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/smp.h>
30 #include <xen/interface/xen.h>
31 #include <xen/interface/physdev.h>
32 #include <xen/interface/vcpu.h>
33 #include <xen/features.h>
36 #include <asm/paravirt.h>
38 #include <asm/xen/hypercall.h>
39 #include <asm/xen/hypervisor.h>
40 #include <asm/fixmap.h>
41 #include <asm/processor.h>
42 #include <asm/setup.h>
44 #include <asm/pgtable.h>
45 #include <asm/tlbflush.h>
49 #include "multicalls.h"
51 EXPORT_SYMBOL_GPL(hypercall_page
);
53 DEFINE_PER_CPU(enum paravirt_lazy_mode
, xen_lazy_mode
);
55 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
56 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
57 DEFINE_PER_CPU(unsigned long, xen_cr3
);
59 struct start_info
*xen_start_info
;
60 EXPORT_SYMBOL_GPL(xen_start_info
);
62 void xen_vcpu_setup(int cpu
)
64 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
67 static void __init
xen_banner(void)
69 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
71 printk(KERN_INFO
"Hypervisor signature: %s\n", xen_start_info
->magic
);
74 static void xen_cpuid(unsigned int *eax
, unsigned int *ebx
,
75 unsigned int *ecx
, unsigned int *edx
)
77 unsigned maskedx
= ~0;
80 * Mask out inconvenient features, to try and disable as many
81 * unsupported kernel subsystems as possible.
84 maskedx
= ~((1 << X86_FEATURE_APIC
) | /* disable APIC */
85 (1 << X86_FEATURE_ACPI
) | /* disable ACPI */
86 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
88 asm(XEN_EMULATE_PREFIX
"cpuid"
93 : "0" (*eax
), "2" (*ecx
));
97 static void xen_set_debugreg(int reg
, unsigned long val
)
99 HYPERVISOR_set_debugreg(reg
, val
);
102 static unsigned long xen_get_debugreg(int reg
)
104 return HYPERVISOR_get_debugreg(reg
);
107 static unsigned long xen_save_fl(void)
109 struct vcpu_info
*vcpu
;
112 vcpu
= x86_read_percpu(xen_vcpu
);
114 /* flag has opposite sense of mask */
115 flags
= !vcpu
->evtchn_upcall_mask
;
117 /* convert to IF type flag
121 return (-flags
) & X86_EFLAGS_IF
;
124 static void xen_restore_fl(unsigned long flags
)
126 struct vcpu_info
*vcpu
;
128 /* convert from IF type flag */
129 flags
= !(flags
& X86_EFLAGS_IF
);
131 /* There's a one instruction preempt window here. We need to
132 make sure we're don't switch CPUs between getting the vcpu
133 pointer and updating the mask. */
135 vcpu
= x86_read_percpu(xen_vcpu
);
136 vcpu
->evtchn_upcall_mask
= flags
;
137 preempt_enable_no_resched();
139 /* Doesn't matter if we get preempted here, because any
140 pending event will get dealt with anyway. */
143 preempt_check_resched();
144 barrier(); /* unmask then check (avoid races) */
145 if (unlikely(vcpu
->evtchn_upcall_pending
))
146 force_evtchn_callback();
150 static void xen_irq_disable(void)
152 /* There's a one instruction preempt window here. We need to
153 make sure we're don't switch CPUs between getting the vcpu
154 pointer and updating the mask. */
156 x86_read_percpu(xen_vcpu
)->evtchn_upcall_mask
= 1;
157 preempt_enable_no_resched();
160 static void xen_irq_enable(void)
162 struct vcpu_info
*vcpu
;
164 /* There's a one instruction preempt window here. We need to
165 make sure we're don't switch CPUs between getting the vcpu
166 pointer and updating the mask. */
168 vcpu
= x86_read_percpu(xen_vcpu
);
169 vcpu
->evtchn_upcall_mask
= 0;
170 preempt_enable_no_resched();
