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>
30 #include <linux/pci.h>
31 #include <linux/gfp.h>
34 #include <xen/interface/xen.h>
35 #include <xen/interface/version.h>
36 #include <xen/interface/physdev.h>
37 #include <xen/interface/vcpu.h>
38 #include <xen/features.h>
40 #include <xen/hvc-console.h>
42 #include <asm/paravirt.h>
45 #include <asm/xen/hypercall.h>
46 #include <asm/xen/hypervisor.h>
47 #include <asm/fixmap.h>
48 #include <asm/processor.h>
49 #include <asm/proto.h>
50 #include <asm/msr-index.h>
51 #include <asm/traps.h>
52 #include <asm/setup.h>
54 #include <asm/pgalloc.h>
55 #include <asm/pgtable.h>
56 #include <asm/tlbflush.h>
57 #include <asm/reboot.h>
58 #include <asm/stackprotector.h>
62 #include "multicalls.h"
64 EXPORT_SYMBOL_GPL(hypercall_page);
66 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
67 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
69 enum xen_domain_type xen_domain_type = XEN_NATIVE;
70 EXPORT_SYMBOL_GPL(xen_domain_type);
72 struct start_info *xen_start_info;
73 EXPORT_SYMBOL_GPL(xen_start_info);
75 struct shared_info xen_dummy_shared_info;
77 void *xen_initial_gdt;
80 * Point at some empty memory to start with. We map the real shared_info
81 * page as soon as fixmap is up and running.
83 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
86 * Flag to determine whether vcpu info placement is available on all
87 * VCPUs. We assume it is to start with, and then set it to zero on
88 * the first failure. This is because it can succeed on some VCPUs
89 * and not others, since it can involve hypervisor memory allocation,
90 * or because the guest failed to guarantee all the appropriate
91 * constraints on all VCPUs (ie buffer can't cross a page boundary).
93 * Note that any particular CPU may be using a placed vcpu structure,
94 * but we can only optimise if the all are.
96 * 0: not available, 1: available
98 static int have_vcpu_info_placement = 1;
100 static void clamp_max_cpus(void)
103 if (setup_max_cpus > MAX_VIRT_CPUS)
104 setup_max_cpus = MAX_VIRT_CPUS;
108 static void xen_vcpu_setup(int cpu)
110 struct vcpu_register_vcpu_info info;
112 struct vcpu_info *vcpup;
114 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
116 if (cpu < MAX_VIRT_CPUS)
117 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
119 if (!have_vcpu_info_placement) {
120 if (cpu >= MAX_VIRT_CPUS)
125 vcpup = &per_cpu(xen_vcpu_info, cpu);
126 info.mfn = arbitrary_virt_to_mfn(vcpup);
127 info.offset = offset_in_page(vcpup);
129 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
130 cpu, vcpup, info.mfn, info.offset);
132 /* Check to see if the hypervisor will put the vcpu_info
133 structure where we want it, which allows direct access via
134 a percpu-variable. */
135 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
138 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
139 have_vcpu_info_placement = 0;
142 /* This cpu is using the registered vcpu info, even if
143 later ones fail to. */
144 per_cpu(xen_vcpu, cpu) = vcpup;
146 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
152 * On restore, set the vcpu placement up again.
153 * If it fails, then we're in a bad state, since
154 * we can't back out from using it...
156 void xen_vcpu_restore(void)
160 for_each_online_cpu(cpu) {
161 bool other_cpu = (cpu != smp_processor_id());
164 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
167 xen_setup_runstate_info(cpu);
169 if (have_vcpu_info_placement)
173 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
178 static void __init xen_banner(void)
180 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
181 struct xen_extraversion extra;
182 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
184 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
186 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
187 version >> 16, version & 0xffff, extra.extraversion,
188 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
191 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
192 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
194 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
195 unsigned int *cx, unsigned int *dx)
197 unsigned maskebx = ~0;
198 unsigned maskecx = ~0;
199 unsigned maskedx = ~0;
202 * Mask out inconvenient features, to try and disable as many
203 * unsupported kernel subsystems as possible.
