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>
35 #include <xen/interface/xen.h>
36 #include <xen/interface/version.h>
37 #include <xen/interface/physdev.h>
38 #include <xen/interface/vcpu.h>
39 #include <xen/interface/memory.h>
40 #include <xen/features.h>
43 #include <xen/hvc-console.h>
45 #include <asm/paravirt.h>
48 #include <asm/xen/pci.h>
49 #include <asm/xen/hypercall.h>
50 #include <asm/xen/hypervisor.h>
51 #include <asm/fixmap.h>
52 #include <asm/processor.h>
53 #include <asm/proto.h>
54 #include <asm/msr-index.h>
55 #include <asm/traps.h>
56 #include <asm/setup.h>
58 #include <asm/pgalloc.h>
59 #include <asm/pgtable.h>
60 #include <asm/tlbflush.h>
61 #include <asm/reboot.h>
62 #include <asm/setup.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 struct start_info *xen_start_info;
79 EXPORT_SYMBOL_GPL(xen_start_info);
81 struct shared_info xen_dummy_shared_info;
83 void *xen_initial_gdt;
85 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
86 __read_mostly int xen_have_vector_callback;
87 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
90 * Point at some empty memory to start with. We map the real shared_info
91 * page as soon as fixmap is up and running.
93 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
96 * Flag to determine whether vcpu info placement is available on all
97 * VCPUs. We assume it is to start with, and then set it to zero on
98 * the first failure. This is because it can succeed on some VCPUs
99 * and not others, since it can involve hypervisor memory allocation,
100 * or because the guest failed to guarantee all the appropriate
101 * constraints on all VCPUs (ie buffer can't cross a page boundary).
103 * Note that any particular CPU may be using a placed vcpu structure,
104 * but we can only optimise if the all are.
106 * 0: not available, 1: available
108 static int have_vcpu_info_placement = 1;
110 static void clamp_max_cpus(void)
113 if (setup_max_cpus > MAX_VIRT_CPUS)
114 setup_max_cpus = MAX_VIRT_CPUS;
118 static void xen_vcpu_setup(int cpu)
120 struct vcpu_register_vcpu_info info;
122 struct vcpu_info *vcpup;
124 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
126 if (cpu < MAX_VIRT_CPUS)
127 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
129 if (!have_vcpu_info_placement) {
130 if (cpu >= MAX_VIRT_CPUS)
135 vcpup = &per_cpu(xen_vcpu_info, cpu);
136 info.mfn = arbitrary_virt_to_mfn(vcpup);
137 info.offset = offset_in_page(vcpup);
139 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
140 cpu, vcpup, info.mfn, info.offset);
142 /* Check to see if the hypervisor will put the vcpu_info
143 structure where we want it, which allows direct access via
144 a percpu-variable. */
145 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
148 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
149 have_vcpu_info_placement = 0;
152 /* This cpu is using the registered vcpu info, even if
153 later ones fail to. */
154 per_cpu(xen_vcpu, cpu) = vcpup;
156 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
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 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
202 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
204 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
205 unsigned int *cx, unsigned int *dx)
207 unsigned maskebx = ~0;
208 unsigned maskecx = ~0;
209 unsigned maskedx = ~0;
212 * Mask out inconvenient features, to try and disable as many
213 * unsupported kernel subsystems as possible.
