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/physdev.h>
32 #include <xen/interface/vcpu.h>
33 #include <xen/features.h>
35 #include <xen/hvc-console.h>
37 #include <asm/paravirt.h>
39 #include <asm/xen/hypercall.h>
40 #include <asm/xen/hypervisor.h>
41 #include <asm/fixmap.h>
42 #include <asm/processor.h>
43 #include <asm/msr-index.h>
44 #include <asm/setup.h>
46 #include <asm/pgtable.h>
47 #include <asm/tlbflush.h>
48 #include <asm/reboot.h>
52 #include "multicalls.h"
54 EXPORT_SYMBOL_GPL(hypercall_page
);
56 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
57 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
60 * Identity map, in addition to plain kernel map. This needs to be
61 * large enough to allocate page table pages to allocate the rest.
62 * Each page can map 2MB.
64 static pte_t level1_ident_pgt
[PTRS_PER_PTE
* 4] __page_aligned_bss
;
67 /* l3 pud for userspace vsyscall mapping */
68 static pud_t level3_user_vsyscall
[PTRS_PER_PUD
] __page_aligned_bss
;
69 #endif /* CONFIG_X86_64 */
72 * Note about cr3 (pagetable base) values:
74 * xen_cr3 contains the current logical cr3 value; it contains the
75 * last set cr3. This may not be the current effective cr3, because
76 * its update may be being lazily deferred. However, a vcpu looking
77 * at its own cr3 can use this value knowing that it everything will
80 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
81 * hypercall to set the vcpu cr3 is complete (so it may be a little
82 * out of date, but it will never be set early). If one vcpu is
83 * looking at another vcpu's cr3 value, it should use this variable.
85 DEFINE_PER_CPU(unsigned long, xen_cr3
); /* cr3 stored as physaddr */
86 DEFINE_PER_CPU(unsigned long, xen_current_cr3
); /* actual vcpu cr3 */
88 struct start_info
*xen_start_info
;
89 EXPORT_SYMBOL_GPL(xen_start_info
);
91 struct shared_info xen_dummy_shared_info
;
94 * Point at some empty memory to start with. We map the real shared_info
95 * page as soon as fixmap is up and running.
97 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
100 * Flag to determine whether vcpu info placement is available on all
101 * VCPUs. We assume it is to start with, and then set it to zero on
102 * the first failure. This is because it can succeed on some VCPUs
103 * and not others, since it can involve hypervisor memory allocation,
104 * or because the guest failed to guarantee all the appropriate
105 * constraints on all VCPUs (ie buffer can't cross a page boundary).
107 * Note that any particular CPU may be using a placed vcpu structure,
108 * but we can only optimise if the all are.
110 * 0: not available, 1: available
112 static int have_vcpu_info_placement
= 1;
114 static void xen_vcpu_setup(int cpu
)
116 struct vcpu_register_vcpu_info info
;
118 struct vcpu_info
*vcpup
;
120 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
121 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
123 if (!have_vcpu_info_placement
)
124 return; /* already tested, not available */
126 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
128 info
.mfn
= virt_to_mfn(vcpup
);
129 info
.offset
= offset_in_page(vcpup
);
131 printk(KERN_DEBUG
"trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
132 cpu
, vcpup
, info
.mfn
, info
.offset
);
134 /* Check to see if the hypervisor will put the vcpu_info
135 structure where we want it, which allows direct access via
136 a percpu-variable. */
137 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
140 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
141 have_vcpu_info_placement
= 0;
143 /* This cpu is using the registered vcpu info, even if
144 later ones fail to. */
145 per_cpu(xen_vcpu
, cpu
) = vcpup
;
147 printk(KERN_DEBUG
"cpu %d using vcpu_info at %p\n",
153 * On restore, set the vcpu placement up again.
154 * If it fails, then we're in a bad state, since
155 * we can't back out from using it...
157 void xen_vcpu_restore(void)
159 if (have_vcpu_info_placement
) {
162 for_each_online_cpu(cpu
) {
163 bool other_cpu
= (cpu
!= smp_processor_id());
166 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
172 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
176 BUG_ON(!have_vcpu_info_placement
);
180 static void __init
xen_banner(void)
182 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
183 struct xen_extraversion extra
;
184 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
186 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
188 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
189 version
>> 16, version
& 0xffff, extra
.extraversion
,
190 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
193 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
194 unsigned int *cx
, unsigned int *dx
)
196 unsigned maskedx
= ~0;
199 * Mask out inconvenient features, to try and disable as many
200 * unsupported kernel subsystems as possible.
