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>
34 #include <linux/edd.h>
36 #ifdef CONFIG_KEXEC_CORE
37 #include <linux/kexec.h>
41 #include <xen/events.h>
42 #include <xen/interface/xen.h>
43 #include <xen/interface/version.h>
44 #include <xen/interface/physdev.h>
45 #include <xen/interface/vcpu.h>
46 #include <xen/interface/memory.h>
47 #include <xen/interface/nmi.h>
48 #include <xen/interface/xen-mca.h>
49 #include <xen/features.h>
52 #include <xen/hvc-console.h>
55 #include <asm/paravirt.h>
58 #include <asm/xen/pci.h>
59 #include <asm/xen/hypercall.h>
60 #include <asm/xen/hypervisor.h>
61 #include <asm/fixmap.h>
62 #include <asm/processor.h>
63 #include <asm/proto.h>
64 #include <asm/msr-index.h>
65 #include <asm/traps.h>
66 #include <asm/setup.h>
68 #include <asm/pgalloc.h>
69 #include <asm/pgtable.h>
70 #include <asm/tlbflush.h>
71 #include <asm/reboot.h>
72 #include <asm/stackprotector.h>
73 #include <asm/hypervisor.h>
74 #include <asm/mach_traps.h>
75 #include <asm/mwait.h>
76 #include <asm/pci_x86.h>
80 #include <linux/acpi.h>
82 #include <acpi/pdc_intel.h>
83 #include <acpi/processor.h>
84 #include <xen/interface/platform.h>
90 #include "multicalls.h"
93 EXPORT_SYMBOL_GPL(hypercall_page);
96 * Pointer to the xen_vcpu_info structure or
97 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
98 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
99 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
100 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
101 * acknowledge pending events.
102 * Also more subtly it is used by the patched version of irq enable/disable
103 * e.g. xen_irq_enable_direct and xen_iret in PV mode.
105 * The desire to be able to do those mask/unmask operations as a single
106 * instruction by using the per-cpu offset held in %gs is the real reason
107 * vcpu info is in a per-cpu pointer and the original reason for this
111 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
114 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
115 * hypercall. This can be used both in PV and PVHVM mode. The structure
116 * overrides the default per_cpu(xen_vcpu, cpu) value.
118 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
120 enum xen_domain_type xen_domain_type = XEN_NATIVE;
121 EXPORT_SYMBOL_GPL(xen_domain_type);
123 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
124 EXPORT_SYMBOL(machine_to_phys_mapping);
125 unsigned long machine_to_phys_nr;
126 EXPORT_SYMBOL(machine_to_phys_nr);
128 struct start_info *xen_start_info;
129 EXPORT_SYMBOL_GPL(xen_start_info);
131 struct shared_info xen_dummy_shared_info;
133 void *xen_initial_gdt;
135 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
136 __read_mostly int xen_have_vector_callback;
137 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
140 * Point at some empty memory to start with. We map the real shared_info
141 * page as soon as fixmap is up and running.
143 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
146 * Flag to determine whether vcpu info placement is available on all
147 * VCPUs. We assume it is to start with, and then set it to zero on
148 * the first failure. This is because it can succeed on some VCPUs
149 * and not others, since it can involve hypervisor memory allocation,
150 * or because the guest failed to guarantee all the appropriate
151 * constraints on all VCPUs (ie buffer can't cross a page boundary).
153 * Note that any particular CPU may be using a placed vcpu structure,
154 * but we can only optimise if the all are.
156 * 0: not available, 1: available
158 static int have_vcpu_info_placement = 1;
161 struct desc_struct desc[3];
165 * Updating the 3 TLS descriptors in the GDT on every task switch is
166 * surprisingly expensive so we avoid updating them if they haven't
167 * changed. Since Xen writes different descriptors than the one
168 * passed in the update_descriptor hypercall we keep shadow copies to
171 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
173 static void clamp_max_cpus(void)
176 if (setup_max_cpus > MAX_VIRT_CPUS)
177 setup_max_cpus = MAX_VIRT_CPUS;
181 static void xen_vcpu_setup(int cpu)
183 struct vcpu_register_vcpu_info info;
185 struct vcpu_info *vcpup;
187 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
190 * This path is called twice on PVHVM - first during bootup via
191 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
192 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
193 * As we can only do the VCPUOP_register_vcpu_info once lets
194 * not over-write its result.
196 * For PV it is called during restore (xen_vcpu_restore) and bootup
197 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
200 if (xen_hvm_domain()) {
201 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
204 if (cpu < MAX_VIRT_CPUS)
205 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
207 if (!have_vcpu_info_placement) {
208 if (cpu >= MAX_VIRT_CPUS)
213 vcpup = &per_cpu(xen_vcpu_info, cpu);
214 info.mfn = arbitrary_virt_to_mfn(vcpup);
215 info.offset = offset_in_page(vcpup);
217 /* Check to see if the hypervisor will put the vcpu_info
218 structure where we want it, which allows direct access via
220 N.B. This hypercall can _only_ be called once per CPU. Subsequent
221 calls will error out with -EINVAL. This is due to the fact that
222 hypervisor has no unregister variant and this hypercall does not
223 allow to over-write info.mfn and info.offset.
225 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
228 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
229 have_vcpu_info_placement = 0;
232 /* This cpu is using the registered vcpu info, even if
233 later ones fail to. */
234 per_cpu(xen_vcpu, cpu) = vcpup;
239 * On restore, set the vcpu placement up again.
