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/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/smp.h>
  29
  30 #include <xen/interface/xen.h>
  31 #include <xen/interface/physdev.h>
  32 #include <xen/interface/vcpu.h>
  33 #include <xen/features.h>
  34 #include <xen/page.h>
  35
  36 #include <asm/paravirt.h>
  37 #include <asm/page.h>
  38 #include <asm/xen/hypercall.h>
  39 #include <asm/xen/hypervisor.h>
  40 #include <asm/fixmap.h>
  41 #include <asm/processor.h>
  42 #include <asm/setup.h>
  43 #include <asm/desc.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/tlbflush.h>
  46
  47 #include "xen-ops.h"
  48 #include "mmu.h"
  49 #include "multicalls.h"
  50
  51 EXPORT_SYMBOL_GPL(hypercall_page);
  52
  53 DEFINE_PER_CPU(enum paravirt_lazy_mode, xen_lazy_mode);
  54
  55 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
  56 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
  57 DEFINE_PER_CPU(unsigned long, xen_cr3);
  58
  59 struct start_info *xen_start_info;
  60 EXPORT_SYMBOL_GPL(xen_start_info);
  61
  62 void xen_vcpu_setup(int cpu)
  63 {
  64         per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
  65 }
  66
  67 static void __init xen_banner(void)
  68 {
  69         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  70                paravirt_ops.name);
  71         printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic);
  72 }
  73
  74 static void xen_cpuid(unsigned int *eax, unsigned int *ebx,
  75                       unsigned int *ecx, unsigned int *edx)
  76 {
  77         unsigned maskedx = ~0;
  78
  79         /*
  80          * Mask out inconvenient features, to try and disable as many
  81          * unsupported kernel subsystems as possible.
  82          */
  83         if (*eax == 1)
  84                 maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
  85                             (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
  86                             (1 << X86_FEATURE_ACC));   /* thermal monitoring */
  87
  88         asm(XEN_EMULATE_PREFIX "cpuid"
  89                 : "=a" (*eax),
  90                   "=b" (*ebx),
  91                   "=c" (*ecx),
  92                   "=d" (*edx)
  93                 : "0" (*eax), "2" (*ecx));
  94         *edx &= maskedx;
  95 }
  96
  97 static void xen_set_debugreg(int reg, unsigned long val)
  98 {
  99         HYPERVISOR_set_debugreg(reg, val);
 100 }
 101
 102 static unsigned long xen_get_debugreg(int reg)
 103 {
 104         return HYPERVISOR_get_debugreg(reg);
 105 }
 106
 107 static unsigned long xen_save_fl(void)
 108 {
 109         struct vcpu_info *vcpu;
 110         unsigned long flags;
 111
 112         vcpu = x86_read_percpu(xen_vcpu);
 113
 114         /* flag has opposite sense of mask */
 115         flags = !vcpu->evtchn_upcall_mask;
 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         /* convert from IF type flag */
 129         flags = !(flags & X86_EFLAGS_IF);
 130
 131         /* There's a one instruction preempt window here.  We need to
 132            make sure we're don't switch CPUs between getting the vcpu
 133            pointer and updating the mask. */
 134         preempt_disable();
 135         vcpu = x86_read_percpu(xen_vcpu);
 136         vcpu->evtchn_upcall_mask = flags;
 137         preempt_enable_no_resched();
 138
 139         /* Doesn't matter if we get preempted here, because any
 140            pending event will get dealt with anyway. */
 141
 142         if (flags == 0) {
 143                 preempt_check_resched();
 144                 barrier(); /* unmask then check (avoid races) */
 145                 if (unlikely(vcpu->evtchn_upcall_pending))
 146                         force_evtchn_callback();
 147         }
 148 }
 149
 150 static void xen_irq_disable(void)
 151 {
 152         /* There's a one instruction preempt window here.  We need to
 153            make sure we're don't switch CPUs between getting the vcpu
 154            pointer and updating the mask. */
 155         preempt_disable();
 156         x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1;
 157         preempt_enable_no_resched();
 158 }
