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/mwait.h>
66 #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>
79 #include "multicalls.h"
81 EXPORT_SYMBOL_GPL(hypercall_page);
83 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
84 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
86 enum xen_domain_type xen_domain_type = XEN_NATIVE;
87 EXPORT_SYMBOL_GPL(xen_domain_type);
89 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
90 EXPORT_SYMBOL(machine_to_phys_mapping);
91 unsigned long machine_to_phys_nr;
92 EXPORT_SYMBOL(machine_to_phys_nr);
94 struct start_info *xen_start_info;
95 EXPORT_SYMBOL_GPL(xen_start_info);
97 struct shared_info xen_dummy_shared_info;
99 void *xen_initial_gdt;
101 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
102 __read_mostly int xen_have_vector_callback;
103 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
106 * Point at some empty memory to start with. We map the real shared_info
107 * page as soon as fixmap is up and running.
109 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
112 * Flag to determine whether vcpu info placement is available on all
113 * VCPUs. We assume it is to start with, and then set it to zero on
114 * the first failure. This is because it can succeed on some VCPUs
115 * and not others, since it can involve hypervisor memory allocation,
116 * or because the guest failed to guarantee all the appropriate
117 * constraints on all VCPUs (ie buffer can't cross a page boundary).
119 * Note that any particular CPU may be using a placed vcpu structure,
120 * but we can only optimise if the all are.
122 * 0: not available, 1: available
124 static int have_vcpu_info_placement = 1;
126 static void clamp_max_cpus(void)
129 if (setup_max_cpus > MAX_VIRT_CPUS)
130 setup_max_cpus = MAX_VIRT_CPUS;
134 static void xen_vcpu_setup(int cpu)
136 struct vcpu_register_vcpu_info info;
138 struct vcpu_info *vcpup;
140 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
143 * This path is called twice on PVHVM - first during bootup via
144 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
145 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
146 * As we can only do the VCPUOP_register_vcpu_info once lets
147 * not over-write its result.
149 * For PV it is called during restore (xen_vcpu_restore) and bootup
150 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
153 if (xen_hvm_domain()) {
154 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
157 if (cpu < MAX_VIRT_CPUS)
158 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
160 if (!have_vcpu_info_placement) {
161 if (cpu >= MAX_VIRT_CPUS)
166 vcpup = &per_cpu(xen_vcpu_info, cpu);
167 info.mfn = arbitrary_virt_to_mfn(vcpup);
168 info.offset = offset_in_page(vcpup);
170 /* Check to see if the hypervisor will put the vcpu_info
171 structure where we want it, which allows direct access via
172 a percpu-variable. */
173 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
176 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
177 have_vcpu_info_placement = 0;
180 /* This cpu is using the registered vcpu info, even if
181 later ones fail to. */
182 per_cpu(xen_vcpu, cpu) = vcpup;
187 * On restore, set the vcpu placement up again.
188 * If it fails, then we're in a bad state, since
189 * we can't back out from using it...
191 void xen_vcpu_restore(void)
195 for_each_online_cpu(cpu) {
196 bool other_cpu = (cpu != smp_processor_id());
199 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
202 xen_setup_runstate_info(cpu);
204 if (have_vcpu_info_placement)
208 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
213 static void __init xen_banner(void)
215 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
216 struct xen_extraversion extra;
217 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
219 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
221 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
222 version >> 16, version & 0xffff, extra.extraversion,
223 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
226 #define CPUID_THERM_POWER_LEAF 6
227 #define APERFMPERF_PRESENT 0
229 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
230 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
232 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
233 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
234 static __read_mostly unsigned int cpuid_leaf5_edx_val;
236 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
237 unsigned int *cx, unsigned int *dx)
239 unsigned maskebx = ~0;
240 unsigned maskecx = ~0;
241 unsigned maskedx = ~0;
244 * Mask out inconvenient features, to try and disable as many
245 * unsupported kernel subsystems as possible.