240 * If it fails, then we're in a bad state, since
241 * we can't back out from using it...
243 void xen_vcpu_restore(void)
247 for_each_possible_cpu(cpu) {
248 bool other_cpu = (cpu != smp_processor_id());
249 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL);
251 if (other_cpu && is_up &&
252 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
255 xen_setup_runstate_info(cpu);
257 if (have_vcpu_info_placement)
260 if (other_cpu && is_up &&
261 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
266 static void __init xen_banner(void)
268 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
269 struct xen_extraversion extra;
270 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
272 pr_info("Booting paravirtualized kernel %son %s\n",
273 xen_feature(XENFEAT_auto_translated_physmap) ?
274 "with PVH extensions " : "", pv_info.name);
275 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
276 version >> 16, version & 0xffff, extra.extraversion,
277 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
279 /* Check if running on Xen version (major, minor) or later */
281 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
283 unsigned int version;
288 version = HYPERVISOR_xen_version(XENVER_version, NULL);
289 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
290 ((version >> 16) > major))
295 #define CPUID_THERM_POWER_LEAF 6
296 #define APERFMPERF_PRESENT 0
298 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
299 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
301 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
302 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
303 static __read_mostly unsigned int cpuid_leaf5_edx_val;
305 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
306 unsigned int *cx, unsigned int *dx)
308 unsigned maskebx = ~0;
309 unsigned maskecx = ~0;
310 unsigned maskedx = ~0;
313 * Mask out inconvenient features, to try and disable as many
314 * unsupported kernel subsystems as possible.
318 maskecx = cpuid_leaf1_ecx_mask;
319 setecx = cpuid_leaf1_ecx_set_mask;
320 maskedx = cpuid_leaf1_edx_mask;
323 case CPUID_MWAIT_LEAF:
324 /* Synthesize the values.. */
327 *cx = cpuid_leaf5_ecx_val;
328 *dx = cpuid_leaf5_edx_val;
331 case CPUID_THERM_POWER_LEAF:
332 /* Disabling APERFMPERF for kernel usage */
333 maskecx = ~(1 << APERFMPERF_PRESENT);
337 /* Suppress extended topology stuff */
342 asm(XEN_EMULATE_PREFIX "cpuid"
347 : "0" (*ax), "2" (*cx));
356 static bool __init xen_check_mwait(void)
359 struct xen_platform_op op = {
360 .cmd = XENPF_set_processor_pminfo,
361 .u.set_pminfo.id = -1,
362 .u.set_pminfo.type = XEN_PM_PDC,
365 unsigned int ax, bx, cx, dx;
366 unsigned int mwait_mask;
368 /* We need to determine whether it is OK to expose the MWAIT
369 * capability to the kernel to harvest deeper than C3 states from ACPI
370 * _CST using the processor_harvest_xen.c module. For this to work, we
371 * need to gather the MWAIT_LEAF values (which the cstate.c code
372 * checks against). The hypervisor won't expose the MWAIT flag because
373 * it would break backwards compatibility; so we will find out directly
374 * from the hardware and hypercall.
376 if (!xen_initial_domain())
380 * When running under platform earlier than Xen4.2, do not expose
381 * mwait, to avoid the risk of loading native acpi pad driver
383 if (!xen_running_on_version_or_later(4, 2))
389 native_cpuid(&ax, &bx, &cx, &dx);
391 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
392 (1 << (X86_FEATURE_MWAIT % 32));
394 if ((cx & mwait_mask) != mwait_mask)
397 /* We need to emulate the MWAIT_LEAF and for that we need both
398 * ecx and edx. The hypercall provides only partial information.
401 ax = CPUID_MWAIT_LEAF;
406 native_cpuid(&ax, &bx, &cx, &dx);
408 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
409 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
411 buf[0] = ACPI_PDC_REVISION_ID;
413 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
415 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
417 if ((HYPERVISOR_dom0_op(&op) == 0) &&
418 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
419 cpuid_leaf5_ecx_val = cx;
420 cpuid_leaf5_edx_val = dx;
427 static void __init xen_init_cpuid_mask(void)
429 unsigned int ax, bx, cx, dx;
430 unsigned int xsave_mask;
432 cpuid_leaf1_edx_mask =
433 ~((1 << X86_FEATURE_MTRR) | /* disable MTRR */
434 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
437 * Xen PV would need some work to support PCID: CR3 handling as well
438 * as xen_flush_tlb_others() would need updating.
440 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_PCID % 32)); /* disable PCID */
442 if (!xen_initial_domain())
443 cpuid_leaf1_edx_mask &=
444 ~((1 << X86_FEATURE_ACPI)); /* disable ACPI */
446 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32));
450 cpuid(1, &ax, &bx, &cx, &dx);
453 (1 << (X86_FEATURE_XSAVE % 32)) |
454 (1 << (X86_FEATURE_OSXSAVE % 32));
456 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
457 if ((cx & xsave_mask) != xsave_mask)
458 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
459 if (xen_check_mwait())
460 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
463 static void __init xen_init_capabilities(void)
466 setup_force_cpu_cap(X86_FEATURE_XENPV);
469 static void xen_set_debugreg(int reg, unsigned long val)
471 HYPERVISOR_set_debugreg(reg, val);
474 static unsigned long xen_get_debugreg(int reg)
476 return HYPERVISOR_get_debugreg(reg);
479 static void xen_end_context_switch(struct task_struct *next)
482 paravirt_end_context_switch(next);
485 static unsigned long xen_store_tr(void)
491 * Set the page permissions for a particular virtual address. If the
492 * address is a vmalloc mapping (or other non-linear mapping), then
493 * find the linear mapping of the page and also set its protections to
496 static void set_aliased_prot(void *v, pgprot_t prot)
505 ptep = lookup_address((unsigned long)v, &level);
506 BUG_ON(ptep == NULL);
508 pfn = pte_pfn(*ptep);
509 page = pfn_to_page(pfn);
511 pte = pfn_pte(pfn, prot);
514 * Careful: update_va_mapping() will fail if the virtual address
515 * we're poking isn't populated in the page tables. We don't
516 * need to worry about the direct map (that's always in the page
517 * tables), but we need to be careful about vmap space. In
518 * particular, the top level page table can lazily propagate
519 * entries between processes, so if we've switched mms since we
520 * vmapped the target in the first place, we might not have the
521 * top-level page table entry populated.
