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>
33 #include <xen/interface/xen.h>
34 #include <xen/interface/version.h>
35 #include <xen/interface/physdev.h>
36 #include <xen/interface/vcpu.h>
37 #include <xen/features.h>
39 #include <xen/hvc-console.h>
41 #include <asm/paravirt.h>
44 #include <asm/xen/hypercall.h>
45 #include <asm/xen/hypervisor.h>
46 #include <asm/fixmap.h>
47 #include <asm/processor.h>
48 #include <asm/proto.h>
49 #include <asm/msr-index.h>
50 #include <asm/traps.h>
51 #include <asm/setup.h>
53 #include <asm/pgalloc.h>
54 #include <asm/pgtable.h>
55 #include <asm/tlbflush.h>
56 #include <asm/reboot.h>
57 #include <asm/stackprotector.h>
61 #include "multicalls.h"
63 EXPORT_SYMBOL_GPL(hypercall_page);
65 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
66 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
68 enum xen_domain_type xen_domain_type = XEN_NATIVE;
69 EXPORT_SYMBOL_GPL(xen_domain_type);
71 struct start_info *xen_start_info;
72 EXPORT_SYMBOL_GPL(xen_start_info);
74 struct shared_info xen_dummy_shared_info;
76 void *xen_initial_gdt;
79 * Point at some empty memory to start with. We map the real shared_info
80 * page as soon as fixmap is up and running.
82 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
85 * Flag to determine whether vcpu info placement is available on all
86 * VCPUs. We assume it is to start with, and then set it to zero on
87 * the first failure. This is because it can succeed on some VCPUs
88 * and not others, since it can involve hypervisor memory allocation,
89 * or because the guest failed to guarantee all the appropriate
90 * constraints on all VCPUs (ie buffer can't cross a page boundary).
92 * Note that any particular CPU may be using a placed vcpu structure,
93 * but we can only optimise if the all are.
95 * 0: not available, 1: available
97 static int have_vcpu_info_placement = 1;
99 static void xen_vcpu_setup(int cpu)
101 struct vcpu_register_vcpu_info info;
103 struct vcpu_info *vcpup;
105 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
106 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
108 if (!have_vcpu_info_placement)
109 return; /* already tested, not available */
111 vcpup = &per_cpu(xen_vcpu_info, cpu);
113 info.mfn = arbitrary_virt_to_mfn(vcpup);
114 info.offset = offset_in_page(vcpup);
116 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
117 cpu, vcpup, info.mfn, info.offset);
119 /* Check to see if the hypervisor will put the vcpu_info
120 structure where we want it, which allows direct access via
121 a percpu-variable. */
122 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
125 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
126 have_vcpu_info_placement = 0;
128 /* This cpu is using the registered vcpu info, even if
129 later ones fail to. */
130 per_cpu(xen_vcpu, cpu) = vcpup;
132 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
138 * On restore, set the vcpu placement up again.
139 * If it fails, then we're in a bad state, since
140 * we can't back out from using it...
142 void xen_vcpu_restore(void)
146 for_each_online_cpu(cpu) {
147 bool other_cpu = (cpu != smp_processor_id());
150 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
153 xen_setup_runstate_info(cpu);
155 if (have_vcpu_info_placement)
159 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
164 static void __init xen_banner(void)
166 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
167 struct xen_extraversion extra;
168 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
170 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
172 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
173 version >> 16, version & 0xffff, extra.extraversion,
174 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
177 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
178 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
180 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
181 unsigned int *cx, unsigned int *dx)
183 unsigned maskebx = ~0;
184 unsigned maskecx = ~0;
185 unsigned maskedx = ~0;
188 * Mask out inconvenient features, to try and disable as many
189 * unsupported kernel subsystems as possible.