172 /* Doesn't matter if we get preempted here, because any
173 pending event will get dealt with anyway. */
175 barrier(); /* unmask then check (avoid races) */
176 if (unlikely(vcpu
->evtchn_upcall_pending
))
177 force_evtchn_callback();
180 static void xen_safe_halt(void)
182 /* Blocking includes an implicit local_irq_enable(). */
183 if (HYPERVISOR_sched_op(SCHEDOP_block
, 0) != 0)
187 static void xen_halt(void)
190 HYPERVISOR_vcpu_op(VCPUOP_down
, smp_processor_id(), NULL
);
195 static void xen_set_lazy_mode(enum paravirt_lazy_mode mode
)
197 BUG_ON(preemptible());
200 case PARAVIRT_LAZY_NONE
:
201 BUG_ON(x86_read_percpu(xen_lazy_mode
) == PARAVIRT_LAZY_NONE
);
204 case PARAVIRT_LAZY_MMU
:
205 case PARAVIRT_LAZY_CPU
:
206 BUG_ON(x86_read_percpu(xen_lazy_mode
) != PARAVIRT_LAZY_NONE
);
209 case PARAVIRT_LAZY_FLUSH
:
210 /* flush if necessary, but don't change state */
211 if (x86_read_percpu(xen_lazy_mode
) != PARAVIRT_LAZY_NONE
)
217 x86_write_percpu(xen_lazy_mode
, mode
);
220 static unsigned long xen_store_tr(void)
225 static void xen_set_ldt(const void *addr
, unsigned entries
)
227 unsigned long linear_addr
= (unsigned long)addr
;
228 struct mmuext_op
*op
;
229 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
232 op
->cmd
= MMUEXT_SET_LDT
;
234 /* ldt my be vmalloced, use arbitrary_virt_to_machine */
236 maddr
= arbitrary_virt_to_machine((unsigned long)addr
);
237 linear_addr
= (unsigned long)maddr
.maddr
;
239 op
->arg1
.linear_addr
= linear_addr
;
240 op
->arg2
.nr_ents
= entries
;
242 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
244 xen_mc_issue(PARAVIRT_LAZY_CPU
);
247 static void xen_load_gdt(const struct Xgt_desc_struct
*dtr
)
249 unsigned long *frames
;
250 unsigned long va
= dtr
->address
;
251 unsigned int size
= dtr
->size
+ 1;
252 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
254 struct multicall_space mcs
;
256 /* A GDT can be up to 64k in size, which corresponds to 8192
257 8-byte entries, or 16 4k pages.. */
259 BUG_ON(size
> 65536);
260 BUG_ON(va
& ~PAGE_MASK
);
262 mcs
= xen_mc_entry(sizeof(*frames
) * pages
);
265 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
266 frames
[f
] = virt_to_mfn(va
);
267 make_lowmem_page_readonly((void *)va
);
270 MULTI_set_gdt(mcs
.mc
, frames
, size
/ sizeof(struct desc_struct
));
272 xen_mc_issue(PARAVIRT_LAZY_CPU
);
275 static void load_TLS_descriptor(struct thread_struct
*t
,
276 unsigned int cpu
, unsigned int i
)
278 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
279 xmaddr_t maddr
= virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
280 struct multicall_space mc
= __xen_mc_entry(0);
282 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
285 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
289 load_TLS_descriptor(t
, cpu
, 0);
290 load_TLS_descriptor(t
, cpu
, 1);
291 load_TLS_descriptor(t
, cpu
, 2);
293 xen_mc_issue(PARAVIRT_LAZY_CPU
);
296 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
297 * it means we're in a context switch, and %gs has just been
298 * saved. This means we can zero it out to prevent faults on
299 * exit from the hypervisor if the next process has no %gs.
300 * Either way, it has been saved, and the new value will get
301 * loaded properly. This will go away as soon as Xen has been
302 * modified to not save/restore %gs for normal hypercalls.
