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/kprobes.h>
24 #include <linux/bootmem.h>
25 #include <linux/module.h>
27 #include <linux/page-flags.h>
28 #include <linux/highmem.h>
29 #include <linux/console.h>
31 #include <xen/interface/xen.h>
32 #include <xen/interface/version.h>
33 #include <xen/interface/physdev.h>
34 #include <xen/interface/vcpu.h>
35 #include <xen/features.h>
37 #include <xen/hvc-console.h>
39 #include <asm/paravirt.h>
42 #include <asm/xen/hypercall.h>
43 #include <asm/xen/hypervisor.h>
44 #include <asm/fixmap.h>
45 #include <asm/processor.h>
46 #include <asm/proto.h>
47 #include <asm/msr-index.h>
48 #include <asm/traps.h>
49 #include <asm/setup.h>
51 #include <asm/pgtable.h>
52 #include <asm/tlbflush.h>
53 #include <asm/reboot.h>
57 #include "multicalls.h"
59 EXPORT_SYMBOL_GPL(hypercall_page);
61 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
62 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
64 enum xen_domain_type xen_domain_type = XEN_NATIVE;
65 EXPORT_SYMBOL_GPL(xen_domain_type);
67 struct start_info *xen_start_info;
68 EXPORT_SYMBOL_GPL(xen_start_info);
70 struct shared_info xen_dummy_shared_info;
72 void *xen_initial_gdt;
75 * Point at some empty memory to start with. We map the real shared_info
76 * page as soon as fixmap is up and running.
78 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
81 * Flag to determine whether vcpu info placement is available on all
82 * VCPUs. We assume it is to start with, and then set it to zero on
83 * the first failure. This is because it can succeed on some VCPUs
84 * and not others, since it can involve hypervisor memory allocation,
85 * or because the guest failed to guarantee all the appropriate
86 * constraints on all VCPUs (ie buffer can't cross a page boundary).
88 * Note that any particular CPU may be using a placed vcpu structure,
89 * but we can only optimise if the all are.
91 * 0: not available, 1: available
93 static int have_vcpu_info_placement = 1;
95 static void xen_vcpu_setup(int cpu)
97 struct vcpu_register_vcpu_info info;
99 struct vcpu_info *vcpup;
101 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
102 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
104 if (!have_vcpu_info_placement)
105 return; /* already tested, not available */
107 vcpup = &per_cpu(xen_vcpu_info, cpu);
109 info.mfn = arbitrary_virt_to_mfn(vcpup);
110 info.offset = offset_in_page(vcpup);
112 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
113 cpu, vcpup, info.mfn, info.offset);
115 /* Check to see if the hypervisor will put the vcpu_info
116 structure where we want it, which allows direct access via
117 a percpu-variable. */
118 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
121 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
122 have_vcpu_info_placement = 0;
124 /* This cpu is using the registered vcpu info, even if
125 later ones fail to. */
126 per_cpu(xen_vcpu, cpu) = vcpup;
128 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
134 * On restore, set the vcpu placement up again.
135 * If it fails, then we're in a bad state, since
136 * we can't back out from using it...
138 void xen_vcpu_restore(void)
140 if (have_vcpu_info_placement) {
143 for_each_online_cpu(cpu) {
144 bool other_cpu = (cpu != smp_processor_id());
147 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
153 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
157 BUG_ON(!have_vcpu_info_placement);
161 static void __init xen_banner(void)
163 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
164 struct xen_extraversion extra;
165 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
167 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
169 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
170 version >> 16, version & 0xffff, extra.extraversion,
171 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
174 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
175 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
177 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
178 unsigned int *cx, unsigned int *dx)
180 unsigned maskecx = ~0;
181 unsigned maskedx = ~0;
184 * Mask out inconvenient features, to try and disable as many
185 * unsupported kernel subsystems as possible.
