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
=
99 static void xen_vcpu_setup(int cpu
)
101 struct vcpu_register_vcpu_info info
;
103 struct vcpu_info
*vcpup
;
105 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
106 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
108 if (!have_vcpu_info_placement
)
109 return; /* already tested, not available */
111 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
113 info
.mfn
= virt_to_mfn(vcpup
);
114 info
.offset
= offset_in_page(vcpup
);
116 printk(KERN_DEBUG
"trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
117 cpu
, vcpup
, info
.mfn
, info
.offset
);
119 /* Check to see if the hypervisor will put the vcpu_info
120 structure where we want it, which allows direct access via
121 a percpu-variable. */
122 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
125 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
126 have_vcpu_info_placement
= 0;
128 /* This cpu is using the registered vcpu info, even if
129 later ones fail to. */
130 per_cpu(xen_vcpu
, cpu
) = vcpup
;
132 printk(KERN_DEBUG
"cpu %d using vcpu_info at %p\n",
138 * On restore, set the vcpu placement up again.
139 * If it fails, then we're in a bad state, since
140 * we can't back out from using it...
142 void xen_vcpu_restore(void)
144 if (have_vcpu_info_placement
) {
147 for_each_online_cpu(cpu
) {
148 bool other_cpu
= (cpu
!= smp_processor_id());
151 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
157 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
161 BUG_ON(!have_vcpu_info_placement
);
165 static void __init
xen_banner(void)
167 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
168 struct xen_extraversion extra
;
169 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
171 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
173 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
174 version
>> 16, version
& 0xffff, extra
.extraversion
,
175 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
178 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
179 unsigned int *cx
, unsigned int *dx
)
181 unsigned maskedx
= ~0;
184 * Mask out inconvenient features, to try and disable as many
185 * unsupported kernel subsystems as possible.
188 maskedx
= ~((1 << X86_FEATURE_APIC
) | /* disable APIC */
189 (1 << X86_FEATURE_ACPI
) | /* disable ACPI */
190 (1 << X86_FEATURE_MCE
) | /* disable MCE */
191 (1 << X86_FEATURE_MCA
) | /* disable MCA */
192 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
194 asm(XEN_EMULATE_PREFIX
"cpuid"
199 : "0" (*ax
), "2" (*cx
));
203 static void xen_set_debugreg(int reg
, unsigned long val
)
205 HYPERVISOR_set_debugreg(reg
, val
);
208 static unsigned long xen_get_debugreg(int reg
)
210 return HYPERVISOR_get_debugreg(reg
);
213 void xen_leave_lazy(void)
215 paravirt_leave_lazy(paravirt_get_lazy_mode());
219 static unsigned long xen_store_tr(void)
225 * Set the page permissions for a particular virtual address. If the
226 * address is a vmalloc mapping (or other non-linear mapping), then
227 * find the linear mapping of the page and also set its protections to
230 static void set_aliased_prot(void *v
, pgprot_t prot
)
238 ptep
= lookup_address((unsigned long)v
, &level
);
239 BUG_ON(ptep
== NULL
);
241 pfn
= pte_pfn(*ptep
);
242 page
= pfn_to_page(pfn
);
244 pte
= pfn_pte(pfn
, prot
);
246 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
249 if (!PageHighMem(page
)) {
250 void *av
= __va(PFN_PHYS(pfn
));
253 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
259 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
261 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
264 for(i
= 0; i
< entries
; i
+= entries_per_page
)
265 set_aliased_prot(ldt
+ i
, PAGE_KERNEL_RO
);
268 static void xen_free_ldt(struct desc_struct
*ldt
, unsigned entries
)
270 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
273 for(i
= 0; i