193 maskecx = cpuid_leaf1_ecx_mask;
194 maskedx = cpuid_leaf1_edx_mask;
198 /* Suppress extended topology stuff */
203 asm(XEN_EMULATE_PREFIX "cpuid"
208 : "0" (*ax), "2" (*cx));
215 static __init void xen_init_cpuid_mask(void)
217 unsigned int ax, bx, cx, dx;
219 cpuid_leaf1_edx_mask =
220 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
221 (1 << X86_FEATURE_MCA) | /* disable MCA */
222 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
224 if (!xen_initial_domain())
225 cpuid_leaf1_edx_mask &=
226 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
227 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
231 xen_cpuid(&ax, &bx, &cx, &dx);
233 /* cpuid claims we support xsave; try enabling it to see what happens */
234 if (cx & (1 << (X86_FEATURE_XSAVE % 32))) {
237 set_in_cr4(X86_CR4_OSXSAVE);
241 if ((cr4 & X86_CR4_OSXSAVE) == 0)
242 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32));
244 clear_in_cr4(X86_CR4_OSXSAVE);
248 static void xen_set_debugreg(int reg, unsigned long val)
250 HYPERVISOR_set_debugreg(reg, val);
253 static unsigned long xen_get_debugreg(int reg)
255 return HYPERVISOR_get_debugreg(reg);
258 static void xen_end_context_switch(struct task_struct *next)
261 paravirt_end_context_switch(next);
264 static unsigned long xen_store_tr(void)
270 * Set the page permissions for a particular virtual address. If the
271 * address is a vmalloc mapping (or other non-linear mapping), then
272 * find the linear mapping of the page and also set its protections to
275 static void set_aliased_prot(void *v, pgprot_t prot)
283 ptep = lookup_address((unsigned long)v, &level);
284 BUG_ON(ptep == NULL);
286 pfn = pte_pfn(*ptep);
287 page = pfn_to_page(pfn);
289 pte = pfn_pte(pfn, prot);
291 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
294 if (!PageHighMem(page)) {
295 void *av = __va(PFN_PHYS(pfn));
298 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
304 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
306 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
309 for(i = 0; i < entries; i += entries_per_page)
310 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
313 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
315 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
318 for(i = 0; i < entries; i += entries_per_page)
319 set_aliased_prot(ldt + i, PAGE_KERNEL);
322 static void xen_set_ldt(const void *addr, unsigned entries)
324 struct mmuext_op *op;
325 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
328 op->cmd = MMUEXT_SET_LDT;
329 op->arg1.linear_addr = (unsigned long)addr;
330 op->arg2.nr_ents = entries;
332 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
334 xen_mc_issue(PARAVIRT_LAZY_CPU);
337 static void xen_load_gdt(const struct desc_ptr *dtr)
339 unsigned long va = dtr->address;
340 unsigned int size = dtr->size + 1;
341 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
342 unsigned long frames[pages];
346 * A GDT can be up to 64k in size, which corresponds to 8192
347 * 8-byte entries, or 16 4k pages..
350 BUG_ON(size > 65536);
351 BUG_ON(va & ~PAGE_MASK);
353 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
356 unsigned long pfn, mfn;
360 * The GDT is per-cpu and is in the percpu data area.
361 * That can be virtually mapped, so we need to do a
362 * page-walk to get the underlying MFN for the
363 * hypercall. The page can also be in the kernel's
364 * linear range, so we need to RO that mapping too.
366 ptep = lookup_address(va, &level);
367 BUG_ON(ptep == NULL);
369 pfn = pte_pfn(*ptep);
370 mfn = pfn_to_mfn(pfn);
371 virt = __va(PFN_PHYS(pfn));
375 make_lowmem_page_readonly((void *)va);
376 make_lowmem_page_readonly(virt);
379 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
384 * load_gdt for early boot, when the gdt is only mapped once
386 static __init void xen_load_gdt_boot(const struct desc_ptr *dtr)
388 unsigned long va = dtr->address;
389 unsigned int size = dtr->size + 1;
390 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
391 unsigned long frames[pages];
395 * A GDT can be up to 64k in size, which corresponds to 8192
396 * 8-byte entries, or 16 4k pages..
399 BUG_ON(size > 65536);
400 BUG_ON(va & ~PAGE_MASK);
402 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
404 unsigned long pfn, mfn;
406 pfn = virt_to_pfn(va);
407 mfn = pfn_to_mfn(pfn);
409 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
411 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
417 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
421 static void load_TLS_descriptor(struct thread_struct *t,
422 unsigned int cpu, unsigned int i)
424 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
425 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
426 struct multicall_space mc = __xen_mc_entry(0);
428 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
431 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
434 * XXX sleazy hack: If we're being called in a lazy-cpu zone
435 * and lazy gs handling is enabled, it means we're in a
436 * context switch, and %gs has just been saved. This means we
437 * can zero it out to prevent faults on exit from the
438 * hypervisor if the next process has no %gs. Either way, it
439 * has been saved, and the new value will get loaded properly.
