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>
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/setup.h>
45 #include <asm/pgtable.h>
46 #include <asm/tlbflush.h>
47 #include <asm/reboot.h>
51 #include "multicalls.h"
53 EXPORT_SYMBOL_GPL(hypercall_page
);
55 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
56 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
59 * Note about cr3 (pagetable base) values:
61 * xen_cr3 contains the current logical cr3 value; it contains the
62 * last set cr3. This may not be the current effective cr3, because
63 * its update may be being lazily deferred. However, a vcpu looking
64 * at its own cr3 can use this value knowing that it everything will
67 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
68 * hypercall to set the vcpu cr3 is complete (so it may be a little
69 * out of date, but it will never be set early). If one vcpu is
70 * looking at another vcpu's cr3 value, it should use this variable.
72 DEFINE_PER_CPU(unsigned long, xen_cr3
); /* cr3 stored as physaddr */
73 DEFINE_PER_CPU(unsigned long, xen_current_cr3
); /* actual vcpu cr3 */
75 struct start_info
*xen_start_info
;
76 EXPORT_SYMBOL_GPL(xen_start_info
);
78 struct shared_info xen_dummy_shared_info
;
81 * Point at some empty memory to start with. We map the real shared_info
82 * page as soon as fixmap is up and running.
84 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
87 * Flag to determine whether vcpu info placement is available on all
88 * VCPUs. We assume it is to start with, and then set it to zero on
89 * the first failure. This is because it can succeed on some VCPUs
90 * and not others, since it can involve hypervisor memory allocation,
91 * or because the guest failed to guarantee all the appropriate
92 * constraints on all VCPUs (ie buffer can't cross a page boundary).
94 * Note that any particular CPU may be using a placed vcpu structure,
95 * but we can only optimise if the all are.
97 * 0: not available, 1: available
99 static int have_vcpu_info_placement
= 1;
101 static void xen_vcpu_setup(int cpu
)
103 struct vcpu_register_vcpu_info info
;
105 struct vcpu_info
*vcpup
;
107 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
108 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
110 if (!have_vcpu_info_placement
)
111 return; /* already tested, not available */
113 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
115 info
.mfn
= virt_to_mfn(vcpup
);
116 info
.offset
= offset_in_page(vcpup
);
118 printk(KERN_DEBUG
"trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
119 cpu
, vcpup
, info
.mfn
, info
.offset
);
121 /* Check to see if the hypervisor will put the vcpu_info
122 structure where we want it, which allows direct access via
123 a percpu-variable. */
124 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
127 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
128 have_vcpu_info_placement
= 0;
130 /* This cpu is using the registered vcpu info, even if
131 later ones fail to. */
132 per_cpu(xen_vcpu
, cpu
) = vcpup
;
134 printk(KERN_DEBUG
"cpu %d using vcpu_info at %p\n",
140 * On restore, set the vcpu placement up again.
141 * If it fails, then we're in a bad state, since
142 * we can't back out from using it...
144 void xen_vcpu_restore(void)
146 if (have_vcpu_info_placement
) {
149 for_each_online_cpu(cpu
) {
150 bool other_cpu
= (cpu
!= smp_processor_id());
153 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
159 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
163 BUG_ON(!have_vcpu_info_placement
);
167 static void __init
xen_banner(void)
169 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
171 printk(KERN_INFO
"Hypervisor signature: %s%s\n",
172 xen_start_info
->magic
,
173 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
176 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
177 unsigned int *cx
, unsigned int *dx
)
179 unsigned maskedx
= ~0;
182 * Mask out inconvenient features, to try and disable as many
183 * unsupported kernel subsystems as possible.