203 maskedx
= ~((1 << X86_FEATURE_APIC
) | /* disable APIC */
204 (1 << X86_FEATURE_ACPI
) | /* disable ACPI */
205 (1 << X86_FEATURE_MCE
) | /* disable MCE */
206 (1 << X86_FEATURE_MCA
) | /* disable MCA */
207 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
209 asm(XEN_EMULATE_PREFIX
"cpuid"
214 : "0" (*ax
), "2" (*cx
));
218 static void xen_set_debugreg(int reg
, unsigned long val
)
220 HYPERVISOR_set_debugreg(reg
, val
);
223 static unsigned long xen_get_debugreg(int reg
)
225 return HYPERVISOR_get_debugreg(reg
);
228 static void xen_leave_lazy(void)
230 paravirt_leave_lazy(paravirt_get_lazy_mode());
234 static unsigned long xen_store_tr(void)
240 * Set the page permissions for a particular virtual address. If the
241 * address is a vmalloc mapping (or other non-linear mapping), then
242 * find the linear mapping of the page and also set its protections to
245 static void set_aliased_prot(void *v
, pgprot_t prot
)
253 ptep
= lookup_address((unsigned long)v
, &level
);
254 BUG_ON(ptep
== NULL
);
256 pfn
= pte_pfn(*ptep
);
257 page
= pfn_to_page(pfn
);
259 pte
= pfn_pte(pfn
, prot
);
261 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
264 if (!PageHighMem(page
)) {
265 void *av
= __va(PFN_PHYS(pfn
));
268 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
274 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
276 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
279 for(i
= 0; i
< entries
; i
+= entries_per_page
)
280 set_aliased_prot(ldt
+ i
, PAGE_KERNEL_RO
);
283 static void xen_free_ldt(struct desc_struct
*ldt
, unsigned entries
)
285 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
288 for(i
= 0; i
< entries
; i
+= entries_per_page
)
289 set_aliased_prot(ldt
+ i
, PAGE_KERNEL
);
292 static void xen_set_ldt(const void *addr
, unsigned entries
)
294 struct mmuext_op
*op
;
295 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
298 op
->cmd
= MMUEXT_SET_LDT
;
299 op
->arg1
.linear_addr
= (unsigned long)addr
;
300 op
->arg2
.nr_ents
= entries
;
302 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
304 xen_mc_issue(PARAVIRT_LAZY_CPU
);
307 static void xen_load_gdt(const struct desc_ptr
*dtr
)
309 unsigned long *frames
;
310 unsigned long va
= dtr
->address
;
311 unsigned int size
= dtr
->size
+ 1;
312 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
314 struct multicall_space mcs
;
316 /* A GDT can be up to 64k in size, which corresponds to 8192
317 8-byte entries, or 16 4k pages.. */
319 BUG_ON(size
> 65536);
320 BUG_ON(va
& ~PAGE_MASK
);
322 mcs
= xen_mc_entry(sizeof(*frames
) * pages
);
325 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
326 frames
[f
] = virt_to_mfn(va
);
327 make_lowmem_page_readonly((void *)va
);
330 MULTI_set_gdt(mcs
.mc
, frames
, size
/ sizeof(struct desc_struct
));
332 xen_mc_issue(PARAVIRT_LAZY_CPU
);
335 static void load_TLS_descriptor(struct thread_struct
*t
,
336 unsigned int cpu
, unsigned int i
)
338 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
339 xmaddr_t maddr
= virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
340 struct multicall_space mc
= __xen_mc_entry(0);
342 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
345 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
348 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
349 * it means we're in a context switch, and %gs has just been
350 * saved. This means we can zero it out to prevent faults on
351 * exit from the hypervisor if the next process has no %gs.
352 * Either way, it has been saved, and the new value will get
353 * loaded properly. This will go away as soon as Xen has been
354 * modified to not save/restore %gs for normal hypercalls.
356 * On x86_64, this hack is not used for %gs, because gs points
357 * to KERNEL_GS_BASE (and uses it for PDA references), so we
358 * must not zero %gs on x86_64
360 * For x86_64, we need to zero %fs, otherwise we may get an
361 * exception between the new %fs descriptor being loaded and
362 * %fs being effectively cleared at __switch_to().
364 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
374 load_TLS_descriptor(t
, cpu
, 0);
375 load_TLS_descriptor(t
, cpu
, 1);
376 load_TLS_descriptor(t
, cpu
, 2);
378 xen_mc_issue(PARAVIRT_LAZY_CPU
);
382 static void xen_load_gs_index(unsigned int idx
)
384 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
389 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
392 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
393 u64 entry
= *(u64
*)ptr
;
398 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
404 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
405 struct trap_info
*info
)
407 if (val
->type
!= 0xf && val
->type
!= 0xe)
410 info
->vector
= vector
;
411 info
->address
= gate_offset(*val
);
412 info
->cs
= gate_segment(*val
);
413 info
->flags
= val
->dpl
;
414 /* interrupt gates clear IF */
415 if (val
->type
== 0xe)
421 /* Locations of each CPU's IDT */
422 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
424 /* Set an IDT entry. If the entry is part of the current IDT, then
