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);
133 * This path is called twice on PVHVM - first during bootup via
134 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
135 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
136 * As we can only do the VCPUOP_register_vcpu_info once lets
137 * not over-write its result.
139 * For PV it is called during restore (xen_vcpu_restore) and bootup
140 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
143 if (xen_hvm_domain()) {
144 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
147 if (cpu < MAX_VIRT_CPUS)
148 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
150 if (!have_vcpu_info_placement) {
151 if (cpu >= MAX_VIRT_CPUS)
156 vcpup = &per_cpu(xen_vcpu_info, cpu);
157 info.mfn = arbitrary_virt_to_mfn(vcpup);
158 info.offset = offset_in_page(vcpup);
160 /* Check to see if the hypervisor will put the vcpu_info
161 structure where we want it, which allows direct access via
162 a percpu-variable. */
163 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
166 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
167 have_vcpu_info_placement = 0;
170 /* This cpu is using the registered vcpu info, even if
171 later ones fail to. */
172 per_cpu(xen_vcpu, cpu) = vcpup;
177 * On restore, set the vcpu placement up again.
178 * If it fails, then we're in a bad state, since
179 * we can't back out from using it...
181 void xen_vcpu_restore(void)
185 for_each_online_cpu(cpu) {
186 bool other_cpu = (cpu != smp_processor_id());
189 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
192 xen_setup_runstate_info(cpu);
194 if (have_vcpu_info_placement)
198 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
203 static void __init xen_banner(void)
205 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
206 struct xen_extraversion extra;
207 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
209 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
211 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
212 version >> 16, version & 0xffff, extra.extraversion,
213 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
216 #define CPUID_THERM_POWER_LEAF 6
217 #define APERFMPERF_PRESENT 0
219 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
220 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
222 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
223 unsigned int *cx, unsigned int *dx)
225 unsigned maskebx = ~0;
226 unsigned maskecx = ~0;
227 unsigned maskedx = ~0;
230 * Mask out inconvenient features, to try and disable as many
231 * unsupported kernel subsystems as possible.
235 maskecx = cpuid_leaf1_ecx_mask;
236 maskedx = cpuid_leaf1_edx_mask;
239 case CPUID_THERM_POWER_LEAF:
240 /* Disabling APERFMPERF for kernel usage */
241 maskecx = ~(1 << APERFMPERF_PRESENT);
245 /* Suppress extended topology stuff */
250 asm(XEN_EMULATE_PREFIX "cpuid"
255 : "0" (*ax), "2" (*cx));
262 static void __init xen_init_cpuid_mask(void)
264 unsigned int ax, bx, cx, dx;
265 unsigned int xsave_mask;
267 cpuid_leaf1_edx_mask =
268 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
269 (1 << X86_FEATURE_MCA) | /* disable MCA */
270 (1 << X86_FEATURE_MTRR) | /* disable MTRR */
271 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
273 if (!xen_initial_domain())
274 cpuid_leaf1_edx_mask &=
275 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
276 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
279 xen_cpuid(&ax, &bx, &cx, &dx);
282 (1 << (X86_FEATURE_XSAVE % 32)) |
283 (1 << (X86_FEATURE_OSXSAVE % 32));
285 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
286 if ((cx & xsave_mask) != xsave_mask)
287 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
290 static void xen_set_debugreg(int reg, unsigned long val)
292 HYPERVISOR_set_debugreg(reg, val);
295 static unsigned long xen_get_debugreg(int reg)
297 return HYPERVISOR_get_debugreg(reg);
300 static void xen_end_context_switch(struct task_struct *next)
303 paravirt_end_context_switch(next);
306 static unsigned long xen_store_tr(void)
312 * Set the page permissions for a particular virtual address. If the
313 * address is a vmalloc mapping (or other non-linear mapping), then
314 * find the linear mapping of the page and also set its protections to
317 static void set_aliased_prot(void *v, pgprot_t prot)
325 ptep = lookup_address((unsigned long)v, &level);
326 BUG_ON(ptep == NULL);
328 pfn = pte_pfn(*ptep);
329 page = pfn_to_page(pfn);
331 pte = pfn_pte(pfn, prot);
333 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
336 if (!PageHighMem(page)) {
337 void *av = __va(PFN_PHYS(pfn));
340 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
346 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
348 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
351 for(i = 0; i < entries; i += entries_per_page)
352 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
355 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
357 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
360 for(i = 0; i < entries; i += entries_per_page)
361 set_aliased_prot(ldt + i, PAGE_KERNEL);
364 static void xen_set_ldt(const void *addr, unsigned entries)
366 struct mmuext_op *op;
367 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
369 trace_xen_cpu_set_ldt(addr, entries);
372 op->cmd = MMUEXT_SET_LDT;
373 op->arg1.linear_addr = (unsigned long)addr;
374 op->arg2.nr_ents = entries;
376 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
378 xen_mc_issue(PARAVIRT_LAZY_CPU);
381 static void xen_load_gdt(const struct desc_ptr *dtr)
383 unsigned long va = dtr->address;
384 unsigned int size = dtr->size + 1;
385 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
386 unsigned long frames[pages];
390 * A GDT can be up to 64k in size, which corresponds to 8192
391 * 8-byte entries, or 16 4k pages..
