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 #include <linux/mm.h>
  25 #include <linux/page-flags.h>
  26 #include <linux/highmem.h>
  27 #include <linux/smp.h>
  28
  29 #include <xen/interface/xen.h>
  30 #include <xen/interface/physdev.h>
  31 #include <xen/interface/vcpu.h>
  32 #include <xen/features.h>
  33 #include <xen/page.h>
  34
  35 #include <asm/paravirt.h>
  36 #include <asm/page.h>
  37 #include <asm/xen/hypercall.h>
  38 #include <asm/xen/hypervisor.h>
  39 #include <asm/fixmap.h>
  40 #include <asm/processor.h>
  41 #include <asm/setup.h>
  42 #include <asm/desc.h>
  43 #include <asm/pgtable.h>
  44 #include <asm/tlbflush.h>
  45
  46 #include "xen-ops.h"
  47 #include "mmu.h"
  48 #include "multicalls.h"
  49
  50 EXPORT_SYMBOL_GPL(hypercall_page);
  51
  52 DEFINE_PER_CPU(enum paravirt_lazy_mode, xen_lazy_mode);
  53
  54 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
  55 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
  56 DEFINE_PER_CPU(unsigned long, xen_cr3);
  57
  58 struct start_info *xen_start_info;
  59 EXPORT_SYMBOL_GPL(xen_start_info);
  60
  61 void xen_vcpu_setup(int cpu)
  62 {
  63         per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
  64 }
  65
  66 static void __init xen_banner(void)
  67 {
  68         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  69                paravirt_ops.name);
  70         printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic);
  71 }
  72
  73 static void xen_cpuid(unsigned int *eax, unsigned int *ebx,
  74                       unsigned int *ecx, unsigned int *edx)
  75 {
  76         unsigned maskedx = ~0;
  77
  78         /*
  79          * Mask out inconvenient features, to try and disable as many
  80          * unsupported kernel subsystems as possible.
  81          */
  82         if (*eax == 1)
  83                 maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
  84                             (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
  85                             (1 << X86_FEATURE_ACC));   /* thermal monitoring */
  86
  87         asm(XEN_EMULATE_PREFIX "cpuid"
  88                 : "=a" (*eax),
  89                   "=b" (*ebx),
  90                   "=c" (*ecx),
  91                   "=d" (*edx)
  92                 : "0" (*eax), "2" (*ecx));
  93         *edx &= maskedx;
  94 }
  95
  96 static void xen_set_debugreg(int reg, unsigned long val)
  97 {
  98         HYPERVISOR_set_debugreg(reg, val);
  99 }
 100
 101 static unsigned long xen_get_debugreg(int reg)
 102 {
 103         return HYPERVISOR_get_debugreg(reg);
 104 }
 105
 106 static unsigned long xen_save_fl(void)
 107 {
 108         struct vcpu_info *vcpu;
 109         unsigned long flags;
 110
 111         preempt_disable();
 112         vcpu = x86_read_percpu(xen_vcpu);
 113         /* flag has opposite sense of mask */
 114         flags = !vcpu->evtchn_upcall_mask;
 115         preempt_enable();
 116
 117         /* convert to IF type flag
 118            -0 -> 0x00000000
 119            -1 -> 0xffffffff
 120         */
 121         return (-flags) & X86_EFLAGS_IF;
 122 }
 123
 124 static void xen_restore_fl(unsigned long flags)
 125 {
 126         struct vcpu_info *vcpu;
 127
 128         preempt_disable();
 129
 130         /* convert from IF type flag */
 131         flags = !(flags & X86_EFLAGS_IF);
 132         vcpu = x86_read_percpu(xen_vcpu);
 133         vcpu->evtchn_upcall_mask = flags;
 134
 135         if (flags == 0) {
 136                 /* Unmask then check (avoid races).  We're only protecting
 137                    against updates by this CPU, so there's no need for
 138                    anything stronger. */
 139                 barrier();
 140
 141                 if (unlikely(vcpu->evtchn_upcall_pending))
 142                         force_evtchn_callback();
 143                 preempt_enable();
 144         } else
 145                 preempt_enable_no_resched();
 146 }
 147
 148 static void xen_irq_disable(void)
 149 {
 150         struct vcpu_info *vcpu;
 151         preempt_disable();
 152         vcpu = x86_read_percpu(xen_vcpu);
 153         vcpu->evtchn_upcall_mask = 1;
 154         preempt_enable_no_resched();
 155 }
 156
 157 static void xen_irq_enable(void)
 158 {
 159         struct vcpu_info *vcpu;
 160
 161         preempt_disable();
 162         vcpu = x86_read_percpu(xen_vcpu);
 163         vcpu->evtchn_upcall_mask = 0;
