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>
48 #include <asm/pgalloc.h>
52 #include "multicalls.h"
54 EXPORT_SYMBOL_GPL(hypercall_page
);
56 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
57 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
60 * Note about cr3 (pagetable base) values:
62 * xen_cr3 contains the current logical cr3 value; it contains the
63 * last set cr3. This may not be the current effective cr3, because
64 * its update may be being lazily deferred. However, a vcpu looking
65 * at its own cr3 can use this value knowing that it everything will
68 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
69 * hypercall to set the vcpu cr3 is complete (so it may be a little
70 * out of date, but it will never be set early). If one vcpu is
71 * looking at another vcpu's cr3 value, it should use this variable.
73 DEFINE_PER_CPU(unsigned long, xen_cr3
); /* cr3 stored as physaddr */
74 DEFINE_PER_CPU(unsigned long, xen_current_cr3
); /* actual vcpu cr3 */
76 struct start_info
*xen_start_info
;
77 EXPORT_SYMBOL_GPL(xen_start_info
);
79 struct shared_info xen_dummy_shared_info
;
82 * Point at some empty memory to start with. We map the real shared_info
83 * page as soon as fixmap is up and running.
85 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
88 * Flag to determine whether vcpu info placement is available on all
89 * VCPUs. We assume it is to start with, and then set it to zero on
90 * the first failure. This is because it can succeed on some VCPUs
91 * and not others, since it can involve hypervisor memory allocation,
92 * or because the guest failed to guarantee all the appropriate
93 * constraints on all VCPUs (ie buffer can't cross a page boundary).
95 * Note that any particular CPU may be using a placed vcpu structure,
96 * but we can only optimise if the all are.
98 * 0: not available, 1: available
100 static int have_vcpu_info_placement
= 1;
102 static void xen_vcpu_setup(int cpu
)
104 struct vcpu_register_vcpu_info info
;
106 struct vcpu_info
*vcpup
;
108 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
109 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
111 if (!have_vcpu_info_placement
)
112 return; /* already tested, not available */
114 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
116 info
.mfn
= virt_to_mfn(vcpup
);
117 info
.offset
= offset_in_page(vcpup
);
119 printk(KERN_DEBUG
"trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
120 cpu
, vcpup
, info
.mfn
, info
.offset
);
122 /* Check to see if the hypervisor will put the vcpu_info
123 structure where we want it, which allows direct access via
124 a percpu-variable. */
125 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
128 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
129 have_vcpu_info_placement
= 0;
131 /* This cpu is using the registered vcpu info, even if
132 later ones fail to. */
133 per_cpu(xen_vcpu
, cpu
) = vcpup
;
135 printk(KERN_DEBUG
"cpu %d using vcpu_info at %p\n",
141 * On restore, set the vcpu placement up again.
142 * If it fails, then we're in a bad state, since
143 * we can't back out from using it...
145 void xen_vcpu_restore(void)
147 if (have_vcpu_info_placement
) {
150 for_each_online_cpu(cpu
) {
151 bool other_cpu
= (cpu
!= smp_processor_id());
154 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
160 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
164 BUG_ON(!have_vcpu_info_placement
);
168 static void __init
xen_banner(void)
170 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
171 struct xen_extraversion extra
;
172 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
174 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
176 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
177 version
>> 16, version
& 0xffff, extra
.extraversion
,
178 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
181 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
182 unsigned int *cx
, unsigned int *dx
)
184 unsigned maskedx
= ~0;
187 * Mask out inconvenient features, to try and disable as many
188 * unsupported kernel subsystems as possible.