186 maskedx
= ~((1 << X86_FEATURE_APIC
) | /* disable APIC */
187 (1 << X86_FEATURE_ACPI
) | /* disable ACPI */
188 (1 << X86_FEATURE_MCE
) | /* disable MCE */
189 (1 << X86_FEATURE_MCA
) | /* disable MCA */
190 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
192 asm(XEN_EMULATE_PREFIX
"cpuid"
197 : "0" (*ax
), "2" (*cx
));
201 static void xen_set_debugreg(int reg
, unsigned long val
)
203 HYPERVISOR_set_debugreg(reg
, val
);
206 static unsigned long xen_get_debugreg(int reg
)
208 return HYPERVISOR_get_debugreg(reg
);
211 static unsigned long xen_save_fl(void)
213 struct vcpu_info
*vcpu
;
216 vcpu
= x86_read_percpu(xen_vcpu
);
218 /* flag has opposite sense of mask */
219 flags
= !vcpu
->evtchn_upcall_mask
;
221 /* convert to IF type flag
225 return (-flags
) & X86_EFLAGS_IF
;
228 static void xen_restore_fl(unsigned long flags
)
230 struct vcpu_info
*vcpu
;
232 /* convert from IF type flag */
233 flags
= !(flags
& X86_EFLAGS_IF
);
235 /* There's a one instruction preempt window here. We need to
236 make sure we're don't switch CPUs between getting the vcpu
237 pointer and updating the mask. */
239 vcpu
= x86_read_percpu(xen_vcpu
);
240 vcpu
->evtchn_upcall_mask
= flags
;
241 preempt_enable_no_resched();
243 /* Doesn't matter if we get preempted here, because any
244 pending event will get dealt with anyway. */
247 preempt_check_resched();
248 barrier(); /* unmask then check (avoid races) */
249 if (unlikely(vcpu
->evtchn_upcall_pending
))
250 force_evtchn_callback();
254 static void xen_irq_disable(void)
256 /* There's a one instruction preempt window here. We need to
257 make sure we're don't switch CPUs between getting the vcpu
258 pointer and updating the mask. */
260 x86_read_percpu(xen_vcpu
)->evtchn_upcall_mask
= 1;
261 preempt_enable_no_resched();
264 static void xen_irq_enable(void)
266 struct vcpu_info
*vcpu
;
268 /* We don't need to worry about being preempted here, since
269 either a) interrupts are disabled, so no preemption, or b)
270 the caller is confused and is trying to re-enable interrupts
271 on an indeterminate processor. */
273 vcpu
= x86_read_percpu(xen_vcpu
);
274 vcpu
->evtchn_upcall_mask
= 0;
276 /* Doesn't matter if we get preempted here, because any
277 pending event will get dealt with anyway. */
279 barrier(); /* unmask then check (avoid races) */
280 if (unlikely(vcpu
->evtchn_upcall_pending
))
281 force_evtchn_callback();
284 static void xen_safe_halt(void)
286 /* Blocking includes an implicit local_irq_enable(). */
287 if (HYPERVISOR_sched_op(SCHEDOP_block
, NULL
) != 0)
291 static void xen_halt(void)
294 HYPERVISOR_vcpu_op(VCPUOP_down
, smp_processor_id(), NULL
);
299 static void xen_leave_lazy(void)
301 paravirt_leave_lazy(paravirt_get_lazy_mode());
305 static unsigned long xen_store_tr(void)
310 static void xen_set_ldt(const void *addr
, unsigned entries
)
312 struct mmuext_op
*op
;
313 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
316 op
->cmd
= MMUEXT_SET_LDT
;
317 op
->arg1
.linear_addr
= (unsigned long)addr
;
318 op
->arg2
.nr_ents
= entries
;
320 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
322 xen_mc_issue(PARAVIRT_LAZY_CPU
);
325 static void xen_load_gdt(const struct desc_ptr
*dtr
)
327 unsigned long *frames
;
328 unsigned long va
= dtr
->address
;
329 unsigned int size
= dtr
->size
+ 1;
330 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
332 struct multicall_space mcs
;
334 /* A GDT can be up to 64k in size, which corresponds to 8192
335 8-byte entries, or 16 4k pages.. */
337 BUG_ON(size
> 65536);
338 BUG_ON(va
& ~PAGE_MASK
);
340 mcs
= xen_mc_entry(sizeof(*frames
) * pages
);
343 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
344 frames
[f
] = virt_to_mfn(va
);
345 make_lowmem_page_readonly((void *)va
);
348 MULTI_set_gdt(mcs
.mc
, frames
, size
/ sizeof(struct desc_struct
));
350 xen_mc_issue(PARAVIRT_LAZY_CPU
);
353 static void load_TLS_descriptor(struct thread_struct
*t
,
354 unsigned int cpu
, unsigned int i
)
356 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
357 xmaddr_t maddr
= virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
358 struct multicall_space mc
= __xen_mc_entry(0);
360 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
363 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
367 load_TLS_descriptor(t
, cpu
, 0);
368 load_TLS_descriptor(t
, cpu
, 1);
369 load_TLS_descriptor(t
, cpu
, 2);
371 xen_mc_issue(PARAVIRT_LAZY_CPU
);
374 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
375 * it means we're in a context switch, and %gs has just been
376 * saved. This means we can zero it out to prevent faults on
377 * exit from the hypervisor if the next process has no %gs.
