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87 * out of date, but it will never be set early).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 cpumask_t
*cpus
, struct mm_struct
*mm
,
644 cpumask_t cpumask
= *cpus
;
645 struct multicall_space mcs
;
648 * A couple of (to be removed) sanity checks:
650 * - current CPU must not be in mask
651 * - mask must exist :)
653 BUG_ON(cpus_empty(cpumask
));
654 BUG_ON(cpu_isset(smp_processor_id(), cpumask
));
657 /* If a CPU which we ran on has gone down, OK. */
658 cpus_and(cpumask
, cpumask
, cpu_online_map
);
659 if (cpus_empty(cpumask
))
662 mcs
= xen_mc_entry(sizeof(*args
));
664 args
->mask
= cpumask
;
665 args
->op
.arg2
.vcpumask
= &args
->mask
;
667 if (va
== TLB_FLUSH_ALL
) {
668 args
->op
.cmd
= MMUEXT_TLB_FLUSH_MULTI
;
670 args
->op
.cmd
= MMUEXT_INVLPG_MULTI
;
671 args
->op
.arg1
.linear_addr
= va
;
674 MULTI_mmuext_op(mcs
.mc
, &args
->op
, 1, NULL
, DOMID_SELF
);
676 xen_mc_issue(PARAVIRT_LAZY_MMU
);
679 static void xen_clts(void)
681 struct multicall_space mcs
;
683 mcs
= xen_mc_entry(0);
685 MULTI_fpu_taskswitch(mcs
.mc
, 0);
687 xen_mc_issue(PARAVIRT_LAZY_CPU
);
690 static void xen_write_cr0(unsigned long cr0
)
692 struct multicall_space mcs
;
694 /* Only pay attention to cr0.TS; everything else is
696 mcs
= xen_mc_entry(0);
698 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
700 xen_mc_issue(PARAVIRT_LAZY_CPU
);
703 static void xen_write_cr2(unsigned long cr2
)
705 x86_read_percpu(xen_vcpu
)->arch
.cr2
= cr2
;
708 static unsigned long xen_read_cr2(void)
710 return x86_read_percpu(xen_vcpu
)->arch
.cr2
;
713 static unsigned long xen_read_cr2_direct(void)
715 return x86_read_percpu(xen_vcpu_info
.arch
.cr2
);
718 static void xen_write_cr4(unsigned long cr4
)
723 native_write_cr4(cr4
);
726 static unsigned long xen_read_cr3(void)
728 return x86_read_percpu(xen_cr3
);
731 static void set_current_cr3(void *v
)
733 x86_write_percpu(xen_current_cr3
, (unsigned long)v
);
736 static void __xen_write_cr3(bool kernel
, unsigned long cr3
)
738 struct mmuext_op
*op
;
739 struct multicall_space mcs
;
743 mfn
= pfn_to_mfn(PFN_DOWN(cr3
));
747 WARN_ON(mfn
== 0 && kernel
);
749 mcs
= __xen_mc_entry(sizeof(*op
));
752 op
->cmd
= kernel
? MMUEXT_NEW_BASEPTR
: MMUEXT_NEW_USER_BASEPTR
;
755 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
758 x86_write_percpu(xen_cr3
, cr3
);
760 /* Update xen_current_cr3 once the batch has actually
762 xen_mc_callback(set_current_cr3
, (void *)cr3
);
766 static void xen_write_cr3(unsigned long cr3
)
768 BUG_ON(preemptible());
770 xen_mc_batch(); /* disables interrupts */
772 /* Update while interrupts are disabled, so its atomic with
774 x86_write_percpu(xen_cr3
, cr3
);
776 __xen_write_cr3(true, cr3
);
780 pgd_t
*user_pgd
= xen_get_user_pgd(__va(cr3
));
782 __xen_write_cr3(false, __pa(user_pgd
));
784 __xen_write_cr3(false, 0);
788 xen_mc_issue(PARAVIRT_LAZY_CPU
); /* interrupts restored */
791 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
802 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
803 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
804 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
