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/events.h>
37 #include <xen/interface/xen.h>
38 #include <xen/interface/version.h>
39 #include <xen/interface/physdev.h>
40 #include <xen/interface/vcpu.h>
41 #include <xen/interface/memory.h>
42 #include <xen/interface/xen-mca.h>
43 #include <xen/features.h>
46 #include <xen/hvc-console.h>
49 #include <asm/paravirt.h>
52 #include <asm/xen/pci.h>
53 #include <asm/xen/hypercall.h>
54 #include <asm/xen/hypervisor.h>
55 #include <asm/fixmap.h>
56 #include <asm/processor.h>
57 #include <asm/proto.h>
58 #include <asm/msr-index.h>
59 #include <asm/traps.h>
60 #include <asm/setup.h>
62 #include <asm/pgalloc.h>
63 #include <asm/pgtable.h>
64 #include <asm/tlbflush.h>
65 #include <asm/reboot.h>
66 #include <asm/stackprotector.h>
67 #include <asm/hypervisor.h>
68 #include <asm/mwait.h>
69 #include <asm/pci_x86.h>
73 #include <linux/acpi.h>
75 #include <acpi/pdc_intel.h>
76 #include <acpi/processor.h>
77 #include <xen/interface/platform.h>
83 #include "multicalls.h"
85 EXPORT_SYMBOL_GPL(hypercall_page);
87 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
88 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
90 enum xen_domain_type xen_domain_type = XEN_NATIVE;
91 EXPORT_SYMBOL_GPL(xen_domain_type);
93 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
94 EXPORT_SYMBOL(machine_to_phys_mapping);
95 unsigned long machine_to_phys_nr;
96 EXPORT_SYMBOL(machine_to_phys_nr);
98 struct start_info *xen_start_info;
99 EXPORT_SYMBOL_GPL(xen_start_info);
101 struct shared_info xen_dummy_shared_info;
103 void *xen_initial_gdt;
105 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
106 __read_mostly int xen_have_vector_callback;
107 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
110 * Point at some empty memory to start with. We map the real shared_info
111 * page as soon as fixmap is up and running.
113 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
116 * Flag to determine whether vcpu info placement is available on all
117 * VCPUs. We assume it is to start with, and then set it to zero on
118 * the first failure. This is because it can succeed on some VCPUs
119 * and not others, since it can involve hypervisor memory allocation,
120 * or because the guest failed to guarantee all the appropriate
121 * constraints on all VCPUs (ie buffer can't cross a page boundary).
123 * Note that any particular CPU may be using a placed vcpu structure,
124 * but we can only optimise if the all are.
126 * 0: not available, 1: available
128 static int have_vcpu_info_placement = 1;
131 struct desc_struct desc[3];
135 * Updating the 3 TLS descriptors in the GDT on every task switch is
136 * surprisingly expensive so we avoid updating them if they haven't
137 * changed. Since Xen writes different descriptors than the one
138 * passed in the update_descriptor hypercall we keep shadow copies to
141 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
143 static void clamp_max_cpus(void)
146 if (setup_max_cpus > MAX_VIRT_CPUS)
147 setup_max_cpus = MAX_VIRT_CPUS;
151 static void xen_vcpu_setup(int cpu)
153 struct vcpu_register_vcpu_info info;
155 struct vcpu_info *vcpup;
157 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
159 if (cpu < MAX_VIRT_CPUS)
160 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
162 if (!have_vcpu_info_placement) {
163 if (cpu >= MAX_VIRT_CPUS)
168 vcpup = &per_cpu(xen_vcpu_info, cpu);
169 info.mfn = arbitrary_virt_to_mfn(vcpup);
170 info.offset = offset_in_page(vcpup);
172 /* Check to see if the hypervisor will put the vcpu_info
173 structure where we want it, which allows direct access via
174 a percpu-variable. */
175 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
178 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
179 have_vcpu_info_placement = 0;
182 /* This cpu is using the registered vcpu info, even if
183 later ones fail to. */
184 per_cpu(xen_vcpu, cpu) = vcpup;
189 * On restore, set the vcpu placement up again.
190 * If it fails, then we're in a bad state, since
191 * we can't back out from using it...