< entries
; i
+= entries_per_page
)
274 set_aliased_prot(ldt
+ i
, PAGE_KERNEL
);
277 static void xen_set_ldt(const void *addr
, unsigned entries
)
279 struct mmuext_op
*op
;
280 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
283 op
->cmd
= MMUEXT_SET_LDT
;
284 op
->arg1
.linear_addr
= (unsigned long)addr
;
285 op
->arg2
.nr_ents
= entries
;
287 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
289 xen_mc_issue(PARAVIRT_LAZY_CPU
);
292 static void xen_load_gdt(const struct desc_ptr
*dtr
)
294 unsigned long *frames
;
295 unsigned long va
= dtr
->address
;
296 unsigned int size
= dtr
->size
+ 1;
297 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
299 struct multicall_space mcs
;
301 /* A GDT can be up to 64k in size, which corresponds to 8192
302 8-byte entries, or 16 4k pages.. */
304 BUG_ON(size
> 65536);
305 BUG_ON(va
& ~PAGE_MASK
);
307 mcs
= xen_mc_entry(sizeof(*frames
) * pages
);
310 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
311 frames
[f
] = virt_to_mfn(va
);
312 make_lowmem_page_readonly((void *)va
);
315 MULTI_set_gdt(mcs
.mc
, frames
, size
/ sizeof(struct desc_struct
));
317 xen_mc_issue(PARAVIRT_LAZY_CPU
);
320 static void load_TLS_descriptor(struct thread_struct
*t
,
321 unsigned int cpu
, unsigned int i
)
323 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
324 xmaddr_t maddr
= virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
325 struct multicall_space mc
= __xen_mc_entry(0);
327 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
330 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
333 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
334 * it means we're in a context switch, and %gs has just been
335 * saved. This means we can zero it out to prevent faults on
336 * exit from the hypervisor if the next process has no %gs.
337 * Either way, it has been saved, and the new value will get
338 * loaded properly. This will go away as soon as Xen has been
339 * modified to not save/restore %gs for normal hypercalls.
341 * On x86_64, this hack is not used for %gs, because gs points
342 * to KERNEL_GS_BASE (and uses it for PDA references), so we
343 * must not zero %gs on x86_64
345 * For x86_64, we need to zero %fs, otherwise we may get an
346 * exception between the new %fs descriptor being loaded and
347 * %fs being effectively cleared at __switch_to().
349 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
359 load_TLS_descriptor(t
, cpu
, 0);
360 load_TLS_descriptor(t
, cpu
, 1);
361 load_TLS_descriptor(t
, cpu
, 2);
363 xen_mc_issue(PARAVIRT_LAZY_CPU
);
367 static void xen_load_gs_index(unsigned int idx
)
369 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
374 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
377 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
378 u64 entry
= *(u64
*)ptr
;
383 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
389 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
390 struct trap_info
*info
)
392 if (val
->type
!= 0xf && val
->type
!= 0xe)
395 info
->vector
= vector
;
396 info
->address
= gate_offset(*val
);
397 info
->cs
= gate_segment(*val
);
398 info
->flags
= val
->dpl
;
399 /* interrupt gates clear IF */
400 if (val
->type
== 0xe)
406 /* Locations of each CPU's IDT */
407 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
409 /* Set an IDT entry. If the entry is part of the current IDT, then
411 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
413 unsigned long p
= (unsigned long)&dt
[entrynum
];
414 unsigned long start
, end
;
418 start
= __get_cpu_var(idt_desc
).address
;
419 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
423 native_write_idt_entry(dt
, entrynum
, g
);
425 if (p
>= start
&& (p
+ 8) <= end
) {
426 struct trap_info info
[2];
430 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
431 if (HYPERVISOR_set_trap_table(info
))
