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/kernel.h>
15 #include <linux/init.h>
16 #include <linux/smp.h>
17 #include <linux/preempt.h>
18 #include <linux/hardirq.h>
19 #include <linux/percpu.h>
20 #include <linux/delay.h>
21 #include <linux/start_kernel.h>
22 #include <linux/sched.h>
23 #include <linux/bootmem.h>
24 #include <linux/module.h>
26 #include <linux/page-flags.h>
27 #include <linux/highmem.h>
28 #include <linux/console.h>
30 #include <xen/interface/xen.h>
31 #include <xen/interface/version.h>
32 #include <xen/interface/physdev.h>
33 #include <xen/interface/vcpu.h>
34 #include <xen/features.h>
36 #include <xen/hvc-console.h>
38 #include <asm/paravirt.h>
41 #include <asm/xen/hypercall.h>
42 #include <asm/xen/hypervisor.h>
43 #include <asm/fixmap.h>
44 #include <asm/processor.h>
45 #include <asm/msr-index.h>
46 #include <asm/setup.h>
48 #include <asm/pgtable.h>
49 #include <asm/tlbflush.h>
50 #include <asm/reboot.h>
54 #include "multicalls.h"
56 EXPORT_SYMBOL_GPL(hypercall_page
);
58 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
59 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
61 enum xen_domain_type xen_domain_type
= XEN_NATIVE
;
62 EXPORT_SYMBOL_GPL(xen_domain_type
);
64 struct start_info
*xen_start_info
;
65 EXPORT_SYMBOL_GPL(xen_start_info
);
67 struct shared_info xen_dummy_shared_info
;
69 void *xen_initial_gdt
;
72 * Point at some empty memory to start with. We map the real shared_info
73 * page as soon as fixmap is up and running.
75 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
78 * Flag to determine whether vcpu info placement is available on all
79 * VCPUs. We assume it is to start with, and then set it to zero on
80 * the first failure. This is because it can succeed on some VCPUs
81 * and not others, since it can involve hypervisor memory allocation,
82 * or because the guest failed to guarantee all the appropriate
83 * constraints on all VCPUs (ie buffer can't cross a page boundary).
85 * Note that any particular CPU may be using a placed vcpu structure,
86 * but we can only optimise if the all are.
88 * 0: not available, 1: available
90 static int have_vcpu_info_placement
= 1;
92 static void xen_vcpu_setup(int cpu
)
94 struct vcpu_register_vcpu_info info
;
96 struct vcpu_info
*vcpup
;
98 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
99 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
101 if (!have_vcpu_info_placement
)
102 return; /* already tested, not available */
104 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
106 info
.mfn
= arbitrary_virt_to_mfn(vcpup
);
107 info
.offset
= offset_in_page(vcpup
);
109 printk(KERN_DEBUG
"trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
110 cpu
, vcpup
, info
.mfn
, info
.offset
);
112 /* Check to see if the hypervisor will put the vcpu_info
113 structure where we want it, which allows direct access via
114 a percpu-variable. */
115 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
118 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
119 have_vcpu_info_placement
= 0;
121 /* This cpu is using the registered vcpu info, even if
122 later ones fail to. */
123 per_cpu(xen_vcpu
, cpu
) = vcpup
;
125 printk(KERN_DEBUG
"cpu %d using vcpu_info at %p\n",
131 * On restore, set the vcpu placement up again.
132 * If it fails, then we're in a bad state, since
133 * we can't back out from using it...
135 void xen_vcpu_restore(void)
137 if (have_vcpu_info_placement
) {
140 for_each_online_cpu(cpu
) {
141 bool other_cpu
= (cpu
!= smp_processor_id());
144 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
150 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
154 BUG_ON(!have_vcpu_info_placement
);
158 static void __init
xen_banner(void)
160 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
161 struct xen_extraversion extra
;
162 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
164 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
166 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
167 version
>> 16, version
& 0xffff, extra
.extraversion
,
168 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
171 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
172 unsigned int *cx
, unsigned int *dx
)
174 unsigned maskedx
= ~0;
177 * Mask out inconvenient features, to try and disable as many
178 * unsupported kernel subsystems as possible.