426 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
428 unsigned long p
= (unsigned long)&dt
[entrynum
];
429 unsigned long start
, end
;
433 start
= __get_cpu_var(idt_desc
).address
;
434 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
438 native_write_idt_entry(dt
, entrynum
, g
);
440 if (p
>= start
&& (p
+ 8) <= end
) {
441 struct trap_info info
[2];
445 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
446 if (HYPERVISOR_set_trap_table(info
))
453 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
454 struct trap_info
*traps
)
456 unsigned in
, out
, count
;
458 count
= (desc
->size
+1) / sizeof(gate_desc
);
461 for (in
= out
= 0; in
< count
; in
++) {
462 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
464 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
467 traps
[out
].address
= 0;
470 void xen_copy_trap_info(struct trap_info
*traps
)
472 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
474 xen_convert_trap_info(desc
, traps
);
477 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
478 hold a spinlock to protect the static traps[] array (static because
479 it avoids allocation, and saves stack space). */
480 static void xen_load_idt(const struct desc_ptr
*desc
)
482 static DEFINE_SPINLOCK(lock
);
483 static struct trap_info traps
[257];
487 __get_cpu_var(idt_desc
) = *desc
;
489 xen_convert_trap_info(desc
, traps
);
492 if (HYPERVISOR_set_trap_table(traps
))
498 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
499 they're handled differently. */
500 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
501 const void *desc
, int type
)
512 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
515 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
524 static void xen_load_sp0(struct tss_struct
*tss
,
525 struct thread_struct
*thread
)
527 struct multicall_space mcs
= xen_mc_entry(0);
528 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
529 xen_mc_issue(PARAVIRT_LAZY_CPU
);
532 static void xen_set_iopl_mask(unsigned mask
)
534 struct physdev_set_iopl set_iopl
;
536 /* Force the change at ring 0. */
537 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
538 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
541 static void xen_io_delay(void)
545 #ifdef CONFIG_X86_LOCAL_APIC
546 static u32
xen_apic_read(unsigned long reg
)
551 static void xen_apic_write(unsigned long reg
, u32 val
)
553 /* Warn to see if there's any stray references */
558 static void xen_flush_tlb(void)
560 struct mmuext_op
*op
;
561 struct multicall_space mcs
;
565 mcs
= xen_mc_entry(sizeof(*op
));
568 op
->cmd
= MMUEXT_TLB_FLUSH_LOCAL
;
569 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
571 xen_mc_issue(PARAVIRT_LAZY_MMU
);
576 static void xen_flush_tlb_single(unsigned long addr
)
578 struct mmuext_op
*op
;
579 struct multicall_space mcs
;
583 mcs
= xen_mc_entry(sizeof(*op
));
585 op
->cmd
= MMUEXT_INVLPG_LOCAL
;
586 op
->arg1
.linear_addr
= addr
& PAGE_MASK
;
587 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
589 xen_mc_issue(PARAVIRT_LAZY_MMU
);
594 static void xen_flush_tlb_others(const cpumask_t
*cpus
, struct mm_struct
*mm
,
601 cpumask_t cpumask
= *cpus
;
602 struct multicall_space mcs
;
605 * A couple of (to be removed) sanity checks:
607 * - current CPU must not be in mask
608 * - mask must exist :)
610 BUG_ON(cpus_empty(cpumask
));
611 BUG_ON(cpu_isset(smp_processor_id(), cpumask
));
614 /* If a CPU which we ran on has gone down, OK. */
615 cpus_and(cpumask
, cpumask
, cpu_online_map
);
616 if (cpus_empty(cpumask
))
619 mcs
= xen_mc_entry(sizeof(*args
));
621 args
->mask
= cpumask
;
622 args
->op
.arg2
.vcpumask
= &args
->mask
;
624 if (va
== TLB_FLUSH_ALL
) {
625 args
->op
.cmd
= MMUEXT_TLB_FLUSH_MULTI
;
627 args
->op
.cmd
= MMUEXT_INVLPG_MULTI
;
628 args
->op
.arg1
.linear_addr
= va
;
631 MULTI_mmuext_op(mcs
.mc
, &args
->op
, 1, NULL
, DOMID_SELF
);
633 xen_mc_issue(PARAVIRT_LAZY_MMU
);
636 static void xen_clts(void)
638 struct multicall_space mcs
;
640 mcs
= xen_mc_entry(0);
642 MULTI_fpu_taskswitch(mcs
.mc
, 0);
644 xen_mc_issue(PARAVIRT_LAZY_CPU
);
647 static void xen_write_cr0(unsigned long cr0
)
649 struct multicall_space mcs
;
651 /* Only pay attention to cr0.TS; everything else is
653 mcs
= xen_mc_entry(0);
655 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
657 xen_mc_issue(PARAVIRT_LAZY_CPU
);
660 static void xen_write_cr2(unsigned long cr2
)
662 x86_read_percpu(xen_vcpu
)->arch
.cr2
= cr2
;
665 static unsigned long xen_read_cr2(void)
667 return x86_read_percpu(xen_vcpu
)->arch
.cr2
;
670 static unsigned long xen_read_cr2_direct(void)
672 return x86_read_percpu(xen_vcpu_info
.arch
.cr2
);
675 static void xen_write_cr4(unsigned long cr4
)
680 native_write_cr4(cr4
);
683 static unsigned long xen_read_cr3(void)
685 return x86_read_percpu(xen_cr3
);
688 static void set_current_cr3(void *v
)
690 x86_write_percpu(xen_current_cr3
, (unsigned long)v
);
693 static void __xen_write_cr3(bool kernel
, unsigned long cr3
)