523 * We disable preemption because we want the same mm active when
524 * we probe the target and when we issue the hypercall. We'll
525 * have the same nominal mm, but if we're a kernel thread, lazy
526 * mm dropping could change our pgd.
528 * Out of an abundance of caution, this uses __get_user() to fault
529 * in the target address just in case there's some obscure case
530 * in which the target address isn't readable.
535 pagefault_disable(); /* Avoid warnings due to being atomic. */
536 __get_user(dummy, (unsigned char __user __force *)v);
539 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
542 if (!PageHighMem(page)) {
543 void *av = __va(PFN_PHYS(pfn));
546 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
554 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
556 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
560 * We need to mark the all aliases of the LDT pages RO. We
561 * don't need to call vm_flush_aliases(), though, since that's
562 * only responsible for flushing aliases out the TLBs, not the
563 * page tables, and Xen will flush the TLB for us if needed.
565 * To avoid confusing future readers: none of this is necessary
566 * to load the LDT. The hypervisor only checks this when the
567 * LDT is faulted in due to subsequent descriptor access.
570 for(i = 0; i < entries; i += entries_per_page)
571 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
574 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
576 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
579 for(i = 0; i < entries; i += entries_per_page)
580 set_aliased_prot(ldt + i, PAGE_KERNEL);
583 static void xen_set_ldt(const void *addr, unsigned entries)
585 struct mmuext_op *op;
586 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
588 trace_xen_cpu_set_ldt(addr, entries);
591 op->cmd = MMUEXT_SET_LDT;
592 op->arg1.linear_addr = (unsigned long)addr;
593 op->arg2.nr_ents = entries;
595 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
597 xen_mc_issue(PARAVIRT_LAZY_CPU);
600 static void xen_load_gdt(const struct desc_ptr *dtr)
602 unsigned long va = dtr->address;
603 unsigned int size = dtr->size + 1;
604 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
605 unsigned long frames[pages];
609 * A GDT can be up to 64k in size, which corresponds to 8192
610 * 8-byte entries, or 16 4k pages..
613 BUG_ON(size > 65536);
614 BUG_ON(va & ~PAGE_MASK);
616 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
619 unsigned long pfn, mfn;
623 * The GDT is per-cpu and is in the percpu data area.
624 * That can be virtually mapped, so we need to do a
625 * page-walk to get the underlying MFN for the
626 * hypercall. The page can also be in the kernel's
627 * linear range, so we need to RO that mapping too.
629 ptep = lookup_address(va, &level);
630 BUG_ON(ptep == NULL);
632 pfn = pte_pfn(*ptep);
633 mfn = pfn_to_mfn(pfn);
634 virt = __va(PFN_PHYS(pfn));
638 make_lowmem_page_readonly((void *)va);
639 make_lowmem_page_readonly(virt);
642 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
647 * load_gdt for early boot, when the gdt is only mapped once
649 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
651 unsigned long va = dtr->address;
652 unsigned int size = dtr->size + 1;
653 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
654 unsigned long frames[pages];
658 * A GDT can be up to 64k in size, which corresponds to 8192
659 * 8-byte entries, or 16 4k pages..
662 BUG_ON(size > 65536);
663 BUG_ON(va & ~PAGE_MASK);
665 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
667 unsigned long pfn, mfn;
669 pfn = virt_to_pfn(va);
670 mfn = pfn_to_mfn(pfn);
672 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
674 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
680 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
684 static inline bool desc_equal(const struct desc_struct *d1,
685 const struct desc_struct *d2)
687 return d1->a == d2->a && d1->b == d2->b;
690 static void load_TLS_descriptor(struct thread_struct *t,
691 unsigned int cpu, unsigned int i)
693 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
694 struct desc_struct *gdt;
696 struct multicall_space mc;
698 if (desc_equal(shadow, &t->tls_array[i]))
701 *shadow = t->tls_array[i];
703 gdt = get_cpu_gdt_table(cpu);
704 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
705 mc = __xen_mc_entry(0);
707 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
710 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
713 * XXX sleazy hack: If we're being called in a lazy-cpu zone
714 * and lazy gs handling is enabled, it means we're in a
715 * context switch, and %gs has just been saved. This means we
716 * can zero it out to prevent faults on exit from the
717 * hypervisor if the next process has no %gs. Either way, it
718 * has been saved, and the new value will get loaded properly.
719 * This will go away as soon as Xen has been modified to not
720 * save/restore %gs for normal hypercalls.
722 * On x86_64, this hack is not used for %gs, because gs points
723 * to KERNEL_GS_BASE (and uses it for PDA references), so we
724 * must not zero %gs on x86_64
726 * For x86_64, we need to zero %fs, otherwise we may get an
727 * exception between the new %fs descriptor being loaded and
728 * %fs being effectively cleared at __switch_to().