 164
 165         /* Unmask then check (avoid races).  We're only protecting
 166            against updates by this CPU, so there's no need for
 167            anything stronger. */
 168         barrier();
 169
 170         if (unlikely(vcpu->evtchn_upcall_pending))
 171                 force_evtchn_callback();
 172         preempt_enable();
 173 }
 174
 175 static void xen_safe_halt(void)
 176 {
 177         /* Blocking includes an implicit local_irq_enable(). */
 178         if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0)
 179                 BUG();
 180 }
 181
 182 static void xen_halt(void)
 183 {
 184         if (irqs_disabled())
 185                 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
 186         else
 187                 xen_safe_halt();
 188 }
 189
 190 static void xen_set_lazy_mode(enum paravirt_lazy_mode mode)
 191 {
 192         switch (mode) {
 193         case PARAVIRT_LAZY_NONE:
 194                 BUG_ON(x86_read_percpu(xen_lazy_mode) == PARAVIRT_LAZY_NONE);
 195                 break;
 196
 197         case PARAVIRT_LAZY_MMU:
 198         case PARAVIRT_LAZY_CPU:
 199                 BUG_ON(x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE);
 200                 break;
 201
 202         case PARAVIRT_LAZY_FLUSH:
 203                 /* flush if necessary, but don't change state */
 204                 if (x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE)
 205                         xen_mc_flush();
 206                 return;
 207         }
 208
 209         xen_mc_flush();
 210         x86_write_percpu(xen_lazy_mode, mode);
 211 }
 212
 213 static unsigned long xen_store_tr(void)
 214 {
 215         return 0;
 216 }
 217
 218 static void xen_set_ldt(const void *addr, unsigned entries)
 219 {
 220         unsigned long linear_addr = (unsigned long)addr;
 221         struct mmuext_op *op;
 222         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 223
 224         op = mcs.args;
 225         op->cmd = MMUEXT_SET_LDT;
 226         if (linear_addr) {
 227                 /* ldt my be vmalloced, use arbitrary_virt_to_machine */
 228                 xmaddr_t maddr;
 229                 maddr = arbitrary_virt_to_machine((unsigned long)addr);
 230                 linear_addr = (unsigned long)maddr.maddr;
 231         }
 232         op->arg1.linear_addr = linear_addr;
 233         op->arg2.nr_ents = entries;
 234
 235         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 236
 237         xen_mc_issue(PARAVIRT_LAZY_CPU);
 238 }
 239
 240 static void xen_load_gdt(const struct Xgt_desc_struct *dtr)
 241 {
 242         unsigned long *frames;
 243         unsigned long va = dtr->address;
 244         unsigned int size = dtr->size + 1;
 245         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 246         int f;
 247         struct multicall_space mcs;
 248
 249         /* A GDT can be up to 64k in size, which corresponds to 8192
 250            8-byte entries, or 16 4k pages.. */
 251
 252         BUG_ON(size > 65536);
 253         BUG_ON(va & ~PAGE_MASK);
 254
 255         mcs = xen_mc_entry(sizeof(*frames) * pages);
 256         frames = mcs.args;
 257
 258         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
 259                 frames[f] = virt_to_mfn(va);
 260                 make_lowmem_page_readonly((void *)va);
 261         }
 262
 263         MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
 264
 265         xen_mc_issue(PARAVIRT_LAZY_CPU);
 266 }
 267
 268 static void load_TLS_descriptor(struct thread_struct *t,
 269                                 unsigned int cpu, unsigned int i)
 270 {
 271         struct desc_struct *gdt = get_cpu_gdt_table(cpu);
 272         xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
 273         struct multicall_space mc = __xen_mc_entry(0);
 274
 275         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
 276 }
 277
 278 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
 279 {
 280         xen_mc_batch();
 281
 282         load_TLS_descriptor(t, cpu, 0);
 283         load_TLS_descriptor(t, cpu, 1);
 284         load_TLS_descriptor(t, cpu, 2);
 285
 286         xen_mc_issue(PARAVIRT_LAZY_CPU);
 287 }
 288
 289 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
 290                                 u32 low, u32 high)
 291 {
 292         unsigned long lp = (unsigned long)&dt[entrynum];
 293         xmaddr_t mach_lp = virt_to_machine(lp);