249 maskecx = cpuid_leaf1_ecx_mask;
250 setecx = cpuid_leaf1_ecx_set_mask;
251 maskedx = cpuid_leaf1_edx_mask;
254 case CPUID_MWAIT_LEAF:
255 /* Synthesize the values.. */
258 *cx = cpuid_leaf5_ecx_val;
259 *dx = cpuid_leaf5_edx_val;
262 case CPUID_THERM_POWER_LEAF:
263 /* Disabling APERFMPERF for kernel usage */
264 maskecx = ~(1 << APERFMPERF_PRESENT);
268 /* Suppress extended topology stuff */
273 asm(XEN_EMULATE_PREFIX "cpuid"
278 : "0" (*ax), "2" (*cx));
287 static bool __init xen_check_mwait(void)
289 #if defined(CONFIG_ACPI) && !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR) && \
290 !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR_MODULE)
291 struct xen_platform_op op = {
292 .cmd = XENPF_set_processor_pminfo,
293 .u.set_pminfo.id = -1,
294 .u.set_pminfo.type = XEN_PM_PDC,
297 unsigned int ax, bx, cx, dx;
298 unsigned int mwait_mask;
300 /* We need to determine whether it is OK to expose the MWAIT
301 * capability to the kernel to harvest deeper than C3 states from ACPI
302 * _CST using the processor_harvest_xen.c module. For this to work, we
303 * need to gather the MWAIT_LEAF values (which the cstate.c code
304 * checks against). The hypervisor won't expose the MWAIT flag because
305 * it would break backwards compatibility; so we will find out directly
306 * from the hardware and hypercall.
308 if (!xen_initial_domain())
314 native_cpuid(&ax, &bx, &cx, &dx);
316 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
317 (1 << (X86_FEATURE_MWAIT % 32));
319 if ((cx & mwait_mask) != mwait_mask)
322 /* We need to emulate the MWAIT_LEAF and for that we need both
323 * ecx and edx. The hypercall provides only partial information.
326 ax = CPUID_MWAIT_LEAF;
331 native_cpuid(&ax, &bx, &cx, &dx);
333 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
334 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
336 buf[0] = ACPI_PDC_REVISION_ID;
338 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
340 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
342 if ((HYPERVISOR_dom0_op(&op) == 0) &&
343 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
344 cpuid_leaf5_ecx_val = cx;
345 cpuid_leaf5_edx_val = dx;
352 static void __init xen_init_cpuid_mask(void)
354 unsigned int ax, bx, cx, dx;
355 unsigned int xsave_mask;
357 cpuid_leaf1_edx_mask =
358 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
359 (1 << X86_FEATURE_MCA) | /* disable MCA */
360 (1 << X86_FEATURE_MTRR) | /* disable MTRR */
361 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
363 if (!xen_initial_domain())
364 cpuid_leaf1_edx_mask &=
365 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
366 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
369 xen_cpuid(&ax, &bx, &cx, &dx);
372 (1 << (X86_FEATURE_XSAVE % 32)) |
373 (1 << (X86_FEATURE_OSXSAVE % 32));
375 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
376 if ((cx & xsave_mask) != xsave_mask)
377 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
378 if (xen_check_mwait())
379 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
382 static void xen_set_debugreg(int reg, unsigned long val)
384 HYPERVISOR_set_debugreg(reg, val);
387 static unsigned long xen_get_debugreg(int reg)
389 return HYPERVISOR_get_debugreg(reg);
392 static void xen_end_context_switch(struct task_struct *next)
395 paravirt_end_context_switch(next);
398 static unsigned long xen_store_tr(void)
404 * Set the page permissions for a particular virtual address. If the
405 * address is a vmalloc mapping (or other non-linear mapping), then
406 * find the linear mapping of the page and also set its protections to
409 static void set_aliased_prot(void *v, pgprot_t prot)
417 ptep = lookup_address((unsigned long)v, &level);
418 BUG_ON(ptep == NULL);
420 pfn = pte_pfn(*ptep);
421 page = pfn_to_page(pfn);
423 pte = pfn_pte(pfn, prot);
425 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
428 if (!PageHighMem(page)) {
429 void *av = __va(PFN_PHYS(pfn));
432 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
438 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
440 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
443 for(i = 0; i < entries; i += entries_per_page)
444 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
447 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
449 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
452 for(i = 0; i < entries; i += entries_per_page)
453 set_aliased_prot(ldt + i, PAGE_KERNEL);
456 static void xen_set_ldt(const void *addr, unsigned entries)
458 struct mmuext_op *op;
459 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
461 trace_xen_cpu_set_ldt(addr, entries);
464 op->cmd = MMUEXT_SET_LDT;
465 op->arg1.linear_addr = (unsigned long)addr;
466 op->arg2.nr_ents = entries;
468 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
470 xen_mc_issue(PARAVIRT_LAZY_CPU);
473 static void xen_load_gdt(const struct desc_ptr *dtr)
475 unsigned long va = dtr->address;
476 unsigned int size = dtr->size + 1;
477 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
478 unsigned long frames[pages];
482 * A GDT can be up to 64k in size, which corresponds to 8192
483 * 8-byte entries, or 16 4k pages..