695 struct mmuext_op
*op
;
696 struct multicall_space mcs
;
700 mfn
= pfn_to_mfn(PFN_DOWN(cr3
));
704 WARN_ON(mfn
== 0 && kernel
);
706 mcs
= __xen_mc_entry(sizeof(*op
));
709 op
->cmd
= kernel
? MMUEXT_NEW_BASEPTR
: MMUEXT_NEW_USER_BASEPTR
;
712 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
715 x86_write_percpu(xen_cr3
, cr3
);
717 /* Update xen_current_cr3 once the batch has actually
719 xen_mc_callback(set_current_cr3
, (void *)cr3
);
723 static void xen_write_cr3(unsigned long cr3
)
725 BUG_ON(preemptible());
727 xen_mc_batch(); /* disables interrupts */
729 /* Update while interrupts are disabled, so its atomic with
731 x86_write_percpu(xen_cr3
, cr3
);
733 __xen_write_cr3(true, cr3
);
737 pgd_t
*user_pgd
= xen_get_user_pgd(__va(cr3
));
739 __xen_write_cr3(false, __pa(user_pgd
));
741 __xen_write_cr3(false, 0);
745 xen_mc_issue(PARAVIRT_LAZY_CPU
); /* interrupts restored */
748 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
759 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
760 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
761 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
764 base
= ((u64
)high
<< 32) | low
;
765 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
773 case MSR_SYSCALL_MASK
:
774 case MSR_IA32_SYSENTER_CS
:
775 case MSR_IA32_SYSENTER_ESP
:
776 case MSR_IA32_SYSENTER_EIP
:
777 /* Fast syscall setup is all done in hypercalls, so
778 these are all ignored. Stub them out here to stop
779 Xen console noise. */
783 ret
= native_write_msr_safe(msr
, low
, high
);
789 /* Early in boot, while setting up the initial pagetable, assume
790 everything is pinned. */
791 static __init
void xen_alloc_pte_init(struct mm_struct
*mm
, u32 pfn
)
793 #ifdef CONFIG_FLATMEM
794 BUG_ON(mem_map
); /* should only be used early */
796 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
799 /* Early release_pte assumes that all pts are pinned, since there's
800 only init_mm and anything attached to that is pinned. */
801 static void xen_release_pte_init(u32 pfn
)
803 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
806 static void pin_pagetable_pfn(unsigned cmd
, unsigned long pfn
)
810 op
.arg1
.mfn
= pfn_to_mfn(pfn
);
811 if (HYPERVISOR_mmuext_op(&op
, 1, NULL
, DOMID_SELF
))
815 /* This needs to make sure the new pte page is pinned iff its being
816 attached to a pinned pagetable. */
817 static void xen_alloc_ptpage(struct mm_struct
*mm
, u32 pfn
, unsigned level
)
819 struct page
*page
= pfn_to_page(pfn
);
821 if (PagePinned(virt_to_page(mm
->pgd
))) {
824 if (!PageHighMem(page
)) {
825 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn
)));
827 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE
, pfn
);
829 /* make sure there are no stray mappings of
835 static void xen_alloc_pte(struct mm_struct
*mm
, u32 pfn
)
837 xen_alloc_ptpage(mm
, pfn
, PT_PTE
);
840 static void xen_alloc_pmd(struct mm_struct
*mm
, u32 pfn
)
842 xen_alloc_ptpage(mm
, pfn
, PT_PMD
);
845 static int xen_pgd_alloc(struct mm_struct
*mm
)
847 pgd_t
*pgd
= mm
->pgd
;
850 BUG_ON(PagePinned(virt_to_page(pgd
)));
854 struct page
*page
= virt_to_page(pgd
);
857 BUG_ON(page
->private != 0);
861 user_pgd
= (pgd_t
*)__get_free_page(GFP_KERNEL
| __GFP_ZERO
);
862 page
->private = (unsigned long)user_pgd
;
864 if (user_pgd
!= NULL
) {
865 user_pgd
[pgd_index(VSYSCALL_START
)] =
866 __pgd(__pa(level3_user_vsyscall
) | _PAGE_TABLE
);
870 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd
))));
877 static void xen_pgd_free(struct mm_struct
*mm
, pgd_t
*pgd
)
880 pgd_t
*user_pgd
= xen_get_user_pgd(pgd
);
883 free_page((unsigned long)user_pgd
);
887 /* This should never happen until we're OK to use struct page */
888 static void xen_release_ptpage(u32 pfn
, unsigned level
)
890 struct page
*page
= pfn_to_page(pfn
);
892 if (PagePinned(page
)) {
893 if (!PageHighMem(page
)) {
895 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, pfn
);
896 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
898 ClearPagePinned(page
);
902 static void xen_release_pte(u32 pfn
)
904 xen_release_ptpage(pfn
, PT_PTE
);
907 static void xen_release_pmd(u32 pfn
)
909 xen_release_ptpage(pfn
, PT_PMD
);
912 #if PAGETABLE_LEVELS == 4
913 static void xen_alloc_pud(struct mm_struct
*mm
, u32 pfn
)
915 xen_alloc_ptpage(mm
, pfn
, PT_PUD
);
918 static void xen_release_pud(u32 pfn
)
920 xen_release_ptpage(pfn
, PT_PUD
);
924 #ifdef CONFIG_HIGHPTE
925 static void *xen_kmap_atomic_pte(struct page
*page
, enum km_type type
)
927 pgprot_t prot
= PAGE_KERNEL
;
929 if (PagePinned(page
))
930 prot
= PAGE_KERNEL_RO
;
932 if (0 && PageHighMem(page
))
933 printk("mapping highpte %lx type %d prot %s\n",
934 page_to_pfn(page
), type
,
935 (unsigned long)pgprot_val(prot
) & _PAGE_RW
? "WRITE" : "READ");
937 return kmap_atomic_prot(page
, type
, prot
);
941 static __init pte_t
mask_rw_pte(pte_t
*ptep
, pte_t pte
)
943 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
944 if (pte_val_ma(*ptep
) & _PAGE_PRESENT
)
945 pte
= __pte_ma(((pte_val_ma(*ptep