193 void xen_vcpu_restore(void)
197 for_each_possible_cpu(cpu) {
198 bool other_cpu = (cpu != smp_processor_id());
199 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL);
201 if (other_cpu && is_up &&
202 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
205 xen_setup_runstate_info(cpu);
207 if (have_vcpu_info_placement)
210 if (other_cpu && is_up &&
211 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
216 static void __init xen_banner(void)
218 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
219 struct xen_extraversion extra;
220 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
222 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
224 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
225 version >> 16, version & 0xffff, extra.extraversion,
226 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
228 /* Check if running on Xen version (major, minor) or later */
230 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
232 unsigned int version;
237 version = HYPERVISOR_xen_version(XENVER_version, NULL);
238 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
239 ((version >> 16) > major))
244 #define CPUID_THERM_POWER_LEAF 6
245 #define APERFMPERF_PRESENT 0
247 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
248 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
250 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
251 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
252 static __read_mostly unsigned int cpuid_leaf5_edx_val;
254 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
255 unsigned int *cx, unsigned int *dx)
257 unsigned maskebx = ~0;
258 unsigned maskecx = ~0;
259 unsigned maskedx = ~0;
262 * Mask out inconvenient features, to try and disable as many
263 * unsupported kernel subsystems as possible.
267 maskecx = cpuid_leaf1_ecx_mask;
268 setecx = cpuid_leaf1_ecx_set_mask;
269 maskedx = cpuid_leaf1_edx_mask;
272 case CPUID_MWAIT_LEAF:
273 /* Synthesize the values.. */
276 *cx = cpuid_leaf5_ecx_val;
277 *dx = cpuid_leaf5_edx_val;
280 case CPUID_THERM_POWER_LEAF:
281 /* Disabling APERFMPERF for kernel usage */
282 maskecx = ~(1 << APERFMPERF_PRESENT);
286 /* Suppress extended topology stuff */
291 asm(XEN_EMULATE_PREFIX "cpuid"
296 : "0" (*ax), "2" (*cx));
305 static bool __init xen_check_mwait(void)
308 struct xen_platform_op op = {
309 .cmd = XENPF_set_processor_pminfo,
310 .u.set_pminfo.id = -1,
311 .u.set_pminfo.type = XEN_PM_PDC,
314 unsigned int ax, bx, cx, dx;
315 unsigned int mwait_mask;
317 /* We need to determine whether it is OK to expose the MWAIT
318 * capability to the kernel to harvest deeper than C3 states from ACPI
319 * _CST using the processor_harvest_xen.c module. For this to work, we
320 * need to gather the MWAIT_LEAF values (which the cstate.c code
321 * checks against). The hypervisor won't expose the MWAIT flag because
322 * it would break backwards compatibility; so we will find out directly
323 * from the hardware and hypercall.
325 if (!xen_initial_domain())
329 * When running under platform earlier than Xen4.2, do not expose
330 * mwait, to avoid the risk of loading native acpi pad driver
332 if (!xen_running_on_version_or_later(4, 2))
338 native_cpuid(&ax, &bx, &cx, &dx);
340 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
341 (1 << (X86_FEATURE_MWAIT % 32));
343 if ((cx & mwait_mask) != mwait_mask)
346 /* We need to emulate the MWAIT_LEAF and for that we need both
347 * ecx and edx. The hypercall provides only partial information.
350 ax = CPUID_MWAIT_LEAF;
355 native_cpuid(&ax, &bx, &cx, &dx);
357 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
358 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
360 buf[0] = ACPI_PDC_REVISION_ID;
362 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
364 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
366 if ((HYPERVISOR_dom0_op(&op) == 0) &&
367 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
368 cpuid_leaf5_ecx_val = cx;
369 cpuid_leaf5_edx_val = dx;
376 static void __init xen_init_cpuid_mask(void)
378 unsigned int ax, bx, cx, dx;
379 unsigned int xsave_mask;
381 cpuid_leaf1_edx_mask =
382 ~((1 << X86_FEATURE_MTRR) | /* disable MTRR */
383 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
385 if (!xen_initial_domain())
386 cpuid_leaf1_edx_mask &=
387 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
388 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
391 xen_cpuid(&ax, &bx, &cx, &dx);
394 (1 << (X86_FEATURE_XSAVE % 32)) |
395 (1 << (X86_FEATURE_OSXSAVE % 32));
397 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
398 if ((cx & xsave_mask) != xsave_mask)
399 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
400 if (xen_check_mwait())
401 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
404 static void xen_set_debugreg(int reg, unsigned long val)
406 HYPERVISOR_set_debugreg(reg, val);
409 static unsigned long xen_get_debugreg(int reg)
