2 * Core of Xen paravirt_ops implementation.
4 * This file contains the xen_paravirt_ops structure itself, and the
6 * - privileged instructions
11 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
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>
26 #include <linux/page-flags.h>
27 #include <linux/highmem.h>
28 #include <linux/console.h>
30 #include <xen/interface/xen.h>
31 #include <xen/interface/physdev.h>
32 #include <xen/interface/vcpu.h>
33 #include <xen/features.h>
35 #include <xen/hvc-console.h>
37 #include <asm/paravirt.h>
39 #include <asm/xen/hypercall.h>
40 #include <asm/xen/hypervisor.h>
41 #include <asm/fixmap.h>
42 #include <asm/processor.h>
43 #include <asm/msr-index.h>
44 #include <asm/setup.h>
46 #include <asm/pgtable.h>
47 #include <asm/tlbflush.h>
48 #include <asm/reboot.h>
52 #include "multicalls.h"
54 EXPORT_SYMBOL_GPL(hypercall_page);
56 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
57 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
60 * Identity map, in addition to plain kernel map. This needs to be
61 * large enough to allocate page table pages to allocate the rest.
62 * Each page can map 2MB.
64 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
67 /* l3 pud for userspace vsyscall mapping */
68 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
69 #endif /* CONFIG_X86_64 */
72 * Note about cr3 (pagetable base) values:
74 * xen_cr3 contains the current logical cr3 value; it contains the
75 * last set cr3. This may not be the current effective cr3, because
76 * its update may be being lazily deferred. However, a vcpu looking
77 * at its own cr3 can use this value knowing that it everything will
80 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
81 * hypercall to set the vcpu cr3 is complete (so it may be a little
82 * out of date, but it will never be set early). If one vcpu is
83 * looking at another vcpu's cr3 value, it should use this variable.
85 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
86 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
88 struct start_info *xen_start_info;
89 EXPORT_SYMBOL_GPL(xen_start_info);
91 struct shared_info xen_dummy_shared_info;
94 * Point at some empty memory to start with. We map the real shared_info
95 * page as soon as fixmap is up and running.
97 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
100 * Flag to determine whether vcpu info placement is available on all
101 * VCPUs. We assume it is to start with, and then set it to zero on
102 * the first failure. This is because it can succeed on some VCPUs
103 * and not others, since it can involve hypervisor memory allocation,
104 * or because the guest failed to guarantee all the appropriate
105 * constraints on all VCPUs (ie buffer can't cross a page boundary).
107 * Note that any particular CPU may be using a placed vcpu structure,
108 * but we can only optimise if the all are.
110 * 0: not available, 1: available
112 static int have_vcpu_info_placement = 1;
114 static void xen_vcpu_setup(int cpu)
116 struct vcpu_register_vcpu_info info;
118 struct vcpu_info *vcpup;
120 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
121 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
123 if (!have_vcpu_info_placement)
124 return; /* already tested, not available */
126 vcpup = &per_cpu(xen_vcpu_info, cpu);
128 info.mfn = virt_to_mfn(vcpup);
129 info.offset = offset_in_page(vcpup);
131 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
132 cpu, vcpup, info.mfn, info.offset);
134 /* Check to see if the hypervisor will put the vcpu_info
135 structure where we want it, which allows direct access via
136 a percpu-variable. */
137 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
140 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
141 have_vcpu_info_placement = 0;
143 /* This cpu is using the registered vcpu info, even if
144 later ones fail to. */
145 per_cpu(xen_vcpu, cpu) = vcpup;
147 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
153 * On restore, set the vcpu placement up again.
154 * If it fails, then we're in a bad state, since
155 * we can't back out from using it...
157 void xen_vcpu_restore(void)
159 if (have_vcpu_info_placement) {
162 for_each_online_cpu(cpu) {
163 bool other_cpu = (cpu != smp_processor_id());
166 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
172 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
176 BUG_ON(!have_vcpu_info_placement);
180 static void __init xen_banner(void)
182 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
183 struct xen_extraversion extra;
184 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
186 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
188 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
189 version >> 16, version & 0xffff, extra.extraversion,
190 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
193 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
194 unsigned int *cx, unsigned int *dx)
196 unsigned maskedx = ~0;
199 * Mask out inconvenient features, to try and disable as many
200 * unsupported kernel subsystems as possible.
