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>
65 #include <asm/pci_x86.h>
69 #include "multicalls.h"
71 EXPORT_SYMBOL_GPL(hypercall_page);
73 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
74 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
76 enum xen_domain_type xen_domain_type = XEN_NATIVE;
77 EXPORT_SYMBOL_GPL(xen_domain_type);
79 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
80 EXPORT_SYMBOL(machine_to_phys_mapping);
81 unsigned long machine_to_phys_nr;
82 EXPORT_SYMBOL(machine_to_phys_nr);
84 struct start_info *xen_start_info;
85 EXPORT_SYMBOL_GPL(xen_start_info);
87 struct shared_info xen_dummy_shared_info;
89 void *xen_initial_gdt;
91 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
92 __read_mostly int xen_have_vector_callback;
93 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
96 * Point at some empty memory to start with. We map the real shared_info
97 * page as soon as fixmap is up and running.
99 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
102 * Flag to determine whether vcpu info placement is available on all
103 * VCPUs. We assume it is to start with, and then set it to zero on
104 * the first failure. This is because it can succeed on some VCPUs
105 * and not others, since it can involve hypervisor memory allocation,
106 * or because the guest failed to guarantee all the appropriate
107 * constraints on all VCPUs (ie buffer can't cross a page boundary).
109 * Note that any particular CPU may be using a placed vcpu structure,
110 * but we can only optimise if the all are.
112 * 0: not available, 1: available
114 static int have_vcpu_info_placement = 1;
116 static void clamp_max_cpus(void)
119 if (setup_max_cpus > MAX_VIRT_CPUS)
120 setup_max_cpus = MAX_VIRT_CPUS;
124 static void xen_vcpu_setup(int cpu)
126 struct vcpu_register_vcpu_info info;
128 struct vcpu_info *vcpup;
130 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
132 if (cpu < MAX_VIRT_CPUS)
133 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
135 if (!have_vcpu_info_placement) {
136 if (cpu >= MAX_VIRT_CPUS)
141 vcpup = &per_cpu(xen_vcpu_info, cpu);
142 info.mfn = arbitrary_virt_to_mfn(vcpup);
143 info.offset = offset_in_page(vcpup);
145 /* Check to see if the hypervisor will put the vcpu_info
146 structure where we want it, which allows direct access via
147 a percpu-variable. */
148 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
151 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
152 have_vcpu_info_placement = 0;
155 /* This cpu is using the registered vcpu info, even if
156 later ones fail to. */
157 per_cpu(xen_vcpu, cpu) = vcpup;
162 * On restore, set the vcpu placement up again.
163 * If it fails, then we're in a bad state, since
164 * we can't back out from using it...
166 void xen_vcpu_restore(void)
170 for_each_online_cpu(cpu) {
171 bool other_cpu = (cpu != smp_processor_id());
174 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
177 xen_setup_runstate_info(cpu);
179 if (have_vcpu_info_placement)
183 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
188 static void __init xen_banner(void)
190 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
191 struct xen_extraversion extra;
192 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
194 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
196 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
197 version >> 16, version & 0xffff, extra.extraversion,
198 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
201 #define CPUID_THERM_POWER_LEAF 6
202 #define APERFMPERF_PRESENT 0
204 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
205 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
207 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
208 unsigned int *cx, unsigned int *dx)
210 unsigned maskebx = ~0;
211 unsigned maskecx = ~0;
212 unsigned maskedx = ~0;
215 * Mask out inconvenient features, to try and disable as many
216 * unsupported kernel subsystems as possible.