807 base
= ((u64
)high
<< 32) | low
;
808 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
816 case MSR_SYSCALL_MASK
:
817 case MSR_IA32_SYSENTER_CS
:
818 case MSR_IA32_SYSENTER_ESP
:
819 case MSR_IA32_SYSENTER_EIP
:
820 /* Fast syscall setup is all done in hypercalls, so
821 these are all ignored. Stub them out here to stop
822 Xen console noise. */
826 ret
= native_write_msr_safe(msr
, low
, high
);
832 /* Early in boot, while setting up the initial pagetable, assume
833 everything is pinned. */
834 static __init
void xen_alloc_pte_init(struct mm_struct
*mm
, unsigned long pfn
)
836 #ifdef CONFIG_FLATMEM
837 BUG_ON(mem_map
); /* should only be used early */
839 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
842 /* Early release_pte assumes that all pts are pinned, since there's
843 only init_mm and anything attached to that is pinned. */
844 static void xen_release_pte_init(unsigned long pfn
)
846 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
849 static void pin_pagetable_pfn(unsigned cmd
, unsigned long pfn
)
853 op
.arg1
.mfn
= pfn_to_mfn(pfn
);
854 if (HYPERVISOR_mmuext_op(&op
, 1, NULL
, DOMID_SELF
))
858 /* This needs to make sure the new pte page is pinned iff its being
859 attached to a pinned pagetable. */
860 static void xen_alloc_ptpage(struct mm_struct
*mm
, unsigned long pfn
, unsigned level
)
862 struct page
*page
= pfn_to_page(pfn
);
864 if (PagePinned(virt_to_page(mm
->pgd
))) {
868 if (!PageHighMem(page
)) {
869 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn
)));
870 if (level
== PT_PTE
&& USE_SPLIT_PTLOCKS
)
871 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE
, pfn
);
873 /* make sure there are no stray mappings of
880 static void xen_alloc_pte(struct mm_struct
*mm
, unsigned long pfn
)
882 xen_alloc_ptpage(mm
, pfn
, PT_PTE
);
885 static void xen_alloc_pmd(struct mm_struct
*mm
, unsigned long pfn
)
887 xen_alloc_ptpage(mm
, pfn
, PT_PMD
);
890 static int xen_pgd_alloc(struct mm_struct
*mm
)
892 pgd_t
*pgd
= mm
->pgd
;
895 BUG_ON(PagePinned(virt_to_page(pgd
)));
899 struct page
*page
= virt_to_page(pgd
);
902 BUG_ON(page
->private != 0);
906 user_pgd
= (pgd_t
*)__get_free_page(GFP_KERNEL
| __GFP_ZERO
);
907 page
->private = (unsigned long)user_pgd
;
909 if (user_pgd
!= NULL
) {
910 user_pgd
[pgd_index(VSYSCALL_START
)] =
911 __pgd(__pa(level3_user_vsyscall
) | _PAGE_TABLE
);
915 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd
))));
922 static void xen_pgd_free(struct mm_struct
*mm
, pgd_t
*pgd
)
925 pgd_t
*user_pgd
= xen_get_user_pgd(pgd
);
928 free_page((unsigned long)user_pgd
);
932 /* This should never happen until we're OK to use struct page */
933 static void xen_release_ptpage(unsigned long pfn
, unsigned level
)
935 struct page
*page
= pfn_to_page(pfn
);
937 if (PagePinned(page
)) {
938 if (!PageHighMem(page
)) {
939 if (level
== PT_PTE
&& USE_SPLIT_PTLOCKS
)
940 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, pfn
);
941 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
943 ClearPagePinned(page
);
947 static void xen_release_pte(unsigned long pfn
)
949 xen_release_ptpage(pfn
, PT_PTE
);
952 static void xen_release_pmd(unsigned long pfn