191 maskedx
= ~((1 << X86_FEATURE_APIC
) | /* disable APIC */
192 (1 << X86_FEATURE_ACPI
) | /* disable ACPI */
193 (1 << X86_FEATURE_MCE
) | /* disable MCE */
194 (1 << X86_FEATURE_MCA
) | /* disable MCA */
195 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
197 asm(XEN_EMULATE_PREFIX
"cpuid"
202 : "0" (*ax
), "2" (*cx
));
206 static void xen_set_debugreg(int reg
, unsigned long val
)
208 HYPERVISOR_set_debugreg(reg
, val
);
211 static unsigned long xen_get_debugreg(int reg
)
213 return HYPERVISOR_get_debugreg(reg
);
216 static unsigned long xen_save_fl(void)
218 struct vcpu_info
*vcpu
;
221 vcpu
= x86_read_percpu(xen_vcpu
);
223 /* flag has opposite sense of mask */
224 flags
= !vcpu
->evtchn_upcall_mask
;
226 /* convert to IF type flag
230 return (-flags
) & X86_EFLAGS_IF
;
233 static void xen_restore_fl(unsigned long flags
)
235 struct vcpu_info
*vcpu
;
237 /* convert from IF type flag */
238 flags
= !(flags
& X86_EFLAGS_IF
);
240 /* There's a one instruction preempt window here. We need to
241 make sure we're don't switch CPUs between getting the vcpu
242 pointer and updating the mask. */
244 vcpu
= x86_read_percpu(xen_vcpu
);
245 vcpu
->evtchn_upcall_mask
= flags
;
246 preempt_enable_no_resched();
248 /* Doesn't matter if we get preempted here, because any
249 pending event will get dealt with anyway. */
252 preempt_check_resched();
253 barrier(); /* unmask then check (avoid races) */
254 if (unlikely(vcpu
->evtchn_upcall_pending
))
255 force_evtchn_callback();
259 static void xen_irq_disable(void)
261 /* There's a one instruction preempt window here. We need to
262 make sure we're don't switch CPUs between getting the vcpu
263 pointer and updating the mask. */
265 x86_read_percpu(xen_vcpu
)->evtchn_upcall_mask
= 1;
266 preempt_enable_no_resched();
269 static void xen_irq_enable(void)
271 struct vcpu_info
*vcpu
;
273 /* We don't need to worry about being preempted here, since
274 either a) interrupts are disabled, so no preemption, or b)
275 the caller is confused and is trying to re-enable interrupts
276 on an indeterminate processor. */
278 vcpu
= x86_read_percpu(xen_vcpu
);
279 vcpu
->evtchn_upcall_mask
= 0;
281 /* Doesn't matter if we get preempted here, because any
282 pending event will get dealt with anyway. */
284 barrier(); /* unmask then check (avoid races) */
285 if (unlikely(vcpu
->evtchn_upcall_pending
))
286 force_evtchn_callback();
289 static void xen_safe_halt(void)
291 /* Blocking includes an implicit local_irq_enable(). */
292 if (HYPERVISOR_sched_op(SCHEDOP_block
, NULL
) != 0)
296 static void xen_halt(void)
299 HYPERVISOR_vcpu_op(VCPUOP_down
, smp_processor_id(), NULL
);
304 static void xen_leave_lazy(void)
306 paravirt_leave_lazy(paravirt_get_lazy_mode());
310 static unsigned long xen_store_tr(void)
315 static void xen_set_ldt(const void *addr
, unsigned entries
)
317 struct mmuext_op
*op
;
318 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
321 op
->cmd
= MMUEXT_SET_LDT
;
322 op
->arg1
.linear_addr
= (unsigned long)addr
;
323 op
->arg2
.nr_ents
= entries
;
325 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
327 xen_mc_issue(PARAVIRT_LAZY_CPU
);
330 static void xen_load_gdt(const struct desc_ptr
*dtr
)
332 unsigned long *frames
;
333 unsigned long va
= dtr
->address
;
334 unsigned int size
= dtr
->size
+ 1;
335 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
337 struct multicall_space mcs
;
339 /* A GDT can be up to 64k in size, which corresponds to 8192
340 8-byte entries, or 16 4k pages.. */
342 BUG_ON(size
> 65536);
343 BUG_ON(va
& ~PAGE_MASK
);
345 mcs
= xen_mc_entry(sizeof(*frames
) * pages
);
348 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
349 frames
[f
] = virt_to_mfn(va
);
350 make_lowmem_page_readonly((void *)va
);
353 MULTI_set_gdt(mcs
.mc
, frames
, size
/ sizeof(struct desc_struct
));
355 xen_mc_issue(PARAVIRT_LAZY_CPU
);
358 static void load_TLS_descriptor(struct thread_struct
*t
,
359 unsigned int cpu
, unsigned int i
)
361 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
362 xmaddr_t maddr
= virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
363 struct multicall_space mc
= __xen_mc_entry(0);
365 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
368 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
372 load_TLS_descriptor(t
, cpu
, 0);
373 load_TLS_descriptor(t
, cpu
, 1);
374 load_TLS_descriptor(t
, cpu
, 2);
376 xen_mc_issue(PARAVIRT_LAZY_CPU
);
379 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
380 * it means we're in a context switch, and %gs has just been
381 * saved. This means we can zero it out to prevent faults on
382 * exit from the hypervisor if the next process has no %gs.
