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
);
65 * Identity map, in addition to plain kernel map. This needs to be
66 * large enough to allocate page table pages to allocate the rest.
67 * Each page can map 2MB.
69 static pte_t level1_ident_pgt
[PTRS_PER_PTE
* 4] __page_aligned_bss
;
72 /* l3 pud for userspace vsyscall mapping */
73 static pud_t level3_user_vsyscall
[PTRS_PER_PUD
] __page_aligned_bss
;
74 #endif /* CONFIG_X86_64 */
77 * Note about cr3 (pagetable base) values:
79 * xen_cr3 contains the current logical cr3 value; it contains the
80 * last set cr3. This may not be the current effective cr3, because
81 * its update may be being lazily deferred. However, a vcpu looking
82 * at its own cr3 can use this value knowing that it everything will
85 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
86 * hypercall to set the vcpu cr3 is complete (so it may be a little
87 * out of date, but it will never be set early). If one vcpu is
88 * looking at another vcpu's cr3 value, it should use this variable.
90 DEFINE_PER_CPU(unsigned long, xen_cr3
); /* cr3 stored as physaddr */
91 DEFINE_PER_CPU(unsigned long, xen_current_cr3
); /* actual vcpu cr3 */
93 struct start_info
*xen_start_info
;
94 EXPORT_SYMBOL_GPL(xen_start_info
);
96 struct shared_info xen_dummy_shared_info
;
99 * Point at some empty memory to start with. We map the real shared_info
100 * page as soon as fixmap is up and running.
102 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
105 * Flag to determine whether vcpu info placement is available on all
106 * VCPUs. We assume it is to start with, and then set it to zero on
107 * the first failure. This is because it can succeed on some VCPUs
108 * and not others, since it can involve hypervisor memory allocation,
109 * or because the guest failed to guarantee all the appropriate
110 * constraints on all VCPUs (ie buffer can't cross a page boundary).
112 * Note that any particular CPU may be using a placed vcpu structure,
113 * but we can only optimise if the all are.
115 * 0: not available, 1: available
117 static int have_vcpu_info_placement
=
126 static void xen_vcpu_setup(int cpu
)
128 struct vcpu_register_vcpu_info info
;
130 struct vcpu_info
*vcpup
;
132 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
133 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
135 if (!have_vcpu_info_placement
)
136 return; /* already tested, not available */
138 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
140 info
.mfn
= virt_to_mfn(vcpup
);
141 info
.offset
= offset_in_page(vcpup
);
143 printk(KERN_DEBUG
"trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
144 cpu
, vcpup
, info
.mfn
, info
.offset
);
146 /* Check to see if the hypervisor will put the vcpu_info
147 structure where we want it, which allows direct access via
148 a percpu-variable. */
149 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
152 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
153 have_vcpu_info_placement
= 0;
155 /* This cpu is using the registered vcpu info, even if
156 later ones fail to. */
157 per_cpu(xen_vcpu
, cpu
) = vcpup
;
159 printk(KERN_DEBUG
"cpu %d using vcpu_info at %p\n",
165 * On restore, set the vcpu placement up again.
166 * If it fails, then we're in a bad state, since
167 * we can't back out from using it...
169 void xen_vcpu_restore(void)
171 if (have_vcpu_info_placement
) {
174 for_each_online_cpu(cpu
) {
175 bool other_cpu
= (cpu
!= smp_processor_id());
178 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
184 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
188 BUG_ON(!have_vcpu_info_placement
);
192 static void __init
xen_banner(void)
194 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
195 struct xen_extraversion extra
;
196 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
198 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
200 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
201 version
>> 16, version
& 0xffff, extra
.extraversion
,
202 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
205 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
206 unsigned int *cx
, unsigned int *dx
)
208 unsigned maskedx
= ~0;
211 * Mask out inconvenient features, to try and disable as many
212 * unsupported kernel subsystems as possible.
215 maskedx
= ~((1 << X86_FEATURE_APIC
) | /* disable APIC */
216 (1 << X86_FEATURE_ACPI
) | /* disable ACPI */
217 (1 << X86_FEATURE_MCE
) | /* disable MCE */
218 (1 << X86_FEATURE_MCA
) | /* disable MCA */
219 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