 294         u64 entry = (u64)high << 32 | low;
 295
 296         xen_mc_flush();
 297         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
 298                 BUG();
 299 }
 300
 301 static int cvt_gate_to_trap(int vector, u32 low, u32 high,
 302                             struct trap_info *info)
 303 {
 304         u8 type, dpl;
 305
 306         type = (high >> 8) & 0x1f;
 307         dpl = (high >> 13) & 3;
 308
 309         if (type != 0xf && type != 0xe)
 310                 return 0;
 311
 312         info->vector = vector;
 313         info->address = (high & 0xffff0000) | (low & 0x0000ffff);
 314         info->cs = low >> 16;
 315         info->flags = dpl;
 316         /* interrupt gates clear IF */
 317         if (type == 0xe)
 318                 info->flags |= 4;
 319
 320         return 1;
 321 }
 322
 323 /* Locations of each CPU's IDT */
 324 static DEFINE_PER_CPU(struct Xgt_desc_struct, idt_desc);
 325
 326 /* Set an IDT entry.  If the entry is part of the current IDT, then
 327    also update Xen. */
 328 static void xen_write_idt_entry(struct desc_struct *dt, int entrynum,
 329                                 u32 low, u32 high)
 330 {
 331
 332         int cpu = smp_processor_id();
 333         unsigned long p = (unsigned long)&dt[entrynum];
 334         unsigned long start = per_cpu(idt_desc, cpu).address;
 335         unsigned long end = start + per_cpu(idt_desc, cpu).size + 1;
 336
 337         xen_mc_flush();
 338
 339         write_dt_entry(dt, entrynum, low, high);
 340
 341         if (p >= start && (p + 8) <= end) {
 342                 struct trap_info info[2];
 343
 344                 info[1].address = 0;
 345
 346                 if (cvt_gate_to_trap(entrynum, low, high, &info[0]))
 347                         if (HYPERVISOR_set_trap_table(info))
 348                                 BUG();
 349         }
 350 }
 351
 352 static void xen_convert_trap_info(const struct Xgt_desc_struct *desc,
 353                                   struct trap_info *traps)
 354 {
 355         unsigned in, out, count;
 356
 357         count = (desc->size+1) / 8;
 358         BUG_ON(count > 256);
 359
 360         for (in = out = 0; in < count; in++) {
 361                 const u32 *entry = (u32 *)(desc->address + in * 8);
 362
 363                 if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out]))
 364                         out++;
 365         }
 366         traps[out].address = 0;
 367 }
 368
 369 void xen_copy_trap_info(struct trap_info *traps)
 370 {
 371         const struct Xgt_desc_struct *desc = &get_cpu_var(idt_desc);
 372
 373         xen_convert_trap_info(desc, traps);
 374
 375         put_cpu_var(idt_desc);
 376 }
 377
 378 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
 379    hold a spinlock to protect the static traps[] array (static because
 380    it avoids allocation, and saves stack space). */
 381 static void xen_load_idt(const struct Xgt_desc_struct *desc)
 382 {
 383         static DEFINE_SPINLOCK(lock);
 384         static struct trap_info traps[257];
 385         int cpu = smp_processor_id();
 386
 387         per_cpu(idt_desc, cpu) = *desc;
 388
 389         spin_lock(&lock);
 390
 391         xen_convert_trap_info(desc, traps);
 392
 393         xen_mc_flush();
 394         if (HYPERVISOR_set_trap_table(traps))
 395                 BUG();
 396
 397         spin_unlock(&lock);
 398 }
 399
 400 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
 401    they're handled differently. */
 402 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
 403                                 u32 low, u32 high)
 404 {
 405         switch ((high >> 8) & 0xff) {
 406         case DESCTYPE_LDT:
 407         case DESCTYPE_TSS:
 408                 /* ignore */
 409                 break;
 410
 411         default: {
 412                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
 413                 u64 desc = (u64)high << 32 | low;
 414
 415                 xen_mc_flush();
 416                 if (HYPERVISOR_update_descriptor(maddr.maddr, desc))
 417                         BUG();
 418         }
 419
 420         }
 421 }
 422
 423 static void xen_load_esp0(struct tss_struct *tss,
 424                                    struct thread_struct *thread)
 425 {
 426         struct multicall_space mcs = xen_mc_entry(0);
 427         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->esp0);
 428         xen_mc_issue(PARAVIRT_LAZY_CPU);
