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/interface/sched.h>
34 #include <xen/features.h>
36 #include <xen/hvc-console.h>
38 #include <asm/paravirt.h>
40 #include <asm/xen/hypercall.h>
41 #include <asm/xen/hypervisor.h>
42 #include <asm/fixmap.h>
43 #include <asm/processor.h>
44 #include <asm/msr-index.h>
45 #include <asm/setup.h>
47 #include <asm/pgtable.h>
48 #include <asm/tlbflush.h>
49 #include <asm/reboot.h>
53 #include "multicalls.h"
55 EXPORT_SYMBOL_GPL(hypercall_page
);
57 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
58 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
61 * Identity map, in addition to plain kernel map. This needs to be
62 * large enough to allocate page table pages to allocate the rest.
63 * Each page can map 2MB.
65 static pte_t level1_ident_pgt
[PTRS_PER_PTE
* 4] __page_aligned_bss
;
68 /* l3 pud for userspace vsyscall mapping */
69 static pud_t level3_user_vsyscall
[PTRS_PER_PUD
] __page_aligned_bss
;
70 #endif /* CONFIG_X86_64 */
73 * Note about cr3 (pagetable base) values:
75 * xen_cr3 contains the current logical cr3 value; it contains the
76 * last set cr3. This may not be the current effective cr3, because
77 * its update may be being lazily deferred. However, a vcpu looking
78 * at its own cr3 can use this value knowing that it everything will
81 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
82 * hypercall to set the vcpu cr3 is complete (so it may be a little
83 * out of date, but it will never be set early). If one vcpu is
84 * looking at another vcpu's cr3 value, it should use this variable.
86 DEFINE_PER_CPU(unsigned long, xen_cr3
); /* cr3 stored as physaddr */
87 DEFINE_PER_CPU(unsigned long, xen_current_cr3
); /* actual vcpu cr3 */
89 struct start_info
*xen_start_info
;
90 EXPORT_SYMBOL_GPL(xen_start_info
);
92 struct shared_info xen_dummy_shared_info
;
95 * Point at some empty memory to start with. We map the real shared_info
96 * page as soon as fixmap is up and running.
98 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
101 * Flag to determine whether vcpu info placement is available on all
102 * VCPUs. We assume it is to start with, and then set it to zero on
103 * the first failure. This is because it can succeed on some VCPUs
104 * and not others, since it can involve hypervisor memory allocation,
105 * or because the guest failed to guarantee all the appropriate
106 * constraints on all VCPUs (ie buffer can't cross a page boundary).
108 * Note that any particular CPU may be using a placed vcpu structure,
109 * but we can only optimise if the all are.
111 * 0: not available, 1: available
113 static int have_vcpu_info_placement
= 1;
115 static void xen_vcpu_setup(int cpu
)
117 struct vcpu_register_vcpu_info info
;
119 struct vcpu_info
*vcpup
;
121 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
122 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
124 if (!have_vcpu_info_placement
)
125 return; /* already tested, not available */
127 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
129 info
.mfn
= virt_to_mfn(vcpup
);
130 info
.offset
= offset_in_page(vcpup
);
132 printk(KERN_DEBUG
"trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
133 cpu
, vcpup
, info
.mfn
, info
.offset
);
135 /* Check to see if the hypervisor will put the vcpu_info
136 structure where we want it, which allows direct access via
137 a percpu-variable. */
138 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
141 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
142 have_vcpu_info_placement
= 0;
144 /* This cpu is using the registered vcpu info, even if
145 later ones fail to. */
146 per_cpu(xen_vcpu
, cpu
) = vcpup
;
148 printk(KERN_DEBUG
"cpu %d using vcpu_info at %p\n",
154 * On restore, set the vcpu placement up again.
155 * If it fails, then we're in a bad state, since
156 * we can't back out from using it...
158 void xen_vcpu_restore(void)
160 if (have_vcpu_info_placement
) {
163 for_each_online_cpu(cpu
) {
164 bool other_cpu
= (cpu
!= smp_processor_id());
167 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
173 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
177 BUG_ON(!have_vcpu_info_placement
);
181 static void __init
xen_banner(void)
183 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
185 printk(KERN_INFO
"Hypervisor signature: %s%s\n",
186 xen_start_info
->magic
,
187 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
190 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
191 unsigned int *cx
, unsigned int *dx
)
193 unsigned maskedx
= ~0;
196 * Mask out inconvenient features, to try and disable as many
197 * unsupported kernel subsystems as possible.