378 * Either way, it has been saved, and the new value will get
379 * loaded properly. This will go away as soon as Xen has been
380 * modified to not save/restore %gs for normal hypercalls.
382 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
)
386 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
389 unsigned long lp
= (unsigned long)&dt
[entrynum
];
390 xmaddr_t mach_lp
= virt_to_machine(lp
);
391 u64 entry
= *(u64
*)ptr
;
396 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
402 static int cvt_gate_to_trap(int vector
, u32 low
, u32 high
,
403 struct trap_info
*info
)
407 type
= (high
>> 8) & 0x1f;
408 dpl
= (high
>> 13) & 3;
410 if (type
!= 0xf && type
!= 0xe)
413 info
->vector
= vector
;
414 info
->address
= (high
& 0xffff0000) | (low
& 0x0000ffff);
415 info
->cs
= low
>> 16;
417 /* interrupt gates clear IF */
424 /* Locations of each CPU's IDT */
425 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
427 /* Set an IDT entry. If the entry is part of the current IDT, then
429 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
431 unsigned long p
= (unsigned long)&dt
[entrynum
];
432 unsigned long start
, end
;
436 start
= __get_cpu_var(idt_desc
).address
;
437 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
441 native_write_idt_entry(dt
, entrynum
, g
);
443 if (p
>= start
&& (p
+ 8) <= end
) {
444 struct trap_info info
[2];
445 u32
*desc
= (u32
*)g
;
449 if (cvt_gate_to_trap(entrynum
, desc
[0], desc
[1], &info
[0]))
450 if (HYPERVISOR_set_trap_table(info
))
457 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
458 struct trap_info
*traps
)
460 unsigned in
, out
, count
;
462 count
= (desc
->size
+1) / 8;
465 for (in
= out
= 0; in
< count
; in
++) {
466 const u32
*entry
= (u32
*)(desc
->address
+ in
* 8);
468 if (cvt_gate_to_trap(in
, entry
[0], entry
[1], &traps
[out
]))
471 traps
[out
].address
= 0;
474 void xen_copy_trap_info(struct trap_info
*traps
)
476 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
478 xen_convert_trap_info(desc
, traps
);
481 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
482 hold a spinlock to protect the static traps[] array (static because
483 it avoids allocation, and saves stack space). */
484 static void xen_load_idt(const struct desc_ptr
*desc
)
486 static DEFINE_SPINLOCK(lock
);
487 static struct trap_info traps
[257];
491 __get_cpu_var(idt_desc
) = *desc
;
493 xen_convert_trap_info(desc
, traps
);
496 if (HYPERVISOR_set_trap_table(traps
))
502 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
503 they're handled differently. */
504 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
505 const void *desc
, int type
)
516 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
519 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
528 static void xen_load_sp0(struct tss_struct
*tss
,
529 struct thread_struct
*thread
)
531 struct multicall_space mcs
= xen_mc_entry(0);
532 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
533 xen_mc_issue(PARAVIRT_LAZY_CPU
);
536 static void xen_set_iopl_mask(unsigned mask
)
538 struct physdev_set_iopl set_iopl
;
540 /* Force the change at ring 0. */
541 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
542 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
545 static void xen_io_delay(void)
549 #ifdef CONFIG_X86_LOCAL_APIC
550 static u32
xen_apic_read(unsigned long reg
)
555 static void xen_apic_write(unsigned long reg
, u32 val
)
557 /* Warn to see if there's any stray references */
562 static void xen_flush_tlb(void)
564 struct mmuext_op
*op
;
565 struct multicall_space mcs
;
569 mcs
= xen_mc_entry(sizeof(*op
));
572 op
->cmd
= MMUEXT_TLB_FLUSH_LOCAL
;
573 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