438 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
439 struct trap_info
*traps
)
441 unsigned in
, out
, count
;
443 count
= (desc
->size
+1) / sizeof(gate_desc
);
446 for (in
= out
= 0; in
< count
; in
++) {
447 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
449 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
452 traps
[out
].address
= 0;
455 void xen_copy_trap_info(struct trap_info
*traps
)
457 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
459 xen_convert_trap_info(desc
, traps
);
462 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
463 hold a spinlock to protect the static traps[] array (static because
464 it avoids allocation, and saves stack space). */
465 static void xen_load_idt(const struct desc_ptr
*desc
)
467 static DEFINE_SPINLOCK(lock
);
468 static struct trap_info traps
[257];
472 __get_cpu_var(idt_desc
) = *desc
;
474 xen_convert_trap_info(desc
, traps
);
477 if (HYPERVISOR_set_trap_table(traps
))
483 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
484 they're handled differently. */
485 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
486 const void *desc
, int type
)
497 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
500 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
509 static void xen_load_sp0(struct tss_struct
*tss
,
510 struct thread_struct
*thread
)
512 struct multicall_space mcs
= xen_mc_entry(0);
513 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
514 xen_mc_issue(PARAVIRT_LAZY_CPU
);
517 static void xen_set_iopl_mask(unsigned mask
)
519 struct physdev_set_iopl set_iopl
;
521 /* Force the change at ring 0. */
522 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
523 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
526 static void xen_io_delay(void)
530 #ifdef CONFIG_X86_LOCAL_APIC
531 static u32
xen_apic_read(u32 reg
)
536 static void xen_apic_write(u32 reg
, u32 val
)
538 /* Warn to see if there's any stray references */
542 static u64
xen_apic_icr_read(void)
547 static void xen_apic_icr_write(u32 low
, u32 id
)
549 /* Warn to see if there's any stray references */
553 static void xen_apic_wait_icr_idle(void)
558 static u32
xen_safe_apic_wait_icr_idle(void)
563 static struct apic_ops xen_basic_apic_ops
= {
564 .read
= xen_apic_read
,
565 .write
= xen_apic_write
,
566 .icr_read
= xen_apic_icr_read
,
567 .icr_write
= xen_apic_icr_write
,
568 .wait_icr_idle
= xen_apic_wait_icr_idle
,
569 .safe_wait_icr_idle
= xen_safe_apic_wait_icr_idle
,
575 static void xen_clts(void)
577 struct multicall_space mcs
;
579 mcs
= xen_mc_entry(0);
581 MULTI_fpu_taskswitch(mcs
.mc
, 0);
583 xen_mc_issue(PARAVIRT_LAZY_CPU
);
586 static void xen_write_cr0(unsigned long cr0
)
588 struct multicall_space mcs
;
590 /* Only pay attention to cr0.TS; everything else is
592 mcs
= xen_mc_entry(0);
594 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
596 xen_mc_issue(PARAVIRT_LAZY_CPU
);
599 static void xen_write_cr4(unsigned long cr4
)
604 native_write_cr4(cr4
);
607 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
618 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
619 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
620 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
623 base
= ((u64
)high
<< 32) | low
;
624 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
632 case MSR_SYSCALL_MASK
:
633 case MSR_IA32_SYSENTER_CS
:
634 case MSR_IA32_SYSENTER_ESP
:
635 case MSR_IA32_SYSENTER_EIP
:
636 /* Fast syscall setup is all done in hypercalls, so
637 these are all ignored. Stub them out here to stop
638 Xen console noise. */
642 ret
= native_write_msr_safe(msr
, low
, high
);
648 void xen_setup_shared_info(void)
650 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
651 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
652 xen_start_info
->shared_info
);
654 HYPERVISOR_shared_info
=
655 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