)
954 xen_release_ptpage(pfn
, PT_PMD
);
957 #if PAGETABLE_LEVELS == 4
958 static void xen_alloc_pud(struct mm_struct
*mm
, unsigned long pfn
)
960 xen_alloc_ptpage(mm
, pfn
, PT_PUD
);
963 static void xen_release_pud(unsigned long pfn
)
965 xen_release_ptpage(pfn
, PT_PUD
);
969 #ifdef CONFIG_HIGHPTE
970 static void *xen_kmap_atomic_pte(struct page
*page
, enum km_type type
)
972 pgprot_t prot
= PAGE_KERNEL
;
974 if (PagePinned(page
))
975 prot
= PAGE_KERNEL_RO
;
977 if (0 && PageHighMem(page
))
978 printk("mapping highpte %lx type %d prot %s\n",
979 page_to_pfn(page
), type
,
980 (unsigned long)pgprot_val(prot
) & _PAGE_RW
? "WRITE" : "READ");
982 return kmap_atomic_prot(page
, type
, prot
);
987 static __init pte_t
mask_rw_pte(pte_t
*ptep
, pte_t pte
)
989 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
990 if (pte_val_ma(*ptep
) & _PAGE_PRESENT
)
991 pte
= __pte_ma(((pte_val_ma(*ptep
) & _PAGE_RW
) | ~_PAGE_RW
) &
997 /* Init-time set_pte while constructing initial pagetables, which
998 doesn't allow RO pagetable pages to be remapped RW */
999 static __init
void xen_set_pte_init(pte_t
*ptep
, pte_t pte
)
1001 pte
= mask_rw_pte(ptep
, pte
);
1003 xen_set_pte(ptep
, pte
);
1007 static __init
void xen_pagetable_setup_start(pgd_t
*base
)
1011 void xen_setup_shared_info(void)
1013 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1014 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
1015 xen_start_info
->shared_info
);
1017 HYPERVISOR_shared_info
=
1018 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
1020 HYPERVISOR_shared_info
=
1021 (struct shared_info
*)__va(xen_start_info
->shared_info
);
1024 /* In UP this is as good a place as any to set up shared info */
1025 xen_setup_vcpu_info_placement();
1028 xen_setup_mfn_list_list();
1031 static __init
void xen_pagetable_setup_done(pgd_t
*base
)
1033 xen_setup_shared_info();
1036 static __init
void xen_post_allocator_init(void)
1038 pv_mmu_ops
.set_pte
= xen_set_pte
;
1039 pv_mmu_ops
.set_pmd
= xen_set_pmd
;
1040 pv_mmu_ops
.set_pud
= xen_set_pud
;
1041 #if PAGETABLE_LEVELS == 4
1042 pv_mmu_ops
.set_pgd
= xen_set_pgd
;
1045 /* This will work as long as patching hasn't happened yet
1046 (which it hasn't) */
1047 pv_mmu_ops
.alloc_pte
= xen_alloc_pte
;
1048 pv_mmu_ops
.alloc_pmd
= xen_alloc_pmd
;
1049 pv_mmu_ops
.release_pte
= xen_release_pte
;
1050 pv_mmu_ops
.release_pmd
= xen_release_pmd
;
1051 #if PAGETABLE_LEVELS == 4
1052 pv_mmu_ops
.alloc_pud
= xen_alloc_pud
;
1053 pv_mmu_ops
.release_pud
= xen_release_pud
;
1056 #ifdef CONFIG_X86_64
1057 SetPagePinned(virt_to_page(level3_user_vsyscall
));
1059 xen_mark_init_mm_pinned();
1062 /* This is called once we have the cpu_possible_map */
1063 void xen_setup_vcpu_info_placement(void)
1067 for_each_possible_cpu(cpu
)
1068 xen_vcpu_setup(cpu
);
1070 /* xen_vcpu_setup managed to place the vcpu_info within the
1071 percpu area for all cpus, so make use of it */
1072 if (have_vcpu_info_placement
) {
1073 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
1075 pv_irq_ops
.save_fl
= xen_save_fl_direct
;
1076 pv_irq_ops
.restore_fl
= xen_restore_fl_direct
;
1077 pv_irq_ops
.irq_disable