203 maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */
204 (1 << X86_FEATURE_ACPI) | /* disable ACPI */
205 (1 << X86_FEATURE_MCE) | /* disable MCE */
206 (1 << X86_FEATURE_MCA) | /* disable MCA */
207 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
209 asm(XEN_EMULATE_PREFIX "cpuid"
214 : "0" (*ax), "2" (*cx));
218 static void xen_set_debugreg(int reg, unsigned long val)
220 HYPERVISOR_set_debugreg(reg, val);
223 static unsigned long xen_get_debugreg(int reg)
225 return HYPERVISOR_get_debugreg(reg);
228 static void xen_leave_lazy(void)
230 paravirt_leave_lazy(paravirt_get_lazy_mode());
234 static unsigned long xen_store_tr(void)
240 * If 'v' is a vmalloc mapping, then find the linear mapping of the
241 * page (if any) and also set its protections to match:
243 static void set_aliased_prot(void *v, pgprot_t prot)
251 ptep = lookup_address((unsigned long)v, &level);
252 BUG_ON(ptep == NULL);
254 pfn = pte_pfn(*ptep);
255 page = pfn_to_page(pfn);
257 pte = pfn_pte(pfn, prot);
259 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
262 if (!PageHighMem(page)) {
263 void *av = __va(PFN_PHYS(pfn));
266 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
272 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
274 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
277 for(i = 0; i < entries; i += entries_per_page)
278 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
281 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
283 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
286 for(i = 0; i < entries; i += entries_per_page)
287 set_aliased_prot(ldt + i, PAGE_KERNEL);
290 static void xen_set_ldt(const void *addr, unsigned entries)
292 struct mmuext_op *op;
293 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
296 op->cmd = MMUEXT_SET_LDT;
297 op->arg1.linear_addr = (unsigned long)addr;
298 op->arg2.nr_ents = entries;
300 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
302 xen_mc_issue(PARAVIRT_LAZY_CPU);
305 static void xen_load_gdt(const struct desc_ptr *dtr)
307 unsigned long *frames;
308 unsigned long va = dtr->address;
309 unsigned int size = dtr->size + 1;
310 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
312 struct multicall_space mcs;
314 /* A GDT can be up to 64k in size, which corresponds to 8192
315 8-byte entries, or 16 4k pages.. */
317 BUG_ON(size > 65536);
318 BUG_ON(va & ~PAGE_MASK);
320 mcs = xen_mc_entry(sizeof(*frames) * pages);
323 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
324 frames[f] = virt_to_mfn(va);
325 make_lowmem_page_readonly((void *)va);
328 MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
330 xen_mc_issue(PARAVIRT_LAZY_CPU);
333 static void load_TLS_descriptor(struct thread_struct *t,
334 unsigned int cpu, unsigned int i)
336 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
337 xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
338 struct multicall_space mc = __xen_mc_entry(0);
340 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
343 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
346 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
347 * it means we're in a context switch, and %gs has just been
348 * saved. This means we can zero it out to prevent faults on
349 * exit from the hypervisor if the next process has no %gs.
350 * Either way, it has been saved, and the new value will get
351 * loaded properly. This will go away as soon as Xen has been
352 * modified to not save/restore %gs for normal hypercalls.
354 * On x86_64, this hack is not used for %gs, because gs points
355 * to KERNEL_GS_BASE (and uses it for PDA references), so we
356 * must not zero %gs on x86_64
358 * For x86_64, we need to zero %fs, otherwise we may get an
359 * exception between the new %fs descriptor being loaded and
360 * %fs being effectively cleared at __switch_to().
362 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
372 load_TLS_descriptor(t, cpu, 0);
373 load_TLS_descriptor(t, cpu, 1);
374 load_TLS_descriptor(t, cpu, 2);
376 xen_mc_issue(PARAVIRT_LAZY_CPU);
380 static void xen_load_gs_index(unsigned int idx)
382 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
387 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
390 unsigned long lp = (unsigned long)&dt[entrynum];
391 xmaddr_t mach_lp = arbitrary_virt_to_machine(lp);
392 u64 entry = *(u64 *)ptr;
397 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
403 static int cvt_gate_to_trap(int vector, const gate_desc *val,
404 struct trap_info *info)
406 if (val->type != 0xf && val->type != 0xe)