657 HYPERVISOR_shared_info
=
658 (struct shared_info
*)__va(xen_start_info
->shared_info
);
661 /* In UP this is as good a place as any to set up shared info */
662 xen_setup_vcpu_info_placement();
665 xen_setup_mfn_list_list();
668 /* This is called once we have the cpu_possible_map */
669 void xen_setup_vcpu_info_placement(void)
673 for_each_possible_cpu(cpu
)
676 /* xen_vcpu_setup managed to place the vcpu_info within the
677 percpu area for all cpus, so make use of it */
678 if (have_vcpu_info_placement
) {
679 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
681 pv_irq_ops
.save_fl
= __PV_IS_CALLEE_SAVE(xen_save_fl_direct
);
682 pv_irq_ops
.restore_fl
= __PV_IS_CALLEE_SAVE(xen_restore_fl_direct
);
683 pv_irq_ops
.irq_disable
= __PV_IS_CALLEE_SAVE(xen_irq_disable_direct
);
684 pv_irq_ops
.irq_enable
= __PV_IS_CALLEE_SAVE(xen_irq_enable_direct
);
685 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
689 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
690 unsigned long addr
, unsigned len
)
692 char *start
, *end
, *reloc
;
695 start
= end
= reloc
= NULL
;
697 #define SITE(op, x) \
698 case PARAVIRT_PATCH(op.x): \
699 if (have_vcpu_info_placement) { \
700 start = (char *)xen_##x##_direct; \
701 end = xen_##x##_direct_end; \
702 reloc = xen_##x##_direct_reloc; \
707 SITE(pv_irq_ops
, irq_enable
);
708 SITE(pv_irq_ops
, irq_disable
);
709 SITE(pv_irq_ops
, save_fl
);
710 SITE(pv_irq_ops
, restore_fl
);
714 if (start
== NULL
|| (end
-start
) > len
)
717 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
719 /* Note: because reloc is assigned from something that
720 appears to be an array, gcc assumes it's non-null,
721 but doesn't know its relationship with start and
723 if (reloc
> start
&& reloc
< end
) {
724 int reloc_off
= reloc
- start
;
725 long *relocp
= (long *)(insnbuf
+ reloc_off
);
726 long delta
= start
- (char *)addr
;
734 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
742 static const struct pv_info xen_info __initdata
= {
743 .paravirt_enabled
= 1,
744 .shared_kernel_pmd
= 0,
749 static const struct pv_init_ops xen_init_ops __initdata
= {
752 .banner
= xen_banner
,
753 .memory_setup
= xen_memory_setup
,
754 .arch_setup
= xen_arch_setup
,
755 .post_allocator_init
= xen_post_allocator_init
,
758 static const struct pv_time_ops xen_time_ops __initdata
= {
759 .time_init
= xen_time_init
,
761 .set_wallclock
= xen_set_wallclock
,
762 .get_wallclock
= xen_get_wallclock
,
763 .get_tsc_khz
= xen_tsc_khz
,
764 .sched_clock
= xen_sched_clock
,
767 static const struct pv_cpu_ops xen_cpu_ops __initdata
= {
770 .set_debugreg
= xen_set_debugreg
,
771 .get_debugreg
= xen_get_debugreg
,
775 .read_cr0
= native_read_cr0
,
776 .write_cr0
= xen_write_cr0
,
778 .read_cr4
= native_read_cr4
,
779 .read_cr4_safe
= native_read_cr4_safe
,
780 .write_cr4
= xen_write_cr4
,
782 .wbinvd
= native_wbinvd
,
784 .read_msr
= native_read_msr_safe
,
785 .write_msr
= xen_write_msr_safe
,
786 .read_tsc
= native_read_tsc
,
787 .read_pmc
= native_read_pmc
,
790 .irq_enable_sysexit
= xen_sysexit
,
792 .usergs_sysret32
= xen_sysret32
,
793 .usergs_sysret64
= xen_sysret64
,
796 .load_tr_desc
= paravirt_nop
,
797 .set_ldt
= xen_set_ldt
,
798 .load_gdt
= xen_load_gdt
,
799 .load_idt
= xen_load_idt
,
800 .load_tls
= xen_load_tls
,
802 .load_gs_index
= xen_load_gs_index
,
805 .alloc_ldt
= xen_alloc_ldt
,
806 .free_ldt
= xen_free_ldt
,
808 .store_gdt
= native_store_gdt
,
809 .store_idt
= native_store_idt
,
810 .store_tr
= xen_store_tr
,
812 .write_ldt_entry
= xen_write_ldt_entry
,
813 .write_gdt_entry
= xen_write_gdt_entry
,
814 .write_idt_entry
= xen_write_idt_entry
,
815 .load_sp0
= xen_load_sp0
,
817 .set_iopl_mask
= xen_set_iopl_mask
,
818 .io_delay
= xen_io_delay
,
820 /* Xen takes care of %gs when switching to usermode for us */
821 .swapgs
= paravirt_nop
,
824 .enter
= paravirt_enter_lazy_cpu
,
825 .leave
= xen_leave_lazy
