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/interface/sched.h>
34 #include <xen/features.h>
36 #include <xen/hvc-console.h>
38 #include <asm/paravirt.h>
40 #include <asm/xen/hypercall.h>
41 #include <asm/xen/hypervisor.h>
42 #include <asm/fixmap.h>
43 #include <asm/processor.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82 * out of date, but it will never be set early).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 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
184 printk(KERN_INFO "Hypervisor signature: %s%s\n",
185 xen_start_info->magic,
186 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
189 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
190 unsigned int *cx, unsigned int *dx)
192 unsigned maskedx = ~0;
195 * Mask out inconvenient features, to try and disable as many
196 * unsupported kernel subsystems as possible.
199 maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */
200 (1 << X86_FEATURE_ACPI) | /* disable ACPI */
201 (1 << X86_FEATURE_MCE) | /* disable MCE */
202 (1 << X86_FEATURE_MCA) | /* disable MCA */
203 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
205 asm(XEN_EMULATE_PREFIX "cpuid"
210 : "0" (*ax), "2" (*cx));
214 static void xen_set_debugreg(int reg, unsigned long val)
216 HYPERVISOR_set_debugreg(reg, val);
219 static unsigned long xen_get_debugreg(int reg)
221 return HYPERVISOR_get_debugreg(reg);
224 static unsigned long xen_save_fl(void)
226 struct vcpu_info *vcpu;
229 vcpu = x86_read_percpu(xen_vcpu);
231 /* flag has opposite sense of mask */
232 flags = !vcpu->evtchn_upcall_mask;
234 /* convert to IF type flag
238 return (-flags) & X86_EFLAGS_IF;
241 static void xen_restore_fl(unsigned long flags)
243 struct vcpu_info *vcpu;
245 /* convert from IF type flag */
246 flags = !(flags & X86_EFLAGS_IF);
248 /* There's a one instruction preempt window here. We need to
249 make sure we're don't switch CPUs between getting the vcpu
250 pointer and updating the mask. */
252 vcpu = x86_read_percpu(xen_vcpu);
253 vcpu->evtchn_upcall_mask = flags;
254 preempt_enable_no_resched();
256 /* Doesn't matter if we get preempted here, because any
257 pending event will get dealt with anyway. */
260 preempt_check_resched();
261 barrier(); /* unmask then check (avoid races) */
262 if (unlikely(vcpu->evtchn_upcall_pending))
263 force_evtchn_callback();
267 static void xen_irq_disable(void)
269 /* There's a one instruction preempt window here. We need to
270 make sure we're don't switch CPUs between getting the vcpu
271 pointer and updating the mask. */
273 x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1;
274 preempt_enable_no_resched();
277 static void xen_irq_enable(void)
279 struct vcpu_info *vcpu;
281 /* We don't need to worry about being preempted here, since
282 either a) interrupts are disabled, so no preemption, or b)
283 the caller is confused and is trying to re-enable interrupts
284 on an indeterminate processor. */
286 vcpu = x86_read_percpu(xen_vcpu);
287 vcpu->evtchn_upcall_mask = 0;
289 /* Doesn't matter if we get preempted here, because any
290 pending event will get dealt with anyway. */
292 barrier(); /* unmask then check (avoid races) */
293 if (unlikely(vcpu->evtchn_upcall_pending))
294 force_evtchn_callback();
297 static void xen_safe_halt(void)
299 /* Blocking includes an implicit local_irq_enable(). */
300 if (HYPERVISOR_sched_op(SCHEDOP_block, NULL) != 0)
304 static void xen_halt(void)
307 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
312 static void xen_leave_lazy(void)
314 paravirt_leave_lazy(paravirt_get_lazy_mode());
318 static unsigned long xen_store_tr(void)
323 static void xen_set_ldt(const void *addr, unsigned entries)
325 struct mmuext_op *op;
326 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
329 op->cmd = MMUEXT_SET_LDT;
330 op->arg1.linear_addr = (unsigned long)addr;
331 op->arg2.nr_ents = entries;
333 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
335 xen_mc_issue(PARAVIRT_LAZY_CPU);
338 static void xen_load_gdt(const struct desc_ptr *dtr)
340 unsigned long *frames;
341 unsigned long va = dtr->address;
342 unsigned int size = dtr->size + 1;
343 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
345 struct multicall_space mcs;
347 /* A GDT can be up to 64k in size, which corresponds to 8192
348 8-byte entries, or 16 4k pages.. */
350 BUG_ON(size > 65536);
351 BUG_ON(va & ~PAGE_MASK);
353 mcs = xen_mc_entry(sizeof(*frames) * pages);
356 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
357 frames[f] = virt_to_mfn(va);
358 make_lowmem_page_readonly((void *)va);
361 MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
363 xen_mc_issue(PARAVIRT_LAZY_CPU);
366 static void load_TLS_descriptor(struct thread_struct *t,
367 unsigned int cpu, unsigned int i)
369 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
370 xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
371 struct multicall_space mc = __xen_mc_entry(0);
373 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
376 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
379 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
380 * it means we're in a context switch, and %gs has just been
381 * saved. This means we can zero it out to prevent faults on
382 * exit from the hypervisor if the next process has no %gs.