207 maskecx = cpuid_leaf1_ecx_mask;
208 maskedx = cpuid_leaf1_edx_mask;
212 /* Suppress extended topology stuff */
217 asm(XEN_EMULATE_PREFIX "cpuid"
222 : "0" (*ax), "2" (*cx));
229 static __init void xen_init_cpuid_mask(void)
231 unsigned int ax, bx, cx, dx;
233 cpuid_leaf1_edx_mask =
234 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
235 (1 << X86_FEATURE_MCA) | /* disable MCA */
236 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
238 if (!xen_initial_domain())
239 cpuid_leaf1_edx_mask &=
240 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
241 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
245 xen_cpuid(&ax, &bx, &cx, &dx);
247 /* cpuid claims we support xsave; try enabling it to see what happens */
248 if (cx & (1 << (X86_FEATURE_XSAVE % 32))) {
251 set_in_cr4(X86_CR4_OSXSAVE);
255 if ((cr4 & X86_CR4_OSXSAVE) == 0)
256 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32));
258 clear_in_cr4(X86_CR4_OSXSAVE);
262 static void xen_set_debugreg(int reg, unsigned long val)
264 HYPERVISOR_set_debugreg(reg, val);
267 static unsigned long xen_get_debugreg(int reg)
269 return HYPERVISOR_get_debugreg(reg);
272 static void xen_end_context_switch(struct task_struct *next)
275 paravirt_end_context_switch(next);
278 static unsigned long xen_store_tr(void)
284 * Set the page permissions for a particular virtual address. If the
285 * address is a vmalloc mapping (or other non-linear mapping), then
286 * find the linear mapping of the page and also set its protections to
289 static void set_aliased_prot(void *v, pgprot_t prot)
297 ptep = lookup_address((unsigned long)v, &level);
298 BUG_ON(ptep == NULL);
300 pfn = pte_pfn(*ptep);
301 page = pfn_to_page(pfn);
303 pte = pfn_pte(pfn, prot);
305 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
308 if (!PageHighMem(page)) {
309 void *av = __va(PFN_PHYS(pfn));
312 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
318 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
320 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
323 for(i = 0; i < entries; i += entries_per_page)
324 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
327 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
329 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
332 for(i = 0; i < entries; i += entries_per_page)
333 set_aliased_prot(ldt + i, PAGE_KERNEL);
336 static void xen_set_ldt(const void *addr, unsigned entries)
338 struct mmuext_op *op;
339 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
342 op->cmd = MMUEXT_SET_LDT;
343 op->arg1.linear_addr = (unsigned long)addr;
344 op->arg2.nr_ents = entries;
346 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
348 xen_mc_issue(PARAVIRT_LAZY_CPU);
351 static void xen_load_gdt(const struct desc_ptr *dtr)
353 unsigned long va = dtr->address;
354 unsigned int size = dtr->size + 1;
355 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
356 unsigned long frames[pages];
360 * A GDT can be up to 64k in size, which corresponds to 8192
361 * 8-byte entries, or 16 4k pages..
364 BUG_ON(size > 65536);
365 BUG_ON(va & ~PAGE_MASK);
367 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
370 unsigned long pfn, mfn;
374 * The GDT is per-cpu and is in the percpu data area.
375 * That can be virtually mapped, so we need to do a
376 * page-walk to get the underlying MFN for the
377 * hypercall. The page can also be in the kernel's
378 * linear range, so we need to RO that mapping too.
380 ptep = lookup_address(va, &level);
381 BUG_ON(ptep == NULL);
383 pfn = pte_pfn(*ptep);
384 mfn = pfn_to_mfn(pfn);
385 virt = __va(PFN_PHYS(pfn));
389 make_lowmem_page_readonly((void *)va);
390 make_lowmem_page_readonly(virt);
393 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
398 * load_gdt for early boot, when the gdt is only mapped once
400 static __init void xen_load_gdt_boot(const struct desc_ptr *dtr)
402 unsigned long va = dtr->address;
403 unsigned int size = dtr->size + 1;
404 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
405 unsigned long frames[pages];
409 * A GDT can be up to 64k in size, which corresponds to 8192
410 * 8-byte entries, or 16 4k pages..
413 BUG_ON(size > 65536);
414 BUG_ON(va & ~PAGE_MASK);
416 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
418 unsigned long pfn, mfn;
420 pfn = virt_to_pfn(va);
421 mfn = pfn_to_mfn(pfn);
423 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
425 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
431 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
435 static void load_TLS_descriptor(struct thread_struct *t,
436 unsigned int cpu, unsigned int i)
438 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
439 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
440 struct multicall_space mc = __xen_mc_entry(0);
442 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
445 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
448 * XXX sleazy hack: If we're being called in a lazy-cpu zone
449 * and lazy gs handling is enabled, it means we're in a
450 * context switch, and %gs has just been saved. This means we
451 * can zero it out to prevent faults on exit from the
452 * hypervisor if the next process has no %gs. Either way, it
453 * has been saved, and the new value will get loaded properly.
