts,
 767
 768         .read_cr0 = native_read_cr0,
 769         .write_cr0 = xen_write_cr0,
 770
 771         .read_cr4 = native_read_cr4,
 772         .read_cr4_safe = native_read_cr4_safe,
 773         .write_cr4 = xen_write_cr4,
 774
 775         .wbinvd = native_wbinvd,
 776
 777         .read_msr = native_read_msr_safe,
 778         .write_msr = xen_write_msr_safe,
 779         .read_tsc = native_read_tsc,
 780         .read_pmc = native_read_pmc,
 781
 782         .iret = xen_iret,
 783         .irq_enable_sysexit = xen_sysexit,
 784 #ifdef CONFIG_X86_64
 785         .usergs_sysret32 = xen_sysret32,
 786         .usergs_sysret64 = xen_sysret64,
 787 #endif
 788
 789         .load_tr_desc = paravirt_nop,
 790         .set_ldt = xen_set_ldt,
 791         .load_gdt = xen_load_gdt,
 792         .load_idt = xen_load_idt,
 793         .load_tls = xen_load_tls,
 794 #ifdef CONFIG_X86_64
 795         .load_gs_index = xen_load_gs_index,
 796 #endif
 797
 798         .alloc_ldt = xen_alloc_ldt,
 799         .free_ldt = xen_free_ldt,
 800
 801         .store_gdt = native_store_gdt,
 802         .store_idt = native_store_idt,
 803         .store_tr = xen_store_tr,
 804
 805         .write_ldt_entry = xen_write_ldt_entry,
 806         .write_gdt_entry = xen_write_gdt_entry,
 807         .write_idt_entry = xen_write_idt_entry,
 808         .load_sp0 = xen_load_sp0,
 809
 810         .set_iopl_mask = xen_set_iopl_mask,
 811         .io_delay = xen_io_delay,
 812
 813         /* Xen takes care of %gs when switching to usermode for us */
 814         .swapgs = paravirt_nop,
 815
 816         .lazy_mode = {
 817                 .enter = paravirt_enter_lazy_cpu,
 818                 .leave = xen_leave_lazy,
 819         },
 820 };
 821
 822 static const struct pv_apic_ops xen_apic_ops __initdata = {
 823 #ifdef CONFIG_X86_LOCAL_APIC
 824         .setup_boot_clock = paravirt_nop,
 825         .setup_secondary_clock = paravirt_nop,
 826         .startup_ipi_hook = paravirt_nop,
 827 #endif
 828 };
 829
 830 static void xen_reboot(int reason)
 831 {
 832         struct sched_shutdown r = { .reason = reason };
 833
 834 #ifdef CONFIG_SMP
 835         smp_send_stop();
 836 #endif
 837
 838         if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
 839                 BUG();
 840 }
 841
 842 static void xen_restart(char *msg)
 843 {
 844         xen_reboot(SHUTDOWN_reboot);
 845 }
 846
 847 static void xen_emergency_restart(void)
 848 {
 849         xen_reboot(SHUTDOWN_reboot);
 850 }
 851
 852 static void xen_machine_halt(void)
 853 {
 854         xen_reboot(SHUTDOWN_poweroff);
 855 }
 856
 857 static void xen_crash_shutdown(struct pt_regs *regs)
 858 {
 859         xen_reboot(SHUTDOWN_crash);
 860 }
 861
 862 static const struct machine_ops __initdata xen_machine_ops = {
 863         .restart = xen_restart,
 864         .halt = xen_machine_halt,
 865         .power_off = xen_machine_halt,
 866         .shutdown = xen_machine_halt,
 867         .crash_shutdown = xen_crash_shutdown,
 868         .emergency_restart = xen_emergency_restart,
 869 };
 870
 871
 872 /* First C function to be called on Xen boot */
 873 asmlinkage void __init xen_start_kernel(void)
 874 {
 875         pgd_t *pgd;
 876
 877         if (!xen_start_info)
 878                 return;
 879
 880         xen_domain_type = XEN_PV_DOMAIN;
 881
 882         BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
 883
 884         xen_setup_features();
 885
 886         /* Install Xen paravirt ops */
 887         pv_info = xen_info;
 888         pv_init_ops = xen_init_ops;
 889         pv_time_ops = xen_time_ops;
 890         pv_cpu_ops = xen_cpu_ops;
 891         pv_apic_ops = xen_apic_ops;
 892         pv_mmu_ops = xen_mmu_ops;
 893
 894         xen_init_irq_ops();
 895
 896 #ifdef CONFIG_X86_LOCAL_APIC
 897         /*
 898          * set up the basic apic ops.
 899          */
 900         apic_ops = &xen_basic_apic_ops;
 901 #endif
 902
 903         if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
 904                 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
 905                 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
 906         }
 907
 908         machine_ops = xen_machine_ops;
 909
 910 #ifdef CONFIG_X86_64
 911         /*
 912          * Setup percpu state.  We only need to do this for 64-bit
 913          * because 32-bit already has %fs set properly.
 914          */
 915         load_percpu_segment(0);
 916 #endif
 917         /*
 918          * The only reliable way to retain the initial address of the
 919          * percpu gdt_page is to remember it here, so we can go and
 920          * mark it RW later, when the initial percpu area is freed.
 921          */
 922         xen_initial_gdt = &per_cpu(gdt_page, 0);
 923
 924         xen_smp_init();
 925
 926         /* Get mfn list */
 927         if (!xen_feature(XENFEAT_auto_translated_physmap))
 928                 xen_build_dynamic_phys_to_machine();
 929
 930         pgd = (pgd_t *)xen_start_info->pt_base;
 931
 932         /* Prevent unwanted bits from being set in PTEs. */
 933         __supported_pte_mask &= ~_PAGE_GLOBAL;
 934         if (!xen_initial_domain())
 935                 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
 936
 937         /* Don't do the full vcpu_info placement stuff until we have a
 938            possible map and a non-dummy shared_info. */
 939         per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
 940
 941         xen_raw_console_write("mapping kernel into physical memory\n");
 942         pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
 943
 944         init_mm.pgd = pgd;
 945
 946         /* keep using Xen gdt for now; no urgent need to change it */
 947
 948         pv_info.kernel_rpl = 1;
 949         if (xen_feature(XENFEAT_supervisor_mode_kernel))
 950                 pv_info.kernel_rpl = 0;
 951
 952         /* set the limit of our address space */
 953         xen_reserve_top();
 954
 955 #ifdef CONFIG_X86_32
 956         /* set up basic CPUID stuff */
 957         cpu_detect(&new_cpu_data);
 958         new_cpu_data.hard_math = 1;
 959         new_cpu_data.x86_capability[0] = cpuid_edx(1);
 960 #endif
 961
 962         /* Poke various useful things into boot_params */
 963         boot_params.hdr.type_of_loader = (9 << 4) | 0;
 964         boot_params.hdr.ramdisk_image = xen_start_info->mod_start
 965                 ? __pa(xen_start_info->mod_start) : 0;
 966         boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
 967         boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
 968
 969         if (!xen_initial_domain()) {
 970                 add_preferred_console("xenboot", 0, NULL);
 971                 add_preferred_console("tty", 0, NULL);
 972                 add_preferred_console("hvc", 0, NULL);
 973         }
 974
 975         xen_raw_console_write("about to get started...\n");
 976
 977         /* Start the world */
 978 #ifdef CONFIG_X86_32
 979         i386_start_kernel();
 980 #else
 981         x86_64_start_reservations((char *)__pa_symbol(&boot_params));
 982 #endif
 983 }