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 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 /* Check to see if the hypervisor will put the vcpu_info
140 structure where we want it, which allows direct access via
141 a percpu-variable. */
142 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
145 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
146 have_vcpu_info_placement = 0;
149 /* This cpu is using the registered vcpu info, even if
150 later ones fail to. */
151 per_cpu(xen_vcpu, cpu) = vcpup;
156 * On restore, set the vcpu placement up again.
157 * If it fails, then we're in a bad state, since
158 * we can't back out from using it...
160 void xen_vcpu_restore(void)
164 for_each_online_cpu(cpu) {
165 bool other_cpu = (cpu != smp_processor_id());
168 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
171 xen_setup_runstate_info(cpu);
173 if (have_vcpu_info_placement)
177 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
182 static void __init xen_banner(void)
184 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
185 struct xen_extraversion extra;
186 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
188 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
190 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
191 version >> 16, version & 0xffff, extra.extraversion,
192 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
195 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
196 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
198 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
199 unsigned int *cx, unsigned int *dx)
201 unsigned maskebx = ~0;
202 unsigned maskecx = ~0;
203 unsigned maskedx = ~0;
206 * Mask out inconvenient features, to try and disable as many
207 * unsupported kernel subsystems as possible.
211 maskecx = cpuid_leaf1_ecx_mask;
212 maskedx = cpuid_leaf1_edx_mask;
216 /* Suppress extended topology stuff */
221 asm(XEN_EMULATE_PREFIX "cpuid"
226 : "0" (*ax), "2" (*cx));
233 static __init void xen_init_cpuid_mask(void)
235 unsigned int ax, bx, cx, dx;
237 cpuid_leaf1_edx_mask =
238 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
239 (1 << X86_FEATURE_MCA) | /* disable MCA */
240 (1 << X86_FEATURE_MTRR) | /* disable MTRR */
241 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
243 if (!xen_initial_domain())
244 cpuid_leaf1_edx_mask &=
245 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
246 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
250 xen_cpuid(&ax, &bx, &cx, &dx);
252 /* cpuid claims we support xsave; try enabling it to see what happens */
253 if (cx & (1 << (X86_FEATURE_XSAVE % 32))) {
256 set_in_cr4(X86_CR4_OSXSAVE);
260 if ((cr4 & X86_CR4_OSXSAVE) == 0)
261 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32));
263 clear_in_cr4(X86_CR4_OSXSAVE);
267 static void xen_set_debugreg(int reg, unsigned long val)
269 HYPERVISOR_set_debugreg(reg, val);
272 static unsigned long xen_get_debugreg(int reg)
274 return HYPERVISOR_get_debugreg(reg);
277 static void xen_end_context_switch(struct task_struct *next)
280 paravirt_end_context_switch(next);
283 static unsigned long xen_store_tr(void)
289 * Set the page permissions for a particular virtual address. If the
290 * address is a vmalloc mapping (or other non-linear mapping), then
291 * find the linear mapping of the page and also set its protections to
294 static void set_aliased_prot(void *v, pgprot_t prot)
302 ptep = lookup_address((unsigned long)v, &level);
303 BUG_ON(ptep == NULL);
305 pfn = pte_pfn(*ptep);
306 page = pfn_to_page(pfn);
308 pte = pfn_pte(pfn, prot);
310 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
313 if (!PageHighMem(page)) {
314 void *av = __va(PFN_PHYS(pfn));
317 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
323 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
325 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
328 for(i = 0; i < entries; i += entries_per_page)
329 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
332 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
334 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
337 for(i = 0; i < entries; i += entries_per_page)
338 set_aliased_prot(ldt + i, PAGE_KERNEL);
341 static void xen_set_ldt(const void *addr, unsigned entries)
343 struct mmuext_op *op;
344 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
347 op->cmd = MMUEXT_SET_LDT;
348 op->arg1.linear_addr = (unsigned long)addr;
349 op->arg2.nr_ents = entries;
351 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
353 xen_mc_issue(PARAVIRT_LAZY_CPU);
356 static void xen_load_gdt(const struct desc_ptr *dtr)
358 unsigned long va = dtr->address;
359 unsigned int size = dtr->size + 1;
360 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
361 unsigned long frames[pages];
365 * A GDT can be up to 64k in size, which corresponds to 8192
366 * 8-byte entries, or 16 4k pages..