454 * This will go away as soon as Xen has been modified to not
455 * save/restore %gs for normal hypercalls.
457 * On x86_64, this hack is not used for %gs, because gs points
458 * to KERNEL_GS_BASE (and uses it for PDA references), so we
459 * must not zero %gs on x86_64
461 * For x86_64, we need to zero %fs, otherwise we may get an
462 * exception between the new %fs descriptor being loaded and
463 * %fs being effectively cleared at __switch_to().
465 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
475 load_TLS_descriptor(t, cpu, 0);
476 load_TLS_descriptor(t, cpu, 1);
477 load_TLS_descriptor(t, cpu, 2);
479 xen_mc_issue(PARAVIRT_LAZY_CPU);
483 static void xen_load_gs_index(unsigned int idx)
485 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
490 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
493 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
494 u64 entry = *(u64 *)ptr;
499 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
505 static int cvt_gate_to_trap(int vector, const gate_desc *val,
506 struct trap_info *info)
510 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
513 info->vector = vector;
515 addr = gate_offset(*val);
518 * Look for known traps using IST, and substitute them
519 * appropriately. The debugger ones are the only ones we care
520 * about. Xen will handle faults like double_fault and
521 * machine_check, so we should never see them. Warn if
522 * there's an unexpected IST-using fault handler.
524 if (addr == (unsigned long)debug)
525 addr = (unsigned long)xen_debug;
526 else if (addr == (unsigned long)int3)
527 addr = (unsigned long)xen_int3;
528 else if (addr == (unsigned long)stack_segment)
529 addr = (unsigned long)xen_stack_segment;
530 else if (addr == (unsigned long)double_fault ||
531 addr == (unsigned long)nmi) {
532 /* Don't need to handle these */
534 #ifdef CONFIG_X86_MCE
535 } else if (addr == (unsigned long)machine_check) {
539 /* Some other trap using IST? */
540 if (WARN_ON(val->ist != 0))
543 #endif /* CONFIG_X86_64 */
544 info->address = addr;
546 info->cs = gate_segment(*val);
547 info->flags = val->dpl;
548 /* interrupt gates clear IF */
549 if (val->type == GATE_INTERRUPT)
550 info->flags |= 1 << 2;
555 /* Locations of each CPU's IDT */
556 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
558 /* Set an IDT entry. If the entry is part of the current IDT, then
560 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
562 unsigned long p = (unsigned long)&dt[entrynum];
563 unsigned long start, end;
567 start = __get_cpu_var(idt_desc).address;
568 end = start + __get_cpu_var(idt_desc).size + 1;
572 native_write_idt_entry(dt, entrynum, g);
574 if (p >= start && (p + 8) <= end) {
575 struct trap_info info[2];
579 if (cvt_gate_to_trap(entrynum, g, &info[0]))
580 if (HYPERVISOR_set_trap_table(info))
587 static void xen_convert_trap_info(const struct desc_ptr *desc,
588 struct trap_info *traps)
590 unsigned in, out, count;
592 count = (desc->size+1) / sizeof(gate_desc);
595 for (in = out = 0; in < count; in++) {
596 gate_desc *entry = (gate_desc*)(desc->address) + in;
598 if (cvt_gate_to_trap(in, entry, &traps[out]))
601 traps[out].address = 0;
604 void xen_copy_trap_info(struct trap_info *traps)
606 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
608 xen_convert_trap_info(desc, traps);
611 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
612 hold a spinlock to protect the static traps[] array (static because
613 it avoids allocation, and saves stack space). */
614 static void xen_load_idt(const struct desc_ptr *desc)
616 static DEFINE_SPINLOCK(lock);
617 static struct trap_info traps[257];
621 __get_cpu_var(idt_desc) = *desc;
623 xen_convert_trap_info(desc, traps);
626 if (HYPERVISOR_set_trap_table(traps))
632 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
633 they're handled differently. */
634 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
635 const void *desc, int type)
646 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
649 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
659 * Version of write_gdt_entry for use at early boot-time needed to
660 * update an entry as simply as possible.