= xen_irq_disable_direct
;
1078 pv_irq_ops
.irq_enable
= xen_irq_enable_direct
;
1079 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1083 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1084 unsigned long addr
, unsigned len
)
1086 char *start
, *end
, *reloc
;
1089 start
= end
= reloc
= NULL
;
1091 #define SITE(op, x) \
1092 case PARAVIRT_PATCH(op.x): \
1093 if (have_vcpu_info_placement) { \
1094 start = (char *)xen_##x##_direct; \
1095 end = xen_##x##_direct_end; \
1096 reloc = xen_##x##_direct_reloc; \
1101 SITE(pv_irq_ops
, irq_enable
);
1102 SITE(pv_irq_ops
, irq_disable
);
1103 SITE(pv_irq_ops
, save_fl
);
1104 SITE(pv_irq_ops
, restore_fl
);
1108 if (start
== NULL
|| (end
-start
) > len
)
1111 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1113 /* Note: because reloc is assigned from something that
1114 appears to be an array, gcc assumes it's non-null,
1115 but doesn't know its relationship with start and
1117 if (reloc
> start
&& reloc
< end
) {
1118 int reloc_off
= reloc
- start
;
1119 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1120 long delta
= start
- (char *)addr
;
1128 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1136 static void xen_set_fixmap(unsigned idx
, unsigned long phys
, pgprot_t prot
)
1140 phys
>>= PAGE_SHIFT
;
1143 case FIX_BTMAP_END
... FIX_BTMAP_BEGIN
:
1144 #ifdef CONFIG_X86_F00F_BUG
1147 #ifdef CONFIG_X86_32
1150 # ifdef CONFIG_HIGHMEM
1151 case FIX_KMAP_BEGIN
... FIX_KMAP_END
:
1154 case VSYSCALL_LAST_PAGE
... VSYSCALL_FIRST_PAGE
:
1156 #ifdef CONFIG_X86_LOCAL_APIC
1157 case FIX_APIC_BASE
: /* maps dummy local APIC */
1159 pte
= pfn_pte(phys
, prot
);
1163 pte
= mfn_pte(phys
, prot
);
1167 __native_set_fixmap(idx
, pte
);
1169 #ifdef CONFIG_X86_64
1170 /* Replicate changes to map the vsyscall page into the user
1171 pagetable vsyscall mapping. */
1172 if (idx
>= VSYSCALL_LAST_PAGE
&& idx
<= VSYSCALL_FIRST_PAGE
) {
1173 unsigned long vaddr
= __fix_to_virt(idx
);
1174 set_pte_vaddr_pud(level3_user_vsyscall
, vaddr
, pte
);
1179 static const struct pv_info xen_info __initdata
= {
1180 .paravirt_enabled
= 1,
1181 .shared_kernel_pmd
= 0,
1186 static const struct pv_init_ops xen_init_ops __initdata
= {
1189 .banner
= xen_banner
,
1190 .memory_setup
= xen_memory_setup
,
1191 .arch_setup
= xen_arch_setup
,
1192 .post_allocator_init
= xen_post_allocator_init
,
1195 static const struct pv_time_ops xen_time_ops __initdata
= {
1196 .time_init
= xen_time_init
,
1198 .set_wallclock
= xen_set_wallclock
,
1199 .get_wallclock
= xen_get_wallclock
,
1200 .get_tsc_khz
= xen_tsc_khz
,
1201 .sched_clock
= xen_sched_clock
,
1204 static const struct pv_cpu_ops xen_cpu_ops __initdata
= {
1207 .set_debugreg
= xen_set_debugreg
,
1208 .get_debugreg
= xen_get_debugreg
,
1212 .read_cr0
= native_read_cr0
,
1213 .write_cr0
= xen_write_cr0
,
1215 .read_cr4
= native_read_cr4
,
1216 .read_cr4_safe
= native_read_cr4_safe
,
1217 .write_cr4
= xen_write_cr4
,
1219 .wbinvd
= native_wbinvd
,
1221 .read_msr
= native_read_msr_safe
,
1222 .write_msr
= xen_write_msr_safe
,
1223 .read_tsc
= native_read_tsc
,
1224 .read_pmc
= native_read_pmc
,
1227 .irq_enable_sysexit
= xen_sysexit
,