486 BUG_ON(size > 65536);
487 BUG_ON(va & ~PAGE_MASK);
489 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
492 unsigned long pfn, mfn;
496 * The GDT is per-cpu and is in the percpu data area.
497 * That can be virtually mapped, so we need to do a
498 * page-walk to get the underlying MFN for the
499 * hypercall. The page can also be in the kernel's
500 * linear range, so we need to RO that mapping too.
502 ptep = lookup_address(va, &level);
503 BUG_ON(ptep == NULL);
505 pfn = pte_pfn(*ptep);
506 mfn = pfn_to_mfn(pfn);
507 virt = __va(PFN_PHYS(pfn));
511 make_lowmem_page_readonly((void *)va);
512 make_lowmem_page_readonly(virt);
515 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
520 * load_gdt for early boot, when the gdt is only mapped once
522 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
524 unsigned long va = dtr->address;
525 unsigned int size = dtr->size + 1;
526 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
527 unsigned long frames[pages];
531 * A GDT can be up to 64k in size, which corresponds to 8192
532 * 8-byte entries, or 16 4k pages..
535 BUG_ON(size > 65536);
536 BUG_ON(va & ~PAGE_MASK);
538 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
540 unsigned long pfn, mfn;
542 pfn = virt_to_pfn(va);
543 mfn = pfn_to_mfn(pfn);
545 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
547 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
553 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
557 static void load_TLS_descriptor(struct thread_struct *t,
558 unsigned int cpu, unsigned int i)
560 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
561 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
562 struct multicall_space mc = __xen_mc_entry(0);
564 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
567 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
570 * XXX sleazy hack: If we're being called in a lazy-cpu zone
571 * and lazy gs handling is enabled, it means we're in a
572 * context switch, and %gs has just been saved. This means we
573 * can zero it out to prevent faults on exit from the
574 * hypervisor if the next process has no %gs. Either way, it
575 * has been saved, and the new value will get loaded properly.
576 * This will go away as soon as Xen has been modified to not
577 * save/restore %gs for normal hypercalls.
579 * On x86_64, this hack is not used for %gs, because gs points
580 * to KERNEL_GS_BASE (and uses it for PDA references), so we
581 * must not zero %gs on x86_64
583 * For x86_64, we need to zero %fs, otherwise we may get an
584 * exception between the new %fs descriptor being loaded and
585 * %fs being effectively cleared at __switch_to().
587 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
597 load_TLS_descriptor(t, cpu, 0);
598 load_TLS_descriptor(t, cpu, 1);
599 load_TLS_descriptor(t, cpu, 2);
601 xen_mc_issue(PARAVIRT_LAZY_CPU);
605 static void xen_load_gs_index(unsigned int idx)
607 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
612 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
615 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
616 u64 entry = *(u64 *)ptr;
618 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
623 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
629 static int cvt_gate_to_trap(int vector, const gate_desc *val,
630 struct trap_info *info)
634 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
637 info->vector = vector;
639 addr = gate_offset(*val);
642 * Look for known traps using IST, and substitute them
643 * appropriately. The debugger ones are the only ones we care
644 * about. Xen will handle faults like double_fault and
645 * machine_check, so we should never see them. Warn if
646 * there's an unexpected IST-using fault handler.