662 static __init void xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
663 const void *desc, int type)
672 xmaddr_t maddr = virt_to_machine(&dt[entry]);
674 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
675 dt[entry] = *(struct desc_struct *)desc;
681 static void xen_load_sp0(struct tss_struct *tss,
682 struct thread_struct *thread)
684 struct multicall_space mcs = xen_mc_entry(0);
685 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
686 xen_mc_issue(PARAVIRT_LAZY_CPU);
689 static void xen_set_iopl_mask(unsigned mask)
691 struct physdev_set_iopl set_iopl;
693 /* Force the change at ring 0. */
694 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
695 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
698 static void xen_io_delay(void)
702 #ifdef CONFIG_X86_LOCAL_APIC
703 static u32 xen_apic_read(u32 reg)
708 static void xen_apic_write(u32 reg, u32 val)
710 /* Warn to see if there's any stray references */
714 static u64 xen_apic_icr_read(void)
719 static void xen_apic_icr_write(u32 low, u32 id)
721 /* Warn to see if there's any stray references */
725 static void xen_apic_wait_icr_idle(void)
730 static u32 xen_safe_apic_wait_icr_idle(void)
735 static void set_xen_basic_apic_ops(void)
737 apic->read = xen_apic_read;
738 apic->write = xen_apic_write;
739 apic->icr_read = xen_apic_icr_read;
740 apic->icr_write = xen_apic_icr_write;
741 apic->wait_icr_idle = xen_apic_wait_icr_idle;
742 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
747 static void xen_clts(void)
749 struct multicall_space mcs;
751 mcs = xen_mc_entry(0);
753 MULTI_fpu_taskswitch(mcs.mc, 0);
755 xen_mc_issue(PARAVIRT_LAZY_CPU);
758 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
760 static unsigned long xen_read_cr0(void)
762 unsigned long cr0 = percpu_read(xen_cr0_value);
764 if (unlikely(cr0 == 0)) {
765 cr0 = native_read_cr0();
766 percpu_write(xen_cr0_value, cr0);
772 static void xen_write_cr0(unsigned long cr0)
774 struct multicall_space mcs;
776 percpu_write(xen_cr0_value, cr0);
778 /* Only pay attention to cr0.TS; everything else is
780 mcs = xen_mc_entry(0);
782 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
784 xen_mc_issue(PARAVIRT_LAZY_CPU);
787 static void xen_write_cr4(unsigned long cr4)
792 native_write_cr4(cr4);
795 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
806 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
807 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
808 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
811 base = ((u64)high << 32) | low;
812 if (HYPERVISOR_set_segment_base(which, base) != 0)
820 case MSR_SYSCALL_MASK:
821 case MSR_IA32_SYSENTER_CS:
822 case MSR_IA32_SYSENTER_ESP:
823 case MSR_IA32_SYSENTER_EIP:
824 /* Fast syscall setup is all done in hypercalls, so
825 these are all ignored. Stub them out here to stop
826 Xen console noise. */
830 ret = native_write_msr_safe(msr, low, high);
836 void xen_setup_shared_info(void)
838 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
839 set_fixmap(FIX_PARAVIRT_BOOTMAP,
840 xen_start_info->shared_info);
842 HYPERVISOR_shared_info =
843 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
845 HYPERVISOR_shared_info =
846 (struct shared_info *)__va(xen_start_info->shared_info);
849 /* In UP this is as good a place as any to set up shared info */
850 xen_setup_vcpu_info_placement();
853 xen_setup_mfn_list_list();
856 /* This is called once we have the cpu_possible_map */
857 void xen_setup_vcpu_info_placement(void)
861 for_each_possible_cpu(cpu)
864 /* xen_vcpu_setup managed to place the vcpu_info within the
865 percpu area for all cpus, so make use of it */
866 if (have_vcpu_info_placement) {
867 printk(KERN_INFO "Xen: using vcpu_info placement\n");
869 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
870 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
871 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
872 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
873 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
