2 * Core of Xen paravirt_ops implementation.
4 * This file contains the xen_paravirt_ops structure itself, and the
6 * - privileged instructions
11 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
14 #include <linux/cpu.h>
15 #include <linux/kernel.h>
16 #include <linux/init.h>
17 #include <linux/smp.h>
18 #include <linux/preempt.h>
19 #include <linux/hardirq.h>
20 #include <linux/percpu.h>
21 #include <linux/delay.h>
22 #include <linux/start_kernel.h>
23 #include <linux/sched.h>
24 #include <linux/kprobes.h>
25 #include <linux/bootmem.h>
26 #include <linux/module.h>
28 #include <linux/page-flags.h>
29 #include <linux/highmem.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/memblock.h>
36 #include <xen/interface/xen.h>
37 #include <xen/interface/version.h>
38 #include <xen/interface/physdev.h>
39 #include <xen/interface/vcpu.h>
40 #include <xen/interface/memory.h>
41 #include <xen/features.h>
44 #include <xen/hvc-console.h>
46 #include <asm/paravirt.h>
49 #include <asm/xen/pci.h>
50 #include <asm/xen/hypercall.h>
51 #include <asm/xen/hypervisor.h>
52 #include <asm/fixmap.h>
53 #include <asm/processor.h>
54 #include <asm/proto.h>
55 #include <asm/msr-index.h>
56 #include <asm/traps.h>
57 #include <asm/setup.h>
59 #include <asm/pgalloc.h>
60 #include <asm/pgtable.h>
61 #include <asm/tlbflush.h>
62 #include <asm/reboot.h>
63 #include <asm/stackprotector.h>
64 #include <asm/hypervisor.h>
65 #include <asm/pci_x86.h>
69 #include "multicalls.h"
71 EXPORT_SYMBOL_GPL(hypercall_page);
73 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
74 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
76 enum xen_domain_type xen_domain_type = XEN_NATIVE;
77 EXPORT_SYMBOL_GPL(xen_domain_type);
79 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
80 EXPORT_SYMBOL(machine_to_phys_mapping);
81 unsigned long machine_to_phys_nr;
82 EXPORT_SYMBOL(machine_to_phys_nr);
84 struct start_info *xen_start_info;
85 EXPORT_SYMBOL_GPL(xen_start_info);
87 struct shared_info xen_dummy_shared_info;
89 void *xen_initial_gdt;
91 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
92 __read_mostly int xen_have_vector_callback;
93 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
96 * Point at some empty memory to start with. We map the real shared_info
97 * page as soon as fixmap is up and running.
99 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
102 * Flag to determine whether vcpu info placement is available on all
103 * VCPUs. We assume it is to start with, and then set it to zero on
104 * the first failure. This is because it can succeed on some VCPUs
105 * and not others, since it can involve hypervisor memory allocation,
106 * or because the guest failed to guarantee all the appropriate
107 * constraints on all VCPUs (ie buffer can't cross a page boundary).
109 * Note that any particular CPU may be using a placed vcpu structure,
110 * but we can only optimise if the all are.
112 * 0: not available, 1: available
114 static int have_vcpu_info_placement = 1;
116 static void clamp_max_cpus(void)
119 if (setup_max_cpus > MAX_VIRT_CPUS)
120 setup_max_cpus = MAX_VIRT_CPUS;
124 static void xen_vcpu_setup(int cpu)
126 struct vcpu_register_vcpu_info info;
128 struct vcpu_info *vcpup;
130 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
133 * This path is called twice on PVHVM - first during bootup via
134 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
135 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
136 * As we can only do the VCPUOP_register_vcpu_info once lets
137 * not over-write its result.
139 * For PV it is called during restore (xen_vcpu_restore) and bootup
140 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
143 if (xen_hvm_domain()) {
144 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
147 if (cpu < MAX_VIRT_CPUS)
148 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
150 if (!have_vcpu_info_placement) {
151 if (cpu >= MAX_VIRT_CPUS)
156 vcpup = &per_cpu(xen_vcpu_info, cpu);
157 info.mfn = arbitrary_virt_to_mfn(vcpup);
158 info.offset = offset_in_page(vcpup);
160 /* Check to see if the hypervisor will put the vcpu_info
161 structure where we want it, which allows direct access via
162 a percpu-variable. */
163 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
166 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
167 have_vcpu_info_placement = 0;
170 /* This cpu is using the registered vcpu info, even if
171 later ones fail to. */
172 per_cpu(xen_vcpu, cpu) = vcpup;
177 * On restore, set the vcpu placement up again.