188 maskecx = cpuid_leaf1_ecx_mask;
189 maskedx = cpuid_leaf1_edx_mask;
192 asm(XEN_EMULATE_PREFIX "cpuid"
197 : "0" (*ax), "2" (*cx));
203 static __init void xen_init_cpuid_mask(void)
205 unsigned int ax, bx, cx, dx;
207 cpuid_leaf1_edx_mask =
208 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
209 (1 << X86_FEATURE_MCA) | /* disable MCA */
210 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
212 if (!xen_initial_domain())
213 cpuid_leaf1_edx_mask &=
214 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
215 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
219 xen_cpuid(&ax, &bx, &cx, &dx);
221 /* cpuid claims we support xsave; try enabling it to see what happens */
222 if (cx & (1 << (X86_FEATURE_XSAVE % 32))) {
225 set_in_cr4(X86_CR4_OSXSAVE);
229 if ((cr4 & X86_CR4_OSXSAVE) == 0)
230 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32));
232 clear_in_cr4(X86_CR4_OSXSAVE);
236 static void xen_set_debugreg(int reg, unsigned long val)
238 HYPERVISOR_set_debugreg(reg, val);
241 static unsigned long xen_get_debugreg(int reg)
243 return HYPERVISOR_get_debugreg(reg);
246 static void xen_end_context_switch(struct task_struct *next)
249 paravirt_end_context_switch(next);
252 static unsigned long xen_store_tr(void)
258 * Set the page permissions for a particular virtual address. If the
259 * address is a vmalloc mapping (or other non-linear mapping), then
260 * find the linear mapping of the page and also set its protections to
263 static void set_aliased_prot(void *v, pgprot_t prot)
271 ptep = lookup_address((unsigned long)v, &level);
272 BUG_ON(ptep == NULL);
274 pfn = pte_pfn(*ptep);
275 page = pfn_to_page(pfn);
277 pte = pfn_pte(pfn, prot);
279 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
282 if (!PageHighMem(page)) {
283 void *av = __va(PFN_PHYS(pfn));
286 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
292 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
294 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
297 for(i = 0; i < entries; i += entries_per_page)
298 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
301 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
303 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
306 for(i = 0; i < entries; i += entries_per_page)
307 set_aliased_prot(ldt + i, PAGE_KERNEL);
310 static void xen_set_ldt(const void *addr, unsigned entries)
312 struct mmuext_op *op;
313 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
316 op->cmd = MMUEXT_SET_LDT;
317 op->arg1.linear_addr = (unsigned long)addr;
318 op->arg2.nr_ents = entries;
320 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
322 xen_mc_issue(PARAVIRT_LAZY_CPU);
325 static void xen_load_gdt(const struct desc_ptr *dtr)
327 unsigned long va = dtr->address;
328 unsigned int size = dtr->size + 1;
329 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
330 unsigned long frames[pages];
333 /* A GDT can be up to 64k in size, which corresponds to 8192
334 8-byte entries, or 16 4k pages.. */
336 BUG_ON(size > 65536);
337 BUG_ON(va & ~PAGE_MASK);
339 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
341 pte_t *ptep = lookup_address(va, &level);
342 unsigned long pfn, mfn;
345 BUG_ON(ptep == NULL);
347 pfn = pte_pfn(*ptep);
348 mfn = pfn_to_mfn(pfn);
349 virt = __va(PFN_PHYS(pfn));
353 make_lowmem_page_readonly((void *)va);
354 make_lowmem_page_readonly(virt);
357 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
361 static void load_TLS_descriptor(struct thread_struct *t,
362 unsigned int cpu, unsigned int i)
364 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
365 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
366 struct multicall_space mc = __xen_mc_entry(0);
368 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
371 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
374 * XXX sleazy hack: If we're being called in a lazy-cpu zone
375 * and lazy gs handling is enabled, it means we're in a
376 * context switch, and %gs has just been saved. This means we
377 * can zero it out to prevent faults on exit from the
378 * hypervisor if the next process has no %gs. Either way, it
379 * has been saved, and the new value will get loaded properly.