648 if (addr == (unsigned long)debug)
649 addr = (unsigned long)xen_debug;
650 else if (addr == (unsigned long)int3)
651 addr = (unsigned long)xen_int3;
652 else if (addr == (unsigned long)stack_segment)
653 addr = (unsigned long)xen_stack_segment;
654 else if (addr == (unsigned long)double_fault ||
655 addr == (unsigned long)nmi) {
656 /* Don't need to handle these */
658 #ifdef CONFIG_X86_MCE
659 } else if (addr == (unsigned long)machine_check) {
663 /* Some other trap using IST? */
664 if (WARN_ON(val->ist != 0))
667 #endif /* CONFIG_X86_64 */
668 info->address = addr;
670 info->cs = gate_segment(*val);
671 info->flags = val->dpl;
672 /* interrupt gates clear IF */
673 if (val->type == GATE_INTERRUPT)
674 info->flags |= 1 << 2;
679 /* Locations of each CPU's IDT */
680 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
682 /* Set an IDT entry. If the entry is part of the current IDT, then
684 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
686 unsigned long p = (unsigned long)&dt[entrynum];
687 unsigned long start, end;
689 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
693 start = __this_cpu_read(idt_desc.address);
694 end = start + __this_cpu_read(idt_desc.size) + 1;
698 native_write_idt_entry(dt, entrynum, g);
700 if (p >= start && (p + 8) <= end) {
701 struct trap_info info[2];
705 if (cvt_gate_to_trap(entrynum, g, &info[0]))
706 if (HYPERVISOR_set_trap_table(info))
713 static void xen_convert_trap_info(const struct desc_ptr *desc,
714 struct trap_info *traps)
716 unsigned in, out, count;
718 count = (desc->size+1) / sizeof(gate_desc);
721 for (in = out = 0; in < count; in++) {
722 gate_desc *entry = (gate_desc*)(desc->address) + in;
724 if (cvt_gate_to_trap(in, entry, &traps[out]))
727 traps[out].address = 0;
730 void xen_copy_trap_info(struct trap_info *traps)
732 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
734 xen_convert_trap_info(desc, traps);
737 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
738 hold a spinlock to protect the static traps[] array (static because
739 it avoids allocation, and saves stack space). */
740 static void xen_load_idt(const struct desc_ptr *desc)
742 static DEFINE_SPINLOCK(lock);
743 static struct trap_info traps[257];
745 trace_xen_cpu_load_idt(desc);
749 __get_cpu_var(idt_desc) = *desc;
751 xen_convert_trap_info(desc, traps);
754 if (HYPERVISOR_set_trap_table(traps))
760 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
761 they're handled differently. */
762 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
763 const void *desc, int type)
765 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
776 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
779 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
789 * Version of write_gdt_entry for use at early boot-time needed to
790 * update an entry as simply as possible.
792 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
793 const void *desc, int type)
795 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
804 xmaddr_t maddr = virt_to_machine(&dt[entry]);
806 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
807 dt[entry] = *(struct desc_struct *)desc;
813 static void xen_load_sp0(struct tss_struct *tss,
814 struct thread_struct *thread)
816 struct multicall_space mcs;
818 mcs = xen_mc_entry(0);
819 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
820 xen_mc_issue(PARAVIRT_LAZY_CPU);
823 static void xen_set_iopl_mask(unsigned mask)
825 struct physdev_set_iopl set_iopl;
827 /* Force the change at ring 0. */
828 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
829 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
832 static void xen_io_delay(void)
836 #ifdef CONFIG_X86_LOCAL_APIC
837 static unsigned long xen_set_apic_id(unsigned int x)
842 static unsigned int xen_get_apic_id(unsigned long x)
844 return ((x)>>24) & 0xFFu;
846 static u32 xen_apic_read(u32 reg)
848 struct xen_platform_op op = {
849 .cmd = XENPF_get_cpuinfo,
850 .interface_version = XENPF_INTERFACE_VERSION,
851 .u.pcpu_info.xen_cpuid = 0,