178 * If it fails, then we're in a bad state, since
179 * we can't back out from using it...
181 void xen_vcpu_restore(void)
185 for_each_online_cpu(cpu) {
186 bool other_cpu = (cpu != smp_processor_id());
189 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
192 xen_setup_runstate_info(cpu);
194 if (have_vcpu_info_placement)
198 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
203 static void __init xen_banner(void)
205 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
206 struct xen_extraversion extra;
207 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
209 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
211 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
212 version >> 16, version & 0xffff, extra.extraversion,
213 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
216 #define CPUID_THERM_POWER_LEAF 6
217 #define APERFMPERF_PRESENT 0
219 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
220 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
222 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
223 unsigned int *cx, unsigned int *dx)
225 unsigned maskebx = ~0;
226 unsigned maskecx = ~0;
227 unsigned maskedx = ~0;
230 * Mask out inconvenient features, to try and disable as many
231 * unsupported kernel subsystems as possible.
235 maskecx = cpuid_leaf1_ecx_mask;
236 maskedx = cpuid_leaf1_edx_mask;
239 case CPUID_THERM_POWER_LEAF:
240 /* Disabling APERFMPERF for kernel usage */
241 maskecx = ~(1 << APERFMPERF_PRESENT);
245 /* Suppress extended topology stuff */
250 asm(XEN_EMULATE_PREFIX "cpuid"
255 : "0" (*ax), "2" (*cx));
262 static void __init xen_init_cpuid_mask(void)
264 unsigned int ax, bx, cx, dx;
265 unsigned int xsave_mask;
267 cpuid_leaf1_edx_mask =
268 ~((1 << X86_FEATURE_MCE) | /* disable MCE */
269 (1 << X86_FEATURE_MCA) | /* disable MCA */
270 (1 << X86_FEATURE_MTRR) | /* disable MTRR */
271 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
274 * Xen PV would need some work to support PCID: CR3 handling as well
275 * as xen_flush_tlb_others() would need updating.
277 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_PCID % 32)); /* disable PCID */
279 if (!xen_initial_domain())
280 cpuid_leaf1_edx_mask &=
281 ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
282 (1 << X86_FEATURE_ACPI)); /* disable ACPI */
285 xen_cpuid(&ax, &bx, &cx, &dx);
288 (1 << (X86_FEATURE_XSAVE % 32)) |
289 (1 << (X86_FEATURE_OSXSAVE % 32));
291 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
292 if ((cx & xsave_mask) != xsave_mask)
293 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
296 static void xen_set_debugreg(int reg, unsigned long val)
298 HYPERVISOR_set_debugreg(reg, val);
301 static unsigned long xen_get_debugreg(int reg)
303 return HYPERVISOR_get_debugreg(reg);
306 static void xen_end_context_switch(struct task_struct *next)
309 paravirt_end_context_switch(next);
312 static unsigned long xen_store_tr(void)
318 * Set the page permissions for a particular virtual address. If the
319 * address is a vmalloc mapping (or other non-linear mapping), then
320 * find the linear mapping of the page and also set its protections to
323 static void set_aliased_prot(void *v, pgprot_t prot)
332 ptep = lookup_address((unsigned long)v, &level);
333 BUG_ON(ptep == NULL);
335 pfn = pte_pfn(*ptep);
336 page = pfn_to_page(pfn);
338 pte = pfn_pte(pfn, prot);
341 * Careful: update_va_mapping() will fail if the virtual address
342 * we're poking isn't populated in the page tables. We don't
343 * need to worry about the direct map (that's always in the page
344 * tables), but we need to be careful about vmap space. In
345 * particular, the top level page table can lazily propagate
346 * entries between processes, so if we've switched mms since we
347 * vmapped the target in the first place, we might not have the
348 * top-level page table entry populated.
350 * We disable preemption because we want the same mm active when
351 * we probe the target and when we issue the hypercall. We'll
352 * have the same nominal mm, but if we're a kernel thread, lazy
353 * mm dropping could change our pgd.