383 * Either way, it has been saved, and the new value will get
384 * loaded properly. This will go away as soon as Xen has been
385 * modified to not save/restore %gs for normal hypercalls.
387 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
)
391 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
394 unsigned long lp
= (unsigned long)&dt
[entrynum
];
395 xmaddr_t mach_lp
= virt_to_machine(lp
);
396 u64 entry
= *(u64
*)ptr
;
401 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
407 static int cvt_gate_to_trap(int vector
, u32 low
, u32 high
,
408 struct trap_info
*info
)
412 type
= (high
>> 8) & 0x1f;
413 dpl
= (high
>> 13) & 3;
415 if (type
!= 0xf && type
!= 0xe)
418 info
->vector
= vector
;
419 info
->address
= (high
& 0xffff0000) | (low
& 0x0000ffff);
420 info
->cs
= low
>> 16;
422 /* interrupt gates clear IF */
429 /* Locations of each CPU's IDT */
430 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
432 /* Set an IDT entry. If the entry is part of the current IDT, then
434 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
436 unsigned long p
= (unsigned long)&dt
[entrynum
];
437 unsigned long start
, end
;
441 start
= __get_cpu_var(idt_desc
).address
;
442 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
446 native_write_idt_entry(dt
, entrynum
, g
);
448 if (p
>= start
&& (p
+ 8) <= end
) {
449 struct trap_info info
[2];
450 u32
*desc
= (u32
*)g
;
454 if (cvt_gate_to_trap(entrynum
, desc
[0], desc
[1], &info
[0]))
455 if (HYPERVISOR_set_trap_table(info
))
462 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
463 struct trap_info
*traps
)
465 unsigned in
, out
, count
;
467 count
= (desc
->size
+1) / 8;
470 for (in
= out
= 0; in
< count
; in
++) {
471 const u32
*entry
= (u32
*)(desc
->address
+ in
* 8);
473 if (cvt_gate_to_trap(in
, entry
[0], entry
[1], &traps
[out
]))
476 traps
[out
].address
= 0;
479 void xen_copy_trap_info(struct trap_info
*traps
)
481 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
483 xen_convert_trap_info(desc
, traps
);
486 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
487 hold a spinlock to protect the static traps[] array (static because
488 it avoids allocation, and saves stack space). */
489 static void xen_load_idt(const struct desc_ptr
*desc
)
491 static DEFINE_SPINLOCK(lock
);
492 static struct trap_info traps
[257];
496 __get_cpu_var(idt_desc
) = *desc
;
498 xen_convert_trap_info(desc
, traps
);
501 if (HYPERVISOR_set_trap_table(traps
))
507 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
508 they're handled differently. */
509 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
510 const void *desc
, int type
)
521 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
524 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
533 static void xen_load_sp0(struct tss_struct
*tss
,
534 struct thread_struct
*thread
)
536 struct multicall_space mcs
= xen_mc_entry(0);