380 * This will go away as soon as Xen has been modified to not
381 * save/restore %gs for normal hypercalls.
383 * On x86_64, this hack is not used for %gs, because gs points
384 * to KERNEL_GS_BASE (and uses it for PDA references), so we
385 * must not zero %gs on x86_64
387 * For x86_64, we need to zero %fs, otherwise we may get an
388 * exception between the new %fs descriptor being loaded and
389 * %fs being effectively cleared at __switch_to().
391 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
401 load_TLS_descriptor(t, cpu, 0);
402 load_TLS_descriptor(t, cpu, 1);
403 load_TLS_descriptor(t, cpu, 2);
405 xen_mc_issue(PARAVIRT_LAZY_CPU);
409 static void xen_load_gs_index(unsigned int idx)
411 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
416 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
419 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
420 u64 entry = *(u64 *)ptr;
425 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
431 static int cvt_gate_to_trap(int vector, const gate_desc *val,
432 struct trap_info *info)
436 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
439 info->vector = vector;
441 addr = gate_offset(*val);
444 * Look for known traps using IST, and substitute them
445 * appropriately. The debugger ones are the only ones we care
446 * about. Xen will handle faults like double_fault and
447 * machine_check, so we should never see them. Warn if
448 * there's an unexpected IST-using fault handler.
450 if (addr == (unsigned long)debug)
451 addr = (unsigned long)xen_debug;
452 else if (addr == (unsigned long)int3)
453 addr = (unsigned long)xen_int3;
454 else if (addr == (unsigned long)stack_segment)
455 addr = (unsigned long)xen_stack_segment;
456 else if (addr == (unsigned long)double_fault ||
457 addr == (unsigned long)nmi) {
458 /* Don't need to handle these */
460 #ifdef CONFIG_X86_MCE
461 } else if (addr == (unsigned long)machine_check) {
465 /* Some other trap using IST? */
466 if (WARN_ON(val->ist != 0))
469 #endif /* CONFIG_X86_64 */
470 info->address = addr;
472 info->cs = gate_segment(*val);
473 info->flags = val->dpl;
474 /* interrupt gates clear IF */
475 if (val->type == GATE_INTERRUPT)
476 info->flags |= 1 << 2;
481 /* Locations of each CPU's IDT */
482 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
484 /* Set an IDT entry. If the entry is part of the current IDT, then
486 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
488 unsigned long p = (unsigned long)&dt[entrynum];
489 unsigned long start, end;
493 start = __get_cpu_var(idt_desc).address;
494 end = start + __get_cpu_var(idt_desc).size + 1;
498 native_write_idt_entry(dt, entrynum, g);
500 if (p >= start && (p + 8) <= end) {
501 struct trap_info info[2];
505 if (cvt_gate_to_trap(entrynum, g, &info[0]))
506 if (HYPERVISOR_set_trap_table(info))
513 static void xen_convert_trap_info(const struct desc_ptr *desc,
514 struct trap_info *traps)
516 unsigned in, out, count;
518 count = (desc->size+1) / sizeof(gate_desc);
521 for (in = out = 0; in < count; in++) {
522 gate_desc *entry = (gate_desc*)(desc->address) + in;
524 if (cvt_gate_to_trap(in, entry, &traps[out]))
527 traps[out].address = 0;
530 void xen_copy_trap_info(struct trap_info *traps)
532 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
534 xen_convert_trap_info(desc, traps);
537 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
538 hold a spinlock to protect the static traps[] array (static because
539 it avoids allocation, and saves stack space). */
540 static void xen_load_idt(const struct desc_ptr *desc)
542 static DEFINE_SPINLOCK(lock);
543 static struct trap_info traps[257];
547 __get_cpu_var(idt_desc) = *desc;