369 BUG_ON(size > 65536);
370 BUG_ON(va & ~PAGE_MASK);
372 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
375 unsigned long pfn, mfn;
379 * The GDT is per-cpu and is in the percpu data area.
380 * That can be virtually mapped, so we need to do a
381 * page-walk to get the underlying MFN for the
382 * hypercall. The page can also be in the kernel's
383 * linear range, so we need to RO that mapping too.
385 ptep = lookup_address(va, &level);
386 BUG_ON(ptep == NULL);
388 pfn = pte_pfn(*ptep);
389 mfn = pfn_to_mfn(pfn);
390 virt = __va(PFN_PHYS(pfn));
394 make_lowmem_page_readonly((void *)va);
395 make_lowmem_page_readonly(virt);
398 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
403 * load_gdt for early boot, when the gdt is only mapped once
405 static __init void xen_load_gdt_boot(const struct desc_ptr *dtr)
407 unsigned long va = dtr->address;
408 unsigned int size = dtr->size + 1;
409 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
410 unsigned long frames[pages];
414 * A GDT can be up to 64k in size, which corresponds to 8192
415 * 8-byte entries, or 16 4k pages..
418 BUG_ON(size > 65536);
419 BUG_ON(va & ~PAGE_MASK);
421 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
423 unsigned long pfn, mfn;
425 pfn = virt_to_pfn(va);
426 mfn = pfn_to_mfn(pfn);
428 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
430 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
436 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
440 static void load_TLS_descriptor(struct thread_struct *t,
441 unsigned int cpu, unsigned int i)
443 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
444 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
445 struct multicall_space mc = __xen_mc_entry(0);
447 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
450 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
453 * XXX sleazy hack: If we're being called in a lazy-cpu zone
454 * and lazy gs handling is enabled, it means we're in a
455 * context switch, and %gs has just been saved. This means we
456 * can zero it out to prevent faults on exit from the
457 * hypervisor if the next process has no %gs. Either way, it
458 * has been saved, and the new value will get loaded properly.
459 * This will go away as soon as Xen has been modified to not
460 * save/restore %gs for normal hypercalls.
462 * On x86_64, this hack is not used for %gs, because gs points
463 * to KERNEL_GS_BASE (and uses it for PDA references), so we
464 * must not zero %gs on x86_64
466 * For x86_64, we need to zero %fs, otherwise we may get an
467 * exception between the new %fs descriptor being loaded and
468 * %fs being effectively cleared at __switch_to().
470 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
480 load_TLS_descriptor(t, cpu, 0);
481 load_TLS_descriptor(t, cpu, 1);
482 load_TLS_descriptor(t, cpu, 2);
484 xen_mc_issue(PARAVIRT_LAZY_CPU);
488 static void xen_load_gs_index(unsigned int idx)
490 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
495 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
498 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
499 u64 entry = *(u64 *)ptr;
504 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
510 static int cvt_gate_to_trap(int vector, const gate_desc *val,
511 struct trap_info *info)
515 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
518 info->vector = vector;
520 addr = gate_offset(*val);
523 * Look for known traps using IST, and substitute them
524 * appropriately. The debugger ones are the only ones we care
525 * about. Xen will handle faults like double_fault and
526 * machine_check, so we should never see them. Warn if
527 * there's an unexpected IST-using fault handler.