221 asm(XEN_EMULATE_PREFIX
"cpuid"
226 : "0" (*ax
), "2" (*cx
));
230 static void xen_set_debugreg(int reg
, unsigned long val
)
232 HYPERVISOR_set_debugreg(reg
, val
);
235 static unsigned long xen_get_debugreg(int reg
)
237 return HYPERVISOR_get_debugreg(reg
);
240 static void xen_leave_lazy(void)
242 paravirt_leave_lazy(paravirt_get_lazy_mode());
246 static unsigned long xen_store_tr(void)
252 * Set the page permissions for a particular virtual address. If the
253 * address is a vmalloc mapping (or other non-linear mapping), then
254 * find the linear mapping of the page and also set its protections to
257 static void set_aliased_prot(void *v
, pgprot_t prot
)
265 ptep
= lookup_address((unsigned long)v
, &level
);
266 BUG_ON(ptep
== NULL
);
268 pfn
= pte_pfn(*ptep
);
269 page
= pfn_to_page(pfn
);
271 pte
= pfn_pte(pfn
, prot
);
273 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
276 if (!PageHighMem(page
)) {
277 void *av
= __va(PFN_PHYS(pfn
));
280 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
286 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
288 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
291 for(i
= 0; i
< entries
; i
+= entries_per_page
)
292 set_aliased_prot(ldt
+ i
, PAGE_KERNEL_RO
);
295 static void xen_free_ldt(struct desc_struct
*ldt
, unsigned entries
)
297 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
300 for(i
= 0; i
< entries
; i
+= entries_per_page
)
301 set_aliased_prot(ldt
+ i
, PAGE_KERNEL
);
304 static void xen_set_ldt(const void *addr
, unsigned entries
)
306 struct mmuext_op
*op
;
307 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
310 op
->cmd
= MMUEXT_SET_LDT
;
311 op
->arg1
.linear_addr
= (unsigned long)addr
;
312 op
->arg2
.nr_ents
= entries
;
314 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
316 xen_mc_issue(PARAVIRT_LAZY_CPU
);
319 static void xen_load_gdt(const struct desc_ptr
*dtr
)
321 unsigned long *frames
;
322 unsigned long va
= dtr
->address
;
323 unsigned int size
= dtr
->size
+ 1;
324 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
326 struct multicall_space mcs
;
328 /* A GDT can be up to 64k in size, which corresponds to 8192
329 8-byte entries, or 16 4k pages.. */
331 BUG_ON(size
> 65536);
332 BUG_ON(va
& ~PAGE_MASK
);
334 mcs
= xen_mc_entry(sizeof(*frames
) * pages
);
337 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
338 frames
[f
] = virt_to_mfn(va
);
339 make_lowmem_page_readonly((void *)va
);
342 MULTI_set_gdt(mcs
.mc
, frames
, size
/ sizeof(struct desc_struct
));
344 xen_mc_issue(PARAVIRT_LAZY_CPU
);
347 static void load_TLS_descriptor(struct thread_struct
*t
,
348 unsigned int cpu
, unsigned int i
)
350 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
351 xmaddr_t maddr
= virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
352 struct multicall_space mc
= __xen_mc_entry(0);
354 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
357 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
360 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
361 * it means we're in a context switch, and %gs has just been
362 * saved. This means we can zero it out to prevent faults on
363 * exit from the hypervisor if the next process has no %gs.
364 * Either way, it has been saved, and the new value will get
365 * loaded properly. This will go away as soon as Xen has been
366 * modified to not save/restore %gs for normal hypercalls.
368 * On x86_64, this hack is not used for %gs, because gs points
369 * to KERNEL_GS_BASE (and uses it for PDA references), so we
370 * must not zero %gs on x86_64
372 * For x86_64, we need to zero %fs, otherwise we may get an
373 * exception between the new %fs descriptor being loaded and
374 * %fs being effectively cleared at __switch_to().
376 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
386 load_TLS_descriptor(t
, cpu
, 0);
387 load_TLS_descriptor(t
, cpu
, 1);
388 load_TLS_descriptor(t
, cpu
, 2);
390 xen_mc_issue(PARAVIRT_LAZY_CPU
);
394 static void xen_load_gs_index(unsigned int idx
)
396 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
401 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
404 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
405 u64 entry
= *(u64
*)ptr
;
410 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
416 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
417 struct trap_info
*info
)
419 if (val
->type
!= 0xf && val
->type
!= 0xe)
422 info
->vector
= vector
;
423 info
->address
= gate_offset(*val
);
424 info
->cs
= gate_segment(*val
);
425 info