355 * Out of an abundance of caution, this uses __get_user() to fault
356 * in the target address just in case there's some obscure case
357 * in which the target address isn't readable.
362 pagefault_disable(); /* Avoid warnings due to being atomic. */
363 __get_user(dummy, (unsigned char __user __force *)v);
366 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
369 if (!PageHighMem(page)) {
370 void *av = __va(PFN_PHYS(pfn));
373 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
381 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
383 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
387 * We need to mark the all aliases of the LDT pages RO. We
388 * don't need to call vm_flush_aliases(), though, since that's
389 * only responsible for flushing aliases out the TLBs, not the
390 * page tables, and Xen will flush the TLB for us if needed.
392 * To avoid confusing future readers: none of this is necessary
393 * to load the LDT. The hypervisor only checks this when the
394 * LDT is faulted in due to subsequent descriptor access.
397 for(i = 0; i < entries; i += entries_per_page)
398 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
401 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
403 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
406 for(i = 0; i < entries; i += entries_per_page)
407 set_aliased_prot(ldt + i, PAGE_KERNEL);
410 static void xen_set_ldt(const void *addr, unsigned entries)
412 struct mmuext_op *op;
413 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
415 trace_xen_cpu_set_ldt(addr, entries);
418 op->cmd = MMUEXT_SET_LDT;
419 op->arg1.linear_addr = (unsigned long)addr;
420 op->arg2.nr_ents = entries;
422 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
424 xen_mc_issue(PARAVIRT_LAZY_CPU);
427 static void xen_load_gdt(const struct desc_ptr *dtr)
429 unsigned long va = dtr->address;
430 unsigned int size = dtr->size + 1;
431 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
432 unsigned long frames[pages];
436 * A GDT can be up to 64k in size, which corresponds to 8192
437 * 8-byte entries, or 16 4k pages..
440 BUG_ON(size > 65536);
441 BUG_ON(va & ~PAGE_MASK);
443 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
446 unsigned long pfn, mfn;
450 * The GDT is per-cpu and is in the percpu data area.
451 * That can be virtually mapped, so we need to do a
452 * page-walk to get the underlying MFN for the
453 * hypercall. The page can also be in the kernel's
454 * linear range, so we need to RO that mapping too.
456 ptep = lookup_address(va, &level);
457 BUG_ON(ptep == NULL);
459 pfn = pte_pfn(*ptep);
460 mfn = pfn_to_mfn(pfn);
461 virt = __va(PFN_PHYS(pfn));
465 make_lowmem_page_readonly((void *)va);
466 make_lowmem_page_readonly(virt);
469 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
474 * load_gdt for early boot, when the gdt is only mapped once
476 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
478 unsigned long va = dtr->address;
479 unsigned int size = dtr->size + 1;
480 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
481 unsigned long frames[pages];
485 * A GDT can be up to 64k in size, which corresponds to 8192
486 * 8-byte entries, or 16 4k pages..
489 BUG_ON(size > 65536);
490 BUG_ON(va & ~PAGE_MASK);
492 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
494 unsigned long pfn, mfn;
496 pfn = virt_to_pfn(va);
497 mfn = pfn_to_mfn(pfn);
499 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
501 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
507 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
511 static void load_TLS_descriptor(struct thread_struct *t,
512 unsigned int cpu, unsigned int i)
514 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
515 xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
516 struct multicall_space mc = __xen_mc_entry(0);
518 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
521 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
524 * XXX sleazy hack: If we're being called in a lazy-cpu zone
525 * and lazy gs handling is enabled, it means we're in a
526 * context switch, and %gs has just been saved. This means we
527 * can zero it out to prevent faults on exit from the
528 * hypervisor if the next process has no %gs. Either way, it
529 * has been saved, and the new value will get loaded properly.
530 * This will go away as soon as Xen has been modified to not
531 * save/restore %gs for normal hypercalls.
533 * On x86_64, this hack is not used for %gs, because gs points
534 * to KERNEL_GS_BASE (and uses it for PDA references), so we
535 * must not zero %gs on x86_64
537 * For x86_64, we need to zero %fs, otherwise we may get an
538 * exception between the new %fs descriptor being loaded and
539 * %fs being effectively cleared at __switch_to().