529 if (addr == (unsigned long)debug)
530 addr = (unsigned long)xen_debug;
531 else if (addr == (unsigned long)int3)
532 addr = (unsigned long)xen_int3;
533 else if (addr == (unsigned long)stack_segment)
534 addr = (unsigned long)xen_stack_segment;
535 else if (addr == (unsigned long)double_fault ||
536 addr == (unsigned long)nmi) {
537 /* Don't need to handle these */
539 #ifdef CONFIG_X86_MCE
540 } else if (addr == (unsigned long)machine_check) {
544 /* Some other trap using IST? */
545 if (WARN_ON(val->ist != 0))
548 #endif /* CONFIG_X86_64 */
549 info->address = addr;
551 info->cs = gate_segment(*val);
552 info->flags = val->dpl;
553 /* interrupt gates clear IF */
554 if (val->type == GATE_INTERRUPT)
555 info->flags |= 1 << 2;
560 /* Locations of each CPU's IDT */
561 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
563 /* Set an IDT entry. If the entry is part of the current IDT, then
565 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
567 unsigned long p = (unsigned long)&dt[entrynum];
568 unsigned long start, end;
572 start = __get_cpu_var(idt_desc).address;
573 end = start + __get_cpu_var(idt_desc).size + 1;
577 native_write_idt_entry(dt, entrynum, g);
579 if (p >= start && (p + 8) <= end) {
580 struct trap_info info[2];
584 if (cvt_gate_to_trap(entrynum, g, &info[0]))
585 if (HYPERVISOR_set_trap_table(info))
592 static void xen_convert_trap_info(const struct desc_ptr *desc,
593 struct trap_info *traps)
595 unsigned in, out, count;
597 count = (desc->size+1) / sizeof(gate_desc);
600 for (in = out = 0; in < count; in++) {
601 gate_desc *entry = (gate_desc*)(desc->address) + in;
603 if (cvt_gate_to_trap(in, entry, &traps[out]))
606 traps[out].address = 0;
609 void xen_copy_trap_info(struct trap_info *traps)
611 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
613 xen_convert_trap_info(desc, traps);
616 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
617 hold a spinlock to protect the static traps[] array (static because
618 it avoids allocation, and saves stack space). */
619 static void xen_load_idt(const struct desc_ptr *desc)
621 static DEFINE_SPINLOCK(lock);
622 static struct trap_info traps[257];
626 __get_cpu_var(idt_desc) = *desc;
628 xen_convert_trap_info(desc, traps);
631 if (HYPERVISOR_set_trap_table(traps))
637 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
638 they're handled differently. */
639 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
640 const void *desc, int type)
651 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
654 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
664 * Version of write_gdt_entry for use at early boot-time needed to
665 * update an entry as simply as possible.
667 static __init void xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
668 const void *desc, int type)
677 xmaddr_t maddr = virt_to_machine(&dt[entry]);
679 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
680 dt[entry] = *(struct desc_struct *)desc;
686 static void xen_load_sp0(struct tss_struct *tss,
687 struct thread_struct *thread)
689 struct multicall_space mcs = xen_mc_entry(0);
690 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
691 xen_mc_issue(PARAVIRT_LAZY_CPU);
694 static void xen_set_iopl_mask(unsigned mask)
696 struct physdev_set_iopl set_iopl;
698 /* Force the change at ring 0. */
699 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
700 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
703 static void xen_io_delay(void)
707 #ifdef CONFIG_X86_LOCAL_APIC
708 static u32 xen_apic_read(u32 reg)
713 static void xen_apic_write(u32 reg, u32 val)
715 /* Warn to see if there's any stray references */
719 static u64 xen_apic_icr_read(void)
724 static void xen_apic_icr_write(u32 low, u32 id)
726 /* Warn to see if there's any stray references */
730 static void xen_apic_wait_icr_idle(void)
735 static u32 xen_safe_apic_wait_icr_idle(void)
740 static void set_xen_basic_apic_ops(void)
742 apic->read = xen_apic_read;
743 apic->write = xen_apic_write;
744 apic->icr_read = xen_apic_icr_read;
745 apic->icr_write = xen_apic_icr_write;
746 apic->wait_icr_idle = xen_apic_wait_icr_idle;
747 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
752 static void xen_clts(void)
754 struct multicall_space mcs;