 429 }
 430
 431 static void xen_set_iopl_mask(unsigned mask)
 432 {
 433         struct physdev_set_iopl set_iopl;
 434
 435         /* Force the change at ring 0. */
 436         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
 437         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 438 }
 439
 440 static void xen_io_delay(void)
 441 {
 442 }
 443
 444 #ifdef CONFIG_X86_LOCAL_APIC
 445 static unsigned long xen_apic_read(unsigned long reg)
 446 {
 447         return 0;
 448 }
 449
 450 static void xen_apic_write(unsigned long reg, unsigned long val)
 451 {
 452         /* Warn to see if there's any stray references */
 453         WARN_ON(1);
 454 }
 455 #endif
 456
 457 static void xen_flush_tlb(void)
 458 {
 459         struct mmuext_op op;
 460
 461         op.cmd = MMUEXT_TLB_FLUSH_LOCAL;
 462         if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
 463                 BUG();
 464 }
 465
 466 static void xen_flush_tlb_single(unsigned long addr)
 467 {
 468         struct mmuext_op op;
 469
 470         op.cmd = MMUEXT_INVLPG_LOCAL;
 471         op.arg1.linear_addr = addr & PAGE_MASK;
 472         if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
 473                 BUG();
 474 }
 475
 476 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
 477                                  unsigned long va)
 478 {
 479         struct mmuext_op op;
 480         cpumask_t cpumask = *cpus;
 481
 482         /*
 483          * A couple of (to be removed) sanity checks:
 484          *
 485          * - current CPU must not be in mask
 486          * - mask must exist :)
 487          */
 488         BUG_ON(cpus_empty(cpumask));
 489         BUG_ON(cpu_isset(smp_processor_id(), cpumask));
 490         BUG_ON(!mm);
 491
 492         /* If a CPU which we ran on has gone down, OK. */
 493         cpus_and(cpumask, cpumask, cpu_online_map);
 494         if (cpus_empty(cpumask))
 495                 return;
 496
 497         if (va == TLB_FLUSH_ALL) {
 498                 op.cmd = MMUEXT_TLB_FLUSH_MULTI;
 499                 op.arg2.vcpumask = (void *)cpus;
 500         } else {
 501                 op.cmd = MMUEXT_INVLPG_MULTI;
 502                 op.arg1.linear_addr = va;
 503                 op.arg2.vcpumask = (void *)cpus;
 504         }
 505
 506         if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
 507                 BUG();
 508 }
 509
 510 static unsigned long xen_read_cr2(void)
 511 {
 512         return x86_read_percpu(xen_vcpu)->arch.cr2;
 513 }
 514
 515 static void xen_write_cr4(unsigned long cr4)
 516 {
 517         /* never allow TSC to be disabled */
 518         native_write_cr4(cr4 & ~X86_CR4_TSD);
 519 }
 520
 521 static unsigned long xen_read_cr3(void)
 522 {
 523         return x86_read_percpu(xen_cr3);
 524 }
 525
 526 static void xen_write_cr3(unsigned long cr3)
 527 {
 528         if (cr3 == x86_read_percpu(xen_cr3)) {
 529                 /* just a simple tlb flush */
 530                 xen_flush_tlb();
 531                 return;
 532         }
 533
 534         x86_write_percpu(xen_cr3, cr3);
 535
 536
 537         {
 538                 struct mmuext_op *op;
 539                 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 540                 unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3));
 541
 542                 op = mcs.args;
 543                 op->cmd = MMUEXT_NEW_BASEPTR;
 544                 op->arg1.mfn = mfn;
 545
 546                 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 547
 548                 xen_mc_issue(PARAVIRT_LAZY_CPU);
 549         }
 550 }
 551
 552 /* Early in boot, while setting up the initial pagetable, assume
 553    everything is pinned. */
 554 static __init void xen_alloc_pt_init(struct mm_struct *mm, u32 pfn)
 555 {
 556         BUG_ON(mem_map);        /* should only be used early */
 557         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
 558 }
 559
 560 /* This needs to make sure the new pte page is pinned iff its being
 561    attached to a pinned pagetable. */
 562 static void xen_alloc_pt(struct mm_struct *mm, u32 pfn)
 563 {
 564         struct page *page = pfn_to_page(pfn);
 565
 566         if (PagePinned(virt_to_page(mm->pgd))) {
 567                 SetPagePinned(page);
 568
 569                 if (!PageHighMem(page))
 570                         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
 571                 else