730 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
740 load_TLS_descriptor(t, cpu, 0);
741 load_TLS_descriptor(t, cpu, 1);
742 load_TLS_descriptor(t, cpu, 2);
744 xen_mc_issue(PARAVIRT_LAZY_CPU);
748 static void xen_load_gs_index(unsigned int idx)
750 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
755 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
758 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
759 u64 entry = *(u64 *)ptr;
761 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
766 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
772 static int cvt_gate_to_trap(int vector, const gate_desc *val,
773 struct trap_info *info)
777 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
780 info->vector = vector;
782 addr = gate_offset(*val);
785 * Look for known traps using IST, and substitute them
786 * appropriately. The debugger ones are the only ones we care
787 * about. Xen will handle faults like double_fault,
788 * so we should never see them. Warn if
789 * there's an unexpected IST-using fault handler.
791 if (addr == (unsigned long)debug)
792 addr = (unsigned long)xen_debug;
793 else if (addr == (unsigned long)int3)
794 addr = (unsigned long)xen_int3;
795 else if (addr == (unsigned long)stack_segment)
796 addr = (unsigned long)xen_stack_segment;
797 else if (addr == (unsigned long)double_fault) {
798 /* Don't need to handle these */
800 #ifdef CONFIG_X86_MCE
801 } else if (addr == (unsigned long)machine_check) {
803 * when xen hypervisor inject vMCE to guest,
804 * use native mce handler to handle it
808 } else if (addr == (unsigned long)nmi)
810 * Use the native version as well.
814 /* Some other trap using IST? */
815 if (WARN_ON(val->ist != 0))
818 #endif /* CONFIG_X86_64 */
819 info->address = addr;
821 info->cs = gate_segment(*val);
822 info->flags = val->dpl;
823 /* interrupt gates clear IF */
824 if (val->type == GATE_INTERRUPT)
825 info->flags |= 1 << 2;
830 /* Locations of each CPU's IDT */
831 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
833 /* Set an IDT entry. If the entry is part of the current IDT, then
835 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
837 unsigned long p = (unsigned long)&dt[entrynum];
838 unsigned long start, end;
840 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
844 start = __this_cpu_read(idt_desc.address);
845 end = start + __this_cpu_read(idt_desc.size) + 1;
849 native_write_idt_entry(dt, entrynum, g);
851 if (p >= start && (p + 8) <= end) {
852 struct trap_info info[2];
856 if (cvt_gate_to_trap(entrynum, g, &info[0]))
857 if (HYPERVISOR_set_trap_table(info))
864 static void xen_convert_trap_info(const struct desc_ptr *desc,
865 struct trap_info *traps)
867 unsigned in, out, count;
869 count = (desc->size+1) / sizeof(gate_desc);
872 for (in = out = 0; in < count; in++) {
873 gate_desc *entry = (gate_desc*)(desc->address) + in;
875 if (cvt_gate_to_trap(in, entry, &traps[out]))
878 traps[out].address = 0;
881 void xen_copy_trap_info(struct trap_info *traps)
883 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
885 xen_convert_trap_info(desc, traps);
888 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
889 hold a spinlock to protect the static traps[] array (static because
890 it avoids allocation, and saves stack space). */
891 static void xen_load_idt(const struct desc_ptr *desc)
893 static DEFINE_SPINLOCK(lock);
894 static struct trap_info traps[257];
896 trace_xen_cpu_load_idt(desc);
900 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
902 xen_convert_trap_info(desc, traps);
905 if (HYPERVISOR_set_trap_table(traps))
911 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
912 they're handled differently. */
913 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
914 const void *desc, int type)
916 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
927 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
930 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
940 * Version of write_gdt_entry for use at early boot-time needed to
941 * update an entry as simply as possible.