549 xen_convert_trap_info(desc, traps);
552 if (HYPERVISOR_set_trap_table(traps))
558 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
559 they're handled differently. */
560 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
561 const void *desc, int type)
572 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
575 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
584 static void xen_load_sp0(struct tss_struct *tss,
585 struct thread_struct *thread)
587 struct multicall_space mcs = xen_mc_entry(0);
588 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
589 xen_mc_issue(PARAVIRT_LAZY_CPU);
592 static void xen_set_iopl_mask(unsigned mask)
594 struct physdev_set_iopl set_iopl;
596 /* Force the change at ring 0. */
597 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
598 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
601 static void xen_io_delay(void)
605 #ifdef CONFIG_X86_LOCAL_APIC
606 static u32 xen_apic_read(u32 reg)
611 static void xen_apic_write(u32 reg, u32 val)
613 /* Warn to see if there's any stray references */
617 static u64 xen_apic_icr_read(void)
622 static void xen_apic_icr_write(u32 low, u32 id)
624 /* Warn to see if there's any stray references */
628 static void xen_apic_wait_icr_idle(void)
633 static u32 xen_safe_apic_wait_icr_idle(void)
638 static void set_xen_basic_apic_ops(void)
640 apic->read = xen_apic_read;
641 apic->write = xen_apic_write;
642 apic->icr_read = xen_apic_icr_read;
643 apic->icr_write = xen_apic_icr_write;
644 apic->wait_icr_idle = xen_apic_wait_icr_idle;
645 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
651 static void xen_clts(void)
653 struct multicall_space mcs;
655 mcs = xen_mc_entry(0);
657 MULTI_fpu_taskswitch(mcs.mc, 0);
659 xen_mc_issue(PARAVIRT_LAZY_CPU);
662 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
664 static unsigned long xen_read_cr0(void)
666 unsigned long cr0 = percpu_read(xen_cr0_value);
668 if (unlikely(cr0 == 0)) {
669 cr0 = native_read_cr0();
670 percpu_write(xen_cr0_value, cr0);
676 static void xen_write_cr0(unsigned long cr0)
678 struct multicall_space mcs;
680 percpu_write(xen_cr0_value, cr0);
682 /* Only pay attention to cr0.TS; everything else is
684 mcs = xen_mc_entry(0);
686 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
688 xen_mc_issue(PARAVIRT_LAZY_CPU);
691 static void xen_write_cr4(unsigned long cr4)
696 native_write_cr4(cr4);
699 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
710 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
711 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
712 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
715 base = ((u64)high << 32) | low;
716 if (HYPERVISOR_set_segment_base(which, base) != 0)
724 case MSR_SYSCALL_MASK:
725 case MSR_IA32_SYSENTER_CS:
726 case MSR_IA32_SYSENTER_ESP:
727 case MSR_IA32_SYSENTER_EIP:
728 /* Fast syscall setup is all done in hypercalls, so
729 these are all ignored. Stub them out here to stop
730 Xen console noise. */
734 ret = native_write_msr_safe(msr, low, high);
740 void xen_setup_shared_info(void)
742 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
743 set_fixmap(FIX_PARAVIRT_BOOTMAP,
744 xen_start_info->shared_info);
746 HYPERVISOR_shared_info =
747 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
749 HYPERVISOR_shared_info =
750 (struct shared_info *)__va(xen_start_info->shared_info);
753 /* In UP this is as good a place as any to set up shared info */
754 xen_setup_vcpu_info_placement();
757 xen_setup_mfn_list_list();
760 /* This is called once we have the cpu_possible_map */
761 void xen_setup_vcpu_info_placement(void)
765 for_each_possible_cpu(cpu)
