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>
47 #include <asm/paravirt.h>
50 #include <asm/xen/pci.h>
51 #include <asm/xen/hypercall.h>
52 #include <asm/xen/hypervisor.h>
53 #include <asm/fixmap.h>
54 #include <asm/processor.h>
55 #include <asm/proto.h>
56 #include <asm/msr-index.h>
57 #include <asm/traps.h>
58 #include <asm/setup.h>
60 #include <asm/pgalloc.h>
61 #include <asm/pgtable.h>
62 #include <asm/tlbflush.h>
63 #include <asm/reboot.h>
64 #include <asm/stackprotector.h>
65 #include <asm/hypervisor.h>
66 #include <asm/mwait.h>
67 #include <asm/pci_x86.h>
70 #include <linux/acpi.h>
72 #include <acpi/pdc_intel.h>
73 #include <acpi/processor.h>
74 #include <xen/interface/platform.h>
80 #include "multicalls.h"
82 EXPORT_SYMBOL_GPL(hypercall_page);
84 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
85 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
87 enum xen_domain_type xen_domain_type = XEN_NATIVE;
88 EXPORT_SYMBOL_GPL(xen_domain_type);
90 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
91 EXPORT_SYMBOL(machine_to_phys_mapping);
92 unsigned long machine_to_phys_nr;
93 EXPORT_SYMBOL(machine_to_phys_nr);
95 struct start_info *xen_start_info;
96 EXPORT_SYMBOL_GPL(xen_start_info);
98 struct shared_info xen_dummy_shared_info;
100 void *xen_initial_gdt;
102 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
103 __read_mostly int xen_have_vector_callback;
104 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
107 * Point at some empty memory to start with. We map the real shared_info
108 * page as soon as fixmap is up and running.
110 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
113 * Flag to determine whether vcpu info placement is available on all
114 * VCPUs. We assume it is to start with, and then set it to zero on
115 * the first failure. This is because it can succeed on some VCPUs
116 * and not others, since it can involve hypervisor memory allocation,
117 * or because the guest failed to guarantee all the appropriate
118 * constraints on all VCPUs (ie buffer can't cross a page boundary).
120 * Note that any particular CPU may be using a placed vcpu structure,
121 * but we can only optimise if the all are.
123 * 0: not available, 1: available
125 static int have_vcpu_info_placement = 1;
127 static void clamp_max_cpus(void)
130 if (setup_max_cpus > MAX_VIRT_CPUS)
131 setup_max_cpus = MAX_VIRT_CPUS;
135 static void xen_vcpu_setup(int cpu)
137 struct vcpu_register_vcpu_info info;
139 struct vcpu_info *vcpup;
141 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
143 if (cpu < MAX_VIRT_CPUS)
144 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
146 if (!have_vcpu_info_placement) {
147 if (cpu >= MAX_VIRT_CPUS)
152 vcpup = &per_cpu(xen_vcpu_info, cpu);
153 info.mfn = arbitrary_virt_to_mfn(vcpup);
154 info.offset = offset_in_page(vcpup);
156 /* Check to see if the hypervisor will put the vcpu_info
157 structure where we want it, which allows direct access via
158 a percpu-variable. */
159 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
162 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
163 have_vcpu_info_placement = 0;
166 /* This cpu is using the registered vcpu info, even if
167 later ones fail to. */
168 per_cpu(xen_vcpu, cpu) = vcpup;
173 * On restore, set the vcpu placement up again.
174 * If it fails, then we're in a bad state, since
175 * we can't back out from using it...