->flags
= val
->dpl
;
426 /* interrupt gates clear IF */
427 if (val
->type
== 0xe)
433 /* Locations of each CPU's IDT */
434 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
436 /* Set an IDT entry. If the entry is part of the current IDT, then
438 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
440 unsigned long p
= (unsigned long)&dt
[entrynum
];
441 unsigned long start
, end
;
445 start
= __get_cpu_var(idt_desc
).address
;
446 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
450 native_write_idt_entry(dt
, entrynum
, g
);
452 if (p
>= start
&& (p
+ 8) <= end
) {
453 struct trap_info info
[2];
457 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
458 if (HYPERVISOR_set_trap_table(info
))
465 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
466 struct trap_info
*traps
)
468 unsigned in
, out
, count
;
470 count
= (desc
->size
+1) / sizeof(gate_desc
);
473 for (in
= out
= 0; in
< count
; in
++) {
474 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
476 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
479 traps
[out
].address
= 0;
482 void xen_copy_trap_info(struct trap_info
*traps
)
484 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
486 xen_convert_trap_info(desc
, traps
);
489 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
490 hold a spinlock to protect the static traps[] array (static because
491 it avoids allocation, and saves stack space). */
492 static void xen_load_idt(const struct desc_ptr
*desc
)
494 static DEFINE_SPINLOCK(lock
);
495 static struct trap_info traps
[257];
499 __get_cpu_var(idt_desc
) = *desc
;
501 xen_convert_trap_info(desc
, traps
);
504 if (HYPERVISOR_set_trap_table(traps
))
510 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
511 they're handled differently. */
512 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
513 const void *desc
, int type
)
524 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
527 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
536 static void xen_load_sp0(struct tss_struct
*tss
,
537 struct thread_struct
*thread
)
539 struct multicall_space mcs
= xen_mc_entry(0);
540 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
541 xen_mc_issue(PARAVIRT_LAZY_CPU
);
544 static void xen_set_iopl_mask(unsigned mask
)
546 struct physdev_set_iopl set_iopl
;
548 /* Force the change at ring 0. */
549 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
550 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
553 static void xen_io_delay(void)
557 #ifdef CONFIG_X86_LOCAL_APIC
558 static u32
xen_apic_read(u32 reg
)
563 static void xen_apic_write(u32 reg
, u32 val
)
565 /* Warn to see if there's any stray references */
569 static u64
xen_apic_icr_read(void)
574 static void xen_apic_icr_write(u32 low
, u32 id
)
576 /* Warn to see if there's any stray references */
580 static void xen_apic_wait_icr_idle(void)
585 static u32
xen_safe_apic_wait_icr_idle(void)
590 static struct apic_ops xen_basic_apic_ops
= {
591 .read
= xen_apic_read
,
592 .write
= xen_apic_write
,
593 .icr_read
= xen_apic_icr_read
,
594 .icr_write
= xen_apic_icr_write
,
595 .wait_icr_idle
= xen_apic_wait_icr_idle
,
596 .safe_wait_icr_idle
= xen_safe_apic_wait_icr_idle
,
601 static void xen_flush_tlb(void)
603 struct mmuext_op
*op
;
604 struct multicall_space mcs
;
608 mcs
= xen_mc_entry(sizeof(*op
));
611 op
->cmd
= MMUEXT_TLB_FLUSH_LOCAL
;
612 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
614 xen_mc_issue(PARAVIRT_LAZY_MMU
);
619 static void xen_flush_tlb_single(unsigned long addr
)
621 struct mmuext_op
*op
;
622 struct multicall_space mcs
;
626 mcs
= xen_mc_entry(sizeof(*op
));
628 op
->cmd
= MMUEXT_INVLPG_LOCAL
;
629 op
->arg1
.linear_addr
= addr
& PAGE_MASK
;
630 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
632 xen_mc_issue(PARAVIRT_LAZY_MMU
);
637 static void xen_flush_tlb_others(const struct cpumask
*cpus
,
638 struct mm_struct
*mm
, unsigned long va
)
642 DECLARE_BITMAP(mask
, NR_CPUS
);
644 struct multicall_space mcs
;
646 BUG_ON(cpumask_empty(cpus
));
649 mcs
= xen_mc_entry(sizeof(*args
));
651 args
->op
.arg2
.vcpumask
= to_cpumask(args
->mask
);
653 /* Remove us, and any offline CPUS. */
654 cpumask_and(to_cpumask(args
->mask
), cpus
, cpu_online_mask
);
655 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args
->mask
));
656 if (unlikely(cpumask_empty(to_cpumask(args
->mask
))))
659 if (va
== TLB_FLUSH_ALL
) {
660 args
->op
.cmd
= MMUEXT_TLB_FLUSH_MULTI
;
662 args
->op
.cmd
= MMUEXT_INVLPG_MULTI
;
663 args
->op
.arg1
.linear_addr
= va
;
666 MULTI_mmuext_op(mcs