768 /* xen_vcpu_setup managed to place the vcpu_info within the
769 percpu area for all cpus, so make use of it */
770 if (have_vcpu_info_placement) {
771 printk(KERN_INFO "Xen: using vcpu_info placement\n");
773 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
774 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
775 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
776 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
777 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
781 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
782 unsigned long addr, unsigned len)
784 char *start, *end, *reloc;
787 start = end = reloc = NULL;
789 #define SITE(op, x) \
790 case PARAVIRT_PATCH(op.x): \
791 if (have_vcpu_info_placement) { \
792 start = (char *)xen_##x##_direct; \
793 end = xen_##x##_direct_end; \
794 reloc = xen_##x##_direct_reloc; \
799 SITE(pv_irq_ops, irq_enable);
800 SITE(pv_irq_ops, irq_disable);
801 SITE(pv_irq_ops, save_fl);
802 SITE(pv_irq_ops, restore_fl);
806 if (start == NULL || (end-start) > len)
809 ret = paravirt_patch_insns(insnbuf, len, start, end);
811 /* Note: because reloc is assigned from something that
812 appears to be an array, gcc assumes it's non-null,
813 but doesn't know its relationship with start and
815 if (reloc > start && reloc < end) {
816 int reloc_off = reloc - start;
817 long *relocp = (long *)(insnbuf + reloc_off);
818 long delta = start - (char *)addr;
826 ret = paravirt_patch_default(type, clobbers, insnbuf,
834 static const struct pv_info xen_info __initdata = {
835 .paravirt_enabled = 1,
836 .shared_kernel_pmd = 0,
841 static const struct pv_init_ops xen_init_ops __initdata = {
844 .banner = xen_banner,
845 .memory_setup = xen_memory_setup,
846 .arch_setup = xen_arch_setup,
847 .post_allocator_init = xen_post_allocator_init,
850 static const struct pv_time_ops xen_time_ops __initdata = {
851 .time_init = xen_time_init,
853 .set_wallclock = xen_set_wallclock,
854 .get_wallclock = xen_get_wallclock,
855 .get_tsc_khz = xen_tsc_khz,
856 .sched_clock = xen_sched_clock,
859 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
862 .set_debugreg = xen_set_debugreg,
863 .get_debugreg = xen_get_debugreg,
867 .read_cr0 = xen_read_cr0,
868 .write_cr0 = xen_write_cr0,
870 .read_cr4 = native_read_cr4,
871 .read_cr4_safe = native_read_cr4_safe,
872 .write_cr4 = xen_write_cr4,
874 .wbinvd = native_wbinvd,
876 .read_msr = native_read_msr_safe,
877 .write_msr = xen_write_msr_safe,
878 .read_tsc = native_read_tsc,
879 .read_pmc = native_read_pmc,
882 .irq_enable_sysexit = xen_sysexit,
884 .usergs_sysret32 = xen_sysret32,
885 .usergs_sysret64 = xen_sysret64,
888 .load_tr_desc = paravirt_nop,
889 .set_ldt = xen_set_ldt,
890 .load_gdt = xen_load_gdt,
891 .load_idt = xen_load_idt,
892 .load_tls = xen_load_tls,
894 .load_gs_index = xen_load_gs_index,
897 .alloc_ldt = xen_alloc_ldt,
898 .free_ldt = xen_free_ldt,
900 .store_gdt = native_store_gdt,
901 .store_idt = native_store_idt,
902 .store_tr = xen_store_tr,
904 .write_ldt_entry = xen_write_ldt_entry,
905 .write_gdt_entry = xen_write_gdt_entry,
906 .write_idt_entry = xen_write_idt_entry,
907 .load_sp0 = xen_load_sp0,
909 .set_iopl_mask = xen_set_iopl_mask,
910 .io_delay = xen_io_delay,
912 /* Xen takes care of %gs when switching to usermode for us */
913 .swapgs = paravirt_nop,
915 .start_context_switch = paravirt_start_context_switch,
916 .end_context_switch = xen_end_context_switch,
919 static const struct pv_apic_ops xen_apic_ops __initdata = {
920 #ifdef CONFIG_X86_LOCAL_APIC
921 .setup_boot_clock = paravirt_nop,
922 .setup_secondary_clock = paravirt_nop,
923 .startup_ipi_hook = paravirt_nop,
927 static void xen_reboot(int reason)