220 maskecx = cpuid_leaf1_ecx_mask;
221 maskedx = cpuid_leaf1_edx_mask;
224 case CPUID_THERM_POWER_LEAF:
225 /* Disabling APERFMPERF for kernel usage */
226 maskecx = ~(1 << APERFMPERF_PRESENT);
230 /* Suppress extended topology stuff */
235 asm(XEN_EMULATE_PREFIX "cpuid"
240 : "0" (*ax), "2" (*cx));
247 static void __init xen_init_cpuid_mask(void)
249 unsigned int ax, bx, cx, dx;
250 unsigned int xsave_mask;
252 cpuid_leaf1_edx_mask =
253 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
254 (1 << X86_FEATURE_MCA) | /* disable MCA */
255 (1 << X86_FEATURE_MTRR) | /* disable MTRR */
256 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
258 if (!xen_initial_domain())
259 cpuid_leaf1_edx_mask &=
260 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
261 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
263 xen_cpuid(&ax, &bx, &cx, &dx);
266 (1 << (X86_FEATURE_XSAVE % 32)) |
267 (1 << (X86_FEATURE_OSXSAVE % 32));
269 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
270 if ((cx & xsave_mask) != xsave_mask)
271 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
274 static void xen_set_debugreg(int reg, unsigned long val)
276 HYPERVISOR_set_debugreg(reg, val);
279 static unsigned long xen_get_debugreg(int reg)
281 return HYPERVISOR_get_debugreg(reg);
284 static void xen_end_context_switch(struct task_struct *next)
287 paravirt_end_context_switch(next);
290 static unsigned long xen_store_tr(void)
296 * Set the page permissions for a particular virtual address. If the
297 * address is a vmalloc mapping (or other non-linear mapping), then
298 * find the linear mapping of the page and also set its protections to
301 static void set_aliased_prot(void *v, pgprot_t prot)
309 ptep = lookup_address((unsigned long)v, &level);
310 BUG_ON(ptep == NULL);
312 pfn = pte_pfn(*ptep);
313 page = pfn_to_page(pfn);
315 pte = pfn_pte(pfn, prot);
317 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
320 if (!PageHighMem(page)) {
321 void *av = __va(PFN_PHYS(pfn));
324 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
330 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
332 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
335 for(i = 0; i < entries; i += entries_per_page)
336 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
339 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
341 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
344 for(i = 0; i < entries; i += entries_per_page)
345 set_aliased_prot(ldt + i, PAGE_KERNEL);
348 static void xen_set_ldt(const void *addr, unsigned entries)
350 struct mmuext_op *op;
351 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
354 op->cmd = MMUEXT_SET_LDT;
355 op->arg1.linear_addr = (unsigned long)addr;
356 op->arg2.nr_ents = entries;
358 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
360 xen_mc_issue(PARAVIRT_LAZY_CPU);
363 static void xen_load_gdt(const struct desc_ptr *dtr)
365 unsigned long va = dtr->address;
366 unsigned int size = dtr->size + 1;
367 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
368 unsigned long frames[pages];
372 * A GDT can be up to 64k in size, which corresponds to 8192
373 * 8-byte entries, or 16 4k pages..
376 BUG_ON(size > 65536);
377 BUG_ON(va & ~PAGE_MASK);
379 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
382 unsigned long pfn, mfn;
386 * The GDT is per-cpu and is in the percpu data area.
387 * That can be virtually mapped, so we need to do a
388 * page-walk to get the underlying MFN for the
389 * hypercall. The page can also be in the kernel's
390 * linear range, so we need to RO that mapping too.
392 ptep = lookup_address(va, &level);
393 BUG_ON(ptep == NULL);
395 pfn = pte_pfn(*ptep);
396 mfn = pfn_to_mfn(pfn);
397 virt = __va(PFN_PHYS(pfn));
401 make_lowmem_page_readonly((void *)va);
402 make_lowmem_page_readonly(virt);
405 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
410 * load_gdt for early boot, when the gdt is only mapped once
412 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
414 unsigned long va = dtr->address;
415 unsigned int size = dtr->size + 1;
416 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
417 unsigned long frames[pages];
421 * A GDT can be up to 64k in size, which corresponds to 8192
422 * 8-byte entries, or 16 4k pages..
425 BUG_ON(size > 65536);
426 BUG_ON(va & ~PAGE_MASK);
428 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
430 unsigned long pfn, mfn;
432 pfn = virt_to_pfn(va);
433 mfn = pfn_to_mfn(pfn);
435 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
437 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
443 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
447 static void load_TLS_descriptor(struct thread_struct *t,
448 unsigned int cpu, unsigned int i)
450 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
451 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
452 struct multicall_space mc = __xen_mc_entry(0);
454 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
457 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
460 * XXX sleazy hack: If we're being called in a lazy-cpu zone
461 * and lazy gs handling is enabled, it means we're in a
462 * context switch, and %gs has just been saved. This means we
463 * can zero it out to prevent faults on exit from the
464 * hypervisor if the next process has no %gs. Either way, it
465 * has been saved, and the new value will get loaded properly.
