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/kprobes.h>
24 #include <linux/bootmem.h>
25 #include <linux/module.h>
27 #include <linux/page-flags.h>
28 #include <linux/highmem.h>
29 #include <linux/console.h>
30 #include <linux/pci.h>
31 #include <linux/gfp.h>
34 #include <xen/interface/xen.h>
35 #include <xen/interface/version.h>
36 #include <xen/interface/physdev.h>
37 #include <xen/interface/vcpu.h>
38 #include <xen/features.h>
40 #include <xen/hvc-console.h>
42 #include <asm/paravirt.h>
45 #include <asm/xen/hypercall.h>
46 #include <asm/xen/hypervisor.h>
47 #include <asm/fixmap.h>
48 #include <asm/processor.h>
49 #include <asm/proto.h>
50 #include <asm/msr-index.h>
51 #include <asm/traps.h>
52 #include <asm/setup.h>
54 #include <asm/pgalloc.h>
55 #include <asm/pgtable.h>
56 #include <asm/tlbflush.h>
57 #include <asm/reboot.h>
58 #include <asm/stackprotector.h>
62 #include "multicalls.h"
64 EXPORT_SYMBOL_GPL(hypercall_page
);
66 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
67 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
69 enum xen_domain_type xen_domain_type
= XEN_NATIVE
;
70 EXPORT_SYMBOL_GPL(xen_domain_type
);
72 struct start_info
*xen_start_info
;
73 EXPORT_SYMBOL_GPL(xen_start_info
);
75 struct shared_info xen_dummy_shared_info
;
77 void *xen_initial_gdt
;
80 * Point at some empty memory to start with. We map the real shared_info
81 * page as soon as fixmap is up and running.
83 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
86 * Flag to determine whether vcpu info placement is available on all
87 * VCPUs. We assume it is to start with, and then set it to zero on
88 * the first failure. This is because it can succeed on some VCPUs
89 * and not others, since it can involve hypervisor memory allocation,
90 * or because the guest failed to guarantee all the appropriate
91 * constraints on all VCPUs (ie buffer can't cross a page boundary).
93 * Note that any particular CPU may be using a placed vcpu structure,
94 * but we can only optimise if the all are.
96 * 0: not available, 1: available
98 static int have_vcpu_info_placement
= 1;
100 static void xen_vcpu_setup(int cpu
)
102 struct vcpu_register_vcpu_info info
;
104 struct vcpu_info
*vcpup
;
106 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
107 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
109 if (!have_vcpu_info_placement
)
110 return; /* already tested, not available */
112 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
114 info
.mfn
= arbitrary_virt_to_mfn(vcpup
);
115 info
.offset
= offset_in_page(vcpup
);
117 printk(KERN_DEBUG
"trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
118 cpu
, vcpup
, info
.mfn
, info
.offset
);
120 /* Check to see if the hypervisor will put the vcpu_info
121 structure where we want it, which allows direct access via
122 a percpu-variable. */
123 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
126 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
127 have_vcpu_info_placement
= 0;
129 /* This cpu is using the registered vcpu info, even if
130 later ones fail to. */
131 per_cpu(xen_vcpu
, cpu
) = vcpup
;
133 printk(KERN_DEBUG
"cpu %d using vcpu_info at %p\n",
139 * On restore, set the vcpu placement up again.
140 * If it fails, then we're in a bad state, since
141 * we can't back out from using it...
143 void xen_vcpu_restore(void)
147 for_each_online_cpu(cpu
) {
148 bool other_cpu
= (cpu
!= smp_processor_id());
151 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
154 xen_setup_runstate_info(cpu
);
156 if (have_vcpu_info_placement
)
160 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
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 __read_mostly
unsigned int cpuid_leaf1_edx_mask
= ~0;
179 static __read_mostly
unsigned int cpuid_leaf1_ecx_mask
= ~0;
181 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
182 unsigned int *cx
, unsigned int *dx
)
184 unsigned maskebx
= ~0;
185 unsigned maskecx
= ~0;
186 unsigned maskedx
= ~0;
189 * Mask out inconvenient features, to try and disable as many
190 * unsupported kernel subsystems as possible.