394 BUG_ON(size > 65536);
395 BUG_ON(va & ~PAGE_MASK);
397 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
400 unsigned long pfn, mfn;
404 * The GDT is per-cpu and is in the percpu data area.
405 * That can be virtually mapped, so we need to do a
406 * page-walk to get the underlying MFN for the
407 * hypercall. The page can also be in the kernel's
408 * linear range, so we need to RO that mapping too.
410 ptep = lookup_address(va, &level);
411 BUG_ON(ptep == NULL);
413 pfn = pte_pfn(*ptep);
414 mfn = pfn_to_mfn(pfn);
415 virt = __va(PFN_PHYS(pfn));
419 make_lowmem_page_readonly((void *)va);
420 make_lowmem_page_readonly(virt);
423 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
428 * load_gdt for early boot, when the gdt is only mapped once
430 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
432 unsigned long va = dtr->address;
433 unsigned int size = dtr->size + 1;
434 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
435 unsigned long frames[pages];
439 * A GDT can be up to 64k in size, which corresponds to 8192
440 * 8-byte entries, or 16 4k pages..
443 BUG_ON(size > 65536);
444 BUG_ON(va & ~PAGE_MASK);
446 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
448 unsigned long pfn, mfn;
450 pfn = virt_to_pfn(va);
451 mfn = pfn_to_mfn(pfn);
453 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
455 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
461 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
465 static void load_TLS_descriptor(struct thread_struct *t,
466 unsigned int cpu, unsigned int i)
468 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
469 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
470 struct multicall_space mc = __xen_mc_entry(0);
472 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
475 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
478 * XXX sleazy hack: If we're being called in a lazy-cpu zone
479 * and lazy gs handling is enabled, it means we're in a
480 * context switch, and %gs has just been saved. This means we
481 * can zero it out to prevent faults on exit from the
482 * hypervisor if the next process has no %gs. Either way, it
483 * has been saved, and the new value will get loaded properly.
484 * This will go away as soon as Xen has been modified to not
485 * save/restore %gs for normal hypercalls.
487 * On x86_64, this hack is not used for %gs, because gs points
488 * to KERNEL_GS_BASE (and uses it for PDA references), so we
489 * must not zero %gs on x86_64
491 * For x86_64, we need to zero %fs, otherwise we may get an
492 * exception between the new %fs descriptor being loaded and
493 * %fs being effectively cleared at __switch_to().
495 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
505 load_TLS_descriptor(t, cpu, 0);
506 load_TLS_descriptor(t, cpu, 1);
507 load_TLS_descriptor(t, cpu, 2);
509 xen_mc_issue(PARAVIRT_LAZY_CPU);
513 static void xen_load_gs_index(unsigned int idx)
515 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
520 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
523 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
524 u64 entry = *(u64 *)ptr;
526 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
531 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
537 static int cvt_gate_to_trap(int vector, const gate_desc *val,
538 struct trap_info *info)
542 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
545 info->vector = vector;
547 addr = gate_offset(*val);
550 * Look for known traps using IST, and substitute them
551 * appropriately. The debugger ones are the only ones we care
552 * about. Xen will handle faults like double_fault and
553 * machine_check, so we should never see them. Warn if
554 * there's an unexpected IST-using fault handler.