929 struct sched_shutdown r = { .reason = reason };
935 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
939 static void xen_restart(char *msg)
941 xen_reboot(SHUTDOWN_reboot);
944 static void xen_emergency_restart(void)
946 xen_reboot(SHUTDOWN_reboot);
949 static void xen_machine_halt(void)
951 xen_reboot(SHUTDOWN_poweroff);
954 static void xen_crash_shutdown(struct pt_regs *regs)
956 xen_reboot(SHUTDOWN_crash);
959 static const struct machine_ops __initdata xen_machine_ops = {
960 .restart = xen_restart,
961 .halt = xen_machine_halt,
962 .power_off = xen_machine_halt,
963 .shutdown = xen_machine_halt,
964 .crash_shutdown = xen_crash_shutdown,
965 .emergency_restart = xen_emergency_restart,
968 /* First C function to be called on Xen boot */
969 asmlinkage void __init xen_start_kernel(void)
976 xen_domain_type = XEN_PV_DOMAIN;
978 /* Install Xen paravirt ops */
980 pv_init_ops = xen_init_ops;
981 pv_time_ops = xen_time_ops;
982 pv_cpu_ops = xen_cpu_ops;
983 pv_apic_ops = xen_apic_ops;
984 pv_mmu_ops = xen_mmu_ops;
988 * Setup percpu state. We only need to do this for 64-bit
989 * because 32-bit already has %fs set properly.
991 load_percpu_segment(0);
995 xen_init_cpuid_mask();
997 #ifdef CONFIG_X86_LOCAL_APIC
999 * set up the basic apic ops.
1001 set_xen_basic_apic_ops();
1004 xen_setup_features();
1006 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1007 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1008 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1011 machine_ops = xen_machine_ops;
1014 * The only reliable way to retain the initial address of the
1015 * percpu gdt_page is to remember it here, so we can go and
1016 * mark it RW later, when the initial percpu area is freed.
1018 xen_initial_gdt = &per_cpu(gdt_page, 0);
1023 if (!xen_feature(XENFEAT_auto_translated_physmap))
1024 xen_build_dynamic_phys_to_machine();
1026 pgd = (pgd_t *)xen_start_info->pt_base;
1028 /* Prevent unwanted bits from being set in PTEs. */
1029 __supported_pte_mask &= ~_PAGE_GLOBAL;
1030 if (!xen_initial_domain())
1031 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1033 #ifdef CONFIG_X86_64
1034 /* Work out if we support NX */
1038 /* Don't do the full vcpu_info placement stuff until we have a
1039 possible map and a non-dummy shared_info. */
1040 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1042 local_irq_disable();
1043 early_boot_irqs_off();
1045 xen_raw_console_write("mapping kernel into physical memory\n");
1046 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1050 /* keep using Xen gdt for now; no urgent need to change it */
1052 pv_info.kernel_rpl = 1;
1053 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1054 pv_info.kernel_rpl = 0;
1056 /* set the limit of our address space */
1059 #ifdef CONFIG_X86_32
1060 /* set up basic CPUID stuff */
1061 cpu_detect(&new_cpu_data);
1062 new_cpu_data.hard_math = 1;
1063 new_cpu_data.wp_works_ok = 1;
1064 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1067 /* Poke various useful things into boot_params */
1068 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1069 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1070 ? __pa(xen_start_info->mod_start) : 0;
1071 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1072 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1074 if (!xen_initial_domain()) {
1075 add_preferred_console("xenboot", 0, NULL);
1076 add_preferred_console("tty", 0, NULL);
1077 add_preferred_console("hvc", 0, NULL);
1080 xen_raw_console_write("about to get started...\n");
1082 /* Start the world */
1083 #ifdef CONFIG_X86_32
1084 i386_start_kernel();
1086 x86_64_start_reservations((char *)__pa_symbol(&boot_params));