194 maskecx
= cpuid_leaf1_ecx_mask
;
195 maskedx
= cpuid_leaf1_edx_mask
;
199 /* Suppress extended topology stuff */
204 asm(XEN_EMULATE_PREFIX
"cpuid"
209 : "0" (*ax
), "2" (*cx
));
216 static __init
void xen_init_cpuid_mask(void)
218 unsigned int ax
, bx
, cx
, dx
;
220 cpuid_leaf1_edx_mask
=
221 ~((1 << X86_FEATURE_MCE
) | /* disable MCE */
222 (1 << X86_FEATURE_MCA
) | /* disable MCA */
223 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
225 if (!xen_initial_domain())
226 cpuid_leaf1_edx_mask
&=
227 ~((1 << X86_FEATURE_APIC
) | /* disable local APIC */
228 (1 << X86_FEATURE_ACPI
)); /* disable ACPI */
232 xen_cpuid(&ax
, &bx
, &cx
, &dx
);
234 /* cpuid claims we support xsave; try enabling it to see what happens */
235 if (cx
& (1 << (X86_FEATURE_XSAVE
% 32))) {
238 set_in_cr4(X86_CR4_OSXSAVE
);
242 if ((cr4
& X86_CR4_OSXSAVE
) == 0)
243 cpuid_leaf1_ecx_mask
&= ~(1 << (X86_FEATURE_XSAVE
% 32));
245 clear_in_cr4(X86_CR4_OSXSAVE
);
249 static void xen_set_debugreg(int reg
, unsigned long val
)
251 HYPERVISOR_set_debugreg(reg
, val
);
254 static unsigned long xen_get_debugreg(int reg
)
256 return HYPERVISOR_get_debugreg(reg
);
259 static void xen_end_context_switch(struct task_struct
*next
)
262 paravirt_end_context_switch(next
);
265 static unsigned long xen_store_tr(void)
271 * Set the page permissions for a particular virtual address. If the
272 * address is a vmalloc mapping (or other non-linear mapping), then
273 * find the linear mapping of the page and also set its protections to
276 static void set_aliased_prot(void *v
, pgprot_t prot
)
284 ptep
= lookup_address((unsigned long)v
, &level
);
285 BUG_ON(ptep
== NULL
);
287 pfn
= pte_pfn(*ptep
);
288 page
= pfn_to_page(pfn
);
290 pte
= pfn_pte(pfn
, prot
);
292 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
295 if (!PageHighMem(page
)) {
296 void *av
= __va(PFN_PHYS(pfn
));
299 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
305 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
307 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
310 for(i
= 0; i
< entries
; i
+= entries_per_page
)
311 set_aliased_prot(ldt
+ i
, PAGE_KERNEL_RO
);
314 static void xen_free_ldt(struct desc_struct
*ldt
, unsigned entries
)
316 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
319 for(i
= 0; i
< entries
; i
+= entries_per_page
)
320 set_aliased_prot(ldt
+ i
, PAGE_KERNEL
);
323 static void xen_set_ldt(const void *addr
, unsigned entries
)
325 struct mmuext_op
*op
;
326 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
329 op
->cmd
= MMUEXT_SET_LDT
;
330 op
->arg1
.linear_addr
= (unsigned long)addr
;
331 op
->arg2
.nr_ents
= entries
;
333 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
335 xen_mc_issue(PARAVIRT_LAZY_CPU
);
338 static void xen_load_gdt(const struct desc_ptr
*dtr
)
340 unsigned long va
= dtr
->address
;
341 unsigned int size
= dtr
->size
+ 1;
342 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
343 unsigned long frames
[pages
];
347 * A GDT can be up to 64k in size, which corresponds to 8192
348 * 8-byte entries, or 16 4k pages..
351 BUG_ON(size
> 65536);
352 BUG_ON(va
& ~PAGE_MASK
);
354 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
357 unsigned long pfn
, mfn
;
361 * The GDT is per-cpu and is in the percpu data area.