217 maskecx = cpuid_leaf1_ecx_mask;
218 maskedx = cpuid_leaf1_edx_mask;
222 /* Suppress extended topology stuff */
227 asm(XEN_EMULATE_PREFIX "cpuid"
232 : "0" (*ax), "2" (*cx));
239 static __init void xen_init_cpuid_mask(void)
241 unsigned int ax, bx, cx, dx;
243 cpuid_leaf1_edx_mask =
244 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
245 (1 << X86_FEATURE_MCA) | /* disable MCA */
246 (1 << X86_FEATURE_MTRR) | /* disable MTRR */
247 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
249 if (!xen_initial_domain())
250 cpuid_leaf1_edx_mask &=
251 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
252 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
256 xen_cpuid(&ax, &bx, &cx, &dx);
258 /* cpuid claims we support xsave; try enabling it to see what happens */
259 if (cx & (1 << (X86_FEATURE_XSAVE % 32))) {
262 set_in_cr4(X86_CR4_OSXSAVE);
266 if ((cr4 & X86_CR4_OSXSAVE) == 0)
267 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32));
269 clear_in_cr4(X86_CR4_OSXSAVE);
273 static void xen_set_debugreg(int reg, unsigned long val)
275 HYPERVISOR_set_debugreg(reg, val);
278 static unsigned long xen_get_debugreg(int reg)
280 return HYPERVISOR_get_debugreg(reg);
283 static void xen_end_context_switch(struct task_struct *next)
286 paravirt_end_context_switch(next);
289 static unsigned long xen_store_tr(void)
295 * Set the page permissions for a particular virtual address. If the
296 * address is a vmalloc mapping (or other non-linear mapping), then
297 * find the linear mapping of the page and also set its protections to
300 static void set_aliased_prot(void *v, pgprot_t prot)
308 ptep = lookup_address((unsigned long)v, &level);
309 BUG_ON(ptep == NULL);
311 pfn = pte_pfn(*ptep);
312 page = pfn_to_page(pfn);
314 pte = pfn_pte(pfn, prot);
316 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
319 if (!PageHighMem(page)) {
320 void *av = __va(PFN_PHYS(pfn));
323 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
329 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
331 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
334 for(i = 0; i < entries; i += entries_per_page)
335 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
338 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
340 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
343 for(i = 0; i < entries; i += entries_per_page)
344 set_aliased_prot(ldt + i, PAGE_KERNEL);
347 static void xen_set_ldt(const void *addr, unsigned entries)
349 struct mmuext_op *op;
350 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
353 op->cmd = MMUEXT_SET_LDT;
354 op->arg1.linear_addr = (unsigned long)addr;
355 op->arg2.nr_ents = entries;
357 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
359 xen_mc_issue(PARAVIRT_LAZY_CPU);
362 static void xen_load_gdt(const struct desc_ptr *dtr)
364 unsigned long va = dtr->address;
365 unsigned int size = dtr->size + 1;
366 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
367 unsigned long frames[pages];
371 * A GDT can be up to 64k in size, which corresponds to 8192
372 * 8-byte entries, or 16 4k pages..
375 BUG_ON(size > 65536);
376 BUG_ON(va & ~PAGE_MASK);
378 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
381 unsigned long pfn, mfn;
385 * The GDT is per-cpu and is in the percpu data area.
386 * That can be virtually mapped, so we need to do a
387 * page-walk to get the underlying MFN for the
388 * hypercall. The page can also be in the kernel's
389 * linear range, so we need to RO that mapping too.
391 ptep = lookup_address(va, &level);
392 BUG_ON(ptep == NULL);
394 pfn = pte_pfn(*ptep);
395 mfn = pfn_to_mfn(pfn);
396 virt = __va(PFN_PHYS(pfn));
400 make_lowmem_page_readonly((void *)va);
401 make_lowmem_page_readonly(virt);
404 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
409 * load_gdt for early boot, when the gdt is only mapped once
411 static __init void xen_load_gdt_boot(const struct desc_ptr *dtr)
413 unsigned long va = dtr->address;
414 unsigned int size = dtr->size + 1;
415 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
416 unsigned long frames[pages];
420 * A GDT can be up to 64k in size, which corresponds to 8192
421 * 8-byte entries, or 16 4k pages..
424 BUG_ON(size > 65536);
425 BUG_ON(va & ~PAGE_MASK);
427 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
429 unsigned long pfn, mfn;
431 pfn = virt_to_pfn(va);
432 mfn = pfn_to_mfn(pfn);
434 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
436 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
442 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
446 static void load_TLS_descriptor(struct thread_struct *t,
447 unsigned int cpu, unsigned int i)
449 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
450 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
451 struct multicall_space mc = __xen_mc_entry(0);
453 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
456 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
459 * XXX sleazy hack: If we're being called in a lazy-cpu zone
460 * and lazy gs handling is enabled, it means we're in a
461 * context switch, and %gs has just been saved. This means we
462 * can zero it out to prevent faults on exit from the
463 * hypervisor if the next process has no %gs. Either way, it
464 * has been saved, and the new value will get loaded properly.
465 * This will go away as soon as Xen has been modified to not
466 * save/restore %gs for normal hypercalls.
468 * On x86_64, this hack is not used for %gs, because gs points
469 * to KERNEL_GS_BASE (and uses it for PDA references), so we
470 * must not zero %gs on x86_64
472 * For x86_64, we need to zero %fs, otherwise we may get an
473 * exception between the new %fs descriptor being loaded and
474 * %fs being effectively cleared at __switch_to().