575 xen_mc_issue(PARAVIRT_LAZY_MMU
);
580 static void xen_flush_tlb_single(unsigned long addr
)
582 struct mmuext_op
*op
;
583 struct multicall_space mcs
;
587 mcs
= xen_mc_entry(sizeof(*op
));
589 op
->cmd
= MMUEXT_INVLPG_LOCAL
;
590 op
->arg1
.linear_addr
= addr
& PAGE_MASK
;
591 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
593 xen_mc_issue(PARAVIRT_LAZY_MMU
);
598 static void xen_flush_tlb_others(const cpumask_t
*cpus
, struct mm_struct
*mm
,
605 cpumask_t cpumask
= *cpus
;
606 struct multicall_space mcs
;
609 * A couple of (to be removed) sanity checks:
611 * - current CPU must not be in mask
612 * - mask must exist :)
614 BUG_ON(cpus_empty(cpumask
));
615 BUG_ON(cpu_isset(smp_processor_id(), cpumask
));
618 /* If a CPU which we ran on has gone down, OK. */
619 cpus_and(cpumask
, cpumask
, cpu_online_map
);
620 if (cpus_empty(cpumask
))
623 mcs
= xen_mc_entry(sizeof(*args
));
625 args
->mask
= cpumask
;
626 args
->op
.arg2
.vcpumask
= &args
->mask
;
628 if (va
== TLB_FLUSH_ALL
) {
629 args
->op
.cmd
= MMUEXT_TLB_FLUSH_MULTI
;
631 args
->op
.cmd
= MMUEXT_INVLPG_MULTI
;
632 args
->op
.arg1
.linear_addr
= va
;
635 MULTI_mmuext_op(mcs
.mc
, &args
->op
, 1, NULL
, DOMID_SELF
);
637 xen_mc_issue(PARAVIRT_LAZY_MMU
);
640 static void xen_clts(void)
642 struct multicall_space mcs
;
644 mcs
= xen_mc_entry(0);
646 MULTI_fpu_taskswitch(mcs
.mc
, 0);
648 xen_mc_issue(PARAVIRT_LAZY_CPU
);
651 static void xen_write_cr0(unsigned long cr0
)
653 struct multicall_space mcs
;
655 /* Only pay attention to cr0.TS; everything else is
657 mcs
= xen_mc_entry(0);
659 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
661 xen_mc_issue(PARAVIRT_LAZY_CPU
);
664 static void xen_write_cr2(unsigned long cr2
)
666 x86_read_percpu(xen_vcpu
)->arch
.cr2
= cr2
;
669 static unsigned long xen_read_cr2(void)
671 return x86_read_percpu(xen_vcpu
)->arch
.cr2
;
674 static unsigned long xen_read_cr2_direct(void)
676 return x86_read_percpu(xen_vcpu_info
.arch
.cr2
);
679 static void xen_write_cr4(unsigned long cr4
)
684 native_write_cr4(cr4
);
687 static unsigned long xen_read_cr3(void)
689 return x86_read_percpu(xen_cr3
);
692 static void set_current_cr3(void *v
)
694 x86_write_percpu(xen_current_cr3
, (unsigned long)v
);
697 static void xen_write_cr3(unsigned long cr3
)
699 struct mmuext_op
*op
;
700 struct multicall_space mcs
;
701 unsigned long mfn
= pfn_to_mfn(PFN_DOWN(cr3
));
703 BUG_ON(preemptible());
705 mcs
= xen_mc_entry(sizeof(*op
)); /* disables interrupts */
707 /* Update while interrupts are disabled, so its atomic with
709 x86_write_percpu(xen_cr3
, cr3
);
712 op
->cmd
= MMUEXT_NEW_BASEPTR
;
715 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
717 /* Update xen_update_cr3 once the batch has actually
719 xen_mc_callback(set_current_cr3
, (void *)cr3
);
721 xen_mc_issue(PARAVIRT_LAZY_CPU
); /* interrupts restored */
724 /* Early in boot, while setting up the initial pagetable, assume
725 everything is pinned. */
726 static __init
void xen_alloc_pte_init(struct mm_struct
*mm
, u32 pfn
)
728 #ifdef CONFIG_FLATMEM
729 BUG_ON(mem_map
); /* should only be used early */
731 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
734 /* Early release_pte assumes that all pts are pinned, since there's
735 only init_mm and anything attached to that is pinned. */
736 static void xen_release_pte_init(u32 pfn
)
738 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
741 static void pin_pagetable_pfn(unsigned cmd
, unsigned long pfn
)
745 op
.arg1
.mfn
= pfn_to_mfn(pfn
);
746 if (HYPERVISOR_mmuext_op(&op
, 1, NULL
, DOMID_SELF
))
750 /* This needs to make sure the new pte page is pinned iff its being
751 attached to a pinned pagetable. */
752 static void xen_alloc_ptpage(struct mm_struct
