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>
54 #include <asm/stackprotector.h>
58 #include "multicalls.h"
60 EXPORT_SYMBOL_GPL(hypercall_page);
62 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
63 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
65 enum xen_domain_type xen_domain_type = XEN_NATIVE;
66 EXPORT_SYMBOL_GPL(xen_domain_type);
68 struct start_info *xen_start_info;
69 EXPORT_SYMBOL_GPL(xen_start_info);
71 struct shared_info xen_dummy_shared_info;
73 void *xen_initial_gdt;
76 * Point at some empty memory to start with. We map the real shared_info
77 * page as soon as fixmap is up and running.
79 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
82 * Flag to determine whether vcpu info placement is available on all
83 * VCPUs. We assume it is to start with, and then set it to zero on
84 * the first failure. This is because it can succeed on some VCPUs
85 * and not others, since it can involve hypervisor memory allocation,
86 * or because the guest failed to guarantee all the appropriate
87 * constraints on all VCPUs (ie buffer can't cross a page boundary).
89 * Note that any particular CPU may be using a placed vcpu structure,
90 * but we can only optimise if the all are.
92 * 0: not available, 1: available
94 static int have_vcpu_info_placement = 1;
96 static void xen_vcpu_setup(int cpu)
98 struct vcpu_register_vcpu_info info;
100 struct vcpu_info *vcpup;
102 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
103 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
105 if (!have_vcpu_info_placement)
106 return; /* already tested, not available */
108 vcpup = &per_cpu(xen_vcpu_info, cpu);
110 info.mfn = arbitrary_virt_to_mfn(vcpup);
111 info.offset = offset_in_page(vcpup);
113 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
114 cpu, vcpup, info.mfn, info.offset);
116 /* Check to see if the hypervisor will put the vcpu_info
117 structure where we want it, which allows direct access via
118 a percpu-variable. */
119 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
122 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
123 have_vcpu_info_placement = 0;
125 /* This cpu is using the registered vcpu info, even if
126 later ones fail to. */
127 per_cpu(xen_vcpu, cpu) = vcpup;
129 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
135 * On restore, set the vcpu placement up again.
136 * If it fails, then we're in a bad state, since
137 * we can't back out from using it...
139 void xen_vcpu_restore(void)
143 for_each_online_cpu(cpu) {
144 bool other_cpu = (cpu != smp_processor_id());
147 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
150 xen_setup_runstate_info(cpu);
152 if (have_vcpu_info_placement)
156 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
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 maskebx = ~0;
181 unsigned maskecx = ~0;
182 unsigned maskedx = ~0;
185 * Mask out inconvenient features, to try and disable as many
186 * unsupported kernel subsystems as possible.
190 maskecx = cpuid_leaf1_ecx_mask;
191 maskedx = cpuid_leaf1_edx_mask;
195 /* Suppress extended topology stuff */
200 asm(XEN_EMULATE_PREFIX "cpuid"
205 : "0" (*ax), "2" (*cx));
212 static __init void xen_init_cpuid_mask(void)
214 unsigned int ax, bx, cx, dx;
216 cpuid_leaf1_edx_mask =
217 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
218 (1 << X86_FEATURE_MCA) | /* disable MCA */
219 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
221 if (!xen_initial_domain())
222 cpuid_leaf1_edx_mask &=
223 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
224 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
228 xen_cpuid(&ax, &bx, &cx, &dx);
230 /* cpuid claims we support xsave; try enabling it to see what happens */
231 if (cx & (1 << (X86_FEATURE_XSAVE % 32))) {
234 set_in_cr4(X86_CR4_OSXSAVE);
238 if ((cr4 & X86_CR4_OSXSAVE) == 0)
239 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32));
241 clear_in_cr4(X86_CR4_OSXSAVE);
245 static void xen_set_debugreg(int reg, unsigned long val)
247 HYPERVISOR_set_debugreg(reg, val);
250 static unsigned long xen_get_debugreg(int reg)
252 return HYPERVISOR_get_debugreg(reg);
255 static void xen_end_context_switch(struct task_struct *next)
258 paravirt_end_context_switch(next);
261 static unsigned long xen_store_tr(void)