411 return HYPERVISOR_get_debugreg(reg);
414 static void xen_end_context_switch(struct task_struct *next)
417 paravirt_end_context_switch(next);
420 static unsigned long xen_store_tr(void)
426 * Set the page permissions for a particular virtual address. If the
427 * address is a vmalloc mapping (or other non-linear mapping), then
428 * find the linear mapping of the page and also set its protections to
431 static void set_aliased_prot(void *v, pgprot_t prot)
439 ptep = lookup_address((unsigned long)v, &level);
440 BUG_ON(ptep == NULL);
442 pfn = pte_pfn(*ptep);
443 page = pfn_to_page(pfn);
445 pte = pfn_pte(pfn, prot);
447 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
450 if (!PageHighMem(page)) {
451 void *av = __va(PFN_PHYS(pfn));
454 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
460 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
462 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
465 for(i = 0; i < entries; i += entries_per_page)
466 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
469 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
471 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
474 for(i = 0; i < entries; i += entries_per_page)
475 set_aliased_prot(ldt + i, PAGE_KERNEL);
478 static void xen_set_ldt(const void *addr, unsigned entries)
480 struct mmuext_op *op;
481 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
483 trace_xen_cpu_set_ldt(addr, entries);
486 op->cmd = MMUEXT_SET_LDT;
487 op->arg1.linear_addr = (unsigned long)addr;
488 op->arg2.nr_ents = entries;
490 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
492 xen_mc_issue(PARAVIRT_LAZY_CPU);
495 static void xen_load_gdt(const struct desc_ptr *dtr)
497 unsigned long va = dtr->address;
498 unsigned int size = dtr->size + 1;
499 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
500 unsigned long frames[pages];
504 * A GDT can be up to 64k in size, which corresponds to 8192
505 * 8-byte entries, or 16 4k pages..
508 BUG_ON(size > 65536);
509 BUG_ON(va & ~PAGE_MASK);
511 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
514 unsigned long pfn, mfn;
518 * The GDT is per-cpu and is in the percpu data area.
519 * That can be virtually mapped, so we need to do a
520 * page-walk to get the underlying MFN for the
521 * hypercall. The page can also be in the kernel's
522 * linear range, so we need to RO that mapping too.
524 ptep = lookup_address(va, &level);
525 BUG_ON(ptep == NULL);
527 pfn = pte_pfn(*ptep);
528 mfn = pfn_to_mfn(pfn);
529 virt = __va(PFN_PHYS(pfn));
533 make_lowmem_page_readonly((void *)va);
534 make_lowmem_page_readonly(virt);
537 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
542 * load_gdt for early boot, when the gdt is only mapped once
544 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
546 unsigned long va = dtr->address;
547 unsigned int size = dtr->size + 1;
548 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
549 unsigned long frames[pages];
553 * A GDT can be up to 64k in size, which corresponds to 8192
554 * 8-byte entries, or 16 4k pages..
557 BUG_ON(size > 65536);
558 BUG_ON(va & ~PAGE_MASK);
560 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
562 unsigned long pfn, mfn;
564 pfn = virt_to_pfn(va);
565 mfn = pfn_to_mfn(pfn);
567 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
569 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
575 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
579 static inline bool desc_equal(const struct desc_struct *d1,
580 const struct desc_struct *d2)
582 return d1->a == d2->a && d1->b == d2->b;
585 static void load_TLS_descriptor(struct thread_struct *t,
586 unsigned int cpu, unsigned int i)
588 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
589 struct desc_struct *gdt;
591 struct multicall_space mc;
593 if (desc_equal(shadow, &t->tls_array[i]))
596 *shadow = t->tls_array[i];
598 gdt = get_cpu_gdt_table(cpu);
599 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
600 mc = __xen_mc_entry(0);
602 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
605 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
608 * XXX sleazy hack: If we're being called in a lazy-cpu zone
609 * and lazy gs handling is enabled, it means we're in a
610 * context switch, and %gs has just been saved. This means we
611 * can zero it out to prevent faults on exit from the
612 * hypervisor if the next process has no %gs. Either way, it
613 * has been saved, and the new value will get loaded properly.
614 * This will go away as soon as Xen has been modified to not
615 * save/restore %gs for normal hypercalls.
617 * On x86_64, this hack is not used for %gs, because gs points
618 * to KERNEL_GS_BASE (and uses it for PDA references), so we
619 * must not zero %gs on x86_64
621 * For x86_64, we need to zero %fs, otherwise we may get an
622 * exception between the new %fs descriptor being loaded and
623 * %fs being effectively cleared at __switch_to().