304 if (xen_get_lazy_mode() == PARAVIRT_LAZY_CPU
)
308 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
311 unsigned long lp
= (unsigned long)&dt
[entrynum
];
312 xmaddr_t mach_lp
= virt_to_machine(lp
);
313 u64 entry
= (u64
)high
<< 32 | low
;
318 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
324 static int cvt_gate_to_trap(int vector
, u32 low
, u32 high
,
325 struct trap_info
*info
)
329 type
= (high
>> 8) & 0x1f;
330 dpl
= (high
>> 13) & 3;
332 if (type
!= 0xf && type
!= 0xe)
335 info
->vector
= vector
;
336 info
->address
= (high
& 0xffff0000) | (low
& 0x0000ffff);
337 info
->cs
= low
>> 16;
339 /* interrupt gates clear IF */
346 /* Locations of each CPU's IDT */
347 static DEFINE_PER_CPU(struct Xgt_desc_struct
, idt_desc
);
349 /* Set an IDT entry. If the entry is part of the current IDT, then
351 static void xen_write_idt_entry(struct desc_struct
*dt
, int entrynum
,
354 unsigned long p
= (unsigned long)&dt
[entrynum
];
355 unsigned long start
, end
;
359 start
= __get_cpu_var(idt_desc
).address
;
360 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
364 write_dt_entry(dt
, entrynum
, low
, high
);
366 if (p
>= start
&& (p
+ 8) <= end
) {
367 struct trap_info info
[2];
371 if (cvt_gate_to_trap(entrynum
, low
, high
, &info
[0]))
372 if (HYPERVISOR_set_trap_table(info
))
379 static void xen_convert_trap_info(const struct Xgt_desc_struct
*desc
,
380 struct trap_info
*traps
)
382 unsigned in
, out
, count
;
384 count
= (desc
->size
+1) / 8;
387 for (in
= out
= 0; in
< count
; in
++) {
388 const u32
*entry
= (u32
*)(desc
->address
+ in
* 8);
390 if (cvt_gate_to_trap(in
, entry
[0], entry
[1], &traps
[out
]))
393 traps
[out
].address
= 0;
396 void xen_copy_trap_info(struct trap_info
*traps
)
398 const struct Xgt_desc_struct
*desc
= &__get_cpu_var(idt_desc
);
400 xen_convert_trap_info(desc
, traps
);
403 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
404 hold a spinlock to protect the static traps[] array (static because
405 it avoids allocation, and saves stack space). */
406 static void xen_load_idt(const struct Xgt_desc_struct
*desc
)
408 static DEFINE_SPINLOCK(lock
);
409 static struct trap_info traps
[257];
413 __get_cpu_var(idt_desc
) = *desc
;
415 xen_convert_trap_info(desc
, traps
);
418 if (HYPERVISOR_set_trap_table(traps
))
424 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
425 they're handled differently. */
426 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
431 switch ((high
>> 8) & 0xff) {
438 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
439 u64 desc
= (u64
)high
<< 32 | low
;
442 if (HYPERVISOR_update_descriptor(maddr
.maddr
, desc
))
451 static void xen_load_esp0(struct tss_struct
*tss
,
452 struct thread_struct
*thread
)
454 struct multicall_space mcs
= xen_mc_entry(0);
455 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->esp0
);
456 xen_mc_issue(PARAVIRT_LAZY_CPU
);
459 static void xen_set_iopl_mask(unsigned mask
)
461 struct physdev_set_iopl set_iopl
;
463 /* Force the change at ring 0. */
464 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