177 void xen_vcpu_restore(void)
181 for_each_online_cpu(cpu) {
182 bool other_cpu = (cpu != smp_processor_id());
185 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
188 xen_setup_runstate_info(cpu);
190 if (have_vcpu_info_placement)
194 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
199 static void __init xen_banner(void)
201 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
202 struct xen_extraversion extra;
203 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
205 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
207 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
208 version >> 16, version & 0xffff, extra.extraversion,
209 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
212 #define CPUID_THERM_POWER_LEAF 6
213 #define APERFMPERF_PRESENT 0
215 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
216 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
218 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
219 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
220 static __read_mostly unsigned int cpuid_leaf5_edx_val;
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 setecx = cpuid_leaf1_ecx_set_mask;
237 maskedx = cpuid_leaf1_edx_mask;
240 case CPUID_MWAIT_LEAF:
241 /* Synthesize the values.. */
244 *cx = cpuid_leaf5_ecx_val;
245 *dx = cpuid_leaf5_edx_val;
248 case CPUID_THERM_POWER_LEAF:
249 /* Disabling APERFMPERF for kernel usage */
250 maskecx = ~(1 << APERFMPERF_PRESENT);
254 /* Suppress extended topology stuff */
259 asm(XEN_EMULATE_PREFIX "cpuid"
264 : "0" (*ax), "2" (*cx));
273 static bool __init xen_check_mwait(void)
275 #if defined(CONFIG_ACPI) && !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR) && \
276 !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR_MODULE)
277 struct xen_platform_op op = {
278 .cmd = XENPF_set_processor_pminfo,
279 .u.set_pminfo.id = -1,
280 .u.set_pminfo.type = XEN_PM_PDC,
283 unsigned int ax, bx, cx, dx;
284 unsigned int mwait_mask;
286 /* We need to determine whether it is OK to expose the MWAIT
287 * capability to the kernel to harvest deeper than C3 states from ACPI
288 * _CST using the processor_harvest_xen.c module. For this to work, we
289 * need to gather the MWAIT_LEAF values (which the cstate.c code
290 * checks against). The hypervisor won't expose the MWAIT flag because
291 * it would break backwards compatibility; so we will find out directly
292 * from the hardware and hypercall.
294 if (!xen_initial_domain())
300 native_cpuid(&ax, &bx, &cx, &dx);
302 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
303 (1 << (X86_FEATURE_MWAIT % 32));
305 if ((cx & mwait_mask) != mwait_mask)
308 /* We need to emulate the MWAIT_LEAF and for that we need both
309 * ecx and edx. The hypercall provides only partial information.
312 ax = CPUID_MWAIT_LEAF;
317 native_cpuid(&ax, &bx, &cx, &dx);
319 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
320 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
322 buf[0] = ACPI_PDC_REVISION_ID;
324 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
326 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
328 if ((HYPERVISOR_dom0_op(&op) == 0) &&
329 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
330 cpuid_leaf5_ecx_val = cx;
331 cpuid_leaf5_edx_val = dx;
338 static void __init xen_init_cpuid_mask(void)
340 unsigned int ax, bx, cx, dx;
341 unsigned int xsave_mask;
343 cpuid_leaf1_edx_mask =
344 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
345 (1 << X86_FEATURE_MCA) | /* disable MCA */
346 (1 << X86_FEATURE_MTRR) | /* disable MTRR */
347 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
349 if (!xen_initial_domain())
350 cpuid_leaf1_edx_mask &=
351 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
352 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
355 xen_cpuid(&ax, &bx, &cx, &dx);
358 (1 << (X86_FEATURE_XSAVE % 32)) |
359 (1 << (X86_FEATURE_OSXSAVE % 32));
361 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
362 if ((cx & xsave_mask) != xsave_mask)
363 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
364 if (xen_check_mwait())
365 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
368 static void xen_set_debugreg(int reg, unsigned long val)
370 HYPERVISOR_set_debugreg(reg, val);
373 static unsigned long xen_get_debugreg(int reg)
375 return HYPERVISOR_get_debugreg(reg);
378 static void xen_end_context_switch(struct task_struct *next)
381 paravirt_end_context_switch(next);
384 static unsigned long xen_store_tr(void)
390 * Set the page permissions for a particular virtual address. If the
391 * address is a vmalloc mapping (or other non-linear mapping), then
392 * find the linear mapping of the page and also set its protections to
395 static void set_aliased_prot(void *v, pgprot_t prot)
403 ptep = lookup_address((unsigned long)v, &level);
404 BUG_ON(ptep == NULL);
406 pfn = pte_pfn(*ptep);
407 page = pfn_to_page(pfn);
409 pte = pfn_pte(pfn, prot);
411 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
414 if (!PageHighMem(page)) {
415 void *av = __va(PFN_PHYS(pfn));
418 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
424 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
426 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
429 for(i = 0; i < entries; i += entries_per_page)
430 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
433 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
435 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
438 for(i = 0; i < entries; i += entries_per_page)
439 set_aliased_prot(ldt + i, PAGE_KERNEL);
442 static void xen_set_ldt(const void *addr, unsigned entries)
444 struct mmuext_op *op;
445 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
447 trace_xen_cpu_set_ldt(addr, entries);
450 op->cmd = MMUEXT_SET_LDT;
451 op->arg1.linear_addr = (unsigned long)addr;
452 op->arg2.nr_ents = entries;
454 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
456 xen_mc_issue(PARAVIRT_LAZY_CPU);
459 static void xen_load_gdt(const struct desc_ptr *dtr)
461 unsigned long va = dtr->address;
462 unsigned int size = dtr->size + 1;
463 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
464 unsigned long frames[pages];
468 * A GDT can be up to 64k in size, which corresponds to 8192
469 * 8-byte entries, or 16 4k pages..