181 maskedx
= ~((1 << X86_FEATURE_APIC
) | /* disable APIC */
182 (1 << X86_FEATURE_ACPI
) | /* disable ACPI */
183 (1 << X86_FEATURE_MCE
) | /* disable MCE */
184 (1 << X86_FEATURE_MCA
) | /* disable MCA */
185 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
187 asm(XEN_EMULATE_PREFIX
"cpuid"
192 : "0" (*ax
), "2" (*cx
));
196 static void xen_set_debugreg(int reg
, unsigned long val
)
198 HYPERVISOR_set_debugreg(reg
, val
);
201 static unsigned long xen_get_debugreg(int reg
)
203 return HYPERVISOR_get_debugreg(reg
);
206 static void xen_end_context_switch(struct task_struct
*next
)
209 paravirt_end_context_switch(next
);
212 static unsigned long xen_store_tr(void)
218 * Set the page permissions for a particular virtual address. If the
219 * address is a vmalloc mapping (or other non-linear mapping), then
220 * find the linear mapping of the page and also set its protections to
223 static void set_aliased_prot(void *v
, pgprot_t prot
)
231 ptep
= lookup_address((unsigned long)v
, &level
);
232 BUG_ON(ptep
== NULL
);
234 pfn
= pte_pfn(*ptep
);
235 page
= pfn_to_page(pfn
);
237 pte
= pfn_pte(pfn
, prot
);
239 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
242 if (!PageHighMem(page
)) {
243 void *av
= __va(PFN_PHYS(pfn
));
246 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
252 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
254 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
257 for(i
= 0; i
< entries
; i
+= entries_per_page
)
258 set_aliased_prot(ldt
+ i
, PAGE_KERNEL_RO
);
261 static void xen_free_ldt(struct desc_struct
*ldt
, unsigned entries
)
263 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
266 for(i
= 0; i
< entries
; i
+= entries_per_page
)
267 set_aliased_prot(ldt
+ i
, PAGE_KERNEL
);
270 static void xen_set_ldt(const void *addr
, unsigned entries
)
272 struct mmuext_op
*op
;
273 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
276 op
->cmd
= MMUEXT_SET_LDT
;
277 op
->arg1
.linear_addr
= (unsigned long)addr
;
278 op
->arg2
.nr_ents
= entries
;
280 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
282 xen_mc_issue(PARAVIRT_LAZY_CPU
);
285 static void xen_load_gdt(const struct desc_ptr
*dtr
)
287 unsigned long va
= dtr
->address
;
288 unsigned int size
= dtr
->size
+ 1;
289 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
290 unsigned long frames
[pages
];
293 /* A GDT can be up to 64k in size, which corresponds to 8192
294 8-byte entries, or 16 4k pages.. */
296 BUG_ON(size
> 65536);
297 BUG_ON(va
& ~PAGE_MASK
);
299 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
301 pte_t
*ptep
= lookup_address(va
, &level
);
302 unsigned long pfn
, mfn
;
305 BUG_ON(ptep
== NULL
);
307 pfn
= pte_pfn(*ptep
);
308 mfn
= pfn_to_mfn(pfn
);
309 virt
= __va(PFN_PHYS(pfn
));
313 printk("xen_load_gdt: %d va=%p mfn=%lx pfn=%lx va'=%p\n",
314 f
, (void *)va
, mfn
, pfn
, virt
);
316 make_lowmem_page_readonly((void *)va
);
317 make_lowmem_page_readonly(virt
);
320 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
324 static void load_TLS_descriptor(struct thread_struct
*t
,
325 unsigned int cpu
, unsigned int i
)
327 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
328 xmaddr_t maddr
= arbitrary_virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
329 struct multicall_space mc
= __xen_mc_entry(0);
331 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
334 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
337 * XXX sleazy hack: If we're being called in a lazy-cpu zone
338 * and lazy gs handling is enabled, it means we're in a
339 * context switch, and %gs has just been saved. This means we
340 * can zero it out to prevent faults on exit from the
341 * hypervisor if the next process has no %gs. Either way, it
342 * has been saved, and the new value will get loaded properly.
343 * This will go away as soon as Xen has been modified to not
344 * save/restore %gs for normal hypercalls.
346 * On x86_64, this hack is not used for %gs, because gs points
347 * to KERNEL_GS_BASE (and uses it for PDA references), so we
348 * must not zero %gs on x86_64
350 * For x86_64, we need to zero %fs, otherwise we may get an
351 * exception between the new %fs descriptor being loaded and
352 * %fs being effectively cleared at __switch_to().