 572                         /* make sure there are no stray mappings of
 573                            this page */
 574                         kmap_flush_unused();
 575         }
 576 }
 577
 578 /* This should never happen until we're OK to use struct page */
 579 static void xen_release_pt(u32 pfn)
 580 {
 581         struct page *page = pfn_to_page(pfn);
 582
 583         if (PagePinned(page)) {
 584                 if (!PageHighMem(page))
 585                         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
 586         }
 587 }
 588
 589 #ifdef CONFIG_HIGHPTE
 590 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
 591 {
 592         pgprot_t prot = PAGE_KERNEL;
 593
 594         if (PagePinned(page))
 595                 prot = PAGE_KERNEL_RO;
 596
 597         if (0 && PageHighMem(page))
 598                 printk("mapping highpte %lx type %d prot %s\n",
 599                        page_to_pfn(page), type,
 600                        (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
 601
 602         return kmap_atomic_prot(page, type, prot);
 603 }
 604 #endif
 605
 606 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
 607 {
 608         /* If there's an existing pte, then don't allow _PAGE_RW to be set */
 609         if (pte_val_ma(*ptep) & _PAGE_PRESENT)
 610                 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
 611                                pte_val_ma(pte));
 612
 613         return pte;
 614 }
 615
 616 /* Init-time set_pte while constructing initial pagetables, which
 617    doesn't allow RO pagetable pages to be remapped RW */
 618 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
 619 {
 620         pte = mask_rw_pte(ptep, pte);
 621
 622         xen_set_pte(ptep, pte);
 623 }
 624
 625 static __init void xen_pagetable_setup_start(pgd_t *base)
 626 {
 627         pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base;
 628
 629         /* special set_pte for pagetable initialization */
 630         paravirt_ops.set_pte = xen_set_pte_init;
 631
 632         init_mm.pgd = base;
 633         /*
 634          * copy top-level of Xen-supplied pagetable into place.  For
 635          * !PAE we can use this as-is, but for PAE it is a stand-in
 636          * while we copy the pmd pages.
 637          */
 638         memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t));
 639
 640         if (PTRS_PER_PMD > 1) {
 641                 int i;
 642                 /*
 643                  * For PAE, need to allocate new pmds, rather than
 644                  * share Xen's, since Xen doesn't like pmd's being
 645                  * shared between address spaces.
 646                  */
 647                 for (i = 0; i < PTRS_PER_PGD; i++) {
 648                         if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
 649                                 pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 650
 651                                 memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
 652                                        PAGE_SIZE);
 653
 654                                 make_lowmem_page_readonly(pmd);
 655
 656                                 set_pgd(&base[i], __pgd(1 + __pa(pmd)));
 657                         } else
 658                                 pgd_clear(&base[i]);
 659                 }
 660         }
 661
 662         /* make sure zero_page is mapped RO so we can use it in pagetables */
 663         make_lowmem_page_readonly(empty_zero_page);
 664         make_lowmem_page_readonly(base);
 665         /*
 666          * Switch to new pagetable.  This is done before
 667          * pagetable_init has done anything so that the new pages
 668          * added to the table can be prepared properly for Xen.
 669          */
 670         xen_write_cr3(__pa(base));
 671 }
 672
 673 static __init void xen_pagetable_setup_done(pgd_t *base)
 674 {
 675         /* This will work as long as patching hasn't happened yet
 676            (which it hasn't) */
 677         paravirt_ops.alloc_pt = xen_alloc_pt;
 678         paravirt_ops.set_pte = xen_set_pte;
 679
 680         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 681                 /*
 682                  * Create a mapping for the shared info page.
 683                  * Should be set_fixmap(), but shared_info is a machine
 684                  * address with no corresponding pseudo-phys address.