440 * This will go away as soon as Xen has been modified to not
441 * save/restore %gs for normal hypercalls.
443 * On x86_64, this hack is not used for %gs, because gs points
444 * to KERNEL_GS_BASE (and uses it for PDA references), so we
445 * must not zero %gs on x86_64
447 * For x86_64, we need to zero %fs, otherwise we may get an
448 * exception between the new %fs descriptor being loaded and
449 * %fs being effectively cleared at __switch_to().
451 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
461 load_TLS_descriptor(t, cpu, 0);
462 load_TLS_descriptor(t, cpu, 1);
463 load_TLS_descriptor(t, cpu, 2);
465 xen_mc_issue(PARAVIRT_LAZY_CPU);
469 static void xen_load_gs_index(unsigned int idx)
471 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
476 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
479 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
480 u64 entry = *(u64 *)ptr;
485 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
491 static int cvt_gate_to_trap(int vector, const gate_desc *val,
492 struct trap_info *info)
496 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
499 info->vector = vector;
501 addr = gate_offset(*val);
504 * Look for known traps using IST, and substitute them
505 * appropriately. The debugger ones are the only ones we care
506 * about. Xen will handle faults like double_fault and
507 * machine_check, so we should never see them. Warn if
508 * there's an unexpected IST-using fault handler.
510 if (addr == (unsigned long)debug)
511 addr = (unsigned long)xen_debug;
512 else if (addr == (unsigned long)int3)
513 addr = (unsigned long)xen_int3;
514 else if (addr == (unsigned long)stack_segment)
515 addr = (unsigned long)xen_stack_segment;
516 else if (addr == (unsigned long)double_fault ||
517 addr == (unsigned long)nmi) {
518 /* Don't need to handle these */
520 #ifdef CONFIG_X86_MCE
521 } else if (addr == (unsigned long)machine_check) {
525 /* Some other trap using IST? */
526 if (WARN_ON(val->ist != 0))
529 #endif /* CONFIG_X86_64 */
530 info->address = addr;
532 info->cs = gate_segment(*val);
533 info->flags = val->dpl;
534 /* interrupt gates clear IF */
535 if (val->type == GATE_INTERRUPT)
536 info->flags |= 1 << 2;
541 /* Locations of each CPU's IDT */
542 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
544 /* Set an IDT entry. If the entry is part of the current IDT, then
546 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
548 unsigned long p = (unsigned long)&dt[entrynum];
549 unsigned long start, end;
553 start = __get_cpu_var(idt_desc).address;
554 end = start + __get_cpu_var(idt_desc).size + 1;
558 native_write_idt_entry(dt, entrynum, g);
560 if (p >= start && (p + 8) <= end) {
561 struct trap_info info[2];
565 if (cvt_gate_to_trap(entrynum, g, &info[0]))
566 if (HYPERVISOR_set_trap_table(info))
573 static void xen_convert_trap_info(const struct desc_ptr *desc,
574 struct trap_info *traps)
576 unsigned in, out, count;
578 count = (desc->size+1) / sizeof(gate_desc);
581 for (in = out = 0; in < count; in++) {
582 gate_desc *entry = (gate_desc*)(desc->address) + in;
584 if (cvt_gate_to_trap(in, entry, &traps[out]))
587 traps[out].address = 0;
590 void xen_copy_trap_info(struct trap_info *traps)
592 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
594 xen_convert_trap_info(desc, traps);
597 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
598 hold a spinlock to protect the static traps[] array (static because
599 it avoids allocation, and saves stack space). */
600 static void xen_load_idt(const struct desc_ptr *desc)
602 static DEFINE_SPINLOCK(lock);
603 static struct trap_info traps[257];
607 __get_cpu_var(idt_desc) = *desc;
609 xen_convert_trap_info(desc, traps);
612 if (HYPERVISOR_set_trap_table(traps))
618 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
619 they're handled differently. */
620 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
621 const void *desc, int type)
632 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
635 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
645 * Version of write_gdt_entry for use at early boot-time needed to
646 * update an entry as simply as possible.