200 maskedx
= ~((1 << X86_FEATURE_APIC
) | /* disable APIC */
201 (1 << X86_FEATURE_ACPI
) | /* disable ACPI */
202 (1 << X86_FEATURE_MCE
) | /* disable MCE */
203 (1 << X86_FEATURE_MCA
) | /* disable MCA */
204 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
206 asm(XEN_EMULATE_PREFIX
"cpuid"
211 : "0" (*ax
), "2" (*cx
));
215 static void xen_set_debugreg(int reg
, unsigned long val
)
217 HYPERVISOR_set_debugreg(reg
, val
);
220 static unsigned long xen_get_debugreg(int reg
)
222 return HYPERVISOR_get_debugreg(reg
);
225 static unsigned long xen_save_fl(void)
227 struct vcpu_info
*vcpu
;
230 vcpu
= x86_read_percpu(xen_vcpu
);
232 /* flag has opposite sense of mask */
233 flags
= !vcpu
->evtchn_upcall_mask
;
235 /* convert to IF type flag
239 return (-flags
) & X86_EFLAGS_IF
;
242 static void xen_restore_fl(unsigned long flags
)
244 struct vcpu_info
*vcpu
;
246 /* convert from IF type flag */
247 flags
= !(flags
& X86_EFLAGS_IF
);
249 /* There's a one instruction preempt window here. We need to
250 make sure we're don't switch CPUs between getting the vcpu
251 pointer and updating the mask. */
253 vcpu
= x86_read_percpu(xen_vcpu
);
254 vcpu
->evtchn_upcall_mask
= flags
;
255 preempt_enable_no_resched();
257 /* Doesn't matter if we get preempted here, because any
258 pending event will get dealt with anyway. */
261 preempt_check_resched();
262 barrier(); /* unmask then check (avoid races) */
263 if (unlikely(vcpu
->evtchn_upcall_pending
))
264 force_evtchn_callback();
268 static void xen_irq_disable(void)
270 /* There's a one instruction preempt window here. We need to
271 make sure we're don't switch CPUs between getting the vcpu
272 pointer and updating the mask. */
274 x86_read_percpu(xen_vcpu
)->evtchn_upcall_mask
= 1;
275 preempt_enable_no_resched();
278 static void xen_irq_enable(void)
280 struct vcpu_info
*vcpu
;
282 /* We don't need to worry about being preempted here, since
283 either a) interrupts are disabled, so no preemption, or b)
284 the caller is confused and is trying to re-enable interrupts
285 on an indeterminate processor. */
287 vcpu
= x86_read_percpu(xen_vcpu
);
288 vcpu
->evtchn_upcall_mask
= 0;
290 /* Doesn't matter if we get preempted here, because any
291 pending event will get dealt with anyway. */
293 barrier(); /* unmask then check (avoid races) */
294 if (unlikely(vcpu
->evtchn_upcall_pending
))
295 force_evtchn_callback();
298 static void xen_safe_halt(void)
300 /* Blocking includes an implicit local_irq_enable(). */
301 if (HYPERVISOR_sched_op(SCHEDOP_block
, NULL
) != 0)
305 static void xen_halt(void)
308 HYPERVISOR_vcpu_op(VCPUOP_down
, smp_processor_id(), NULL
);
313 static void xen_leave_lazy(void)
315 paravirt_leave_lazy(paravirt_get_lazy_mode());
319 static unsigned long xen_store_tr(void)
324 static void xen_set_ldt(const void *addr
, unsigned entries
)
326 struct mmuext_op
*op
;
327 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
330 op
->cmd
= MMUEXT_SET_LDT
;
331 op
->arg1
.linear_addr
= (unsigned long)addr
;
332 op
->arg2
.nr_ents
= entries
;
334 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
336 xen_mc_issue(PARAVIRT_LAZY_CPU
);
339 static void xen_load_gdt(const struct desc_ptr
*dtr
)
341 unsigned long *frames
;
342 unsigned long va
= dtr
->address
;
343 unsigned int size
= dtr
->size
+ 1;
344 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
346 struct multicall_space mcs
;
348 /* A GDT can be up to 64k in size, which corresponds to 8192
349 8-byte entries, or 16 4k pages.. */
351 BUG_ON(size
> 65536);
352 BUG_ON(va
& ~PAGE_MASK
);
354 mcs
= xen_mc_entry(sizeof(*frames
) * pages
);
357 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
358 frames
[f
] = virt_to_mfn(va
);
359 make_lowmem_page_readonly((void *)va
);
362 MULTI_set_gdt(mcs
.mc
, frames
, size
/ sizeof(struct desc_struct
));
364 xen_mc_issue(PARAVIRT_LAZY_CPU
);
367 static void load_TLS_descriptor(struct thread_struct
*t
,
368 unsigned int cpu
, unsigned int i
)
370 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
371 xmaddr_t maddr
= virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
372 struct multicall_space mc
= __xen_mc_entry(0);
374 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
377 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
380 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
381 * it means we're in a context switch, and %gs has just been
382 * saved. This means we can zero it out to prevent faults on
383 * exit from the hypervisor if the next process has no %gs.
384 * Either way, it has been saved, and the new value will get
385 * loaded properly. This will go away as soon as Xen has been
386 * modified to not save/restore %gs for normal hypercalls.
388 * On x86_64, this hack is not used for %gs, because gs points
389 * to KERNEL_GS_BASE (and uses it for PDA references), so we
390 * must not zero %gs on x86_64
392 * For x86_64, we need to zero %fs, otherwise we may get an
393 * exception between the new %fs descriptor being loaded and
394 * %fs being effectively cleared at __switch_to().
396 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
406 load_TLS_descriptor(t
, cpu
, 0);
407 load_TLS_descriptor(t
, cpu
, 1);
408 load_TLS_descriptor(t
, cpu
, 2);
410 xen_mc_issue(PARAVIRT_LAZY_CPU
);
414 static void xen_load_gs_index(unsigned int idx