362 * That can be virtually mapped, so we need to do a
363 * page-walk to get the underlying MFN for the
364 * hypercall. The page can also be in the kernel's
365 * linear range, so we need to RO that mapping too.
367 ptep
= lookup_address(va
, &level
);
368 BUG_ON(ptep
== NULL
);
370 pfn
= pte_pfn(*ptep
);
371 mfn
= pfn_to_mfn(pfn
);
372 virt
= __va(PFN_PHYS(pfn
));
376 make_lowmem_page_readonly((void *)va
);
377 make_lowmem_page_readonly(virt
);
380 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
385 * load_gdt for early boot, when the gdt is only mapped once
387 static __init
void xen_load_gdt_boot(const struct desc_ptr
*dtr
)
389 unsigned long va
= dtr
->address
;
390 unsigned int size
= dtr
->size
+ 1;
391 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
392 unsigned long frames
[pages
];
396 * A GDT can be up to 64k in size, which corresponds to 8192
397 * 8-byte entries, or 16 4k pages..
400 BUG_ON(size
> 65536);
401 BUG_ON(va
& ~PAGE_MASK
);
403 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
405 unsigned long pfn
, mfn
;
407 pfn
= virt_to_pfn(va
);
408 mfn
= pfn_to_mfn(pfn
);
410 pte
= pfn_pte(pfn
, PAGE_KERNEL_RO
);
412 if (HYPERVISOR_update_va_mapping((unsigned long)va
, pte
, 0))
418 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
422 static void load_TLS_descriptor(struct thread_struct
*t
,
423 unsigned int cpu
, unsigned int i
)
425 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
426 xmaddr_t maddr
= arbitrary_virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
427 struct multicall_space mc
= __xen_mc_entry(0);
429 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
432 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
435 * XXX sleazy hack: If we're being called in a lazy-cpu zone
436 * and lazy gs handling is enabled, it means we're in a
437 * context switch, and %gs has just been saved. This means we
438 * can zero it out to prevent faults on exit from the
439 * hypervisor if the next process has no %gs. Either way, it
440 * has been saved, and the new value will get loaded properly.
441 * This will go away as soon as Xen has been modified to not
442 * save/restore %gs for normal hypercalls.
444 * On x86_64, this hack is not used for %gs, because gs points
445 * to KERNEL_GS_BASE (and uses it for PDA references), so we
446 * must not zero %gs on x86_64
448 * For x86_64, we need to zero %fs, otherwise we may get an
449 * exception between the new %fs descriptor being loaded and
450 * %fs being effectively cleared at __switch_to().
452 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
462 load_TLS_descriptor(t
, cpu
, 0);
463 load_TLS_descriptor(t
, cpu
, 1);
464 load_TLS_descriptor(t
, cpu
, 2);
466 xen_mc_issue(PARAVIRT_LAZY_CPU
);
470 static void xen_load_gs_index(unsigned int idx
)
472 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
477 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
480 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
481 u64 entry
= *(u64
*)ptr
;
486 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
492 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
493 struct trap_info
*info
)
497 if (val
->type
!= GATE_TRAP
&& val
->type
!= GATE_INTERRUPT
)
500 info
->vector
= vector
;
502 addr
= gate_offset(*val
);
505 * Look for known traps using IST, and substitute them
506 * appropriately. The debugger ones are the only ones we care
507 * about. Xen will handle faults like double_fault and
508 * machine_check, so we should never see them. Warn if
509 * there's an unexpected IST-using fault handler.