476 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
486 load_TLS_descriptor(t, cpu, 0);
487 load_TLS_descriptor(t, cpu, 1);
488 load_TLS_descriptor(t, cpu, 2);
490 xen_mc_issue(PARAVIRT_LAZY_CPU);
494 static void xen_load_gs_index(unsigned int idx)
496 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
501 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
504 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
505 u64 entry = *(u64 *)ptr;
510 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
516 static int cvt_gate_to_trap(int vector, const gate_desc *val,
517 struct trap_info *info)
521 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
524 info->vector = vector;
526 addr = gate_offset(*val);
529 * Look for known traps using IST, and substitute them
530 * appropriately. The debugger ones are the only ones we care
531 * about. Xen will handle faults like double_fault and
532 * machine_check, so we should never see them. Warn if
533 * there's an unexpected IST-using fault handler.
535 if (addr == (unsigned long)debug)
536 addr = (unsigned long)xen_debug;
537 else if (addr == (unsigned long)int3)
538 addr = (unsigned long)xen_int3;
539 else if (addr == (unsigned long)stack_segment)
540 addr = (unsigned long)xen_stack_segment;
541 else if (addr == (unsigned long)double_fault ||
542 addr == (unsigned long)nmi) {
543 /* Don't need to handle these */
545 #ifdef CONFIG_X86_MCE
546 } else if (addr == (unsigned long)machine_check) {
550 /* Some other trap using IST? */
551 if (WARN_ON(val->ist != 0))
554 #endif /* CONFIG_X86_64 */
555 info->address = addr;
557 info->cs = gate_segment(*val);
558 info->flags = val->dpl;
559 /* interrupt gates clear IF */
560 if (val->type == GATE_INTERRUPT)
561 info->flags |= 1 << 2;
566 /* Locations of each CPU's IDT */
567 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
569 /* Set an IDT entry. If the entry is part of the current IDT, then
571 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
573 unsigned long p = (unsigned long)&dt[entrynum];
574 unsigned long start, end;
578 start = __get_cpu_var(idt_desc).address;
579 end = start + __get_cpu_var(idt_desc).size + 1;
583 native_write_idt_entry(dt, entrynum, g);
585 if (p >= start && (p + 8) <= end) {
586 struct trap_info info[2];
590 if (cvt_gate_to_trap(entrynum, g, &info[0]))
591 if (HYPERVISOR_set_trap_table(info))
598 static void xen_convert_trap_info(const struct desc_ptr *desc,
599 struct trap_info *traps)
601 unsigned in, out, count;
603 count = (desc->size+1) / sizeof(gate_desc);
606 for (in = out = 0; in < count; in++) {
607 gate_desc *entry = (gate_desc*)(desc->address) + in;
609 if (cvt_gate_to_trap(in, entry, &traps[out]))
612 traps[out].address = 0;
615 void xen_copy_trap_info(struct trap_info *traps)
617 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
619 xen_convert_trap_info(desc, traps);
622 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
623 hold a spinlock to protect the static traps[] array (static because
624 it avoids allocation, and saves stack space). */
625 static void xen_load_idt(const struct desc_ptr *desc)
627 static DEFINE_SPINLOCK(lock);
628 static struct trap_info traps[257];
632 __get_cpu_var(idt_desc) = *desc;
634 xen_convert_trap_info(desc, traps);
637 if (HYPERVISOR_set_trap_table(traps))
643 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
644 they're handled differently. */
645 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
646 const void *desc, int type)
657 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
660 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
670 * Version of write_gdt_entry for use at early boot-time needed to
671 * update an entry as simply as possible.