472 BUG_ON(size > 65536);
473 BUG_ON(va & ~PAGE_MASK);
475 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
478 unsigned long pfn, mfn;
482 * The GDT is per-cpu and is in the percpu data area.
483 * That can be virtually mapped, so we need to do a
484 * page-walk to get the underlying MFN for the
485 * hypercall. The page can also be in the kernel's
486 * linear range, so we need to RO that mapping too.
488 ptep = lookup_address(va, &level);
489 BUG_ON(ptep == NULL);
491 pfn = pte_pfn(*ptep);
492 mfn = pfn_to_mfn(pfn);
493 virt = __va(PFN_PHYS(pfn));
497 make_lowmem_page_readonly((void *)va);
498 make_lowmem_page_readonly(virt);
501 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
506 * load_gdt for early boot, when the gdt is only mapped once
508 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
510 unsigned long va = dtr->address;
511 unsigned int size = dtr->size + 1;
512 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
513 unsigned long frames[pages];
517 * A GDT can be up to 64k in size, which corresponds to 8192
518 * 8-byte entries, or 16 4k pages..
521 BUG_ON(size > 65536);
522 BUG_ON(va & ~PAGE_MASK);
524 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
526 unsigned long pfn, mfn;
528 pfn = virt_to_pfn(va);
529 mfn = pfn_to_mfn(pfn);
531 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
533 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
539 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
543 static void load_TLS_descriptor(struct thread_struct *t,
544 unsigned int cpu, unsigned int i)
546 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
547 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
548 struct multicall_space mc = __xen_mc_entry(0);
550 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
553 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
556 * XXX sleazy hack: If we're being called in a lazy-cpu zone
557 * and lazy gs handling is enabled, it means we're in a
558 * context switch, and %gs has just been saved. This means we
559 * can zero it out to prevent faults on exit from the
560 * hypervisor if the next process has no %gs. Either way, it
561 * has been saved, and the new value will get loaded properly.
562 * This will go away as soon as Xen has been modified to not
563 * save/restore %gs for normal hypercalls.
565 * On x86_64, this hack is not used for %gs, because gs points
566 * to KERNEL_GS_BASE (and uses it for PDA references), so we
567 * must not zero %gs on x86_64
569 * For x86_64, we need to zero %fs, otherwise we may get an
570 * exception between the new %fs descriptor being loaded and
571 * %fs being effectively cleared at __switch_to().
573 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
583 load_TLS_descriptor(t, cpu, 0);
584 load_TLS_descriptor(t, cpu, 1);
585 load_TLS_descriptor(t, cpu, 2);
587 xen_mc_issue(PARAVIRT_LAZY_CPU);
591 static void xen_load_gs_index(unsigned int idx)
593 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
598 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
601 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
602 u64 entry = *(u64 *)ptr;
604 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
609 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
615 static int cvt_gate_to_trap(int vector, const gate_desc *val,
616 struct trap_info *info)
620 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
623 info->vector = vector;
625 addr = gate_offset(*val);
628 * Look for known traps using IST, and substitute them
629 * appropriately. The debugger ones are the only ones we care
630 * about. Xen will handle faults like double_fault and
631 * machine_check, so we should never see them. Warn if
632 * there's an unexpected IST-using fault handler.