465 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
468 static void xen_io_delay(void)
472 #ifdef CONFIG_X86_LOCAL_APIC
473 static unsigned long xen_apic_read(unsigned long reg
)
478 static void xen_apic_write(unsigned long reg
, unsigned long val
)
480 /* Warn to see if there's any stray references */
485 static void xen_flush_tlb(void)
487 struct mmuext_op
*op
;
488 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
491 op
->cmd
= MMUEXT_TLB_FLUSH_LOCAL
;
492 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
494 xen_mc_issue(PARAVIRT_LAZY_MMU
);
497 static void xen_flush_tlb_single(unsigned long addr
)
499 struct mmuext_op
*op
;
500 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
503 op
->cmd
= MMUEXT_INVLPG_LOCAL
;
504 op
->arg1
.linear_addr
= addr
& PAGE_MASK
;
505 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
507 xen_mc_issue(PARAVIRT_LAZY_MMU
);
510 static void xen_flush_tlb_others(const cpumask_t
*cpus
, struct mm_struct
*mm
,
517 cpumask_t cpumask
= *cpus
;
518 struct multicall_space mcs
;
521 * A couple of (to be removed) sanity checks:
523 * - current CPU must not be in mask
524 * - mask must exist :)
526 BUG_ON(cpus_empty(cpumask
));
527 BUG_ON(cpu_isset(smp_processor_id(), cpumask
));
530 /* If a CPU which we ran on has gone down, OK. */
531 cpus_and(cpumask
, cpumask
, cpu_online_map
);
532 if (cpus_empty(cpumask
))
535 mcs
= xen_mc_entry(sizeof(*args
));
537 args
->mask
= cpumask
;
538 args
->op
.arg2
.vcpumask
= &args
->mask
;
540 if (va
== TLB_FLUSH_ALL
) {
541 args
->op
.cmd
= MMUEXT_TLB_FLUSH_MULTI
;
543 args
->op
.cmd
= MMUEXT_INVLPG_MULTI
;
544 args
->op
.arg1
.linear_addr
= va
;
547 MULTI_mmuext_op(mcs
.mc
, &args
->op
, 1, NULL
, DOMID_SELF
);
549 xen_mc_issue(PARAVIRT_LAZY_MMU
);
552 static unsigned long xen_read_cr2(void)
554 return x86_read_percpu(xen_vcpu
)->arch
.cr2
;
557 static void xen_write_cr4(unsigned long cr4
)
559 /* never allow TSC to be disabled */
560 native_write_cr4(cr4
& ~X86_CR4_TSD
);
563 static unsigned long xen_read_cr3(void)
565 return x86_read_percpu(xen_cr3
);
568 static void xen_write_cr3(unsigned long cr3
)
570 BUG_ON(preemptible());
572 if (cr3
== x86_read_percpu(xen_cr3
)) {
573 /* just a simple tlb flush */
578 x86_write_percpu(xen_cr3
, cr3
);
582 struct mmuext_op
*op
;
583 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
584 unsigned long mfn
= pfn_to_mfn(PFN_DOWN(cr3
));
587 op
->cmd
= MMUEXT_NEW_BASEPTR
;
590 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
592 xen_mc_issue(PARAVIRT_LAZY_CPU
);
596 /* Early in boot, while setting up the initial pagetable, assume
597 everything is pinned. */
598 static __init
void xen_alloc_pt_init(struct mm_struct
*mm
, u32 pfn
)
600 BUG_ON(mem_map
); /* should only be used early */
601 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
604 /* This needs to make sure the new pte page is pinned iff its being
605 attached to a pinned pagetable. */
606 static void xen_alloc_pt(struct mm_struct
*mm
, u32 pfn
)
608 struct page
*page
= pfn_to_page(pfn
);
610 if (PagePinned(virt_to_page(mm
