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/cpu.h>
15 #include <linux/kernel.h>
16 #include <linux/init.h>
17 #include <linux/smp.h>
18 #include <linux/preempt.h>
19 #include <linux/hardirq.h>
20 #include <linux/percpu.h>
21 #include <linux/delay.h>
22 #include <linux/start_kernel.h>
23 #include <linux/sched.h>
24 #include <linux/kprobes.h>
25 #include <linux/bootmem.h>
26 #include <linux/module.h>
28 #include <linux/page-flags.h>
29 #include <linux/highmem.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/memblock.h>
36 #include <xen/interface/xen.h>
37 #include <xen/interface/version.h>
38 #include <xen/interface/physdev.h>
39 #include <xen/interface/vcpu.h>
40 #include <xen/interface/memory.h>
41 #include <xen/features.h>
44 #include <xen/hvc-console.h>
46 #include <asm/paravirt.h>
49 #include <asm/xen/pci.h>
50 #include <asm/xen/hypercall.h>
51 #include <asm/xen/hypervisor.h>
52 #include <asm/fixmap.h>
53 #include <asm/processor.h>
54 #include <asm/proto.h>
55 #include <asm/msr-index.h>
56 #include <asm/traps.h>
57 #include <asm/setup.h>
59 #include <asm/pgalloc.h>
60 #include <asm/pgtable.h>
61 #include <asm/tlbflush.h>
62 #include <asm/reboot.h>
63 #include <asm/stackprotector.h>
64 #include <asm/hypervisor.h>
68 #include "multicalls.h"
70 EXPORT_SYMBOL_GPL(hypercall_page);
72 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
73 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
75 enum xen_domain_type xen_domain_type = XEN_NATIVE;
76 EXPORT_SYMBOL_GPL(xen_domain_type);
78 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
79 EXPORT_SYMBOL(machine_to_phys_mapping);
80 unsigned int machine_to_phys_order;
81 EXPORT_SYMBOL(machine_to_phys_order);
83 struct start_info *xen_start_info;
84 EXPORT_SYMBOL_GPL(xen_start_info);
86 struct shared_info xen_dummy_shared_info;
88 void *xen_initial_gdt;
90 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
91 __read_mostly int xen_have_vector_callback;
92 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
95 * Point at some empty memory to start with. We map the real shared_info
96 * page as soon as fixmap is up and running.
98 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
101 * Flag to determine whether vcpu info placement is available on all
102 * VCPUs. We assume it is to start with, and then set it to zero on
103 * the first failure. This is because it can succeed on some VCPUs
104 * and not others, since it can involve hypervisor memory allocation,
105 * or because the guest failed to guarantee all the appropriate
106 * constraints on all VCPUs (ie buffer can't cross a page boundary).
108 * Note that any particular CPU may be using a placed vcpu structure,
109 * but we can only optimise if the all are.
111 * 0: not available, 1: available
113 static int have_vcpu_info_placement = 1;
115 static void clamp_max_cpus(void)
118 if (setup_max_cpus > MAX_VIRT_CPUS)
119 setup_max_cpus = MAX_VIRT_CPUS;
123 static void xen_vcpu_setup(int cpu)
125 struct vcpu_register_vcpu_info info;
127 struct vcpu_info *vcpup;
129 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
131 if (cpu < MAX_VIRT_CPUS)
132 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
134 if (!have_vcpu_info_placement) {
135 if (cpu >= MAX_VIRT_CPUS)
140 vcpup = &per_cpu(xen_vcpu_info, cpu);
141 info.mfn = arbitrary_virt_to_mfn(vcpup);
142 info.offset = offset_in_page(vcpup);
144 /* Check to see if the hypervisor will put the vcpu_info
145 structure where we want it, which allows direct access via
146 a percpu-variable. */
147 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
150 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
151 have_vcpu_info_placement = 0;
154 /* This cpu is using the registered vcpu info, even if
155 later ones fail to. */
156 per_cpu(xen_vcpu, cpu) = vcpup;
161 * On restore, set the vcpu placement up again.
162 * If it fails, then we're in a bad state, since
163 * we can't back out from using it...