409 info->vector = vector;
410 info->address = gate_offset(*val);
411 info->cs = gate_segment(*val);
412 info->flags = val->dpl;
413 /* interrupt gates clear IF */
414 if (val->type == 0xe)
420 /* Locations of each CPU's IDT */
421 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
423 /* Set an IDT entry. If the entry is part of the current IDT, then
425 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
427 unsigned long p = (unsigned long)&dt[entrynum];
428 unsigned long start, end;
432 start = __get_cpu_var(idt_desc).address;
433 end = start + __get_cpu_var(idt_desc).size + 1;
437 native_write_idt_entry(dt, entrynum, g);
439 if (p >= start && (p + 8) <= end) {
440 struct trap_info info[2];
444 if (cvt_gate_to_trap(entrynum, g, &info[0]))
445 if (HYPERVISOR_set_trap_table(info))
452 static void xen_convert_trap_info(const struct desc_ptr *desc,
453 struct trap_info *traps)
455 unsigned in, out, count;
457 count = (desc->size+1) / sizeof(gate_desc);
460 for (in = out = 0; in < count; in++) {
461 gate_desc *entry = (gate_desc*)(desc->address) + in;
463 if (cvt_gate_to_trap(in, entry, &traps[out]))
466 traps[out].address = 0;
469 void xen_copy_trap_info(struct trap_info *traps)
471 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
473 xen_convert_trap_info(desc, traps);
476 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
477 hold a spinlock to protect the static traps[] array (static because
478 it avoids allocation, and saves stack space). */
479 static void xen_load_idt(const struct desc_ptr *desc)
481 static DEFINE_SPINLOCK(lock);
482 static struct trap_info traps[257];
486 __get_cpu_var(idt_desc) = *desc;
488 xen_convert_trap_info(desc, traps);
491 if (HYPERVISOR_set_trap_table(traps))
497 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
498 they're handled differently. */
499 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
500 const void *desc, int type)
511 xmaddr_t maddr = virt_to_machine(&dt[entry]);
514 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
523 static void xen_load_sp0(struct tss_struct *tss,
524 struct thread_struct *thread)
526 struct multicall_space mcs = xen_mc_entry(0);
527 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
528 xen_mc_issue(PARAVIRT_LAZY_CPU);
531 static void xen_set_iopl_mask(unsigned mask)
533 struct physdev_set_iopl set_iopl;
535 /* Force the change at ring 0. */
536 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
537 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
540 static void xen_io_delay(void)
544 #ifdef CONFIG_X86_LOCAL_APIC
545 static u32 xen_apic_read(unsigned long reg)
550 static void xen_apic_write(unsigned long reg, u32 val)
552 /* Warn to see if there's any stray references */
557 static void xen_flush_tlb(void)
559 struct mmuext_op *op;
560 struct multicall_space mcs;
564 mcs = xen_mc_entry(sizeof(*op));
567 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
568 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
570 xen_mc_issue(PARAVIRT_LAZY_MMU);
575 static void xen_flush_tlb_single(unsigned long addr)
577 struct mmuext_op *op;
578 struct multicall_space mcs;
582 mcs = xen_mc_entry(sizeof(*op));
584 op->cmd = MMUEXT_INVLPG_LOCAL;
585 op->arg1.linear_addr = addr & PAGE_MASK;
586 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
588 xen_mc_issue(PARAVIRT_LAZY_MMU);
593 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
600 cpumask_t cpumask = *cpus;
601 struct multicall_space mcs;
604 * A couple of (to be removed) sanity checks:
606 * - current CPU must not be in mask
607 * - mask must exist :)
609 BUG_ON(cpus_empty(cpumask));
610 BUG_ON(cpu_isset(smp_processor_id(), cpumask));
613 /* If a CPU which we ran on has gone down, OK. */
614 cpus_and(cpumask, cpumask, cpu_online_map);
615 if (cpus_empty(cpumask))
618 mcs = xen_mc_entry(sizeof(*args));
620 args->mask = cpumask;
621 args->op.arg2.vcpumask = &args->mask;
623 if (va == TLB_FLUSH_ALL) {
624 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
626 args->op.cmd = MMUEXT_INVLPG_MULTI;
627 args->op.arg1.linear_addr = va;
630 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
632 xen_mc_issue(PARAVIRT_LAZY_MMU);
635 static void xen_clts(void)
637 struct multicall_space mcs;
639 mcs = xen_mc_entry(0);
641 MULTI_fpu_taskswitch(mcs.mc, 0);
643 xen_mc_issue(PARAVIRT_LAZY_CPU);
646 static void xen_write_cr0(unsigned long cr0)
648 struct multicall_space mcs;
650 /* Only pay attention to cr0.TS; everything else is
652 mcs = xen_mc_entry(0);
654 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