756 mcs = xen_mc_entry(0);
758 MULTI_fpu_taskswitch(mcs.mc, 0);
760 xen_mc_issue(PARAVIRT_LAZY_CPU);
763 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
765 static unsigned long xen_read_cr0(void)
767 unsigned long cr0 = percpu_read(xen_cr0_value);
769 if (unlikely(cr0 == 0)) {
770 cr0 = native_read_cr0();
771 percpu_write(xen_cr0_value, cr0);
777 static void xen_write_cr0(unsigned long cr0)
779 struct multicall_space mcs;
781 percpu_write(xen_cr0_value, cr0);
783 /* Only pay attention to cr0.TS; everything else is
785 mcs = xen_mc_entry(0);
787 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
789 xen_mc_issue(PARAVIRT_LAZY_CPU);
792 static void xen_write_cr4(unsigned long cr4)
797 native_write_cr4(cr4);
800 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
811 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
812 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
813 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
816 base = ((u64)high << 32) | low;
817 if (HYPERVISOR_set_segment_base(which, base) != 0)
825 case MSR_SYSCALL_MASK:
826 case MSR_IA32_SYSENTER_CS:
827 case MSR_IA32_SYSENTER_ESP:
828 case MSR_IA32_SYSENTER_EIP:
829 /* Fast syscall setup is all done in hypercalls, so
830 these are all ignored. Stub them out here to stop
831 Xen console noise. */
834 case MSR_IA32_CR_PAT:
835 if (smp_processor_id() == 0)
836 xen_set_pat(((u64)high << 32) | low);
840 ret = native_write_msr_safe(msr, low, high);
846 void xen_setup_shared_info(void)
848 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
849 set_fixmap(FIX_PARAVIRT_BOOTMAP,
850 xen_start_info->shared_info);
852 HYPERVISOR_shared_info =
853 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
855 HYPERVISOR_shared_info =
856 (struct shared_info *)__va(xen_start_info->shared_info);
859 /* In UP this is as good a place as any to set up shared info */
860 xen_setup_vcpu_info_placement();
863 xen_setup_mfn_list_list();
866 /* This is called once we have the cpu_possible_map */
867 void xen_setup_vcpu_info_placement(void)
871 for_each_possible_cpu(cpu)
874 /* xen_vcpu_setup managed to place the vcpu_info within the
875 percpu area for all cpus, so make use of it */
876 if (have_vcpu_info_placement) {
877 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
878 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
879 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
880 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
881 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
885 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
886 unsigned long addr, unsigned len)
888 char *start, *end, *reloc;
891 start = end = reloc = NULL;
893 #define SITE(op, x) \
894 case PARAVIRT_PATCH(op.x): \
895 if (have_vcpu_info_placement) { \
896 start = (char *)xen_##x##_direct; \
897 end = xen_##x##_direct_end; \
898 reloc = xen_##x##_direct_reloc; \
903 SITE(pv_irq_ops, irq_enable);
904 SITE(pv_irq_ops, irq_disable);
905 SITE(pv_irq_ops, save_fl);
906 SITE(pv_irq_ops, restore_fl);
910 if (start == NULL || (end-start) > len)
913 ret = paravirt_patch_insns(insnbuf, len, start, end);
915 /* Note: because reloc is assigned from something that
916 appears to be an array, gcc assumes it's non-null,
917 but doesn't know its relationship with start and
919 if (reloc > start && reloc < end) {
920 int reloc_off = reloc - start;
921 long *relocp = (long *)(insnbuf + reloc_off);
922 long delta = start - (char *)addr;
930 ret = paravirt_patch_default(type, clobbers, insnbuf,
938 static const struct pv_info xen_info __initdata = {
939 .paravirt_enabled = 1,
940 .shared_kernel_pmd = 0,
945 static const struct pv_init_ops xen_init_ops __initdata = {
949 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
952 .set_debugreg = xen_set_debugreg,
953 .get_debugreg = xen_get_debugreg,
957 .read_cr0 = xen_read_cr0,
958 .write_cr0 = xen_write_cr0,
960 .read_cr4 = native_read_cr4,
961 .read_cr4_safe = native_read_cr4_safe,
962 .write_cr4 = xen_write_cr4,
964 .wbinvd = native_wbinvd,
966 .read_msr = native_read_msr_safe,
967 .write_msr = xen_write_msr_safe,
968 .read_tsc = native_read_tsc,
969 .read_pmc = native_read_pmc,
972 .irq_enable_sysexit = xen_sysexit,
974 .usergs_sysret32 = xen_sysret32,
975 .usergs_sysret64 = xen_sysret64,