165 void xen_vcpu_restore(void)
169 for_each_online_cpu(cpu) {
170 bool other_cpu = (cpu != smp_processor_id());
173 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
176 xen_setup_runstate_info(cpu);
178 if (have_vcpu_info_placement)
182 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
187 static void __init xen_banner(void)
189 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
190 struct xen_extraversion extra;
191 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
193 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
195 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
196 version >> 16, version & 0xffff, extra.extraversion,
197 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
200 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
201 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
203 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
204 unsigned int *cx, unsigned int *dx)
206 unsigned maskebx = ~0;
207 unsigned maskecx = ~0;
208 unsigned maskedx = ~0;
211 * Mask out inconvenient features, to try and disable as many
212 * unsupported kernel subsystems as possible.
216 maskecx = cpuid_leaf1_ecx_mask;
217 maskedx = cpuid_leaf1_edx_mask;
221 /* Suppress extended topology stuff */
226 asm(XEN_EMULATE_PREFIX "cpuid"
231 : "0" (*ax), "2" (*cx));
238 static __init void xen_init_cpuid_mask(void)
240 unsigned int ax, bx, cx, dx;
242 cpuid_leaf1_edx_mask =
243 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
244 (1 << X86_FEATURE_MCA) | /* disable MCA */
245 (1 << X86_FEATURE_MTRR) | /* disable MTRR */
246 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
248 if (!xen_initial_domain())
249 cpuid_leaf1_edx_mask &=
250 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
251 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
253 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32)); /* disable XSAVE */
256 static void xen_set_debugreg(int reg, unsigned long val)
258 HYPERVISOR_set_debugreg(reg, val);
261 static unsigned long xen_get_debugreg(int reg)
263 return HYPERVISOR_get_debugreg(reg);
266 static void xen_end_context_switch(struct task_struct *next)
269 paravirt_end_context_switch(next);
272 static unsigned long xen_store_tr(void)
278 * Set the page permissions for a particular virtual address. If the
279 * address is a vmalloc mapping (or other non-linear mapping), then
280 * find the linear mapping of the page and also set its protections to
283 static void set_aliased_prot(void *v, pgprot_t prot)
291 ptep = lookup_address((unsigned long)v, &level);
292 BUG_ON(ptep == NULL);
294 pfn = pte_pfn(*ptep);
295 page = pfn_to_page(pfn);
297 pte = pfn_pte(pfn, prot);
299 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
302 if (!PageHighMem(page)) {
303 void *av = __va(PFN_PHYS(pfn));
306 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
312 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
314 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
317 for(i = 0; i < entries; i += entries_per_page)
318 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
321 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
323 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
326 for(i = 0; i < entries; i += entries_per_page)
327 set_aliased_prot(ldt + i, PAGE_KERNEL);
330 static void xen_set_ldt(const void *addr, unsigned entries)
332 struct mmuext_op *op;
333 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
336 op->cmd = MMUEXT_SET_LDT;
337 op->arg1.linear_addr = (unsigned long)addr;
338 op->arg2.nr_ents = entries;
340 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
342 xen_mc_issue(PARAVIRT_LAZY_CPU);
345 static void xen_load_gdt(const struct desc_ptr *dtr)
347 unsigned long va = dtr->address;
348 unsigned int size = dtr->size + 1;
349 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
350 unsigned long frames[pages];
354 * A GDT can be up to 64k in size, which corresponds to 8192
355 * 8-byte entries, or 16 4k pages..
358 BUG_ON(size > 65536);
359 BUG_ON(va & ~PAGE_MASK);
361 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
364 unsigned long pfn, mfn;
368 * The GDT is per-cpu and is in the percpu data area.
369 * That can be virtually mapped, so we need to do a
370 * page-walk to get the underlying MFN for the
371 * hypercall. The page can also be in the kernel's
372 * linear range, so we need to RO that mapping too.
374 ptep = lookup_address(va, &level);
375 BUG_ON(ptep == NULL);
377 pfn = pte_pfn(*ptep);
378 mfn = pfn_to_mfn(pfn);
379 virt = __va(PFN_PHYS(pfn));
383 make_lowmem_page_readonly((void *)va);
384 make_lowmem_page_readonly(virt);
387 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
392 * load_gdt for early boot, when the gdt is only mapped once
394 static __init void xen_load_gdt_boot(const struct desc_ptr *dtr)
396 unsigned long va = dtr->address;
397 unsigned int size = dtr->size + 1;
398 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
399 unsigned long frames[pages];
403 * A GDT can be up to 64k in size, which corresponds to 8192
404 * 8-byte entries, or 16 4k pages..