383 * Either way, it has been saved, and the new value will get
384 * loaded properly. This will go away as soon as Xen has been
385 * modified to not save/restore %gs for normal hypercalls.
387 * On x86_64, this hack is not used for %gs, because gs points
388 * to KERNEL_GS_BASE (and uses it for PDA references), so we
389 * must not zero %gs on x86_64
391 * For x86_64, we need to zero %fs, otherwise we may get an
392 * exception between the new %fs descriptor being loaded and
393 * %fs being effectively cleared at __switch_to().
395 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
405 load_TLS_descriptor(t, cpu, 0);
406 load_TLS_descriptor(t, cpu, 1);
407 load_TLS_descriptor(t, cpu, 2);
409 xen_mc_issue(PARAVIRT_LAZY_CPU);
413 static void xen_load_gs_index(unsigned int idx)
415 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
420 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
423 unsigned long lp = (unsigned long)&dt[entrynum];
424 xmaddr_t mach_lp = virt_to_machine(lp);
425 u64 entry = *(u64 *)ptr;
430 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
436 static int cvt_gate_to_trap(int vector, const gate_desc *val,
437 struct trap_info *info)
439 if (val->type != 0xf && val->type != 0xe)
442 info->vector = vector;
443 info->address = gate_offset(*val);
444 info->cs = gate_segment(*val);
445 info->flags = val->dpl;
446 /* interrupt gates clear IF */
447 if (val->type == 0xe)
453 /* Locations of each CPU's IDT */
454 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
456 /* Set an IDT entry. If the entry is part of the current IDT, then
458 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
460 unsigned long p = (unsigned long)&dt[entrynum];
461 unsigned long start, end;
465 start = __get_cpu_var(idt_desc).address;
466 end = start + __get_cpu_var(idt_desc).size + 1;
470 native_write_idt_entry(dt, entrynum, g);
472 if (p >= start && (p + 8) <= end) {
473 struct trap_info info[2];
477 if (cvt_gate_to_trap(entrynum, g, &info[0]))
478 if (HYPERVISOR_set_trap_table(info))
485 static void xen_convert_trap_info(const struct desc_ptr *desc,
486 struct trap_info *traps)
488 unsigned in, out, count;
490 count = (desc->size+1) / sizeof(gate_desc);
493 for (in = out = 0; in < count; in++) {
494 gate_desc *entry = (gate_desc*)(desc->address) + in;
496 if (cvt_gate_to_trap(in, entry, &traps[out]))
499 traps[out].address = 0;
502 void xen_copy_trap_info(struct trap_info *traps)
504 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
506 xen_convert_trap_info(desc, traps);
509 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
510 hold a spinlock to protect the static traps[] array (static because
511 it avoids allocation, and saves stack space). */
512 static void xen_load_idt(const struct desc_ptr *desc)
514 static DEFINE_SPINLOCK(lock);
515 static struct trap_info traps[257];
519 __get_cpu_var(idt_desc) = *desc;
521 xen_convert_trap_info(desc, traps);
524 if (HYPERVISOR_set_trap_table(traps))
530 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
531 they're handled differently. */
532 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
533 const void *desc, int type)
544 xmaddr_t maddr = virt_to_machine(&dt[entry]);
547 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
556 static void xen_load_sp0(struct tss_struct *tss,
557 struct thread_struct *thread)
559 struct multicall_space mcs = xen_mc_entry(0);
560 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
561 xen_mc_issue(PARAVIRT_LAZY_CPU);
564 static void xen_set_iopl_mask(unsigned mask)
566 struct physdev_set_iopl set_iopl;
568 /* Force the change at ring 0. */
569 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
570 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
573 static void xen_io_delay(void)
577 #ifdef CONFIG_X86_LOCAL_APIC
578 static u32 xen_apic_read(unsigned long reg)
583 static void xen_apic_write(unsigned long reg, u32 val)
585 /* Warn to see if there's any stray references */
590 static void xen_flush_tlb(void)
592 struct mmuext_op *op;
593 struct multicall_space mcs;
597 mcs = xen_mc_entry(sizeof(*op));
600 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
601 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