354 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
364 load_TLS_descriptor(t
, cpu
, 0);
365 load_TLS_descriptor(t
, cpu
, 1);
366 load_TLS_descriptor(t
, cpu
, 2);
368 xen_mc_issue(PARAVIRT_LAZY_CPU
);
372 static void xen_load_gs_index(unsigned int idx
)
374 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
379 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
382 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
383 u64 entry
= *(u64
*)ptr
;
388 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
394 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
395 struct trap_info
*info
)
397 if (val
->type
!= 0xf && val
->type
!= 0xe)
400 info
->vector
= vector
;
401 info
->address
= gate_offset(*val
);
402 info
->cs
= gate_segment(*val
);
403 info
->flags
= val
->dpl
;
404 /* interrupt gates clear IF */
405 if (val
->type
== 0xe)
411 /* Locations of each CPU's IDT */
412 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
414 /* Set an IDT entry. If the entry is part of the current IDT, then
416 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
418 unsigned long p
= (unsigned long)&dt
[entrynum
];
419 unsigned long start
, end
;
423 start
= __get_cpu_var(idt_desc
).address
;
424 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
428 native_write_idt_entry(dt
, entrynum
, g
);
430 if (p
>= start
&& (p
+ 8) <= end
) {
431 struct trap_info info
[2];
435 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
436 if (HYPERVISOR_set_trap_table(info
))
443 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
444 struct trap_info
*traps
)
446 unsigned in
, out
, count
;
448 count
= (desc
->size
+1) / sizeof(gate_desc
);
451 for (in
= out
= 0; in
< count
; in
++) {
452 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
454 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
457 traps
[out
].address
= 0;
460 void xen_copy_trap_info(struct trap_info
*traps
)
462 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
464 xen_convert_trap_info(desc
, traps
);
467 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
468 hold a spinlock to protect the static traps[] array (static because
469 it avoids allocation, and saves stack space). */
470 static void xen_load_idt(const struct desc_ptr
*desc
)
472 static DEFINE_SPINLOCK(lock
);
473 static struct trap_info traps
[257];
477 __get_cpu_var(idt_desc
) = *desc
;
479 xen_convert_trap_info(desc
, traps
);
482 if (HYPERVISOR_set_trap_table(traps
))
488 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
489 they're handled differently. */
490 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
491 const void *desc
, int type
)
502 xmaddr_t maddr
= arbitrary_virt_to_machine(&dt
[entry
]);
505 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
514 static void xen_load_sp0(struct tss_struct
*tss
,
515 struct thread_struct
*thread
)
517 struct multicall_space mcs
= xen_mc_entry(0);
518 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
519 xen_mc_issue(PARAVIRT_LAZY_CPU
);
522 static void xen_set_iopl_mask(unsigned mask
)
524 struct physdev_set_iopl set_iopl
;
526 /* Force the change at ring 0. */
527 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
528 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
531 static void xen_io_delay(void)
535 #ifdef CONFIG_X86_LOCAL_APIC
536 static u32
xen_apic_read(u32 reg
)
541 static void xen_apic_write(u32 reg
, u32 val
)
543 /* Warn to see if there's any stray references */
547 static u64
xen_apic_icr_read(void)
552 static void xen_apic_icr_write(u32 low
, u32 id
)
554 /* Warn to see if there's any stray references */
558 static void xen_apic_wait_icr_idle(void)
563 static u32
xen_safe_apic_wait_icr_idle(void)
568 static void set_xen_basic_apic_ops(void)
570 apic
->read
= xen_apic_read
;
571 apic
->write
= xen_apic_write
;
572 apic
->icr_read
= xen_apic_icr_read
;
573 apic
->icr_write
= xen_apic_icr_write
;
574 apic
->wait_icr_idle
= xen_apic_wait_icr_idle
;
575 apic
->safe_wait_icr_idle
= xen_safe_apic_wait_icr_idle
;
581 static void xen_clts(void)
583 struct multicall_space mcs
;
585 mcs
= xen_mc_entry(0);
587 MULTI_fpu_taskswitch(mcs
.mc
, 0);
589 xen_mc_issue(PARAVIRT_LAZY_CPU
);
592 static void xen_write_cr0(unsigned long cr0
)
594 struct multicall_space mcs
;
596 /* Only pay attention to cr0.TS; everything else is
598 mcs
= xen_mc_entry(0);
600 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