877 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
878 unsigned long addr, unsigned len)
880 char *start, *end, *reloc;
883 start = end = reloc = NULL;
885 #define SITE(op, x) \
886 case PARAVIRT_PATCH(op.x): \
887 if (have_vcpu_info_placement) { \
888 start = (char *)xen_##x##_direct; \
889 end = xen_##x##_direct_end; \
890 reloc = xen_##x##_direct_reloc; \
895 SITE(pv_irq_ops, irq_enable);
896 SITE(pv_irq_ops, irq_disable);
897 SITE(pv_irq_ops, save_fl);
898 SITE(pv_irq_ops, restore_fl);
902 if (start == NULL || (end-start) > len)
905 ret = paravirt_patch_insns(insnbuf, len, start, end);
907 /* Note: because reloc is assigned from something that
908 appears to be an array, gcc assumes it's non-null,
909 but doesn't know its relationship with start and
911 if (reloc > start && reloc < end) {
912 int reloc_off = reloc - start;
913 long *relocp = (long *)(insnbuf + reloc_off);
914 long delta = start - (char *)addr;
922 ret = paravirt_patch_default(type, clobbers, insnbuf,
930 static const struct pv_info xen_info __initdata = {
931 .paravirt_enabled = 1,
932 .shared_kernel_pmd = 0,
937 static const struct pv_init_ops xen_init_ops __initdata = {
941 static const struct pv_time_ops xen_time_ops __initdata = {
942 .sched_clock = xen_clocksource_read,
945 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
948 .set_debugreg = xen_set_debugreg,
949 .get_debugreg = xen_get_debugreg,
953 .read_cr0 = xen_read_cr0,
954 .write_cr0 = xen_write_cr0,
956 .read_cr4 = native_read_cr4,
957 .read_cr4_safe = native_read_cr4_safe,
958 .write_cr4 = xen_write_cr4,
960 .wbinvd = native_wbinvd,
962 .read_msr = native_read_msr_safe,
963 .write_msr = xen_write_msr_safe,
964 .read_tsc = native_read_tsc,
965 .read_pmc = native_read_pmc,
968 .irq_enable_sysexit = xen_sysexit,
970 .usergs_sysret32 = xen_sysret32,
971 .usergs_sysret64 = xen_sysret64,
974 .load_tr_desc = paravirt_nop,
975 .set_ldt = xen_set_ldt,
976 .load_gdt = xen_load_gdt,
977 .load_idt = xen_load_idt,
978 .load_tls = xen_load_tls,
980 .load_gs_index = xen_load_gs_index,
983 .alloc_ldt = xen_alloc_ldt,
984 .free_ldt = xen_free_ldt,
986 .store_gdt = native_store_gdt,
987 .store_idt = native_store_idt,
988 .store_tr = xen_store_tr,
990 .write_ldt_entry = xen_write_ldt_entry,
991 .write_gdt_entry = xen_write_gdt_entry,
992 .write_idt_entry = xen_write_idt_entry,
993 .load_sp0 = xen_load_sp0,
995 .set_iopl_mask = xen_set_iopl_mask,
996 .io_delay = xen_io_delay,
998 /* Xen takes care of %gs when switching to usermode for us */
999 .swapgs = paravirt_nop,
1001 .start_context_switch = paravirt_start_context_switch,
1002 .end_context_switch = xen_end_context_switch,
1005 static const struct pv_apic_ops xen_apic_ops __initdata = {
1006 #ifdef CONFIG_X86_LOCAL_APIC
1007 .startup_ipi_hook = paravirt_nop,
1011 static void xen_reboot(int reason)
1013 struct sched_shutdown r = { .reason = reason };
1019 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1023 static void xen_restart(char *msg)
1025 xen_reboot(SHUTDOWN_reboot);
1028 static void xen_emergency_restart(void)
1030 xen_reboot(SHUTDOWN_reboot);
1033 static void xen_machine_halt(void)
1035 xen_reboot(SHUTDOWN_poweroff);
1038 static void xen_crash_shutdown(struct pt_regs *regs)
1040 xen_reboot(SHUTDOWN_crash);
1044 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1046 xen_reboot(SHUTDOWN_crash);
1050 static struct notifier_block xen_panic_block = {
1051 .notifier_call= xen_panic_event,
1054 int xen_panic_handler_init(void)
1056 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1060 static const struct machine_ops __initdata xen_machine_ops = {
1061 .restart = xen_restart,
1062 .halt = xen_machine_halt,
1063 .power_off = xen_machine_halt,
1064 .shutdown = xen_machine_halt,
1065 .crash_shutdown = xen_crash_shutdown,
1066 .emergency_restart = xen_emergency_restart,
1070 * Set up the GDT and segment registers for -fstack-protector. Until
1071 * we do this, we have to be careful not to call any stack-protected
1072 * function, which is most of the kernel.