267 * Set the page permissions for a particular virtual address. If the
268 * address is a vmalloc mapping (or other non-linear mapping), then
269 * find the linear mapping of the page and also set its protections to
272 static void set_aliased_prot(void *v, pgprot_t prot)
280 ptep = lookup_address((unsigned long)v, &level);
281 BUG_ON(ptep == NULL);
283 pfn = pte_pfn(*ptep);
284 page = pfn_to_page(pfn);
286 pte = pfn_pte(pfn, prot);
288 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
291 if (!PageHighMem(page)) {
292 void *av = __va(PFN_PHYS(pfn));
295 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
301 static void xen_alloc_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_RO);
310 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
312 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
315 for(i = 0; i < entries; i += entries_per_page)
316 set_aliased_prot(ldt + i, PAGE_KERNEL);
319 static void xen_set_ldt(const void *addr, unsigned entries)
321 struct mmuext_op *op;
322 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
325 op->cmd = MMUEXT_SET_LDT;
326 op->arg1.linear_addr = (unsigned long)addr;
327 op->arg2.nr_ents = entries;
329 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
331 xen_mc_issue(PARAVIRT_LAZY_CPU);
334 static void xen_load_gdt(const struct desc_ptr *dtr)
336 unsigned long va = dtr->address;
337 unsigned int size = dtr->size + 1;
338 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
339 unsigned long frames[pages];
343 * A GDT can be up to 64k in size, which corresponds to 8192
344 * 8-byte entries, or 16 4k pages..
347 BUG_ON(size > 65536);
348 BUG_ON(va & ~PAGE_MASK);
350 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
353 unsigned long pfn, mfn;
357 * The GDT is per-cpu and is in the percpu data area.
358 * That can be virtually mapped, so we need to do a
359 * page-walk to get the underlying MFN for the
360 * hypercall. The page can also be in the kernel's
361 * linear range, so we need to RO that mapping too.
363 ptep = lookup_address(va, &level);
364 BUG_ON(ptep == NULL);
366 pfn = pte_pfn(*ptep);
367 mfn = pfn_to_mfn(pfn);
368 virt = __va(PFN_PHYS(pfn));
372 make_lowmem_page_readonly((void *)va);
373 make_lowmem_page_readonly(virt);
376 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
381 * load_gdt for early boot, when the gdt is only mapped once
383 static __init void xen_load_gdt_boot(const struct desc_ptr *dtr)
385 unsigned long va = dtr->address;
386 unsigned int size = dtr->size + 1;
387 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
388 unsigned long frames[pages];
392 * A GDT can be up to 64k in size, which corresponds to 8192
393 * 8-byte entries, or 16 4k pages..
396 BUG_ON(size > 65536);
397 BUG_ON(va & ~PAGE_MASK);
399 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
401 unsigned long pfn, mfn;
403 pfn = virt_to_pfn(va);
404 mfn = pfn_to_mfn(pfn);
406 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
408 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
414 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
418 static void load_TLS_descriptor(struct thread_struct *t,
419 unsigned int cpu, unsigned int i)
421 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
422 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
423 struct multicall_space mc = __xen_mc_entry(0);
425 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
428 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
431 * XXX sleazy hack: If we're being called in a lazy-cpu zone
432 * and lazy gs handling is enabled, it means we're in a
433 * context switch, and %gs has just been saved. This means we
434 * can zero it out to prevent faults on exit from the
435 * hypervisor if the next process has no %gs. Either way, it
436 * has been saved, and the new value will get loaded properly.
437 * This will go away as soon as Xen has been modified to not
438 * save/restore %gs for normal hypercalls.
440 * On x86_64, this hack is not used for %gs, because gs points
441 * to KERNEL_GS_BASE (and uses it for PDA references), so we
442 * must not zero %gs on x86_64
444 * For x86_64, we need to zero %fs, otherwise we may get an
445 * exception between the new %fs descriptor being loaded and
446 * %fs being effectively cleared at __switch_to().