)
416 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
421 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
424 unsigned long lp
= (unsigned long)&dt
[entrynum
];
425 xmaddr_t mach_lp
= virt_to_machine(lp
);
426 u64 entry
= *(u64
*)ptr
;
431 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
437 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
438 struct trap_info
*info
)
440 if (val
->type
!= 0xf && val
->type
!= 0xe)
443 info
->vector
= vector
;
444 info
->address
= gate_offset(*val
);
445 info
->cs
= gate_segment(*val
);
446 info
->flags
= val
->dpl
;
447 /* interrupt gates clear IF */
448 if (val
->type
== 0xe)
454 /* Locations of each CPU's IDT */
455 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
457 /* Set an IDT entry. If the entry is part of the current IDT, then
459 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
461 unsigned long p
= (unsigned long)&dt
[entrynum
];
462 unsigned long start
, end
;
466 start
= __get_cpu_var(idt_desc
).address
;
467 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
471 native_write_idt_entry(dt
, entrynum
, g
);
473 if (p
>= start
&& (p
+ 8) <= end
) {
474 struct trap_info info
[2];
478 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
479 if (HYPERVISOR_set_trap_table(info
))
486 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
487 struct trap_info
*traps
)
489 unsigned in
, out
, count
;
491 count
= (desc
->size
+1) / sizeof(gate_desc
);
494 for (in
= out
= 0; in
< count
; in
++) {
495 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
497 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
500 traps
[out
].address
= 0;
503 void xen_copy_trap_info(struct trap_info
*traps
)
505 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
507 xen_convert_trap_info(desc
, traps
);
510 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
511 hold a spinlock to protect the static traps[] array (static because
512 it avoids allocation, and saves stack space). */
513 static void xen_load_idt(const struct desc_ptr
*desc
)
515 static DEFINE_SPINLOCK(lock
);
516 static struct trap_info traps
[257];
520 __get_cpu_var(idt_desc
) = *desc
;
522 xen_convert_trap_info(desc
, traps
);
525 if (HYPERVISOR_set_trap_table(traps
))
531 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
532 they're handled differently. */
533 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
534 const void *desc
, int type
)
545 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
548 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
557 static void xen_load_sp0(struct tss_struct
*tss
,
558 struct thread_struct
*thread
)
560 struct multicall_space mcs
= xen_mc_entry(0);
561 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
562 xen_mc_issue(PARAVIRT_LAZY_CPU
);
565 static void xen_set_iopl_mask(unsigned mask
)
567 struct physdev_set_iopl set_iopl
;
569 /* Force the change at ring 0. */
570 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
571 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
574 static void xen_io_delay(void)
578 #ifdef CONFIG_X86_LOCAL_APIC
579 static u32
xen_apic_read(unsigned long reg
)
584 static void xen_apic_write(unsigned long reg
, u32 val
)
586 /* Warn to see if there's any stray references */
591 static void xen_flush_tlb(void)
593 struct mmuext_op
*op
;
594 struct multicall_space mcs
;
598 mcs
= xen_mc_entry(sizeof(*op
));
601 op
->cmd
= MMUEXT_TLB_FLUSH_LOCAL
;
602 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
604 xen_mc_issue(PARAVIRT_LAZY_MMU
);
609 static void xen_flush_tlb_single(unsigned long addr
)
611 struct mmuext_op
*op
;
612 struct multicall_space mcs
;
616 mcs
= xen_mc_entry(sizeof(*op
));
618 op
->cmd
= MMUEXT_INVLPG_LOCAL
;
619 op
->arg1
.linear_addr
= addr
& PAGE_MASK
;
620 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
622 xen_mc_issue(PARAVIRT_LAZY_MMU
);
627 static void xen_flush_tlb_others(const cpumask_t
*cpus
, struct mm_struct
*mm
,
634 cpumask_t cpumask
= *cpus
;
635 struct multicall_space mcs
;
638 * A couple of (to be removed) sanity checks:
640 * - current CPU must not be in mask
641 * - mask must exist :)
643 BUG_ON(cpus_empty(cpumask
));
644 BUG_ON(cpu_isset(smp_processor_id(), cpumask
));
647 /* If a CPU which we ran on has gone down, OK. */
648 cpus_and(cpumask
, cpumask
, cpu_online_map
);
649 if (cpus_empty(cpumask
))
652 mcs
= xen_mc_entry(sizeof(*args
));
654 args
->mask
= cpumask
;
655 args
->op
.arg2
.vcpumask
= &args
->mask
;
657 if (va
== TLB_FLUSH_ALL
) {
658 args
->op
.cmd
= MMUEXT_TLB_FLUSH_MULTI
;
660 args
->op
.cmd
= MMUEXT_INVLPG_MULTI
;
661 args
->op
.arg1
.linear_addr
= va
;
664 MULTI_mmuext_op(mcs
.mc
, &args
->op
, 1, NULL
, DOMID_SELF
);
666 xen_mc_issue(PARAVIRT_LAZY_MMU
);
669 static void xen_clts(void)
671 struct multicall_space mcs
;
673 mcs
= xen_mc_entry(0);
675 MULTI_fpu_taskswitch(mcs
.mc
, 0);
677 xen_mc_issue(PARAVIRT_LAZY_CPU
);
680 static void xen_write_cr0(unsigned long cr0
)
682 struct multicall_space mcs
;
684 /* Only pay attention to cr0.TS; everything else is
686 mcs
= xen_mc_entry(0);
688 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
690 xen_mc_issue(PARAVIRT_LAZY_CPU
);
693 static void xen_write_cr2(unsigned long cr2
)
695 x86_read_percpu(xen_vcpu
)->arch
.cr2
= cr2
;
698 static unsigned long xen_read_cr2(void)