1074 static void __init xen_setup_stackprotector(void)
1076 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1077 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1079 setup_stack_canary_segment(0);
1080 switch_to_new_gdt(0);
1082 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1083 pv_cpu_ops.load_gdt = xen_load_gdt;
1086 /* First C function to be called on Xen boot */
1087 asmlinkage void __init xen_start_kernel(void)
1091 if (!xen_start_info)
1094 xen_domain_type = XEN_PV_DOMAIN;
1096 /* Install Xen paravirt ops */
1098 pv_init_ops = xen_init_ops;
1099 pv_time_ops = xen_time_ops;
1100 pv_cpu_ops = xen_cpu_ops;
1101 pv_apic_ops = xen_apic_ops;
1103 x86_init.resources.memory_setup = xen_memory_setup;
1104 x86_init.oem.arch_setup = xen_arch_setup;
1105 x86_init.oem.banner = xen_banner;
1107 x86_init.timers.timer_init = xen_time_init;
1108 x86_init.timers.setup_percpu_clockev = x86_init_noop;
1109 x86_cpuinit.setup_percpu_clockev = x86_init_noop;
1111 x86_platform.calibrate_tsc = xen_tsc_khz;
1112 x86_platform.get_wallclock = xen_get_wallclock;
1113 x86_platform.set_wallclock = xen_set_wallclock;
1116 * Set up some pagetable state before starting to set any ptes.
1121 /* Prevent unwanted bits from being set in PTEs. */
1122 __supported_pte_mask &= ~_PAGE_GLOBAL;
1123 if (!xen_initial_domain())
1124 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1126 __supported_pte_mask |= _PAGE_IOMAP;
1129 * Prevent page tables from being allocated in highmem, even
1130 * if CONFIG_HIGHPTE is enabled.
1132 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1134 /* Work out if we support NX */
1137 xen_setup_features();
1140 if (!xen_feature(XENFEAT_auto_translated_physmap))
1141 xen_build_dynamic_phys_to_machine();
1144 * Set up kernel GDT and segment registers, mainly so that
1145 * -fstack-protector code can be executed.
1147 xen_setup_stackprotector();
1150 xen_init_cpuid_mask();
1152 #ifdef CONFIG_X86_LOCAL_APIC
1154 * set up the basic apic ops.
1156 set_xen_basic_apic_ops();
1159 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1160 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1161 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1164 machine_ops = xen_machine_ops;
1167 * The only reliable way to retain the initial address of the
1168 * percpu gdt_page is to remember it here, so we can go and
1169 * mark it RW later, when the initial percpu area is freed.
1171 xen_initial_gdt = &per_cpu(gdt_page, 0);
1175 pgd = (pgd_t *)xen_start_info->pt_base;
1177 /* Don't do the full vcpu_info placement stuff until we have a
1178 possible map and a non-dummy shared_info. */
1179 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1181 local_irq_disable();
1182 early_boot_irqs_off();
1184 xen_raw_console_write("mapping kernel into physical memory\n");
1185 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1189 /* keep using Xen gdt for now; no urgent need to change it */
1191 #ifdef CONFIG_X86_32
1192 pv_info.kernel_rpl = 1;
1193 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1194 pv_info.kernel_rpl = 0;
1196 pv_info.kernel_rpl = 0;
1199 /* set the limit of our address space */
1202 #ifdef CONFIG_X86_32
1203 /* set up basic CPUID stuff */
1204 cpu_detect(&new_cpu_data);
1205 new_cpu_data.hard_math = 1;
1206 new_cpu_data.wp_works_ok = 1;
1207 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1210 /* Poke various useful things into boot_params */
1211 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1212 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1213 ? __pa(xen_start_info->mod_start) : 0;
1214 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1215 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1217 if (!xen_initial_domain()) {
1218 add_preferred_console("xenboot", 0, NULL);
1219 add_preferred_console("tty", 0, NULL);
1220 add_preferred_console("hvc", 0, NULL);
1222 /* Make sure ACS will be enabled */
1227 xen_raw_console_write("about to get started...\n");
1229 xen_setup_runstate_info(0);
1231 /* Start the world */
1232 #ifdef CONFIG_X86_32
1233 i386_start_kernel();
1235 x86_64_start_reservations((char *)__pa_symbol(&boot_params));