648 static __init void xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
649 const void *desc, int type)
658 xmaddr_t maddr = virt_to_machine(&dt[entry]);
660 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
661 dt[entry] = *(struct desc_struct *)desc;
667 static void xen_load_sp0(struct tss_struct *tss,
668 struct thread_struct *thread)
670 struct multicall_space mcs = xen_mc_entry(0);
671 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
672 xen_mc_issue(PARAVIRT_LAZY_CPU);
675 static void xen_set_iopl_mask(unsigned mask)
677 struct physdev_set_iopl set_iopl;
679 /* Force the change at ring 0. */
680 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
681 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
684 static void xen_io_delay(void)
688 #ifdef CONFIG_X86_LOCAL_APIC
689 static u32 xen_apic_read(u32 reg)
694 static void xen_apic_write(u32 reg, u32 val)
696 /* Warn to see if there's any stray references */
700 static u64 xen_apic_icr_read(void)
705 static void xen_apic_icr_write(u32 low, u32 id)
707 /* Warn to see if there's any stray references */
711 static void xen_apic_wait_icr_idle(void)
716 static u32 xen_safe_apic_wait_icr_idle(void)
721 static void set_xen_basic_apic_ops(void)
723 apic->read = xen_apic_read;
724 apic->write = xen_apic_write;
725 apic->icr_read = xen_apic_icr_read;
726 apic->icr_write = xen_apic_icr_write;
727 apic->wait_icr_idle = xen_apic_wait_icr_idle;
728 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
734 static void xen_clts(void)
736 struct multicall_space mcs;
738 mcs = xen_mc_entry(0);
740 MULTI_fpu_taskswitch(mcs.mc, 0);
742 xen_mc_issue(PARAVIRT_LAZY_CPU);
745 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
747 static unsigned long xen_read_cr0(void)
749 unsigned long cr0 = percpu_read(xen_cr0_value);
751 if (unlikely(cr0 == 0)) {
752 cr0 = native_read_cr0();
753 percpu_write(xen_cr0_value, cr0);
759 static void xen_write_cr0(unsigned long cr0)
761 struct multicall_space mcs;
763 percpu_write(xen_cr0_value, cr0);
765 /* Only pay attention to cr0.TS; everything else is
767 mcs = xen_mc_entry(0);
769 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
771 xen_mc_issue(PARAVIRT_LAZY_CPU);
774 static void xen_write_cr4(unsigned long cr4)
779 native_write_cr4(cr4);
782 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
793 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
794 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
795 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
798 base = ((u64)high << 32) | low;
799 if (HYPERVISOR_set_segment_base(which, base) != 0)
807 case MSR_SYSCALL_MASK:
808 case MSR_IA32_SYSENTER_CS:
809 case MSR_IA32_SYSENTER_ESP:
810 case MSR_IA32_SYSENTER_EIP:
811 /* Fast syscall setup is all done in hypercalls, so
812 these are all ignored. Stub them out here to stop
813 Xen console noise. */
817 ret = native_write_msr_safe(msr, low, high);
823 void xen_setup_shared_info(void)
825 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
826 set_fixmap(FIX_PARAVIRT_BOOTMAP,
827 xen_start_info->shared_info);
829 HYPERVISOR_shared_info =
830 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
832 HYPERVISOR_shared_info =
833 (struct shared_info *)__va(xen_start_info->shared_info);
836 /* In UP this is as good a place as any to set up shared info */
837 xen_setup_vcpu_info_placement();
840 xen_setup_mfn_list_list();
843 /* This is called once we have the cpu_possible_map */
844 void xen_setup_vcpu_info_placement(void)
848 for_each_possible_cpu(cpu)
851 /* xen_vcpu_setup managed to place the vcpu_info within the
852 percpu area for all cpus, so make use of it */
853 if (have_vcpu_info_placement) {
854 printk(KERN_INFO "Xen: using vcpu_info placement\n");
856 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
857 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
858 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