943 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
944 const void *desc, int type)
946 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
955 xmaddr_t maddr = virt_to_machine(&dt[entry]);
957 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
958 dt[entry] = *(struct desc_struct *)desc;
964 static void xen_load_sp0(struct tss_struct *tss,
965 struct thread_struct *thread)
967 struct multicall_space mcs;
969 mcs = xen_mc_entry(0);
970 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
971 xen_mc_issue(PARAVIRT_LAZY_CPU);
972 tss->x86_tss.sp0 = thread->sp0;
975 void xen_set_iopl_mask(unsigned mask)
977 struct physdev_set_iopl set_iopl;
979 /* Force the change at ring 0. */
980 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
981 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
984 static void xen_io_delay(void)
988 static void xen_clts(void)
990 struct multicall_space mcs;
992 mcs = xen_mc_entry(0);
994 MULTI_fpu_taskswitch(mcs.mc, 0);
996 xen_mc_issue(PARAVIRT_LAZY_CPU);
999 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
1001 static unsigned long xen_read_cr0(void)
1003 unsigned long cr0 = this_cpu_read(xen_cr0_value);
1005 if (unlikely(cr0 == 0)) {
1006 cr0 = native_read_cr0();
1007 this_cpu_write(xen_cr0_value, cr0);
1013 static void xen_write_cr0(unsigned long cr0)
1015 struct multicall_space mcs;
1017 this_cpu_write(xen_cr0_value, cr0);
1019 /* Only pay attention to cr0.TS; everything else is
1021 mcs = xen_mc_entry(0);
1023 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1025 xen_mc_issue(PARAVIRT_LAZY_CPU);
1028 static void xen_write_cr4(unsigned long cr4)
1030 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1032 native_write_cr4(cr4);
1034 #ifdef CONFIG_X86_64
1035 static inline unsigned long xen_read_cr8(void)
1039 static inline void xen_write_cr8(unsigned long val)
1045 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1049 if (pmu_msr_read(msr, &val, err))
1052 val = native_read_msr_safe(msr, err);
1054 case MSR_IA32_APICBASE:
1055 #ifdef CONFIG_X86_X2APIC
1056 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
1058 val &= ~X2APIC_ENABLE;
1064 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1071 #ifdef CONFIG_X86_64
1075 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
1076 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
1077 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
1080 base = ((u64)high << 32) | low;
1081 if (HYPERVISOR_set_segment_base(which, base) != 0)
1089 case MSR_SYSCALL_MASK:
1090 case MSR_IA32_SYSENTER_CS:
1091 case MSR_IA32_SYSENTER_ESP:
1092 case MSR_IA32_SYSENTER_EIP:
1093 /* Fast syscall setup is all done in hypercalls, so
1094 these are all ignored. Stub them out here to stop
1095 Xen console noise. */
1099 if (!pmu_msr_write(msr, low, high, &ret))
1100 ret = native_write_msr_safe(msr, low, high);
1106 void xen_setup_shared_info(void)
1108 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1109 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1110 xen_start_info->shared_info);
1112 HYPERVISOR_shared_info =
1113 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1115 HYPERVISOR_shared_info =
1116 (struct shared_info *)__va(xen_start_info->shared_info);
1119 /* In UP this is as good a place as any to set up shared info */
1120 xen_setup_vcpu_info_placement();
1123 xen_setup_mfn_list_list();
1126 /* This is called once we have the cpu_possible_mask */
1127 void xen_setup_vcpu_info_placement(void)
1131 for_each_possible_cpu(cpu)
1132 xen_vcpu_setup(cpu);
1134 /* xen_vcpu_setup managed to place the vcpu_info within the
1135 * percpu area for all cpus, so make use of it. Note that for
1136 * PVH we want to use native IRQ mechanism. */
1137 if (have_vcpu_info_placement && !xen_pvh_domain()) {
1138 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1139 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1140 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1141 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1142 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1146 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1147 unsigned long addr, unsigned len)
1149 char *start, *end, *reloc;
1152 start = end = reloc = NULL;
1154 #define SITE(op, x) \
1155 case PARAVIRT_PATCH(op.x): \
1156 if (have_vcpu_info_placement) { \
1157 start = (char *)xen_##x##_direct; \
1158 end = xen_##x##_direct_end; \
1159 reloc = xen_##x##_direct_reloc; \
1164 SITE(pv_irq_ops, irq_enable);
1165 SITE(pv_irq_ops, irq_disable);
1166 SITE(pv_irq_ops, save_fl);
1167 SITE(pv_irq_ops, restore_fl);
1171 if (start == NULL || (end-start) > len)
1174 ret = paravirt_patch_insns(insnbuf, len, start, end);
1176 /* Note: because reloc is assigned from something that
1177 appears to be an array, gcc assumes it's non-null,
1178 but doesn't know its relationship with start and
1180 if (reloc > start && reloc < end) {
1181 int reloc_off = reloc - start;
1182 long *relocp = (long *)(insnbuf + reloc_off);
1183 long delta = start - (char *)addr;
1191 ret = paravirt_patch_default(type, clobbers, insnbuf,
1199 static const struct pv_info xen_info __initconst = {
1200 .paravirt_enabled = 1,
1201 .shared_kernel_pmd = 0,
1203 #ifdef CONFIG_X86_64
1204 .extra_user_64bit_cs = FLAT_USER_CS64,
1210 static const struct pv_init_ops xen_init_ops __initconst = {
1214 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1217 .set_debugreg = xen_set_debugreg,
1218 .get_debugreg = xen_get_debugreg,
1222 .read_cr0 = xen_read_cr0,
1223 .write_cr0 = xen_write_cr0,
1225 .read_cr4 = native_read_cr4,
1226 .read_cr4_safe = native_read_cr4_safe,
1227 .write_cr4 = xen_write_cr4,
1229 #ifdef CONFIG_X86_64
1230 .read_cr8 = xen_read_cr8,
1231 .write_cr8 = xen_write_cr8,
1234 .wbinvd = native_wbinvd,
1236 .read_msr = xen_read_msr_safe,
1237 .write_msr = xen_write_msr_safe,