656 xen_mc_issue(PARAVIRT_LAZY_CPU);
659 static void xen_write_cr2(unsigned long cr2)
661 x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
664 static unsigned long xen_read_cr2(void)
666 return x86_read_percpu(xen_vcpu)->arch.cr2;
669 static unsigned long xen_read_cr2_direct(void)
671 return x86_read_percpu(xen_vcpu_info.arch.cr2);
674 static void xen_write_cr4(unsigned long cr4)
679 native_write_cr4(cr4);
682 static unsigned long xen_read_cr3(void)
684 return x86_read_percpu(xen_cr3);
687 static void set_current_cr3(void *v)
689 x86_write_percpu(xen_current_cr3, (unsigned long)v);
692 static void __xen_write_cr3(bool kernel, unsigned long cr3)
694 struct mmuext_op *op;
695 struct multicall_space mcs;
699 mfn = pfn_to_mfn(PFN_DOWN(cr3));
703 WARN_ON(mfn == 0 && kernel);
705 mcs = __xen_mc_entry(sizeof(*op));
708 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
711 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
714 x86_write_percpu(xen_cr3, cr3);
716 /* Update xen_current_cr3 once the batch has actually
718 xen_mc_callback(set_current_cr3, (void *)cr3);
722 static void xen_write_cr3(unsigned long cr3)
724 BUG_ON(preemptible());
726 xen_mc_batch(); /* disables interrupts */
728 /* Update while interrupts are disabled, so its atomic with
730 x86_write_percpu(xen_cr3, cr3);
732 __xen_write_cr3(true, cr3);
736 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
738 __xen_write_cr3(false, __pa(user_pgd));
740 __xen_write_cr3(false, 0);
744 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
747 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
758 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
759 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
760 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
763 base = ((u64)high << 32) | low;
764 if (HYPERVISOR_set_segment_base(which, base) != 0)
772 case MSR_SYSCALL_MASK:
773 case MSR_IA32_SYSENTER_CS:
774 case MSR_IA32_SYSENTER_ESP:
775 case MSR_IA32_SYSENTER_EIP:
776 /* Fast syscall setup is all done in hypercalls, so
777 these are all ignored. Stub them out here to stop
778 Xen console noise. */
782 ret = native_write_msr_safe(msr, low, high);
788 /* Early in boot, while setting up the initial pagetable, assume
789 everything is pinned. */
790 static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn)
792 #ifdef CONFIG_FLATMEM
793 BUG_ON(mem_map); /* should only be used early */
795 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
798 /* Early release_pte assumes that all pts are pinned, since there's
799 only init_mm and anything attached to that is pinned. */
800 static void xen_release_pte_init(u32 pfn)
802 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
805 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
809 op.arg1.mfn = pfn_to_mfn(pfn);
810 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
814 /* This needs to make sure the new pte page is pinned iff its being
815 attached to a pinned pagetable. */
816 static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
818 struct page *page = pfn_to_page(pfn);
820 if (PagePinned(virt_to_page(mm->pgd))) {
823 if (!PageHighMem(page)) {
824 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
826 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
828 /* make sure there are no stray mappings of
834 static void xen_alloc_pte(struct mm_struct *mm, u32 pfn)
836 xen_alloc_ptpage(mm, pfn, PT_PTE);
839 static void xen_alloc_pmd(struct mm_struct *mm, u32 pfn)
841 xen_alloc_ptpage(mm, pfn, PT_PMD);
844 static int xen_pgd_alloc(struct mm_struct *mm)
846 pgd_t *pgd = mm->pgd;
849 BUG_ON(PagePinned(virt_to_page(pgd)));
853 struct page *page = virt_to_page(pgd);
856 BUG_ON(page->private != 0);
860 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
861 page->private = (unsigned long)user_pgd;
863 if (user_pgd != NULL) {
864 user_pgd[pgd_index(VSYSCALL_START)] =
865 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
869 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
876 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
879 pgd_t *user_pgd = xen_get_user_pgd(pgd);
882 free_page((unsigned long)user_pgd);
886 /* This should never happen until we're OK to use struct page */
887 static void xen_release_ptpage(u32 pfn, unsigned level)
889 struct page *page = pfn_to_page(pfn);
891 if (PagePinned(page)) {
892 if (!PageHighMem(page)) {
894 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
895 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
897 ClearPagePinned(page);