407 BUG_ON(size > 65536);
408 BUG_ON(va & ~PAGE_MASK);
410 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
412 unsigned long pfn, mfn;
414 pfn = virt_to_pfn(va);
415 mfn = pfn_to_mfn(pfn);
417 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
419 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
425 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
429 static void load_TLS_descriptor(struct thread_struct *t,
430 unsigned int cpu, unsigned int i)
432 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
433 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
434 struct multicall_space mc = __xen_mc_entry(0);
436 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
439 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
442 * XXX sleazy hack: If we're being called in a lazy-cpu zone
443 * and lazy gs handling is enabled, it means we're in a
444 * context switch, and %gs has just been saved. This means we
445 * can zero it out to prevent faults on exit from the
446 * hypervisor if the next process has no %gs. Either way, it
447 * has been saved, and the new value will get loaded properly.
448 * This will go away as soon as Xen has been modified to not
449 * save/restore %gs for normal hypercalls.
451 * On x86_64, this hack is not used for %gs, because gs points
452 * to KERNEL_GS_BASE (and uses it for PDA references), so we
453 * must not zero %gs on x86_64
455 * For x86_64, we need to zero %fs, otherwise we may get an
456 * exception between the new %fs descriptor being loaded and
457 * %fs being effectively cleared at __switch_to().
459 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
469 load_TLS_descriptor(t, cpu, 0);
470 load_TLS_descriptor(t, cpu, 1);
471 load_TLS_descriptor(t, cpu, 2);
473 xen_mc_issue(PARAVIRT_LAZY_CPU);
477 static void xen_load_gs_index(unsigned int idx)
479 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
484 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
487 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
488 u64 entry = *(u64 *)ptr;
493 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
499 static int cvt_gate_to_trap(int vector, const gate_desc *val,
500 struct trap_info *info)
504 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
507 info->vector = vector;
509 addr = gate_offset(*val);
512 * Look for known traps using IST, and substitute them
513 * appropriately. The debugger ones are the only ones we care
514 * about. Xen will handle faults like double_fault and
515 * machine_check, so we should never see them. Warn if
516 * there's an unexpected IST-using fault handler.
518 if (addr == (unsigned long)debug)
519 addr = (unsigned long)xen_debug;
520 else if (addr == (unsigned long)int3)
521 addr = (unsigned long)xen_int3;
522 else if (addr == (unsigned long)stack_segment)
523 addr = (unsigned long)xen_stack_segment;
524 else if (addr == (unsigned long)double_fault ||
525 addr == (unsigned long)nmi) {
526 /* Don't need to handle these */
528 #ifdef CONFIG_X86_MCE
529 } else if (addr == (unsigned long)machine_check) {
533 /* Some other trap using IST? */
534 if (WARN_ON(val->ist != 0))
537 #endif /* CONFIG_X86_64 */
538 info->address = addr;
540 info->cs = gate_segment(*val);
541 info->flags = val->dpl;
542 /* interrupt gates clear IF */
543 if (val->type == GATE_INTERRUPT)
544 info->flags |= 1 << 2;
549 /* Locations of each CPU's IDT */
550 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
552 /* Set an IDT entry. If the entry is part of the current IDT, then
554 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
556 unsigned long p = (unsigned long)&dt[entrynum];
557 unsigned long start, end;
561 start = __this_cpu_read(idt_desc.address);
562 end = start + __this_cpu_read(idt_desc.size) + 1;
566 native_write_idt_entry(dt, entrynum, g);
568 if (p >= start && (p + 8) <= end) {
569 struct trap_info info[2];
573 if (cvt_gate_to_trap(entrynum, g, &info[0]))
574 if (HYPERVISOR_set_trap_table(info))
581 static void xen_convert_trap_info(const struct desc_ptr *desc,
582 struct trap_info *traps)
584 unsigned in, out, count;
586 count = (desc->size+1) / sizeof(gate_desc);
589 for (in = out = 0; in < count; in++) {
590 gate_desc *entry = (gate_desc*)(desc->address) + in;
592 if (cvt_gate_to_trap(in, entry, &traps[out]))
595 traps[out].address = 0;
598 void xen_copy_trap_info(struct trap_info *traps)
600 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
602 xen_convert_trap_info(desc, traps);
605 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
606 hold a spinlock to protect the static traps[] array (static because
607 it avoids allocation, and saves stack space). */
608 static void xen_load_idt(const struct desc_ptr *desc)
610 static DEFINE_SPINLOCK(lock);
611 static struct trap_info traps[257];
615 __get_cpu_var(idt_desc) = *desc;
617 xen_convert_trap_info(desc, traps);
620 if (HYPERVISOR_set_trap_table(traps))
626 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
627 they're handled differently. */
628 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
629 const void *desc, int type)
640 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
643 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
653 * Version of write_gdt_entry for use at early boot-time needed to
654 * update an entry as simply as possible.