603 xen_mc_issue(PARAVIRT_LAZY_MMU);
608 static void xen_flush_tlb_single(unsigned long addr)
610 struct mmuext_op *op;
611 struct multicall_space mcs;
615 mcs = xen_mc_entry(sizeof(*op));
617 op->cmd = MMUEXT_INVLPG_LOCAL;
618 op->arg1.linear_addr = addr & PAGE_MASK;
619 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
621 xen_mc_issue(PARAVIRT_LAZY_MMU);
626 static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
633 cpumask_t cpumask = *cpus;
634 struct multicall_space mcs;
637 * A couple of (to be removed) sanity checks:
639 * - current CPU must not be in mask
640 * - mask must exist :)
642 BUG_ON(cpus_empty(cpumask));
643 BUG_ON(cpu_isset(smp_processor_id(), cpumask));
646 /* If a CPU which we ran on has gone down, OK. */
647 cpus_and(cpumask, cpumask, cpu_online_map);
648 if (cpus_empty(cpumask))
651 mcs = xen_mc_entry(sizeof(*args));
653 args->mask = cpumask;
654 args->op.arg2.vcpumask = &args->mask;
656 if (va == TLB_FLUSH_ALL) {
657 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
659 args->op.cmd = MMUEXT_INVLPG_MULTI;
660 args->op.arg1.linear_addr = va;
663 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
665 xen_mc_issue(PARAVIRT_LAZY_MMU);
668 static void xen_clts(void)
670 struct multicall_space mcs;
672 mcs = xen_mc_entry(0);
674 MULTI_fpu_taskswitch(mcs.mc, 0);
676 xen_mc_issue(PARAVIRT_LAZY_CPU);
679 static void xen_write_cr0(unsigned long cr0)
681 struct multicall_space mcs;
683 /* Only pay attention to cr0.TS; everything else is
685 mcs = xen_mc_entry(0);
687 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
689 xen_mc_issue(PARAVIRT_LAZY_CPU);
692 static void xen_write_cr2(unsigned long cr2)
694 x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
697 static unsigned long xen_read_cr2(void)
699 return x86_read_percpu(xen_vcpu)->arch.cr2;
702 static unsigned long xen_read_cr2_direct(void)
704 return x86_read_percpu(xen_vcpu_info.arch.cr2);
707 static void xen_write_cr4(unsigned long cr4)
712 native_write_cr4(cr4);
715 static unsigned long xen_read_cr3(void)
717 return x86_read_percpu(xen_cr3);
720 static void set_current_cr3(void *v)
722 x86_write_percpu(xen_current_cr3, (unsigned long)v);
725 static void __xen_write_cr3(bool kernel, unsigned long cr3)
727 struct mmuext_op *op;
728 struct multicall_space mcs;
732 mfn = pfn_to_mfn(PFN_DOWN(cr3));
736 WARN_ON(mfn == 0 && kernel);
738 mcs = __xen_mc_entry(sizeof(*op));
741 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
744 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
747 x86_write_percpu(xen_cr3, cr3);
749 /* Update xen_current_cr3 once the batch has actually
751 xen_mc_callback(set_current_cr3, (void *)cr3);
755 static void xen_write_cr3(unsigned long cr3)
757 BUG_ON(preemptible());
759 xen_mc_batch(); /* disables interrupts */
761 /* Update while interrupts are disabled, so its atomic with
763 x86_write_percpu(xen_cr3, cr3);
765 __xen_write_cr3(true, cr3);
769 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
771 __xen_write_cr3(false, __pa(user_pgd));
773 __xen_write_cr3(false, 0);
777 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
780 /* Early in boot, while setting up the initial pagetable, assume
781 everything is pinned. */
782 static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn)
784 #ifdef CONFIG_FLATMEM
785 BUG_ON(mem_map); /* should only be used early */
787 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
790 /* Early release_pte assumes that all pts are pinned, since there's
791 only init_mm and anything attached to that is pinned. */
792 static void xen_release_pte_init(u32 pfn)
794 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
797 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
801 op.arg1.mfn = pfn_to_mfn(pfn);
802 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
806 /* This needs to make sure the new pte page is pinned iff its being
807 attached to a pinned pagetable. */
808 static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
810 struct page *page = pfn_to_page(pfn);
812 if (PagePinned(virt_to_page(mm->pgd))) {
815 if (!PageHighMem(page)) {
816 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
818 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