978 .load_tr_desc = paravirt_nop,
979 .set_ldt = xen_set_ldt,
980 .load_gdt = xen_load_gdt,
981 .load_idt = xen_load_idt,
982 .load_tls = xen_load_tls,
984 .load_gs_index = xen_load_gs_index,
987 .alloc_ldt = xen_alloc_ldt,
988 .free_ldt = xen_free_ldt,
990 .store_gdt = native_store_gdt,
991 .store_idt = native_store_idt,
992 .store_tr = xen_store_tr,
994 .write_ldt_entry = xen_write_ldt_entry,
995 .write_gdt_entry = xen_write_gdt_entry,
996 .write_idt_entry = xen_write_idt_entry,
997 .load_sp0 = xen_load_sp0,
999 .set_iopl_mask = xen_set_iopl_mask,
1000 .io_delay = xen_io_delay,
1002 /* Xen takes care of %gs when switching to usermode for us */
1003 .swapgs = paravirt_nop,
1005 .start_context_switch = paravirt_start_context_switch,
1006 .end_context_switch = xen_end_context_switch,
1009 static const struct pv_apic_ops xen_apic_ops __initdata = {
1010 #ifdef CONFIG_X86_LOCAL_APIC
1011 .startup_ipi_hook = paravirt_nop,
1015 static void xen_reboot(int reason)
1017 struct sched_shutdown r = { .reason = reason };
1019 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1023 static void xen_restart(char *msg)
1025 xen_reboot(SHUTDOWN_reboot);
1028 static void xen_emergency_restart(void)
1030 xen_reboot(SHUTDOWN_reboot);
1033 static void xen_machine_halt(void)
1035 xen_reboot(SHUTDOWN_poweroff);
1038 static void xen_crash_shutdown(struct pt_regs *regs)
1040 xen_reboot(SHUTDOWN_crash);
1044 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1046 xen_reboot(SHUTDOWN_crash);
1050 static struct notifier_block xen_panic_block = {
1051 .notifier_call= xen_panic_event,
1054 int xen_panic_handler_init(void)
1056 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1060 static const struct machine_ops __initdata xen_machine_ops = {
1061 .restart = xen_restart,
1062 .halt = xen_machine_halt,
1063 .power_off = xen_machine_halt,
1064 .shutdown = xen_machine_halt,
1065 .crash_shutdown = xen_crash_shutdown,
1066 .emergency_restart = xen_emergency_restart,
1070 * Set up the GDT and segment registers for -fstack-protector. Until
1071 * we do this, we have to be careful not to call any stack-protected
1072 * function, which is most of the kernel.
1074 static void __init xen_setup_stackprotector(void)
1076 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1077 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1079 setup_stack_canary_segment(0);
1080 switch_to_new_gdt(0);
1082 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1083 pv_cpu_ops.load_gdt = xen_load_gdt;
1086 /* First C function to be called on Xen boot */
1087 asmlinkage void __init xen_start_kernel(void)
1091 if (!xen_start_info)
1094 xen_domain_type = XEN_PV_DOMAIN;
1096 /* Install Xen paravirt ops */
1098 pv_init_ops = xen_init_ops;
1099 pv_cpu_ops = xen_cpu_ops;
1100 pv_apic_ops = xen_apic_ops;
1102 x86_init.resources.memory_setup = xen_memory_setup;
1103 x86_init.oem.arch_setup = xen_arch_setup;
1104 x86_init.oem.banner = xen_banner;
1106 xen_init_time_ops();
1109 * Set up some pagetable state before starting to set any ptes.
1114 /* Prevent unwanted bits from being set in PTEs. */
1115 __supported_pte_mask &= ~_PAGE_GLOBAL;
1116 if (!xen_initial_domain())
1117 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1119 __supported_pte_mask |= _PAGE_IOMAP;
1122 * Prevent page tables from being allocated in highmem, even
1123 * if CONFIG_HIGHPTE is enabled.
1125 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1127 /* Work out if we support NX */
1130 xen_setup_features();
1133 if (!xen_feature(XENFEAT_auto_translated_physmap))
1134 xen_build_dynamic_phys_to_machine();
1137 * Set up kernel GDT and segment registers, mainly so that
1138 * -fstack-protector code can be executed.
1140 xen_setup_stackprotector();
1143 xen_init_cpuid_mask();
1145 #ifdef CONFIG_X86_LOCAL_APIC
1147 * set up the basic apic ops.
1149 set_xen_basic_apic_ops();
1152 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1153 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1154 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1157 machine_ops = xen_machine_ops;
1160 * The only reliable way to retain the initial address of the
1161 * percpu gdt_page is to remember it here, so we can go and
1162 * mark it RW later, when the initial percpu area is freed.