.mc
, &args
->op
, 1, NULL
, DOMID_SELF
);
669 xen_mc_issue(PARAVIRT_LAZY_MMU
);
672 static void xen_clts(void)
674 struct multicall_space mcs
;
676 mcs
= xen_mc_entry(0);
678 MULTI_fpu_taskswitch(mcs
.mc
, 0);
680 xen_mc_issue(PARAVIRT_LAZY_CPU
);
683 static void xen_write_cr0(unsigned long cr0
)
685 struct multicall_space mcs
;
687 /* Only pay attention to cr0.TS; everything else is
689 mcs
= xen_mc_entry(0);
691 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
693 xen_mc_issue(PARAVIRT_LAZY_CPU
);
696 static void xen_write_cr2(unsigned long cr2
)
698 percpu_read(xen_vcpu
)->arch
.cr2
= cr2
;
701 static unsigned long xen_read_cr2(void)
703 return percpu_read(xen_vcpu
)->arch
.cr2
;
706 static unsigned long xen_read_cr2_direct(void)
708 return percpu_read(xen_vcpu_info
.arch
.cr2
);
711 static void xen_write_cr4(unsigned long cr4
)
716 native_write_cr4(cr4
);
719 static unsigned long xen_read_cr3(void)
721 return percpu_read(xen_cr3
);
724 static void set_current_cr3(void *v
)
726 percpu_write(xen_current_cr3
, (unsigned long)v
);
729 static void __xen_write_cr3(bool kernel
, unsigned long cr3
)
731 struct mmuext_op
*op
;
732 struct multicall_space mcs
;
736 mfn
= pfn_to_mfn(PFN_DOWN(cr3
));
740 WARN_ON(mfn
== 0 && kernel
);
742 mcs
= __xen_mc_entry(sizeof(*op
));
745 op
->cmd
= kernel
? MMUEXT_NEW_BASEPTR
: MMUEXT_NEW_USER_BASEPTR
;
748 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
751 percpu_write(xen_cr3
, cr3
);
753 /* Update xen_current_cr3 once the batch has actually
755 xen_mc_callback(set_current_cr3
, (void *)cr3
);
759 static void xen_write_cr3(unsigned long cr3
)
761 BUG_ON(preemptible());
763 xen_mc_batch(); /* disables interrupts */
765 /* Update while interrupts are disabled, so its atomic with
767 percpu_write(xen_cr3
, cr3
);
769 __xen_write_cr3(true, cr3
);
773 pgd_t
*user_pgd
= xen_get_user_pgd(__va(cr3
));
775 __xen_write_cr3(false, __pa(user_pgd
));
777 __xen_write_cr3(false, 0);
781 xen_mc_issue(PARAVIRT_LAZY_CPU
); /* interrupts restored */
784 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
795 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
796 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
797 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
800 base
= ((u64
)high
<< 32) | low
;
801 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
809 case MSR_SYSCALL_MASK
:
810 case MSR_IA32_SYSENTER_CS
:
811 case MSR_IA32_SYSENTER_ESP
:
812 case MSR_IA32_SYSENTER_EIP
:
813 /* Fast syscall setup is all done in hypercalls, so
814 these are all ignored. Stub them out here to stop
815 Xen console noise. */
819 ret
= native_write_msr_safe(msr
, low
, high
);
825 /* Early in boot, while setting up the initial pagetable, assume
826 everything is pinned. */
827 static __init
void xen_alloc_pte_init(struct mm_struct
*mm
, unsigned long pfn
)
829 #ifdef CONFIG_FLATMEM
830 BUG_ON(mem_map
); /* should only be used early */
832 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
835 /* Early release_pte assumes that all pts are pinned, since there's
836 only init_mm and anything attached to that is pinned. */
837 static void xen_release_pte_init(unsigned long pfn
)
839 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
842 static void pin_pagetable_pfn(unsigned cmd
, unsigned long pfn
)
846 op
.arg1
.mfn
= pfn_to_mfn(pfn
);
847 if (HYPERVISOR_mmuext_op(&op
, 1, NULL
, DOMID_SELF
))
851 /* This needs to make sure the new pte page is pinned iff its being
852 attached to a pinned pagetable. */
853 static void xen_alloc_ptpage(struct mm_struct
*mm
, unsigned long pfn
, unsigned level
)
855 struct page
*page
= pfn_to_page(pfn
);
857 if (PagePinned(virt_to_page(mm
->pgd
))) {
861 if (!PageHighMem(page
)) {
862 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn
)));
863 if (level
== PT_PTE
&& USE_SPLIT_PTLOCKS
)
864 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE
, pfn
);
866 /* make sure there are no stray mappings of
873 static void xen_alloc_pte(struct mm_struct
*mm
, unsigned long pfn
)
875 xen_alloc_ptpage(mm
, pfn
, PT_PTE
);
878 static void xen_alloc_pmd(struct mm_struct
*mm
, unsigned long pfn
)
880 xen_alloc_ptpage(mm
, pfn
, PT_PMD
);
883 static int xen_pgd_alloc(struct mm_struct
*mm
)
885 pgd_t
*pgd
= mm
->pgd
;
888 BUG_ON(PagePinned(virt_to_page(pgd
)));
892 struct page
*page
= virt_to_page(pgd
);
895 BUG_ON(page
->private != 0);
899 user_pgd
= (pgd_t
*)__get_free_page(GFP_KERNEL
| __GFP_ZERO
);
900 page
->private = (unsigned long)user_pgd
;
902 if (user_pgd
!= NULL
) {
903 user_pgd