->pgd
))) {
613 if (!PageHighMem(page
))
614 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
616 /* make sure there are no stray mappings of
622 /* This should never happen until we're OK to use struct page */
623 static void xen_release_pt(u32 pfn
)
625 struct page
*page
= pfn_to_page(pfn
);
627 if (PagePinned(page
)) {
628 if (!PageHighMem(page
))
629 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
633 #ifdef CONFIG_HIGHPTE
634 static void *xen_kmap_atomic_pte(struct page
*page
, enum km_type type
)
636 pgprot_t prot
= PAGE_KERNEL
;
638 if (PagePinned(page
))
639 prot
= PAGE_KERNEL_RO
;
641 if (0 && PageHighMem(page
))
642 printk("mapping highpte %lx type %d prot %s\n",
643 page_to_pfn(page
), type
,
644 (unsigned long)pgprot_val(prot
) & _PAGE_RW
? "WRITE" : "READ");
646 return kmap_atomic_prot(page
, type
, prot
);
650 static __init pte_t
mask_rw_pte(pte_t
*ptep
, pte_t pte
)
652 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
653 if (pte_val_ma(*ptep
) & _PAGE_PRESENT
)
654 pte
= __pte_ma(((pte_val_ma(*ptep
) & _PAGE_RW
) | ~_PAGE_RW
) &
660 /* Init-time set_pte while constructing initial pagetables, which
661 doesn't allow RO pagetable pages to be remapped RW */
662 static __init
void xen_set_pte_init(pte_t
*ptep
, pte_t pte
)
664 pte
= mask_rw_pte(ptep
, pte
);
666 xen_set_pte(ptep
, pte
);
669 static __init
void xen_pagetable_setup_start(pgd_t
*base
)
671 pgd_t
*xen_pgd
= (pgd_t
*)xen_start_info
->pt_base
;
673 /* special set_pte for pagetable initialization */
674 paravirt_ops
.set_pte
= xen_set_pte_init
;
678 * copy top-level of Xen-supplied pagetable into place. For
679 * !PAE we can use this as-is, but for PAE it is a stand-in
680 * while we copy the pmd pages.
682 memcpy(base
, xen_pgd
, PTRS_PER_PGD
* sizeof(pgd_t
));
684 if (PTRS_PER_PMD
> 1) {
687 * For PAE, need to allocate new pmds, rather than
688 * share Xen's, since Xen doesn't like pmd's being
689 * shared between address spaces.
691 for (i
= 0; i
< PTRS_PER_PGD
; i
++) {
692 if (pgd_val_ma(xen_pgd
[i
]) & _PAGE_PRESENT
) {
693 pmd_t
*pmd
= (pmd_t
*)alloc_bootmem_low_pages(PAGE_SIZE
);
695 memcpy(pmd
, (void *)pgd_page_vaddr(xen_pgd
[i
]),
698 make_lowmem_page_readonly(pmd
);
700 set_pgd(&base
[i
], __pgd(1 + __pa(pmd
)));
706 /* make sure zero_page is mapped RO so we can use it in pagetables */
707 make_lowmem_page_readonly(empty_zero_page
);
708 make_lowmem_page_readonly(base
);
710 * Switch to new pagetable. This is done before
711 * pagetable_init has done anything so that the new pages
712 * added to the table can be prepared properly for Xen.
714 xen_write_cr3(__pa(base
));
717 static __init
void xen_pagetable_setup_done(pgd_t
*base
)
719 /* This will work as long as patching hasn't happened yet
721 paravirt_ops
.alloc_pt
= xen_alloc_pt
;
722 paravirt_ops
.set_pte
= xen_set_pte
;
724 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
726 * Create a mapping for the shared info page.
727 * Should be set_fixmap(), but shared_info is a machine
728 * address with no corresponding pseudo-phys address.