1164 xen_initial_gdt = &per_cpu(gdt_page, 0);
1168 pgd = (pgd_t *)xen_start_info->pt_base;
1170 if (!xen_initial_domain())
1171 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1173 __supported_pte_mask |= _PAGE_IOMAP;
1174 /* Don't do the full vcpu_info placement stuff until we have a
1175 possible map and a non-dummy shared_info. */
1176 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1178 local_irq_disable();
1179 early_boot_irqs_off();
1183 xen_raw_console_write("mapping kernel into physical memory\n");
1184 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1185 xen_ident_map_ISA();
1187 /* Allocate and initialize top and mid mfn levels for p2m structure */
1188 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;
1202 /* set the limit of our address space */
1205 #ifdef CONFIG_X86_32
1206 /* set up basic CPUID stuff */
1207 cpu_detect(&new_cpu_data);
1208 new_cpu_data.hard_math = 1;
1209 new_cpu_data.wp_works_ok = 1;
1210 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1213 /* Poke various useful things into boot_params */
1214 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1215 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1216 ? __pa(xen_start_info->mod_start) : 0;
1217 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1218 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1220 if (!xen_initial_domain()) {
1221 add_preferred_console("xenboot", 0, NULL);
1222 add_preferred_console("tty", 0, NULL);
1223 add_preferred_console("hvc", 0, NULL);
1225 x86_init.pci.arch_init = pci_xen_init;
1227 /* Make sure ACS will be enabled */
1232 xen_raw_console_write("about to get started...\n");
1234 xen_setup_runstate_info(0);
1236 /* Start the world */
1237 #ifdef CONFIG_X86_32
1238 i386_start_kernel();
1240 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1244 static uint32_t xen_cpuid_base(void)
1246 uint32_t base, eax, ebx, ecx, edx;
1249 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
1250 cpuid(base, &eax, &ebx, &ecx, &edx);
1251 *(uint32_t *)(signature + 0) = ebx;
1252 *(uint32_t *)(signature + 4) = ecx;
1253 *(uint32_t *)(signature + 8) = edx;
1256 if (!strcmp("XenVMMXenVMM", signature) && ((eax - base) >= 2))
1263 static int init_hvm_pv_info(int *major, int *minor)
1265 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1268 base = xen_cpuid_base();
1269 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1272 *minor = eax & 0xffff;
1273 printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1275 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1277 pfn = __pa(hypercall_page);
1278 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1280 xen_setup_features();
1283 pv_info.kernel_rpl = 0;
1285 xen_domain_type = XEN_HVM_DOMAIN;
1290 void xen_hvm_init_shared_info(void)
1293 struct xen_add_to_physmap xatp;
1294 static struct shared_info *shared_info_page = 0;
1296 if (!shared_info_page)
1297 shared_info_page = (struct shared_info *)
1298 extend_brk(PAGE_SIZE, PAGE_SIZE);
1299 xatp.domid = DOMID_SELF;
1301 xatp.space = XENMAPSPACE_shared_info;
1302 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1303 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1306 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1308 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1309 * page, we use it in the event channel upcall and in some pvclock
1310 * related functions. We don't need the vcpu_info placement
1311 * optimizations because we don't use any pv_mmu or pv_irq op on
1313 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1314 * online but xen_hvm_init_shared_info is run at resume time too and
1315 * in that case multiple vcpus might be online. */
1316 for_each_online_cpu(cpu) {
1317 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1321 #ifdef CONFIG_XEN_PVHVM
1322 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1323 unsigned long action, void *hcpu)
1325 int cpu = (long)hcpu;
1327 case CPU_UP_PREPARE:
1328 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1336 static struct notifier_block __cpuinitdata xen_hvm_cpu_notifier = {
1337 .notifier_call = xen_hvm_cpu_notify,
1340 static void __init xen_hvm_guest_init(void)
1345 r = init_hvm_pv_info(&major, &minor);
1349 xen_hvm_init_shared_info();
1351 if (xen_feature(XENFEAT_hvm_callback_vector))
1352 xen_have_vector_callback = 1;
1353 register_cpu_notifier(&xen_hvm_cpu_notifier);
1354 xen_unplug_emulated_devices();
1355 have_vcpu_info_placement = 0;
1356 x86_init.irqs.intr_init = xen_init_IRQ;
1357 xen_hvm_init_time_ops();
1358 xen_hvm_init_mmu_ops();
1361 static bool __init xen_hvm_platform(void)
1363 if (xen_pv_domain())
1366 if (!xen_cpuid_base())
1372 const __refconst struct hypervisor_x86 x86_hyper_xen_hvm = {
1374 .detect = xen_hvm_platform,
1375 .init_platform = xen_hvm_guest_init,
1377 EXPORT_SYMBOL(x86_hyper_xen_hvm);