1239 .read_pmc = xen_read_pmc,
1242 #ifdef CONFIG_X86_64
1243 .usergs_sysret64 = xen_sysret64,
1246 .load_tr_desc = paravirt_nop,
1247 .set_ldt = xen_set_ldt,
1248 .load_gdt = xen_load_gdt,
1249 .load_idt = xen_load_idt,
1250 .load_tls = xen_load_tls,
1251 #ifdef CONFIG_X86_64
1252 .load_gs_index = xen_load_gs_index,
1255 .alloc_ldt = xen_alloc_ldt,
1256 .free_ldt = xen_free_ldt,
1258 .store_idt = native_store_idt,
1259 .store_tr = xen_store_tr,
1261 .write_ldt_entry = xen_write_ldt_entry,
1262 .write_gdt_entry = xen_write_gdt_entry,
1263 .write_idt_entry = xen_write_idt_entry,
1264 .load_sp0 = xen_load_sp0,
1266 .set_iopl_mask = xen_set_iopl_mask,
1267 .io_delay = xen_io_delay,
1269 /* Xen takes care of %gs when switching to usermode for us */
1270 .swapgs = paravirt_nop,
1272 .start_context_switch = paravirt_start_context_switch,
1273 .end_context_switch = xen_end_context_switch,
1276 static const struct pv_apic_ops xen_apic_ops __initconst = {
1277 #ifdef CONFIG_X86_LOCAL_APIC
1278 .startup_ipi_hook = paravirt_nop,
1282 static void xen_reboot(int reason)
1284 struct sched_shutdown r = { .reason = reason };
1287 for_each_online_cpu(cpu)
1288 xen_pmu_finish(cpu);
1290 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1294 static void xen_restart(char *msg)
1296 xen_reboot(SHUTDOWN_reboot);
1299 static void xen_emergency_restart(void)
1301 xen_reboot(SHUTDOWN_reboot);
1304 static void xen_machine_halt(void)
1306 xen_reboot(SHUTDOWN_poweroff);
1309 static void xen_machine_power_off(void)
1313 xen_reboot(SHUTDOWN_poweroff);
1316 static void xen_crash_shutdown(struct pt_regs *regs)
1318 xen_reboot(SHUTDOWN_crash);
1322 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1324 xen_reboot(SHUTDOWN_crash);
1328 static struct notifier_block xen_panic_block = {
1329 .notifier_call= xen_panic_event,
1333 int xen_panic_handler_init(void)
1335 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1339 static const struct machine_ops xen_machine_ops __initconst = {
1340 .restart = xen_restart,
1341 .halt = xen_machine_halt,
1342 .power_off = xen_machine_power_off,
1343 .shutdown = xen_machine_halt,
1344 .crash_shutdown = xen_crash_shutdown,
1345 .emergency_restart = xen_emergency_restart,
1348 static unsigned char xen_get_nmi_reason(void)
1350 unsigned char reason = 0;
1352 /* Construct a value which looks like it came from port 0x61. */
1353 if (test_bit(_XEN_NMIREASON_io_error,
1354 &HYPERVISOR_shared_info->arch.nmi_reason))
1355 reason |= NMI_REASON_IOCHK;
1356 if (test_bit(_XEN_NMIREASON_pci_serr,
1357 &HYPERVISOR_shared_info->arch.nmi_reason))
1358 reason |= NMI_REASON_SERR;
1363 static void __init xen_boot_params_init_edd(void)
1365 #if IS_ENABLED(CONFIG_EDD)
1366 struct xen_platform_op op;
1367 struct edd_info *edd_info;
1372 edd_info = boot_params.eddbuf;
1373 mbr_signature = boot_params.edd_mbr_sig_buffer;
1375 op.cmd = XENPF_firmware_info;
1377 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1378 for (nr = 0; nr < EDDMAXNR; nr++) {
1379 struct edd_info *info = edd_info + nr;
1381 op.u.firmware_info.index = nr;
1382 info->params.length = sizeof(info->params);
1383 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1385 ret = HYPERVISOR_dom0_op(&op);
1389 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1392 C(interface_support);
1393 C(legacy_max_cylinder);
1395 C(legacy_sectors_per_track);
1398 boot_params.eddbuf_entries = nr;
1400 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1401 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1402 op.u.firmware_info.index = nr;
1403 ret = HYPERVISOR_dom0_op(&op);
1406 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1408 boot_params.edd_mbr_sig_buf_entries = nr;
1413 * Set up the GDT and segment registers for -fstack-protector. Until
1414 * we do this, we have to be careful not to call any stack-protected
1415 * function, which is most of the kernel.
1417 * Note, that it is __ref because the only caller of this after init
1418 * is PVH which is not going to use xen_load_gdt_boot or other
1421 static void __ref xen_setup_gdt(int cpu)
1423 if (xen_feature(XENFEAT_auto_translated_physmap)) {
1424 #ifdef CONFIG_X86_64
1425 unsigned long dummy;
1427 load_percpu_segment(cpu); /* We need to access per-cpu area */
1428 switch_to_new_gdt(cpu); /* GDT and GS set */
1430 /* We are switching of the Xen provided GDT to our HVM mode
1431 * GDT. The new GDT has __KERNEL_CS with CS.L = 1
1432 * and we are jumping to reload it.
1434 asm volatile ("pushq %0\n"
1435 "leaq 1f(%%rip),%0\n"
1439 : "=&r" (dummy) : "0" (__KERNEL_CS));
1442 * While not needed, we also set the %es, %ds, and %fs
1443 * to zero. We don't care about %ss as it is NULL.
1444 * Strictly speaking this is not needed as Xen zeros those
1445 * out (and also MSR_FS_BASE, MSR_GS_BASE, MSR_KERNEL_GS_BASE)
1447 * Linux zeros them in cpu_init() and in secondary_startup_64
1454 /* PVH: TODO Implement. */
1457 return; /* PVH does not need any PV GDT ops. */
1459 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1460 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1462 setup_stack_canary_segment(0);
1463 switch_to_new_gdt(0);
1465 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1466 pv_cpu_ops.load_gdt = xen_load_gdt;
1469 #ifdef CONFIG_XEN_PVH
1471 * A PV guest starts with default flags that are not set for PVH, set them
1474 static void xen_pvh_set_cr_flags(int cpu)
1477 /* Some of these are setup in 'secondary_startup_64'. The others:
1478 * X86_CR0_TS, X86_CR0_PE, X86_CR0_ET are set by Xen for HVM guests
1479 * (which PVH shared codepaths), while X86_CR0_PG is for PVH. */
1480 write_cr0(read_cr0() | X86_CR0_MP | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM);
1485 * For BSP, PSE PGE are set in probe_page_size_mask(), for APs
1486 * set them here. For all, OSFXSR OSXMMEXCPT are set in fpu__init_cpu().