602 xen_mc_issue(PARAVIRT_LAZY_CPU
);
605 static void xen_write_cr4(unsigned long cr4
)
610 native_write_cr4(cr4
);
613 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
624 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
625 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
626 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
629 base
= ((u64
)high
<< 32) | low
;
630 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
638 case MSR_SYSCALL_MASK
:
639 case MSR_IA32_SYSENTER_CS
:
640 case MSR_IA32_SYSENTER_ESP
:
641 case MSR_IA32_SYSENTER_EIP
:
642 /* Fast syscall setup is all done in hypercalls, so
643 these are all ignored. Stub them out here to stop
644 Xen console noise. */
648 ret
= native_write_msr_safe(msr
, low
, high
);
654 void xen_setup_shared_info(void)
656 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
657 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
658 xen_start_info
->shared_info
);
660 HYPERVISOR_shared_info
=
661 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
663 HYPERVISOR_shared_info
=
664 (struct shared_info
*)__va(xen_start_info
->shared_info
);
667 /* In UP this is as good a place as any to set up shared info */
668 xen_setup_vcpu_info_placement();
671 xen_setup_mfn_list_list();
674 /* This is called once we have the cpu_possible_map */
675 void xen_setup_vcpu_info_placement(void)
679 for_each_possible_cpu(cpu
)
682 /* xen_vcpu_setup managed to place the vcpu_info within the
683 percpu area for all cpus, so make use of it */
684 if (have_vcpu_info_placement
) {
685 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
687 pv_irq_ops
.save_fl
= __PV_IS_CALLEE_SAVE(xen_save_fl_direct
);
688 pv_irq_ops
.restore_fl
= __PV_IS_CALLEE_SAVE(xen_restore_fl_direct
);
689 pv_irq_ops
.irq_disable
= __PV_IS_CALLEE_SAVE(xen_irq_disable_direct
);
690 pv_irq_ops
.irq_enable
= __PV_IS_CALLEE_SAVE(xen_irq_enable_direct
);
691 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
695 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
696 unsigned long addr
, unsigned len
)
698 char *start
, *end
, *reloc
;
701 start
= end
= reloc
= NULL
;
703 #define SITE(op, x) \
704 case PARAVIRT_PATCH(op.x): \
705 if (have_vcpu_info_placement) { \
706 start = (char *)xen_##x##_direct; \
707 end = xen_##x##_direct_end; \
708 reloc = xen_##x##_direct_reloc; \
713 SITE(pv_irq_ops
, irq_enable
);
714 SITE(pv_irq_ops
, irq_disable
);
715 SITE(pv_irq_ops
, save_fl
);
716 SITE(pv_irq_ops
, restore_fl
);
720 if (start
== NULL
|| (end
-start
) > len
)
723 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
725 /* Note: because reloc is assigned from something that
726 appears to be an array, gcc assumes it's non-null,
727 but doesn't know its relationship with start and
729 if (reloc
> start
&& reloc
< end
) {
730 int reloc_off
= reloc
- start
;
731 long *relocp
= (long *)(insnbuf
+ reloc_off
);
732 long delta
= start
- (char *)addr
;
740 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
748 static const struct pv_info xen_info __initdata
= {
749 .paravirt_enabled
= 1,
750 .shared_kernel_pmd
= 0,
755 static const struct pv_init_ops xen_init_ops __initdata
= {
758 .banner
= xen_banner
,
759 .memory_setup
= xen_memory_setup
,
760 .arch_setup
= xen_arch_setup
,
761 .post_allocator_init
= xen_post_allocator_init
,
764 static const struct pv_time_ops xen_time_ops __initdata
= {
765 .time_init
= xen_time_init
,
767 .set_wallclock
= xen_set_wallclock
,
768 .get_wallclock
= xen_get_wallclock
,
769 .get_tsc_khz
= xen_tsc_khz
,
770 .sched_clock
= xen_sched_clock
,
773 static const struct pv_cpu_ops xen_cpu_ops __initdata
= {
776 .set_debugreg
= xen_set_debugreg
,
777 .get_debugreg
= xen_get_debugreg
,
781 .read_cr0
= native_read_cr0
,
782 .write_cr0
= xen_write_cr0
,
784 .read_cr4
= native_read_cr4
,
785 .read_cr4_safe
= native_read_cr4_safe
,
786 .write_cr4
= xen_write_cr4
,
788 .wbinvd
= native_wbinvd
,
790 .read_msr
= native_read_msr_safe
,
791 .write_msr
= xen_write_msr_safe
,
792 .read_tsc
= native_read_tsc
,
793 .read_pmc
= native_read_pmc
,
796 .irq_enable_sysexit
= xen_sysexit
,
798 .usergs_sysret32
= xen_sysret32
,
799 .usergs_sysret64
= xen_sysret64
,
802 .load_tr_desc
= paravirt_nop
,
803 .set_ldt
= xen_set_ldt
,
804 .load_gdt
= xen_load_gdt
,
805 .load_idt
= xen_load_idt
,
806 .load_tls
= xen_load_tls
,
808 .load_gs_index
= xen_load_gs_index
,
811 .alloc_ldt
= xen_alloc_ldt