448 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
458 load_TLS_descriptor(t, cpu, 0);
459 load_TLS_descriptor(t, cpu, 1);
460 load_TLS_descriptor(t, cpu, 2);
462 xen_mc_issue(PARAVIRT_LAZY_CPU);
466 static void xen_load_gs_index(unsigned int idx)
468 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
473 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
476 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
477 u64 entry = *(u64 *)ptr;
482 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
488 static int cvt_gate_to_trap(int vector, const gate_desc *val,
489 struct trap_info *info)
493 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
496 info->vector = vector;
498 addr = gate_offset(*val);
501 * Look for known traps using IST, and substitute them
502 * appropriately. The debugger ones are the only ones we care
503 * about. Xen will handle faults like double_fault and
504 * machine_check, so we should never see them. Warn if
505 * there's an unexpected IST-using fault handler.
507 if (addr == (unsigned long)debug)
508 addr = (unsigned long)xen_debug;
509 else if (addr == (unsigned long)int3)
510 addr = (unsigned long)xen_int3;
511 else if (addr == (unsigned long)stack_segment)
512 addr = (unsigned long)xen_stack_segment;
513 else if (addr == (unsigned long)double_fault ||
514 addr == (unsigned long)nmi) {
515 /* Don't need to handle these */
517 #ifdef CONFIG_X86_MCE
518 } else if (addr == (unsigned long)machine_check) {
522 /* Some other trap using IST? */
523 if (WARN_ON(val->ist != 0))
526 #endif /* CONFIG_X86_64 */
527 info->address = addr;
529 info->cs = gate_segment(*val);
530 info->flags = val->dpl;
531 /* interrupt gates clear IF */
532 if (val->type == GATE_INTERRUPT)
533 info->flags |= 1 << 2;
538 /* Locations of each CPU's IDT */
539 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
541 /* Set an IDT entry. If the entry is part of the current IDT, then
543 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
545 unsigned long p = (unsigned long)&dt[entrynum];
546 unsigned long start, end;
550 start = __get_cpu_var(idt_desc).address;
551 end = start + __get_cpu_var(idt_desc).size + 1;
555 native_write_idt_entry(dt, entrynum, g);
557 if (p >= start && (p + 8) <= end) {
558 struct trap_info info[2];
562 if (cvt_gate_to_trap(entrynum, g, &info[0]))
563 if (HYPERVISOR_set_trap_table(info))
570 static void xen_convert_trap_info(const struct desc_ptr *desc,
571 struct trap_info *traps)
573 unsigned in, out, count;
575 count = (desc->size+1) / sizeof(gate_desc);
578 for (in = out = 0; in < count; in++) {
579 gate_desc *entry = (gate_desc*)(desc->address) + in;
581 if (cvt_gate_to_trap(in, entry, &traps[out]))
584 traps[out].address = 0;
587 void xen_copy_trap_info(struct trap_info *traps)
589 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
591 xen_convert_trap_info(desc, traps);
594 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
595 hold a spinlock to protect the static traps[] array (static because
596 it avoids allocation, and saves stack space). */
597 static void xen_load_idt(const struct desc_ptr *desc)
599 static DEFINE_SPINLOCK(lock);
600 static struct trap_info traps[257];
604 __get_cpu_var(idt_desc) = *desc;
606 xen_convert_trap_info(desc, traps);
609 if (HYPERVISOR_set_trap_table(traps))
615 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
616 they're handled differently. */
617 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
618 const void *desc, int type)
629 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
632 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
642 * Version of write_gdt_entry for use at early boot-time needed to
643 * update an entry as simply as possible.