1228 #ifdef CONFIG_X86_64
1229 .usergs_sysret32
= xen_sysret32
,
1230 .usergs_sysret64
= xen_sysret64
,
1233 .load_tr_desc
= paravirt_nop
,
1234 .set_ldt
= xen_set_ldt
,
1235 .load_gdt
= xen_load_gdt
,
1236 .load_idt
= xen_load_idt
,
1237 .load_tls
= xen_load_tls
,
1238 #ifdef CONFIG_X86_64
1239 .load_gs_index
= xen_load_gs_index
,
1242 .alloc_ldt
= xen_alloc_ldt
,
1243 .free_ldt
= xen_free_ldt
,
1245 .store_gdt
= native_store_gdt
,
1246 .store_idt
= native_store_idt
,
1247 .store_tr
= xen_store_tr
,
1249 .write_ldt_entry
= xen_write_ldt_entry
,
1250 .write_gdt_entry
= xen_write_gdt_entry
,
1251 .write_idt_entry
= xen_write_idt_entry
,
1252 .load_sp0
= xen_load_sp0
,
1254 .set_iopl_mask
= xen_set_iopl_mask
,
1255 .io_delay
= xen_io_delay
,
1257 /* Xen takes care of %gs when switching to usermode for us */
1258 .swapgs
= paravirt_nop
,
1261 .enter
= paravirt_enter_lazy_cpu
,
1262 .leave
= xen_leave_lazy
,
1266 static const struct pv_apic_ops xen_apic_ops __initdata
= {
1267 #ifdef CONFIG_X86_LOCAL_APIC
1268 .setup_boot_clock
= paravirt_nop
,
1269 .setup_secondary_clock
= paravirt_nop
,
1270 .startup_ipi_hook
= paravirt_nop
,
1274 static const struct pv_mmu_ops xen_mmu_ops __initdata
= {
1275 .pagetable_setup_start
= xen_pagetable_setup_start
,
1276 .pagetable_setup_done
= xen_pagetable_setup_done
,
1278 .read_cr2
= xen_read_cr2
,
1279 .write_cr2
= xen_write_cr2
,
1281 .read_cr3
= xen_read_cr3
,
1282 .write_cr3
= xen_write_cr3
,
1284 .flush_tlb_user
= xen_flush_tlb
,
1285 .flush_tlb_kernel
= xen_flush_tlb
,
1286 .flush_tlb_single
= xen_flush_tlb_single
,
1287 .flush_tlb_others
= xen_flush_tlb_others
,
1289 .pte_update
= paravirt_nop
,
1290 .pte_update_defer
= paravirt_nop
,
1292 .pgd_alloc
= xen_pgd_alloc
,
1293 .pgd_free
= xen_pgd_free
,
1295 .alloc_pte
= xen_alloc_pte_init
,
1296 .release_pte
= xen_release_pte_init
,
1297 .alloc_pmd
= xen_alloc_pte_init
,
1298 .alloc_pmd_clone
= paravirt_nop
,
1299 .release_pmd
= xen_release_pte_init
,
1301 #ifdef CONFIG_HIGHPTE
1302 .kmap_atomic_pte
= xen_kmap_atomic_pte
,
1305 #ifdef CONFIG_X86_64
1306 .set_pte
= xen_set_pte
,
1308 .set_pte
= xen_set_pte_init
,
1310 .set_pte_at
= xen_set_pte_at
,
1311 .set_pmd
= xen_set_pmd_hyper
,
1313 .ptep_modify_prot_start
= __ptep_modify_prot_start
,
1314 .ptep_modify_prot_commit
= __ptep_modify_prot_commit
,
1316 .pte_val
= xen_pte_val
,
1317 .pte_flags
= native_pte_flags
,
1318 .pgd_val
= xen_pgd_val
,
1320 .make_pte
= xen_make_pte
,
1321 .make_pgd
= xen_make_pgd
,
1323 #ifdef CONFIG_X86_PAE
1324 .set_pte_atomic
= xen_set_pte_atomic
,
1325 .set_pte_present
= xen_set_pte_at
,
1326 .pte_clear
= xen_pte_clear
,
1327 .pmd_clear
= xen_pmd_clear
,
1328 #endif /* CONFIG_X86_PAE */
1329 .set_pud
= xen_set_pud_hyper
,
1331 .make_pmd
= xen_make_pmd
,
1332 .pmd_val
= xen_pmd_val
,
1334 #if PAGETABLE_LEVELS == 4
1335 .pud_val
= xen_pud_val
,
1336 .make_pud
= xen_make_pud
,
1337 .set_pgd
= xen_set_pgd_hyper
,
1339 .alloc_pud
= xen_alloc_pte_init
,
1340 .release_pud
= xen_release_pte_init
,
1341 #endif /* PAGETABLE_LEVELS == 4 */
1343 .activate_mm
= xen_activate_mm
,
1344 .dup_mmap
= xen_dup_mmap
,
1345 .exit_mmap
= xen_exit_mmap
,
1348 .enter