634 if (addr == (unsigned long)debug)
635 addr = (unsigned long)xen_debug;
636 else if (addr == (unsigned long)int3)
637 addr = (unsigned long)xen_int3;
638 else if (addr == (unsigned long)stack_segment)
639 addr = (unsigned long)xen_stack_segment;
640 else if (addr == (unsigned long)double_fault ||
641 addr == (unsigned long)nmi) {
642 /* Don't need to handle these */
644 #ifdef CONFIG_X86_MCE
645 } else if (addr == (unsigned long)machine_check) {
649 /* Some other trap using IST? */
650 if (WARN_ON(val->ist != 0))
653 #endif /* CONFIG_X86_64 */
654 info->address = addr;
656 info->cs = gate_segment(*val);
657 info->flags = val->dpl;
658 /* interrupt gates clear IF */
659 if (val->type == GATE_INTERRUPT)
660 info->flags |= 1 << 2;
665 /* Locations of each CPU's IDT */
666 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
668 /* Set an IDT entry. If the entry is part of the current IDT, then
670 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
672 unsigned long p = (unsigned long)&dt[entrynum];
673 unsigned long start, end;
675 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
679 start = __this_cpu_read(idt_desc.address);
680 end = start + __this_cpu_read(idt_desc.size) + 1;
684 native_write_idt_entry(dt, entrynum, g);
686 if (p >= start && (p + 8) <= end) {
687 struct trap_info info[2];
691 if (cvt_gate_to_trap(entrynum, g, &info[0]))
692 if (HYPERVISOR_set_trap_table(info))
699 static void xen_convert_trap_info(const struct desc_ptr *desc,
700 struct trap_info *traps)
702 unsigned in, out, count;
704 count = (desc->size+1) / sizeof(gate_desc);
707 for (in = out = 0; in < count; in++) {
708 gate_desc *entry = (gate_desc*)(desc->address) + in;
710 if (cvt_gate_to_trap(in, entry, &traps[out]))
713 traps[out].address = 0;
716 void xen_copy_trap_info(struct trap_info *traps)
718 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
720 xen_convert_trap_info(desc, traps);
723 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
724 hold a spinlock to protect the static traps[] array (static because
725 it avoids allocation, and saves stack space). */
726 static void xen_load_idt(const struct desc_ptr *desc)
728 static DEFINE_SPINLOCK(lock);
729 static struct trap_info traps[257];
731 trace_xen_cpu_load_idt(desc);
735 __get_cpu_var(idt_desc) = *desc;
737 xen_convert_trap_info(desc, traps);
740 if (HYPERVISOR_set_trap_table(traps))
746 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
747 they're handled differently. */
748 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
749 const void *desc, int type)
751 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
762 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
765 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
775 * Version of write_gdt_entry for use at early boot-time needed to
776 * update an entry as simply as possible.
778 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
779 const void *desc, int type)
781 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
790 xmaddr_t maddr = virt_to_machine(&dt[entry]);
792 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
793 dt[entry] = *(struct desc_struct *)desc;
799 static void xen_load_sp0(struct tss_struct *tss,
800 struct thread_struct *thread)
802 struct multicall_space mcs;
804 mcs = xen_mc_entry(0);
805 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
806 xen_mc_issue(PARAVIRT_LAZY_CPU);
809 static void xen_set_iopl_mask(unsigned mask)
811 struct physdev_set_iopl set_iopl;
813 /* Force the change at ring 0. */
814 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
815 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
818 static void xen_io_delay(void)
822 #ifdef CONFIG_X86_LOCAL_APIC
823 static unsigned long xen_set_apic_id(unsigned int x)
828 static unsigned int xen_get_apic_id(unsigned long x)
830 return ((x)>>24) & 0xFFu;
832 static u32 xen_apic_read(u32 reg)
834 struct xen_platform_op op = {
835 .cmd = XENPF_get_cpuinfo,
836 .interface_version = XENPF_INTERFACE_VERSION,
837 .u.pcpu_info.xen_cpuid = 0,