) & _PAGE_RW
) | ~_PAGE_RW
) &
951 /* Init-time set_pte while constructing initial pagetables, which
952 doesn't allow RO pagetable pages to be remapped RW */
953 static __init
void xen_set_pte_init(pte_t
*ptep
, pte_t pte
)
955 pte
= mask_rw_pte(ptep
, pte
);
957 xen_set_pte(ptep
, pte
);
960 static __init
void xen_pagetable_setup_start(pgd_t
*base
)
964 void xen_setup_shared_info(void)
966 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
967 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
968 xen_start_info
->shared_info
);
970 HYPERVISOR_shared_info
=
971 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
973 HYPERVISOR_shared_info
=
974 (struct shared_info
*)__va(xen_start_info
->shared_info
);
977 /* In UP this is as good a place as any to set up shared info */
978 xen_setup_vcpu_info_placement();
981 xen_setup_mfn_list_list();
984 static __init
void xen_pagetable_setup_done(pgd_t
*base
)
986 xen_setup_shared_info();
989 static __init
void xen_post_allocator_init(void)
991 pv_mmu_ops
.set_pte
= xen_set_pte
;
992 pv_mmu_ops
.set_pmd
= xen_set_pmd
;
993 pv_mmu_ops
.set_pud
= xen_set_pud
;
994 #if PAGETABLE_LEVELS == 4
995 pv_mmu_ops
.set_pgd
= xen_set_pgd
;
998 /* This will work as long as patching hasn't happened yet
1000 pv_mmu_ops
.alloc_pte
= xen_alloc_pte
;
1001 pv_mmu_ops
.alloc_pmd
= xen_alloc_pmd
;
1002 pv_mmu_ops
.release_pte
= xen_release_pte
;
1003 pv_mmu_ops
.release_pmd
= xen_release_pmd
;
1004 #if PAGETABLE_LEVELS == 4
1005 pv_mmu_ops
.alloc_pud
= xen_alloc_pud
;
1006 pv_mmu_ops
.release_pud
= xen_release_pud
;
1009 #ifdef CONFIG_X86_64
1010 SetPagePinned(virt_to_page(level3_user_vsyscall
));
1012 xen_mark_init_mm_pinned();
1015 /* This is called once we have the cpu_possible_map */
1016 void xen_setup_vcpu_info_placement(void)
1020 for_each_possible_cpu(cpu
)
1021 xen_vcpu_setup(cpu
);
1023 /* xen_vcpu_setup managed to place the vcpu_info within the
1024 percpu area for all cpus, so make use of it */
1025 #ifdef CONFIG_X86_32
1026 if (have_vcpu_info_placement
) {
1027 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
1029 pv_irq_ops
.save_fl
= xen_save_fl_direct
;
1030 pv_irq_ops
.restore_fl
= xen_restore_fl_direct
;
1031 pv_irq_ops
.irq_disable
= xen_irq_disable_direct
;
1032 pv_irq_ops
.irq_enable
= xen_irq_enable_direct
;
1033 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1038 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1039 unsigned long addr
, unsigned len
)
1041 char *start
, *end
, *reloc
;
1044 start
= end
= reloc
= NULL
;
1046 #define SITE(op, x) \
1047 case PARAVIRT_PATCH(op.x): \
1048 if (have_vcpu_info_placement) { \
1049 start = (char *)xen_##x##_direct; \
1050 end = xen_##x##_direct_end; \
1051 reloc = xen_##x##_direct_reloc; \
1056 #ifdef CONFIG_X86_32
1057 SITE(pv_irq_ops
, irq_enable
);
1058 SITE(pv_irq_ops
, irq_disable
);
1059 SITE(pv_irq_ops
, save_fl
);
1060 SITE(pv_irq_ops
, restore_fl
);
1061 #endif /* CONFIG_X86_32 */
1065 if (start
== NULL
|| (end
-start
) > len
)
1068 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1070 /* Note: because reloc is assigned from something that
1071 appears to be an array, gcc assumes it's non-null,
1072 but doesn't know its relationship with start and
1074 if (reloc
> start
&& reloc
< end
) {
1075 int reloc_off
= reloc
- start
;
1076 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1077 long delta
= start
- (char *)addr
;
1085 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1093 static void xen_set_fixmap(unsigned idx
, unsigned long phys
, pgprot_t prot
)
1097 phys
>>= PAGE_SHIFT
;
1100 case FIX_BTMAP_END
... FIX_BTMAP_BEGIN
:
1101 #ifdef CONFIG_X86_F00F_BUG
1104 #ifdef CONFIG_X86_32
1107 # ifdef CONFIG_HIGHMEM
1108 case FIX_KMAP_BEGIN
... FIX_KMAP_END
:
1111 case VSYSCALL_LAST_PAGE
... VSYSCALL_FIRST_PAGE
:
1113 #ifdef CONFIG_X86_LOCAL_APIC
1114 case FIX_APIC_BASE
: /* maps dummy local APIC */
1116 pte
= pfn_pte(phys
, prot
);
1120 pte
= mfn_pte(phys
, prot
);
1124 __native_set_fixmap(idx
, pte
);
1126 #ifdef CONFIG_X86_64
1127 /* Replicate changes to map the vsyscall page into the user
1128 pagetable vsyscall mapping. */
1129 if (idx
>= VSYSCALL_LAST_PAGE
&& idx
<= VSYSCALL_FIRST_PAGE
) {
1130 unsigned long vaddr
= __fix_to_virt(idx
);
1131 set_pte_vaddr_pud(level3_user_vsyscall
, vaddr
, pte
);
1136 static const struct pv_info xen_info __initdata
= {
1137 .paravirt_enabled
= 1,
1138 .shared_kernel_pmd
= 0,
1143 static const struct pv_init_ops xen_init_ops __initdata
= {
1146 .banner
= xen_banner
,
1147 .memory_setup
= xen_memory_setup
,
1148 .arch_setup
= xen_arch_setup
,
1149 .post_allocator_init
= xen_post_allocator_init
,
1152 static const struct pv_time_ops xen_time_ops __initdata
= {
1153 .time_init
= xen_time_init
,
1155 .set_wallclock
= xen_set_wallclock
,
1156 .get_wallclock
= xen_get_wallclock
,
1157 .get_tsc_khz
= xen_tsc_khz
,
1158 .sched_clock
= xen_sched_clock
,
1161 static const struct pv_cpu_ops xen_cpu_ops __initdata
= {
1164 .set_debugreg
= xen_set_debugreg
,
1165 .get_debugreg