673 static __init void xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
674 const void *desc, int type)
683 xmaddr_t maddr = virt_to_machine(&dt[entry]);
685 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
686 dt[entry] = *(struct desc_struct *)desc;
692 static void xen_load_sp0(struct tss_struct *tss,
693 struct thread_struct *thread)
695 struct multicall_space mcs = xen_mc_entry(0);
696 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
697 xen_mc_issue(PARAVIRT_LAZY_CPU);
700 static void xen_set_iopl_mask(unsigned mask)
702 struct physdev_set_iopl set_iopl;
704 /* Force the change at ring 0. */
705 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
706 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
709 static void xen_io_delay(void)
713 #ifdef CONFIG_X86_LOCAL_APIC
714 static u32 xen_apic_read(u32 reg)
719 static void xen_apic_write(u32 reg, u32 val)
721 /* Warn to see if there's any stray references */
725 static u64 xen_apic_icr_read(void)
730 static void xen_apic_icr_write(u32 low, u32 id)
732 /* Warn to see if there's any stray references */
736 static void xen_apic_wait_icr_idle(void)
741 static u32 xen_safe_apic_wait_icr_idle(void)
746 static void set_xen_basic_apic_ops(void)
748 apic->read = xen_apic_read;
749 apic->write = xen_apic_write;
750 apic->icr_read = xen_apic_icr_read;
751 apic->icr_write = xen_apic_icr_write;
752 apic->wait_icr_idle = xen_apic_wait_icr_idle;
753 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
758 static void xen_clts(void)
760 struct multicall_space mcs;
762 mcs = xen_mc_entry(0);
764 MULTI_fpu_taskswitch(mcs.mc, 0);
766 xen_mc_issue(PARAVIRT_LAZY_CPU);
769 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
771 static unsigned long xen_read_cr0(void)
773 unsigned long cr0 = percpu_read(xen_cr0_value);
775 if (unlikely(cr0 == 0)) {
776 cr0 = native_read_cr0();
777 percpu_write(xen_cr0_value, cr0);
783 static void xen_write_cr0(unsigned long cr0)
785 struct multicall_space mcs;
787 percpu_write(xen_cr0_value, cr0);
789 /* Only pay attention to cr0.TS; everything else is
791 mcs = xen_mc_entry(0);
793 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
795 xen_mc_issue(PARAVIRT_LAZY_CPU);
798 static void xen_write_cr4(unsigned long cr4)
803 native_write_cr4(cr4);
806 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
817 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
818 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
819 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
822 base = ((u64)high << 32) | low;
823 if (HYPERVISOR_set_segment_base(which, base) != 0)
831 case MSR_SYSCALL_MASK:
832 case MSR_IA32_SYSENTER_CS:
833 case MSR_IA32_SYSENTER_ESP:
834 case MSR_IA32_SYSENTER_EIP:
835 /* Fast syscall setup is all done in hypercalls, so
836 these are all ignored. Stub them out here to stop
837 Xen console noise. */
841 ret = native_write_msr_safe(msr, low, high);
847 void xen_setup_shared_info(void)
849 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
850 set_fixmap(FIX_PARAVIRT_BOOTMAP,
851 xen_start_info->shared_info);
853 HYPERVISOR_shared_info =
854 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
856 HYPERVISOR_shared_info =
857 (struct shared_info *)__va(xen_start_info->shared_info);
860 /* In UP this is as good a place as any to set up shared info */
861 xen_setup_vcpu_info_placement();
864 xen_setup_mfn_list_list();
867 /* This is called once we have the cpu_possible_map */
868 void xen_setup_vcpu_info_placement(void)
872 for_each_possible_cpu(cpu)
875 /* xen_vcpu_setup managed to place the vcpu_info within the
876 percpu area for all cpus, so make use of it */
877 if (have_vcpu_info_placement) {
878 printk(KERN_INFO "Xen: using vcpu_info placement\n");
880 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
881 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