 159
 160 static void xen_irq_enable(void)
 161 {
 162         struct vcpu_info *vcpu;
 163
 164         /* There's a one instruction preempt window here.  We need to
 165            make sure we're don't switch CPUs between getting the vcpu
 166            pointer and updating the mask. */
 167         preempt_disable();
 168         vcpu = x86_read_percpu(xen_vcpu);
 169         vcpu->evtchn_upcall_mask = 0;
 170         preempt_enable_no_resched();
 171
 172         /* Doesn't matter if we get preempted here, because any
 173            pending event will get dealt with anyway. */
 174
 175         barrier(); /* unmask then check (avoid races) */
 176         if (unlikely(vcpu->evtchn_upcall_pending))
 177                 force_evtchn_callback();
 178 }
 179
 180 static void xen_safe_halt(void)
 181 {
 182         /* Blocking includes an implicit local_irq_enable(). */
 183         if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0)
 184                 BUG();
 185 }
 186
 187 static void xen_halt(void)
 188 {
 189         if (irqs_disabled())
 190                 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
 191         else
 192                 xen_safe_halt();
 193 }
 194
 195 static void xen_set_lazy_mode(enum paravirt_lazy_mode mode)
 196 {
 197         BUG_ON(preemptible());
 198
 199         switch (mode) {
 200         case PARAVIRT_LAZY_NONE:
 201                 BUG_ON(x86_read_percpu(xen_lazy_mode) == PARAVIRT_LAZY_NONE);
 202                 break;
 203
 204         case PARAVIRT_LAZY_MMU:
 205         case PARAVIRT_LAZY_CPU:
 206                 BUG_ON(x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE);
 207                 break;
 208
 209         case PARAVIRT_LAZY_FLUSH:
 210                 /* flush if necessary, but don't change state */
 211                 if (x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE)
 212                         xen_mc_flush();
 213                 return;
 214         }
 215
 216         xen_mc_flush();
 217         x86_write_percpu(xen_lazy_mode, mode);
 218 }
 219
 220 static unsigned long xen_store_tr(void)
 221 {
 222         return 0;
 223 }
 224
 225 static void xen_set_ldt(const void *addr, unsigned entries)
 226 {
 227         unsigned long linear_addr = (unsigned long)addr;
 228         struct mmuext_op *op;
 229         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 230
 231         op = mcs.args;
 232         op->cmd = MMUEXT_SET_LDT;
 233         if (linear_addr) {
 234                 /* ldt my be vmalloced, use arbitrary_virt_to_machine */
 235                 xmaddr_t maddr;
 236                 maddr = arbitrary_virt_to_machine((unsigned long)addr);
 237                 linear_addr = (unsigned long)maddr.maddr;
 238         }
 239         op->arg1.linear_addr = linear_addr;
 240         op->arg2.nr_ents = entries;
 241
 242         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 243
 244         xen_mc_issue(PARAVIRT_LAZY_CPU);
 245 }
 246
 247 static void xen_load_gdt(const struct Xgt_desc_struct *dtr)
 248 {
 249         unsigned long *frames;
 250         unsigned long va = dtr->address;
 251         unsigned int size = dtr->size + 1;
 252         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 253         int f;
 254         struct multicall_space mcs;
 255
 256         /* A GDT can be up to 64k in size, which corresponds to 8192
 257            8-byte entries, or 16 4k pages.. */
 258
 259         BUG_ON(size > 65536);
 260         BUG_ON(va & ~PAGE_MASK);
 261
 262         mcs = xen_mc_entry(sizeof(*frames) * pages);
 263         frames = mcs.args;
 264
 265         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
 266                 frames[f] = virt_to_mfn(va);
 267                 make_lowmem_page_readonly((void *)va);
 268         }
 269
 270         MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
 271
 272         xen_mc_issue(PARAVIRT_LAZY_CPU);
 273 }
 274
 275 static void load_TLS_descriptor(struct thread_struct *t,
 276                                 unsigned int cpu, unsigned int i)
 277 {
 278         struct desc_struct *gdt = get_cpu_gdt_table(cpu);