466 * This will go away as soon as Xen has been modified to not
467 * save/restore %gs for normal hypercalls.
469 * On x86_64, this hack is not used for %gs, because gs points
470 * to KERNEL_GS_BASE (and uses it for PDA references), so we
471 * must not zero %gs on x86_64
473 * For x86_64, we need to zero %fs, otherwise we may get an
474 * exception between the new %fs descriptor being loaded and
475 * %fs being effectively cleared at __switch_to().
477 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
487 load_TLS_descriptor(t, cpu, 0);
488 load_TLS_descriptor(t, cpu, 1);
489 load_TLS_descriptor(t, cpu, 2);
491 xen_mc_issue(PARAVIRT_LAZY_CPU);
495 static void xen_load_gs_index(unsigned int idx)
497 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
502 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
505 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
506 u64 entry = *(u64 *)ptr;
511 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
517 static int cvt_gate_to_trap(int vector, const gate_desc *val,
518 struct trap_info *info)
522 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
525 info->vector = vector;
527 addr = gate_offset(*val);
530 * Look for known traps using IST, and substitute them
531 * appropriately. The debugger ones are the only ones we care
532 * about. Xen will handle faults like double_fault and
533 * machine_check, so we should never see them. Warn if
534 * there's an unexpected IST-using fault handler.
536 if (addr == (unsigned long)debug)
537 addr = (unsigned long)xen_debug;
538 else if (addr == (unsigned long)int3)
539 addr = (unsigned long)xen_int3;
540 else if (addr == (unsigned long)stack_segment)
541 addr = (unsigned long)xen_stack_segment;
542 else if (addr == (unsigned long)double_fault ||
543 addr == (unsigned long)nmi) {
544 /* Don't need to handle these */
546 #ifdef CONFIG_X86_MCE
547 } else if (addr == (unsigned long)machine_check) {
551 /* Some other trap using IST? */
552 if (WARN_ON(val->ist != 0))
555 #endif /* CONFIG_X86_64 */
556 info->address = addr;
558 info->cs = gate_segment(*val);
559 info->flags = val->dpl;
560 /* interrupt gates clear IF */
561 if (val->type == GATE_INTERRUPT)
562 info->flags |= 1 << 2;
567 /* Locations of each CPU's IDT */
568 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
570 /* Set an IDT entry. If the entry is part of the current IDT, then
572 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
574 unsigned long p = (unsigned long)&dt[entrynum];
575 unsigned long start, end;
579 start = __this_cpu_read(idt_desc.address);
580 end = start + __this_cpu_read(idt_desc.size) + 1;
584 native_write_idt_entry(dt, entrynum, g);
586 if (p >= start && (p + 8) <= end) {
587 struct trap_info info[2];
591 if (cvt_gate_to_trap(entrynum, g, &info[0]))
592 if (HYPERVISOR_set_trap_table(info))
599 static void xen_convert_trap_info(const struct desc_ptr *desc,
600 struct trap_info *traps)
602 unsigned in, out, count;
604 count = (desc->size+1) / sizeof(gate_desc);
607 for (in = out = 0; in < count; in++) {
608 gate_desc *entry = (gate_desc*)(desc->address) + in;
610 if (cvt_gate_to_trap(in, entry, &traps[out]))
613 traps[out].address = 0;
616 void xen_copy_trap_info(struct trap_info *traps)
618 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
620 xen_convert_trap_info(desc, traps);
623 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
624 hold a spinlock to protect the static traps[] array (static because
625 it avoids allocation, and saves stack space). */
626 static void xen_load_idt(const struct desc_ptr *desc)
628 static DEFINE_SPINLOCK(lock);
629 static struct trap_info traps[257];
633 __get_cpu_var(idt_desc) = *desc;
635 xen_convert_trap_info(desc, traps);
638 if (HYPERVISOR_set_trap_table(traps))
644 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
645 they're handled differently. */
646 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
647 const void *desc, int type)
658 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
661 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
671 * Version of write_gdt_entry for use at early boot-time needed to
672 * update an entry as simply as possible.