730 set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP
),
731 PFN_DOWN(xen_start_info
->shared_info
),
734 HYPERVISOR_shared_info
=
735 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
738 HYPERVISOR_shared_info
=
739 (struct shared_info
*)__va(xen_start_info
->shared_info
);
741 /* Actually pin the pagetable down, but we can't set PG_pinned
742 yet because the page structures don't exist yet. */
745 #ifdef CONFIG_X86_PAE
746 op
.cmd
= MMUEXT_PIN_L3_TABLE
;
748 op
.cmd
= MMUEXT_PIN_L3_TABLE
;
750 op
.arg1
.mfn
= pfn_to_mfn(PFN_DOWN(__pa(base
)));
751 if (HYPERVISOR_mmuext_op(&op
, 1, NULL
, DOMID_SELF
))
755 xen_vcpu_setup(smp_processor_id());
758 static const struct paravirt_ops xen_paravirt_ops __initdata
= {
759 .paravirt_enabled
= 1,
760 .shared_kernel_pmd
= 0,
763 .banner
= xen_banner
,
765 .patch
= paravirt_patch_default
,
767 .memory_setup
= xen_memory_setup
,
768 .arch_setup
= xen_arch_setup
,
769 .init_IRQ
= xen_init_IRQ
,
770 .post_allocator_init
= xen_mark_init_mm_pinned
,
772 .time_init
= xen_time_init
,
773 .set_wallclock
= xen_set_wallclock
,
774 .get_wallclock
= xen_get_wallclock
,
775 .get_cpu_khz
= xen_cpu_khz
,
776 .sched_clock
= xen_sched_clock
,
780 .set_debugreg
= xen_set_debugreg
,
781 .get_debugreg
= xen_get_debugreg
,
785 .read_cr0
= native_read_cr0
,
786 .write_cr0
= native_write_cr0
,
788 .read_cr2
= xen_read_cr2
,
789 .write_cr2
= native_write_cr2
,
791 .read_cr3
= xen_read_cr3
,
792 .write_cr3
= xen_write_cr3
,
794 .read_cr4
= native_read_cr4
,
795 .read_cr4_safe
= native_read_cr4_safe
,
796 .write_cr4
= xen_write_cr4
,
798 .save_fl
= xen_save_fl
,
799 .restore_fl
= xen_restore_fl
,
800 .irq_disable
= xen_irq_disable
,
801 .irq_enable
= xen_irq_enable
,
802 .safe_halt
= xen_safe_halt
,
804 .wbinvd
= native_wbinvd
,
806 .read_msr
= native_read_msr_safe
,
807 .write_msr
= native_write_msr_safe
,
808 .read_tsc
= native_read_tsc
,
809 .read_pmc
= native_read_pmc
,
811 .iret
= (void *)&hypercall_page
[__HYPERVISOR_iret
],
812 .irq_enable_sysexit
= NULL
, /* never called */
814 .load_tr_desc
= paravirt_nop
,
815 .set_ldt
= xen_set_ldt
,
816 .load_gdt
= xen_load_gdt
,
817 .load_idt
= xen_load_idt
,
818 .load_tls
= xen_load_tls
,
820 .store_gdt
= native_store_gdt
,
821 .store_idt
= native_store_idt
,
822 .store_tr
= xen_store_tr
,
824 .write_ldt_entry
= xen_write_ldt_entry
,
825 .write_gdt_entry
= xen_write_gdt_entry
,
826 .write_idt_entry
= xen_write_idt_entry
,
827 .load_esp0
= xen_load_esp0
,
829 .set_iopl_mask
= xen_set_iopl_mask
,
830 .io_delay
= xen_io_delay
,
832 #ifdef CONFIG_X86_LOCAL_APIC
833 .apic_write
= xen_apic_write
,
834 .apic_write_atomic
= xen_apic_write
,
835 .apic_read
= xen_apic_read
,
836 .setup_boot_clock
= paravirt_nop
,
837 .setup_secondary_clock
= paravirt_nop
,
838 .startup_ipi_hook
= paravirt_nop
,
841 .flush_tlb_user
= xen_flush_tlb
,
842 .flush_tlb_kernel
= xen_flush_tlb
,
843 .flush_tlb_single
= xen_flush_tlb_single