*mm
, u32 pfn
, unsigned level
)
754 struct page
*page
= pfn_to_page(pfn
);
756 if (PagePinned(virt_to_page(mm
->pgd
))) {
759 if (!PageHighMem(page
)) {
760 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
762 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE
, pfn
);
764 /* make sure there are no stray mappings of
770 static void xen_alloc_pte(struct mm_struct
*mm
, u32 pfn
)
772 xen_alloc_ptpage(mm
, pfn
, PT_PTE
);
775 static void xen_alloc_pmd(struct mm_struct
*mm
, u32 pfn
)
777 xen_alloc_ptpage(mm
, pfn
, PT_PMD
);
780 /* This should never happen until we're OK to use struct page */
781 static void xen_release_ptpage(u32 pfn
, unsigned level
)
783 struct page
*page
= pfn_to_page(pfn
);
785 if (PagePinned(page
)) {
786 if (!PageHighMem(page
)) {
788 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, pfn
);
789 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
791 ClearPagePinned(page
);
795 static void xen_release_pte(u32 pfn
)
797 xen_release_ptpage(pfn
, PT_PTE
);
800 static void xen_release_pmd(u32 pfn
)
802 xen_release_ptpage(pfn
, PT_PMD
);
805 #ifdef CONFIG_HIGHPTE
806 static void *xen_kmap_atomic_pte(struct page
*page
, enum km_type type
)
808 pgprot_t prot
= PAGE_KERNEL
;
810 if (PagePinned(page
))
811 prot
= PAGE_KERNEL_RO
;
813 if (0 && PageHighMem(page
))
814 printk("mapping highpte %lx type %d prot %s\n",
815 page_to_pfn(page
), type
,
816 (unsigned long)pgprot_val(prot
) & _PAGE_RW
? "WRITE" : "READ");
818 return kmap_atomic_prot(page
, type
, prot
);
822 static __init pte_t
mask_rw_pte(pte_t
*ptep
, pte_t pte
)
824 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
825 if (pte_val_ma(*ptep
) & _PAGE_PRESENT
)
826 pte
= __pte_ma(((pte_val_ma(*ptep
) & _PAGE_RW
) | ~_PAGE_RW
) &
832 /* Init-time set_pte while constructing initial pagetables, which
833 doesn't allow RO pagetable pages to be remapped RW */
834 static __init
void xen_set_pte_init(pte_t
*ptep
, pte_t pte
)
836 pte
= mask_rw_pte(ptep
, pte
);
838 xen_set_pte(ptep
, pte
);
841 static __init
void xen_pagetable_setup_start(pgd_t
*base
)
843 pgd_t
*xen_pgd
= (pgd_t
*)xen_start_info
->pt_base
;
846 /* special set_pte for pagetable initialization */
847 pv_mmu_ops
.set_pte
= xen_set_pte_init
;
851 * copy top-level of Xen-supplied pagetable into place. This
852 * is a stand-in while we copy the pmd pages.
854 memcpy(base
, xen_pgd
, PTRS_PER_PGD
* sizeof(pgd_t
));
857 * For PAE, need to allocate new pmds, rather than
858 * share Xen's, since Xen doesn't like pmd's being
859 * shared between address spaces.
861 for (i
= 0; i
< PTRS_PER_PGD
; i
++) {
862 if (pgd_val_ma(xen_pgd
[i
]) & _PAGE_PRESENT
) {
863 pmd_t
*pmd
= (pmd_t
*)alloc_bootmem_low_pages(PAGE_SIZE
);
865 memcpy(pmd
, (void *)pgd_page_vaddr(xen_pgd
[i
]),
868 make_lowmem_page_readonly(pmd
);
870 set_pgd(&base
[i
], __pgd(1 + __pa(pmd
)));
875 /* make sure zero_page is mapped RO so we can use it in pagetables */
876 make_lowmem_page_readonly(empty_zero_page
);
877 make_lowmem_page_readonly(base
);
879 * Switch to new pagetable. This is done before
880 * pagetable_init has done anything so that the new pages
881 * added to the table can be prepared properly for Xen.
883 xen_write_cr3(__pa(base
));
885 /* Unpin initial Xen pagetable */
886 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
,
887 PFN_DOWN(__pa(xen_start_info
->pt_base
)));
890 void xen_setup_shared_info(void)
892 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
893 unsigned long addr
= fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
896 * Create a mapping for the shared info page.
897 * Should be set_fixmap(), but shared_info is a machine
898 * address with no corresponding pseudo-phys address.