855 /* Shouldn't need this as APIC is turned off for PV, and we only
856 * get called on the bootup processor. But just in case. */
857 if (!xen_initial_domain() || smp_processor_id())
866 ret = HYPERVISOR_dom0_op(&op);
870 return op.u.pcpu_info.apic_id << 24;
873 static void xen_apic_write(u32 reg, u32 val)
875 /* Warn to see if there's any stray references */
879 static u64 xen_apic_icr_read(void)
884 static void xen_apic_icr_write(u32 low, u32 id)
886 /* Warn to see if there's any stray references */
890 static void xen_apic_wait_icr_idle(void)
895 static u32 xen_safe_apic_wait_icr_idle(void)
900 static void set_xen_basic_apic_ops(void)
902 apic->read = xen_apic_read;
903 apic->write = xen_apic_write;
904 apic->icr_read = xen_apic_icr_read;
905 apic->icr_write = xen_apic_icr_write;
906 apic->wait_icr_idle = xen_apic_wait_icr_idle;
907 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
908 apic->set_apic_id = xen_set_apic_id;
909 apic->get_apic_id = xen_get_apic_id;
914 static void xen_clts(void)
916 struct multicall_space mcs;
918 mcs = xen_mc_entry(0);
920 MULTI_fpu_taskswitch(mcs.mc, 0);
922 xen_mc_issue(PARAVIRT_LAZY_CPU);
925 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
927 static unsigned long xen_read_cr0(void)
929 unsigned long cr0 = this_cpu_read(xen_cr0_value);
931 if (unlikely(cr0 == 0)) {
932 cr0 = native_read_cr0();
933 this_cpu_write(xen_cr0_value, cr0);
939 static void xen_write_cr0(unsigned long cr0)
941 struct multicall_space mcs;
943 this_cpu_write(xen_cr0_value, cr0);
945 /* Only pay attention to cr0.TS; everything else is
947 mcs = xen_mc_entry(0);
949 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
951 xen_mc_issue(PARAVIRT_LAZY_CPU);
954 static void xen_write_cr4(unsigned long cr4)
959 native_write_cr4(cr4);
962 static inline unsigned long xen_read_cr8(void)
966 static inline void xen_write_cr8(unsigned long val)
971 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
982 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
983 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
984 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
987 base = ((u64)high << 32) | low;
988 if (HYPERVISOR_set_segment_base(which, base) != 0)
996 case MSR_SYSCALL_MASK:
997 case MSR_IA32_SYSENTER_CS:
998 case MSR_IA32_SYSENTER_ESP:
999 case MSR_IA32_SYSENTER_EIP:
1000 /* Fast syscall setup is all done in hypercalls, so
1001 these are all ignored. Stub them out here to stop
1002 Xen console noise. */
1005 case MSR_IA32_CR_PAT:
1006 if (smp_processor_id() == 0)
1007 xen_set_pat(((u64)high << 32) | low);
1011 ret = native_write_msr_safe(msr, low, high);
1017 void xen_setup_shared_info(void)
1019 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1020 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1021 xen_start_info->shared_info);
1023 HYPERVISOR_shared_info =
1024 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1026 HYPERVISOR_shared_info =
1027 (struct shared_info *)__va(xen_start_info->shared_info);
1030 /* In UP this is as good a place as any to set up shared info */
1031 xen_setup_vcpu_info_placement();
1034 xen_setup_mfn_list_list();
1037 /* This is called once we have the cpu_possible_mask */
1038 void xen_setup_vcpu_info_placement(void)
1042 for_each_possible_cpu(cpu)
1043 xen_vcpu_setup(cpu);
1045 /* xen_vcpu_setup managed to place the vcpu_info within the
1046 percpu area for all cpus, so make use of it */
1047 if (have_vcpu_info_placement) {
1048 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1049 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1050 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1051 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1052 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1056 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1057 unsigned long addr, unsigned len)
1059 char *start, *end, *reloc;
1062 start = end = reloc = NULL;
1064 #define SITE(op, x) \
1065 case PARAVIRT_PATCH(op.x): \
1066 if (have_vcpu_info_placement) { \
1067 start = (char *)xen_##x##_direct; \
1068 end = xen_##x##_direct_end; \