700 return x86_read_percpu(xen_vcpu
)->arch
.cr2
;
703 static unsigned long xen_read_cr2_direct(void)
705 return x86_read_percpu(xen_vcpu_info
.arch
.cr2
);
708 static void xen_write_cr4(unsigned long cr4
)
713 native_write_cr4(cr4
);
716 static unsigned long xen_read_cr3(void)
718 return x86_read_percpu(xen_cr3
);
721 static void set_current_cr3(void *v
)
723 x86_write_percpu(xen_current_cr3
, (unsigned long)v
);
726 static void __xen_write_cr3(bool kernel
, unsigned long cr3
)
728 struct mmuext_op
*op
;
729 struct multicall_space mcs
;
733 mfn
= pfn_to_mfn(PFN_DOWN(cr3
));
737 WARN_ON(mfn
== 0 && kernel
);
739 mcs
= __xen_mc_entry(sizeof(*op
));
742 op
->cmd
= kernel
? MMUEXT_NEW_BASEPTR
: MMUEXT_NEW_USER_BASEPTR
;
745 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
748 x86_write_percpu(xen_cr3
, cr3
);
750 /* Update xen_current_cr3 once the batch has actually
752 xen_mc_callback(set_current_cr3
, (void *)cr3
);
756 static void xen_write_cr3(unsigned long cr3
)
758 BUG_ON(preemptible());
760 xen_mc_batch(); /* disables interrupts */
762 /* Update while interrupts are disabled, so its atomic with
764 x86_write_percpu(xen_cr3
, cr3
);
766 __xen_write_cr3(true, cr3
);
770 pgd_t
*user_pgd
= xen_get_user_pgd(__va(cr3
));
772 __xen_write_cr3(false, __pa(user_pgd
));
774 __xen_write_cr3(false, 0);
778 xen_mc_issue(PARAVIRT_LAZY_CPU
); /* interrupts restored */
781 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
792 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
793 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
794 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
797 base
= ((u64
)high
<< 32) | low
;
798 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
803 ret
= native_write_msr_safe(msr
, low
, high
);
809 /* Early in boot, while setting up the initial pagetable, assume
810 everything is pinned. */
811 static __init
void xen_alloc_pte_init(struct mm_struct
*mm
, u32 pfn
)
813 #ifdef CONFIG_FLATMEM
814 BUG_ON(mem_map
); /* should only be used early */
816 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
819 /* Early release_pte assumes that all pts are pinned, since there's
820 only init_mm and anything attached to that is pinned. */
821 static void xen_release_pte_init(u32 pfn
)
823 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
826 static void pin_pagetable_pfn(unsigned cmd
, unsigned long pfn
)
830 op
.arg1
.mfn
= pfn_to_mfn(pfn
);
831 if (HYPERVISOR_mmuext_op(&op
, 1, NULL
, DOMID_SELF
))
835 /* This needs to make sure the new pte page is pinned iff its being
836 attached to a pinned pagetable. */
837 static void xen_alloc_ptpage(struct mm_struct
*mm
, u32 pfn
, unsigned level
)
839 struct page
*page
= pfn_to_page(pfn
);
841 if (PagePinned(virt_to_page(mm
->pgd
))) {
844 if (!PageHighMem(page
)) {
845 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
847 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE
, pfn
);
849 /* make sure there are no stray mappings of
855 static void xen_alloc_pte(struct mm_struct
*mm
, u32 pfn
)
857 xen_alloc_ptpage(mm
, pfn
, PT_PTE
);
860 static void xen_alloc_pmd(struct mm_struct
*mm
, u32 pfn
)
862 xen_alloc_ptpage(mm
, pfn
, PT_PMD
);
865 static int xen_pgd_alloc(struct mm_struct
*mm
)
867 pgd_t
*pgd
= mm
->pgd
;
870 BUG_ON(PagePinned(virt_to_page(pgd
)));
874 struct page
*page
= virt_to_page(pgd
);
877 BUG_ON(page
->private != 0);
881 user_pgd
= (pgd_t
*)__get_free_page(GFP_KERNEL
| __GFP_ZERO
);
882 page
->private = (unsigned long)user_pgd
;
884 if (user_pgd
!= NULL
) {
885 user_pgd
[pgd_index(VSYSCALL_START
)] =
886 __pgd(__pa(level3_user_vsyscall
) | _PAGE_TABLE
);
890 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd
))));
897 static void xen_pgd_free(struct mm_struct
*mm
, pgd_t
*pgd
)
900 pgd_t
*user_pgd
= xen_get_user_pgd(pgd
);
903 free_page((unsigned long)user_pgd
);
907 /* This should never happen until we're OK to use struct page */
908 static void xen_release_ptpage(u32 pfn
, unsigned level
)
910 struct page
*page
= pfn_to_page(pfn
);
912 if (PagePinned(page
)) {
913 if (!PageHighMem(page
)) {
915 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, pfn
);
916 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
918 ClearPagePinned(page
);
922 static void xen_release_pte(u32 pfn
)
924 xen_release_ptpage(pfn
, PT_PTE
);
927 static void xen_release_pmd(u32 pfn
)
929 xen_release_ptpage(pfn
, PT_PMD
);
932 #if PAGETABLE_LEVELS == 4
933 static void xen_alloc_pud(struct mm_struct
*mm
, u32 pfn
)
935 xen_alloc_ptpage(mm
, pfn
, PT_PUD
);
938 static void xen_release_pud(u32 pfn
)
940 xen_release_ptpage(pfn
, PT_PUD
);
944 #ifdef CONFIG_HIGHPTE
945 static void *xen_kmap_atomic_pte(struct page
*page
, enum km_type type
)
947 pgprot_t prot
= PAGE_KERNEL
;
949 if (PagePinned(page
))
950 prot
= PAGE_KERNEL_RO
;
952 if (0 && PageHighMem(page
))
953 printk("mapping highpte %lx type %d prot %s\n",
954 page_to_pfn(page
), type
,
955 (unsigned long)pgprot_val(prot
) & _PAGE_RW
? "WRITE" : "READ");
957 return kmap_atomic_prot(page
, type
, prot
);
961 static __init pte_t
mask_rw_pte(pte_t
*ptep
, pte_t pte
)
963 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
964 if (pte_val_ma(*ptep
) & _PAGE_PRESENT
)
965 pte
= __pte_ma(((pte_val_ma(*ptep
) & _PAGE_RW
) | ~_PAGE_RW
) &