 279         xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
 280         struct multicall_space mc = __xen_mc_entry(0);
 281
 282         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
 283 }
 284
 285 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
 286 {
 287         xen_mc_batch();
 288
 289         load_TLS_descriptor(t, cpu, 0);
 290         load_TLS_descriptor(t, cpu, 1);
 291         load_TLS_descriptor(t, cpu, 2);
 292
 293         xen_mc_issue(PARAVIRT_LAZY_CPU);
 294 }
 295
 296 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
 297                                 u32 low, u32 high)
 298 {
 299         unsigned long lp = (unsigned long)&dt[entrynum];
 300         xmaddr_t mach_lp = virt_to_machine(lp);
 301         u64 entry = (u64)high << 32 | low;
 302
 303         preempt_disable();
 304
 305         xen_mc_flush();
 306         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
 307                 BUG();
 308
 309         preempt_enable();
 310 }
 311
 312 static int cvt_gate_to_trap(int vector, u32 low, u32 high,
 313                             struct trap_info *info)
 314 {
 315         u8 type, dpl;
 316
 317         type = (high >> 8) & 0x1f;
 318         dpl = (high >> 13) & 3;
 319
 320         if (type != 0xf && type != 0xe)
 321                 return 0;
 322
 323         info->vector = vector;
 324         info->address = (high & 0xffff0000) | (low & 0x0000ffff);
 325         info->cs = low >> 16;
 326         info->flags = dpl;
 327         /* interrupt gates clear IF */
 328         if (type == 0xe)
 329                 info->flags |= 4;
 330
 331         return 1;
 332 }
 333
 334 /* Locations of each CPU's IDT */
 335 static DEFINE_PER_CPU(struct Xgt_desc_struct, idt_desc);
 336
 337 /* Set an IDT entry.  If the entry is part of the current IDT, then
 338    also update Xen. */
 339 static void xen_write_idt_entry(struct desc_struct *dt, int entrynum,
 340                                 u32 low, u32 high)
 341 {
 342         unsigned long p = (unsigned long)&dt[entrynum];
 343         unsigned long start, end;
 344
 345         preempt_disable();
 346
 347         start = __get_cpu_var(idt_desc).address;
 348         end = start + __get_cpu_var(idt_desc).size + 1;
 349
 350         xen_mc_flush();
 351
 352         write_dt_entry(dt, entrynum, low, high);
 353
 354         if (p >= start && (p + 8) <= end) {
 355                 struct trap_info info[2];
 356
 357                 info[1].address = 0;
 358
 359                 if (cvt_gate_to_trap(entrynum, low, high, &info[0]))
 360                         if (HYPERVISOR_set_trap_table(info))
 361                                 BUG();
 362         }
 363
 364         preempt_enable();
 365 }
 366
 367 static void xen_convert_trap_info(const struct Xgt_desc_struct *desc,
 368                                   struct trap_info *traps)
 369 {
 370         unsigned in, out, count;
 371
 372         count = (desc->size+1) / 8;
 373         BUG_ON(count > 256);
 374
 375         for (in = out = 0; in < count; in++) {
 376                 const u32 *entry = (u32 *)(desc->address + in * 8);
 377
 378                 if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out]))
 379                         out++;
 380         }
 381         traps[out].address = 0;
 382 }
 383
 384 void xen_copy_trap_info(struct trap_info *traps)
 385 {
 386         const struct Xgt_desc_struct *desc = &__get_cpu_var(idt_desc);
 387
 388         xen_convert_trap_info(desc, traps);
 389 }
 390
 391 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
 392    hold a spinlock to protect the static traps[] array (static because
 393    it avoids allocation, and saves stack space). */
 394 static void xen_load_idt(const struct Xgt_desc_struct *desc)
 395 {
 396         static DEFINE_SPINLOCK(lock);
 397         static struct trap_info traps[257];
 398
 399         spin_lock(&lock);
 400
 401         __get_cpu_var(idt_desc) = *desc;
 402
 403         xen_convert_trap_info(desc, traps);
 404
 405         xen_mc_flush();
 406         if (HYPERVISOR_set_trap_table(traps))
 407                 BUG();
 408
 409         spin_unlock(&lock);
 410 }
 411