645 static __init void xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
646 const void *desc, int type)
655 xmaddr_t maddr = virt_to_machine(&dt[entry]);
657 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
658 dt[entry] = *(struct desc_struct *)desc;
664 static void xen_load_sp0(struct tss_struct *tss,
665 struct thread_struct *thread)
667 struct multicall_space mcs = xen_mc_entry(0);
668 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
669 xen_mc_issue(PARAVIRT_LAZY_CPU);
672 static void xen_set_iopl_mask(unsigned mask)
674 struct physdev_set_iopl set_iopl;
676 /* Force the change at ring 0. */
677 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
678 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
681 static void xen_io_delay(void)
685 #ifdef CONFIG_X86_LOCAL_APIC
686 static u32 xen_apic_read(u32 reg)
691 static void xen_apic_write(u32 reg, u32 val)
693 /* Warn to see if there's any stray references */
697 static u64 xen_apic_icr_read(void)
702 static void xen_apic_icr_write(u32 low, u32 id)
704 /* Warn to see if there's any stray references */
708 static void xen_apic_wait_icr_idle(void)
713 static u32 xen_safe_apic_wait_icr_idle(void)
718 static void set_xen_basic_apic_ops(void)
720 apic->read = xen_apic_read;
721 apic->write = xen_apic_write;
722 apic->icr_read = xen_apic_icr_read;
723 apic->icr_write = xen_apic_icr_write;
724 apic->wait_icr_idle = xen_apic_wait_icr_idle;
725 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
731 static void xen_clts(void)
733 struct multicall_space mcs;
735 mcs = xen_mc_entry(0);
737 MULTI_fpu_taskswitch(mcs.mc, 0);
739 xen_mc_issue(PARAVIRT_LAZY_CPU);
742 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
744 static unsigned long xen_read_cr0(void)
746 unsigned long cr0 = percpu_read(xen_cr0_value);
748 if (unlikely(cr0 == 0)) {
749 cr0 = native_read_cr0();
750 percpu_write(xen_cr0_value, cr0);
756 static void xen_write_cr0(unsigned long cr0)
758 struct multicall_space mcs;
760 percpu_write(xen_cr0_value, cr0);
762 /* Only pay attention to cr0.TS; everything else is
764 mcs = xen_mc_entry(0);
766 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
768 xen_mc_issue(PARAVIRT_LAZY_CPU);
771 static void xen_write_cr4(unsigned long cr4)
776 native_write_cr4(cr4);
779 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
790 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
791 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
792 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
795 base = ((u64)high << 32) | low;
796 if (HYPERVISOR_set_segment_base(which, base) != 0)
804 case MSR_SYSCALL_MASK:
805 case MSR_IA32_SYSENTER_CS:
806 case MSR_IA32_SYSENTER_ESP:
807 case MSR_IA32_SYSENTER_EIP:
808 /* Fast syscall setup is all done in hypercalls, so
809 these are all ignored. Stub them out here to stop
810 Xen console noise. */
814 ret = native_write_msr_safe(msr, low, high);
820 void xen_setup_shared_info(void)
822 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
823 set_fixmap(FIX_PARAVIRT_BOOTMAP,
824 xen_start_info->shared_info);
826 HYPERVISOR_shared_info =
827 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
829 HYPERVISOR_shared_info =
830 (struct shared_info *)__va(xen_start_info->shared_info);
833 /* In UP this is as good a place as any to set up shared info */
834 xen_setup_vcpu_info_placement();
837 xen_setup_mfn_list_list();
840 /* This is called once we have the cpu_possible_map */
841 void xen_setup_vcpu_info_placement(void)
845 for_each_possible_cpu(cpu)
848 /* xen_vcpu_setup managed to place the vcpu_info within the
849 percpu area for all cpus, so make use of it */
850 if (have_vcpu_info_placement) {
851 printk(KERN_INFO "Xen: using vcpu_info placement\n");
853 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
854 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
855 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