656 static __init void xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
657 const void *desc, int type)
666 xmaddr_t maddr = virt_to_machine(&dt[entry]);
668 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
669 dt[entry] = *(struct desc_struct *)desc;
675 static void xen_load_sp0(struct tss_struct *tss,
676 struct thread_struct *thread)
678 struct multicall_space mcs = xen_mc_entry(0);
679 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
680 xen_mc_issue(PARAVIRT_LAZY_CPU);
683 static void xen_set_iopl_mask(unsigned mask)
685 struct physdev_set_iopl set_iopl;
687 /* Force the change at ring 0. */
688 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
689 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
692 static void xen_io_delay(void)
696 #ifdef CONFIG_X86_LOCAL_APIC
697 static u32 xen_apic_read(u32 reg)
702 static void xen_apic_write(u32 reg, u32 val)
704 /* Warn to see if there's any stray references */
708 static u64 xen_apic_icr_read(void)
713 static void xen_apic_icr_write(u32 low, u32 id)
715 /* Warn to see if there's any stray references */
719 static void xen_apic_wait_icr_idle(void)
724 static u32 xen_safe_apic_wait_icr_idle(void)
729 static void set_xen_basic_apic_ops(void)
731 apic->read = xen_apic_read;
732 apic->write = xen_apic_write;
733 apic->icr_read = xen_apic_icr_read;
734 apic->icr_write = xen_apic_icr_write;
735 apic->wait_icr_idle = xen_apic_wait_icr_idle;
736 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
741 static void xen_clts(void)
743 struct multicall_space mcs;
745 mcs = xen_mc_entry(0);
747 MULTI_fpu_taskswitch(mcs.mc, 0);
749 xen_mc_issue(PARAVIRT_LAZY_CPU);
752 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
754 static unsigned long xen_read_cr0(void)
756 unsigned long cr0 = percpu_read(xen_cr0_value);
758 if (unlikely(cr0 == 0)) {
759 cr0 = native_read_cr0();
760 percpu_write(xen_cr0_value, cr0);
766 static void xen_write_cr0(unsigned long cr0)
768 struct multicall_space mcs;
770 percpu_write(xen_cr0_value, cr0);
772 /* Only pay attention to cr0.TS; everything else is
774 mcs = xen_mc_entry(0);
776 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
778 xen_mc_issue(PARAVIRT_LAZY_CPU);
781 static void xen_write_cr4(unsigned long cr4)
786 native_write_cr4(cr4);
789 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
800 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
801 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
802 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
805 base = ((u64)high << 32) | low;
806 if (HYPERVISOR_set_segment_base(which, base) != 0)
814 case MSR_SYSCALL_MASK:
815 case MSR_IA32_SYSENTER_CS:
816 case MSR_IA32_SYSENTER_ESP:
817 case MSR_IA32_SYSENTER_EIP:
818 /* Fast syscall setup is all done in hypercalls, so
819 these are all ignored. Stub them out here to stop
820 Xen console noise. */
823 case MSR_IA32_CR_PAT:
824 if (smp_processor_id() == 0)
825 xen_set_pat(((u64)high << 32) | low);
829 ret = native_write_msr_safe(msr, low, high);
835 void xen_setup_shared_info(void)
837 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
838 set_fixmap(FIX_PARAVIRT_BOOTMAP,
839 xen_start_info->shared_info);
841 HYPERVISOR_shared_info =
842 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
844 HYPERVISOR_shared_info =
845 (struct shared_info *)__va(xen_start_info->shared_info);
848 /* In UP this is as good a place as any to set up shared info */
849 xen_setup_vcpu_info_placement();
852 xen_setup_mfn_list_list();
855 /* This is called once we have the cpu_possible_map */
856 void xen_setup_vcpu_info_placement(void)
860 for_each_possible_cpu(cpu)
863 /* xen_vcpu_setup managed to place the vcpu_info within the
864 percpu area for all cpus, so make use of it */
865 if (have_vcpu_info_placement) {
866 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
867 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
868 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
869 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