 412 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
 413    they're handled differently. */
 414 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
 415                                 u32 low, u32 high)
 416 {
 417         preempt_disable();
 418
 419         switch ((high >> 8) & 0xff) {
 420         case DESCTYPE_LDT:
 421         case DESCTYPE_TSS:
 422                 /* ignore */
 423                 break;
 424
 425         default: {
 426                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
 427                 u64 desc = (u64)high << 32 | low;
 428
 429                 xen_mc_flush();
 430                 if (HYPERVISOR_update_descriptor(maddr.maddr, desc))
 431                         BUG();
 432         }
 433
 434         }
 435
 436         preempt_enable();
 437 }
 438
 439 static void xen_load_esp0(struct tss_struct *tss,
 440                           struct thread_struct *thread)
 441 {
 442         struct multicall_space mcs = xen_mc_entry(0);
 443         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->esp0);
 444         xen_mc_issue(PARAVIRT_LAZY_CPU);
 445 }
 446
 447 static void xen_set_iopl_mask(unsigned mask)
 448 {
 449         struct physdev_set_iopl set_iopl;
 450
 451         /* Force the change at ring 0. */
 452         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
 453         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 454 }
 455
 456 static void xen_io_delay(void)
 457 {
 458 }
 459
 460 #ifdef CONFIG_X86_LOCAL_APIC
 461 static unsigned long xen_apic_read(unsigned long reg)
 462 {
 463         return 0;
 464 }
 465
 466 static void xen_apic_write(unsigned long reg, unsigned long val)
 467 {
 468         /* Warn to see if there's any stray references */
 469         WARN_ON(1);
 470 }
 471 #endif
 472
 473 static void xen_flush_tlb(void)
 474 {
 475         struct mmuext_op *op;
 476         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 477
 478         op = mcs.args;
 479         op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
 480         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 481
 482         xen_mc_issue(PARAVIRT_LAZY_MMU);
 483 }
 484
 485 static void xen_flush_tlb_single(unsigned long addr)
 486 {
 487         struct mmuext_op *op;
 488         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 489
 490         op = mcs.args;
 491         op->cmd = MMUEXT_INVLPG_LOCAL;
 492         op->arg1.linear_addr = addr & PAGE_MASK;
 493         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 494
 495         xen_mc_issue(PARAVIRT_LAZY_MMU);
 496 }
 497
 498 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
 499                                  unsigned long va)
 500 {
 501         struct {
 502                 struct mmuext_op op;
 503                 cpumask_t mask;
 504         } *args;
 505         cpumask_t cpumask = *cpus;
 506         struct multicall_space mcs;
 507
 508         /*
 509          * A couple of (to be removed) sanity checks:
 510          *
 511          * - current CPU must not be in mask
 512          * - mask must exist :)
 513          */
 514         BUG_ON(cpus_empty(cpumask));
 515         BUG_ON(cpu_isset(smp_processor_id(), cpumask));
 516         BUG_ON(!mm);
 517
 518         /* If a CPU which we ran on has gone down, OK. */
 519         cpus_and(cpumask, cpumask, cpu_online_map);
 520         if (cpus_empty(cpumask))
 521                 return;
 522
 523         mcs = xen_mc_entry(sizeof(*args));
 524         args = mcs.args;
 525         args->mask = cpumask;
 526         args->op.arg2.vcpumask = &args->mask;
 527
 528         if (va == TLB_FLUSH_ALL) {
 529                 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
 530         } else {
 531                 args->op.cmd = MMUEXT_INVLPG_MULTI;
 532                 args->op.arg1.linear_addr = va;
 533         }
 534
 535         MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
 536
 537         xen_mc_issue(PARAVIRT_LAZY_MMU);
 538 }
 539
 540 static unsigned long xen_read_cr2(void)
 541 {
 542         return x86_read_percpu(xen_vcpu)->arch.cr2;
 543 }
 544
 545 static void xen_write_cr4(unsigned long cr4)
 546 {