971 /* Init-time set_pte while constructing initial pagetables, which
972 doesn't allow RO pagetable pages to be remapped RW */
973 static __init
void xen_set_pte_init(pte_t
*ptep
, pte_t pte
)
975 pte
= mask_rw_pte(ptep
, pte
);
977 xen_set_pte(ptep
, pte
);
980 static __init
void xen_pagetable_setup_start(pgd_t
*base
)
984 void xen_setup_shared_info(void)
986 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
987 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
988 xen_start_info
->shared_info
);
990 HYPERVISOR_shared_info
=
991 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
993 HYPERVISOR_shared_info
=
994 (struct shared_info
*)__va(xen_start_info
->shared_info
);
997 /* In UP this is as good a place as any to set up shared info */
998 xen_setup_vcpu_info_placement();
1001 xen_setup_mfn_list_list();
1004 static __init
void xen_pagetable_setup_done(pgd_t
*base
)
1006 xen_setup_shared_info();
1009 static __init
void xen_post_allocator_init(void)
1011 pv_mmu_ops
.set_pte
= xen_set_pte
;
1012 pv_mmu_ops
.set_pmd
= xen_set_pmd
;
1013 pv_mmu_ops
.set_pud
= xen_set_pud
;
1014 #if PAGETABLE_LEVELS == 4
1015 pv_mmu_ops
.set_pgd
= xen_set_pgd
;
1018 /* This will work as long as patching hasn't happened yet
1019 (which it hasn't) */
1020 pv_mmu_ops
.alloc_pte
= xen_alloc_pte
;
1021 pv_mmu_ops
.alloc_pmd
= xen_alloc_pmd
;
1022 pv_mmu_ops
.release_pte
= xen_release_pte
;
1023 pv_mmu_ops
.release_pmd
= xen_release_pmd
;
1024 #if PAGETABLE_LEVELS == 4
1025 pv_mmu_ops
.alloc_pud
= xen_alloc_pud
;
1026 pv_mmu_ops
.release_pud
= xen_release_pud
;
1029 #ifdef CONFIG_X86_64
1030 SetPagePinned(virt_to_page(level3_user_vsyscall
));
1032 xen_mark_init_mm_pinned();
1035 /* This is called once we have the cpu_possible_map */
1036 void xen_setup_vcpu_info_placement(void)
1040 for_each_possible_cpu(cpu
)
1041 xen_vcpu_setup(cpu
);
1043 /* xen_vcpu_setup managed to place the vcpu_info within the
1044 percpu area for all cpus, so make use of it */
1045 #ifdef CONFIG_X86_32
1046 if (have_vcpu_info_placement
) {
1047 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
1049 pv_irq_ops
.save_fl
= xen_save_fl_direct
;
1050 pv_irq_ops
.restore_fl
= xen_restore_fl_direct
;
1051 pv_irq_ops
.irq_disable
= xen_irq_disable_direct
;
1052 pv_irq_ops
.irq_enable
= xen_irq_enable_direct
;
1053 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1058 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1059 unsigned long addr
, unsigned len
)
1061 char *start
, *end
, *reloc
;
1064 start
= end
= reloc
= NULL
;
1066 #define SITE(op, x) \
1067 case PARAVIRT_PATCH(op.x): \
1068 if (have_vcpu_info_placement) { \
1069 start = (char *)xen_##x##_direct; \
1070 end = xen_##x##_direct_end; \
1071 reloc = xen_##x##_direct_reloc; \
1076 #ifdef CONFIG_X86_32
1077 SITE(pv_irq_ops
, irq_enable
);
1078 SITE(pv_irq_ops
, irq_disable
);
1079 SITE(pv_irq_ops
, save_fl
);
1080 SITE(pv_irq_ops
, restore_fl
);
1081 #endif /* CONFIG_X86_32 */
1085 if (start
== NULL
|| (end
-start
) > len
)
1088 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1090 /* Note: because reloc is assigned from something that
1091 appears to be an array, gcc assumes it's non-null,
1092 but doesn't know its relationship with start and
1094 if (reloc
> start
&& reloc
< end
) {
1095 int reloc_off
= reloc
- start
;
1096 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1097 long delta
= start
- (char *)addr
;
1105 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1113 static void xen_set_fixmap(unsigned idx
, unsigned long phys
, pgprot_t prot
)
1117 phys
>>= PAGE_SHIFT
;
1120 case FIX_BTMAP_END
... FIX_BTMAP_BEGIN
:
1121 #ifdef CONFIG_X86_F00F_BUG
1124 #ifdef CONFIG_X86_32
1127 # ifdef CONFIG_HIGHMEM
1128 case FIX_KMAP_BEGIN
... FIX_KMAP_END
:
1131 case VSYSCALL_LAST_PAGE
... VSYSCALL_FIRST_PAGE
:
1133 #ifdef CONFIG_X86_LOCAL_APIC
1134 case FIX_APIC_BASE
: /* maps dummy local APIC */
1136 pte
= pfn_pte(phys
, prot
);
1140 pte
= mfn_pte(phys
, prot
);
1144 __native_set_fixmap(idx
, pte
);
1146 #ifdef CONFIG_X86_64
1147 /* Replicate changes to map the vsyscall page into the user
1148 pagetable vsyscall mapping. */
1149 if (idx
>= VSYSCALL_LAST_PAGE
&& idx
<= VSYSCALL_FIRST_PAGE
) {
1150 unsigned long vaddr
= __fix_to_virt(idx
);
1151 set_pte_vaddr_pud(level3_user_vsyscall
, vaddr
, pte
);
1156 static const struct pv_info xen_info __initdata
= {
1157 .paravirt_enabled
= 1,
1158 .shared_kernel_pmd
= 0,
1163 static const struct pv_init_ops xen_init_ops __initdata
= {
1166 .banner
= xen_banner
,
1167 .memory_setup
= xen_memory_setup
,
1168 .arch_setup
= xen_arch_setup
,
1169 .post_allocator_init
= xen_post_allocator_init
,
1172 static const struct pv_time_ops xen_time_ops __initdata
= {
1173 .time_init
= xen_time_init
,
1175 .set_wallclock
= xen_set_wallclock
,
1176 .get_wallclock
= xen_get_wallclock
,
1177 .get_tsc_khz
= xen_tsc_khz
,
1178 .sched_clock
= xen_sched_clock
,
1181 static const struct pv_cpu_ops xen_cpu_ops __initdata
= {
1184 .set_debugreg
= xen_set_debugreg
,
1185 .get_debugreg
= xen_get_debugreg
,
1189 .read_cr0
= native_read_cr0
,
1190 .write_cr0
= xen_write_cr0
,
1192 .read_cr4
= native_read_cr4
,
1193 .read_cr4_safe
= native_read_cr4_safe
,
1194 .write_cr4
= xen_write_cr4
,
1196 .wbinvd
= native_wbinvd
,
1198 .read_msr
= native_read_msr_safe
,
1199 .write_msr
= xen_write_msr_safe