= paravirt_enter_lazy_mmu
,
1349 .leave
= xen_leave_lazy
,
1352 .set_fixmap
= xen_set_fixmap
,
1355 static void xen_reboot(int reason
)
1357 struct sched_shutdown r
= { .reason
= reason
};
1363 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1367 static void xen_restart(char *msg
)
1369 xen_reboot(SHUTDOWN_reboot
);
1372 static void xen_emergency_restart(void)
1374 xen_reboot(SHUTDOWN_reboot
);
1377 static void xen_machine_halt(void)
1379 xen_reboot(SHUTDOWN_poweroff
);
1382 static void xen_crash_shutdown(struct pt_regs
*regs
)
1384 xen_reboot(SHUTDOWN_crash
);
1387 static const struct machine_ops __initdata xen_machine_ops
= {
1388 .restart
= xen_restart
,
1389 .halt
= xen_machine_halt
,
1390 .power_off
= xen_machine_halt
,
1391 .shutdown
= xen_machine_halt
,
1392 .crash_shutdown
= xen_crash_shutdown
,
1393 .emergency_restart
= xen_emergency_restart
,
1397 static void __init
xen_reserve_top(void)
1399 #ifdef CONFIG_X86_32
1400 unsigned long top
= HYPERVISOR_VIRT_START
;
1401 struct xen_platform_parameters pp
;
1403 if (HYPERVISOR_xen_version(XENVER_platform_parameters
, &pp
) == 0)
1404 top
= pp
.virt_start
;
1406 reserve_top_address(-top
);
1407 #endif /* CONFIG_X86_32 */
1411 * Like __va(), but returns address in the kernel mapping (which is
1412 * all we have until the physical memory mapping has been set up.
1414 static void *__ka(phys_addr_t paddr
)
1416 #ifdef CONFIG_X86_64
1417 return (void *)(paddr
+ __START_KERNEL_map
);
1423 /* Convert a machine address to physical address */
1424 static unsigned long m2p(phys_addr_t maddr
)
1428 maddr
&= PTE_PFN_MASK
;
1429 paddr
= mfn_to_pfn(maddr
>> PAGE_SHIFT
) << PAGE_SHIFT
;
1434 /* Convert a machine address to kernel virtual */
1435 static void *m2v(phys_addr_t maddr
)
1437 return __ka(m2p(maddr
));
1440 static void set_page_prot(void *addr
, pgprot_t prot
)
1442 unsigned long pfn
= __pa(addr
) >> PAGE_SHIFT
;
1443 pte_t pte
= pfn_pte(pfn
, prot
);
1445 if (HYPERVISOR_update_va_mapping((unsigned long)addr
, pte
, 0))
1449 static __init
void xen_map_identity_early(pmd_t
*pmd
, unsigned long max_pfn
)
1451 unsigned pmdidx
, pteidx
;
1457 for (pmdidx
= 0; pmdidx
< PTRS_PER_PMD
&& pfn
< max_pfn
; pmdidx
++) {
1460 /* Reuse or allocate a page of ptes */
1461 if (pmd_present(pmd
[pmdidx
]))
1462 pte_page
= m2v(pmd
[pmdidx
].pmd
);
1464 /* Check for free pte pages */
1465 if (ident_pte
== ARRAY_SIZE(level1_ident_pgt
))
1468 pte_page
= &level1_ident_pgt
[ident_pte
];
1469 ident_pte
+= PTRS_PER_PTE
;
1471 pmd
[pmdidx
] = __pmd(__pa(pte_page
) | _PAGE_TABLE
);
1474 /* Install mappings */
1475 for (pteidx
= 0; pteidx
< PTRS_PER_PTE
; pteidx
++, pfn
++) {
1478 if (pfn
> max_pfn_mapped
)
1479 max_pfn_mapped
= pfn
;
1481 if (!pte_none(pte_page
[pteidx
]))
1484 pte
= pfn_pte(pfn
, PAGE_KERNEL_EXEC
);
1485 pte_page
[pteidx
] = pte
;
1489 for (pteidx
= 0; pteidx
< ident_pte
; pteidx
+= PTRS_PER_PTE
)
1490 set_page_prot(&level1_ident_pgt
[pteidx
], PAGE_KERNEL_RO
);
1492 set_page_prot(pmd
, PAGE_KERNEL_RO
);
1495 #ifdef CONFIG_X86_64
1496 static void convert_pfn_mfn(void *v
)
1501 /* All levels are converted the same way, so just treat them
1503 for (i
= 0; i
< PTRS_PER_PTE