882 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
883 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
884 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
888 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
889 unsigned long addr, unsigned len)
891 char *start, *end, *reloc;
894 start = end = reloc = NULL;
896 #define SITE(op, x) \
897 case PARAVIRT_PATCH(op.x): \
898 if (have_vcpu_info_placement) { \
899 start = (char *)xen_##x##_direct; \
900 end = xen_##x##_direct_end; \
901 reloc = xen_##x##_direct_reloc; \
906 SITE(pv_irq_ops, irq_enable);
907 SITE(pv_irq_ops, irq_disable);
908 SITE(pv_irq_ops, save_fl);
909 SITE(pv_irq_ops, restore_fl);
913 if (start == NULL || (end-start) > len)
916 ret = paravirt_patch_insns(insnbuf, len, start, end);
918 /* Note: because reloc is assigned from something that
919 appears to be an array, gcc assumes it's non-null,
920 but doesn't know its relationship with start and
922 if (reloc > start && reloc < end) {
923 int reloc_off = reloc - start;
924 long *relocp = (long *)(insnbuf + reloc_off);
925 long delta = start - (char *)addr;
933 ret = paravirt_patch_default(type, clobbers, insnbuf,
941 static const struct pv_info xen_info __initdata = {
942 .paravirt_enabled = 1,
943 .shared_kernel_pmd = 0,
948 static const struct pv_init_ops xen_init_ops __initdata = {
952 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
955 .set_debugreg = xen_set_debugreg,
956 .get_debugreg = xen_get_debugreg,
960 .read_cr0 = xen_read_cr0,
961 .write_cr0 = xen_write_cr0,
963 .read_cr4 = native_read_cr4,
964 .read_cr4_safe = native_read_cr4_safe,
965 .write_cr4 = xen_write_cr4,
967 .wbinvd = native_wbinvd,
969 .read_msr = native_read_msr_safe,
970 .write_msr = xen_write_msr_safe,
971 .read_tsc = native_read_tsc,
972 .read_pmc = native_read_pmc,
975 .irq_enable_sysexit = xen_sysexit,
977 .usergs_sysret32 = xen_sysret32,
978 .usergs_sysret64 = xen_sysret64,
981 .load_tr_desc = paravirt_nop,
982 .set_ldt = xen_set_ldt,
983 .load_gdt = xen_load_gdt,
984 .load_idt = xen_load_idt,
985 .load_tls = xen_load_tls,
987 .load_gs_index = xen_load_gs_index,
990 .alloc_ldt = xen_alloc_ldt,
991 .free_ldt = xen_free_ldt,
993 .store_gdt = native_store_gdt,
994 .store_idt = native_store_idt,
995 .store_tr = xen_store_tr,
997 .write_ldt_entry = xen_write_ldt_entry,
998 .write_gdt_entry = xen_write_gdt_entry,
999 .write_idt_entry = xen_write_idt_entry,
1000 .load_sp0 = xen_load_sp0,
1002 .set_iopl_mask = xen_set_iopl_mask,
1003 .io_delay = xen_io_delay,
1005 /* Xen takes care of %gs when switching to usermode for us */
1006 .swapgs = paravirt_nop,
1008 .start_context_switch = paravirt_start_context_switch,
1009 .end_context_switch = xen_end_context_switch,
1012 static const struct pv_apic_ops xen_apic_ops __initdata = {
1013 #ifdef CONFIG_X86_LOCAL_APIC
1014 .startup_ipi_hook = paravirt_nop,
1018 static void xen_reboot(int reason)
1020 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_crash_shutdown(struct pt_regs *regs)
1047 xen_reboot(SHUTDOWN_crash);
1051 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1053 xen_reboot(SHUTDOWN_crash);
1057 static struct notifier_block xen_panic_block = {
1058 .notifier_call= xen_panic_event,
1061 int xen_panic_handler_init(void)
1063 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1067 static const struct machine_ops __initdata xen_machine_ops = {
1068 .restart = xen_restart,
1069 .halt = xen_machine_halt,
1070 .power_off = xen_machine_halt,
1071 .shutdown = xen_machine_halt,
1072 .crash_shutdown = xen_crash_shutdown,
1073 .emergency_restart = xen_emergency_restart,
1077 * Set up the GDT and segment registers for -fstack-protector. Until
1078 * we do this, we have to be careful not to call any stack-protected
1079 * function, which is most of the kernel.