674 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
675 const void *desc, int type)
684 xmaddr_t maddr = virt_to_machine(&dt[entry]);
686 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
687 dt[entry] = *(struct desc_struct *)desc;
693 static void xen_load_sp0(struct tss_struct *tss,
694 struct thread_struct *thread)
696 struct multicall_space mcs = xen_mc_entry(0);
697 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
698 xen_mc_issue(PARAVIRT_LAZY_CPU);
701 static void xen_set_iopl_mask(unsigned mask)
703 struct physdev_set_iopl set_iopl;
705 /* Force the change at ring 0. */
706 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
707 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
710 static void xen_io_delay(void)
714 #ifdef CONFIG_X86_LOCAL_APIC
715 static u32 xen_apic_read(u32 reg)
720 static void xen_apic_write(u32 reg, u32 val)
722 /* Warn to see if there's any stray references */
726 static u64 xen_apic_icr_read(void)
731 static void xen_apic_icr_write(u32 low, u32 id)
733 /* Warn to see if there's any stray references */
737 static void xen_apic_wait_icr_idle(void)
742 static u32 xen_safe_apic_wait_icr_idle(void)
747 static void set_xen_basic_apic_ops(void)
749 apic->read = xen_apic_read;
750 apic->write = xen_apic_write;
751 apic->icr_read = xen_apic_icr_read;
752 apic->icr_write = xen_apic_icr_write;
753 apic->wait_icr_idle = xen_apic_wait_icr_idle;
754 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
759 static void xen_clts(void)
761 struct multicall_space mcs;
763 mcs = xen_mc_entry(0);
765 MULTI_fpu_taskswitch(mcs.mc, 0);
767 xen_mc_issue(PARAVIRT_LAZY_CPU);
770 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
772 static unsigned long xen_read_cr0(void)
774 unsigned long cr0 = percpu_read(xen_cr0_value);
776 if (unlikely(cr0 == 0)) {
777 cr0 = native_read_cr0();
778 percpu_write(xen_cr0_value, cr0);
784 static void xen_write_cr0(unsigned long cr0)
786 struct multicall_space mcs;
788 percpu_write(xen_cr0_value, cr0);
790 /* Only pay attention to cr0.TS; everything else is
792 mcs = xen_mc_entry(0);
794 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
796 xen_mc_issue(PARAVIRT_LAZY_CPU);
799 static void xen_write_cr4(unsigned long cr4)
804 native_write_cr4(cr4);
807 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
818 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
819 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
820 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
823 base = ((u64)high << 32) | low;
824 if (HYPERVISOR_set_segment_base(which, base) != 0)
832 case MSR_SYSCALL_MASK:
833 case MSR_IA32_SYSENTER_CS:
834 case MSR_IA32_SYSENTER_ESP:
835 case MSR_IA32_SYSENTER_EIP:
836 /* Fast syscall setup is all done in hypercalls, so
837 these are all ignored. Stub them out here to stop
838 Xen console noise. */
841 case MSR_IA32_CR_PAT:
842 if (smp_processor_id() == 0)
843 xen_set_pat(((u64)high << 32) | low);
847 ret = native_write_msr_safe(msr, low, high);
853 void xen_setup_shared_info(void)
855 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
856 set_fixmap(FIX_PARAVIRT_BOOTMAP,
857 xen_start_info->shared_info);
859 HYPERVISOR_shared_info =
860 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
862 HYPERVISOR_shared_info =
863 (struct shared_info *)__va(xen_start_info->shared_info);
866 /* In UP this is as good a place as any to set up shared info */
867 xen_setup_vcpu_info_placement();
870 xen_setup_mfn_list_list();
873 /* This is called once we have the cpu_possible_map */
874 void xen_setup_vcpu_info_placement(void)
878 for_each_possible_cpu(cpu)
881 /* xen_vcpu_setup managed to place the vcpu_info within the
882 percpu area for all cpus, so make use of it */
883 if (have_vcpu_info_placement) {
884 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
885 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
886 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
887 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
888 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