; i
++)
1504 pte
[i
] = xen_make_pte(pte
[i
].pte
);
1508 * Set up the inital kernel pagetable.
1510 * We can construct this by grafting the Xen provided pagetable into
1511 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1512 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1513 * means that only the kernel has a physical mapping to start with -
1514 * but that's enough to get __va working. We need to fill in the rest
1515 * of the physical mapping once some sort of allocator has been set
1518 static __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
,
1519 unsigned long max_pfn
)
1524 /* Zap identity mapping */
1525 init_level4_pgt
[0] = __pgd(0);
1527 /* Pre-constructed entries are in pfn, so convert to mfn */
1528 convert_pfn_mfn(init_level4_pgt
);
1529 convert_pfn_mfn(level3_ident_pgt
);
1530 convert_pfn_mfn(level3_kernel_pgt
);
1532 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
)].pgd
);
1533 l2
= m2v(l3
[pud_index(__START_KERNEL_map
)].pud
);
1535 memcpy(level2_ident_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1536 memcpy(level2_kernel_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1538 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
+ PMD_SIZE
)].pgd
);
1539 l2
= m2v(l3
[pud_index(__START_KERNEL_map
+ PMD_SIZE
)].pud
);
1540 memcpy(level2_fixmap_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1542 /* Set up identity map */
1543 xen_map_identity_early(level2_ident_pgt
, max_pfn
);
1545 /* Make pagetable pieces RO */
1546 set_page_prot(init_level4_pgt
, PAGE_KERNEL_RO
);
1547 set_page_prot(level3_ident_pgt
, PAGE_KERNEL_RO
);
1548 set_page_prot(level3_kernel_pgt
, PAGE_KERNEL_RO
);
1549 set_page_prot(level3_user_vsyscall
, PAGE_KERNEL_RO
);
1550 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1551 set_page_prot(level2_fixmap_pgt
, PAGE_KERNEL_RO
);
1553 /* Pin down new L4 */
1554 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE
,
1555 PFN_DOWN(__pa_symbol(init_level4_pgt
)));
1557 /* Unpin Xen-provided one */
1558 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1561 pgd
= init_level4_pgt
;
1564 * At this stage there can be no user pgd, and no page
1565 * structure to attach it to, so make sure we just set kernel
1569 __xen_write_cr3(true, __pa(pgd
));
1570 xen_mc_issue(PARAVIRT_LAZY_CPU
);
1572 reserve_early(__pa(xen_start_info
->pt_base
),
1573 __pa(xen_start_info
->pt_base
+
1574 xen_start_info
->nr_pt_frames
* PAGE_SIZE
),
1579 #else /* !CONFIG_X86_64 */
1580 static pmd_t level2_kernel_pgt
[PTRS_PER_PMD
] __page_aligned_bss
;
1582 static __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
,
1583 unsigned long max_pfn
)
1587 init_pg_tables_start
= __pa(pgd
);
1588 init_pg_tables_end
= __pa(pgd
) + xen_start_info
->nr_pt_frames
*PAGE_SIZE
;
1589 max_pfn_mapped
= PFN_DOWN(init_pg_tables_end
+ 512*1024);
1591 kernel_pmd
= m2v(pgd
[KERNEL_PGD_BOUNDARY
].pgd
);
1592 memcpy(level2_kernel_pgt
, kernel_pmd
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1594 xen_map_identity_early(level2_kernel_pgt
, max_pfn
);
1596 memcpy(swapper_pg_dir
, pgd
, sizeof(pgd_t
) * PTRS_PER_PGD
);
1597 set_pgd(&swapper_pg_dir
[KERNEL_PGD_BOUNDARY
],