= xen_get_debugreg
,
1169 .read_cr0
= native_read_cr0
,
1170 .write_cr0
= xen_write_cr0
,
1172 .read_cr4
= native_read_cr4
,
1173 .read_cr4_safe
= native_read_cr4_safe
,
1174 .write_cr4
= xen_write_cr4
,
1176 .wbinvd
= native_wbinvd
,
1178 .read_msr
= native_read_msr_safe
,
1179 .write_msr
= xen_write_msr_safe
,
1180 .read_tsc
= native_read_tsc
,
1181 .read_pmc
= native_read_pmc
,
1184 .irq_enable_sysexit
= xen_sysexit
,
1185 #ifdef CONFIG_X86_64
1186 .usergs_sysret32
= xen_sysret32
,
1187 .usergs_sysret64
= xen_sysret64
,
1190 .load_tr_desc
= paravirt_nop
,
1191 .set_ldt
= xen_set_ldt
,
1192 .load_gdt
= xen_load_gdt
,
1193 .load_idt
= xen_load_idt
,
1194 .load_tls
= xen_load_tls
,
1195 #ifdef CONFIG_X86_64
1196 .load_gs_index
= xen_load_gs_index
,
1199 .alloc_ldt
= xen_alloc_ldt
,
1200 .free_ldt
= xen_free_ldt
,
1202 .store_gdt
= native_store_gdt
,
1203 .store_idt
= native_store_idt
,
1204 .store_tr
= xen_store_tr
,
1206 .write_ldt_entry
= xen_write_ldt_entry
,
1207 .write_gdt_entry
= xen_write_gdt_entry
,
1208 .write_idt_entry
= xen_write_idt_entry
,
1209 .load_sp0
= xen_load_sp0
,
1211 .set_iopl_mask
= xen_set_iopl_mask
,
1212 .io_delay
= xen_io_delay
,
1214 /* Xen takes care of %gs when switching to usermode for us */
1215 .swapgs
= paravirt_nop
,
1218 .enter
= paravirt_enter_lazy_cpu
,
1219 .leave
= xen_leave_lazy
,
1223 static const struct pv_apic_ops xen_apic_ops __initdata
= {
1224 #ifdef CONFIG_X86_LOCAL_APIC
1225 .apic_write
= xen_apic_write
,
1226 .apic_read
= xen_apic_read
,
1227 .setup_boot_clock
= paravirt_nop
,
1228 .setup_secondary_clock
= paravirt_nop
,
1229 .startup_ipi_hook
= paravirt_nop
,
1233 static const struct pv_mmu_ops xen_mmu_ops __initdata
= {
1234 .pagetable_setup_start
= xen_pagetable_setup_start
,
1235 .pagetable_setup_done
= xen_pagetable_setup_done
,
1237 .read_cr2
= xen_read_cr2
,
1238 .write_cr2
= xen_write_cr2
,
1240 .read_cr3
= xen_read_cr3
,
1241 .write_cr3
= xen_write_cr3
,
1243 .flush_tlb_user
= xen_flush_tlb
,
1244 .flush_tlb_kernel
= xen_flush_tlb
,
1245 .flush_tlb_single
= xen_flush_tlb_single
,
1246 .flush_tlb_others
= xen_flush_tlb_others
,
1248 .pte_update
= paravirt_nop
,
1249 .pte_update_defer
= paravirt_nop
,
1251 .pgd_alloc
= xen_pgd_alloc
,
1252 .pgd_free
= xen_pgd_free
,
1254 .alloc_pte
= xen_alloc_pte_init
,
1255 .release_pte
= xen_release_pte_init
,
1256 .alloc_pmd
= xen_alloc_pte_init
,
1257 .alloc_pmd_clone
= paravirt_nop
,
1258 .release_pmd
= xen_release_pte_init
,
1260 #ifdef CONFIG_HIGHPTE
1261 .kmap_atomic_pte
= xen_kmap_atomic_pte
,
1264 #ifdef CONFIG_X86_64
1265 .set_pte
= xen_set_pte
,
1267 .set_pte
= xen_set_pte_init
,
1269 .set_pte_at
= xen_set_pte_at
,
1270 .set_pmd
= xen_set_pmd_hyper
,
1272 .ptep_modify_prot_start
= __ptep_modify_prot_start
,
1273 .ptep_modify_prot_commit
= __ptep_modify_prot_commit
,
1275 .pte_val
= xen_pte_val
,
1276 .pte_flags
= native_pte_flags
,
1277 .pgd_val
= xen_pgd_val
,
1279 .make_pte
= xen_make_pte
,
1280 .make_pgd
= xen_make_pgd
,
1282 #ifdef CONFIG_X86_PAE
1283 .set_pte_atomic
= xen_set_pte_atomic
,
1284 .set_pte_present
= xen_set_pte_at
,
1285 .pte_clear
= xen_pte_clear
,
1286 .pmd_clear
= xen_pmd_clear
,
1287 #endif /* CONFIG_X86_PAE */
1288 .set_pud
= xen_set_pud_hyper
,
1290 .make_pmd
= xen_make_pmd
,
1291 .pmd_val
= xen_pmd_val
,
1293 #if PAGETABLE_LEVELS == 4
1294 .pud_val
= xen_pud_val
,
1295 .make_pud
= xen_make_pud
,
1296 .set_pgd
= xen_set_pgd_hyper
,
1298 .alloc_pud
= xen_alloc_pte_init
,
1299 .release_pud
= xen_release_pte_init
,
1300 #endif /* PAGETABLE_LEVELS == 4 */
1302 .activate_mm
= xen_activate_mm
,
1303 .dup_mmap
= xen_dup_mmap
,
1304 .exit_mmap
= xen_exit_mmap
,
1307 .enter
= paravirt_enter_lazy_mmu
,
1308 .leave
= xen_leave_lazy
,
1311 .set_fixmap
= xen_set_fixmap
,
1314 static void xen_reboot(int reason
)
1316 struct sched_shutdown r
= { .reason
= reason
};
1322 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1326 static void xen_restart(char *msg
)
1328 xen_reboot(SHUTDOWN_reboot
);
1331 static void xen_emergency_restart(void)
1333 xen_reboot(SHUTDOWN_reboot
);
1336 static void xen_machine_halt(void)
1338 xen_reboot(SHUTDOWN_poweroff
);
1341 static void xen_crash_shutdown(struct pt_regs
*regs
)
1343 xen_reboot(SHUTDOWN_crash
);
1346 static const struct machine_ops __initdata xen_machine_ops
= {
1347 .restart
= xen_restart
,
1348 .halt
= xen_machine_halt
,
1349 .power_off
= xen_machine_halt
,
1350 .shutdown
= xen_machine_halt
,
1351 .crash_shutdown
= xen_crash_shutdown
,
1352 .emergency_restart
= xen_emergency_restart
,
1356 static void __init
xen_reserve_top(void)
1358 #ifdef CONFIG_X86_32
1359 unsigned long top
= HYPERVISOR_VIRT_START
;
1360 struct xen_platform_parameters pp
;
1362 if (HYPERVISOR_xen_version(XENVER_platform_parameters
, &pp
) == 0)
1363 top
= pp
.virt_start
;
1365 reserve_top_address(-top
+ 2 * PAGE_SIZE
);
1366 #endif /* CONFIG_X86_32 */
1370 * Like __va(), but returns address in the kernel mapping (which is
1371 * all we have until the physical memory mapping has been set up.