511 if (addr
== (unsigned long)debug
)
512 addr
= (unsigned long)xen_debug
;
513 else if (addr
== (unsigned long)int3
)
514 addr
= (unsigned long)xen_int3
;
515 else if (addr
== (unsigned long)stack_segment
)
516 addr
= (unsigned long)xen_stack_segment
;
517 else if (addr
== (unsigned long)double_fault
||
518 addr
== (unsigned long)nmi
) {
519 /* Don't need to handle these */
521 #ifdef CONFIG_X86_MCE
522 } else if (addr
== (unsigned long)machine_check
) {
526 /* Some other trap using IST? */
527 if (WARN_ON(val
->ist
!= 0))
530 #endif /* CONFIG_X86_64 */
531 info
->address
= addr
;
533 info
->cs
= gate_segment(*val
);
534 info
->flags
= val
->dpl
;
535 /* interrupt gates clear IF */
536 if (val
->type
== GATE_INTERRUPT
)
537 info
->flags
|= 1 << 2;
542 /* Locations of each CPU's IDT */
543 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
545 /* Set an IDT entry. If the entry is part of the current IDT, then
547 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
549 unsigned long p
= (unsigned long)&dt
[entrynum
];
550 unsigned long start
, end
;
554 start
= __get_cpu_var(idt_desc
).address
;
555 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
559 native_write_idt_entry(dt
, entrynum
, g
);
561 if (p
>= start
&& (p
+ 8) <= end
) {
562 struct trap_info info
[2];
566 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
567 if (HYPERVISOR_set_trap_table(info
))
574 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
575 struct trap_info
*traps
)
577 unsigned in
, out
, count
;
579 count
= (desc
->size
+1) / sizeof(gate_desc
);
582 for (in
= out
= 0; in
< count
; in
++) {
583 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
585 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
588 traps
[out
].address
= 0;
591 void xen_copy_trap_info(struct trap_info
*traps
)
593 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
595 xen_convert_trap_info(desc
, traps
);
598 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
599 hold a spinlock to protect the static traps[] array (static because
600 it avoids allocation, and saves stack space). */
601 static void xen_load_idt(const struct desc_ptr
*desc
)
603 static DEFINE_SPINLOCK(lock
);
604 static struct trap_info traps
[257];
608 __get_cpu_var(idt_desc
) = *desc
;
610 xen_convert_trap_info(desc
, traps
);
613 if (HYPERVISOR_set_trap_table(traps
))
619 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
620 they're handled differently. */
621 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
622 const void *desc
, int type
)
633 xmaddr_t maddr
= arbitrary_virt_to_machine(&dt
[entry
]);
636 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
646 * Version of write_gdt_entry for use at early boot-time needed to
647 * update an entry as simply as possible.
649 static __init
void xen_write_gdt_entry_boot(struct desc_struct
*dt
, int entry
,
650 const void *desc
, int type
)
659 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
661 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
662 dt
[entry
] = *(struct desc_struct
*)desc
;
668 static void xen_load_sp0(struct tss_struct
*tss
,
669 struct thread_struct
*thread
)
671 struct multicall_space mcs
= xen_mc_entry(0);
672 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
673 xen_mc_issue(PARAVIRT_LAZY_CPU
);
676 static void xen_set_iopl_mask(unsigned mask
)
678 struct physdev_set_iopl set_iopl
;
680 /* Force the change at ring 0. */
681 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
682 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
685 static void xen_io_delay(void)