[pgd_index(VSYSCALL_START
)] =
904 __pgd(__pa(level3_user_vsyscall
) | _PAGE_TABLE
);
908 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd
))));
915 static void xen_pgd_free(struct mm_struct
*mm
, pgd_t
*pgd
)
918 pgd_t
*user_pgd
= xen_get_user_pgd(pgd
);
921 free_page((unsigned long)user_pgd
);
925 /* This should never happen until we're OK to use struct page */
926 static void xen_release_ptpage(unsigned long pfn
, unsigned level
)
928 struct page
*page
= pfn_to_page(pfn
);
930 if (PagePinned(page
)) {
931 if (!PageHighMem(page
)) {
932 if (level
== PT_PTE
&& USE_SPLIT_PTLOCKS
)
933 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, pfn
);
934 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
936 ClearPagePinned(page
);
940 static void xen_release_pte(unsigned long pfn
)
942 xen_release_ptpage(pfn
, PT_PTE
);
945 static void xen_release_pmd(unsigned long pfn
)
947 xen_release_ptpage(pfn
, PT_PMD
);
950 #if PAGETABLE_LEVELS == 4
951 static void xen_alloc_pud(struct mm_struct
*mm
, unsigned long pfn
)
953 xen_alloc_ptpage(mm
, pfn
, PT_PUD
);
956 static void xen_release_pud(unsigned long pfn
)
958 xen_release_ptpage(pfn
, PT_PUD
);
962 #ifdef CONFIG_HIGHPTE
963 static void *xen_kmap_atomic_pte(struct page
*page
, enum km_type type
)
965 pgprot_t prot
= PAGE_KERNEL
;
967 if (PagePinned(page
))
968 prot
= PAGE_KERNEL_RO
;
970 if (0 && PageHighMem(page
))
971 printk("mapping highpte %lx type %d prot %s\n",
972 page_to_pfn(page
), type
,
973 (unsigned long)pgprot_val(prot
) & _PAGE_RW
? "WRITE" : "READ");
975 return kmap_atomic_prot(page
, type
, prot
);
980 static __init pte_t
mask_rw_pte(pte_t
*ptep
, pte_t pte
)
982 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
983 if (pte_val_ma(*ptep
) & _PAGE_PRESENT
)
984 pte
= __pte_ma(((pte_val_ma(*ptep
) & _PAGE_RW
) | ~_PAGE_RW
) &
990 /* Init-time set_pte while constructing initial pagetables, which
991 doesn't allow RO pagetable pages to be remapped RW */
992 static __init
void xen_set_pte_init(pte_t
*ptep
, pte_t pte
)
994 pte
= mask_rw_pte(ptep
, pte
);
996 xen_set_pte(ptep
, pte
);
1000 static __init
void xen_pagetable_setup_start(pgd_t
*base
)
1004 void xen_setup_shared_info(void)
1006 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1007 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
1008 xen_start_info
->shared_info
);
1010 HYPERVISOR_shared_info
=
1011 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
1013 HYPERVISOR_shared_info
=
1014 (struct shared_info
*)__va(xen_start_info
->shared_info
);
1017 /* In UP this is as good a place as any to set up shared info */
1018 xen_setup_vcpu_info_placement();
1021 xen_setup_mfn_list_list();
1024 static __init
void xen_pagetable_setup_done(pgd_t
*base
)
1026 xen_setup_shared_info();
1029 static __init
void xen_post_allocator_init(void)
1031 pv_mmu_ops
.set_pte
= xen_set_pte
;
1032 pv_mmu_ops
.set_pmd
= xen_set_pmd
;
1033 pv_mmu_ops
.set_pud
= xen_set_pud
;
1034 #if PAGETABLE_LEVELS == 4
1035 pv_mmu_ops
.set_pgd
= xen_set_pgd
;
1038 /* This will work as long as patching hasn't happened yet
1039 (which it hasn't) */
1040 pv_mmu_ops
.alloc_pte
= xen_alloc_pte
;
1041 pv_mmu_ops
.alloc_pmd
= xen_alloc_pmd
;
1042 pv_mmu_ops
.release_pte
= xen_release_pte
;
1043 pv_mmu_ops
.release_pmd
= xen_release_pmd
;
1044 #if PAGETABLE_LEVELS == 4
1045 pv_mmu_ops
.alloc_pud
= xen_alloc_pud
;
1046 pv_mmu_ops
.release_pud
= xen_release_pud
;
1049 #ifdef CONFIG_X86_64
1050 SetPagePinned(virt_to_page(level3_user_vsyscall
));
1052 xen_mark_init_mm_pinned();
1055 /* This is called once we have the cpu_possible_map */
1056 void xen_setup_vcpu_info_placement(void)
1060 for_each_possible_cpu(cpu
)
1061 xen_vcpu_setup(cpu
);
1063 /* xen_vcpu_setup managed to place the vcpu_info within the
1064 percpu area for all cpus, so make use of it */
1065 if (have_vcpu_info_placement
) {
1066 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
1068 pv_irq_ops
.save_fl
= xen_save_fl_direct
;
1069 pv_irq_ops
.restore_fl
= xen_restore_fl_direct
;
1070 pv_irq_ops
.irq_disable
= xen_irq_disable_direct
;
1071 pv_irq_ops
.irq_enable
= xen_irq_enable_direct
;
1072 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1076 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1077 unsigned long addr
, unsigned len
)
1079 char *start
, *end
, *reloc
;
1082 start
= end
= reloc
= NULL
;
1084 #define SITE(op, x) \
1085 case PARAVIRT_PATCH(op.x): \
1086 if (have_vcpu_info_placement) { \
1087 start = (char *)xen_##x##_direct; \
1088 end = xen_##x##_direct_end; \
1089 reloc = xen_##x##_direct_reloc; \