1373 static void *__ka(phys_addr_t paddr
)
1375 #ifdef CONFIG_X86_64
1376 return (void *)(paddr
+ __START_KERNEL_map
);
1382 /* Convert a machine address to physical address */
1383 static unsigned long m2p(phys_addr_t maddr
)
1387 maddr
&= PTE_PFN_MASK
;
1388 paddr
= mfn_to_pfn(maddr
>> PAGE_SHIFT
) << PAGE_SHIFT
;
1393 /* Convert a machine address to kernel virtual */
1394 static void *m2v(phys_addr_t maddr
)
1396 return __ka(m2p(maddr
));
1399 #ifdef CONFIG_X86_64
1400 static void walk(pgd_t
*pgd
, unsigned long addr
)
1402 unsigned l4idx
= pgd_index(addr
);
1403 unsigned l3idx
= pud_index(addr
);
1404 unsigned l2idx
= pmd_index(addr
);
1405 unsigned l1idx
= pte_index(addr
);
1411 xen_raw_printk("walk %p, %lx -> %d %d %d %d\n",
1412 pgd
, addr
, l4idx
, l3idx
, l2idx
, l1idx
);
1415 xen_raw_printk(" l4: %016lx\n", l4
.pgd
);
1416 xen_raw_printk(" %016lx\n", pgd_val(l4
));
1418 l3
= ((pud_t
*)(m2v(l4
.pgd
)))[l3idx
];
1419 xen_raw_printk(" l3: %016lx\n", l3
.pud
);
1420 xen_raw_printk(" %016lx\n", pud_val(l3
));
1422 l2
= ((pmd_t
*)(m2v(l3
.pud
)))[l2idx
];
1423 xen_raw_printk(" l2: %016lx\n", l2
.pmd
);
1424 xen_raw_printk(" %016lx\n", pmd_val(l2
));
1426 l1
= ((pte_t
*)(m2v(l2
.pmd
)))[l1idx
];
1427 xen_raw_printk(" l1: %016lx\n", l1
.pte
);
1428 xen_raw_printk(" %016lx\n", pte_val(l1
));
1432 static void set_page_prot(void *addr
, pgprot_t prot
)
1434 unsigned long pfn
= __pa(addr
) >> PAGE_SHIFT
;
1435 pte_t pte
= pfn_pte(pfn
, prot
);
1437 xen_raw_printk("addr=%p pfn=%lx mfn=%lx prot=%016llx pte=%016llx\n",
1438 addr
, pfn
, get_phys_to_machine(pfn
),
1439 pgprot_val(prot
), pte
.pte
);
1441 if (HYPERVISOR_update_va_mapping((unsigned long)addr
, pte
, 0))
1445 static __init
void xen_map_identity_early(pmd_t
*pmd
, unsigned long max_pfn
)
1447 unsigned pmdidx
, pteidx
;
1453 for(pmdidx
= 0; pmdidx
< PTRS_PER_PMD
&& pfn
< max_pfn
; pmdidx
++) {
1456 /* Reuse or allocate a page of ptes */
1457 if (pmd_present(pmd
[pmdidx
]))
1458 pte_page
= m2v(pmd
[pmdidx
].pmd
);
1460 /* Check for free pte pages */
1461 if (ident_pte
== ARRAY_SIZE(level1_ident_pgt
))
1464 pte_page
= &level1_ident_pgt
[ident_pte
];
1465 ident_pte
+= PTRS_PER_PTE
;
1467 pmd
[pmdidx
] = __pmd(__pa(pte_page
) | _PAGE_TABLE
);
1470 /* Install mappings */
1471 for(pteidx
= 0; pteidx
< PTRS_PER_PTE
; pteidx
++, pfn
++) {
1474 if (pfn
> max_pfn_mapped
)
1475 max_pfn_mapped
= pfn
;
1477 if (!pte_none(pte_page
[pteidx
]))
1480 pte
= pfn_pte(pfn
, PAGE_KERNEL_EXEC
);
1481 pte_page
[pteidx
] = pte
;
1485 for(pteidx
= 0; pteidx
< ident_pte
; pteidx
+= PTRS_PER_PTE
)
1486 set_page_prot(&level1_ident_pgt
[pteidx
], PAGE_KERNEL_RO
);
1488 set_page_prot(pmd
, PAGE_KERNEL_RO
);
1491 #ifdef CONFIG_X86_64
1492 static void convert_pfn_mfn(void *v
)
1497 /* All levels are converted the same way, so just treat them
1499 for(i
= 0; i
< PTRS_PER_PTE
; i
++)
1500 pte
[i
] = xen_make_pte(pte
[i
].pte
);
1504 * Set up the inital kernel pagetable.