556 if (addr == (unsigned long)debug)
557 addr = (unsigned long)xen_debug;
558 else if (addr == (unsigned long)int3)
559 addr = (unsigned long)xen_int3;
560 else if (addr == (unsigned long)stack_segment)
561 addr = (unsigned long)xen_stack_segment;
562 else if (addr == (unsigned long)double_fault ||
563 addr == (unsigned long)nmi) {
564 /* Don't need to handle these */
566 #ifdef CONFIG_X86_MCE
567 } else if (addr == (unsigned long)machine_check) {
571 /* Some other trap using IST? */
572 if (WARN_ON(val->ist != 0))
575 #endif /* CONFIG_X86_64 */
576 info->address = addr;
578 info->cs = gate_segment(*val);
579 info->flags = val->dpl;
580 /* interrupt gates clear IF */
581 if (val->type == GATE_INTERRUPT)
582 info->flags |= 1 << 2;
587 /* Locations of each CPU's IDT */
588 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
590 /* Set an IDT entry. If the entry is part of the current IDT, then
592 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
594 unsigned long p = (unsigned long)&dt[entrynum];
595 unsigned long start, end;
597 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
601 start = __this_cpu_read(idt_desc.address);
602 end = start + __this_cpu_read(idt_desc.size) + 1;
606 native_write_idt_entry(dt, entrynum, g);
608 if (p >= start && (p + 8) <= end) {
609 struct trap_info info[2];
613 if (cvt_gate_to_trap(entrynum, g, &info[0]))
614 if (HYPERVISOR_set_trap_table(info))
621 static void xen_convert_trap_info(const struct desc_ptr *desc,
622 struct trap_info *traps)
624 unsigned in, out, count;
626 count = (desc->size+1) / sizeof(gate_desc);
629 for (in = out = 0; in < count; in++) {
630 gate_desc *entry = (gate_desc*)(desc->address) + in;
632 if (cvt_gate_to_trap(in, entry, &traps[out]))
635 traps[out].address = 0;
638 void xen_copy_trap_info(struct trap_info *traps)
640 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
642 xen_convert_trap_info(desc, traps);
645 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
646 hold a spinlock to protect the static traps[] array (static because
647 it avoids allocation, and saves stack space). */
648 static void xen_load_idt(const struct desc_ptr *desc)
650 static DEFINE_SPINLOCK(lock);
651 static struct trap_info traps[257];
653 trace_xen_cpu_load_idt(desc);
657 __get_cpu_var(idt_desc) = *desc;
659 xen_convert_trap_info(desc, traps);
662 if (HYPERVISOR_set_trap_table(traps))
668 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
669 they're handled differently. */
670 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
671 const void *desc, int type)
673 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
684 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
687 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
697 * Version of write_gdt_entry for use at early boot-time needed to
698 * update an entry as simply as possible.