820 /* make sure there are no stray mappings of
826 static void xen_alloc_pte(struct mm_struct *mm, u32 pfn)
828 xen_alloc_ptpage(mm, pfn, PT_PTE);
831 static void xen_alloc_pmd(struct mm_struct *mm, u32 pfn)
833 xen_alloc_ptpage(mm, pfn, PT_PMD);
836 static int xen_pgd_alloc(struct mm_struct *mm)
838 pgd_t *pgd = mm->pgd;
841 BUG_ON(PagePinned(virt_to_page(pgd)));
845 struct page *page = virt_to_page(pgd);
848 BUG_ON(page->private != 0);
852 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
853 page->private = (unsigned long)user_pgd;
855 if (user_pgd != NULL) {
856 user_pgd[pgd_index(VSYSCALL_START)] =
857 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
861 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
868 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
871 pgd_t *user_pgd = xen_get_user_pgd(pgd);
874 free_page((unsigned long)user_pgd);
878 /* This should never happen until we're OK to use struct page */
879 static void xen_release_ptpage(u32 pfn, unsigned level)
881 struct page *page = pfn_to_page(pfn);
883 if (PagePinned(page)) {
884 if (!PageHighMem(page)) {
886 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
887 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
889 ClearPagePinned(page);
893 static void xen_release_pte(u32 pfn)
895 xen_release_ptpage(pfn, PT_PTE);
898 static void xen_release_pmd(u32 pfn)
900 xen_release_ptpage(pfn, PT_PMD);
903 #if PAGETABLE_LEVELS == 4
904 static void xen_alloc_pud(struct mm_struct *mm, u32 pfn)
906 xen_alloc_ptpage(mm, pfn, PT_PUD);
909 static void xen_release_pud(u32 pfn)
911 xen_release_ptpage(pfn, PT_PUD);
915 #ifdef CONFIG_HIGHPTE
916 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
918 pgprot_t prot = PAGE_KERNEL;
920 if (PagePinned(page))
921 prot = PAGE_KERNEL_RO;
923 if (0 && PageHighMem(page))
924 printk("mapping highpte %lx type %d prot %s\n",
925 page_to_pfn(page), type,
926 (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
928 return kmap_atomic_prot(page, type, prot);
932 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
934 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
935 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
936 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
942 /* Init-time set_pte while constructing initial pagetables, which
943 doesn't allow RO pagetable pages to be remapped RW */
944 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
946 pte = mask_rw_pte(ptep, pte);
948 xen_set_pte(ptep, pte);
951 static __init void xen_pagetable_setup_start(pgd_t *base)
955 void xen_setup_shared_info(void)
957 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
958 set_fixmap(FIX_PARAVIRT_BOOTMAP,
959 xen_start_info->shared_info);
961 HYPERVISOR_shared_info =
962 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
964 HYPERVISOR_shared_info =
965 (struct shared_info *)__va(xen_start_info->shared_info);
968 /* In UP this is as good a place as any to set up shared info */
969 xen_setup_vcpu_info_placement();
972 xen_setup_mfn_list_list();
975 static __init void xen_pagetable_setup_done(pgd_t *base)
977 xen_setup_shared_info();
980 static __init void xen_post_allocator_init(void)
982 pv_mmu_ops.set_pte = xen_set_pte;
983 pv_mmu_ops.set_pmd = xen_set_pmd;
984 pv_mmu_ops.set_pud = xen_set_pud;
985 #if PAGETABLE_LEVELS == 4
986 pv_mmu_ops.set_pgd = xen_set_pgd;
989 /* This will work as long as patching hasn't happened yet
991 pv_mmu_ops.alloc_pte = xen_alloc_pte;
992 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
993 pv_mmu_ops.release_pte = xen_release_pte;
994 pv_mmu_ops.release_pmd = xen_release_pmd;
995 #if PAGETABLE_LEVELS == 4
996 pv_mmu_ops.alloc_pud = xen_alloc_pud;
997 pv_mmu_ops.release_pud = xen_release_pud;
1000 #ifdef CONFIG_X86_64
1001 SetPagePinned(virt_to_page(level3_user_vsyscall));
1003 xen_mark_init_mm_pinned();
1006 /* This is called once we have the cpu_possible_map */
1007 void xen_setup_vcpu_info_placement(void)
1011 for_each_possible_cpu(cpu)
1012 xen_vcpu_setup(cpu);
1014 /* xen_vcpu_setup managed to place the vcpu_info within the
1015 percpu area for all cpus, so make use of it */
1016 #ifdef CONFIG_X86_32
1017 if (have_vcpu_info_placement) {