1506 * We can construct this by grafting the Xen provided pagetable into
1507 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1508 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1509 * means that only the kernel has a physical mapping to start with -
1510 * but that's enough to get __va working. We need to fill in the rest
1511 * of the physical mapping once some sort of allocator has been set
1514 static __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
, unsigned long max_pfn
)
1519 /* Zap identity mapping */
1520 init_level4_pgt
[0] = __pgd(0);
1522 /* Pre-constructed entries are in pfn, so convert to mfn */
1523 convert_pfn_mfn(init_level4_pgt
);
1524 convert_pfn_mfn(level3_ident_pgt
);
1525 convert_pfn_mfn(level3_kernel_pgt
);
1527 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
)].pgd
);
1528 l2
= m2v(l3
[pud_index(__START_KERNEL_map
)].pud
);
1530 memcpy(level2_ident_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1531 memcpy(level2_kernel_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1533 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
+ PMD_SIZE
)].pgd
);
1534 l2
= m2v(l3
[pud_index(__START_KERNEL_map
+ PMD_SIZE
)].pud
);
1535 memcpy(level2_fixmap_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1537 /* Set up identity map */
1538 xen_map_identity_early(level2_ident_pgt
, max_pfn
);
1540 /* Make pagetable pieces RO */
1541 set_page_prot(init_level4_pgt
, PAGE_KERNEL_RO
);
1542 set_page_prot(level3_ident_pgt
, PAGE_KERNEL_RO
);
1543 set_page_prot(level3_kernel_pgt
, PAGE_KERNEL_RO
);
1544 set_page_prot(level3_user_vsyscall
, PAGE_KERNEL_RO
);
1545 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1546 set_page_prot(level2_fixmap_pgt
, PAGE_KERNEL_RO
);
1548 /* Pin down new L4 */
1549 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE
,
1550 PFN_DOWN(__pa_symbol(init_level4_pgt
)));
1552 /* Unpin Xen-provided one */
1553 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1556 pgd
= init_level4_pgt
;
1559 * At this stage there can be no user pgd, and no page
1560 * structure to attach it to, so make sure we just set kernel
1564 __xen_write_cr3(true, __pa(pgd
));
1565 xen_mc_issue(PARAVIRT_LAZY_CPU
);
1567 reserve_early(__pa(xen_start_info
->pt_base
),
1568 __pa(xen_start_info
->pt_base
+
1569 xen_start_info
->nr_pt_frames
* PAGE_SIZE
),
1574 #else /* !CONFIG_X86_64 */
1575 static pmd_t level2_kernel_pgt
[PTRS_PER_PMD
] __page_aligned_bss
;
1577 static __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
, unsigned long max_pfn
)
1581 init_pg_tables_start
= __pa(pgd
);
1582 init_pg_tables_end
= __pa(pgd
) + xen_start_info
->nr_pt_frames
*PAGE_SIZE
;
1583 max_pfn_mapped
= PFN_DOWN(init_pg_tables_end
+ 512*1024);
1585 kernel_pmd
= m2v(pgd
[KERNEL_PGD_BOUNDARY
].pgd
);
1586 memcpy(level2_kernel_pgt
, kernel_pmd
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1588 xen_map_identity_early(level2_kernel_pgt
, max_pfn
);
1590 memcpy(swapper_pg_dir
, pgd
, sizeof(pgd_t
) * PTRS_PER_PGD
);
1591 set_pgd(&swapper_pg_dir
[KERNEL_PGD_BOUNDARY
],
1592 __pgd(__pa(level2_kernel_pgt
) | _PAGE_PRESENT
));
1594 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1595 set_page_prot(swapper_pg_dir
, PAGE_KERNEL_RO
);
1596 set_page_prot(empty_zero_page
, PAGE_KERNEL_RO
);
1598 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1600 xen_write_cr3(__pa(swapper_pg_dir
));
1602 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE
, PFN_DOWN(__pa(swapper_pg_dir
)));
1604 return swapper_pg_dir
;
1606 #endif /* CONFIG_X86_64 */
1608 /* First C function to be called on Xen boot */
1609 asmlinkage
void __init
xen_start_kernel(void)
1613 if (!xen_start_info
)
1616 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3", 5) != 0);
1618 xen_setup_features();
1620 /* Install Xen paravirt ops */
1622 pv_init_ops
= xen_init_ops
;
1623 pv_time_ops
= xen_time_ops
;
1624 pv_cpu_ops
= xen_cpu_ops
;
1625 pv_apic_ops
= xen_apic_ops
;
1626 pv_mmu_ops
= xen_mmu_ops
;
1630 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1631 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1632 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1635 machine_ops
= xen_machine_ops
;
1637 #ifdef CONFIG_X86_64
1638 /* Disable until direct per-cpu data access. */
1639 have_vcpu_info_placement
= 0;
1646 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1647 xen_build_dynamic_phys_to_machine();
1649 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1651 /* Prevent unwanted bits from being set in PTEs. */
1652 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1653 if (!is_initial_xendomain())
1654 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1656 /* Don't do the full vcpu_info placement stuff until we have a
1657 possible map and a non-dummy shared_info. */
1658 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1660 xen_raw_console_write("mapping kernel into physical memory\n");
1661 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
1665 /* keep using Xen gdt for now; no urgent need to change it */
1667 pv_info
.kernel_rpl
= 1;
1668 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1669 pv_info
.kernel_rpl
= 0;
1671 /* set the limit of our address space */
1674 #ifdef CONFIG_X86_32
1675 /* set up basic CPUID stuff */
1676 cpu_detect(&new_cpu_data
);
1677 new_cpu_data
.hard_math
= 1;
1678 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1681 /* Poke various useful things into boot_params */
1682 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1683 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1684 ? __pa(xen_start_info
->mod_start
) : 0;
1685 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1686 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1688 if (!is_initial_xendomain()) {
1689 add_preferred_console("xenboot", 0, NULL
);
1690 add_preferred_console("tty", 0, NULL
);
1691 add_preferred_console("hvc", 0, NULL
);
1694 xen_raw_console_write("about to get started...\n");
1697 xen_raw_printk("&boot_params=%p __pa(&boot_params)=%lx __va(__pa(&boot_params))=%lx\n",
1698 &boot_params
, __pa_symbol(&boot_params
),
1699 __va(__pa_symbol(&boot_params
)));
1701 walk(pgd
, &boot_params
);
1702 walk(pgd
, __va(__pa(&boot_params
)));
1705 /* Start the world */
1706 #ifdef CONFIG_X86_32
1707 i386_start_kernel();
1709 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));