870 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
874 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
875 unsigned long addr, unsigned len)
877 char *start, *end, *reloc;
880 start = end = reloc = NULL;
882 #define SITE(op, x) \
883 case PARAVIRT_PATCH(op.x): \
884 if (have_vcpu_info_placement) { \
885 start = (char *)xen_##x##_direct; \
886 end = xen_##x##_direct_end; \
887 reloc = xen_##x##_direct_reloc; \
892 SITE(pv_irq_ops, irq_enable);
893 SITE(pv_irq_ops, irq_disable);
894 SITE(pv_irq_ops, save_fl);
895 SITE(pv_irq_ops, restore_fl);
899 if (start == NULL || (end-start) > len)
902 ret = paravirt_patch_insns(insnbuf, len, start, end);
904 /* Note: because reloc is assigned from something that
905 appears to be an array, gcc assumes it's non-null,
906 but doesn't know its relationship with start and
908 if (reloc > start && reloc < end) {
909 int reloc_off = reloc - start;
910 long *relocp = (long *)(insnbuf + reloc_off);
911 long delta = start - (char *)addr;
919 ret = paravirt_patch_default(type, clobbers, insnbuf,
927 static const struct pv_info xen_info __initdata = {
928 .paravirt_enabled = 1,
929 .shared_kernel_pmd = 0,
934 static const struct pv_init_ops xen_init_ops __initdata = {
938 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
941 .set_debugreg = xen_set_debugreg,
942 .get_debugreg = xen_get_debugreg,
946 .read_cr0 = xen_read_cr0,
947 .write_cr0 = xen_write_cr0,
949 .read_cr4 = native_read_cr4,
950 .read_cr4_safe = native_read_cr4_safe,
951 .write_cr4 = xen_write_cr4,
953 .wbinvd = native_wbinvd,
955 .read_msr = native_read_msr_safe,
956 .write_msr = xen_write_msr_safe,
957 .read_tsc = native_read_tsc,
958 .read_pmc = native_read_pmc,
961 .irq_enable_sysexit = xen_sysexit,
963 .usergs_sysret32 = xen_sysret32,
964 .usergs_sysret64 = xen_sysret64,
967 .load_tr_desc = paravirt_nop,
968 .set_ldt = xen_set_ldt,
969 .load_gdt = xen_load_gdt,
970 .load_idt = xen_load_idt,
971 .load_tls = xen_load_tls,
973 .load_gs_index = xen_load_gs_index,
976 .alloc_ldt = xen_alloc_ldt,
977 .free_ldt = xen_free_ldt,
979 .store_gdt = native_store_gdt,
980 .store_idt = native_store_idt,
981 .store_tr = xen_store_tr,
983 .write_ldt_entry = xen_write_ldt_entry,
984 .write_gdt_entry = xen_write_gdt_entry,
985 .write_idt_entry = xen_write_idt_entry,
986 .load_sp0 = xen_load_sp0,
988 .set_iopl_mask = xen_set_iopl_mask,
989 .io_delay = xen_io_delay,
991 /* Xen takes care of %gs when switching to usermode for us */
992 .swapgs = paravirt_nop,
994 .start_context_switch = paravirt_start_context_switch,
995 .end_context_switch = xen_end_context_switch,
998 static const struct pv_apic_ops xen_apic_ops __initdata = {
999 #ifdef CONFIG_X86_LOCAL_APIC
1000 .startup_ipi_hook = paravirt_nop,
1004 static void xen_reboot(int reason)
1006 struct sched_shutdown r = { .reason = reason };
1008 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1012 static void xen_restart(char *msg)
1014 xen_reboot(SHUTDOWN_reboot);
1017 static void xen_emergency_restart(void)
1019 xen_reboot(SHUTDOWN_reboot);
1022 static void xen_machine_halt(void)
1024 xen_reboot(SHUTDOWN_poweroff);
1027 static void xen_crash_shutdown(struct pt_regs *regs)
1029 xen_reboot(SHUTDOWN_crash);
1033 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1035 xen_reboot(SHUTDOWN_crash);
1039 static struct notifier_block xen_panic_block = {
1040 .notifier_call= xen_panic_event,
1043 int xen_panic_handler_init(void)
1045 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1049 static const struct machine_ops __initdata xen_machine_ops = {
1050 .restart = xen_restart,
1051 .halt = xen_machine_halt,
1052 .power_off = xen_machine_halt,
1053 .shutdown = xen_machine_halt,
1054 .crash_shutdown = xen_crash_shutdown,
1055 .emergency_restart = xen_emergency_restart,
1059 * Set up the GDT and segment registers for -fstack-protector. Until
1060 * we do this, we have to be careful not to call any stack-protected
1061 * function, which is most of the kernel.