689 #ifdef CONFIG_X86_LOCAL_APIC
690 static u32
xen_apic_read(u32 reg
)
695 static void xen_apic_write(u32 reg
, u32 val
)
697 /* Warn to see if there's any stray references */
701 static u64
xen_apic_icr_read(void)
706 static void xen_apic_icr_write(u32 low
, u32 id
)
708 /* Warn to see if there's any stray references */
712 static void xen_apic_wait_icr_idle(void)
717 static u32
xen_safe_apic_wait_icr_idle(void)
722 static void set_xen_basic_apic_ops(void)
724 apic
->read
= xen_apic_read
;
725 apic
->write
= xen_apic_write
;
726 apic
->icr_read
= xen_apic_icr_read
;
727 apic
->icr_write
= xen_apic_icr_write
;
728 apic
->wait_icr_idle
= xen_apic_wait_icr_idle
;
729 apic
->safe_wait_icr_idle
= xen_safe_apic_wait_icr_idle
;
735 static void xen_clts(void)
737 struct multicall_space mcs
;
739 mcs
= xen_mc_entry(0);
741 MULTI_fpu_taskswitch(mcs
.mc
, 0);
743 xen_mc_issue(PARAVIRT_LAZY_CPU
);
746 static DEFINE_PER_CPU(unsigned long, xen_cr0_value
);
748 static unsigned long xen_read_cr0(void)
750 unsigned long cr0
= percpu_read(xen_cr0_value
);
752 if (unlikely(cr0
== 0)) {
753 cr0
= native_read_cr0();
754 percpu_write(xen_cr0_value
, cr0
);
760 static void xen_write_cr0(unsigned long cr0
)
762 struct multicall_space mcs
;
764 percpu_write(xen_cr0_value
, cr0
);
766 /* Only pay attention to cr0.TS; everything else is
768 mcs
= xen_mc_entry(0);
770 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
772 xen_mc_issue(PARAVIRT_LAZY_CPU
);
775 static void xen_write_cr4(unsigned long cr4
)
780 native_write_cr4(cr4
);
783 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
794 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
795 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
796 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
799 base
= ((u64
)high
<< 32) | low
;
800 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
808 case MSR_SYSCALL_MASK
:
809 case MSR_IA32_SYSENTER_CS
:
810 case MSR_IA32_SYSENTER_ESP
:
811 case MSR_IA32_SYSENTER_EIP
:
812 /* Fast syscall setup is all done in hypercalls, so
813 these are all ignored. Stub them out here to stop
814 Xen console noise. */
818 ret
= native_write_msr_safe(msr
, low
, high
);
824 void xen_setup_shared_info(void)
826 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
827 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
828 xen_start_info
->shared_info
);
830 HYPERVISOR_shared_info
=
831 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
833 HYPERVISOR_shared_info
=
834 (struct shared_info
*)__va(xen_start_info
->shared_info
);
837 /* In UP this is as good a place as any to set up shared info */
838 xen_setup_vcpu_info_placement();
841 xen_setup_mfn_list_list();
844 /* This is called once we have the cpu_possible_map */
845 void xen_setup_vcpu_info_placement(void)
849 for_each_possible_cpu(cpu
)
852 /* xen_vcpu_setup managed to place the vcpu_info within the
853 percpu area for all cpus, so make use of it */
854 if (have_vcpu_info_placement
) {
855 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
857 pv_irq_ops
.save_fl
= __PV_IS_CALLEE_SAVE(xen_save_fl_direct
);
858 pv_irq_ops
.restore_fl
= __PV_IS_CALLEE_SAVE(xen_restore_fl_direct
);
859 pv_irq_ops
.irq_disable
= __PV_IS_CALLEE_SAVE(xen_irq_disable_direct
);
860 pv_irq_ops
.irq_enable
= __PV_IS_CALLEE_SAVE(xen_irq_enable_direct
);
861 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
865 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
866 unsigned long addr
, unsigned len
)
868 char *start
, *end
, *reloc
;
871 start
= end
= reloc
= NULL
;
873 #define SITE(op, x) \
874 case PARAVIRT_PATCH(op.x): \