901 PFN_DOWN(xen_start_info
->shared_info
),
904 HYPERVISOR_shared_info
= (struct shared_info
*)addr
;
906 HYPERVISOR_shared_info
=
907 (struct shared_info
*)__va(xen_start_info
->shared_info
);
910 /* In UP this is as good a place as any to set up shared info */
911 xen_setup_vcpu_info_placement();
914 xen_setup_mfn_list_list();
917 static __init
void xen_pagetable_setup_done(pgd_t
*base
)
919 /* This will work as long as patching hasn't happened yet
921 pv_mmu_ops
.alloc_pte
= xen_alloc_pte
;
922 pv_mmu_ops
.alloc_pmd
= xen_alloc_pmd
;
923 pv_mmu_ops
.release_pte
= xen_release_pte
;
924 pv_mmu_ops
.release_pmd
= xen_release_pmd
;
925 pv_mmu_ops
.set_pte
= xen_set_pte
;
927 xen_setup_shared_info();
929 /* Actually pin the pagetable down, but we can't set PG_pinned
930 yet because the page structures don't exist yet. */
931 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE
, PFN_DOWN(__pa(base
)));
934 static __init
void xen_post_allocator_init(void)
936 pv_mmu_ops
.set_pmd
= xen_set_pmd
;
937 pv_mmu_ops
.set_pud
= xen_set_pud
;
939 xen_mark_init_mm_pinned();
942 /* This is called once we have the cpu_possible_map */
943 void xen_setup_vcpu_info_placement(void)
947 for_each_possible_cpu(cpu
)
950 /* xen_vcpu_setup managed to place the vcpu_info within the
951 percpu area for all cpus, so make use of it */
952 if (have_vcpu_info_placement
) {
953 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
955 pv_irq_ops
.save_fl
= xen_save_fl_direct
;
956 pv_irq_ops
.restore_fl
= xen_restore_fl_direct
;
957 pv_irq_ops
.irq_disable
= xen_irq_disable_direct
;
958 pv_irq_ops
.irq_enable
= xen_irq_enable_direct
;
959 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
963 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
964 unsigned long addr
, unsigned len
)
966 char *start
, *end
, *reloc
;
969 start
= end
= reloc
= NULL
;
971 #define SITE(op, x) \
972 case PARAVIRT_PATCH(op.x): \
973 if (have_vcpu_info_placement) { \
974 start = (char *)xen_##x##_direct; \
975 end = xen_##x##_direct_end; \
976 reloc = xen_##x##_direct_reloc; \
981 SITE(pv_irq_ops
, irq_enable
);
982 SITE(pv_irq_ops
, irq_disable
);
983 SITE(pv_irq_ops
, save_fl
);
984 SITE(pv_irq_ops
, restore_fl
);
988 if (start
== NULL
|| (end
-start
) > len
)
991 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
993 /* Note: because reloc is assigned from something that
994 appears to be an array, gcc assumes it's non-null,
995 but doesn't know its relationship with start and
997 if (reloc
> start
&& reloc
< end
) {
998 int reloc_off
= reloc
- start
;
999 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1000 long delta
= start
- (char *)addr
;
1008 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1016 static const struct pv_info xen_info __initdata
= {
1017 .paravirt_enabled
= 1,
1018 .shared_kernel_pmd
= 0,
1023 static const struct pv_init_ops xen_init_ops __initdata
= {
1026 .banner
= xen_banner
,
1027 .memory_setup
= xen_memory_setup
,
1028 .arch_setup
= xen_arch_setup
,
1029 .post_allocator_init
= xen_post_allocator_init
,
1032 static const struct pv_time_ops xen_time_ops __initdata
= {
1033 .time_init
= xen_time_init
,
1035 .set_wallclock
= xen_set_wallclock
,
1036 .get_wallclock
= xen_get_wallclock
,
1037 .get_cpu_khz
= xen_cpu_khz
,
1038 .sched_clock
= xen_sched_clock
,
1041 static const struct pv_cpu_ops xen_cpu_ops __initdata
= {
1044 .set_debugreg
= xen_set_debugreg
,
1045 .get_debugreg
= xen_get_debugreg
,
1049 .read_cr0
= native_read_cr0
,
1050 .write_cr0
= xen_write_cr0
,
1052 .read_cr4