700 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
701 const void *desc, int type)
703 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
712 xmaddr_t maddr = virt_to_machine(&dt[entry]);
714 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
715 dt[entry] = *(struct desc_struct *)desc;
721 static void xen_load_sp0(struct tss_struct *tss,
722 struct thread_struct *thread)
724 struct multicall_space mcs;
726 mcs = xen_mc_entry(0);
727 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
728 xen_mc_issue(PARAVIRT_LAZY_CPU);
731 static void xen_set_iopl_mask(unsigned mask)
733 struct physdev_set_iopl set_iopl;
735 /* Force the change at ring 0. */
736 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
737 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
740 static void xen_io_delay(void)
744 #ifdef CONFIG_X86_LOCAL_APIC
745 static u32 xen_apic_read(u32 reg)
750 static void xen_apic_write(u32 reg, u32 val)
752 /* Warn to see if there's any stray references */
756 static u64 xen_apic_icr_read(void)
761 static void xen_apic_icr_write(u32 low, u32 id)
763 /* Warn to see if there's any stray references */
767 static void xen_apic_wait_icr_idle(void)
772 static u32 xen_safe_apic_wait_icr_idle(void)
777 static void set_xen_basic_apic_ops(void)
779 apic->read = xen_apic_read;
780 apic->write = xen_apic_write;
781 apic->icr_read = xen_apic_icr_read;
782 apic->icr_write = xen_apic_icr_write;
783 apic->wait_icr_idle = xen_apic_wait_icr_idle;
784 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
789 static void xen_clts(void)
791 struct multicall_space mcs;
793 mcs = xen_mc_entry(0);
795 MULTI_fpu_taskswitch(mcs.mc, 0);
797 xen_mc_issue(PARAVIRT_LAZY_CPU);
800 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
802 static unsigned long xen_read_cr0(void)
804 unsigned long cr0 = percpu_read(xen_cr0_value);
806 if (unlikely(cr0 == 0)) {
807 cr0 = native_read_cr0();
808 percpu_write(xen_cr0_value, cr0);
814 static void xen_write_cr0(unsigned long cr0)
816 struct multicall_space mcs;
818 percpu_write(xen_cr0_value, cr0);
820 /* Only pay attention to cr0.TS; everything else is
822 mcs = xen_mc_entry(0);
824 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
826 xen_mc_issue(PARAVIRT_LAZY_CPU);
829 static void xen_write_cr4(unsigned long cr4)
834 native_write_cr4(cr4);
837 static inline unsigned long xen_read_cr8(void)
841 static inline void xen_write_cr8(unsigned long val)
846 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
857 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
858 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
859 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
862 base = ((u64)high << 32) | low;
863 if (HYPERVISOR_set_segment_base(which, base) != 0)
871 case MSR_SYSCALL_MASK:
872 case MSR_IA32_SYSENTER_CS:
873 case MSR_IA32_SYSENTER_ESP:
874 case MSR_IA32_SYSENTER_EIP:
875 /* Fast syscall setup is all done in hypercalls, so
876 these are all ignored. Stub them out here to stop
877 Xen console noise. */
880 case MSR_IA32_CR_PAT:
881 if (smp_processor_id() == 0)
882 xen_set_pat(((u64)high << 32) | low);
886 ret = native_write_msr_safe(msr, low, high);
892 void xen_setup_shared_info(void)
894 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
895 set_fixmap(FIX_PARAVIRT_BOOTMAP,
896 xen_start_info->shared_info);
898 HYPERVISOR_shared_info =
899 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
901 HYPERVISOR_shared_info =
902 (struct shared_info *)__va(xen_start_info->shared_info);
905 /* In UP this is as good a place as any to set up shared info */
906 xen_setup_vcpu_info_placement();
909 xen_setup_mfn_list_list();
912 /* This is called once we have the cpu_possible_map */
913 void xen_setup_vcpu_info_placement(void)
917 for_each_possible_cpu(cpu)
920 /* xen_vcpu_setup managed to place the vcpu_info within the
921 percpu area for all cpus, so make use of it */
922 if (have_vcpu_info_placement) {
923 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
924 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
925 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
926 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
927 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
931 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