875 if (have_vcpu_info_placement) { \
876 start = (char *)xen_##x##_direct; \
877 end = xen_##x##_direct_end; \
878 reloc = xen_##x##_direct_reloc; \
883 SITE(pv_irq_ops
, irq_enable
);
884 SITE(pv_irq_ops
, irq_disable
);
885 SITE(pv_irq_ops
, save_fl
);
886 SITE(pv_irq_ops
, restore_fl
);
890 if (start
== NULL
|| (end
-start
) > len
)
893 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
895 /* Note: because reloc is assigned from something that
896 appears to be an array, gcc assumes it's non-null,
897 but doesn't know its relationship with start and
899 if (reloc
> start
&& reloc
< end
) {
900 int reloc_off
= reloc
- start
;
901 long *relocp
= (long *)(insnbuf
+ reloc_off
);
902 long delta
= start
- (char *)addr
;
910 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
918 static const struct pv_info xen_info __initdata
= {
919 .paravirt_enabled
= 1,
920 .shared_kernel_pmd
= 0,
925 static const struct pv_init_ops xen_init_ops __initdata
= {
929 static const struct pv_time_ops xen_time_ops __initdata
= {
930 .sched_clock
= xen_sched_clock
,
933 static const struct pv_cpu_ops xen_cpu_ops __initdata
= {
936 .set_debugreg
= xen_set_debugreg
,
937 .get_debugreg
= xen_get_debugreg
,
941 .read_cr0
= xen_read_cr0
,
942 .write_cr0
= xen_write_cr0
,
944 .read_cr4
= native_read_cr4
,
945 .read_cr4_safe
= native_read_cr4_safe
,
946 .write_cr4
= xen_write_cr4
,
948 .wbinvd
= native_wbinvd
,
950 .read_msr
= native_read_msr_safe
,
951 .write_msr
= xen_write_msr_safe
,
952 .read_tsc
= native_read_tsc
,
953 .read_pmc
= native_read_pmc
,
956 .irq_enable_sysexit
= xen_sysexit
,
958 .usergs_sysret32
= xen_sysret32
,
959 .usergs_sysret64
= xen_sysret64
,
962 .load_tr_desc
= paravirt_nop
,
963 .set_ldt
= xen_set_ldt
,
964 .load_gdt
= xen_load_gdt
,
965 .load_idt
= xen_load_idt
,
966 .load_tls
= xen_load_tls
,
968 .load_gs_index
= xen_load_gs_index
,
971 .alloc_ldt
= xen_alloc_ldt
,
972 .free_ldt
= xen_free_ldt
,
974 .store_gdt
= native_store_gdt
,
975 .store_idt
= native_store_idt
,
976 .store_tr
= xen_store_tr
,
978 .write_ldt_entry
= xen_write_ldt_entry
,
979 .write_gdt_entry
= xen_write_gdt_entry
,
980 .write_idt_entry
= xen_write_idt_entry
,
981 .load_sp0
= xen_load_sp0
,
983 .set_iopl_mask
= xen_set_iopl_mask
,
984 .io_delay
= xen_io_delay
,
986 /* Xen takes care of %gs when switching to usermode for us */
987 .swapgs
= paravirt_nop
,
989 .start_context_switch
= paravirt_start_context_switch
,
990 .end_context_switch
= xen_end_context_switch
,
993 static const struct pv_apic_ops xen_apic_ops __initdata
= {
994 #ifdef CONFIG_X86_LOCAL_APIC
995 .startup_ipi_hook
= paravirt_nop
,
999 static void xen_reboot(int reason
)
1001 struct sched_shutdown r
= { .reason
= reason
};
1007 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1011 static void xen_restart(char *msg
)
1013 xen_reboot(SHUTDOWN_reboot
);
1016 static void xen_emergency_restart(void)
1018 xen_reboot(SHUTDOWN_reboot
);
1021 static void xen_machine_halt(void)
1023 xen_reboot(SHUTDOWN_poweroff
);
1026 static void xen_crash_shutdown(struct pt_regs
*regs
)
1028 xen_reboot(SHUTDOWN_crash
);
1031 static const struct machine_ops __initdata xen_machine_ops
= {
1032 .restart
= xen_restart
,
1033 .halt
= xen_machine_halt
,
1034 .power_off
= xen_machine_halt
,
1035 .shutdown
= xen_machine_halt
,
1036 .crash_shutdown
= xen_crash_shutdown
,
1037 .emergency_restart
= xen_emergency_restart
,
1041 * Set up the GDT and segment registers for -fstack-protector. Until
1042 * we do this, we have to be careful not to call any stack-protected
1043 * function, which is most of the kernel.