 547         /* never allow TSC to be disabled */
 548         native_write_cr4(cr4 & ~X86_CR4_TSD);
 549 }
 550
 551 static unsigned long xen_read_cr3(void)
 552 {
 553         return x86_read_percpu(xen_cr3);
 554 }
 555
 556 static void xen_write_cr3(unsigned long cr3)
 557 {
 558         BUG_ON(preemptible());
 559
 560         if (cr3 == x86_read_percpu(xen_cr3)) {
 561                 /* just a simple tlb flush */
 562                 xen_flush_tlb();
 563                 return;
 564         }
 565
 566         x86_write_percpu(xen_cr3, cr3);
 567
 568
 569         {
 570                 struct mmuext_op *op;
 571                 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 572                 unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3));
 573
 574                 op = mcs.args;
 575                 op->cmd = MMUEXT_NEW_BASEPTR;
 576                 op->arg1.mfn = mfn;
 577
 578                 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 579
 580                 xen_mc_issue(PARAVIRT_LAZY_CPU);
 581         }
 582 }
 583
 584 /* Early in boot, while setting up the initial pagetable, assume
 585    everything is pinned. */
 586 static __init void xen_alloc_pt_init(struct mm_struct *mm, u32 pfn)
 587 {
 588         BUG_ON(mem_map);        /* should only be used early */
 589         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
 590 }
 591
 592 /* This needs to make sure the new pte page is pinned iff its being
 593    attached to a pinned pagetable. */
 594 static void xen_alloc_pt(struct mm_struct *mm, u32 pfn)
 595 {
 596         struct page *page = pfn_to_page(pfn);
 597
 598         if (PagePinned(virt_to_page(mm->pgd))) {
 599                 SetPagePinned(page);
 600
 601                 if (!PageHighMem(page))
 602                         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
 603                 else
 604                         /* make sure there are no stray mappings of
 605                            this page */
 606                         kmap_flush_unused();
 607         }
 608 }
 609
 610 /* This should never happen until we're OK to use struct page */
 611 static void xen_release_pt(u32 pfn)
 612 {
 613         struct page *page = pfn_to_page(pfn);
 614
 615         if (PagePinned(page)) {
 616                 if (!PageHighMem(page))
 617                         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
 618         }
 619 }
 620
 621 #ifdef CONFIG_HIGHPTE
 622 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
 623 {
 624         pgprot_t prot = PAGE_KERNEL;
 625
 626         if (PagePinned(page))
 627                 prot = PAGE_KERNEL_RO;
 628
 629         if (0 && PageHighMem(page))
 630                 printk("mapping highpte %lx type %d prot %s\n",
 631                        page_to_pfn(page), type,
 632                        (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
 633
 634         return kmap_atomic_prot(page, type, prot);
 635 }
 636 #endif
 637
 638 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
 639 {
 640         /* If there's an existing pte, then don't allow _PAGE_RW to be set */
 641         if (pte_val_ma(*ptep) & _PAGE_PRESENT)
 642                 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
 643                                pte_val_ma(pte));
 644
 645         return pte;
 646 }
 647
 648 /* Init-time set_pte while constructing initial pagetables, which
 649    doesn't allow RO pagetable pages to be remapped RW */
 650 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
 651 {
 652         pte = mask_rw_pte(ptep, pte);
 653
 654         xen_set_pte(ptep, pte);
 655 }
 656
 657 static __init void xen_pagetable_setup_start(pgd_t *base)
 658 {
 659         pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base;
 660
 661         /* special set_pte for pagetable initialization */
 662         paravirt_ops.set_pte = xen_set_pte_init;
 663
 664         init_mm.pgd = base;
 665         /*
 666          * copy top-level of Xen-supplied pagetable into place.  For
 667          * !PAE we can use this as-is, but for PAE it is a stand-in
 668          * while we copy the pmd pages.