1094 SITE(pv_irq_ops
, irq_enable
);
1095 SITE(pv_irq_ops
, irq_disable
);
1096 SITE(pv_irq_ops
, save_fl
);
1097 SITE(pv_irq_ops
, restore_fl
);
1101 if (start
== NULL
|| (end
-start
) > len
)
1104 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1106 /* Note: because reloc is assigned from something that
1107 appears to be an array, gcc assumes it's non-null,
1108 but doesn't know its relationship with start and
1110 if (reloc
> start
&& reloc
< end
) {
1111 int reloc_off
= reloc
- start
;
1112 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1113 long delta
= start
- (char *)addr
;
1121 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1129 static void xen_set_fixmap(unsigned idx
, unsigned long phys
, pgprot_t prot
)
1133 phys
>>= PAGE_SHIFT
;
1136 case FIX_BTMAP_END
... FIX_BTMAP_BEGIN
:
1137 #ifdef CONFIG_X86_F00F_BUG
1140 #ifdef CONFIG_X86_32
1143 # ifdef CONFIG_HIGHMEM
1144 case FIX_KMAP_BEGIN
... FIX_KMAP_END
:
1147 case VSYSCALL_LAST_PAGE
... VSYSCALL_FIRST_PAGE
:
1149 #ifdef CONFIG_X86_LOCAL_APIC
1150 case FIX_APIC_BASE
: /* maps dummy local APIC */
1152 pte
= pfn_pte(phys
, prot
);
1156 pte
= mfn_pte(phys
, prot
);
1160 __native_set_fixmap(idx
, pte
);
1162 #ifdef CONFIG_X86_64
1163 /* Replicate changes to map the vsyscall page into the user
1164 pagetable vsyscall mapping. */
1165 if (idx
>= VSYSCALL_LAST_PAGE
&& idx
<= VSYSCALL_FIRST_PAGE
) {
1166 unsigned long vaddr
= __fix_to_virt(idx
);
1167 set_pte_vaddr_pud(level3_user_vsyscall
, vaddr
, pte
);
1172 static const struct pv_info xen_info __initdata
= {
1173 .paravirt_enabled
= 1,
1174 .shared_kernel_pmd
= 0,
1179 static const struct pv_init_ops xen_init_ops __initdata
= {
1182 .banner
= xen_banner
,
1183 .memory_setup
= xen_memory_setup
,
1184 .arch_setup
= xen_arch_setup
,
1185 .post_allocator_init
= xen_post_allocator_init
,
1188 static const struct pv_time_ops xen_time_ops __initdata
= {
1189 .time_init
= xen_time_init
,
1191 .set_wallclock
= xen_set_wallclock
,
1192 .get_wallclock
= xen_get_wallclock
,
1193 .get_tsc_khz
= xen_tsc_khz
,
1194 .sched_clock
= xen_sched_clock
,
1197 static const struct pv_cpu_ops xen_cpu_ops __initdata
= {
1200 .set_debugreg
= xen_set_debugreg
,
1201 .get_debugreg
= xen_get_debugreg
,
1205 .read_cr0
= native_read_cr0
,
1206 .write_cr0
= xen_write_cr0
,
1208 .read_cr4
= native_read_cr4
,
1209 .read_cr4_safe
= native_read_cr4_safe
,
1210 .write_cr4
= xen_write_cr4
,
1212 .wbinvd
= native_wbinvd
,
1214 .read_msr
= native_read_msr_safe
,
1215 .write_msr
= xen_write_msr_safe
,
1216 .read_tsc
= native_read_tsc
,
1217 .read_pmc
= native_read_pmc
,
1220 .irq_enable_sysexit
= xen_sysexit
,
1221 #ifdef CONFIG_X86_64
1222 .usergs_sysret32
= xen_sysret32
,
1223 .usergs_sysret64
= xen_sysret64
,
1226 .load_tr_desc
= paravirt_nop
,
1227 .set_ldt
= xen_set_ldt
,
1228 .load_gdt
= xen_load_gdt
,
1229 .load_idt
= xen_load_idt
,
1230 .load_tls
= xen_load_tls
,
1231 #ifdef CONFIG_X86_64
1232 .load_gs_index
= xen_load_gs_index
,
1235 .alloc_ldt
= xen_alloc_ldt
,
1236 .free_ldt
= xen_free_ldt
,
1238 .store_gdt
= native_store_gdt
,
1239 .store_idt
= native_store_idt
,
1240 .store_tr
= xen_store_tr
,
1242 .write_ldt_entry
= xen_write_ldt_entry
,
1243 .write_gdt_entry
= xen_write_gdt_entry
,
1244 .write_idt_entry
= xen_write_idt_entry
,
1245 .load_sp0
= xen_load_sp0
,
1247 .set_iopl_mask
= xen_set_iopl_mask
,
1248 .io_delay
= xen_io_delay
,
1250 /* Xen takes care of %gs when switching to usermode for us */
1251 .swapgs
= paravirt_nop
,
1254 .enter
= paravirt_enter_lazy_cpu
,
1255 .leave
= xen_leave_lazy
,
1259 static const struct pv_apic_ops xen_apic_ops __initdata
= {
1260 #ifdef CONFIG_X86_LOCAL_APIC
1261 .setup_boot_clock
= paravirt_nop
,
1262 .setup_secondary_clock
= paravirt_nop
,
1263 .startup_ipi_hook
= paravirt_nop
,
1267 static const struct pv_mmu_ops xen_mmu_ops __initdata
= {
1268 .pagetable_setup_start
= xen_pagetable_setup_start
,
1269 .pagetable_setup_done
= xen_pagetable_setup_done
,
1271 .read_cr2
= xen_read_cr2
,
1272 .write_cr2
= xen_write_cr2
,
1274 .read_cr3
= xen_read_cr3
,
1275 .write_cr3
= xen_write_cr3
,
1277 .flush_tlb_user
= xen_flush_tlb
,
1278 .flush_tlb_kernel
= xen_flush_tlb
,
1279 .flush_tlb_single
= xen_flush_tlb_single
,
1280 .flush_tlb_others
= xen_flush_tlb_others
,
1282 .pte_update
= paravirt_nop
,
1283 .pte_update_defer
= paravirt_nop
,
1285 .pgd_alloc
= xen_pgd_alloc
,
1286 .pgd_free
= xen_pgd_free
,
1288 .alloc_pte
= xen_alloc_pte_init
,
1289 .release_pte
= xen_release_pte_init
,
1290 .alloc_pmd
= xen_alloc_pte_init
,