932 unsigned long addr, unsigned len)
934 char *start, *end, *reloc;
937 start = end = reloc = NULL;
939 #define SITE(op, x) \
940 case PARAVIRT_PATCH(op.x): \
941 if (have_vcpu_info_placement) { \
942 start = (char *)xen_##x##_direct; \
943 end = xen_##x##_direct_end; \
944 reloc = xen_##x##_direct_reloc; \
949 SITE(pv_irq_ops, irq_enable);
950 SITE(pv_irq_ops, irq_disable);
951 SITE(pv_irq_ops, save_fl);
952 SITE(pv_irq_ops, restore_fl);
956 if (start == NULL || (end-start) > len)
959 ret = paravirt_patch_insns(insnbuf, len, start, end);
961 /* Note: because reloc is assigned from something that
962 appears to be an array, gcc assumes it's non-null,
963 but doesn't know its relationship with start and
965 if (reloc > start && reloc < end) {
966 int reloc_off = reloc - start;
967 long *relocp = (long *)(insnbuf + reloc_off);
968 long delta = start - (char *)addr;
976 ret = paravirt_patch_default(type, clobbers, insnbuf,
984 static const struct pv_info xen_info __initconst = {
985 .paravirt_enabled = 1,
986 .shared_kernel_pmd = 0,
989 .extra_user_64bit_cs = FLAT_USER_CS64,
995 static const struct pv_init_ops xen_init_ops __initconst = {
999 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1002 .set_debugreg = xen_set_debugreg,
1003 .get_debugreg = xen_get_debugreg,
1007 .read_cr0 = xen_read_cr0,
1008 .write_cr0 = xen_write_cr0,
1010 .read_cr4 = native_read_cr4,
1011 .read_cr4_safe = native_read_cr4_safe,
1012 .write_cr4 = xen_write_cr4,
1014 #ifdef CONFIG_X86_64
1015 .read_cr8 = xen_read_cr8,
1016 .write_cr8 = xen_write_cr8,
1019 .wbinvd = native_wbinvd,
1021 .read_msr = native_read_msr_safe,
1022 .rdmsr_regs = native_rdmsr_safe_regs,
1023 .write_msr = xen_write_msr_safe,
1024 .wrmsr_regs = native_wrmsr_safe_regs,
1026 .read_tsc = native_read_tsc,
1027 .read_pmc = native_read_pmc,
1029 .read_tscp = native_read_tscp,
1032 .irq_enable_sysexit = xen_sysexit,
1033 #ifdef CONFIG_X86_64
1034 .usergs_sysret32 = xen_sysret32,
1035 .usergs_sysret64 = xen_sysret64,
1038 .load_tr_desc = paravirt_nop,
1039 .set_ldt = xen_set_ldt,
1040 .load_gdt = xen_load_gdt,
1041 .load_idt = xen_load_idt,
1042 .load_tls = xen_load_tls,
1043 #ifdef CONFIG_X86_64
1044 .load_gs_index = xen_load_gs_index,
1047 .alloc_ldt = xen_alloc_ldt,
1048 .free_ldt = xen_free_ldt,
1050 .store_gdt = native_store_gdt,
1051 .store_idt = native_store_idt,
1052 .store_tr = xen_store_tr,
1054 .write_ldt_entry = xen_write_ldt_entry,
1055 .write_gdt_entry = xen_write_gdt_entry,
1056 .write_idt_entry = xen_write_idt_entry,
1057 .load_sp0 = xen_load_sp0,
1059 .set_iopl_mask = xen_set_iopl_mask,
1060 .io_delay = xen_io_delay,
1062 /* Xen takes care of %gs when switching to usermode for us */
1063 .swapgs = paravirt_nop,
1065 .start_context_switch = paravirt_start_context_switch,
1066 .end_context_switch = xen_end_context_switch,
1069 static const struct pv_apic_ops xen_apic_ops __initconst = {
1070 #ifdef CONFIG_X86_LOCAL_APIC
1071 .startup_ipi_hook = paravirt_nop,
1075 static void xen_reboot(int reason)
1077 struct sched_shutdown r = { .reason = reason };
1079 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1083 static void xen_restart(char *msg)
1085 xen_reboot(SHUTDOWN_reboot);
1088 static void xen_emergency_restart(void)
1090 xen_reboot(SHUTDOWN_reboot);
1093 static void xen_machine_halt(void)
1095 xen_reboot(SHUTDOWN_poweroff);
1098 static void xen_machine_power_off(void)
1102 xen_reboot(SHUTDOWN_poweroff);
1105 static void xen_crash_shutdown(struct pt_regs *regs)
1107 xen_reboot(SHUTDOWN_crash);
1111 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1113 xen_reboot(SHUTDOWN_crash);
1117 static struct notifier_block xen_panic_block = {
1118 .notifier_call= xen_panic_event,
1121 int xen_panic_handler_init(void)
1123 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1127 static const struct machine_ops xen_machine_ops __initconst = {
1128 .restart = xen_restart,
1129 .halt = xen_machine_halt,
1130 .power_off = xen_machine_power_off,
1131 .shutdown = xen_machine_halt,
1132 .crash_shutdown = xen_crash_shutdown,
1133 .emergency_restart = xen_emergency_restart,
1137 * Set up the GDT and segment registers for -fstack-protector. Until
1138 * we do this, we have to be careful not to call any stack-protected
1139 * function, which is most of the kernel.