841 /* Shouldn't need this as APIC is turned off for PV, and we only
842 * get called on the bootup processor. But just in case. */
843 if (!xen_initial_domain() || smp_processor_id())
852 ret = HYPERVISOR_dom0_op(&op);
856 return op.u.pcpu_info.apic_id << 24;
859 static void xen_apic_write(u32 reg, u32 val)
861 /* Warn to see if there's any stray references */
865 static u64 xen_apic_icr_read(void)
870 static void xen_apic_icr_write(u32 low, u32 id)
872 /* Warn to see if there's any stray references */
876 static void xen_apic_wait_icr_idle(void)
881 static u32 xen_safe_apic_wait_icr_idle(void)
886 static void set_xen_basic_apic_ops(void)
888 apic->read = xen_apic_read;
889 apic->write = xen_apic_write;
890 apic->icr_read = xen_apic_icr_read;
891 apic->icr_write = xen_apic_icr_write;
892 apic->wait_icr_idle = xen_apic_wait_icr_idle;
893 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
894 apic->set_apic_id = xen_set_apic_id;
895 apic->get_apic_id = xen_get_apic_id;
898 apic->send_IPI_allbutself = xen_send_IPI_allbutself;
899 apic->send_IPI_mask_allbutself = xen_send_IPI_mask_allbutself;
900 apic->send_IPI_mask = xen_send_IPI_mask;
901 apic->send_IPI_all = xen_send_IPI_all;
902 apic->send_IPI_self = xen_send_IPI_self;
908 static void xen_clts(void)
910 struct multicall_space mcs;
912 mcs = xen_mc_entry(0);
914 MULTI_fpu_taskswitch(mcs.mc, 0);
916 xen_mc_issue(PARAVIRT_LAZY_CPU);
919 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
921 static unsigned long xen_read_cr0(void)
923 unsigned long cr0 = this_cpu_read(xen_cr0_value);
925 if (unlikely(cr0 == 0)) {
926 cr0 = native_read_cr0();
927 this_cpu_write(xen_cr0_value, cr0);
933 static void xen_write_cr0(unsigned long cr0)
935 struct multicall_space mcs;
937 this_cpu_write(xen_cr0_value, cr0);
939 /* Only pay attention to cr0.TS; everything else is
941 mcs = xen_mc_entry(0);
943 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
945 xen_mc_issue(PARAVIRT_LAZY_CPU);
948 static void xen_write_cr4(unsigned long cr4)
953 native_write_cr4(cr4);
956 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
967 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
968 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
969 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
972 base = ((u64)high << 32) | low;
973 if (HYPERVISOR_set_segment_base(which, base) != 0)
981 case MSR_SYSCALL_MASK:
982 case MSR_IA32_SYSENTER_CS:
983 case MSR_IA32_SYSENTER_ESP:
984 case MSR_IA32_SYSENTER_EIP:
985 /* Fast syscall setup is all done in hypercalls, so
986 these are all ignored. Stub them out here to stop
987 Xen console noise. */
990 case MSR_IA32_CR_PAT:
991 if (smp_processor_id() == 0)
992 xen_set_pat(((u64)high << 32) | low);
996 ret = native_write_msr_safe(msr, low, high);
1002 void xen_setup_shared_info(void)
1004 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1005 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1006 xen_start_info->shared_info);
1008 HYPERVISOR_shared_info =
1009 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1011 HYPERVISOR_shared_info =
1012 (struct shared_info *)__va(xen_start_info->shared_info);
1015 /* In UP this is as good a place as any to set up shared info */
1016 xen_setup_vcpu_info_placement();
1019 xen_setup_mfn_list_list();
1022 /* This is called once we have the cpu_possible_mask */
1023 void xen_setup_vcpu_info_placement(void)
1027 for_each_possible_cpu(cpu)
1028 xen_vcpu_setup(cpu);
1030 /* xen_vcpu_setup managed to place the vcpu_info within the
1031 percpu area for all cpus, so make use of it */
1032 if (have_vcpu_info_placement) {
1033 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1034 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1035 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1036 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1037 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1041 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1042 unsigned long addr, unsigned len)
1044 char *start, *end, *reloc;
1047 start = end = reloc = NULL;
1049 #define SITE(op, x) \
1050 case PARAVIRT_PATCH(op.x): \