1063 static void __init xen_setup_stackprotector(void)
1065 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1066 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1068 setup_stack_canary_segment(0);
1069 switch_to_new_gdt(0);
1071 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1072 pv_cpu_ops.load_gdt = xen_load_gdt;
1075 /* First C function to be called on Xen boot */
1076 asmlinkage void __init xen_start_kernel(void)
1078 struct physdev_set_iopl set_iopl;
1082 if (!xen_start_info)
1085 xen_domain_type = XEN_PV_DOMAIN;
1087 xen_setup_machphys_mapping();
1089 /* Install Xen paravirt ops */
1091 pv_init_ops = xen_init_ops;
1092 pv_cpu_ops = xen_cpu_ops;
1093 pv_apic_ops = xen_apic_ops;
1095 x86_init.resources.memory_setup = xen_memory_setup;
1096 x86_init.oem.arch_setup = xen_arch_setup;
1097 x86_init.oem.banner = xen_banner;
1099 xen_init_time_ops();
1102 * Set up some pagetable state before starting to set any ptes.
1107 /* Prevent unwanted bits from being set in PTEs. */
1108 __supported_pte_mask &= ~_PAGE_GLOBAL;
1109 if (!xen_initial_domain())
1110 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1112 __supported_pte_mask |= _PAGE_IOMAP;
1115 * Prevent page tables from being allocated in highmem, even
1116 * if CONFIG_HIGHPTE is enabled.
1118 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1120 /* Work out if we support NX */
1123 xen_setup_features();
1126 if (!xen_feature(XENFEAT_auto_translated_physmap))
1127 xen_build_dynamic_phys_to_machine();
1130 * Set up kernel GDT and segment registers, mainly so that
1131 * -fstack-protector code can be executed.
1133 xen_setup_stackprotector();
1136 xen_init_cpuid_mask();
1138 #ifdef CONFIG_X86_LOCAL_APIC
1140 * set up the basic apic ops.
1142 set_xen_basic_apic_ops();
1145 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1146 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1147 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1150 machine_ops = xen_machine_ops;
1153 * The only reliable way to retain the initial address of the
1154 * percpu gdt_page is to remember it here, so we can go and
1155 * mark it RW later, when the initial percpu area is freed.
1157 xen_initial_gdt = &per_cpu(gdt_page, 0);
1161 #ifdef CONFIG_ACPI_NUMA
1163 * The pages we from Xen are not related to machine pages, so
1164 * any NUMA information the kernel tries to get from ACPI will
1165 * be meaningless. Prevent it from trying.
1170 pgd = (pgd_t *)xen_start_info->pt_base;
1172 if (!xen_initial_domain())
1173 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1175 __supported_pte_mask |= _PAGE_IOMAP;
1176 /* Don't do the full vcpu_info placement stuff until we have a
1177 possible map and a non-dummy shared_info. */
1178 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1180 local_irq_disable();
1181 early_boot_irqs_disabled = true;
1185 xen_raw_console_write("mapping kernel into physical memory\n");
1186 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1187 xen_ident_map_ISA();
1189 /* Allocate and initialize top and mid mfn levels for p2m structure */
1190 xen_build_mfn_list_list();
1192 /* keep using Xen gdt for now; no urgent need to change it */
1194 #ifdef CONFIG_X86_32
1195 pv_info.kernel_rpl = 1;
1196 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1197 pv_info.kernel_rpl = 0;
1199 pv_info.kernel_rpl = 0;
1201 /* set the limit of our address space */
1204 /* We used to do this in xen_arch_setup, but that is too late on AMD
1205 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1206 * which pokes 0xcf8 port.