 685                  */
 686                 set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP),
 687                             PFN_DOWN(xen_start_info->shared_info),
 688                             PAGE_KERNEL);
 689
 690                 HYPERVISOR_shared_info =
 691                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
 692
 693         } else
 694                 HYPERVISOR_shared_info =
 695                         (struct shared_info *)__va(xen_start_info->shared_info);
 696
 697         /* Actually pin the pagetable down, but we can't set PG_pinned
 698            yet because the page structures don't exist yet. */
 699         {
 700                 struct mmuext_op op;
 701 #ifdef CONFIG_X86_PAE
 702                 op.cmd = MMUEXT_PIN_L3_TABLE;
 703 #else
 704                 op.cmd = MMUEXT_PIN_L3_TABLE;
 705 #endif
 706                 op.arg1.mfn = pfn_to_mfn(PFN_DOWN(__pa(base)));
 707                 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
 708                         BUG();
 709         }
 710
 711         xen_vcpu_setup(smp_processor_id());
 712 }
 713
 714 static const struct paravirt_ops xen_paravirt_ops __initdata = {
 715         .paravirt_enabled = 1,
 716         .shared_kernel_pmd = 0,
 717
 718         .name = "Xen",
 719         .banner = xen_banner,
 720
 721         .patch = paravirt_patch_default,
 722
 723         .memory_setup = xen_memory_setup,
 724         .arch_setup = xen_arch_setup,
 725         .init_IRQ = xen_init_IRQ,
 726         .post_allocator_init = xen_mark_init_mm_pinned,
 727
 728         .time_init = xen_time_init,
 729         .set_wallclock = xen_set_wallclock,
 730         .get_wallclock = xen_get_wallclock,
 731         .get_cpu_khz = xen_cpu_khz,
 732         .sched_clock = xen_sched_clock,
 733
 734         .cpuid = xen_cpuid,
 735
 736         .set_debugreg = xen_set_debugreg,
 737         .get_debugreg = xen_get_debugreg,
 738
 739         .clts = native_clts,
 740
 741         .read_cr0 = native_read_cr0,
 742         .write_cr0 = native_write_cr0,
 743
 744         .read_cr2 = xen_read_cr2,
 745         .write_cr2 = native_write_cr2,
 746
 747         .read_cr3 = xen_read_cr3,
 748         .write_cr3 = xen_write_cr3,
 749
 750         .read_cr4 = native_read_cr4,
 751         .read_cr4_safe = native_read_cr4_safe,
 752         .write_cr4 = xen_write_cr4,
 753
 754         .save_fl = xen_save_fl,
 755         .restore_fl = xen_restore_fl,
 756         .irq_disable = xen_irq_disable,
 757         .irq_enable = xen_irq_enable,
 758         .safe_halt = xen_safe_halt,
 759         .halt = xen_halt,
 760         .wbinvd = native_wbinvd,
 761
 762         .read_msr = native_read_msr_safe,
 763         .write_msr = native_write_msr_safe,
 764         .read_tsc = native_read_tsc,
 765         .read_pmc = native_read_pmc,
 766
 767         .iret = (void *)&hypercall_page[__HYPERVISOR_iret],
 768         .irq_enable_sysexit = NULL,  /* never called */
 769
 770         .load_tr_desc = paravirt_nop,
 771         .set_ldt = xen_set_ldt,
 772         .load_gdt = xen_load_gdt,
 773         .load_idt = xen_load_idt,
 774         .load_tls = xen_load_tls,
 775
 776         .store_gdt = native_store_gdt,
 777         .store_idt = native_store_idt,
 778         .store_tr = xen_store_tr,
 779
 780         .write_ldt_entry = xen_write_ldt_entry,
 781         .write_gdt_entry = xen_write_gdt_entry,
 782         .write_idt_entry = xen_write_idt_entry,
 783         .load_esp0 = xen_load_esp0,
 784
 785         .set_iopl_mask = xen_set_iopl_mask,
 786         .io_delay = xen_io_delay,
 787
 788 #ifdef CONFIG_X86_LOCAL_APIC
 789         .apic_write = xen_apic_write,
 790         .apic_write_atomic = xen_apic_write,
 791         .apic_read = xen_apic_read,
 792         .setup_boot_clock = paravirt_nop,
 793         .setup_secondary_clock = paravirt_nop,
 794         .startup_ipi_hook = paravirt_nop,
 795 #endif
 796