625 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
635 load_TLS_descriptor(t, cpu, 0);
636 load_TLS_descriptor(t, cpu, 1);
637 load_TLS_descriptor(t, cpu, 2);
639 xen_mc_issue(PARAVIRT_LAZY_CPU);
643 static void xen_load_gs_index(unsigned int idx)
645 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
650 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
653 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
654 u64 entry = *(u64 *)ptr;
656 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
661 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
667 static int cvt_gate_to_trap(int vector, const gate_desc *val,
668 struct trap_info *info)
672 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
675 info->vector = vector;
677 addr = gate_offset(*val);
680 * Look for known traps using IST, and substitute them
681 * appropriately. The debugger ones are the only ones we care
682 * about. Xen will handle faults like double_fault,
683 * so we should never see them. Warn if
684 * there's an unexpected IST-using fault handler.
686 if (addr == (unsigned long)debug)
687 addr = (unsigned long)xen_debug;
688 else if (addr == (unsigned long)int3)
689 addr = (unsigned long)xen_int3;
690 else if (addr == (unsigned long)stack_segment)
691 addr = (unsigned long)xen_stack_segment;
692 else if (addr == (unsigned long)double_fault ||
693 addr == (unsigned long)nmi) {
694 /* Don't need to handle these */
696 #ifdef CONFIG_X86_MCE
697 } else if (addr == (unsigned long)machine_check) {
699 * when xen hypervisor inject vMCE to guest,
700 * use native mce handler to handle it
705 /* Some other trap using IST? */
706 if (WARN_ON(val->ist != 0))
709 #endif /* CONFIG_X86_64 */
710 info->address = addr;
712 info->cs = gate_segment(*val);
713 info->flags = val->dpl;
714 /* interrupt gates clear IF */
715 if (val->type == GATE_INTERRUPT)
716 info->flags |= 1 << 2;
721 /* Locations of each CPU's IDT */
722 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
724 /* Set an IDT entry. If the entry is part of the current IDT, then
726 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
728 unsigned long p = (unsigned long)&dt[entrynum];
729 unsigned long start, end;
731 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
735 start = __this_cpu_read(idt_desc.address);
736 end = start + __this_cpu_read(idt_desc.size) + 1;
740 native_write_idt_entry(dt, entrynum, g);
742 if (p >= start && (p + 8) <= end) {
743 struct trap_info info[2];
747 if (cvt_gate_to_trap(entrynum, g, &info[0]))
748 if (HYPERVISOR_set_trap_table(info))
755 static void xen_convert_trap_info(const struct desc_ptr *desc,
756 struct trap_info *traps)
758 unsigned in, out, count;
760 count = (desc->size+1) / sizeof(gate_desc);
763 for (in = out = 0; in < count; in++) {
764 gate_desc *entry = (gate_desc*)(desc->address) + in;
766 if (cvt_gate_to_trap(in, entry, &traps[out]))
769 traps[out].address = 0;
772 void xen_copy_trap_info(struct trap_info *traps)
774 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
776 xen_convert_trap_info(desc, traps);
779 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
780 hold a spinlock to protect the static traps[] array (static because
781 it avoids allocation, and saves stack space). */
782 static void xen_load_idt(const struct desc_ptr *desc)
784 static DEFINE_SPINLOCK(lock);
785 static struct trap_info traps[257];
787 trace_xen_cpu_load_idt(desc);
791 __get_cpu_var(idt_desc) = *desc;
793 xen_convert_trap_info(desc, traps);
796 if (HYPERVISOR_set_trap_table(traps))
802 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
803 they're handled differently. */
804 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
805 const void *desc, int type)
807 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
818 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
821 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
831 * Version of write_gdt_entry for use at early boot-time needed to
832 * update an entry as simply as possible.