541 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
551 load_TLS_descriptor(t, cpu, 0);
552 load_TLS_descriptor(t, cpu, 1);
553 load_TLS_descriptor(t, cpu, 2);
555 xen_mc_issue(PARAVIRT_LAZY_CPU);
559 static void xen_load_gs_index(unsigned int idx)
561 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
566 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
569 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
570 u64 entry = *(u64 *)ptr;
572 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
577 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
583 static int cvt_gate_to_trap(int vector, const gate_desc *val,
584 struct trap_info *info)
588 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
591 info->vector = vector;
593 addr = gate_offset(*val);
596 * Look for known traps using IST, and substitute them
597 * appropriately. The debugger ones are the only ones we care
598 * about. Xen will handle faults like double_fault and
599 * machine_check, so we should never see them. Warn if
600 * there's an unexpected IST-using fault handler.
602 if (addr == (unsigned long)debug)
603 addr = (unsigned long)xen_debug;
604 else if (addr == (unsigned long)int3)
605 addr = (unsigned long)xen_int3;
606 else if (addr == (unsigned long)stack_segment)
607 addr = (unsigned long)xen_stack_segment;
608 else if (addr == (unsigned long)double_fault ||
609 addr == (unsigned long)nmi) {
610 /* Don't need to handle these */
612 #ifdef CONFIG_X86_MCE
613 } else if (addr == (unsigned long)machine_check) {
617 /* Some other trap using IST? */
618 if (WARN_ON(val->ist != 0))
621 #endif /* CONFIG_X86_64 */
622 info->address = addr;
624 info->cs = gate_segment(*val);
625 info->flags = val->dpl;
626 /* interrupt gates clear IF */
627 if (val->type == GATE_INTERRUPT)
628 info->flags |= 1 << 2;
633 /* Locations of each CPU's IDT */
634 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
636 /* Set an IDT entry. If the entry is part of the current IDT, then
638 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
640 unsigned long p = (unsigned long)&dt[entrynum];
641 unsigned long start, end;
643 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
647 start = __this_cpu_read(idt_desc.address);
648 end = start + __this_cpu_read(idt_desc.size) + 1;
652 native_write_idt_entry(dt, entrynum, g);
654 if (p >= start && (p + 8) <= end) {
655 struct trap_info info[2];
659 if (cvt_gate_to_trap(entrynum, g, &info[0]))
660 if (HYPERVISOR_set_trap_table(info))
667 static void xen_convert_trap_info(const struct desc_ptr *desc,
668 struct trap_info *traps)
670 unsigned in, out, count;
672 count = (desc->size+1) / sizeof(gate_desc);
675 for (in = out = 0; in < count; in++) {
676 gate_desc *entry = (gate_desc*)(desc->address) + in;
678 if (cvt_gate_to_trap(in, entry, &traps[out]))
681 traps[out].address = 0;
684 void xen_copy_trap_info(struct trap_info *traps)
686 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
688 xen_convert_trap_info(desc, traps);
691 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
692 hold a spinlock to protect the static traps[] array (static because
693 it avoids allocation, and saves stack space). */
694 static void xen_load_idt(const struct desc_ptr *desc)
696 static DEFINE_SPINLOCK(lock);
697 static struct trap_info traps[257];
699 trace_xen_cpu_load_idt(desc);
703 __get_cpu_var(idt_desc) = *desc;
705 xen_convert_trap_info(desc, traps);
708 if (HYPERVISOR_set_trap_table(traps))
714 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
715 they're handled differently. */
716 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
717 const void *desc, int type)
719 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
730 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
733 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
743 * Version of write_gdt_entry for use at early boot-time needed to
744 * update an entry as simply as possible.