,
812 .free_ldt
= xen_free_ldt
,
814 .store_gdt
= native_store_gdt
,
815 .store_idt
= native_store_idt
,
816 .store_tr
= xen_store_tr
,
818 .write_ldt_entry
= xen_write_ldt_entry
,
819 .write_gdt_entry
= xen_write_gdt_entry
,
820 .write_idt_entry
= xen_write_idt_entry
,
821 .load_sp0
= xen_load_sp0
,
823 .set_iopl_mask
= xen_set_iopl_mask
,
824 .io_delay
= xen_io_delay
,
826 /* Xen takes care of %gs when switching to usermode for us */
827 .swapgs
= paravirt_nop
,
829 .start_context_switch
= paravirt_start_context_switch
,
830 .end_context_switch
= xen_end_context_switch
,
833 static const struct pv_apic_ops xen_apic_ops __initdata
= {
834 #ifdef CONFIG_X86_LOCAL_APIC
835 .setup_boot_clock
= paravirt_nop
,
836 .setup_secondary_clock
= paravirt_nop
,
837 .startup_ipi_hook
= paravirt_nop
,
841 static void xen_reboot(int reason
)
843 struct sched_shutdown r
= { .reason
= reason
};
849 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
853 static void xen_restart(char *msg
)
855 xen_reboot(SHUTDOWN_reboot
);
858 static void xen_emergency_restart(void)
860 xen_reboot(SHUTDOWN_reboot
);
863 static void xen_machine_halt(void)
865 xen_reboot(SHUTDOWN_poweroff
);
868 static void xen_crash_shutdown(struct pt_regs
*regs
)
870 xen_reboot(SHUTDOWN_crash
);
873 static const struct machine_ops __initdata xen_machine_ops
= {
874 .restart
= xen_restart
,
875 .halt
= xen_machine_halt
,
876 .power_off
= xen_machine_halt
,
877 .shutdown
= xen_machine_halt
,
878 .crash_shutdown
= xen_crash_shutdown
,
879 .emergency_restart
= xen_emergency_restart
,
883 /* First C function to be called on Xen boot */
884 asmlinkage
void __init
xen_start_kernel(void)
891 xen_domain_type
= XEN_PV_DOMAIN
;
893 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3", 5) != 0);
895 xen_setup_features();
897 /* Install Xen paravirt ops */
899 pv_init_ops
= xen_init_ops
;
900 pv_time_ops
= xen_time_ops
;
901 pv_cpu_ops
= xen_cpu_ops
;
902 pv_apic_ops
= xen_apic_ops
;
903 pv_mmu_ops
= xen_mmu_ops
;
907 #ifdef CONFIG_X86_LOCAL_APIC
909 * set up the basic apic ops.
911 set_xen_basic_apic_ops();
914 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
915 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
916 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
919 machine_ops
= xen_machine_ops
;
923 * Setup percpu state. We only need to do this for 64-bit
924 * because 32-bit already has %fs set properly.
926 load_percpu_segment(0);
929 * The only reliable way to retain the initial address of the
930 * percpu gdt_page is to remember it here, so we can go and
931 * mark it RW later, when the initial percpu area is freed.
933 xen_initial_gdt
= &per_cpu(gdt_page
, 0);
938 if (!xen_feature(XENFEAT_auto_translated_physmap
))
939 xen_build_dynamic_phys_to_machine();
941 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
943 /* Prevent unwanted bits from being set in PTEs. */
944 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
945 if (!xen_initial_domain())
946 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
948 /* Don't do the full vcpu_info placement stuff until we have a
949 possible map and a non-dummy shared_info. */
950 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
953 early_boot_irqs_off();
955 xen_raw_console_write("mapping kernel into physical memory\n");
956 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
960 /* keep using Xen gdt for now; no urgent need to change it */
962 pv_info
.kernel_rpl
= 1;
963 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
964 pv_info
.kernel_rpl
= 0;
966 /* set the limit of our address space */
970 /* set up basic CPUID stuff */
971 cpu_detect(&new_cpu_data
);
972 new_cpu_data
.hard_math
= 1;
973 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
976 /* Poke various useful things into boot_params */
977 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
978 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
979 ? __pa(xen_start_info
->mod_start
) : 0;
980 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
981 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
983 if (!xen_initial_domain()) {
984 add_preferred_console("xenboot", 0, NULL
);
985 add_preferred_console("tty", 0, NULL
);
986 add_preferred_console("hvc", 0, NULL
);
989 xen_raw_console_write("about to get started...\n");
991 /* Start the world */
995 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));