,
1200 .read_tsc
= native_read_tsc
,
1201 .read_pmc
= native_read_pmc
,
1204 .irq_enable_sysexit
= xen_sysexit
,
1205 #ifdef CONFIG_X86_64
1206 .usergs_sysret32
= xen_sysret32
,
1207 .usergs_sysret64
= xen_sysret64
,
1210 .load_tr_desc
= paravirt_nop
,
1211 .set_ldt
= xen_set_ldt
,
1212 .load_gdt
= xen_load_gdt
,
1213 .load_idt
= xen_load_idt
,
1214 .load_tls
= xen_load_tls
,
1215 #ifdef CONFIG_X86_64
1216 .load_gs_index
= xen_load_gs_index
,
1219 .store_gdt
= native_store_gdt
,
1220 .store_idt
= native_store_idt
,
1221 .store_tr
= xen_store_tr
,
1223 .write_ldt_entry
= xen_write_ldt_entry
,
1224 .write_gdt_entry
= xen_write_gdt_entry
,
1225 .write_idt_entry
= xen_write_idt_entry
,
1226 .load_sp0
= xen_load_sp0
,
1228 .set_iopl_mask
= xen_set_iopl_mask
,
1229 .io_delay
= xen_io_delay
,
1231 /* Xen takes care of %gs when switching to usermode for us */
1232 .swapgs
= paravirt_nop
,
1235 .enter
= paravirt_enter_lazy_cpu
,
1236 .leave
= xen_leave_lazy
,
1240 static void __init
__xen_init_IRQ(void)
1242 #ifdef CONFIG_X86_64
1245 /* Create identity vector->irq map */
1246 for(i
= 0; i
< NR_VECTORS
; i
++) {
1249 for_each_possible_cpu(cpu
)
1250 per_cpu(vector_irq
, cpu
)[i
] = i
;
1252 #endif /* CONFIG_X86_64 */
1257 static const struct pv_irq_ops xen_irq_ops __initdata
= {
1258 .init_IRQ
= __xen_init_IRQ
,
1259 .save_fl
= xen_save_fl
,
1260 .restore_fl
= xen_restore_fl
,
1261 .irq_disable
= xen_irq_disable
,
1262 .irq_enable
= xen_irq_enable
,
1263 .safe_halt
= xen_safe_halt
,
1265 #ifdef CONFIG_X86_64
1266 .adjust_exception_frame
= xen_adjust_exception_frame
,
1270 static const struct pv_apic_ops xen_apic_ops __initdata
= {
1271 #ifdef CONFIG_X86_LOCAL_APIC
1272 .apic_write
= xen_apic_write
,
1273 .apic_write_atomic
= xen_apic_write
,
1274 .apic_read
= xen_apic_read
,
1275 .setup_boot_clock
= paravirt_nop
,
1276 .setup_secondary_clock
= paravirt_nop
,
1277 .startup_ipi_hook
= paravirt_nop
,
1281 static const struct pv_mmu_ops xen_mmu_ops __initdata
= {
1282 .pagetable_setup_start
= xen_pagetable_setup_start
,
1283 .pagetable_setup_done
= xen_pagetable_setup_done
,
1285 .read_cr2
= xen_read_cr2
,
1286 .write_cr2
= xen_write_cr2
,
1288 .read_cr3
= xen_read_cr3
,
1289 .write_cr3
= xen_write_cr3
,
1291 .flush_tlb_user
= xen_flush_tlb
,
1292 .flush_tlb_kernel
= xen_flush_tlb
,
1293 .flush_tlb_single
= xen_flush_tlb_single
,
1294 .flush_tlb_others
= xen_flush_tlb_others
,
1296 .pte_update
= paravirt_nop
,
1297 .pte_update_defer
= paravirt_nop
,
1299 .pgd_alloc
= xen_pgd_alloc
,
1300 .pgd_free
= xen_pgd_free
,
1302 .alloc_pte
= xen_alloc_pte_init
,
1303 .release_pte
= xen_release_pte_init
,
1304 .alloc_pmd
= xen_alloc_pte_init
,
1305 .alloc_pmd_clone
= paravirt_nop
,
1306 .release_pmd
= xen_release_pte_init
,
1308 #ifdef CONFIG_HIGHPTE
1309 .kmap_atomic_pte
= xen_kmap_atomic_pte
,
1312 #ifdef CONFIG_X86_64
1313 .set_pte
= xen_set_pte
,
1315 .set_pte
= xen_set_pte_init
,
1317 .set_pte_at
= xen_set_pte_at
,
1318 .set_pmd
= xen_set_pmd_hyper
,
1320 .ptep_modify_prot_start
= __ptep_modify_prot_start
,
1321 .ptep_modify_prot_commit
= __ptep_modify_prot_commit
,
1323 .pte_val
= xen_pte_val
,
1324 .pte_flags
= native_pte_val
,
1325 .pgd_val
= xen_pgd_val
,
1327 .make_pte
= xen_make_pte
,
1328 .make_pgd
= xen_make_pgd
,
1330 #ifdef CONFIG_X86_PAE
1331 .set_pte_atomic
= xen_set_pte_atomic
,
1332 .set_pte_present
= xen_set_pte_at
,
1333 .pte_clear
= xen_pte_clear
,
1334 .pmd_clear
= xen_pmd_clear
,
1335 #endif /* CONFIG_X86_PAE */
1336 .set_pud
= xen_set_pud_hyper
,
1338 .make_pmd
= xen_make_pmd
,
1339 .pmd_val
= xen_pmd_val
,
1341 #if PAGETABLE_LEVELS == 4
1342 .pud_val
= xen_pud_val
,
1343 .make_pud
= xen_make_pud
,
1344 .set_pgd
= xen_set_pgd_hyper
,
1346 .alloc_pud
= xen_alloc_pte_init
,
1347 .release_pud
= xen_release_pte_init
,
1348 #endif /* PAGETABLE_LEVELS == 4 */
1350 .activate_mm
= xen_activate_mm
,
1351 .dup_mmap
= xen_dup_mmap
,
1352 .exit_mmap
= xen_exit_mmap
,
1355 .enter
= paravirt_enter_lazy_mmu
,
1356 .leave
= xen_leave_lazy
,
1359 .set_fixmap
= xen_set_fixmap
,
1362 static void xen_reboot(int reason
)
1364 struct sched_shutdown r
= { .reason
= reason
};
1370 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1374 static void xen_restart(char *msg
)
1376 xen_reboot(SHUTDOWN_reboot
);
1379 static void xen_emergency_restart(void)
1381 xen_reboot(SHUTDOWN_reboot
);
1384 static void xen_machine_halt(void)
1386 xen_reboot(SHUTDOWN_poweroff
);
1389 static void xen_crash_shutdown(struct pt_regs
*regs
)
1391 xen_reboot(SHUTDOWN_crash
);
1394 static const struct machine_ops __initdata xen_machine_ops
= {
1395 .restart
= xen_restart
,
1396 .halt
= xen_machine_halt
,
1397 .power_off
= xen_machine_halt
,
1398 .shutdown
= xen_machine_halt
,
1399 .crash_shutdown
= xen_crash_shutdown
,
1400 .emergency_restart
= xen_emergency_restart
,
1404 static void __init
xen_reserve_top(void)
1406 #ifdef CONFIG_X86_32
1407 unsigned long top
= HYPERVISOR_VIRT_START
;
1408 struct xen_platform_parameters pp
;
1410 if (HYPERVISOR_xen_version(XENVER_platform_parameters
, &pp
) == 0)
1411 top
= pp
.virt_start
;
1413 reserve_top_address(-top
+ 2 * PAGE_SIZE
);
1414 #endif /* CONFIG_X86_32 */
1418 * Like __va(), but returns address in the kernel mapping (which is
1419 * all we have until the physical memory mapping has been set up.