= native_read_cr4
,
1053 .read_cr4_safe
= native_read_cr4_safe
,
1054 .write_cr4
= xen_write_cr4
,
1056 .wbinvd
= native_wbinvd
,
1058 .read_msr
= native_read_msr_safe
,
1059 .write_msr
= native_write_msr_safe
,
1060 .read_tsc
= native_read_tsc
,
1061 .read_pmc
= native_read_pmc
,
1064 .irq_enable_syscall_ret
= xen_sysexit
,
1066 .load_tr_desc
= paravirt_nop
,
1067 .set_ldt
= xen_set_ldt
,
1068 .load_gdt
= xen_load_gdt
,
1069 .load_idt
= xen_load_idt
,
1070 .load_tls
= xen_load_tls
,
1072 .store_gdt
= native_store_gdt
,
1073 .store_idt
= native_store_idt
,
1074 .store_tr
= xen_store_tr
,
1076 .write_ldt_entry
= xen_write_ldt_entry
,
1077 .write_gdt_entry
= xen_write_gdt_entry
,
1078 .write_idt_entry
= xen_write_idt_entry
,
1079 .load_sp0
= xen_load_sp0
,
1081 .set_iopl_mask
= xen_set_iopl_mask
,
1082 .io_delay
= xen_io_delay
,
1085 .enter
= paravirt_enter_lazy_cpu
,
1086 .leave
= xen_leave_lazy
,
1090 static const struct pv_irq_ops xen_irq_ops __initdata
= {
1091 .init_IRQ
= xen_init_IRQ
,
1092 .save_fl
= xen_save_fl
,
1093 .restore_fl
= xen_restore_fl
,
1094 .irq_disable
= xen_irq_disable
,
1095 .irq_enable
= xen_irq_enable
,
1096 .safe_halt
= xen_safe_halt
,
1100 static const struct pv_apic_ops xen_apic_ops __initdata
= {
1101 #ifdef CONFIG_X86_LOCAL_APIC
1102 .apic_write
= xen_apic_write
,
1103 .apic_write_atomic
= xen_apic_write
,
1104 .apic_read
= xen_apic_read
,
1105 .setup_boot_clock
= paravirt_nop
,
1106 .setup_secondary_clock
= paravirt_nop
,
1107 .startup_ipi_hook
= paravirt_nop
,
1111 static const struct pv_mmu_ops xen_mmu_ops __initdata
= {
1112 .pagetable_setup_start
= xen_pagetable_setup_start
,
1113 .pagetable_setup_done
= xen_pagetable_setup_done
,
1115 .read_cr2
= xen_read_cr2
,
1116 .write_cr2
= xen_write_cr2
,
1118 .read_cr3
= xen_read_cr3
,
1119 .write_cr3
= xen_write_cr3
,
1121 .flush_tlb_user
= xen_flush_tlb
,
1122 .flush_tlb_kernel
= xen_flush_tlb
,
1123 .flush_tlb_single
= xen_flush_tlb_single
,
1124 .flush_tlb_others
= xen_flush_tlb_others
,
1126 .pte_update
= paravirt_nop
,
1127 .pte_update_defer
= paravirt_nop
,
1129 .alloc_pte
= xen_alloc_pte_init
,
1130 .release_pte
= xen_release_pte_init
,
1131 .alloc_pmd
= xen_alloc_pte_init
,
1132 .alloc_pmd_clone
= paravirt_nop
,
1133 .release_pmd
= xen_release_pte_init
,
1135 #ifdef CONFIG_HIGHPTE
1136 .kmap_atomic_pte
= xen_kmap_atomic_pte
,
1139 .set_pte
= NULL
, /* see xen_pagetable_setup_* */
1140 .set_pte_at
= xen_set_pte_at
,
1141 .set_pmd
= xen_set_pmd_hyper
,
1143 .ptep_modify_prot_start
= __ptep_modify_prot_start
,
1144 .ptep_modify_prot_commit
= __ptep_modify_prot_commit
,
1146 .pte_val
= xen_pte_val
,
1147 .pte_flags
= native_pte_val
,
1148 .pgd_val
= xen_pgd_val
,
1150 .make_pte
= xen_make_pte
,
1151 .make_pgd
= xen_make_pgd
,
1153 .set_pte_atomic
= xen_set_pte_atomic
,
1154 .set_pte_present
= xen_set_pte_at
,
1155 .set_pud
= xen_set_pud_hyper
,
1156 .pte_clear
= xen_pte_clear
,
1157 .pmd_clear
= xen_pmd_clear
,
1159 .make_pmd
= xen_make_pmd
,
1160 .pmd_val
= xen_pmd_val
,
1162 .activate_mm
= xen_activate_mm
,
1163 .dup_mmap
= xen_dup_mmap
,
1164 .exit_mmap
= xen_exit_mmap
,
1167 .enter
= paravirt_enter_lazy_mmu
,
1168 .leave
= xen_leave_lazy
,
1173 static const struct smp_ops xen_smp_ops __initdata
= {
1174 .smp_prepare_boot_cpu
= xen_smp_prepare_boot_cpu
,
1175 .smp_prepare_cpus
= xen_smp_prepare_cpus
,
1176 .cpu_up
= xen_cpu_up
,
1177 .smp_cpus_done
= xen_smp_cpus_done
,
1179 .smp_send_stop
= xen_smp_send_stop
,
1180 .smp_send_reschedule
= xen_smp_send_reschedule
,
1181 .smp_call_function_mask