892 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
893 unsigned long addr, unsigned len)
895 char *start, *end, *reloc;
898 start = end = reloc = NULL;
900 #define SITE(op, x) \
901 case PARAVIRT_PATCH(op.x): \
902 if (have_vcpu_info_placement) { \
903 start = (char *)xen_##x##_direct; \
904 end = xen_##x##_direct_end; \
905 reloc = xen_##x##_direct_reloc; \
910 SITE(pv_irq_ops, irq_enable);
911 SITE(pv_irq_ops, irq_disable);
912 SITE(pv_irq_ops, save_fl);
913 SITE(pv_irq_ops, restore_fl);
917 if (start == NULL || (end-start) > len)
920 ret = paravirt_patch_insns(insnbuf, len, start, end);
922 /* Note: because reloc is assigned from something that
923 appears to be an array, gcc assumes it's non-null,
924 but doesn't know its relationship with start and
926 if (reloc > start && reloc < end) {
927 int reloc_off = reloc - start;
928 long *relocp = (long *)(insnbuf + reloc_off);
929 long delta = start - (char *)addr;
937 ret = paravirt_patch_default(type, clobbers, insnbuf,
945 static const struct pv_info xen_info __initconst = {
946 .paravirt_enabled = 1,
947 .shared_kernel_pmd = 0,
952 static const struct pv_init_ops xen_init_ops __initconst = {
956 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
959 .set_debugreg = xen_set_debugreg,
960 .get_debugreg = xen_get_debugreg,
964 .read_cr0 = xen_read_cr0,
965 .write_cr0 = xen_write_cr0,
967 .read_cr4 = native_read_cr4,
968 .read_cr4_safe = native_read_cr4_safe,
969 .write_cr4 = xen_write_cr4,
971 .wbinvd = native_wbinvd,
973 .read_msr = native_read_msr_safe,
974 .write_msr = xen_write_msr_safe,
975 .read_tsc = native_read_tsc,
976 .read_pmc = native_read_pmc,
979 .irq_enable_sysexit = xen_sysexit,
981 .usergs_sysret32 = xen_sysret32,
982 .usergs_sysret64 = xen_sysret64,
985 .load_tr_desc = paravirt_nop,
986 .set_ldt = xen_set_ldt,
987 .load_gdt = xen_load_gdt,
988 .load_idt = xen_load_idt,
989 .load_tls = xen_load_tls,
991 .load_gs_index = xen_load_gs_index,
994 .alloc_ldt = xen_alloc_ldt,
995 .free_ldt = xen_free_ldt,
997 .store_gdt = native_store_gdt,
998 .store_idt = native_store_idt,
999 .store_tr = xen_store_tr,
1001 .write_ldt_entry = xen_write_ldt_entry,
1002 .write_gdt_entry = xen_write_gdt_entry,
1003 .write_idt_entry = xen_write_idt_entry,
1004 .load_sp0 = xen_load_sp0,
1006 .set_iopl_mask = xen_set_iopl_mask,
1007 .io_delay = xen_io_delay,
1009 /* Xen takes care of %gs when switching to usermode for us */
1010 .swapgs = paravirt_nop,
1012 .start_context_switch = paravirt_start_context_switch,
1013 .end_context_switch = xen_end_context_switch,
1016 static const struct pv_apic_ops xen_apic_ops __initconst = {
1017 #ifdef CONFIG_X86_LOCAL_APIC
1018 .startup_ipi_hook = paravirt_nop,
1022 static void xen_reboot(int reason)
1024 struct sched_shutdown r = { .reason = reason };
1026 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1030 static void xen_restart(char *msg)
1032 xen_reboot(SHUTDOWN_reboot);
1035 static void xen_emergency_restart(void)
1037 xen_reboot(SHUTDOWN_reboot);
1040 static void xen_machine_halt(void)
1042 xen_reboot(SHUTDOWN_poweroff);
1045 static void xen_machine_power_off(void)
1049 xen_reboot(SHUTDOWN_poweroff);
1052 static void xen_crash_shutdown(struct pt_regs *regs)
1054 xen_reboot(SHUTDOWN_crash);
1058 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1060 xen_reboot(SHUTDOWN_crash);
1064 static struct notifier_block xen_panic_block = {
1065 .notifier_call= xen_panic_event,
1068 int xen_panic_handler_init(void)
1070 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1074 static const struct machine_ops xen_machine_ops __initconst = {
1075 .restart = xen_restart,
1076 .halt = xen_machine_halt,
1077 .power_off = xen_machine_power_off,
1078 .shutdown = xen_machine_halt,
1079 .crash_shutdown = xen_crash_shutdown,
1080 .emergency_restart = xen_emergency_restart,
1084 * Set up the GDT and segment registers for -fstack-protector. Until
1085 * we do this, we have to be careful not to call any stack-protected
1086 * function, which is most of the kernel.