746 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
747 const void *desc, int type)
749 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
758 xmaddr_t maddr = virt_to_machine(&dt[entry]);
760 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
761 dt[entry] = *(struct desc_struct *)desc;
767 static void xen_load_sp0(struct tss_struct *tss,
768 struct thread_struct *thread)
770 struct multicall_space mcs;
772 mcs = xen_mc_entry(0);
773 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
774 xen_mc_issue(PARAVIRT_LAZY_CPU);
777 void xen_set_iopl_mask(unsigned mask)
779 struct physdev_set_iopl set_iopl;
781 /* Force the change at ring 0. */
782 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
783 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
786 static void xen_io_delay(void)
790 #ifdef CONFIG_X86_LOCAL_APIC
791 static u32 xen_apic_read(u32 reg)
796 static void xen_apic_write(u32 reg, u32 val)
798 /* Warn to see if there's any stray references */
802 static u64 xen_apic_icr_read(void)
807 static void xen_apic_icr_write(u32 low, u32 id)
809 /* Warn to see if there's any stray references */
813 static void xen_apic_wait_icr_idle(void)
818 static u32 xen_safe_apic_wait_icr_idle(void)
823 static void set_xen_basic_apic_ops(void)
825 apic->read = xen_apic_read;
826 apic->write = xen_apic_write;
827 apic->icr_read = xen_apic_icr_read;
828 apic->icr_write = xen_apic_icr_write;
829 apic->wait_icr_idle = xen_apic_wait_icr_idle;
830 apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
835 static void xen_clts(void)
837 struct multicall_space mcs;
839 mcs = xen_mc_entry(0);
841 MULTI_fpu_taskswitch(mcs.mc, 0);
843 xen_mc_issue(PARAVIRT_LAZY_CPU);
846 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
848 static unsigned long xen_read_cr0(void)
850 unsigned long cr0 = percpu_read(xen_cr0_value);
852 if (unlikely(cr0 == 0)) {
853 cr0 = native_read_cr0();
854 percpu_write(xen_cr0_value, cr0);
860 static void xen_write_cr0(unsigned long cr0)
862 struct multicall_space mcs;
864 percpu_write(xen_cr0_value, cr0);
866 /* Only pay attention to cr0.TS; everything else is
868 mcs = xen_mc_entry(0);
870 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
872 xen_mc_issue(PARAVIRT_LAZY_CPU);
875 static void xen_write_cr4(unsigned long cr4)
880 native_write_cr4(cr4);
883 static inline unsigned long xen_read_cr8(void)
887 static inline void xen_write_cr8(unsigned long val)
892 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
903 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
904 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
905 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
908 base = ((u64)high << 32) | low;
909 if (HYPERVISOR_set_segment_base(which, base) != 0)
917 case MSR_SYSCALL_MASK:
918 case MSR_IA32_SYSENTER_CS:
919 case MSR_IA32_SYSENTER_ESP:
920 case MSR_IA32_SYSENTER_EIP:
921 /* Fast syscall setup is all done in hypercalls, so
922 these are all ignored. Stub them out here to stop
923 Xen console noise. */
926 case MSR_IA32_CR_PAT:
927 if (smp_processor_id() == 0)
928 xen_set_pat(((u64)high << 32) | low);
932 ret = native_write_msr_safe(msr, low, high);
938 void xen_setup_shared_info(void)
940 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
941 set_fixmap(FIX_PARAVIRT_BOOTMAP,
942 xen_start_info->shared_info);
944 HYPERVISOR_shared_info =
945 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
947 HYPERVISOR_shared_info =
948 (struct shared_info *)__va(xen_start_info->shared_info);
951 /* In UP this is as good a place as any to set up shared info */
952 xen_setup_vcpu_info_placement();
955 xen_setup_mfn_list_list();
958 /* This is called once we have the cpu_possible_map */
959 void xen_setup_vcpu_info_placement(void)
963 for_each_possible_cpu(cpu)
966 /* xen_vcpu_setup managed to place the vcpu_info within the
967 percpu area for all cpus, so make use of it */
968 if (have_vcpu_info_placement) {
969 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
970 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
971 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
972 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
973 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
977 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
978 unsigned long addr, unsigned len)