1141 static void __init xen_setup_stackprotector(void)
1143 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1144 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1146 setup_stack_canary_segment(0);
1147 switch_to_new_gdt(0);
1149 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1150 pv_cpu_ops.load_gdt = xen_load_gdt;
1153 /* First C function to be called on Xen boot */
1154 asmlinkage void __init xen_start_kernel(void)
1156 struct physdev_set_iopl set_iopl;
1160 if (!xen_start_info)
1163 xen_domain_type = XEN_PV_DOMAIN;
1165 xen_setup_machphys_mapping();
1167 /* Install Xen paravirt ops */
1169 pv_init_ops = xen_init_ops;
1170 pv_cpu_ops = xen_cpu_ops;
1171 pv_apic_ops = xen_apic_ops;
1173 x86_init.resources.memory_setup = xen_memory_setup;
1174 x86_init.oem.arch_setup = xen_arch_setup;
1175 x86_init.oem.banner = xen_banner;
1177 xen_init_time_ops();
1180 * Set up some pagetable state before starting to set any ptes.
1185 /* Prevent unwanted bits from being set in PTEs. */
1186 __supported_pte_mask &= ~_PAGE_GLOBAL;
1187 if (!xen_initial_domain())
1188 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1190 __supported_pte_mask |= _PAGE_IOMAP;
1193 * Prevent page tables from being allocated in highmem, even
1194 * if CONFIG_HIGHPTE is enabled.
1196 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1198 /* Work out if we support NX */
1201 xen_setup_features();
1204 if (!xen_feature(XENFEAT_auto_translated_physmap))
1205 xen_build_dynamic_phys_to_machine();
1208 * Set up kernel GDT and segment registers, mainly so that
1209 * -fstack-protector code can be executed.
1211 xen_setup_stackprotector();
1214 xen_init_cpuid_mask();
1216 #ifdef CONFIG_X86_LOCAL_APIC
1218 * set up the basic apic ops.
1220 set_xen_basic_apic_ops();
1223 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1224 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1225 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1228 machine_ops = xen_machine_ops;
1231 * The only reliable way to retain the initial address of the
1232 * percpu gdt_page is to remember it here, so we can go and
1233 * mark it RW later, when the initial percpu area is freed.
1235 xen_initial_gdt = &per_cpu(gdt_page, 0);
1239 #ifdef CONFIG_ACPI_NUMA
1241 * The pages we from Xen are not related to machine pages, so
1242 * any NUMA information the kernel tries to get from ACPI will
1243 * be meaningless. Prevent it from trying.
1248 pgd = (pgd_t *)xen_start_info->pt_base;
1250 if (!xen_initial_domain())
1251 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1253 __supported_pte_mask |= _PAGE_IOMAP;
1254 /* Don't do the full vcpu_info placement stuff until we have a
1255 possible map and a non-dummy shared_info. */
1256 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1258 local_irq_disable();
1259 early_boot_irqs_disabled = true;
1263 xen_raw_console_write("mapping kernel into physical memory\n");
1264 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1265 xen_ident_map_ISA();
1267 /* Allocate and initialize top and mid mfn levels for p2m structure */
1268 xen_build_mfn_list_list();
1270 /* keep using Xen gdt for now; no urgent need to change it */
1272 #ifdef CONFIG_X86_32
1273 pv_info.kernel_rpl = 1;
1274 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1275 pv_info.kernel_rpl = 0;
1277 pv_info.kernel_rpl = 0;
1279 /* set the limit of our address space */
1282 /* We used to do this in xen_arch_setup, but that is too late on AMD
1283 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1284 * which pokes 0xcf8 port.