1421 static void *__ka(phys_addr_t paddr
)
1423 #ifdef CONFIG_X86_64
1424 return (void *)(paddr
+ __START_KERNEL_map
);
1430 /* Convert a machine address to physical address */
1431 static unsigned long m2p(phys_addr_t maddr
)
1436 paddr
= mfn_to_pfn(maddr
>> PAGE_SHIFT
) << PAGE_SHIFT
;
1441 /* Convert a machine address to kernel virtual */
1442 static void *m2v(phys_addr_t maddr
)
1444 return __ka(m2p(maddr
));
1447 #ifdef CONFIG_X86_64
1448 static void walk(pgd_t
*pgd
, unsigned long addr
)
1450 unsigned l4idx
= pgd_index(addr
);
1451 unsigned l3idx
= pud_index(addr
);
1452 unsigned l2idx
= pmd_index(addr
);
1453 unsigned l1idx
= pte_index(addr
);
1459 xen_raw_printk("walk %p, %lx -> %d %d %d %d\n",
1460 pgd
, addr
, l4idx
, l3idx
, l2idx
, l1idx
);
1463 xen_raw_printk(" l4: %016lx\n", l4
.pgd
);
1464 xen_raw_printk(" %016lx\n", pgd_val(l4
));
1466 l3
= ((pud_t
*)(m2v(l4
.pgd
)))[l3idx
];
1467 xen_raw_printk(" l3: %016lx\n", l3
.pud
);
1468 xen_raw_printk(" %016lx\n", pud_val(l3
));
1470 l2
= ((pmd_t
*)(m2v(l3
.pud
)))[l2idx
];
1471 xen_raw_printk(" l2: %016lx\n", l2
.pmd
);
1472 xen_raw_printk(" %016lx\n", pmd_val(l2
));
1474 l1
= ((pte_t
*)(m2v(l2
.pmd
)))[l1idx
];
1475 xen_raw_printk(" l1: %016lx\n", l1
.pte
);
1476 xen_raw_printk(" %016lx\n", pte_val(l1
));
1480 static void set_page_prot(void *addr
, pgprot_t prot
)
1482 unsigned long pfn
= __pa(addr
) >> PAGE_SHIFT
;
1483 pte_t pte
= pfn_pte(pfn
, prot
);
1485 xen_raw_printk("addr=%p pfn=%lx mfn=%lx prot=%016llx pte=%016llx\n",
1486 addr
, pfn
, get_phys_to_machine(pfn
),
1487 pgprot_val(prot
), pte
.pte
);
1489 if (HYPERVISOR_update_va_mapping((unsigned long)addr
, pte
, 0))
1493 static __init
void xen_map_identity_early(pmd_t
*pmd
, unsigned long max_pfn
)
1495 unsigned pmdidx
, pteidx
;
1501 for(pmdidx
= 0; pmdidx
< PTRS_PER_PMD
&& pfn
< max_pfn
; pmdidx
++) {
1504 /* Reuse or allocate a page of ptes */
1505 if (pmd_present(pmd
[pmdidx
]))
1506 pte_page
= m2v(pmd
[pmdidx
].pmd
);
1508 /* Check for free pte pages */
1509 if (ident_pte
== ARRAY_SIZE(level1_ident_pgt
))
1512 pte_page
= &level1_ident_pgt
[ident_pte
];
1513 ident_pte
+= PTRS_PER_PTE
;
1515 pmd
[pmdidx
] = __pmd(__pa(pte_page
) | _PAGE_TABLE
);
1518 /* Install mappings */
1519 for(pteidx
= 0; pteidx
< PTRS_PER_PTE
; pteidx
++, pfn
++) {
1522 if (pfn
> max_pfn_mapped
)
1523 max_pfn_mapped
= pfn
;
1525 if (!pte_none(pte_page
[pteidx
]))
1528 pte
= pfn_pte(pfn
, PAGE_KERNEL_EXEC
);
1529 pte_page
[pteidx
] = pte
;
1533 for(pteidx
= 0; pteidx
< ident_pte
; pteidx
+= PTRS_PER_PTE
)
1534 set_page_prot(&level1_ident_pgt
[pteidx
], PAGE_KERNEL_RO
);
1536 set_page_prot(pmd
, PAGE_KERNEL_RO
);
1539 #ifdef CONFIG_X86_64
1540 static void convert_pfn_mfn(void *v
)
1545 /* All levels are converted the same way, so just treat them
1547 for(i
= 0; i
< PTRS_PER_PTE
; i
++)
1548 pte
[i
] = xen_make_pte(pte
[i
].pte
);
1552 * Set up the inital kernel pagetable.