834 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
835 const void *desc, int type)
837 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
846 xmaddr_t maddr = virt_to_machine(&dt[entry]);
848 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
849 dt[entry] = *(struct desc_struct *)desc;
855 static void xen_load_sp0(struct tss_struct *tss,
856 struct thread_struct *thread)
858 struct multicall_space mcs;
860 mcs = xen_mc_entry(0);
861 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
862 xen_mc_issue(PARAVIRT_LAZY_CPU);
865 static void xen_set_iopl_mask(unsigned mask)
867 struct physdev_set_iopl set_iopl;
869 /* Force the change at ring 0. */
870 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
871 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
874 static void xen_io_delay(void)
878 #ifdef CONFIG_X86_LOCAL_APIC
879 static unsigned long xen_set_apic_id(unsigned int x)
884 static unsigned int xen_get_apic_id(unsigned long x)
886 return ((x)>>24) & 0xFFu;
888 static u32 xen_apic_read(u32 reg)
890 struct xen_platform_op op = {
891 .cmd = XENPF_get_cpuinfo,
892 .interface_version = XENPF_INTERFACE_VERSION,
893 .u.pcpu_info.xen_cpuid = 0,
897 /* Shouldn't need this as APIC is turned off for PV, and we only
898 * get called on the bootup processor. But just in case. */
899 if (!xen_initial_domain() || smp_processor_id())
908 ret = HYPERVISOR_dom0_op(&op);
912 return op.u.pcpu_info.apic_id << 24;
915 static void xen_apic_write(u32 reg, u32 val)
917 /* Warn to see if there's any stray references */
921 static u64 xen_apic_icr_read(void)
926 static void xen_apic_icr_write(u32 low, u32 id)
928 /* Warn to see if there's any stray references */
932 static void xen_apic_wait_icr_idle(void)
937 static u32 xen_safe_apic_wait_icr_idle(void)
942 static void set_xen_basic_apic_ops(void)
944 apic->read = xen_apic_read;
945 apic->write = xen_apic_write;
946 apic->icr_read = xen_apic_icr_read;
947 apic->icr_write = xen_apic_icr_write;
948 apic->wait_icr_idle = xen_apic_wait_icr_idle;
949 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
950 apic->set_apic_id = xen_set_apic_id;
951 apic->get_apic_id = xen_get_apic_id;
954 apic->send_IPI_allbutself = xen_send_IPI_allbutself;
955 apic->send_IPI_mask_allbutself = xen_send_IPI_mask_allbutself;
956 apic->send_IPI_mask = xen_send_IPI_mask;
957 apic->send_IPI_all = xen_send_IPI_all;
958 apic->send_IPI_self = xen_send_IPI_self;
964 static void xen_clts(void)
966 struct multicall_space mcs;
968 mcs = xen_mc_entry(0);
970 MULTI_fpu_taskswitch(mcs.mc, 0);
972 xen_mc_issue(PARAVIRT_LAZY_CPU);
975 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
977 static unsigned long xen_read_cr0(void)
979 unsigned long cr0 = this_cpu_read(xen_cr0_value);
981 if (unlikely(cr0 == 0)) {
982 cr0 = native_read_cr0();
983 this_cpu_write(xen_cr0_value, cr0);
989 static void xen_write_cr0(unsigned long cr0)
991 struct multicall_space mcs;
993 this_cpu_write(xen_cr0_value, cr0);
995 /* Only pay attention to cr0.TS; everything else is
997 mcs = xen_mc_entry(0);
999 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1001 xen_mc_issue(PARAVIRT_LAZY_CPU);
1004 static void xen_write_cr4(unsigned long cr4)
1006 cr4 &= ~X86_CR4_PGE;
1007 cr4 &= ~X86_CR4_PSE;
1009 native_write_cr4(cr4);
1011 #ifdef CONFIG_X86_64
1012 static inline unsigned long xen_read_cr8(void)
1016 static inline void xen_write_cr8(unsigned long val)
1021 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1028 #ifdef CONFIG_X86_64
1032 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
1033 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
1034 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
1037 base = ((u64)high << 32) | low;
1038 if (HYPERVISOR_set_segment_base(which, base) != 0)
1046 case MSR_SYSCALL_MASK:
1047 case MSR_IA32_SYSENTER_CS:
1048 case MSR_IA32_SYSENTER_ESP:
1049 case MSR_IA32_SYSENTER_EIP:
1050 /* Fast syscall setup is all done in hypercalls, so
1051 these are all ignored. Stub them out here to stop
1052 Xen console noise. */
1055 case MSR_IA32_CR_PAT:
1056 if (smp_processor_id() == 0)
1057 xen_set_pat(((u64)high << 32) | low);
1061 ret = native_write_msr_safe(msr, low, high);
1067 void xen_setup_shared_info(void)
1069 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1070 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1071 xen_start_info->shared_info);
1073 HYPERVISOR_shared_info =
1074 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1076 HYPERVISOR_shared_info =
1077 (struct shared_info *)__va(xen_start_info->shared_info);
1080 /* In UP this is as good a place as any to set up shared info */
1081 xen_setup_vcpu_info_placement();
1084 xen_setup_mfn_list_list();
1087 /* This is called once we have the cpu_possible_mask */
1088 void xen_setup_vcpu_info_placement(void)