 669          */
 670         memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t));
 671
 672         if (PTRS_PER_PMD > 1) {
 673                 int i;
 674                 /*
 675                  * For PAE, need to allocate new pmds, rather than
 676                  * share Xen's, since Xen doesn't like pmd's being
 677                  * shared between address spaces.
 678                  */
 679                 for (i = 0; i < PTRS_PER_PGD; i++) {
 680                         if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
 681                                 pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 682
 683                                 memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
 684                                        PAGE_SIZE);
 685
 686                                 make_lowmem_page_readonly(pmd);
 687
 688                                 set_pgd(&base[i], __pgd(1 + __pa(pmd)));
 689                         } else
 690                                 pgd_clear(&base[i]);
 691                 }
 692         }
 693
 694         /* make sure zero_page is mapped RO so we can use it in pagetables */
 695         make_lowmem_page_readonly(empty_zero_page);
 696         make_lowmem_page_readonly(base);
 697         /*
 698          * Switch to new pagetable.  This is done before
 699          * pagetable_init has done anything so that the new pages
 700          * added to the table can be prepared properly for Xen.
 701          */
 702         xen_write_cr3(__pa(base));
 703 }
 704
 705 static __init void xen_pagetable_setup_done(pgd_t *base)
 706 {
 707         /* This will work as long as patching hasn't happened yet
 708            (which it hasn't) */
 709         paravirt_ops.alloc_pt = xen_alloc_pt;
 710         paravirt_ops.set_pte = xen_set_pte;
 711
 712         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 713                 /*
 714                  * Create a mapping for the shared info page.
 715                  * Should be set_fixmap(), but shared_info is a machine
 716                  * address with no corresponding pseudo-phys address.
 717                  */
 718                 set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP),
 719                             PFN_DOWN(xen_start_info->shared_info),
 720                             PAGE_KERNEL);
 721
 722                 HYPERVISOR_shared_info =
 723                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
 724
 725         } else
 726                 HYPERVISOR_shared_info =
 727                         (struct shared_info *)__va(xen_start_info->shared_info);
 728
 729         /* Actually pin the pagetable down, but we can't set PG_pinned
 730            yet because the page structures don't exist yet. */
 731         {
 732                 struct mmuext_op op;
 733 #ifdef CONFIG_X86_PAE
 734                 op.cmd = MMUEXT_PIN_L3_TABLE;
 735 #else
 736                 op.cmd = MMUEXT_PIN_L3_TABLE;
 737 #endif
 738                 op.arg1.mfn = pfn_to_mfn(PFN_DOWN(__pa(base)));
 739                 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
 740                         BUG();
 741         }
 742
 743         xen_vcpu_setup(smp_processor_id());
 744 }
 745
 746 static const struct paravirt_ops xen_paravirt_ops __initdata = {