1018 printk(KERN_INFO "Xen: using vcpu_info placement\n");
1020 pv_irq_ops.save_fl = xen_save_fl_direct;
1021 pv_irq_ops.restore_fl = xen_restore_fl_direct;
1022 pv_irq_ops.irq_disable = xen_irq_disable_direct;
1023 pv_irq_ops.irq_enable = xen_irq_enable_direct;
1024 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1029 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1030 unsigned long addr, unsigned len)
1032 char *start, *end, *reloc;
1035 start = end = reloc = NULL;
1037 #define SITE(op, x) \
1038 case PARAVIRT_PATCH(op.x): \
1039 if (have_vcpu_info_placement) { \
1040 start = (char *)xen_##x##_direct; \
1041 end = xen_##x##_direct_end; \
1042 reloc = xen_##x##_direct_reloc; \
1047 #ifdef CONFIG_X86_32
1048 SITE(pv_irq_ops, irq_enable);
1049 SITE(pv_irq_ops, irq_disable);
1050 SITE(pv_irq_ops, save_fl);
1051 SITE(pv_irq_ops, restore_fl);
1052 #endif /* CONFIG_X86_32 */
1056 if (start == NULL || (end-start) > len)
1059 ret = paravirt_patch_insns(insnbuf, len, start, end);
1061 /* Note: because reloc is assigned from something that
1062 appears to be an array, gcc assumes it's non-null,
1063 but doesn't know its relationship with start and
1065 if (reloc > start && reloc < end) {
1066 int reloc_off = reloc - start;
1067 long *relocp = (long *)(insnbuf + reloc_off);
1068 long delta = start - (char *)addr;
1076 ret = paravirt_patch_default(type, clobbers, insnbuf,
1084 static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot)
1088 phys >>= PAGE_SHIFT;
1091 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1092 #ifdef CONFIG_X86_F00F_BUG
1095 #ifdef CONFIG_X86_32
1098 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1100 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1102 #ifdef CONFIG_X86_LOCAL_APIC
1103 case FIX_APIC_BASE: /* maps dummy local APIC */
1105 pte = pfn_pte(phys, prot);
1109 pte = mfn_pte(phys, prot);
1113 __native_set_fixmap(idx, pte);
1115 #ifdef CONFIG_X86_64
1116 /* Replicate changes to map the vsyscall page into the user
1117 pagetable vsyscall mapping. */
1118 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1119 unsigned long vaddr = __fix_to_virt(idx);
1120 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1125 static const struct pv_info xen_info __initdata = {
1126 .paravirt_enabled = 1,
1127 .shared_kernel_pmd = 0,
1132 static const struct pv_init_ops xen_init_ops __initdata = {
1135 .banner = xen_banner,
1136 .memory_setup = xen_memory_setup,
1137 .arch_setup = xen_arch_setup,
1138 .post_allocator_init = xen_post_allocator_init,
1141 static const struct pv_time_ops xen_time_ops __initdata = {
1142 .time_init = xen_time_init,
1144 .set_wallclock = xen_set_wallclock,
1145 .get_wallclock = xen_get_wallclock,
1146 .get_tsc_khz = xen_tsc_khz,
1147 .sched_clock = xen_sched_clock,
1150 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
1153 .set_debugreg = xen_set_debugreg,
1154 .get_debugreg = xen_get_debugreg,
1158 .read_cr0 = native_read_cr0,
1159 .write_cr0 = xen_write_cr0,
1161 .read_cr4 = native_read_cr4,
1162 .read_cr4_safe = native_read_cr4_safe,
1163 .write_cr4 = xen_write_cr4,
1165 .wbinvd = native_wbinvd,
1167 .read_msr = native_read_msr_safe,
1168 .write_msr = native_write_msr_safe,
1169 .read_tsc = native_read_tsc,
1170 .read_pmc = native_read_pmc,
1173 .irq_enable_sysexit = xen_sysexit,
1174 #ifdef CONFIG_X86_64
1175 .usergs_sysret32 = xen_sysret32,
1176 .usergs_sysret64 = xen_sysret64,
1179 .load_tr_desc = paravirt_nop,
1180 .set_ldt = xen_set_ldt,
1181 .load_gdt = xen_load_gdt,
1182 .load_idt = xen_load_idt,
1183 .load_tls = xen_load_tls,
1184 #ifdef CONFIG_X86_64
1185 .load_gs_index = xen_load_gs_index,
1188 .store_gdt = native_store_gdt,
1189 .store_idt = native_store_idt,
1190 .store_tr = xen_store_tr,
1192 .write_ldt_entry = xen_write_ldt_entry,
1193 .write_gdt_entry = xen_write_gdt_entry,
1194 .write_idt_entry = xen_write_idt_entry,
1195 .load_sp0 = xen_load_sp0,
1197 .set_iopl_mask = xen_set_iopl_mask,
1198 .io_delay = xen_io_delay,
1200 /* Xen takes care of %gs when switching to usermode for us */
1201 .swapgs = paravirt_nop,
1204 .enter = paravirt_enter_lazy_cpu,
1205 .leave = xen_leave_lazy,
1209 static void __init __xen_init_IRQ(void)