 797         .flush_tlb_user = xen_flush_tlb,
 798         .flush_tlb_kernel = xen_flush_tlb,
 799         .flush_tlb_single = xen_flush_tlb_single,
 800         .flush_tlb_others = xen_flush_tlb_others,
 801
 802         .pte_update = paravirt_nop,
 803         .pte_update_defer = paravirt_nop,
 804
 805         .pagetable_setup_start = xen_pagetable_setup_start,
 806         .pagetable_setup_done = xen_pagetable_setup_done,
 807
 808         .alloc_pt = xen_alloc_pt_init,
 809         .release_pt = xen_release_pt,
 810         .alloc_pd = paravirt_nop,
 811         .alloc_pd_clone = paravirt_nop,
 812         .release_pd = paravirt_nop,
 813
 814 #ifdef CONFIG_HIGHPTE
 815         .kmap_atomic_pte = xen_kmap_atomic_pte,
 816 #endif
 817
 818         .set_pte = NULL,        /* see xen_pagetable_setup_* */
 819         .set_pte_at = xen_set_pte_at,
 820         .set_pmd = xen_set_pmd,
 821
 822         .pte_val = xen_pte_val,
 823         .pgd_val = xen_pgd_val,
 824
 825         .make_pte = xen_make_pte,
 826         .make_pgd = xen_make_pgd,
 827
 828 #ifdef CONFIG_X86_PAE
 829         .set_pte_atomic = xen_set_pte_atomic,
 830         .set_pte_present = xen_set_pte_at,
 831         .set_pud = xen_set_pud,
 832         .pte_clear = xen_pte_clear,
 833         .pmd_clear = xen_pmd_clear,
 834
 835         .make_pmd = xen_make_pmd,
 836         .pmd_val = xen_pmd_val,
 837 #endif  /* PAE */
 838
 839         .activate_mm = xen_activate_mm,
 840         .dup_mmap = xen_dup_mmap,
 841         .exit_mmap = xen_exit_mmap,
 842
 843         .set_lazy_mode = xen_set_lazy_mode,
 844 };
 845
 846 #ifdef CONFIG_SMP
 847 static const struct smp_ops xen_smp_ops __initdata = {
 848         .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
 849         .smp_prepare_cpus = xen_smp_prepare_cpus,
 850         .cpu_up = xen_cpu_up,
 851         .smp_cpus_done = xen_smp_cpus_done,
 852
 853         .smp_send_stop = xen_smp_send_stop,
 854         .smp_send_reschedule = xen_smp_send_reschedule,
 855         .smp_call_function_mask = xen_smp_call_function_mask,
 856 };
 857 #endif  /* CONFIG_SMP */
 858
 859 /* First C function to be called on Xen boot */
 860 asmlinkage void __init xen_start_kernel(void)
 861 {
 862         pgd_t *pgd;
 863
 864         if (!xen_start_info)
 865                 return;
 866
 867         BUG_ON(memcmp(xen_start_info->magic, "xen-3.0", 7) != 0);
 868
 869         /* Install Xen paravirt ops */
 870         paravirt_ops = xen_paravirt_ops;
 871 #ifdef CONFIG_SMP
 872         smp_ops = xen_smp_ops;
 873 #endif
 874
 875         xen_setup_features();
 876
 877         /* Get mfn list */
 878         if (!xen_feature(XENFEAT_auto_translated_physmap))
 879                 phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list;
 880
 881         pgd = (pgd_t *)xen_start_info->pt_base;
 882
 883         init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
 884
 885         init_mm.pgd = pgd; /* use the Xen pagetables to start */
 886
 887         /* keep using Xen gdt for now; no urgent need to change it */
 888
 889         x86_write_percpu(xen_cr3, __pa(pgd));
 890         xen_vcpu_setup(0);
 891
 892         paravirt_ops.kernel_rpl = 1;
 893         if (xen_feature(XENFEAT_supervisor_mode_kernel))
 894                 paravirt_ops.kernel_rpl = 0;
 895
 896         /* set the limit of our address space */
 897         reserve_top_address(-HYPERVISOR_VIRT_START + 2 * PAGE_SIZE);
 898
 899         /* set up basic CPUID stuff */
 900         cpu_detect(&new_cpu_data);
 901         new_cpu_data.hard_math = 1;
 902         new_cpu_data.x86_capability[0] = cpuid_edx(1);
 903
 904         /* Poke various useful things into boot_params */
 905         LOADER_TYPE = (9 << 4) | 0;
 906         INITRD_START = xen_start_info->mod_start ? __pa(xen_start_info->mod_start) : 0;
 907         INITRD_SIZE = xen_start_info->mod_len;
 908
 909         /* Start the world */
 910         start_kernel();
 911 }