 747         .paravirt_enabled = 1,
 748         .shared_kernel_pmd = 0,
 749
 750         .name = "Xen",
 751         .banner = xen_banner,
 752
 753         .patch = paravirt_patch_default,
 754
 755         .memory_setup = xen_memory_setup,
 756         .arch_setup = xen_arch_setup,
 757         .init_IRQ = xen_init_IRQ,
 758         .post_allocator_init = xen_mark_init_mm_pinned,
 759
 760         .time_init = xen_time_init,
 761         .set_wallclock = xen_set_wallclock,
 762         .get_wallclock = xen_get_wallclock,
 763         .get_cpu_khz = xen_cpu_khz,
 764         .sched_clock = xen_sched_clock,
 765
 766         .cpuid = xen_cpuid,
 767
 768         .set_debugreg = xen_set_debugreg,
 769         .get_debugreg = xen_get_debugreg,
 770
 771         .clts = native_clts,
 772
 773         .read_cr0 = native_read_cr0,
 774         .write_cr0 = native_write_cr0,
 775
 776         .read_cr2 = xen_read_cr2,
 777         .write_cr2 = native_write_cr2,
 778
 779         .read_cr3 = xen_read_cr3,
 780         .write_cr3 = xen_write_cr3,
 781
 782         .read_cr4 = native_read_cr4,
 783         .read_cr4_safe = native_read_cr4_safe,
 784         .write_cr4 = xen_write_cr4,
 785
 786         .save_fl = xen_save_fl,
 787         .restore_fl = xen_restore_fl,
 788         .irq_disable = xen_irq_disable,
 789         .irq_enable = xen_irq_enable,
 790         .safe_halt = xen_safe_halt,
 791         .halt = xen_halt,
 792         .wbinvd = native_wbinvd,
 793
 794         .read_msr = native_read_msr_safe,
 795         .write_msr = native_write_msr_safe,
 796         .read_tsc = native_read_tsc,
 797         .read_pmc = native_read_pmc,
 798
 799         .iret = (void *)&hypercall_page[__HYPERVISOR_iret],
 800         .irq_enable_sysexit = NULL,  /* never called */
 801
 802         .load_tr_desc = paravirt_nop,
 803         .set_ldt = xen_set_ldt,
 804         .load_gdt = xen_load_gdt,
 805         .load_idt = xen_load_idt,
 806         .load_tls = xen_load_tls,
 807
 808         .store_gdt = native_store_gdt,
 809         .store_idt = native_store_idt,
 810         .store_tr = xen_store_tr,
 811
 812         .write_ldt_entry = xen_write_ldt_entry,
 813         .write_gdt_entry = xen_write_gdt_entry,
 814         .write_idt_entry = xen_write_idt_entry,
 815         .load_esp0 = xen_load_esp0,
 816
 817         .set_iopl_mask = xen_set_iopl_mask,
 818         .io_delay = xen_io_delay,
 819
 820 #ifdef CONFIG_X86_LOCAL_APIC
 821         .apic_write = xen_apic_write,
 822         .apic_write_atomic = xen_apic_write,
 823         .apic_read = xen_apic_read,
 824         .setup_boot_clock = paravirt_nop,
 825         .setup_secondary_clock = paravirt_nop,
 826         .startup_ipi_hook = paravirt_nop,
 827 #endif
 828
 829         .flush_tlb_user = xen_flush_tlb,
 830         .flush_tlb_kernel = xen_flush_tlb,
 831         .flush_tlb_single = xen_flush_tlb_single,
 832         .flush_tlb_others = xen_flush_tlb_others,
 833
 834         .pte_update = paravirt_nop,
 835         .pte_update_defer = paravirt_nop,
 836
 837         .pagetable_setup_start = xen_pagetable_setup_start,
 838         .pagetable_setup_done = xen_pagetable_setup_done,
 839
 840         .alloc_pt = xen_alloc_pt_init,
 841         .release_pt = xen_release_pt,
 842         .alloc_pd = paravirt_nop,
 843         .alloc_pd_clone = paravirt_nop,