,
844 .flush_tlb_others
= xen_flush_tlb_others
,
846 .pte_update
= paravirt_nop
,
847 .pte_update_defer
= paravirt_nop
,
849 .pagetable_setup_start
= xen_pagetable_setup_start
,
850 .pagetable_setup_done
= xen_pagetable_setup_done
,
852 .alloc_pt
= xen_alloc_pt_init
,
853 .release_pt
= xen_release_pt
,
854 .alloc_pd
= paravirt_nop
,
855 .alloc_pd_clone
= paravirt_nop
,
856 .release_pd
= paravirt_nop
,
858 #ifdef CONFIG_HIGHPTE
859 .kmap_atomic_pte
= xen_kmap_atomic_pte
,
862 .set_pte
= NULL
, /* see xen_pagetable_setup_* */
863 .set_pte_at
= xen_set_pte_at
,
864 .set_pmd
= xen_set_pmd
,
866 .pte_val
= xen_pte_val
,
867 .pgd_val
= xen_pgd_val
,
869 .make_pte
= xen_make_pte
,
870 .make_pgd
= xen_make_pgd
,
872 #ifdef CONFIG_X86_PAE
873 .set_pte_atomic
= xen_set_pte_atomic
,
874 .set_pte_present
= xen_set_pte_at
,
875 .set_pud
= xen_set_pud
,
876 .pte_clear
= xen_pte_clear
,
877 .pmd_clear
= xen_pmd_clear
,
879 .make_pmd
= xen_make_pmd
,
880 .pmd_val
= xen_pmd_val
,
883 .activate_mm
= xen_activate_mm
,
884 .dup_mmap
= xen_dup_mmap
,
885 .exit_mmap
= xen_exit_mmap
,
887 .set_lazy_mode
= xen_set_lazy_mode
,
891 static const struct smp_ops xen_smp_ops __initdata
= {
892 .smp_prepare_boot_cpu
= xen_smp_prepare_boot_cpu
,
893 .smp_prepare_cpus
= xen_smp_prepare_cpus
,
894 .cpu_up
= xen_cpu_up
,
895 .smp_cpus_done
= xen_smp_cpus_done
,
897 .smp_send_stop
= xen_smp_send_stop
,
898 .smp_send_reschedule
= xen_smp_send_reschedule
,
899 .smp_call_function_mask
= xen_smp_call_function_mask
,
901 #endif /* CONFIG_SMP */
903 /* First C function to be called on Xen boot */
904 asmlinkage
void __init
xen_start_kernel(void)
911 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3.0", 7) != 0);
913 /* Install Xen paravirt ops */
914 paravirt_ops
= xen_paravirt_ops
;
916 smp_ops
= xen_smp_ops
;
919 xen_setup_features();
922 if (!xen_feature(XENFEAT_auto_translated_physmap
))
923 phys_to_machine_mapping
= (unsigned long *)xen_start_info
->mfn_list
;
925 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
927 init_pg_tables_end
= __pa(pgd
) + xen_start_info
->nr_pt_frames
*PAGE_SIZE
;
929 init_mm
.pgd
= pgd
; /* use the Xen pagetables to start */
931 /* keep using Xen gdt for now; no urgent need to change it */
933 x86_write_percpu(xen_cr3
, __pa(pgd
));
936 paravirt_ops
.kernel_rpl
= 1;
937 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
938 paravirt_ops
.kernel_rpl
= 0;
940 /* set the limit of our address space */
941 reserve_top_address(-HYPERVISOR_VIRT_START
+ 2 * PAGE_SIZE
);
943 /* set up basic CPUID stuff */
944 cpu_detect(&new_cpu_data
);
945 new_cpu_data
.hard_math
= 1;
946 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
948 /* Poke various useful things into boot_params */
949 LOADER_TYPE
= (9 << 4) | 0;
950 INITRD_START
= xen_start_info
->mod_start
? __pa(xen_start_info
->mod_start
) : 0;
951 INITRD_SIZE
= xen_start_info
->mod_len
;
953 /* Start the world */