1051 if (have_vcpu_info_placement) { \
1052 start = (char *)xen_##x##_direct; \
1053 end = xen_##x##_direct_end; \
1054 reloc = xen_##x##_direct_reloc; \
1059 SITE(pv_irq_ops, irq_enable);
1060 SITE(pv_irq_ops, irq_disable);
1061 SITE(pv_irq_ops, save_fl);
1062 SITE(pv_irq_ops, restore_fl);
1066 if (start == NULL || (end-start) > len)
1069 ret = paravirt_patch_insns(insnbuf, len, start, end);
1071 /* Note: because reloc is assigned from something that
1072 appears to be an array, gcc assumes it's non-null,
1073 but doesn't know its relationship with start and
1075 if (reloc > start && reloc < end) {
1076 int reloc_off = reloc - start;
1077 long *relocp = (long *)(insnbuf + reloc_off);
1078 long delta = start - (char *)addr;
1086 ret = paravirt_patch_default(type, clobbers, insnbuf,
1094 static const struct pv_info xen_info __initconst = {
1095 .paravirt_enabled = 1,
1096 .shared_kernel_pmd = 0,
1098 #ifdef CONFIG_X86_64
1099 .extra_user_64bit_cs = FLAT_USER_CS64,
1105 static const struct pv_init_ops xen_init_ops __initconst = {
1109 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1112 .set_debugreg = xen_set_debugreg,
1113 .get_debugreg = xen_get_debugreg,
1117 .read_cr0 = xen_read_cr0,
1118 .write_cr0 = xen_write_cr0,
1120 .read_cr4 = native_read_cr4,
1121 .read_cr4_safe = native_read_cr4_safe,
1122 .write_cr4 = xen_write_cr4,
1124 .wbinvd = native_wbinvd,
1126 .read_msr = native_read_msr_safe,
1127 .rdmsr_regs = native_rdmsr_safe_regs,
1128 .write_msr = xen_write_msr_safe,
1129 .wrmsr_regs = native_wrmsr_safe_regs,
1131 .read_tsc = native_read_tsc,
1132 .read_pmc = native_read_pmc,
1135 .irq_enable_sysexit = xen_sysexit,
1136 #ifdef CONFIG_X86_64
1137 .usergs_sysret32 = xen_sysret32,
1138 .usergs_sysret64 = xen_sysret64,
1141 .load_tr_desc = paravirt_nop,
1142 .set_ldt = xen_set_ldt,
1143 .load_gdt = xen_load_gdt,
1144 .load_idt = xen_load_idt,
1145 .load_tls = xen_load_tls,
1146 #ifdef CONFIG_X86_64
1147 .load_gs_index = xen_load_gs_index,
1150 .alloc_ldt = xen_alloc_ldt,
1151 .free_ldt = xen_free_ldt,
1153 .store_gdt = native_store_gdt,
1154 .store_idt = native_store_idt,
1155 .store_tr = xen_store_tr,
1157 .write_ldt_entry = xen_write_ldt_entry,
1158 .write_gdt_entry = xen_write_gdt_entry,
1159 .write_idt_entry = xen_write_idt_entry,
1160 .load_sp0 = xen_load_sp0,
1162 .set_iopl_mask = xen_set_iopl_mask,
1163 .io_delay = xen_io_delay,
1165 /* Xen takes care of %gs when switching to usermode for us */
1166 .swapgs = paravirt_nop,
1168 .start_context_switch = paravirt_start_context_switch,
1169 .end_context_switch = xen_end_context_switch,
1172 static const struct pv_apic_ops xen_apic_ops __initconst = {
1173 #ifdef CONFIG_X86_LOCAL_APIC
1174 .startup_ipi_hook = paravirt_nop,
1178 static void xen_reboot(int reason)
1180 struct sched_shutdown r = { .reason = reason };
1182 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1186 static void xen_restart(char *msg)
1188 xen_reboot(SHUTDOWN_reboot);
1191 static void xen_emergency_restart(void)
1193 xen_reboot(SHUTDOWN_reboot);
1196 static void xen_machine_halt(void)
1198 xen_reboot(SHUTDOWN_poweroff);
1201 static void xen_machine_power_off(void)
1205 xen_reboot(SHUTDOWN_poweroff);
1208 static void xen_crash_shutdown(struct pt_regs *regs)
1210 xen_reboot(SHUTDOWN_crash);
1214 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1216 xen_reboot(SHUTDOWN_crash);
1220 static struct notifier_block xen_panic_block = {
1221 .notifier_call= xen_panic_event,
1224 int xen_panic_handler_init(void)
1226 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1230 static const struct machine_ops xen_machine_ops __initconst = {
1231 .restart = xen_restart,
1232 .halt = xen_machine_halt,
1233 .power_off = xen_machine_power_off,
1234 .shutdown = xen_machine_halt,
1235 .crash_shutdown = xen_crash_shutdown,
1236 .emergency_restart = xen_emergency_restart,
1240 * Set up the GDT and segment registers for -fstack-protector. Until
1241 * we do this, we have to be careful not to call any stack-protected
1242 * function, which is most of the kernel.