1489 cr4_set_bits_and_update_boot(X86_CR4_PSE);
1492 cr4_set_bits_and_update_boot(X86_CR4_PGE);
1496 * Note, that it is ref - because the only caller of this after init
1497 * is PVH which is not going to use xen_load_gdt_boot or other
1500 void __ref xen_pvh_secondary_vcpu_init(int cpu)
1503 xen_pvh_set_cr_flags(cpu);
1506 static void __init xen_pvh_early_guest_init(void)
1508 if (!xen_feature(XENFEAT_auto_translated_physmap))
1511 if (!xen_feature(XENFEAT_hvm_callback_vector))
1514 xen_have_vector_callback = 1;
1516 xen_pvh_early_cpu_init(0, false);
1517 xen_pvh_set_cr_flags(0);
1519 #ifdef CONFIG_X86_32
1520 BUG(); /* PVH: Implement proper support. */
1523 #endif /* CONFIG_XEN_PVH */
1525 /* First C function to be called on Xen boot */
1526 asmlinkage __visible void __init xen_start_kernel(void)
1528 struct physdev_set_iopl set_iopl;
1529 unsigned long initrd_start = 0;
1532 if (!xen_start_info)
1535 xen_domain_type = XEN_PV_DOMAIN;
1537 xen_setup_features();
1538 #ifdef CONFIG_XEN_PVH
1539 xen_pvh_early_guest_init();
1541 xen_setup_machphys_mapping();
1543 /* Install Xen paravirt ops */
1545 if (xen_initial_domain())
1546 pv_info.features |= PV_SUPPORTED_RTC;
1547 pv_init_ops = xen_init_ops;
1548 pv_apic_ops = xen_apic_ops;
1549 if (!xen_pvh_domain()) {
1550 pv_cpu_ops = xen_cpu_ops;
1552 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1555 if (xen_feature(XENFEAT_auto_translated_physmap))
1556 x86_init.resources.memory_setup = xen_auto_xlated_memory_setup;
1558 x86_init.resources.memory_setup = xen_memory_setup;
1559 x86_init.oem.arch_setup = xen_arch_setup;
1560 x86_init.oem.banner = xen_banner;
1562 xen_init_time_ops();
1565 * Set up some pagetable state before starting to set any ptes.
1570 /* Prevent unwanted bits from being set in PTEs. */
1571 __supported_pte_mask &= ~_PAGE_GLOBAL;
1574 * Prevent page tables from being allocated in highmem, even
1575 * if CONFIG_HIGHPTE is enabled.
1577 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1579 /* Work out if we support NX */
1583 xen_build_dynamic_phys_to_machine();
1586 * Set up kernel GDT and segment registers, mainly so that
1587 * -fstack-protector code can be executed.
1592 xen_init_cpuid_mask();
1593 xen_init_capabilities();
1595 #ifdef CONFIG_X86_LOCAL_APIC
1597 * set up the basic apic ops.
1602 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1603 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1604 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1607 machine_ops = xen_machine_ops;
1610 * The only reliable way to retain the initial address of the
1611 * percpu gdt_page is to remember it here, so we can go and
1612 * mark it RW later, when the initial percpu area is freed.
1614 xen_initial_gdt = &per_cpu(gdt_page, 0);
1618 #ifdef CONFIG_ACPI_NUMA
1620 * The pages we from Xen are not related to machine pages, so
1621 * any NUMA information the kernel tries to get from ACPI will
1622 * be meaningless. Prevent it from trying.
1626 /* Don't do the full vcpu_info placement stuff until we have a
1627 possible map and a non-dummy shared_info. */
1628 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1630 local_irq_disable();
1631 early_boot_irqs_disabled = true;
1633 xen_raw_console_write("mapping kernel into physical memory\n");
1634 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1635 xen_start_info->nr_pages);
1636 xen_reserve_special_pages();
1638 /* keep using Xen gdt for now; no urgent need to change it */
1640 #ifdef CONFIG_X86_32
1641 pv_info.kernel_rpl = 1;
1642 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1643 pv_info.kernel_rpl = 0;
1645 pv_info.kernel_rpl = 0;
1647 /* set the limit of our address space */
1650 /* PVH: runs at default kernel iopl of 0 */
1651 if (!xen_pvh_domain()) {
1653 * We used to do this in xen_arch_setup, but that is too late
1654 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1655 * early_amd_init which pokes 0xcf8 port.