859 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
860 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
864 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
865 unsigned long addr, unsigned len)
867 char *start, *end, *reloc;
870 start = end = reloc = NULL;
872 #define SITE(op, x) \
873 case PARAVIRT_PATCH(op.x): \
874 if (have_vcpu_info_placement) { \
875 start = (char *)xen_##x##_direct; \
876 end = xen_##x##_direct_end; \
877 reloc = xen_##x##_direct_reloc; \
882 SITE(pv_irq_ops, irq_enable);
883 SITE(pv_irq_ops, irq_disable);
884 SITE(pv_irq_ops, save_fl);
885 SITE(pv_irq_ops, restore_fl);
889 if (start == NULL || (end-start) > len)
892 ret = paravirt_patch_insns(insnbuf, len, start, end);
894 /* Note: because reloc is assigned from something that
895 appears to be an array, gcc assumes it's non-null,
896 but doesn't know its relationship with start and
898 if (reloc > start && reloc < end) {
899 int reloc_off = reloc - start;
900 long *relocp = (long *)(insnbuf + reloc_off);
901 long delta = start - (char *)addr;
909 ret = paravirt_patch_default(type, clobbers, insnbuf,
917 static const struct pv_info xen_info __initdata = {
918 .paravirt_enabled = 1,
919 .shared_kernel_pmd = 0,
924 static const struct pv_init_ops xen_init_ops __initdata = {
928 static const struct pv_time_ops xen_time_ops __initdata = {
929 .sched_clock = xen_sched_clock,
932 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
935 .set_debugreg = xen_set_debugreg,
936 .get_debugreg = xen_get_debugreg,
940 .read_cr0 = xen_read_cr0,
941 .write_cr0 = xen_write_cr0,
943 .read_cr4 = native_read_cr4,
944 .read_cr4_safe = native_read_cr4_safe,
945 .write_cr4 = xen_write_cr4,
947 .wbinvd = native_wbinvd,
949 .read_msr = native_read_msr_safe,
950 .write_msr = xen_write_msr_safe,
951 .read_tsc = native_read_tsc,
952 .read_pmc = native_read_pmc,
955 .irq_enable_sysexit = xen_sysexit,
957 .usergs_sysret32 = xen_sysret32,
958 .usergs_sysret64 = xen_sysret64,
961 .load_tr_desc = paravirt_nop,
962 .set_ldt = xen_set_ldt,
963 .load_gdt = xen_load_gdt,
964 .load_idt = xen_load_idt,
965 .load_tls = xen_load_tls,
967 .load_gs_index = xen_load_gs_index,
970 .alloc_ldt = xen_alloc_ldt,
971 .free_ldt = xen_free_ldt,
973 .store_gdt = native_store_gdt,
974 .store_idt = native_store_idt,
975 .store_tr = xen_store_tr,
977 .write_ldt_entry = xen_write_ldt_entry,
978 .write_gdt_entry = xen_write_gdt_entry,
979 .write_idt_entry = xen_write_idt_entry,
980 .load_sp0 = xen_load_sp0,
982 .set_iopl_mask = xen_set_iopl_mask,
983 .io_delay = xen_io_delay,
985 /* Xen takes care of %gs when switching to usermode for us */
986 .swapgs = paravirt_nop,
988 .start_context_switch = paravirt_start_context_switch,
989 .end_context_switch = xen_end_context_switch,
992 static const struct pv_apic_ops xen_apic_ops __initdata = {
993 #ifdef CONFIG_X86_LOCAL_APIC
994 .startup_ipi_hook = paravirt_nop,
998 static void xen_reboot(int reason)
1000 struct sched_shutdown r = { .reason = reason };
1006 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1010 static void xen_restart(char *msg)
1012 xen_reboot(SHUTDOWN_reboot);
1015 static void xen_emergency_restart(void)
1017 xen_reboot(SHUTDOWN_reboot);
1020 static void xen_machine_halt(void)
1022 xen_reboot(SHUTDOWN_poweroff);
1025 static void xen_crash_shutdown(struct pt_regs *regs)
1027 xen_reboot(SHUTDOWN_crash);
1030 static const struct machine_ops __initdata xen_machine_ops = {
1031 .restart = xen_restart,
1032 .halt = xen_machine_halt,
1033 .power_off = xen_machine_halt,
1034 .shutdown = xen_machine_halt,
1035 .crash_shutdown = xen_crash_shutdown,
1036 .emergency_restart = xen_emergency_restart,
1040 * Set up the GDT and segment registers for -fstack-protector. Until
1041 * we do this, we have to be careful not to call any stack-protected
1042 * function, which is most of the kernel.