,
829 static const struct pv_apic_ops xen_apic_ops __initdata
= {
830 #ifdef CONFIG_X86_LOCAL_APIC
831 .setup_boot_clock
= paravirt_nop
,
832 .setup_secondary_clock
= paravirt_nop
,
833 .startup_ipi_hook
= paravirt_nop
,
837 static void xen_reboot(int reason
)
839 struct sched_shutdown r
= { .reason
= reason
};
845 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
849 static void xen_restart(char *msg
)
851 xen_reboot(SHUTDOWN_reboot
);
854 static void xen_emergency_restart(void)
856 xen_reboot(SHUTDOWN_reboot
);
859 static void xen_machine_halt(void)
861 xen_reboot(SHUTDOWN_poweroff
);
864 static void xen_crash_shutdown(struct pt_regs
*regs
)
866 xen_reboot(SHUTDOWN_crash
);
869 static const struct machine_ops __initdata xen_machine_ops
= {
870 .restart
= xen_restart
,
871 .halt
= xen_machine_halt
,
872 .power_off
= xen_machine_halt
,
873 .shutdown
= xen_machine_halt
,
874 .crash_shutdown
= xen_crash_shutdown
,
875 .emergency_restart
= xen_emergency_restart
,
879 /* First C function to be called on Xen boot */
880 asmlinkage
void __init
xen_start_kernel(void)
887 xen_domain_type
= XEN_PV_DOMAIN
;
889 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3", 5) != 0);
891 xen_setup_features();
893 /* Install Xen paravirt ops */
895 pv_init_ops
= xen_init_ops
;
896 pv_time_ops
= xen_time_ops
;
897 pv_cpu_ops
= xen_cpu_ops
;
898 pv_apic_ops
= xen_apic_ops
;
899 pv_mmu_ops
= xen_mmu_ops
;
903 #ifdef CONFIG_X86_LOCAL_APIC
905 * set up the basic apic ops.
907 apic_ops
= &xen_basic_apic_ops
;
910 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
911 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
912 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
915 machine_ops
= xen_machine_ops
;
918 /* Disable until direct per-cpu data access. */
919 have_vcpu_info_placement
= 0;
924 * Setup percpu state. We only need to do this for 64-bit
925 * because 32-bit already has %fs set properly.
927 load_percpu_segment(0);
930 * The only reliable way to retain the initial address of the
931 * percpu gdt_page is to remember it here, so we can go and
932 * mark it RW later, when the initial percpu area is freed.
934 xen_initial_gdt
= &per_cpu(gdt_page
, 0);
939 if (!xen_feature(XENFEAT_auto_translated_physmap
))
940 xen_build_dynamic_phys_to_machine();
942 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
944 /* Prevent unwanted bits from being set in PTEs. */
945 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
946 if (!xen_initial_domain())
947 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
949 /* Don't do the full vcpu_info placement stuff until we have a
950 possible map and a non-dummy shared_info. */
951 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
953 xen_raw_console_write("mapping kernel into physical memory\n");
954 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
958 /* keep using Xen gdt for now; no urgent need to change it */
960 pv_info
.kernel_rpl
= 1;
961 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
962 pv_info
.kernel_rpl
= 0;
964 /* set the limit of our address space */
968 /* set up basic CPUID stuff */
969 cpu_detect(&new_cpu_data
);
970 new_cpu_data
.hard_math
= 1;
971 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
974 /* Poke various useful things into boot_params */
975 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
976 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
977 ? __pa(xen_start_info
->mod_start
) : 0;
978 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
979 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
981 if (!xen_initial_domain()) {
982 add_preferred_console("xenboot", 0, NULL
);
983 add_preferred_console("tty", 0, NULL
);
984 add_preferred_console("hvc", 0, NULL
);
987 xen_raw_console_write("about to get started...\n");
989 /* Start the world */
993 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));