1291 .alloc_pmd_clone
= paravirt_nop
,
1292 .release_pmd
= xen_release_pte_init
,
1294 #ifdef CONFIG_HIGHPTE
1295 .kmap_atomic_pte
= xen_kmap_atomic_pte
,
1298 #ifdef CONFIG_X86_64
1299 .set_pte
= xen_set_pte
,
1301 .set_pte
= xen_set_pte_init
,
1303 .set_pte_at
= xen_set_pte_at
,
1304 .set_pmd
= xen_set_pmd_hyper
,
1306 .ptep_modify_prot_start
= __ptep_modify_prot_start
,
1307 .ptep_modify_prot_commit
= __ptep_modify_prot_commit
,
1309 .pte_val
= xen_pte_val
,
1310 .pte_flags
= native_pte_flags
,
1311 .pgd_val
= xen_pgd_val
,
1313 .make_pte
= xen_make_pte
,
1314 .make_pgd
= xen_make_pgd
,
1316 #ifdef CONFIG_X86_PAE
1317 .set_pte_atomic
= xen_set_pte_atomic
,
1318 .set_pte_present
= xen_set_pte_at
,
1319 .pte_clear
= xen_pte_clear
,
1320 .pmd_clear
= xen_pmd_clear
,
1321 #endif /* CONFIG_X86_PAE */
1322 .set_pud
= xen_set_pud_hyper
,
1324 .make_pmd
= xen_make_pmd
,
1325 .pmd_val
= xen_pmd_val
,
1327 #if PAGETABLE_LEVELS == 4
1328 .pud_val
= xen_pud_val
,
1329 .make_pud
= xen_make_pud
,
1330 .set_pgd
= xen_set_pgd_hyper
,
1332 .alloc_pud
= xen_alloc_pte_init
,
1333 .release_pud
= xen_release_pte_init
,
1334 #endif /* PAGETABLE_LEVELS == 4 */
1336 .activate_mm
= xen_activate_mm
,
1337 .dup_mmap
= xen_dup_mmap
,
1338 .exit_mmap
= xen_exit_mmap
,
1341 .enter
= paravirt_enter_lazy_mmu
,
1342 .leave
= xen_leave_lazy
,
1345 .set_fixmap
= xen_set_fixmap
,
1348 static void xen_reboot(int reason
)
1350 struct sched_shutdown r
= { .reason
= reason
};
1356 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1360 static void xen_restart(char *msg
)
1362 xen_reboot(SHUTDOWN_reboot
);
1365 static void xen_emergency_restart(void)
1367 xen_reboot(SHUTDOWN_reboot
);
1370 static void xen_machine_halt(void)
1372 xen_reboot(SHUTDOWN_poweroff
);
1375 static void xen_crash_shutdown(struct pt_regs
*regs
)
1377 xen_reboot(SHUTDOWN_crash
);
1380 static const struct machine_ops __initdata xen_machine_ops
= {
1381 .restart
= xen_restart
,
1382 .halt
= xen_machine_halt
,
1383 .power_off
= xen_machine_halt
,
1384 .shutdown
= xen_machine_halt
,
1385 .crash_shutdown
= xen_crash_shutdown
,
1386 .emergency_restart
= xen_emergency_restart
,
1390 static void __init
xen_reserve_top(void)
1392 #ifdef CONFIG_X86_32
1393 unsigned long top
= HYPERVISOR_VIRT_START
;
1394 struct xen_platform_parameters pp
;
1396 if (HYPERVISOR_xen_version(XENVER_platform_parameters
, &pp
) == 0)
1397 top
= pp
.virt_start
;
1399 reserve_top_address(-top
);
1400 #endif /* CONFIG_X86_32 */
1404 * Like __va(), but returns address in the kernel mapping (which is
1405 * all we have until the physical memory mapping has been set up.
1407 static void *__ka(phys_addr_t paddr
)
1409 #ifdef CONFIG_X86_64
1410 return (void *)(paddr
+ __START_KERNEL_map
);
1416 /* Convert a machine address to physical address */
1417 static unsigned long m2p(phys_addr_t maddr
)
1421 maddr
&= PTE_PFN_MASK
;
1422 paddr
= mfn_to_pfn(maddr
>> PAGE_SHIFT
) << PAGE_SHIFT
;
1427 /* Convert a machine address to kernel virtual */
1428 static void *m2v(phys_addr_t maddr
)
1430 return __ka(m2p(maddr
));
1433 static void set_page_prot(void *addr
, pgprot_t prot
)
1435 unsigned long pfn
= __pa(addr
) >> PAGE_SHIFT
;
1436 pte_t pte
= pfn_pte(pfn
, prot
);
1438 if (HYPERVISOR_update_va_mapping((unsigned long)addr
, pte
, 0))
1442 static __init
void xen_map_identity_early(pmd_t
*pmd
, unsigned long max_pfn
)
1444 unsigned pmdidx
, pteidx
;
1450 for (pmdidx
= 0; pmdidx
< PTRS_PER_PMD
&& pfn
< max_pfn
; pmdidx
++) {
1453 /* Reuse or allocate a page of ptes */
1454 if (pmd_present(pmd
[pmdidx
]))
1455 pte_page
= m2v(pmd
[pmdidx
].pmd
);
1457 /* Check for free pte pages */
1458 if (ident_pte
== ARRAY_SIZE(level1_ident_pgt
))
1461 pte_page
= &level1_ident_pgt
[ident_pte
];
1462 ident_pte
+= PTRS_PER_PTE
;
1464 pmd
[pmdidx
] = __pmd(__pa(pte_page
) | _PAGE_TABLE
);
1467 /* Install mappings */
1468 for (pteidx
= 0; pteidx
< PTRS_PER_PTE
; pteidx
++, pfn
++) {
1471 if (pfn
> max_pfn_mapped
)
1472 max_pfn_mapped
= pfn
;
1474 if (!pte_none(pte_page
[pteidx
]))
1477 pte
= pfn_pte(pfn
, PAGE_KERNEL_EXEC
);
1478 pte_page
[pteidx
] = pte
;
1482 for (pteidx
= 0; pteidx
< ident_pte
; pteidx
+= PTRS_PER_PTE
)
1483 set_page_prot(&level1_ident_pgt
[pteidx
], PAGE_KERNEL_RO
);
1485 set_page_prot(pmd
, PAGE_KERNEL_RO
);
1488 #ifdef CONFIG_X86_64
1489 static void convert_pfn_mfn(void *v
)
1494 /* All levels are converted the same way, so just treat them
1496 for (i
= 0; i
< PTRS_PER_PTE
; i
++)
1497 pte
[i
] = xen_make_pte(pte
[i
].pte
);
1501 * Set up the inital kernel pagetable.