901 static void xen_release_pte(u32 pfn)
903 xen_release_ptpage(pfn, PT_PTE);
906 static void xen_release_pmd(u32 pfn)
908 xen_release_ptpage(pfn, PT_PMD);
911 #if PAGETABLE_LEVELS == 4
912 static void xen_alloc_pud(struct mm_struct *mm, u32 pfn)
914 xen_alloc_ptpage(mm, pfn, PT_PUD);
917 static void xen_release_pud(u32 pfn)
919 xen_release_ptpage(pfn, PT_PUD);
923 #ifdef CONFIG_HIGHPTE
924 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
926 pgprot_t prot = PAGE_KERNEL;
928 if (PagePinned(page))
929 prot = PAGE_KERNEL_RO;
931 if (0 && PageHighMem(page))
932 printk("mapping highpte %lx type %d prot %s\n",
933 page_to_pfn(page), type,
934 (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
936 return kmap_atomic_prot(page, type, prot);
940 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
942 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
943 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
944 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
950 /* Init-time set_pte while constructing initial pagetables, which
951 doesn't allow RO pagetable pages to be remapped RW */
952 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
954 pte = mask_rw_pte(ptep, pte);
956 xen_set_pte(ptep, pte);
959 static __init void xen_pagetable_setup_start(pgd_t *base)
963 void xen_setup_shared_info(void)
965 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
966 set_fixmap(FIX_PARAVIRT_BOOTMAP,
967 xen_start_info->shared_info);
969 HYPERVISOR_shared_info =
970 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
972 HYPERVISOR_shared_info =
973 (struct shared_info *)__va(xen_start_info->shared_info);
976 /* In UP this is as good a place as any to set up shared info */
977 xen_setup_vcpu_info_placement();
980 xen_setup_mfn_list_list();
983 static __init void xen_pagetable_setup_done(pgd_t *base)
985 xen_setup_shared_info();
988 static __init void xen_post_allocator_init(void)
990 pv_mmu_ops.set_pte = xen_set_pte;
991 pv_mmu_ops.set_pmd = xen_set_pmd;
992 pv_mmu_ops.set_pud = xen_set_pud;
993 #if PAGETABLE_LEVELS == 4
994 pv_mmu_ops.set_pgd = xen_set_pgd;
997 /* This will work as long as patching hasn't happened yet
999 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1000 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1001 pv_mmu_ops.release_pte = xen_release_pte;
1002 pv_mmu_ops.release_pmd = xen_release_pmd;
1003 #if PAGETABLE_LEVELS == 4
1004 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1005 pv_mmu_ops.release_pud = xen_release_pud;
1008 #ifdef CONFIG_X86_64
1009 SetPagePinned(virt_to_page(level3_user_vsyscall));
1011 xen_mark_init_mm_pinned();
1014 /* This is called once we have the cpu_possible_map */
1015 void xen_setup_vcpu_info_placement(void)
1019 for_each_possible_cpu(cpu)
1020 xen_vcpu_setup(cpu);
1022 /* xen_vcpu_setup managed to place the vcpu_info within the
1023 percpu area for all cpus, so make use of it */
1024 #ifdef CONFIG_X86_32
1025 if (have_vcpu_info_placement) {
1026 printk(KERN_INFO "Xen: using vcpu_info placement\n");
1028 pv_irq_ops.save_fl = xen_save_fl_direct;
1029 pv_irq_ops.restore_fl = xen_restore_fl_direct;
1030 pv_irq_ops.irq_disable = xen_irq_disable_direct;
1031 pv_irq_ops.irq_enable = xen_irq_enable_direct;
1032 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1037 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1038 unsigned long addr, unsigned len)
1040 char *start, *end, *reloc;
1043 start = end = reloc = NULL;
1045 #define SITE(op, x) \
1046 case PARAVIRT_PATCH(op.x): \
1047 if (have_vcpu_info_placement) { \
1048 start = (char *)xen_##x##_direct; \
1049 end = xen_##x##_direct_end; \
1050 reloc = xen_##x##_direct_reloc; \
1055 #ifdef CONFIG_X86_32
1056 SITE(pv_irq_ops, irq_enable);
1057 SITE(pv_irq_ops, irq_disable);
1058 SITE(pv_irq_ops, save_fl);
1059 SITE(pv_irq_ops, restore_fl);
1060 #endif /* CONFIG_X86_32 */
1064 if (start == NULL || (end-start) > len)
1067 ret = paravirt_patch_insns(insnbuf, len, start, end);
1069 /* Note: because reloc is assigned from something that
1070 appears to be an array, gcc assumes it's non-null,
1071 but doesn't know its relationship with start and
1073 if (reloc > start && reloc < end) {
1074 int reloc_off = reloc - start;
1075 long *relocp = (long *)(insnbuf + reloc_off);
1076 long delta = start - (char *)addr;
1084 ret = paravirt_patch_default(type, clobbers, insnbuf,
1092 static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot)
1096 phys >>= PAGE_SHIFT;
1099 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1100 #ifdef CONFIG_X86_F00F_BUG