980 char *start, *end, *reloc;
983 start = end = reloc = NULL;
985 #define SITE(op, x) \
986 case PARAVIRT_PATCH(op.x): \
987 if (have_vcpu_info_placement) { \
988 start = (char *)xen_##x##_direct; \
989 end = xen_##x##_direct_end; \
990 reloc = xen_##x##_direct_reloc; \
995 SITE(pv_irq_ops, irq_enable);
996 SITE(pv_irq_ops, irq_disable);
997 SITE(pv_irq_ops, save_fl);
998 SITE(pv_irq_ops, restore_fl);
1002 if (start == NULL || (end-start) > len)
1005 ret = paravirt_patch_insns(insnbuf, len, start, end);
1007 /* Note: because reloc is assigned from something that
1008 appears to be an array, gcc assumes it's non-null,
1009 but doesn't know its relationship with start and
1011 if (reloc > start && reloc < end) {
1012 int reloc_off = reloc - start;
1013 long *relocp = (long *)(insnbuf + reloc_off);
1014 long delta = start - (char *)addr;
1022 ret = paravirt_patch_default(type, clobbers, insnbuf,
1030 static const struct pv_info xen_info __initconst = {
1031 .paravirt_enabled = 1,
1032 .shared_kernel_pmd = 0,
1034 #ifdef CONFIG_X86_64
1035 .extra_user_64bit_cs = FLAT_USER_CS64,
1041 static const struct pv_init_ops xen_init_ops __initconst = {
1045 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1048 .set_debugreg = xen_set_debugreg,
1049 .get_debugreg = xen_get_debugreg,
1053 .read_cr0 = xen_read_cr0,
1054 .write_cr0 = xen_write_cr0,
1056 .read_cr4 = native_read_cr4,
1057 .read_cr4_safe = native_read_cr4_safe,
1058 .write_cr4 = xen_write_cr4,
1060 #ifdef CONFIG_X86_64
1061 .read_cr8 = xen_read_cr8,
1062 .write_cr8 = xen_write_cr8,
1065 .wbinvd = native_wbinvd,
1067 .read_msr = native_read_msr_safe,
1068 .rdmsr_regs = native_rdmsr_safe_regs,
1069 .write_msr = xen_write_msr_safe,
1070 .wrmsr_regs = native_wrmsr_safe_regs,
1072 .read_tsc = native_read_tsc,
1073 .read_pmc = native_read_pmc,
1075 .read_tscp = native_read_tscp,
1078 .irq_enable_sysexit = xen_sysexit,
1079 #ifdef CONFIG_X86_64
1080 .usergs_sysret32 = xen_sysret32,
1081 .usergs_sysret64 = xen_sysret64,
1084 .load_tr_desc = paravirt_nop,
1085 .set_ldt = xen_set_ldt,
1086 .load_gdt = xen_load_gdt,
1087 .load_idt = xen_load_idt,
1088 .load_tls = xen_load_tls,
1089 #ifdef CONFIG_X86_64
1090 .load_gs_index = xen_load_gs_index,
1093 .alloc_ldt = xen_alloc_ldt,
1094 .free_ldt = xen_free_ldt,
1096 .store_gdt = native_store_gdt,
1097 .store_idt = native_store_idt,
1098 .store_tr = xen_store_tr,
1100 .write_ldt_entry = xen_write_ldt_entry,
1101 .write_gdt_entry = xen_write_gdt_entry,
1102 .write_idt_entry = xen_write_idt_entry,
1103 .load_sp0 = xen_load_sp0,
1105 .set_iopl_mask = xen_set_iopl_mask,
1106 .io_delay = xen_io_delay,
1108 /* Xen takes care of %gs when switching to usermode for us */
1109 .swapgs = paravirt_nop,
1111 .start_context_switch = paravirt_start_context_switch,
1112 .end_context_switch = xen_end_context_switch,
1115 static const struct pv_apic_ops xen_apic_ops __initconst = {
1116 #ifdef CONFIG_X86_LOCAL_APIC
1117 .startup_ipi_hook = paravirt_nop,
1121 static void xen_reboot(int reason)
1123 struct sched_shutdown r = { .reason = reason };
1125 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1129 static void xen_restart(char *msg)
1131 xen_reboot(SHUTDOWN_reboot);
1134 static void xen_emergency_restart(void)
1136 xen_reboot(SHUTDOWN_reboot);
1139 static void xen_machine_halt(void)
1141 xen_reboot(SHUTDOWN_poweroff);
1144 static void xen_machine_power_off(void)
1148 xen_reboot(SHUTDOWN_poweroff);
1151 static void xen_crash_shutdown(struct pt_regs *regs)
1153 xen_reboot(SHUTDOWN_crash);
1157 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1159 xen_reboot(SHUTDOWN_crash);
1163 static struct notifier_block xen_panic_block = {
1164 .notifier_call= xen_panic_event,
1167 int xen_panic_handler_init(void)
1169 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1173 static const struct machine_ops xen_machine_ops __initconst = {
1174 .restart = xen_restart,
1175 .halt = xen_machine_halt,
1176 .power_off = xen_machine_power_off,
1177 .shutdown = xen_machine_halt,
1178 .crash_shutdown = xen_crash_shutdown,
1179 .emergency_restart = xen_emergency_restart,
1183 * Set up the GDT and segment registers for -fstack-protector. Until
1184 * we do this, we have to be careful not to call any stack-protected
1185 * function, which is most of the kernel.
