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