1103 #ifdef CONFIG_X86_32
1106 # ifdef CONFIG_HIGHMEM
1107 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1110 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1112 #ifdef CONFIG_X86_LOCAL_APIC
1113 case FIX_APIC_BASE: /* maps dummy local APIC */
1115 pte = pfn_pte(phys, prot);
1119 pte = mfn_pte(phys, prot);
1123 __native_set_fixmap(idx, pte);
1125 #ifdef CONFIG_X86_64
1126 /* Replicate changes to map the vsyscall page into the user
1127 pagetable vsyscall mapping. */
1128 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1129 unsigned long vaddr = __fix_to_virt(idx);
1130 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1135 static const struct pv_info xen_info __initdata = {
1136 .paravirt_enabled = 1,
1137 .shared_kernel_pmd = 0,
1142 static const struct pv_init_ops xen_init_ops __initdata = {
1145 .banner = xen_banner,
1146 .memory_setup = xen_memory_setup,
1147 .arch_setup = xen_arch_setup,
1148 .post_allocator_init = xen_post_allocator_init,
1151 static const struct pv_time_ops xen_time_ops __initdata = {
1152 .time_init = xen_time_init,
1154 .set_wallclock = xen_set_wallclock,
1155 .get_wallclock = xen_get_wallclock,
1156 .get_tsc_khz = xen_tsc_khz,
1157 .sched_clock = xen_sched_clock,
1160 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
1163 .set_debugreg = xen_set_debugreg,
1164 .get_debugreg = xen_get_debugreg,
1168 .read_cr0 = native_read_cr0,
1169 .write_cr0 = xen_write_cr0,
1171 .read_cr4 = native_read_cr4,
1172 .read_cr4_safe = native_read_cr4_safe,
1173 .write_cr4 = xen_write_cr4,
1175 .wbinvd = native_wbinvd,
1177 .read_msr = native_read_msr_safe,
1178 .write_msr = xen_write_msr_safe,
1179 .read_tsc = native_read_tsc,
1180 .read_pmc = native_read_pmc,
1183 .irq_enable_sysexit = xen_sysexit,
1184 #ifdef CONFIG_X86_64
1185 .usergs_sysret32 = xen_sysret32,
1186 .usergs_sysret64 = xen_sysret64,
1189 .load_tr_desc = paravirt_nop,
1190 .set_ldt = xen_set_ldt,
1191 .load_gdt = xen_load_gdt,
1192 .load_idt = xen_load_idt,
1193 .load_tls = xen_load_tls,
1194 #ifdef CONFIG_X86_64
1195 .load_gs_index = xen_load_gs_index,
1198 .alloc_ldt = xen_alloc_ldt,
1199 .free_ldt = xen_free_ldt,
1201 .store_gdt = native_store_gdt,
1202 .store_idt = native_store_idt,
1203 .store_tr = xen_store_tr,
1205 .write_ldt_entry = xen_write_ldt_entry,
1206 .write_gdt_entry = xen_write_gdt_entry,
1207 .write_idt_entry = xen_write_idt_entry,
1208 .load_sp0 = xen_load_sp0,
1210 .set_iopl_mask = xen_set_iopl_mask,
1211 .io_delay = xen_io_delay,
1213 /* Xen takes care of %gs when switching to usermode for us */
1214 .swapgs = paravirt_nop,
1217 .enter = paravirt_enter_lazy_cpu,
1218 .leave = xen_leave_lazy,
1222 static const struct pv_apic_ops xen_apic_ops __initdata = {
1223 #ifdef CONFIG_X86_LOCAL_APIC
1224 .apic_write = xen_apic_write,
1225 .apic_read = xen_apic_read,
1226 .setup_boot_clock = paravirt_nop,
1227 .setup_secondary_clock = paravirt_nop,
1228 .startup_ipi_hook = paravirt_nop,
1232 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1233 .pagetable_setup_start = xen_pagetable_setup_start,
1234 .pagetable_setup_done = xen_pagetable_setup_done,
1236 .read_cr2 = xen_read_cr2,
1237 .write_cr2 = xen_write_cr2,
1239 .read_cr3 = xen_read_cr3,
1240 .write_cr3 = xen_write_cr3,
1242 .flush_tlb_user = xen_flush_tlb,
1243 .flush_tlb_kernel = xen_flush_tlb,
1244 .flush_tlb_single = xen_flush_tlb_single,
1245 .flush_tlb_others = xen_flush_tlb_others,
1247 .pte_update = paravirt_nop,
1248 .pte_update_defer = paravirt_nop,
1250 .pgd_alloc = xen_pgd_alloc,
1251 .pgd_free = xen_pgd_free,
1253 .alloc_pte = xen_alloc_pte_init,
1254 .release_pte = xen_release_pte_init,
1255 .alloc_pmd = xen_alloc_pte_init,
1256 .alloc_pmd_clone = paravirt_nop,
1257 .release_pmd = xen_release_pte_init,
1259 #ifdef CONFIG_HIGHPTE
1260 .kmap_atomic_pte = xen_kmap_atomic_pte,
1263 #ifdef CONFIG_X86_64
1264 .set_pte = xen_set_pte,
1266 .set_pte = xen_set_pte_init,
1268 .set_pte_at = xen_set_pte_at,
1269 .set_pmd = xen_set_pmd_hyper,
1271 .ptep_modify_prot_start = __ptep_modify_prot_start,
1272 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1274 .pte_val = xen_pte_val,
1275 .pte_flags = native_pte_flags,
1276 .pgd_val = xen_pgd_val,
1278 .make_pte = xen_make_pte,
1279 .make_pgd = xen_make_pgd,
1281 #ifdef CONFIG_X86_PAE
1282 .set_pte_atomic = xen_set_pte_atomic,
1283 .set_pte_present = xen_set_pte_at,
1284 .pte_clear = xen_pte_clear,
1285 .pmd_clear = xen_pmd_clear,
1286 #endif /* CONFIG_X86_PAE */
1287 .set_pud = xen_set_pud_hyper,
1289 .make_pmd = xen_make_pmd,