1187 static void __init xen_setup_stackprotector(void)
1189 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1190 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1192 setup_stack_canary_segment(0);
1193 switch_to_new_gdt(0);
1195 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1196 pv_cpu_ops.load_gdt = xen_load_gdt;
1199 /* First C function to be called on Xen boot */
1200 asmlinkage void __init xen_start_kernel(void)
1202 struct physdev_set_iopl set_iopl;
1206 if (!xen_start_info)
1209 xen_domain_type = XEN_PV_DOMAIN;
1211 xen_setup_machphys_mapping();
1213 /* Install Xen paravirt ops */
1215 pv_init_ops = xen_init_ops;
1216 pv_cpu_ops = xen_cpu_ops;
1217 pv_apic_ops = xen_apic_ops;
1219 x86_init.resources.memory_setup = xen_memory_setup;
1220 x86_init.oem.arch_setup = xen_arch_setup;
1221 x86_init.oem.banner = xen_banner;
1223 xen_init_time_ops();
1226 * Set up some pagetable state before starting to set any ptes.
1231 /* Prevent unwanted bits from being set in PTEs. */
1232 __supported_pte_mask &= ~_PAGE_GLOBAL;
1233 if (!xen_initial_domain())
1234 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1236 __supported_pte_mask |= _PAGE_IOMAP;
1239 * Prevent page tables from being allocated in highmem, even
1240 * if CONFIG_HIGHPTE is enabled.
1242 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1244 /* Work out if we support NX */
1247 xen_setup_features();
1250 if (!xen_feature(XENFEAT_auto_translated_physmap))
1251 xen_build_dynamic_phys_to_machine();
1254 * Set up kernel GDT and segment registers, mainly so that
1255 * -fstack-protector code can be executed.
1257 xen_setup_stackprotector();
1260 xen_init_cpuid_mask();
1262 #ifdef CONFIG_X86_LOCAL_APIC
1264 * set up the basic apic ops.
1266 set_xen_basic_apic_ops();
1269 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1270 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1271 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1274 machine_ops = xen_machine_ops;
1277 * The only reliable way to retain the initial address of the
1278 * percpu gdt_page is to remember it here, so we can go and
1279 * mark it RW later, when the initial percpu area is freed.
1281 xen_initial_gdt = &per_cpu(gdt_page, 0);
1285 #ifdef CONFIG_ACPI_NUMA
1287 * The pages we from Xen are not related to machine pages, so
1288 * any NUMA information the kernel tries to get from ACPI will
1289 * be meaningless. Prevent it from trying.
1294 pgd = (pgd_t *)xen_start_info->pt_base;
1296 if (!xen_initial_domain())
1297 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1299 __supported_pte_mask |= _PAGE_IOMAP;
1300 /* Don't do the full vcpu_info placement stuff until we have a
1301 possible map and a non-dummy shared_info. */
1302 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1304 local_irq_disable();
1305 early_boot_irqs_disabled = true;
1309 xen_raw_console_write("mapping kernel into physical memory\n");
1310 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1311 xen_ident_map_ISA();
1313 /* Allocate and initialize top and mid mfn levels for p2m structure */
1314 xen_build_mfn_list_list();
1316 /* keep using Xen gdt for now; no urgent need to change it */
1318 #ifdef CONFIG_X86_32
1319 pv_info.kernel_rpl = 1;
1320 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1321 pv_info.kernel_rpl = 0;
1323 pv_info.kernel_rpl = 0;
1325 /* set the limit of our address space */
1328 /* We used to do this in xen_arch_setup, but that is too late on AMD
1329 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1330 * which pokes 0xcf8 port.