1088 static void __init xen_setup_stackprotector(void)
1090 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1091 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1093 setup_stack_canary_segment(0);
1094 switch_to_new_gdt(0);
1096 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1097 pv_cpu_ops.load_gdt = xen_load_gdt;
1100 /* First C function to be called on Xen boot */
1101 asmlinkage void __init xen_start_kernel(void)
1103 struct physdev_set_iopl set_iopl;
1107 if (!xen_start_info)
1110 xen_domain_type = XEN_PV_DOMAIN;
1112 xen_setup_machphys_mapping();
1114 /* Install Xen paravirt ops */
1116 pv_init_ops = xen_init_ops;
1117 pv_cpu_ops = xen_cpu_ops;
1118 pv_apic_ops = xen_apic_ops;
1120 x86_init.resources.memory_setup = xen_memory_setup;
1121 x86_init.oem.arch_setup = xen_arch_setup;
1122 x86_init.oem.banner = xen_banner;
1124 xen_init_time_ops();
1127 * Set up some pagetable state before starting to set any ptes.
1132 /* Prevent unwanted bits from being set in PTEs. */
1133 __supported_pte_mask &= ~_PAGE_GLOBAL;
1134 if (!xen_initial_domain())
1135 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1137 __supported_pte_mask |= _PAGE_IOMAP;
1140 * Prevent page tables from being allocated in highmem, even
1141 * if CONFIG_HIGHPTE is enabled.
1143 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1145 /* Work out if we support NX */
1148 xen_setup_features();
1151 if (!xen_feature(XENFEAT_auto_translated_physmap))
1152 xen_build_dynamic_phys_to_machine();
1155 * Set up kernel GDT and segment registers, mainly so that
1156 * -fstack-protector code can be executed.
1158 xen_setup_stackprotector();
1161 xen_init_cpuid_mask();
1163 #ifdef CONFIG_X86_LOCAL_APIC
1165 * set up the basic apic ops.
1167 set_xen_basic_apic_ops();
1170 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1171 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1172 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1175 machine_ops = xen_machine_ops;
1178 * The only reliable way to retain the initial address of the
1179 * percpu gdt_page is to remember it here, so we can go and
1180 * mark it RW later, when the initial percpu area is freed.
1182 xen_initial_gdt = &per_cpu(gdt_page, 0);
1186 #ifdef CONFIG_ACPI_NUMA
1188 * The pages we from Xen are not related to machine pages, so
1189 * any NUMA information the kernel tries to get from ACPI will
1190 * be meaningless. Prevent it from trying.
1195 pgd = (pgd_t *)xen_start_info->pt_base;
1197 if (!xen_initial_domain())
1198 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1200 __supported_pte_mask |= _PAGE_IOMAP;
1201 /* Don't do the full vcpu_info placement stuff until we have a
1202 possible map and a non-dummy shared_info. */
1203 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1205 local_irq_disable();
1206 early_boot_irqs_disabled = true;
1210 xen_raw_console_write("mapping kernel into physical memory\n");
1211 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1212 xen_ident_map_ISA();
1214 /* Allocate and initialize top and mid mfn levels for p2m structure */
1215 xen_build_mfn_list_list();
1217 /* keep using Xen gdt for now; no urgent need to change it */
1219 #ifdef CONFIG_X86_32
1220 pv_info.kernel_rpl = 1;
1221 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1222 pv_info.kernel_rpl = 0;
1224 pv_info.kernel_rpl = 0;
1226 /* set the limit of our address space */
1229 /* We used to do this in xen_arch_setup, but that is too late on AMD
1230 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1231 * which pokes 0xcf8 port.