1503 * We can construct this by grafting the Xen provided pagetable into
1504 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1505 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1506 * means that only the kernel has a physical mapping to start with -
1507 * but that's enough to get __va working. We need to fill in the rest
1508 * of the physical mapping once some sort of allocator has been set
1511 static __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
,
1512 unsigned long max_pfn
)
1517 /* Zap identity mapping */
1518 init_level4_pgt
[0] = __pgd(0);
1520 /* Pre-constructed entries are in pfn, so convert to mfn */
1521 convert_pfn_mfn(init_level4_pgt
);
1522 convert_pfn_mfn(level3_ident_pgt
);
1523 convert_pfn_mfn(level3_kernel_pgt
);
1525 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
)].pgd
);
1526 l2
= m2v(l3
[pud_index(__START_KERNEL_map
)].pud
);
1528 memcpy(level2_ident_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1529 memcpy(level2_kernel_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1531 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
+ PMD_SIZE
)].pgd
);
1532 l2
= m2v(l3
[pud_index(__START_KERNEL_map
+ PMD_SIZE
)].pud
);
1533 memcpy(level2_fixmap_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1535 /* Set up identity map */
1536 xen_map_identity_early(level2_ident_pgt
, max_pfn
);
1538 /* Make pagetable pieces RO */
1539 set_page_prot(init_level4_pgt
, PAGE_KERNEL_RO
);
1540 set_page_prot(level3_ident_pgt
, PAGE_KERNEL_RO
);
1541 set_page_prot(level3_kernel_pgt
, PAGE_KERNEL_RO
);
1542 set_page_prot(level3_user_vsyscall
, PAGE_KERNEL_RO
);
1543 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1544 set_page_prot(level2_fixmap_pgt
, PAGE_KERNEL_RO
);
1546 /* Pin down new L4 */
1547 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE
,
1548 PFN_DOWN(__pa_symbol(init_level4_pgt
)));
1550 /* Unpin Xen-provided one */
1551 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1554 pgd
= init_level4_pgt
;
1557 * At this stage there can be no user pgd, and no page
1558 * structure to attach it to, so make sure we just set kernel
1562 __xen_write_cr3(true, __pa(pgd
));
1563 xen_mc_issue(PARAVIRT_LAZY_CPU
);
1565 reserve_early(__pa(xen_start_info
->pt_base
),
1566 __pa(xen_start_info
->pt_base
+
1567 xen_start_info
->nr_pt_frames
* PAGE_SIZE
),
1572 #else /* !CONFIG_X86_64 */
1573 static pmd_t level2_kernel_pgt
[PTRS_PER_PMD
] __page_aligned_bss
;
1575 static __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
,
1576 unsigned long max_pfn
)
1580 init_pg_tables_start
= __pa(pgd
);
1581 init_pg_tables_end
= __pa(pgd
) + xen_start_info
->nr_pt_frames
*PAGE_SIZE
;
1582 max_pfn_mapped
= PFN_DOWN(init_pg_tables_end
+ 512*1024);
1584 kernel_pmd
= m2v(pgd
[KERNEL_PGD_BOUNDARY
].pgd
);
1585 memcpy(level2_kernel_pgt
, kernel_pmd
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1587 xen_map_identity_early(level2_kernel_pgt
, max_pfn
);
1589 memcpy(swapper_pg_dir
, pgd
, sizeof(pgd_t
) * PTRS_PER_PGD
);
1590 set_pgd(&swapper_pg_dir
[KERNEL_PGD_BOUNDARY
],
1591 __pgd(__pa(level2_kernel_pgt
) | _PAGE_PRESENT
));
1593 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1594 set_page_prot(swapper_pg_dir
, PAGE_KERNEL_RO
);
1595 set_page_prot(empty_zero_page
, PAGE_KERNEL_RO
);
1597 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1599 xen_write_cr3(__pa(swapper_pg_dir
));
1601 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE
, PFN_DOWN(__pa(swapper_pg_dir
)));
1603 return swapper_pg_dir
;
1605 #endif /* CONFIG_X86_64 */
1607 /* First C function to be called on Xen boot */
1608 asmlinkage
void __init
xen_start_kernel(void)
1612 if (!xen_start_info
)
1615 xen_domain_type
= XEN_PV_DOMAIN
;
1617 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3", 5) != 0);
1619 xen_setup_features();
1621 /* Install Xen paravirt ops */
1623 pv_init_ops
= xen_init_ops
;
1624 pv_time_ops
= xen_time_ops
;
1625 pv_cpu_ops
= xen_cpu_ops
;
1626 pv_apic_ops
= xen_apic_ops
;
1627 pv_mmu_ops
= xen_mmu_ops
;
1631 #ifdef CONFIG_X86_LOCAL_APIC
1633 * set up the basic apic ops.
1635 apic_ops
= &xen_basic_apic_ops
;
1638 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1639 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1640 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1643 machine_ops
= xen_machine_ops
;
1645 #ifdef CONFIG_X86_64
1646 /* Disable until direct per-cpu data access. */
1647 have_vcpu_info_placement
= 0;
1653 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1654 xen_build_dynamic_phys_to_machine();
1656 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1658 /* Prevent unwanted bits from being set in PTEs. */
1659 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1660 if (!xen_initial_domain())
1661 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1663 /* Don't do the full vcpu_info placement stuff until we have a
1664 possible map and a non-dummy shared_info. */
1665 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1667 xen_raw_console_write("mapping kernel into physical memory\n");
1668 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
1672 /* keep using Xen gdt for now; no urgent need to change it */
1674 pv_info
.kernel_rpl
= 1;
1675 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1676 pv_info
.kernel_rpl
= 0;
1678 /* set the limit of our address space */
1681 #ifdef CONFIG_X86_32
1682 /* set up basic CPUID stuff */
1683 cpu_detect(&new_cpu_data
);
1684 new_cpu_data
.hard_math
= 1;
1685 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1688 /* Poke various useful things into boot_params */
1689 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1690 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1691 ? __pa(xen_start_info
->mod_start
) : 0;
1692 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1693 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1695 if (!xen_initial_domain()) {
1696 add_preferred_console("xenboot", 0, NULL
);
1697 add_preferred_console("tty", 0, NULL
);
1698 add_preferred_console("hvc", 0, NULL
);
1701 xen_raw_console_write("about to get started...\n");
1703 /* Start the world */
1704 #ifdef CONFIG_X86_32
1705 i386_start_kernel();
1707 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));