1658 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1660 xen_raw_printk("physdev_op failed %d\n", rc);
1663 #ifdef CONFIG_X86_32
1664 /* set up basic CPUID stuff */
1665 cpu_detect(&new_cpu_data);
1666 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1667 new_cpu_data.wp_works_ok = 1;
1668 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1671 if (xen_start_info->mod_start) {
1672 if (xen_start_info->flags & SIF_MOD_START_PFN)
1673 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1675 initrd_start = __pa(xen_start_info->mod_start);
1678 /* Poke various useful things into boot_params */
1679 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1680 boot_params.hdr.ramdisk_image = initrd_start;
1681 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1682 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1684 if (!xen_initial_domain()) {
1685 add_preferred_console("xenboot", 0, NULL);
1686 add_preferred_console("tty", 0, NULL);
1687 add_preferred_console("hvc", 0, NULL);
1689 x86_init.pci.arch_init = pci_xen_init;
1691 const struct dom0_vga_console_info *info =
1692 (void *)((char *)xen_start_info +
1693 xen_start_info->console.dom0.info_off);
1694 struct xen_platform_op op = {
1695 .cmd = XENPF_firmware_info,
1696 .interface_version = XENPF_INTERFACE_VERSION,
1697 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1700 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1701 xen_start_info->console.domU.mfn = 0;
1702 xen_start_info->console.domU.evtchn = 0;
1704 if (HYPERVISOR_dom0_op(&op) == 0)
1705 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1707 /* Make sure ACS will be enabled */
1710 xen_acpi_sleep_register();
1712 /* Avoid searching for BIOS MP tables */
1713 x86_init.mpparse.find_smp_config = x86_init_noop;
1714 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1716 xen_boot_params_init_edd();
1719 /* PCI BIOS service won't work from a PV guest. */
1720 pci_probe &= ~PCI_PROBE_BIOS;
1722 xen_raw_console_write("about to get started...\n");
1724 xen_setup_runstate_info(0);
1728 /* Start the world */
1729 #ifdef CONFIG_X86_32
1730 i386_start_kernel();
1732 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1733 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1737 void __ref xen_hvm_init_shared_info(void)
1740 struct xen_add_to_physmap xatp;
1741 static struct shared_info *shared_info_page = 0;
1743 if (!shared_info_page)
1744 shared_info_page = (struct shared_info *)
1745 extend_brk(PAGE_SIZE, PAGE_SIZE);
1746 xatp.domid = DOMID_SELF;
1748 xatp.space = XENMAPSPACE_shared_info;
1749 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1750 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1753 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1755 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1756 * page, we use it in the event channel upcall and in some pvclock
1757 * related functions. We don't need the vcpu_info placement
1758 * optimizations because we don't use any pv_mmu or pv_irq op on
1760 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1761 * online but xen_hvm_init_shared_info is run at resume time too and
1762 * in that case multiple vcpus might be online. */
1763 for_each_online_cpu(cpu) {
1764 /* Leave it to be NULL. */
1765 if (cpu >= MAX_VIRT_CPUS)
1767 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1771 #ifdef CONFIG_XEN_PVHVM
1772 static void __init init_hvm_pv_info(void)
1775 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1778 base = xen_cpuid_base();
1779 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1782 minor = eax & 0xffff;
1783 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1785 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1787 pfn = __pa(hypercall_page);
1788 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1790 xen_setup_features();
1792 pv_info.name = "Xen HVM";
1794 xen_domain_type = XEN_HVM_DOMAIN;
1797 static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action,
1800 int cpu = (long)hcpu;
1802 case CPU_UP_PREPARE:
1803 xen_vcpu_setup(cpu);
1804 if (xen_have_vector_callback) {
1805 if (xen_feature(XENFEAT_hvm_safe_pvclock))
1806 xen_setup_timer(cpu);
1815 static struct notifier_block xen_hvm_cpu_notifier = {
1816 .notifier_call = xen_hvm_cpu_notify,
1819 #ifdef CONFIG_KEXEC_CORE
1820 static void xen_hvm_shutdown(void)
1822 native_machine_shutdown();
1823 if (kexec_in_progress)
1824 xen_reboot(SHUTDOWN_soft_reset);
1827 static void xen_hvm_crash_shutdown(struct pt_regs *regs)
1829 native_machine_crash_shutdown(regs);
1830 xen_reboot(SHUTDOWN_soft_reset);
1834 static void __init xen_hvm_guest_init(void)
1836 if (xen_pv_domain())
1841 xen_hvm_init_shared_info();
1843 xen_panic_handler_init();
1845 if (xen_feature(XENFEAT_hvm_callback_vector))
1846 xen_have_vector_callback = 1;
1848 register_cpu_notifier(&xen_hvm_cpu_notifier);
1849 xen_unplug_emulated_devices();
1850 x86_init.irqs.intr_init = xen_init_IRQ;
1851 xen_hvm_init_time_ops();
1852 xen_hvm_init_mmu_ops();
1853 #ifdef CONFIG_KEXEC_CORE
1854 machine_ops.shutdown = xen_hvm_shutdown;
1855 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
1860 static bool xen_nopv = false;
1861 static __init int xen_parse_nopv(char *arg)
1866 early_param("xen_nopv", xen_parse_nopv);
1868 static uint32_t __init xen_platform(void)
1873 return xen_cpuid_base();
1876 bool xen_hvm_need_lapic(void)
1880 if (xen_pv_domain())
1882 if (!xen_hvm_domain())
1884 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1888 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1890 const struct hypervisor_x86 x86_hyper_xen = {
1892 .detect = xen_platform,
1893 #ifdef CONFIG_XEN_PVHVM
1894 .init_platform = xen_hvm_guest_init,
1896 .x2apic_available = xen_x2apic_para_available,
1898 EXPORT_SYMBOL(x86_hyper_xen);
1900 #ifdef CONFIG_HOTPLUG_CPU
1901 void xen_arch_register_cpu(int num)
1903 arch_register_cpu(num);
1905 EXPORT_SYMBOL(xen_arch_register_cpu);
1907 void xen_arch_unregister_cpu(int num)
1909 arch_unregister_cpu(num);
1911 EXPORT_SYMBOL(xen_arch_unregister_cpu);