1092 for_each_possible_cpu(cpu)
1093 xen_vcpu_setup(cpu);
1095 /* xen_vcpu_setup managed to place the vcpu_info within the
1096 percpu area for all cpus, so make use of it */
1097 if (have_vcpu_info_placement) {
1098 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1099 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1100 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1101 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1102 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1106 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1107 unsigned long addr, unsigned len)
1109 char *start, *end, *reloc;
1112 start = end = reloc = NULL;
1114 #define SITE(op, x) \
1115 case PARAVIRT_PATCH(op.x): \
1116 if (have_vcpu_info_placement) { \
1117 start = (char *)xen_##x##_direct; \
1118 end = xen_##x##_direct_end; \
1119 reloc = xen_##x##_direct_reloc; \
1124 SITE(pv_irq_ops, irq_enable);
1125 SITE(pv_irq_ops, irq_disable);
1126 SITE(pv_irq_ops, save_fl);
1127 SITE(pv_irq_ops, restore_fl);
1131 if (start == NULL || (end-start) > len)
1134 ret = paravirt_patch_insns(insnbuf, len, start, end);
1136 /* Note: because reloc is assigned from something that
1137 appears to be an array, gcc assumes it's non-null,
1138 but doesn't know its relationship with start and
1140 if (reloc > start && reloc < end) {
1141 int reloc_off = reloc - start;
1142 long *relocp = (long *)(insnbuf + reloc_off);
1143 long delta = start - (char *)addr;
1151 ret = paravirt_patch_default(type, clobbers, insnbuf,
1159 static const struct pv_info xen_info __initconst = {
1160 .paravirt_enabled = 1,
1161 .shared_kernel_pmd = 0,
1163 #ifdef CONFIG_X86_64
1164 .extra_user_64bit_cs = FLAT_USER_CS64,
1170 static const struct pv_init_ops xen_init_ops __initconst = {
1174 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1177 .set_debugreg = xen_set_debugreg,
1178 .get_debugreg = xen_get_debugreg,
1182 .read_cr0 = xen_read_cr0,
1183 .write_cr0 = xen_write_cr0,
1185 .read_cr4 = native_read_cr4,
1186 .read_cr4_safe = native_read_cr4_safe,
1187 .write_cr4 = xen_write_cr4,
1189 #ifdef CONFIG_X86_64
1190 .read_cr8 = xen_read_cr8,
1191 .write_cr8 = xen_write_cr8,
1194 .wbinvd = native_wbinvd,
1196 .read_msr = native_read_msr_safe,
1197 .write_msr = xen_write_msr_safe,
1199 .read_tsc = native_read_tsc,
1200 .read_pmc = native_read_pmc,
1202 .read_tscp = native_read_tscp,
1205 .irq_enable_sysexit = xen_sysexit,
1206 #ifdef CONFIG_X86_64
1207 .usergs_sysret32 = xen_sysret32,
1208 .usergs_sysret64 = xen_sysret64,
1211 .load_tr_desc = paravirt_nop,
1212 .set_ldt = xen_set_ldt,
1213 .load_gdt = xen_load_gdt,
1214 .load_idt = xen_load_idt,
1215 .load_tls = xen_load_tls,
1216 #ifdef CONFIG_X86_64
1217 .load_gs_index = xen_load_gs_index,
1220 .alloc_ldt = xen_alloc_ldt,
1221 .free_ldt = xen_free_ldt,
1223 .store_gdt = native_store_gdt,
1224 .store_idt = native_store_idt,
1225 .store_tr = xen_store_tr,
1227 .write_ldt_entry = xen_write_ldt_entry,
1228 .write_gdt_entry = xen_write_gdt_entry,
1229 .write_idt_entry = xen_write_idt_entry,
1230 .load_sp0 = xen_load_sp0,
1232 .set_iopl_mask = xen_set_iopl_mask,
1233 .io_delay = xen_io_delay,
1235 /* Xen takes care of %gs when switching to usermode for us */
1236 .swapgs = paravirt_nop,
1238 .start_context_switch = paravirt_start_context_switch,
1239 .end_context_switch = xen_end_context_switch,
1242 static const struct pv_apic_ops xen_apic_ops __initconst = {
1243 #ifdef CONFIG_X86_LOCAL_APIC
1244 .startup_ipi_hook = paravirt_nop,
1248 static void xen_reboot(int reason)
1250 struct sched_shutdown r = { .reason = reason };
1252 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1256 static void xen_restart(char *msg)
1258 xen_reboot(SHUTDOWN_reboot);
1261 static void xen_emergency_restart(void)
1263 xen_reboot(SHUTDOWN_reboot);
1266 static void xen_machine_halt(void)
1268 xen_reboot(SHUTDOWN_poweroff);
1271 static void xen_machine_power_off(void)
1275 xen_reboot(SHUTDOWN_poweroff);
1278 static void xen_crash_shutdown(struct pt_regs *regs)
1280 xen_reboot(SHUTDOWN_crash);
1284 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1286 xen_reboot(SHUTDOWN_crash);
1290 static struct notifier_block xen_panic_block = {
1291 .notifier_call= xen_panic_event,
1294 int xen_panic_handler_init(void)
1296 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1300 static const struct machine_ops xen_machine_ops __initconst = {
1301 .restart = xen_restart,
1302 .halt = xen_machine_halt,
1303 .power_off = xen_machine_power_off,
1304 .shutdown = xen_machine_halt,
1305 .crash_shutdown = xen_crash_shutdown,
1306 .emergency_restart = xen_emergency_restart,
1310 * Set up the GDT and segment registers for -fstack-protector. Until
1311 * we do this, we have to be careful not to call any stack-protected
1312 * function, which is most of the kernel.