1290 .pmd_val = xen_pmd_val,
1292 #if PAGETABLE_LEVELS == 4
1293 .pud_val = xen_pud_val,
1294 .make_pud = xen_make_pud,
1295 .set_pgd = xen_set_pgd_hyper,
1297 .alloc_pud = xen_alloc_pte_init,
1298 .release_pud = xen_release_pte_init,
1299 #endif /* PAGETABLE_LEVELS == 4 */
1301 .activate_mm = xen_activate_mm,
1302 .dup_mmap = xen_dup_mmap,
1303 .exit_mmap = xen_exit_mmap,
1306 .enter = paravirt_enter_lazy_mmu,
1307 .leave = xen_leave_lazy,
1310 .set_fixmap = xen_set_fixmap,
1313 static void xen_reboot(int reason)
1315 struct sched_shutdown r = { .reason = reason };
1321 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1325 static void xen_restart(char *msg)
1327 xen_reboot(SHUTDOWN_reboot);
1330 static void xen_emergency_restart(void)
1332 xen_reboot(SHUTDOWN_reboot);
1335 static void xen_machine_halt(void)
1337 xen_reboot(SHUTDOWN_poweroff);
1340 static void xen_crash_shutdown(struct pt_regs *regs)
1342 xen_reboot(SHUTDOWN_crash);
1345 static const struct machine_ops __initdata xen_machine_ops = {
1346 .restart = xen_restart,
1347 .halt = xen_machine_halt,
1348 .power_off = xen_machine_halt,
1349 .shutdown = xen_machine_halt,
1350 .crash_shutdown = xen_crash_shutdown,
1351 .emergency_restart = xen_emergency_restart,
1355 static void __init xen_reserve_top(void)
1357 #ifdef CONFIG_X86_32
1358 unsigned long top = HYPERVISOR_VIRT_START;
1359 struct xen_platform_parameters pp;
1361 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1362 top = pp.virt_start;
1364 reserve_top_address(-top + 2 * PAGE_SIZE);
1365 #endif /* CONFIG_X86_32 */
1369 * Like __va(), but returns address in the kernel mapping (which is
1370 * all we have until the physical memory mapping has been set up.
1372 static void *__ka(phys_addr_t paddr)
1374 #ifdef CONFIG_X86_64
1375 return (void *)(paddr + __START_KERNEL_map);
1381 /* Convert a machine address to physical address */
1382 static unsigned long m2p(phys_addr_t maddr)
1386 maddr &= PTE_PFN_MASK;
1387 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1392 /* Convert a machine address to kernel virtual */
1393 static void *m2v(phys_addr_t maddr)
1395 return __ka(m2p(maddr));
1398 #ifdef CONFIG_X86_64
1399 static void walk(pgd_t *pgd, unsigned long addr)
1401 unsigned l4idx = pgd_index(addr);
1402 unsigned l3idx = pud_index(addr);
1403 unsigned l2idx = pmd_index(addr);
1404 unsigned l1idx = pte_index(addr);
1410 xen_raw_printk("walk %p, %lx -> %d %d %d %d\n",
1411 pgd, addr, l4idx, l3idx, l2idx, l1idx);
1414 xen_raw_printk(" l4: %016lx\n", l4.pgd);
1415 xen_raw_printk(" %016lx\n", pgd_val(l4));
1417 l3 = ((pud_t *)(m2v(l4.pgd)))[l3idx];
1418 xen_raw_printk(" l3: %016lx\n", l3.pud);
1419 xen_raw_printk(" %016lx\n", pud_val(l3));
1421 l2 = ((pmd_t *)(m2v(l3.pud)))[l2idx];
1422 xen_raw_printk(" l2: %016lx\n", l2.pmd);
1423 xen_raw_printk(" %016lx\n", pmd_val(l2));
1425 l1 = ((pte_t *)(m2v(l2.pmd)))[l1idx];
1426 xen_raw_printk(" l1: %016lx\n", l1.pte);
1427 xen_raw_printk(" %016lx\n", pte_val(l1));
1431 static void set_page_prot(void *addr, pgprot_t prot)
1433 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1434 pte_t pte = pfn_pte(pfn, prot);
1436 xen_raw_printk("addr=%p pfn=%lx mfn=%lx prot=%016llx pte=%016llx\n",
1437 addr, pfn, get_phys_to_machine(pfn),
1438 pgprot_val(prot), pte.pte);
1440 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1444 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1446 unsigned pmdidx, pteidx;
1452 for(pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1455 /* Reuse or allocate a page of ptes */
1456 if (pmd_present(pmd[pmdidx]))
1457 pte_page = m2v(pmd[pmdidx].pmd);
1459 /* Check for free pte pages */
1460 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1463 pte_page = &level1_ident_pgt[ident_pte];
1464 ident_pte += PTRS_PER_PTE;
1466 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1469 /* Install mappings */
1470 for(pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1473 if (pfn > max_pfn_mapped)
1474 max_pfn_mapped = pfn;
1476 if (!pte_none(pte_page[pteidx]))
1479 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1480 pte_page[pteidx] = pte;
1484 for(pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1485 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1487 set_page_prot(pmd, PAGE_KERNEL_RO);
1490 #ifdef CONFIG_X86_64
1491 static void convert_pfn_mfn(void *v)
1496 /* All levels are converted the same way, so just treat them
1498 for(i = 0; i < PTRS_PER_PTE; i++)
1499 pte[i] = xen_make_pte(pte[i].pte);
1503 * Set up the inital kernel pagetable.