= xen_smp_call_function_mask
,
1183 #endif /* CONFIG_SMP */
1185 static void xen_reboot(int reason
)
1187 struct sched_shutdown r
= { .reason
= reason
};
1193 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1197 static void xen_restart(char *msg
)
1199 xen_reboot(SHUTDOWN_reboot
);
1202 static void xen_emergency_restart(void)
1204 xen_reboot(SHUTDOWN_reboot
);
1207 static void xen_machine_halt(void)
1209 xen_reboot(SHUTDOWN_poweroff
);
1212 static void xen_crash_shutdown(struct pt_regs
*regs
)
1214 xen_reboot(SHUTDOWN_crash
);
1217 static const struct machine_ops __initdata xen_machine_ops
= {
1218 .restart
= xen_restart
,
1219 .halt
= xen_machine_halt
,
1220 .power_off
= xen_machine_halt
,
1221 .shutdown
= xen_machine_halt
,
1222 .crash_shutdown
= xen_crash_shutdown
,
1223 .emergency_restart
= xen_emergency_restart
,
1227 static void __init
xen_reserve_top(void)
1229 unsigned long top
= HYPERVISOR_VIRT_START
;
1230 struct xen_platform_parameters pp
;
1232 if (HYPERVISOR_xen_version(XENVER_platform_parameters
, &pp
) == 0)
1233 top
= pp
.virt_start
;
1235 reserve_top_address(-top
+ 2 * PAGE_SIZE
);
1238 /* First C function to be called on Xen boot */
1239 asmlinkage
void __init
xen_start_kernel(void)
1243 if (!xen_start_info
)
1246 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3", 5) != 0);
1248 xen_setup_features();
1250 /* Install Xen paravirt ops */
1252 pv_init_ops
= xen_init_ops
;
1253 pv_time_ops
= xen_time_ops
;
1254 pv_cpu_ops
= xen_cpu_ops
;
1255 pv_irq_ops
= xen_irq_ops
;
1256 pv_apic_ops
= xen_apic_ops
;
1257 pv_mmu_ops
= xen_mmu_ops
;
1259 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1260 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1261 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1264 machine_ops
= xen_machine_ops
;
1267 smp_ops
= xen_smp_ops
;
1271 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1272 xen_build_dynamic_phys_to_machine();
1274 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1276 init_pg_tables_start
= __pa(pgd
);
1277 init_pg_tables_end
= __pa(pgd
) + xen_start_info
->nr_pt_frames
*PAGE_SIZE
;
1279 init_mm
.pgd
= pgd
; /* use the Xen pagetables to start */
1281 /* keep using Xen gdt for now; no urgent need to change it */
1283 x86_write_percpu(xen_cr3
, __pa(pgd
));
1284 x86_write_percpu(xen_current_cr3
, __pa(pgd
));
1286 /* Don't do the full vcpu_info placement stuff until we have a
1287 possible map and a non-dummy shared_info. */
1288 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1290 pv_info
.kernel_rpl
= 1;
1291 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1292 pv_info
.kernel_rpl
= 0;
1294 /* Prevent unwanted bits from being set in PTEs. */
1295 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1296 if (!is_initial_xendomain())
1297 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1299 /* set the limit of our address space */
1302 /* set up basic CPUID stuff */
1303 cpu_detect(&new_cpu_data
);
1304 new_cpu_data
.hard_math
= 1;
1305 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1307 /* Poke various useful things into boot_params */
1308 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1309 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1310 ? __pa(xen_start_info
->mod_start
) : 0;
1311 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1313 if (!is_initial_xendomain()) {
1314 add_preferred_console("xenboot", 0, NULL
);
1315 add_preferred_console("tty", 0, NULL
);
1316 add_preferred_console("hvc", 0, NULL
);
1319 /* Start the world */
1320 i386_start_kernel();