1211 #ifdef CONFIG_X86_64
1214 /* Create identity vector->irq map */
1215 for(i = 0; i < NR_VECTORS; i++) {
1218 for_each_possible_cpu(cpu)
1219 per_cpu(vector_irq, cpu)[i] = i;
1221 #endif /* CONFIG_X86_64 */
1226 static const struct pv_irq_ops xen_irq_ops __initdata = {
1227 .init_IRQ = __xen_init_IRQ,
1228 .save_fl = xen_save_fl,
1229 .restore_fl = xen_restore_fl,
1230 .irq_disable = xen_irq_disable,
1231 .irq_enable = xen_irq_enable,
1232 .safe_halt = xen_safe_halt,
1234 #ifdef CONFIG_X86_64
1235 .adjust_exception_frame = xen_adjust_exception_frame,
1239 static const struct pv_apic_ops xen_apic_ops __initdata = {
1240 #ifdef CONFIG_X86_LOCAL_APIC
1241 .apic_write = xen_apic_write,
1242 .apic_write_atomic = xen_apic_write,
1243 .apic_read = xen_apic_read,
1244 .setup_boot_clock = paravirt_nop,
1245 .setup_secondary_clock = paravirt_nop,
1246 .startup_ipi_hook = paravirt_nop,
1250 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1251 .pagetable_setup_start = xen_pagetable_setup_start,
1252 .pagetable_setup_done = xen_pagetable_setup_done,
1254 .read_cr2 = xen_read_cr2,
1255 .write_cr2 = xen_write_cr2,
1257 .read_cr3 = xen_read_cr3,
1258 .write_cr3 = xen_write_cr3,
1260 .flush_tlb_user = xen_flush_tlb,
1261 .flush_tlb_kernel = xen_flush_tlb,
1262 .flush_tlb_single = xen_flush_tlb_single,
1263 .flush_tlb_others = xen_flush_tlb_others,
1265 .pte_update = paravirt_nop,
1266 .pte_update_defer = paravirt_nop,
1268 .pgd_alloc = xen_pgd_alloc,
1269 .pgd_free = xen_pgd_free,
1271 .alloc_pte = xen_alloc_pte_init,
1272 .release_pte = xen_release_pte_init,
1273 .alloc_pmd = xen_alloc_pte_init,
1274 .alloc_pmd_clone = paravirt_nop,
1275 .release_pmd = xen_release_pte_init,
1277 #ifdef CONFIG_HIGHPTE
1278 .kmap_atomic_pte = xen_kmap_atomic_pte,
1281 #ifdef CONFIG_X86_64
1282 .set_pte = xen_set_pte,
1284 .set_pte = xen_set_pte_init,
1286 .set_pte_at = xen_set_pte_at,
1287 .set_pmd = xen_set_pmd_hyper,
1289 .ptep_modify_prot_start = __ptep_modify_prot_start,
1290 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1292 .pte_val = xen_pte_val,
1293 .pte_flags = native_pte_val,
1294 .pgd_val = xen_pgd_val,
1296 .make_pte = xen_make_pte,
1297 .make_pgd = xen_make_pgd,
1299 #ifdef CONFIG_X86_PAE
1300 .set_pte_atomic = xen_set_pte_atomic,
1301 .set_pte_present = xen_set_pte_at,
1302 .pte_clear = xen_pte_clear,
1303 .pmd_clear = xen_pmd_clear,
1304 #endif /* CONFIG_X86_PAE */
1305 .set_pud = xen_set_pud_hyper,
1307 .make_pmd = xen_make_pmd,
1308 .pmd_val = xen_pmd_val,
1310 #if PAGETABLE_LEVELS == 4
1311 .pud_val = xen_pud_val,
1312 .make_pud = xen_make_pud,
1313 .set_pgd = xen_set_pgd_hyper,
1315 .alloc_pud = xen_alloc_pte_init,
1316 .release_pud = xen_release_pte_init,
1317 #endif /* PAGETABLE_LEVELS == 4 */
1319 .activate_mm = xen_activate_mm,
1320 .dup_mmap = xen_dup_mmap,
1321 .exit_mmap = xen_exit_mmap,
1324 .enter = paravirt_enter_lazy_mmu,
1325 .leave = xen_leave_lazy,
1328 .set_fixmap = xen_set_fixmap,
1331 static void xen_reboot(int reason)
1333 struct sched_shutdown r = { .reason = reason };
1339 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1343 static void xen_restart(char *msg)
1345 xen_reboot(SHUTDOWN_reboot);
1348 static void xen_emergency_restart(void)
1350 xen_reboot(SHUTDOWN_reboot);
1353 static void xen_machine_halt(void)
1355 xen_reboot(SHUTDOWN_poweroff);
1358 static void xen_crash_shutdown(struct pt_regs *regs)
1360 xen_reboot(SHUTDOWN_crash);
1363 static const struct machine_ops __initdata xen_machine_ops = {
1364 .restart = xen_restart,
1365 .halt = xen_machine_halt,
1366 .power_off = xen_machine_halt,
1367 .shutdown = xen_machine_halt,
1368 .crash_shutdown = xen_crash_shutdown,
1369 .emergency_restart = xen_emergency_restart,
1373 static void __init xen_reserve_top(void)
1375 #ifdef CONFIG_X86_32
1376 unsigned long top = HYPERVISOR_VIRT_START;
1377 struct xen_platform_parameters pp;
1379 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1380 top = pp.virt_start;
1382 reserve_top_address(-top + 2 * PAGE_SIZE);
1383 #endif /* CONFIG_X86_32 */
1387 * Like __va(), but returns address in the kernel mapping (which is
1388 * all we have until the physical memory mapping has been set up.