1505 * We can construct this by grafting the Xen provided pagetable into
1506 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1507 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1508 * means that only the kernel has a physical mapping to start with -
1509 * but that's enough to get __va working. We need to fill in the rest
1510 * of the physical mapping once some sort of allocator has been set
1513 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1518 /* Zap identity mapping */
1519 init_level4_pgt[0] = __pgd(0);
1521 /* Pre-constructed entries are in pfn, so convert to mfn */
1522 convert_pfn_mfn(init_level4_pgt);
1523 convert_pfn_mfn(level3_ident_pgt);
1524 convert_pfn_mfn(level3_kernel_pgt);
1526 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1527 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1529 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1530 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1532 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1533 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1534 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1536 /* Set up identity map */
1537 xen_map_identity_early(level2_ident_pgt, max_pfn);
1539 /* Make pagetable pieces RO */
1540 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1541 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1542 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1543 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1544 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1545 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1547 /* Pin down new L4 */
1548 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1549 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1551 /* Unpin Xen-provided one */
1552 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1555 pgd = init_level4_pgt;
1558 * At this stage there can be no user pgd, and no page
1559 * structure to attach it to, so make sure we just set kernel
1563 __xen_write_cr3(true, __pa(pgd));
1564 xen_mc_issue(PARAVIRT_LAZY_CPU);
1566 reserve_early(__pa(xen_start_info->pt_base),
1567 __pa(xen_start_info->pt_base +
1568 xen_start_info->nr_pt_frames * PAGE_SIZE),
1573 #else /* !CONFIG_X86_64 */
1574 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1576 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1580 init_pg_tables_start = __pa(pgd);
1581 init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1582 max_pfn_mapped = PFN_DOWN(init_pg_tables_end + 512*1024);
1584 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1585 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1587 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1589 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1590 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1591 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1593 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1594 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1595 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1597 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1599 xen_write_cr3(__pa(swapper_pg_dir));
1601 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1603 return swapper_pg_dir;
1605 #endif /* CONFIG_X86_64 */
1607 /* First C function to be called on Xen boot */
1608 asmlinkage void __init xen_start_kernel(void)
1612 if (!xen_start_info)
1615 BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
1617 xen_setup_features();
1619 /* Install Xen paravirt ops */
1621 pv_init_ops = xen_init_ops;
1622 pv_time_ops = xen_time_ops;
1623 pv_cpu_ops = xen_cpu_ops;
1624 pv_apic_ops = xen_apic_ops;
1625 pv_mmu_ops = xen_mmu_ops;
1629 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1630 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1631 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1634 machine_ops = xen_machine_ops;
1636 #ifdef CONFIG_X86_64
1637 /* Disable until direct per-cpu data access. */
1638 have_vcpu_info_placement = 0;
1645 if (!xen_feature(XENFEAT_auto_translated_physmap))
1646 xen_build_dynamic_phys_to_machine();
1648 pgd = (pgd_t *)xen_start_info->pt_base;
1650 /* Prevent unwanted bits from being set in PTEs. */
1651 __supported_pte_mask &= ~_PAGE_GLOBAL;
1652 if (!is_initial_xendomain())
1653 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1655 /* Don't do the full vcpu_info placement stuff until we have a
1656 possible map and a non-dummy shared_info. */
1657 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1659 xen_raw_console_write("mapping kernel into physical memory\n");
1660 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1664 /* keep using Xen gdt for now; no urgent need to change it */
1666 pv_info.kernel_rpl = 1;
1667 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1668 pv_info.kernel_rpl = 0;
1670 /* set the limit of our address space */
1673 #ifdef CONFIG_X86_32
1674 /* set up basic CPUID stuff */
1675 cpu_detect(&new_cpu_data);
1676 new_cpu_data.hard_math = 1;
1677 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1680 /* Poke various useful things into boot_params */
1681 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1682 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1683 ? __pa(xen_start_info->mod_start) : 0;
1684 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1685 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1687 if (!is_initial_xendomain()) {
1688 add_preferred_console("xenboot", 0, NULL);
1689 add_preferred_console("tty", 0, NULL);
1690 add_preferred_console("hvc", 0, NULL);
1693 xen_raw_console_write("about to get started...\n");
1696 xen_raw_printk("&boot_params=%p __pa(&boot_params)=%lx __va(__pa(&boot_params))=%lx\n",
1697 &boot_params, __pa_symbol(&boot_params),
1698 __va(__pa_symbol(&boot_params)));
1700 walk(pgd, &boot_params);
1701 walk(pgd, __va(__pa(&boot_params)));
1704 /* Start the world */
1705 #ifdef CONFIG_X86_32
1706 i386_start_kernel();
1708 x86_64_start_reservations((char *)__pa_symbol(&boot_params));