1314 static void __init xen_setup_stackprotector(void)
1316 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1317 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1319 setup_stack_canary_segment(0);
1320 switch_to_new_gdt(0);
1322 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1323 pv_cpu_ops.load_gdt = xen_load_gdt;
1326 /* First C function to be called on Xen boot */
1327 asmlinkage void __init xen_start_kernel(void)
1329 struct physdev_set_iopl set_iopl;
1332 if (!xen_start_info)
1335 xen_domain_type = XEN_PV_DOMAIN;
1337 xen_setup_machphys_mapping();
1339 /* Install Xen paravirt ops */
1341 pv_init_ops = xen_init_ops;
1342 pv_cpu_ops = xen_cpu_ops;
1343 pv_apic_ops = xen_apic_ops;
1345 x86_init.resources.memory_setup = xen_memory_setup;
1346 x86_init.oem.arch_setup = xen_arch_setup;
1347 x86_init.oem.banner = xen_banner;
1349 xen_init_time_ops();
1352 * Set up some pagetable state before starting to set any ptes.
1357 /* Prevent unwanted bits from being set in PTEs. */
1358 __supported_pte_mask &= ~_PAGE_GLOBAL;
1360 if (!xen_initial_domain())
1362 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1364 __supported_pte_mask |= _PAGE_IOMAP;
1367 * Prevent page tables from being allocated in highmem, even
1368 * if CONFIG_HIGHPTE is enabled.
1370 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1372 /* Work out if we support NX */
1375 xen_setup_features();
1378 if (!xen_feature(XENFEAT_auto_translated_physmap))
1379 xen_build_dynamic_phys_to_machine();
1382 * Set up kernel GDT and segment registers, mainly so that
1383 * -fstack-protector code can be executed.
1385 xen_setup_stackprotector();
1388 xen_init_cpuid_mask();
1390 #ifdef CONFIG_X86_LOCAL_APIC
1392 * set up the basic apic ops.
1394 set_xen_basic_apic_ops();
1397 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1398 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1399 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1402 machine_ops = xen_machine_ops;
1405 * The only reliable way to retain the initial address of the
1406 * percpu gdt_page is to remember it here, so we can go and
1407 * mark it RW later, when the initial percpu area is freed.
1409 xen_initial_gdt = &per_cpu(gdt_page, 0);
1413 #ifdef CONFIG_ACPI_NUMA
1415 * The pages we from Xen are not related to machine pages, so
1416 * any NUMA information the kernel tries to get from ACPI will
1417 * be meaningless. Prevent it from trying.
1421 #ifdef CONFIG_X86_PAT
1423 * For right now disable the PAT. We should remove this once
1424 * git commit 8eaffa67b43e99ae581622c5133e20b0f48bcef1
1425 * (xen/pat: Disable PAT support for now) is reverted.
1429 /* Don't do the full vcpu_info placement stuff until we have a
1430 possible map and a non-dummy shared_info. */
1431 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1433 local_irq_disable();
1434 early_boot_irqs_disabled = true;
1436 xen_raw_console_write("mapping kernel into physical memory\n");
1437 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base, xen_start_info->nr_pages);
1439 /* Allocate and initialize top and mid mfn levels for p2m structure */
1440 xen_build_mfn_list_list();
1442 /* keep using Xen gdt for now; no urgent need to change it */
1444 #ifdef CONFIG_X86_32
1445 pv_info.kernel_rpl = 1;
1446 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1447 pv_info.kernel_rpl = 0;
1449 pv_info.kernel_rpl = 0;
1451 /* set the limit of our address space */
1454 /* We used to do this in xen_arch_setup, but that is too late on AMD
1455 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1456 * which pokes 0xcf8 port.