1390 static void *__ka(phys_addr_t paddr)
1392 #ifdef CONFIG_X86_64
1393 return (void *)(paddr + __START_KERNEL_map);
1399 /* Convert a machine address to physical address */
1400 static unsigned long m2p(phys_addr_t maddr)
1405 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1410 /* Convert a machine address to kernel virtual */
1411 static void *m2v(phys_addr_t maddr)
1413 return __ka(m2p(maddr));
1416 #ifdef CONFIG_X86_64
1417 static void walk(pgd_t *pgd, unsigned long addr)
1419 unsigned l4idx = pgd_index(addr);
1420 unsigned l3idx = pud_index(addr);
1421 unsigned l2idx = pmd_index(addr);
1422 unsigned l1idx = pte_index(addr);
1428 xen_raw_printk("walk %p, %lx -> %d %d %d %d\n",
1429 pgd, addr, l4idx, l3idx, l2idx, l1idx);
1432 xen_raw_printk(" l4: %016lx\n", l4.pgd);
1433 xen_raw_printk(" %016lx\n", pgd_val(l4));
1435 l3 = ((pud_t *)(m2v(l4.pgd)))[l3idx];
1436 xen_raw_printk(" l3: %016lx\n", l3.pud);
1437 xen_raw_printk(" %016lx\n", pud_val(l3));
1439 l2 = ((pmd_t *)(m2v(l3.pud)))[l2idx];
1440 xen_raw_printk(" l2: %016lx\n", l2.pmd);
1441 xen_raw_printk(" %016lx\n", pmd_val(l2));
1443 l1 = ((pte_t *)(m2v(l2.pmd)))[l1idx];
1444 xen_raw_printk(" l1: %016lx\n", l1.pte);
1445 xen_raw_printk(" %016lx\n", pte_val(l1));
1449 static void set_page_prot(void *addr, pgprot_t prot)
1451 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1452 pte_t pte = pfn_pte(pfn, prot);
1454 xen_raw_printk("addr=%p pfn=%lx mfn=%lx prot=%016llx pte=%016llx\n",
1455 addr, pfn, get_phys_to_machine(pfn),
1456 pgprot_val(prot), pte.pte);
1458 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1462 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1464 unsigned pmdidx, pteidx;
1470 for(pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1473 /* Reuse or allocate a page of ptes */
1474 if (pmd_present(pmd[pmdidx]))
1475 pte_page = m2v(pmd[pmdidx].pmd);
1477 /* Check for free pte pages */
1478 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1481 pte_page = &level1_ident_pgt[ident_pte];
1482 ident_pte += PTRS_PER_PTE;
1484 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1487 /* Install mappings */
1488 for(pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1491 if (pfn > max_pfn_mapped)
1492 max_pfn_mapped = pfn;
1494 if (!pte_none(pte_page[pteidx]))
1497 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1498 pte_page[pteidx] = pte;
1502 for(pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1503 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1505 set_page_prot(pmd, PAGE_KERNEL_RO);
1508 #ifdef CONFIG_X86_64
1509 static void convert_pfn_mfn(void *v)
1514 /* All levels are converted the same way, so just treat them
1516 for(i = 0; i < PTRS_PER_PTE; i++)
1517 pte[i] = xen_make_pte(pte[i].pte);
1521 * Set up the inital kernel pagetable.