 844         .release_pd = paravirt_nop,
 845
 846 #ifdef CONFIG_HIGHPTE
 847         .kmap_atomic_pte = xen_kmap_atomic_pte,
 848 #endif
 849
 850         .set_pte = NULL,        /* see xen_pagetable_setup_* */
 851         .set_pte_at = xen_set_pte_at,
 852         .set_pmd = xen_set_pmd,
 853
 854         .pte_val = xen_pte_val,
 855         .pgd_val = xen_pgd_val,
 856
 857         .make_pte = xen_make_pte,
 858         .make_pgd = xen_make_pgd,
 859
 860 #ifdef CONFIG_X86_PAE
 861         .set_pte_atomic = xen_set_pte_atomic,
 862         .set_pte_present = xen_set_pte_at,
 863         .set_pud = xen_set_pud,
 864         .pte_clear = xen_pte_clear,
 865         .pmd_clear = xen_pmd_clear,
 866
 867         .make_pmd = xen_make_pmd,
 868         .pmd_val = xen_pmd_val,
 869 #endif  /* PAE */
 870
 871         .activate_mm = xen_activate_mm,
 872         .dup_mmap = xen_dup_mmap,
 873         .exit_mmap = xen_exit_mmap,
 874
 875         .set_lazy_mode = xen_set_lazy_mode,
 876 };
 877
 878 #ifdef CONFIG_SMP
 879 static const struct smp_ops xen_smp_ops __initdata = {
 880         .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
 881         .smp_prepare_cpus = xen_smp_prepare_cpus,
 882         .cpu_up = xen_cpu_up,
 883         .smp_cpus_done = xen_smp_cpus_done,
 884
 885         .smp_send_stop = xen_smp_send_stop,
 886         .smp_send_reschedule = xen_smp_send_reschedule,
 887         .smp_call_function_mask = xen_smp_call_function_mask,
 888 };
 889 #endif  /* CONFIG_SMP */
 890
 891 /* First C function to be called on Xen boot */
 892 asmlinkage void __init xen_start_kernel(void)
 893 {
 894         pgd_t *pgd;
 895
 896         if (!xen_start_info)
 897                 return;
 898
 899         BUG_ON(memcmp(xen_start_info->magic, "xen-3.0", 7) != 0);
 900
 901         /* Install Xen paravirt ops */
 902         paravirt_ops = xen_paravirt_ops;
 903 #ifdef CONFIG_SMP
 904         smp_ops = xen_smp_ops;
 905 #endif
 906
 907         xen_setup_features();
 908
 909         /* Get mfn list */
 910         if (!xen_feature(XENFEAT_auto_translated_physmap))
 911                 phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list;
 912
 913         pgd = (pgd_t *)xen_start_info->pt_base;
 914
 915         init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
 916
 917         init_mm.pgd = pgd; /* use the Xen pagetables to start */
 918
 919         /* keep using Xen gdt for now; no urgent need to change it */
 920
 921         x86_write_percpu(xen_cr3, __pa(pgd));
 922         xen_vcpu_setup(0);
 923
 924         paravirt_ops.kernel_rpl = 1;
 925         if (xen_feature(XENFEAT_supervisor_mode_kernel))
 926                 paravirt_ops.kernel_rpl = 0;
 927
 928         /* set the limit of our address space */
 929         reserve_top_address(-HYPERVISOR_VIRT_START + 2 * PAGE_SIZE);
 930
 931         /* set up basic CPUID stuff */
 932         cpu_detect(&new_cpu_data);
 933         new_cpu_data.hard_math = 1;
 934         new_cpu_data.x86_capability[0] = cpuid_edx(1);
 935
 936         /* Poke various useful things into boot_params */
 937         LOADER_TYPE = (9 << 4) | 0;
 938         INITRD_START = xen_start_info->mod_start ? __pa(xen_start_info->mod_start) : 0;
 939         INITRD_SIZE = xen_start_info->mod_len;
 940
 941         /* Start the world */
 942         start_kernel();
 943 }