1209 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1211 xen_raw_printk("physdev_op failed %d\n", rc);
1213 #ifdef CONFIG_X86_32
1214 /* set up basic CPUID stuff */
1215 cpu_detect(&new_cpu_data);
1216 new_cpu_data.hard_math = 1;
1217 new_cpu_data.wp_works_ok = 1;
1218 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1221 /* Poke various useful things into boot_params */
1222 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1223 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1224 ? __pa(xen_start_info->mod_start) : 0;
1225 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1226 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1228 if (!xen_initial_domain()) {
1229 add_preferred_console("xenboot", 0, NULL);
1230 add_preferred_console("tty", 0, NULL);
1231 add_preferred_console("hvc", 0, NULL);
1233 x86_init.pci.arch_init = pci_xen_init;
1235 /* Make sure ACS will be enabled */
1240 xen_raw_console_write("about to get started...\n");
1242 xen_setup_runstate_info(0);
1244 /* Start the world */
1245 #ifdef CONFIG_X86_32
1246 i386_start_kernel();
1248 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1252 static int init_hvm_pv_info(int *major, int *minor)
1254 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1257 base = xen_cpuid_base();
1258 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1261 *minor = eax & 0xffff;
1262 printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1264 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1266 pfn = __pa(hypercall_page);
1267 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1269 xen_setup_features();
1271 pv_info.name = "Xen HVM";
1273 xen_domain_type = XEN_HVM_DOMAIN;
1278 void __ref xen_hvm_init_shared_info(void)
1281 struct xen_add_to_physmap xatp;
1282 static struct shared_info *shared_info_page = 0;
1284 if (!shared_info_page)
1285 shared_info_page = (struct shared_info *)
1286 extend_brk(PAGE_SIZE, PAGE_SIZE);
1287 xatp.domid = DOMID_SELF;
1289 xatp.space = XENMAPSPACE_shared_info;
1290 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1291 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1294 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1296 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1297 * page, we use it in the event channel upcall and in some pvclock
1298 * related functions. We don't need the vcpu_info placement
1299 * optimizations because we don't use any pv_mmu or pv_irq op on
1301 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1302 * online but xen_hvm_init_shared_info is run at resume time too and
1303 * in that case multiple vcpus might be online. */
1304 for_each_online_cpu(cpu) {
1305 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1309 #ifdef CONFIG_XEN_PVHVM
1310 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1311 unsigned long action, void *hcpu)
1313 int cpu = (long)hcpu;
1315 case CPU_UP_PREPARE:
1316 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1317 if (xen_have_vector_callback)
1318 xen_init_lock_cpu(cpu);
1326 static struct notifier_block __cpuinitdata xen_hvm_cpu_notifier = {
1327 .notifier_call = xen_hvm_cpu_notify,
1330 static void __init xen_hvm_guest_init(void)
1335 r = init_hvm_pv_info(&major, &minor);
1339 xen_hvm_init_shared_info();
1341 if (xen_feature(XENFEAT_hvm_callback_vector))
1342 xen_have_vector_callback = 1;
1344 register_cpu_notifier(&xen_hvm_cpu_notifier);
1345 xen_unplug_emulated_devices();
1346 have_vcpu_info_placement = 0;
1347 x86_init.irqs.intr_init = xen_init_IRQ;
1348 xen_hvm_init_time_ops();
1349 xen_hvm_init_mmu_ops();
1352 static bool __init xen_hvm_platform(void)
1354 if (xen_pv_domain())
1357 if (!xen_cpuid_base())
1363 bool xen_hvm_need_lapic(void)
1365 if (xen_pv_domain())
1367 if (!xen_hvm_domain())
1369 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1373 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1375 const __refconst struct hypervisor_x86 x86_hyper_xen_hvm = {
1377 .detect = xen_hvm_platform,
1378 .init_platform = xen_hvm_guest_init,
1380 EXPORT_SYMBOL(x86_hyper_xen_hvm);