1234 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1236 xen_raw_printk("physdev_op failed %d\n", rc);
1238 #ifdef CONFIG_X86_32
1239 /* set up basic CPUID stuff */
1240 cpu_detect(&new_cpu_data);
1241 new_cpu_data.hard_math = 1;
1242 new_cpu_data.wp_works_ok = 1;
1243 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1246 /* Poke various useful things into boot_params */
1247 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1248 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1249 ? __pa(xen_start_info->mod_start) : 0;
1250 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1251 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1253 if (!xen_initial_domain()) {
1254 add_preferred_console("xenboot", 0, NULL);
1255 add_preferred_console("tty", 0, NULL);
1256 add_preferred_console("hvc", 0, NULL);
1258 x86_init.pci.arch_init = pci_xen_init;
1260 const struct dom0_vga_console_info *info =
1261 (void *)((char *)xen_start_info +
1262 xen_start_info->console.dom0.info_off);
1264 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1265 xen_start_info->console.domU.mfn = 0;
1266 xen_start_info->console.domU.evtchn = 0;
1268 /* Make sure ACS will be enabled */
1272 /* PCI BIOS service won't work from a PV guest. */
1273 pci_probe &= ~PCI_PROBE_BIOS;
1275 xen_raw_console_write("about to get started...\n");
1277 xen_setup_runstate_info(0);
1279 /* Start the world */
1280 #ifdef CONFIG_X86_32
1281 i386_start_kernel();
1283 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1287 static int init_hvm_pv_info(int *major, int *minor)
1289 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1292 base = xen_cpuid_base();
1293 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1296 *minor = eax & 0xffff;
1297 printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1299 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1301 pfn = __pa(hypercall_page);
1302 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1304 xen_setup_features();
1306 pv_info.name = "Xen HVM";
1308 xen_domain_type = XEN_HVM_DOMAIN;
1313 void __ref xen_hvm_init_shared_info(void)
1316 struct xen_add_to_physmap xatp;
1317 static struct shared_info *shared_info_page = 0;
1319 if (!shared_info_page)
1320 shared_info_page = (struct shared_info *)
1321 extend_brk(PAGE_SIZE, PAGE_SIZE);
1322 xatp.domid = DOMID_SELF;
1324 xatp.space = XENMAPSPACE_shared_info;
1325 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1326 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1329 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1331 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1332 * page, we use it in the event channel upcall and in some pvclock
1333 * related functions. We don't need the vcpu_info placement
1334 * optimizations because we don't use any pv_mmu or pv_irq op on
1336 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1337 * online but xen_hvm_init_shared_info is run at resume time too and
1338 * in that case multiple vcpus might be online. */
1339 for_each_online_cpu(cpu) {
1340 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1344 #ifdef CONFIG_XEN_PVHVM
1345 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1346 unsigned long action, void *hcpu)
1348 int cpu = (long)hcpu;
1350 case CPU_UP_PREPARE:
1351 xen_vcpu_setup(cpu);
1352 if (xen_have_vector_callback)
1353 xen_init_lock_cpu(cpu);
1361 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1362 .notifier_call = xen_hvm_cpu_notify,
1365 static void __init xen_hvm_guest_init(void)
1370 r = init_hvm_pv_info(&major, &minor);
1374 xen_hvm_init_shared_info();
1376 if (xen_feature(XENFEAT_hvm_callback_vector))
1377 xen_have_vector_callback = 1;
1379 register_cpu_notifier(&xen_hvm_cpu_notifier);
1380 xen_unplug_emulated_devices();
1381 x86_init.irqs.intr_init = xen_init_IRQ;
1382 xen_hvm_init_time_ops();
1383 xen_hvm_init_mmu_ops();
1386 static bool __init xen_hvm_platform(void)
1388 if (xen_pv_domain())
1391 if (!xen_cpuid_base())
1397 bool xen_hvm_need_lapic(void)
1399 if (xen_pv_domain())
1401 if (!xen_hvm_domain())
1403 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1407 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1409 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1411 .detect = xen_hvm_platform,
1412 .init_platform = xen_hvm_guest_init,
1414 EXPORT_SYMBOL(x86_hyper_xen_hvm);