1244 static void __init xen_setup_stackprotector(void)
1246 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1247 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1249 setup_stack_canary_segment(0);
1250 switch_to_new_gdt(0);
1252 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1253 pv_cpu_ops.load_gdt = xen_load_gdt;
1256 /* First C function to be called on Xen boot */
1257 asmlinkage void __init xen_start_kernel(void)
1259 struct physdev_set_iopl set_iopl;
1263 if (!xen_start_info)
1266 xen_domain_type = XEN_PV_DOMAIN;
1268 xen_setup_machphys_mapping();
1270 /* Install Xen paravirt ops */
1272 pv_init_ops = xen_init_ops;
1273 pv_cpu_ops = xen_cpu_ops;
1274 pv_apic_ops = xen_apic_ops;
1276 x86_init.resources.memory_setup = xen_memory_setup;
1277 x86_init.oem.arch_setup = xen_arch_setup;
1278 x86_init.oem.banner = xen_banner;
1280 xen_init_time_ops();
1283 * Set up some pagetable state before starting to set any ptes.
1288 /* Prevent unwanted bits from being set in PTEs. */
1289 __supported_pte_mask &= ~_PAGE_GLOBAL;
1291 if (!xen_initial_domain())
1293 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1295 __supported_pte_mask |= _PAGE_IOMAP;
1298 * Prevent page tables from being allocated in highmem, even
1299 * if CONFIG_HIGHPTE is enabled.
1301 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1303 /* Work out if we support NX */
1306 xen_setup_features();
1309 if (!xen_feature(XENFEAT_auto_translated_physmap))
1310 xen_build_dynamic_phys_to_machine();
1313 * Set up kernel GDT and segment registers, mainly so that
1314 * -fstack-protector code can be executed.
1316 xen_setup_stackprotector();
1319 xen_init_cpuid_mask();
1321 #ifdef CONFIG_X86_LOCAL_APIC
1323 * set up the basic apic ops.
1325 set_xen_basic_apic_ops();
1328 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1329 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1330 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1333 machine_ops = xen_machine_ops;
1336 * The only reliable way to retain the initial address of the
1337 * percpu gdt_page is to remember it here, so we can go and
1338 * mark it RW later, when the initial percpu area is freed.
1340 xen_initial_gdt = &per_cpu(gdt_page, 0);
1344 #ifdef CONFIG_ACPI_NUMA
1346 * The pages we from Xen are not related to machine pages, so
1347 * any NUMA information the kernel tries to get from ACPI will
1348 * be meaningless. Prevent it from trying.
1353 pgd = (pgd_t *)xen_start_info->pt_base;
1355 /* Don't do the full vcpu_info placement stuff until we have a
1356 possible map and a non-dummy shared_info. */
1357 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1359 local_irq_disable();
1360 early_boot_irqs_disabled = true;
1362 xen_raw_console_write("mapping kernel into physical memory\n");
1363 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1365 /* Allocate and initialize top and mid mfn levels for p2m structure */
1366 xen_build_mfn_list_list();
1368 /* keep using Xen gdt for now; no urgent need to change it */
1370 #ifdef CONFIG_X86_32
1371 pv_info.kernel_rpl = 1;
1372 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1373 pv_info.kernel_rpl = 0;
1375 pv_info.kernel_rpl = 0;
1377 /* set the limit of our address space */
1380 /* We used to do this in xen_arch_setup, but that is too late on AMD
1381 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1382 * which pokes 0xcf8 port.