1081 static void __init xen_setup_stackprotector(void)
1083 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1084 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1086 setup_stack_canary_segment(0);
1087 switch_to_new_gdt(0);
1089 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1090 pv_cpu_ops.load_gdt = xen_load_gdt;
1093 /* First C function to be called on Xen boot */
1094 asmlinkage void __init xen_start_kernel(void)
1098 if (!xen_start_info)
1101 xen_domain_type = XEN_PV_DOMAIN;
1103 /* Install Xen paravirt ops */
1105 pv_init_ops = xen_init_ops;
1106 pv_cpu_ops = xen_cpu_ops;
1107 pv_apic_ops = xen_apic_ops;
1109 x86_init.resources.memory_setup = xen_memory_setup;
1110 x86_init.oem.arch_setup = xen_arch_setup;
1111 x86_init.oem.banner = xen_banner;
1113 xen_init_time_ops();
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 if (!xen_initial_domain())
1178 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1180 __supported_pte_mask |= _PAGE_IOMAP;
1181 /* Don't do the full vcpu_info placement stuff until we have a
1182 possible map and a non-dummy shared_info. */
1183 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1185 local_irq_disable();
1186 early_boot_irqs_off();
1188 xen_raw_console_write("mapping kernel into physical memory\n");
1189 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1190 xen_ident_map_ISA();
1194 /* keep using Xen gdt for now; no urgent need to change it */
1196 #ifdef CONFIG_X86_32
1197 pv_info.kernel_rpl = 1;
1198 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1199 pv_info.kernel_rpl = 0;
1201 pv_info.kernel_rpl = 0;
1204 /* set the limit of our address space */
1207 #ifdef CONFIG_X86_32
1208 /* set up basic CPUID stuff */
1209 cpu_detect(&new_cpu_data);
1210 new_cpu_data.hard_math = 1;
1211 new_cpu_data.wp_works_ok = 1;
1212 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1215 /* Poke various useful things into boot_params */
1216 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1217 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1218 ? __pa(xen_start_info->mod_start) : 0;
1219 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1220 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1222 if (!xen_initial_domain()) {
1223 add_preferred_console("xenboot", 0, NULL);
1224 add_preferred_console("tty", 0, NULL);
1225 add_preferred_console("hvc", 0, NULL);
1227 x86_init.pci.arch_init = pci_xen_init;
1229 /* Make sure ACS will be enabled */
1234 xen_raw_console_write("about to get started...\n");
1236 xen_setup_runstate_info(0);
1238 /* Start the world */
1239 #ifdef CONFIG_X86_32
1240 i386_start_kernel();
1242 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1246 static uint32_t xen_cpuid_base(void)
1248 uint32_t base, eax, ebx, ecx, edx;
1251 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
1252 cpuid(base, &eax, &ebx, &ecx, &edx);
1253 *(uint32_t *)(signature + 0) = ebx;
1254 *(uint32_t *)(signature + 4) = ecx;
1255 *(uint32_t *)(signature + 8) = edx;
1258 if (!strcmp("XenVMMXenVMM", signature) && ((eax - base) >= 2))
1265 static int init_hvm_pv_info(int *major, int *minor)
1267 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1270 base = xen_cpuid_base();
1271 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1274 *minor = eax & 0xffff;
1275 printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1277 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1279 pfn = __pa(hypercall_page);
1280 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1282 xen_setup_features();
1285 pv_info.kernel_rpl = 0;
1287 xen_domain_type = XEN_HVM_DOMAIN;
1292 void xen_hvm_init_shared_info(void)
1295 struct xen_add_to_physmap xatp;
1296 static struct shared_info *shared_info_page = 0;
1298 if (!shared_info_page)
1299 shared_info_page = (struct shared_info *)
1300 extend_brk(PAGE_SIZE, PAGE_SIZE);
1301 xatp.domid = DOMID_SELF;
1303 xatp.space = XENMAPSPACE_shared_info;
1304 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1305 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1308 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1310 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1311 * page, we use it in the event channel upcall and in some pvclock
1312 * related functions. We don't need the vcpu_info placement
1313 * optimizations because we don't use any pv_mmu or pv_irq op on
1315 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1316 * online but xen_hvm_init_shared_info is run at resume time too and
1317 * in that case multiple vcpus might be online. */
1318 for_each_online_cpu(cpu) {
1319 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1323 #ifdef CONFIG_XEN_PVHVM
1324 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1325 unsigned long action, void *hcpu)
1327 int cpu = (long)hcpu;
1329 case CPU_UP_PREPARE:
1330 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1338 static struct notifier_block __cpuinitdata xen_hvm_cpu_notifier = {
1339 .notifier_call = xen_hvm_cpu_notify,
1342 static void __init xen_hvm_guest_init(void)
1347 r = init_hvm_pv_info(&major, &minor);
1351 xen_hvm_init_shared_info();
1353 if (xen_feature(XENFEAT_hvm_callback_vector))
1354 xen_have_vector_callback = 1;
1355 register_cpu_notifier(&xen_hvm_cpu_notifier);
1356 xen_unplug_emulated_devices();
1357 have_vcpu_info_placement = 0;
1358 x86_init.irqs.intr_init = xen_init_IRQ;
1359 xen_hvm_init_time_ops();
1360 xen_hvm_init_mmu_ops();
1363 static bool __init xen_hvm_platform(void)
1365 if (xen_pv_domain())
1368 if (!xen_cpuid_base())
1374 const __refconst struct hypervisor_x86 x86_hyper_xen_hvm = {
1376 .detect = xen_hvm_platform,
1377 .init_platform = xen_hvm_guest_init,
1379 EXPORT_SYMBOL(x86_hyper_xen_hvm);