1459 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1461 xen_raw_printk("physdev_op failed %d\n", rc);
1463 #ifdef CONFIG_X86_32
1464 /* set up basic CPUID stuff */
1465 cpu_detect(&new_cpu_data);
1466 new_cpu_data.hard_math = 1;
1467 new_cpu_data.wp_works_ok = 1;
1468 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1471 /* Poke various useful things into boot_params */
1472 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1473 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1474 ? __pa(xen_start_info->mod_start) : 0;
1475 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1476 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1478 if (!xen_initial_domain()) {
1479 add_preferred_console("xenboot", 0, NULL);
1480 add_preferred_console("tty", 0, NULL);
1481 add_preferred_console("hvc", 0, NULL);
1483 x86_init.pci.arch_init = pci_xen_init;
1485 const struct dom0_vga_console_info *info =
1486 (void *)((char *)xen_start_info +
1487 xen_start_info->console.dom0.info_off);
1488 struct xen_platform_op op = {
1489 .cmd = XENPF_firmware_info,
1490 .interface_version = XENPF_INTERFACE_VERSION,
1491 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1494 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1495 xen_start_info->console.domU.mfn = 0;
1496 xen_start_info->console.domU.evtchn = 0;
1498 if (HYPERVISOR_dom0_op(&op) == 0)
1499 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1503 /* Make sure ACS will be enabled */
1506 xen_acpi_sleep_register();
1508 /* Avoid searching for BIOS MP tables */
1509 x86_init.mpparse.find_smp_config = x86_init_noop;
1510 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1513 /* PCI BIOS service won't work from a PV guest. */
1514 pci_probe &= ~PCI_PROBE_BIOS;
1516 xen_raw_console_write("about to get started...\n");
1518 xen_setup_runstate_info(0);
1520 /* Start the world */
1521 #ifdef CONFIG_X86_32
1522 i386_start_kernel();
1524 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1528 void __ref xen_hvm_init_shared_info(void)
1531 struct xen_add_to_physmap xatp;
1532 static struct shared_info *shared_info_page = 0;
1534 if (!shared_info_page)
1535 shared_info_page = (struct shared_info *)
1536 extend_brk(PAGE_SIZE, PAGE_SIZE);
1537 xatp.domid = DOMID_SELF;
1539 xatp.space = XENMAPSPACE_shared_info;
1540 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1541 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1544 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1546 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1547 * page, we use it in the event channel upcall and in some pvclock
1548 * related functions. We don't need the vcpu_info placement
1549 * optimizations because we don't use any pv_mmu or pv_irq op on
1551 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1552 * online but xen_hvm_init_shared_info is run at resume time too and
1553 * in that case multiple vcpus might be online. */
1554 for_each_online_cpu(cpu) {
1555 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1559 #ifdef CONFIG_XEN_PVHVM
1560 static void __init init_hvm_pv_info(void)
1563 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1566 base = xen_cpuid_base();
1567 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1570 minor = eax & 0xffff;
1571 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1573 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1575 pfn = __pa(hypercall_page);
1576 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1578 xen_setup_features();
1580 pv_info.name = "Xen HVM";
1582 xen_domain_type = XEN_HVM_DOMAIN;
1585 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1586 unsigned long action, void *hcpu)
1588 int cpu = (long)hcpu;
1590 case CPU_UP_PREPARE:
1591 xen_vcpu_setup(cpu);
1592 if (xen_have_vector_callback)
1593 xen_init_lock_cpu(cpu);
1601 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1602 .notifier_call = xen_hvm_cpu_notify,
1605 static void __init xen_hvm_guest_init(void)
1609 xen_hvm_init_shared_info();
1611 if (xen_feature(XENFEAT_hvm_callback_vector))
1612 xen_have_vector_callback = 1;
1614 register_cpu_notifier(&xen_hvm_cpu_notifier);
1615 xen_unplug_emulated_devices();
1616 x86_init.irqs.intr_init = xen_init_IRQ;
1617 xen_hvm_init_time_ops();
1618 xen_hvm_init_mmu_ops();
1621 static bool __init xen_hvm_platform(void)
1623 if (xen_pv_domain())
1626 if (!xen_cpuid_base())
1632 bool xen_hvm_need_lapic(void)
1634 if (xen_pv_domain())
1636 if (!xen_hvm_domain())
1638 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1642 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1644 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1646 .detect = xen_hvm_platform,
1647 .init_platform = xen_hvm_guest_init,
1648 .x2apic_available = xen_x2apic_para_available,
1650 EXPORT_SYMBOL(x86_hyper_xen_hvm);