856 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
857 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
861 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
862 unsigned long addr, unsigned len)
864 char *start, *end, *reloc;
867 start = end = reloc = NULL;
869 #define SITE(op, x) \
870 case PARAVIRT_PATCH(op.x): \
871 if (have_vcpu_info_placement) { \
872 start = (char *)xen_##x##_direct; \
873 end = xen_##x##_direct_end; \
874 reloc = xen_##x##_direct_reloc; \
879 SITE(pv_irq_ops, irq_enable);
880 SITE(pv_irq_ops, irq_disable);
881 SITE(pv_irq_ops, save_fl);
882 SITE(pv_irq_ops, restore_fl);
886 if (start == NULL || (end-start) > len)
889 ret = paravirt_patch_insns(insnbuf, len, start, end);
891 /* Note: because reloc is assigned from something that
892 appears to be an array, gcc assumes it's non-null,
893 but doesn't know its relationship with start and
895 if (reloc > start && reloc < end) {
896 int reloc_off = reloc - start;
897 long *relocp = (long *)(insnbuf + reloc_off);
898 long delta = start - (char *)addr;
906 ret = paravirt_patch_default(type, clobbers, insnbuf,
914 static const struct pv_info xen_info __initdata = {
915 .paravirt_enabled = 1,
916 .shared_kernel_pmd = 0,
921 static const struct pv_init_ops xen_init_ops __initdata = {
925 static const struct pv_time_ops xen_time_ops __initdata = {
926 .sched_clock = xen_sched_clock,
929 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
932 .set_debugreg = xen_set_debugreg,
933 .get_debugreg = xen_get_debugreg,
937 .read_cr0 = xen_read_cr0,
938 .write_cr0 = xen_write_cr0,
940 .read_cr4 = native_read_cr4,
941 .read_cr4_safe = native_read_cr4_safe,
942 .write_cr4 = xen_write_cr4,
944 .wbinvd = native_wbinvd,
946 .read_msr = native_read_msr_safe,
947 .write_msr = xen_write_msr_safe,
948 .read_tsc = native_read_tsc,
949 .read_pmc = native_read_pmc,
952 .irq_enable_sysexit = xen_sysexit,
954 .usergs_sysret32 = xen_sysret32,
955 .usergs_sysret64 = xen_sysret64,
958 .load_tr_desc = paravirt_nop,
959 .set_ldt = xen_set_ldt,
960 .load_gdt = xen_load_gdt,
961 .load_idt = xen_load_idt,
962 .load_tls = xen_load_tls,
964 .load_gs_index = xen_load_gs_index,
967 .alloc_ldt = xen_alloc_ldt,
968 .free_ldt = xen_free_ldt,
970 .store_gdt = native_store_gdt,
971 .store_idt = native_store_idt,
972 .store_tr = xen_store_tr,
974 .write_ldt_entry = xen_write_ldt_entry,
975 .write_gdt_entry = xen_write_gdt_entry,
976 .write_idt_entry = xen_write_idt_entry,
977 .load_sp0 = xen_load_sp0,
979 .set_iopl_mask = xen_set_iopl_mask,
980 .io_delay = xen_io_delay,
982 /* Xen takes care of %gs when switching to usermode for us */
983 .swapgs = paravirt_nop,
985 .start_context_switch = paravirt_start_context_switch,
986 .end_context_switch = xen_end_context_switch,
989 static const struct pv_apic_ops xen_apic_ops __initdata = {
990 #ifdef CONFIG_X86_LOCAL_APIC
991 .startup_ipi_hook = paravirt_nop,
995 static void xen_reboot(int reason)
997 struct sched_shutdown r = { .reason = reason };
1003 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1007 static void xen_restart(char *msg)
1009 xen_reboot(SHUTDOWN_reboot);
1012 static void xen_emergency_restart(void)
1014 xen_reboot(SHUTDOWN_reboot);
1017 static void xen_machine_halt(void)
1019 xen_reboot(SHUTDOWN_poweroff);
1022 static void xen_crash_shutdown(struct pt_regs *regs)
1024 xen_reboot(SHUTDOWN_crash);
1027 static const struct machine_ops __initdata xen_machine_ops = {
1028 .restart = xen_restart,
1029 .halt = xen_machine_halt,
1030 .power_off = xen_machine_halt,
1031 .shutdown = xen_machine_halt,
1032 .crash_shutdown = xen_crash_shutdown,
1033 .emergency_restart = xen_emergency_restart,
1037 * Set up the GDT and segment registers for -fstack-protector. Until
1038 * we do this, we have to be careful not to call any stack-protected
1039 * function, which is most of the kernel.