1044 static void __init xen_setup_stackprotector(void)
1046 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1047 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1049 setup_stack_canary_segment(0);
1050 switch_to_new_gdt(0);
1052 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1053 pv_cpu_ops.load_gdt = xen_load_gdt;
1056 /* First C function to be called on Xen boot */
1057 asmlinkage void __init xen_start_kernel(void)
1061 if (!xen_start_info)
1064 xen_domain_type = XEN_PV_DOMAIN;
1066 /* Install Xen paravirt ops */
1068 pv_init_ops = xen_init_ops;
1069 pv_time_ops = xen_time_ops;
1070 pv_cpu_ops = xen_cpu_ops;
1071 pv_apic_ops = xen_apic_ops;
1073 x86_init.resources.memory_setup = xen_memory_setup;
1074 x86_init.oem.arch_setup = xen_arch_setup;
1075 x86_init.oem.banner = xen_banner;
1077 x86_init.timers.timer_init = xen_time_init;
1078 x86_init.timers.setup_percpu_clockev = x86_init_noop;
1079 x86_cpuinit.setup_percpu_clockev = x86_init_noop;
1081 x86_platform.calibrate_tsc = xen_tsc_khz;
1082 x86_platform.get_wallclock = xen_get_wallclock;
1083 x86_platform.set_wallclock = xen_set_wallclock;
1086 * Set up some pagetable state before starting to set any ptes.
1091 /* Prevent unwanted bits from being set in PTEs. */
1092 __supported_pte_mask &= ~_PAGE_GLOBAL;
1093 if (!xen_initial_domain())
1094 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1096 __supported_pte_mask |= _PAGE_IOMAP;
1099 * Prevent page tables from being allocated in highmem, even
1100 * if CONFIG_HIGHPTE is enabled.
1102 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1104 /* Work out if we support NX */
1107 xen_setup_features();
1110 if (!xen_feature(XENFEAT_auto_translated_physmap))
1111 xen_build_dynamic_phys_to_machine();
1114 * Set up kernel GDT and segment registers, mainly so that
1115 * -fstack-protector code can be executed.
1117 xen_setup_stackprotector();
1120 xen_init_cpuid_mask();
1122 #ifdef CONFIG_X86_LOCAL_APIC
1124 * set up the basic apic ops.
1126 set_xen_basic_apic_ops();
1129 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1130 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1131 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1134 machine_ops = xen_machine_ops;
1137 * The only reliable way to retain the initial address of the
1138 * percpu gdt_page is to remember it here, so we can go and
1139 * mark it RW later, when the initial percpu area is freed.
1141 xen_initial_gdt = &per_cpu(gdt_page, 0);
1145 pgd = (pgd_t *)xen_start_info->pt_base;
1147 /* Don't do the full vcpu_info placement stuff until we have a
1148 possible map and a non-dummy shared_info. */
1149 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1151 local_irq_disable();
1152 early_boot_irqs_off();
1154 xen_raw_console_write("mapping kernel into physical memory\n");
1155 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1159 /* keep using Xen gdt for now; no urgent need to change it */
1161 #ifdef CONFIG_X86_32
1162 pv_info.kernel_rpl = 1;
1163 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1164 pv_info.kernel_rpl = 0;
1166 pv_info.kernel_rpl = 0;
1169 /* set the limit of our address space */
1172 #ifdef CONFIG_X86_32
1173 /* set up basic CPUID stuff */
1174 cpu_detect(&new_cpu_data);
1175 new_cpu_data.hard_math = 1;
1176 new_cpu_data.wp_works_ok = 1;
1177 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1180 /* Poke various useful things into boot_params */
1181 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1182 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1183 ? __pa(xen_start_info->mod_start) : 0;
1184 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1185 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1187 if (!xen_initial_domain()) {
1188 add_preferred_console("xenboot", 0, NULL);
1189 add_preferred_console("tty", 0, NULL);
1190 add_preferred_console("hvc", 0, NULL);
1192 /* Make sure ACS will be enabled */
1197 xen_raw_console_write("about to get started...\n");
1199 xen_setup_runstate_info(0);
1201 /* Start the world */
1202 #ifdef CONFIG_X86_32
1203 i386_start_kernel();
1205 x86_64_start_reservations((char *)__pa_symbol(&boot_params));