1598 __pgd(__pa(level2_kernel_pgt
) | _PAGE_PRESENT
));
1600 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1601 set_page_prot(swapper_pg_dir
, PAGE_KERNEL_RO
);
1602 set_page_prot(empty_zero_page
, PAGE_KERNEL_RO
);
1604 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1606 xen_write_cr3(__pa(swapper_pg_dir
));
1608 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE
, PFN_DOWN(__pa(swapper_pg_dir
)));
1610 return swapper_pg_dir
;
1612 #endif /* CONFIG_X86_64 */
1614 /* First C function to be called on Xen boot */
1615 asmlinkage
void __init
xen_start_kernel(void)
1619 if (!xen_start_info
)
1622 xen_domain_type
= XEN_PV_DOMAIN
;
1624 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3", 5) != 0);
1626 xen_setup_features();
1628 /* Install Xen paravirt ops */
1630 pv_init_ops
= xen_init_ops
;
1631 pv_time_ops
= xen_time_ops
;
1632 pv_cpu_ops
= xen_cpu_ops
;
1633 pv_apic_ops
= xen_apic_ops
;
1634 pv_mmu_ops
= xen_mmu_ops
;
1638 #ifdef CONFIG_X86_LOCAL_APIC
1640 * set up the basic apic ops.
1642 apic_ops
= &xen_basic_apic_ops
;
1645 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1646 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1647 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1650 machine_ops
= xen_machine_ops
;
1652 #ifdef CONFIG_X86_64
1653 /* Disable until direct per-cpu data access. */
1654 have_vcpu_info_placement
= 0;
1661 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1662 xen_build_dynamic_phys_to_machine();
1664 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1666 /* Prevent unwanted bits from being set in PTEs. */
1667 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1668 if (!xen_initial_domain())
1669 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1671 /* Don't do the full vcpu_info placement stuff until we have a
1672 possible map and a non-dummy shared_info. */
1673 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1675 xen_raw_console_write("mapping kernel into physical memory\n");
1676 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
1680 /* keep using Xen gdt for now; no urgent need to change it */
1682 pv_info
.kernel_rpl
= 1;
1683 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1684 pv_info
.kernel_rpl
= 0;
1686 /* set the limit of our address space */
1689 #ifdef CONFIG_X86_32
1690 /* set up basic CPUID stuff */
1691 cpu_detect(&new_cpu_data
);
1692 new_cpu_data
.hard_math
= 1;
1693 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1696 /* Poke various useful things into boot_params */
1697 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1698 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1699 ? __pa(xen_start_info
->mod_start
) : 0;
1700 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1701 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1703 if (!xen_initial_domain()) {
1704 add_preferred_console("xenboot", 0, NULL
);
1705 add_preferred_console("tty", 0, NULL
);
1706 add_preferred_console("hvc", 0, NULL
);
1709 xen_raw_console_write("about to get started...\n");
1711 /* Start the world */
1712 #ifdef CONFIG_X86_32
1713 i386_start_kernel();
1715 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));