1069 reloc = xen_##x##_direct_reloc; \
1074 SITE(pv_irq_ops, irq_enable);
1075 SITE(pv_irq_ops, irq_disable);
1076 SITE(pv_irq_ops, save_fl);
1077 SITE(pv_irq_ops, restore_fl);
1081 if (start == NULL || (end-start) > len)
1084 ret = paravirt_patch_insns(insnbuf, len, start, end);
1086 /* Note: because reloc is assigned from something that
1087 appears to be an array, gcc assumes it's non-null,
1088 but doesn't know its relationship with start and
1090 if (reloc > start && reloc < end) {
1091 int reloc_off = reloc - start;
1092 long *relocp = (long *)(insnbuf + reloc_off);
1093 long delta = start - (char *)addr;
1101 ret = paravirt_patch_default(type, clobbers, insnbuf,
1109 static const struct pv_info xen_info __initconst = {
1110 .paravirt_enabled = 1,
1111 .shared_kernel_pmd = 0,
1113 #ifdef CONFIG_X86_64
1114 .extra_user_64bit_cs = FLAT_USER_CS64,
1120 static const struct pv_init_ops xen_init_ops __initconst = {
1124 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1127 .set_debugreg = xen_set_debugreg,
1128 .get_debugreg = xen_get_debugreg,
1132 .read_cr0 = xen_read_cr0,
1133 .write_cr0 = xen_write_cr0,
1135 .read_cr4 = native_read_cr4,
1136 .read_cr4_safe = native_read_cr4_safe,
1137 .write_cr4 = xen_write_cr4,
1139 #ifdef CONFIG_X86_64
1140 .read_cr8 = xen_read_cr8,
1141 .write_cr8 = xen_write_cr8,
1144 .wbinvd = native_wbinvd,
1146 .read_msr = native_read_msr_safe,
1147 .rdmsr_regs = native_rdmsr_safe_regs,
1148 .write_msr = xen_write_msr_safe,
1149 .wrmsr_regs = native_wrmsr_safe_regs,
1151 .read_tsc = native_read_tsc,
1152 .read_pmc = native_read_pmc,
1154 .read_tscp = native_read_tscp,
1157 .irq_enable_sysexit = xen_sysexit,
1158 #ifdef CONFIG_X86_64
1159 .usergs_sysret32 = xen_sysret32,
1160 .usergs_sysret64 = xen_sysret64,
1163 .load_tr_desc = paravirt_nop,
1164 .set_ldt = xen_set_ldt,
1165 .load_gdt = xen_load_gdt,
1166 .load_idt = xen_load_idt,
1167 .load_tls = xen_load_tls,
1168 #ifdef CONFIG_X86_64
1169 .load_gs_index = xen_load_gs_index,
1172 .alloc_ldt = xen_alloc_ldt,
1173 .free_ldt = xen_free_ldt,
1175 .store_gdt = native_store_gdt,
1176 .store_idt = native_store_idt,
1177 .store_tr = xen_store_tr,
1179 .write_ldt_entry = xen_write_ldt_entry,
1180 .write_gdt_entry = xen_write_gdt_entry,
1181 .write_idt_entry = xen_write_idt_entry,
1182 .load_sp0 = xen_load_sp0,
1184 .set_iopl_mask = xen_set_iopl_mask,
1185 .io_delay = xen_io_delay,
1187 /* Xen takes care of %gs when switching to usermode for us */
1188 .swapgs = paravirt_nop,
1190 .start_context_switch = paravirt_start_context_switch,
1191 .end_context_switch = xen_end_context_switch,
1194 static const struct pv_apic_ops xen_apic_ops __initconst = {
1195 #ifdef CONFIG_X86_LOCAL_APIC
1196 .startup_ipi_hook = paravirt_nop,
1200 static void xen_reboot(int reason)
1202 struct sched_shutdown r = { .reason = reason };
1204 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1208 static void xen_restart(char *msg)
1210 xen_reboot(SHUTDOWN_reboot);
1213 static void xen_emergency_restart(void)
1215 xen_reboot(SHUTDOWN_reboot);
1218 static void xen_machine_halt(void)
1220 xen_reboot(SHUTDOWN_poweroff);
1223 static void xen_machine_power_off(void)
1227 xen_reboot(SHUTDOWN_poweroff);
1230 static void xen_crash_shutdown(struct pt_regs *regs)
1232 xen_reboot(SHUTDOWN_crash);
1236 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1238 xen_reboot(SHUTDOWN_crash);
1242 static struct notifier_block xen_panic_block = {
1243 .notifier_call= xen_panic_event,
1246 int xen_panic_handler_init(void)
1248 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1252 static const struct machine_ops xen_machine_ops __initconst = {
1253 .restart = xen_restart,
1254 .halt = xen_machine_halt,
1255 .power_off = xen_machine_power_off,
1256 .shutdown = xen_machine_halt,
1257 .crash_shutdown = xen_crash_shutdown,
1258 .emergency_restart = xen_emergency_restart,
1262 * Set up the GDT and segment registers for -fstack-protector. Until
1263 * we do this, we have to be careful not to call any stack-protected
1264 * function, which is most of the kernel.