1041 static void __init xen_setup_stackprotector(void)
1043 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1044 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1046 setup_stack_canary_segment(0);
1047 switch_to_new_gdt(0);
1049 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1050 pv_cpu_ops.load_gdt = xen_load_gdt;
1053 /* First C function to be called on Xen boot */
1054 asmlinkage void __init xen_start_kernel(void)
1058 if (!xen_start_info)
1061 xen_domain_type = XEN_PV_DOMAIN;
1063 /* Install Xen paravirt ops */
1065 pv_init_ops = xen_init_ops;
1066 pv_time_ops = xen_time_ops;
1067 pv_cpu_ops = xen_cpu_ops;
1068 pv_apic_ops = xen_apic_ops;
1070 x86_init.resources.memory_setup = xen_memory_setup;
1071 x86_init.oem.arch_setup = xen_arch_setup;
1072 x86_init.oem.banner = xen_banner;
1074 x86_init.timers.timer_init = xen_time_init;
1075 x86_init.timers.setup_percpu_clockev = x86_init_noop;
1076 x86_cpuinit.setup_percpu_clockev = x86_init_noop;
1078 x86_platform.calibrate_tsc = xen_tsc_khz;
1079 x86_platform.get_wallclock = xen_get_wallclock;
1080 x86_platform.set_wallclock = xen_set_wallclock;
1083 * Set up some pagetable state before starting to set any ptes.
1088 /* Prevent unwanted bits from being set in PTEs. */
1089 __supported_pte_mask &= ~_PAGE_GLOBAL;
1090 if (!xen_initial_domain())
1091 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1093 __supported_pte_mask |= _PAGE_IOMAP;
1095 /* Work out if we support NX */
1098 xen_setup_features();
1101 if (!xen_feature(XENFEAT_auto_translated_physmap))
1102 xen_build_dynamic_phys_to_machine();
1105 * Set up kernel GDT and segment registers, mainly so that
1106 * -fstack-protector code can be executed.
1108 xen_setup_stackprotector();
1111 xen_init_cpuid_mask();
1113 #ifdef CONFIG_X86_LOCAL_APIC
1115 * set up the basic apic ops.
1117 set_xen_basic_apic_ops();
1120 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1121 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1122 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1125 machine_ops = xen_machine_ops;
1128 * The only reliable way to retain the initial address of the
1129 * percpu gdt_page is to remember it here, so we can go and
1130 * mark it RW later, when the initial percpu area is freed.
1132 xen_initial_gdt = &per_cpu(gdt_page, 0);
1136 pgd = (pgd_t *)xen_start_info->pt_base;
1138 /* Don't do the full vcpu_info placement stuff until we have a
1139 possible map and a non-dummy shared_info. */
1140 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1142 local_irq_disable();
1143 early_boot_irqs_off();
1145 xen_raw_console_write("mapping kernel into physical memory\n");
1146 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1150 /* keep using Xen gdt for now; no urgent need to change it */
1152 pv_info.kernel_rpl = 1;
1153 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1154 pv_info.kernel_rpl = 0;
1156 /* set the limit of our address space */
1159 #ifdef CONFIG_X86_32
1160 /* set up basic CPUID stuff */
1161 cpu_detect(&new_cpu_data);
1162 new_cpu_data.hard_math = 1;
1163 new_cpu_data.wp_works_ok = 1;
1164 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1167 /* Poke various useful things into boot_params */
1168 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1169 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1170 ? __pa(xen_start_info->mod_start) : 0;
1171 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1172 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1174 if (!xen_initial_domain()) {
1175 add_preferred_console("xenboot", 0, NULL);
1176 add_preferred_console("tty", 0, NULL);
1177 add_preferred_console("hvc", 0, NULL);
1180 xen_raw_console_write("about to get started...\n");
1182 xen_setup_runstate_info(0);
1184 /* Start the world */
1185 #ifdef CONFIG_X86_32
1186 i386_start_kernel();
1188 x86_64_start_reservations((char *)__pa_symbol(&boot_params));