1333 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1335 xen_raw_printk("physdev_op failed %d\n", rc);
1337 #ifdef CONFIG_X86_32
1338 /* set up basic CPUID stuff */
1339 cpu_detect(&new_cpu_data);
1340 new_cpu_data.hard_math = 1;
1341 new_cpu_data.wp_works_ok = 1;
1342 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1345 /* Poke various useful things into boot_params */
1346 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1347 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1348 ? __pa(xen_start_info->mod_start) : 0;
1349 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1350 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1352 if (!xen_initial_domain()) {
1353 add_preferred_console("xenboot", 0, NULL);
1354 add_preferred_console("tty", 0, NULL);
1355 add_preferred_console("hvc", 0, NULL);
1357 x86_init.pci.arch_init = pci_xen_init;
1359 const struct dom0_vga_console_info *info =
1360 (void *)((char *)xen_start_info +
1361 xen_start_info->console.dom0.info_off);
1363 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1364 xen_start_info->console.domU.mfn = 0;
1365 xen_start_info->console.domU.evtchn = 0;
1367 /* Make sure ACS will be enabled */
1370 /* Avoid searching for BIOS MP tables */
1371 x86_init.mpparse.find_smp_config = x86_init_noop;
1372 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1375 /* PCI BIOS service won't work from a PV guest. */
1376 pci_probe &= ~PCI_PROBE_BIOS;
1378 xen_raw_console_write("about to get started...\n");
1380 xen_setup_runstate_info(0);
1382 /* Start the world */
1383 #ifdef CONFIG_X86_32
1384 i386_start_kernel();
1386 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1390 static int init_hvm_pv_info(int *major, int *minor)
1392 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1395 base = xen_cpuid_base();
1396 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1399 *minor = eax & 0xffff;
1400 printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1402 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1404 pfn = __pa(hypercall_page);
1405 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1407 xen_setup_features();
1409 pv_info.name = "Xen HVM";
1411 xen_domain_type = XEN_HVM_DOMAIN;
1416 void __ref xen_hvm_init_shared_info(void)
1419 struct xen_add_to_physmap xatp;
1420 static struct shared_info *shared_info_page = 0;
1422 if (!shared_info_page)
1423 shared_info_page = (struct shared_info *)
1424 extend_brk(PAGE_SIZE, PAGE_SIZE);
1425 xatp.domid = DOMID_SELF;
1427 xatp.space = XENMAPSPACE_shared_info;
1428 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1429 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1432 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1434 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1435 * page, we use it in the event channel upcall and in some pvclock
1436 * related functions. We don't need the vcpu_info placement
1437 * optimizations because we don't use any pv_mmu or pv_irq op on
1439 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1440 * online but xen_hvm_init_shared_info is run at resume time too and
1441 * in that case multiple vcpus might be online. */
1442 for_each_online_cpu(cpu) {
1443 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1447 #ifdef CONFIG_XEN_PVHVM
1448 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1449 unsigned long action, void *hcpu)
1451 int cpu = (long)hcpu;
1453 case CPU_UP_PREPARE:
1454 xen_vcpu_setup(cpu);
1455 if (xen_have_vector_callback) {
1456 xen_init_lock_cpu(cpu);
1457 if (xen_feature(XENFEAT_hvm_safe_pvclock))
1458 xen_setup_timer(cpu);
1467 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1468 .notifier_call = xen_hvm_cpu_notify,
1471 static void __init xen_hvm_guest_init(void)
1476 r = init_hvm_pv_info(&major, &minor);
1480 xen_hvm_init_shared_info();
1482 if (xen_feature(XENFEAT_hvm_callback_vector))
1483 xen_have_vector_callback = 1;
1485 register_cpu_notifier(&xen_hvm_cpu_notifier);
1486 xen_unplug_emulated_devices();
1487 x86_init.irqs.intr_init = xen_init_IRQ;
1488 xen_hvm_init_time_ops();
1489 xen_hvm_init_mmu_ops();
1492 static bool __init xen_hvm_platform(void)
1494 if (xen_pv_domain())
1497 if (!xen_cpuid_base())
1503 bool xen_hvm_need_lapic(void)
1505 if (xen_pv_domain())
1507 if (!xen_hvm_domain())
1509 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1513 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1515 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1517 .detect = xen_hvm_platform,
1518 .init_platform = xen_hvm_guest_init,
1520 EXPORT_SYMBOL(x86_hyper_xen_hvm);