1554 * We can construct this by grafting the Xen provided pagetable into
1555 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1556 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1557 * means that only the kernel has a physical mapping to start with -
1558 * but that's enough to get __va working. We need to fill in the rest
1559 * of the physical mapping once some sort of allocator has been set
1562 static __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
, unsigned long max_pfn
)
1567 /* Zap identity mapping */
1568 init_level4_pgt
[0] = __pgd(0);
1570 /* Pre-constructed entries are in pfn, so convert to mfn */
1571 convert_pfn_mfn(init_level4_pgt
);
1572 convert_pfn_mfn(level3_ident_pgt
);
1573 convert_pfn_mfn(level3_kernel_pgt
);
1575 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
)].pgd
);
1576 l2
= m2v(l3
[pud_index(__START_KERNEL_map
)].pud
);
1578 memcpy(level2_ident_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1579 memcpy(level2_kernel_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1581 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
+ PMD_SIZE
)].pgd
);
1582 l2
= m2v(l3
[pud_index(__START_KERNEL_map
+ PMD_SIZE
)].pud
);
1583 memcpy(level2_fixmap_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1585 /* Set up identity map */
1586 xen_map_identity_early(level2_ident_pgt
, max_pfn
);
1588 /* Make pagetable pieces RO */
1589 set_page_prot(init_level4_pgt
, PAGE_KERNEL_RO
);
1590 set_page_prot(level3_ident_pgt
, PAGE_KERNEL_RO
);
1591 set_page_prot(level3_kernel_pgt
, PAGE_KERNEL_RO
);
1592 set_page_prot(level3_user_vsyscall
, PAGE_KERNEL_RO
);
1593 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1594 set_page_prot(level2_fixmap_pgt
, PAGE_KERNEL_RO
);
1596 /* Pin down new L4 */
1597 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE
,
1598 PFN_DOWN(__pa_symbol(init_level4_pgt
)));
1600 /* Unpin Xen-provided one */
1601 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1604 pgd
= init_level4_pgt
;
1607 * At this stage there can be no user pgd, and no page
1608 * structure to attach it to, so make sure we just set kernel
1612 __xen_write_cr3(true, __pa(pgd
));
1613 xen_mc_issue(PARAVIRT_LAZY_CPU
);
1615 reserve_early(__pa(xen_start_info
->pt_base
),
1616 __pa(xen_start_info
->pt_base
+
1617 xen_start_info
->nr_pt_frames
* PAGE_SIZE
),
1622 #else /* !CONFIG_X86_64 */
1623 static pmd_t level2_kernel_pgt
[PTRS_PER_PMD
] __page_aligned_bss
;
1625 static __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
, unsigned long max_pfn
)
1629 init_pg_tables_start
= __pa(pgd
);
1630 init_pg_tables_end
= __pa(pgd
) + xen_start_info
->nr_pt_frames
*PAGE_SIZE
;
1631 max_pfn_mapped
= PFN_DOWN(init_pg_tables_end
+ 512*1024);
1633 kernel_pmd
= m2v(pgd
[KERNEL_PGD_BOUNDARY
].pgd
);
1634 memcpy(level2_kernel_pgt
, kernel_pmd
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1636 xen_map_identity_early(level2_kernel_pgt
, max_pfn
);
1638 memcpy(swapper_pg_dir
, pgd
, sizeof(pgd_t
) * PTRS_PER_PGD
);
1639 set_pgd(&swapper_pg_dir
[KERNEL_PGD_BOUNDARY
],
1640 __pgd(__pa(level2_kernel_pgt
) | _PAGE_PRESENT
));
1642 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1643 set_page_prot(swapper_pg_dir
, PAGE_KERNEL_RO
);
1644 set_page_prot(empty_zero_page
, PAGE_KERNEL_RO
);
1646 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1648 xen_write_cr3(__pa(swapper_pg_dir
));
1650 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE
, PFN_DOWN(__pa(swapper_pg_dir
)));
1652 return swapper_pg_dir
;
1654 #endif /* CONFIG_X86_64 */
1656 /* First C function to be called on Xen boot */
1657 asmlinkage
void __init
xen_start_kernel(void)
1661 if (!xen_start_info
)
1664 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3", 5) != 0);
1666 xen_setup_features();
1668 /* Install Xen paravirt ops */
1670 pv_init_ops
= xen_init_ops
;
1671 pv_time_ops
= xen_time_ops
;
1672 pv_cpu_ops
= xen_cpu_ops
;
1673 pv_irq_ops
= xen_irq_ops
;
1674 pv_apic_ops
= xen_apic_ops
;
1675 pv_mmu_ops
= xen_mmu_ops
;
1677 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1678 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1679 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1682 machine_ops
= xen_machine_ops
;
1684 #ifdef CONFIG_X86_64
1685 /* Disable until direct per-cpu data access. */
1686 have_vcpu_info_placement
= 0;
1693 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1694 xen_build_dynamic_phys_to_machine();
1696 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1698 /* Prevent unwanted bits from being set in PTEs. */
1699 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1700 if (!is_initial_xendomain())
1701 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1703 /* Don't do the full vcpu_info placement stuff until we have a
1704 possible map and a non-dummy shared_info. */
1705 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1707 xen_raw_console_write("mapping kernel into physical memory\n");
1708 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
1712 /* keep using Xen gdt for now; no urgent need to change it */
1714 pv_info
.kernel_rpl
= 1;
1715 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1716 pv_info
.kernel_rpl
= 0;
1718 /* set the limit of our address space */
1721 #ifdef CONFIG_X86_32
1722 /* set up basic CPUID stuff */
1723 cpu_detect(&new_cpu_data
);
1724 new_cpu_data
.hard_math
= 1;
1725 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1728 /* Poke various useful things into boot_params */
1729 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1730 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1731 ? __pa(xen_start_info
->mod_start
) : 0;
1732 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1733 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1735 if (!is_initial_xendomain()) {
1736 add_preferred_console("xenboot", 0, NULL
);
1737 add_preferred_console("tty", 0, NULL
);
1738 add_preferred_console("hvc", 0, NULL
);
1741 xen_raw_console_write("about to get started...\n");
1744 xen_raw_printk("&boot_params=%p __pa(&boot_params)=%lx __va(__pa(&boot_params))=%lx\n",
1745 &boot_params
, __pa_symbol(&boot_params
),
1746 __va(__pa_symbol(&boot_params
)));
1748 walk(pgd
, &boot_params
);
1749 walk(pgd
, __va(__pa(&boot_params
)));
1752 /* Start the world */
1753 #ifdef CONFIG_X86_32
1754 i386_start_kernel();
1756 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));