1266 static void __init xen_setup_stackprotector(void)
1268 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1269 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1271 setup_stack_canary_segment(0);
1272 switch_to_new_gdt(0);
1274 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1275 pv_cpu_ops.load_gdt = xen_load_gdt;
1278 /* First C function to be called on Xen boot */
1279 asmlinkage void __init xen_start_kernel(void)
1281 struct physdev_set_iopl set_iopl;
1285 if (!xen_start_info)
1288 xen_domain_type = XEN_PV_DOMAIN;
1290 xen_setup_machphys_mapping();
1292 /* Install Xen paravirt ops */
1294 pv_init_ops = xen_init_ops;
1295 pv_cpu_ops = xen_cpu_ops;
1296 pv_apic_ops = xen_apic_ops;
1298 x86_init.resources.memory_setup = xen_memory_setup;
1299 x86_init.oem.arch_setup = xen_arch_setup;
1300 x86_init.oem.banner = xen_banner;
1302 xen_init_time_ops();
1305 * Set up some pagetable state before starting to set any ptes.
1310 /* Prevent unwanted bits from being set in PTEs. */
1311 __supported_pte_mask &= ~_PAGE_GLOBAL;
1313 if (!xen_initial_domain())
1315 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1317 __supported_pte_mask |= _PAGE_IOMAP;
1320 * Prevent page tables from being allocated in highmem, even
1321 * if CONFIG_HIGHPTE is enabled.
1323 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1325 /* Work out if we support NX */
1328 xen_setup_features();
1331 if (!xen_feature(XENFEAT_auto_translated_physmap))
1332 xen_build_dynamic_phys_to_machine();
1335 * Set up kernel GDT and segment registers, mainly so that
1336 * -fstack-protector code can be executed.
1338 xen_setup_stackprotector();
1341 xen_init_cpuid_mask();
1343 #ifdef CONFIG_X86_LOCAL_APIC
1345 * set up the basic apic ops.
1347 set_xen_basic_apic_ops();
1350 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1351 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1352 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1355 machine_ops = xen_machine_ops;
1358 * The only reliable way to retain the initial address of the
1359 * percpu gdt_page is to remember it here, so we can go and
1360 * mark it RW later, when the initial percpu area is freed.
1362 xen_initial_gdt = &per_cpu(gdt_page, 0);
1366 #ifdef CONFIG_ACPI_NUMA
1368 * The pages we from Xen are not related to machine pages, so
1369 * any NUMA information the kernel tries to get from ACPI will
1370 * be meaningless. Prevent it from trying.
1374 #ifdef CONFIG_X86_PAT
1376 * For right now disable the PAT. We should remove this once
1377 * git commit 8eaffa67b43e99ae581622c5133e20b0f48bcef1
1378 * (xen/pat: Disable PAT support for now) is reverted.
1382 pgd = (pgd_t *)xen_start_info->pt_base;
1384 /* Don't do the full vcpu_info placement stuff until we have a
1385 possible map and a non-dummy shared_info. */
1386 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1388 local_irq_disable();
1389 early_boot_irqs_disabled = true;
1391 xen_raw_console_write("mapping kernel into physical memory\n");
1392 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1393 xen_ident_map_ISA();
1395 /* Allocate and initialize top and mid mfn levels for p2m structure */
1396 xen_build_mfn_list_list();
1398 /* keep using Xen gdt for now; no urgent need to change it */
1400 #ifdef CONFIG_X86_32
1401 pv_info.kernel_rpl = 1;
1402 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1403 pv_info.kernel_rpl = 0;
1405 pv_info.kernel_rpl = 0;
1407 /* set the limit of our address space */
1410 /* We used to do this in xen_arch_setup, but that is too late on AMD
1411 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1412 * which pokes 0xcf8 port.