1287 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1289 xen_raw_printk("physdev_op failed %d\n", rc);
1291 #ifdef CONFIG_X86_32
1292 /* set up basic CPUID stuff */
1293 cpu_detect(&new_cpu_data);
1294 new_cpu_data.hard_math = 1;
1295 new_cpu_data.wp_works_ok = 1;
1296 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1299 /* Poke various useful things into boot_params */
1300 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1301 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1302 ? __pa(xen_start_info->mod_start) : 0;
1303 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1304 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1306 if (!xen_initial_domain()) {
1307 add_preferred_console("xenboot", 0, NULL);
1308 add_preferred_console("tty", 0, NULL);
1309 add_preferred_console("hvc", 0, NULL);
1311 x86_init.pci.arch_init = pci_xen_init;
1313 const struct dom0_vga_console_info *info =
1314 (void *)((char *)xen_start_info +
1315 xen_start_info->console.dom0.info_off);
1317 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1318 xen_start_info->console.domU.mfn = 0;
1319 xen_start_info->console.domU.evtchn = 0;
1321 /* Make sure ACS will be enabled */
1324 /* Avoid searching for BIOS MP tables */
1325 x86_init.mpparse.find_smp_config = x86_init_noop;
1326 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1329 /* PCI BIOS service won't work from a PV guest. */
1330 pci_probe &= ~PCI_PROBE_BIOS;
1332 xen_raw_console_write("about to get started...\n");
1334 xen_setup_runstate_info(0);
1336 /* Start the world */
1337 #ifdef CONFIG_X86_32
1338 i386_start_kernel();
1340 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1344 static int init_hvm_pv_info(int *major, int *minor)
1346 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1349 base = xen_cpuid_base();
1350 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1353 *minor = eax & 0xffff;
1354 printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1356 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1358 pfn = __pa(hypercall_page);
1359 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1361 xen_setup_features();
1363 pv_info.name = "Xen HVM";
1365 xen_domain_type = XEN_HVM_DOMAIN;
1370 void __ref xen_hvm_init_shared_info(void)
1373 struct xen_add_to_physmap xatp;
1374 static struct shared_info *shared_info_page = 0;
1376 if (!shared_info_page)
1377 shared_info_page = (struct shared_info *)
1378 extend_brk(PAGE_SIZE, PAGE_SIZE);
1379 xatp.domid = DOMID_SELF;
1381 xatp.space = XENMAPSPACE_shared_info;
1382 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1383 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1386 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1388 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1389 * page, we use it in the event channel upcall and in some pvclock
1390 * related functions. We don't need the vcpu_info placement
1391 * optimizations because we don't use any pv_mmu or pv_irq op on
1393 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1394 * online but xen_hvm_init_shared_info is run at resume time too and
1395 * in that case multiple vcpus might be online. */
1396 for_each_online_cpu(cpu) {
1397 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1401 #ifdef CONFIG_XEN_PVHVM
1402 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1403 unsigned long action, void *hcpu)
1405 int cpu = (long)hcpu;
1407 case CPU_UP_PREPARE:
1408 xen_vcpu_setup(cpu);
1409 if (xen_have_vector_callback) {
1410 xen_init_lock_cpu(cpu);
1411 if (xen_feature(XENFEAT_hvm_safe_pvclock))
1412 xen_setup_timer(cpu);
1421 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1422 .notifier_call = xen_hvm_cpu_notify,
1425 static void __init xen_hvm_guest_init(void)
1430 r = init_hvm_pv_info(&major, &minor);
1434 xen_hvm_init_shared_info();
1436 if (xen_feature(XENFEAT_hvm_callback_vector))
1437 xen_have_vector_callback = 1;
1439 register_cpu_notifier(&xen_hvm_cpu_notifier);
1440 xen_unplug_emulated_devices();
1441 x86_init.irqs.intr_init = xen_init_IRQ;
1442 xen_hvm_init_time_ops();
1443 xen_hvm_init_mmu_ops();
1446 static bool __init xen_hvm_platform(void)
1448 if (xen_pv_domain())
1451 if (!xen_cpuid_base())
1457 bool xen_hvm_need_lapic(void)
1459 if (xen_pv_domain())
1461 if (!xen_hvm_domain())
1463 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1467 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1469 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1471 .detect = xen_hvm_platform,
1472 .init_platform = xen_hvm_guest_init,
1474 EXPORT_SYMBOL(x86_hyper_xen_hvm);