1523 * We can construct this by grafting the Xen provided pagetable into
1524 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1525 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1526 * means that only the kernel has a physical mapping to start with -
1527 * but that's enough to get __va working. We need to fill in the rest
1528 * of the physical mapping once some sort of allocator has been set
1531 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1536 /* Zap identity mapping */
1537 init_level4_pgt[0] = __pgd(0);
1539 /* Pre-constructed entries are in pfn, so convert to mfn */
1540 convert_pfn_mfn(init_level4_pgt);
1541 convert_pfn_mfn(level3_ident_pgt);
1542 convert_pfn_mfn(level3_kernel_pgt);
1544 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1545 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1547 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1548 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1550 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1551 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1552 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1554 /* Set up identity map */
1555 xen_map_identity_early(level2_ident_pgt, max_pfn);
1557 /* Make pagetable pieces RO */
1558 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1559 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1560 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1561 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1562 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1563 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1565 /* Pin down new L4 */
1566 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1567 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1569 /* Unpin Xen-provided one */
1570 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1573 pgd = init_level4_pgt;
1576 * At this stage there can be no user pgd, and no page
1577 * structure to attach it to, so make sure we just set kernel
1581 __xen_write_cr3(true, __pa(pgd));
1582 xen_mc_issue(PARAVIRT_LAZY_CPU);
1584 reserve_early(__pa(xen_start_info->pt_base),
1585 __pa(xen_start_info->pt_base +
1586 xen_start_info->nr_pt_frames * PAGE_SIZE),
1591 #else /* !CONFIG_X86_64 */
1592 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1594 static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1598 init_pg_tables_start = __pa(pgd);
1599 init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1600 max_pfn_mapped = PFN_DOWN(init_pg_tables_end + 512*1024);
1602 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1603 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1605 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1607 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1608 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1609 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1611 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1612 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1613 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1615 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1617 xen_write_cr3(__pa(swapper_pg_dir));
1619 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1621 return swapper_pg_dir;
1623 #endif /* CONFIG_X86_64 */
1625 /* First C function to be called on Xen boot */
1626 asmlinkage void __init xen_start_kernel(void)
1630 if (!xen_start_info)
1633 BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
1635 xen_setup_features();
1637 /* Install Xen paravirt ops */
1639 pv_init_ops = xen_init_ops;
1640 pv_time_ops = xen_time_ops;
1641 pv_cpu_ops = xen_cpu_ops;
1642 pv_irq_ops = xen_irq_ops;
1643 pv_apic_ops = xen_apic_ops;
1644 pv_mmu_ops = xen_mmu_ops;
1646 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1647 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1648 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1651 machine_ops = xen_machine_ops;
1653 #ifdef CONFIG_X86_64
1654 /* Disable until direct per-cpu data access. */
1655 have_vcpu_info_placement = 0;
1662 if (!xen_feature(XENFEAT_auto_translated_physmap))
1663 xen_build_dynamic_phys_to_machine();
1665 pgd = (pgd_t *)xen_start_info->pt_base;
1667 /* Prevent unwanted bits from being set in PTEs. */
1668 __supported_pte_mask &= ~_PAGE_GLOBAL;
1669 if (!is_initial_xendomain())
1670 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1672 /* Don't do the full vcpu_info placement stuff until we have a
1673 possible map and a non-dummy shared_info. */
1674 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1676 xen_raw_console_write("mapping kernel into physical memory\n");
1677 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1681 /* keep using Xen gdt for now; no urgent need to change it */
1683 pv_info.kernel_rpl = 1;
1684 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1685 pv_info.kernel_rpl = 0;
1687 /* set the limit of our address space */
1690 #ifdef CONFIG_X86_32
1691 /* set up basic CPUID stuff */
1692 cpu_detect(&new_cpu_data);
1693 new_cpu_data.hard_math = 1;
1694 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1697 /* Poke various useful things into boot_params */
1698 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1699 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1700 ? __pa(xen_start_info->mod_start) : 0;
1701 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1702 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1704 if (!is_initial_xendomain()) {
1705 add_preferred_console("xenboot", 0, NULL);
1706 add_preferred_console("tty", 0, NULL);
1707 add_preferred_console("hvc", 0, NULL);
1710 xen_raw_console_write("about to get started...\n");
1713 xen_raw_printk("&boot_params=%p __pa(&boot_params)=%lx __va(__pa(&boot_params))=%lx\n",
1714 &boot_params, __pa_symbol(&boot_params),
1715 __va(__pa_symbol(&boot_params)));
1717 walk(pgd, &boot_params);
1718 walk(pgd, __va(__pa(&boot_params)));
1721 /* Start the world */
1722 #ifdef CONFIG_X86_32
1723 i386_start_kernel();
1725 x86_64_start_reservations((char *)__pa_symbol(&boot_params));