1385 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1387 xen_raw_printk("physdev_op failed %d\n", rc);
1389 #ifdef CONFIG_X86_32
1390 /* set up basic CPUID stuff */
1391 cpu_detect(&new_cpu_data);
1392 new_cpu_data.hard_math = 1;
1393 new_cpu_data.wp_works_ok = 1;
1394 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1397 /* Poke various useful things into boot_params */
1398 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1399 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1400 ? __pa(xen_start_info->mod_start) : 0;
1401 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1402 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1404 if (!xen_initial_domain()) {
1405 add_preferred_console("xenboot", 0, NULL);
1406 add_preferred_console("tty", 0, NULL);
1407 add_preferred_console("hvc", 0, NULL);
1409 x86_init.pci.arch_init = pci_xen_init;
1411 const struct dom0_vga_console_info *info =
1412 (void *)((char *)xen_start_info +
1413 xen_start_info->console.dom0.info_off);
1415 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1416 xen_start_info->console.domU.mfn = 0;
1417 xen_start_info->console.domU.evtchn = 0;
1421 /* Make sure ACS will be enabled */
1424 xen_acpi_sleep_register();
1427 /* PCI BIOS service won't work from a PV guest. */
1428 pci_probe &= ~PCI_PROBE_BIOS;
1430 xen_raw_console_write("about to get started...\n");
1432 xen_setup_runstate_info(0);
1434 /* Start the world */
1435 #ifdef CONFIG_X86_32
1436 i386_start_kernel();
1438 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1442 static int init_hvm_pv_info(int *major, int *minor)
1444 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1447 base = xen_cpuid_base();
1448 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1451 *minor = eax & 0xffff;
1452 printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1454 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1456 pfn = __pa(hypercall_page);
1457 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1459 xen_setup_features();
1461 pv_info.name = "Xen HVM";
1463 xen_domain_type = XEN_HVM_DOMAIN;
1468 void __ref xen_hvm_init_shared_info(void)
1471 struct xen_add_to_physmap xatp;
1472 static struct shared_info *shared_info_page = 0;
1474 if (!shared_info_page)
1475 shared_info_page = (struct shared_info *)
1476 extend_brk(PAGE_SIZE, PAGE_SIZE);
1477 xatp.domid = DOMID_SELF;
1479 xatp.space = XENMAPSPACE_shared_info;
1480 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1481 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1484 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1486 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1487 * page, we use it in the event channel upcall and in some pvclock
1488 * related functions. We don't need the vcpu_info placement
1489 * optimizations because we don't use any pv_mmu or pv_irq op on
1491 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1492 * online but xen_hvm_init_shared_info is run at resume time too and
1493 * in that case multiple vcpus might be online. */
1494 for_each_online_cpu(cpu) {
1495 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1499 #ifdef CONFIG_XEN_PVHVM
1500 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1501 unsigned long action, void *hcpu)
1503 int cpu = (long)hcpu;
1505 case CPU_UP_PREPARE:
1506 xen_vcpu_setup(cpu);
1507 if (xen_have_vector_callback)
1508 xen_init_lock_cpu(cpu);
1516 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1517 .notifier_call = xen_hvm_cpu_notify,
1520 static void __init xen_hvm_guest_init(void)
1525 r = init_hvm_pv_info(&major, &minor);
1529 xen_hvm_init_shared_info();
1531 if (xen_feature(XENFEAT_hvm_callback_vector))
1532 xen_have_vector_callback = 1;
1534 register_cpu_notifier(&xen_hvm_cpu_notifier);
1535 xen_unplug_emulated_devices();
1536 x86_init.irqs.intr_init = xen_init_IRQ;
1537 xen_hvm_init_time_ops();
1538 xen_hvm_init_mmu_ops();
1541 static bool __init xen_hvm_platform(void)
1543 if (xen_pv_domain())
1546 if (!xen_cpuid_base())
1552 bool xen_hvm_need_lapic(void)
1554 if (xen_pv_domain())
1556 if (!xen_hvm_domain())
1558 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1562 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1564 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1566 .detect = xen_hvm_platform,
1567 .init_platform = xen_hvm_guest_init,
1569 EXPORT_SYMBOL(x86_hyper_xen_hvm);