1415 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1417 xen_raw_printk("physdev_op failed %d\n", rc);
1419 #ifdef CONFIG_X86_32
1420 /* set up basic CPUID stuff */
1421 cpu_detect(&new_cpu_data);
1422 new_cpu_data.hard_math = 1;
1423 new_cpu_data.wp_works_ok = 1;
1424 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1427 /* Poke various useful things into boot_params */
1428 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1429 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1430 ? __pa(xen_start_info->mod_start) : 0;
1431 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1432 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1434 if (!xen_initial_domain()) {
1435 add_preferred_console("xenboot", 0, NULL);
1436 add_preferred_console("tty", 0, NULL);
1437 add_preferred_console("hvc", 0, NULL);
1439 x86_init.pci.arch_init = pci_xen_init;
1441 const struct dom0_vga_console_info *info =
1442 (void *)((char *)xen_start_info +
1443 xen_start_info->console.dom0.info_off);
1445 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1446 xen_start_info->console.domU.mfn = 0;
1447 xen_start_info->console.domU.evtchn = 0;
1449 /* Make sure ACS will be enabled */
1453 /* PCI BIOS service won't work from a PV guest. */
1454 pci_probe &= ~PCI_PROBE_BIOS;
1456 xen_raw_console_write("about to get started...\n");
1458 xen_setup_runstate_info(0);
1460 /* Start the world */
1461 #ifdef CONFIG_X86_32
1462 i386_start_kernel();
1464 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1468 static int init_hvm_pv_info(int *major, int *minor)
1470 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1473 base = xen_cpuid_base();
1474 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1477 *minor = eax & 0xffff;
1478 printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1480 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1482 pfn = __pa(hypercall_page);
1483 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1485 xen_setup_features();
1487 pv_info.name = "Xen HVM";
1489 xen_domain_type = XEN_HVM_DOMAIN;
1494 void __ref xen_hvm_init_shared_info(void)
1497 struct xen_add_to_physmap xatp;
1498 static struct shared_info *shared_info_page = 0;
1500 if (!shared_info_page)
1501 shared_info_page = (struct shared_info *)
1502 extend_brk(PAGE_SIZE, PAGE_SIZE);
1503 xatp.domid = DOMID_SELF;
1505 xatp.space = XENMAPSPACE_shared_info;
1506 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1507 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1510 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1512 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1513 * page, we use it in the event channel upcall and in some pvclock
1514 * related functions. We don't need the vcpu_info placement
1515 * optimizations because we don't use any pv_mmu or pv_irq op on
1517 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1518 * online but xen_hvm_init_shared_info is run at resume time too and
1519 * in that case multiple vcpus might be online. */
1520 for_each_online_cpu(cpu) {
1521 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1525 #ifdef CONFIG_XEN_PVHVM
1526 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1527 unsigned long action, void *hcpu)
1529 int cpu = (long)hcpu;
1531 case CPU_UP_PREPARE:
1532 xen_vcpu_setup(cpu);
1533 if (xen_have_vector_callback) {
1534 xen_init_lock_cpu(cpu);
1535 if (xen_feature(XENFEAT_hvm_safe_pvclock))
1536 xen_setup_timer(cpu);
1545 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1546 .notifier_call = xen_hvm_cpu_notify,
1549 static void __init xen_hvm_guest_init(void)
1554 r = init_hvm_pv_info(&major, &minor);
1558 xen_hvm_init_shared_info();
1560 if (xen_feature(XENFEAT_hvm_callback_vector))
1561 xen_have_vector_callback = 1;
1563 register_cpu_notifier(&xen_hvm_cpu_notifier);
1564 xen_unplug_emulated_devices();
1565 x86_init.irqs.intr_init = xen_init_IRQ;
1566 xen_hvm_init_time_ops();
1567 xen_hvm_init_mmu_ops();
1570 static bool __init xen_hvm_platform(void)
1572 if (xen_pv_domain())
1575 if (!xen_cpuid_base())
1581 bool xen_hvm_need_lapic(void)
1583 if (xen_pv_domain())
1585 if (!xen_hvm_domain())
1587 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1591 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1593 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1595 .detect = xen_hvm_platform,
1596 .init_platform = xen_hvm_guest_init,
1598 EXPORT_SYMBOL(x86_hyper_xen_hvm);