1045 static void __init
xen_setup_stackprotector(void)
1047 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry_boot
;
1048 pv_cpu_ops
.load_gdt
= xen_load_gdt_boot
;
1050 setup_stack_canary_segment(0);
1051 switch_to_new_gdt(0);
1053 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry
;
1054 pv_cpu_ops
.load_gdt
= xen_load_gdt
;
1057 /* First C function to be called on Xen boot */
1058 asmlinkage
void __init
xen_start_kernel(void)
1062 if (!xen_start_info
)
1065 xen_domain_type
= XEN_PV_DOMAIN
;
1067 /* Install Xen paravirt ops */
1069 pv_init_ops
= xen_init_ops
;
1070 pv_time_ops
= xen_time_ops
;
1071 pv_cpu_ops
= xen_cpu_ops
;
1072 pv_apic_ops
= xen_apic_ops
;
1074 x86_init
.resources
.memory_setup
= xen_memory_setup
;
1075 x86_init
.oem
.arch_setup
= xen_arch_setup
;
1076 x86_init
.oem
.banner
= xen_banner
;
1078 x86_init
.timers
.timer_init
= xen_time_init
;
1079 x86_init
.timers
.setup_percpu_clockev
= x86_init_noop
;
1080 x86_cpuinit
.setup_percpu_clockev
= x86_init_noop
;
1082 x86_platform
.calibrate_tsc
= xen_tsc_khz
;
1083 x86_platform
.get_wallclock
= xen_get_wallclock
;
1084 x86_platform
.set_wallclock
= xen_set_wallclock
;
1087 * Set up some pagetable state before starting to set any ptes.
1092 /* Prevent unwanted bits from being set in PTEs. */
1093 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1094 if (!xen_initial_domain())
1095 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1097 __supported_pte_mask
|= _PAGE_IOMAP
;
1100 * Prevent page tables from being allocated in highmem, even
1101 * if CONFIG_HIGHPTE is enabled.
1103 __userpte_alloc_gfp
&= ~__GFP_HIGHMEM
;
1105 /* Work out if we support NX */
1108 xen_setup_features();
1111 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1112 xen_build_dynamic_phys_to_machine();
1115 * Set up kernel GDT and segment registers, mainly so that
1116 * -fstack-protector code can be executed.
1118 xen_setup_stackprotector();
1121 xen_init_cpuid_mask();
1123 #ifdef CONFIG_X86_LOCAL_APIC
1125 * set up the basic apic ops.
1127 set_xen_basic_apic_ops();
1130 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1131 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1132 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1135 machine_ops
= xen_machine_ops
;
1138 * The only reliable way to retain the initial address of the
1139 * percpu gdt_page is to remember it here, so we can go and
1140 * mark it RW later, when the initial percpu area is freed.
1142 xen_initial_gdt
= &per_cpu(gdt_page
, 0);
1146 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1148 if (!xen_initial_domain())
1149 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1151 __supported_pte_mask
|= _PAGE_IOMAP
;
1152 /* Don't do the full vcpu_info placement stuff until we have a
1153 possible map and a non-dummy shared_info. */
1154 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1156 local_irq_disable();
1157 early_boot_irqs_off();
1159 xen_raw_console_write("mapping kernel into physical memory\n");
1160 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
1164 /* keep using Xen gdt for now; no urgent need to change it */
1166 #ifdef CONFIG_X86_32
1167 pv_info
.kernel_rpl
= 1;
1168 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1169 pv_info
.kernel_rpl
= 0;
1171 pv_info
.kernel_rpl
= 0;
1174 /* set the limit of our address space */
1177 #ifdef CONFIG_X86_32
1178 /* set up basic CPUID stuff */
1179 cpu_detect(&new_cpu_data
);
1180 new_cpu_data
.hard_math
= 1;
1181 new_cpu_data
.wp_works_ok
= 1;
1182 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1185 /* Poke various useful things into boot_params */
1186 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1187 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1188 ? __pa(xen_start_info
->mod_start
) : 0;
1189 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1190 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1192 if (!xen_initial_domain()) {
1193 add_preferred_console("xenboot", 0, NULL
);
1194 add_preferred_console("tty", 0, NULL
);
1195 add_preferred_console("hvc", 0, NULL
);
1197 /* Make sure ACS will be enabled */
1202 xen_raw_console_write("about to get started...\n");
1204 xen_setup_runstate_info(0);
1206 /* Start the world */
1207 #ifdef CONFIG_X86_32
1208 i386_start_kernel();
1210 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));