537 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
538 xen_mc_issue(PARAVIRT_LAZY_CPU
);
541 static void xen_set_iopl_mask(unsigned mask
)
543 struct physdev_set_iopl set_iopl
;
545 /* Force the change at ring 0. */
546 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
547 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
550 static void xen_io_delay(void)
554 #ifdef CONFIG_X86_LOCAL_APIC
555 static u32
xen_apic_read(unsigned long reg
)
560 static void xen_apic_write(unsigned long reg
, u32 val
)
562 /* Warn to see if there's any stray references */
567 static void xen_flush_tlb(void)
569 struct mmuext_op
*op
;
570 struct multicall_space mcs
;
574 mcs
= xen_mc_entry(sizeof(*op
));
577 op
->cmd
= MMUEXT_TLB_FLUSH_LOCAL
;
578 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
580 xen_mc_issue(PARAVIRT_LAZY_MMU
);
585 static void xen_flush_tlb_single(unsigned long addr
)
587 struct mmuext_op
*op
;
588 struct multicall_space mcs
;
592 mcs
= xen_mc_entry(sizeof(*op
));
594 op
->cmd
= MMUEXT_INVLPG_LOCAL
;
595 op
->arg1
.linear_addr
= addr
& PAGE_MASK
;
596 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
598 xen_mc_issue(PARAVIRT_LAZY_MMU
);
603 static void xen_flush_tlb_others(const cpumask_t
*cpus
, struct mm_struct
*mm
,
610 cpumask_t cpumask
= *cpus
;
611 struct multicall_space mcs
;
614 * A couple of (to be removed) sanity checks:
616 * - current CPU must not be in mask
617 * - mask must exist :)
619 BUG_ON(cpus_empty(cpumask
));
620 BUG_ON(cpu_isset(smp_processor_id(), cpumask
));
623 /* If a CPU which we ran on has gone down, OK. */
624 cpus_and(cpumask
, cpumask
, cpu_online_map
);
625 if (cpus_empty(cpumask
))
628 mcs
= xen_mc_entry(sizeof(*args
));
630 args
->mask
= cpumask
;
631 args
->op
.arg2
.vcpumask
= &args
->mask
;
633 if (va
== TLB_FLUSH_ALL
) {
634 args
->op
.cmd
= MMUEXT_TLB_FLUSH_MULTI
;
636 args
->op
.cmd
= MMUEXT_INVLPG_MULTI
;
637 args
->op
.arg1
.linear_addr
= va
;
640 MULTI_mmuext_op(mcs
.mc
, &args
->op
, 1, NULL
, DOMID_SELF
);
642 xen_mc_issue(PARAVIRT_LAZY_MMU
);
645 static void xen_clts(void)
647 struct multicall_space mcs
;
649 mcs
= xen_mc_entry(0);
651 MULTI_fpu_taskswitch(mcs
.mc
, 0);
653 xen_mc_issue(PARAVIRT_LAZY_CPU
);
656 static void xen_write_cr0(unsigned long cr0
)
658 struct multicall_space mcs
;
660 /* Only pay attention to cr0.TS; everything else is
662 mcs
= xen_mc_entry(0);
664 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
666 xen_mc_issue(PARAVIRT_LAZY_CPU
);
669 static void xen_write_cr2(unsigned long cr2
)
671 x86_read_percpu(xen_vcpu
)->arch
.cr2
= cr2
;
674 static unsigned long xen_read_cr2(void)
676 return x86_read_percpu(xen_vcpu
)->arch
.cr2
;
679 static unsigned long xen_read_cr2_direct(void)
681 return x86_read_percpu(xen_vcpu_info
.arch
.cr2
);
684 static void xen_write_cr4(unsigned long cr4
)
689 native_write_cr4(cr4
);
692 static unsigned long xen_read_cr3(void)
694 return x86_read_percpu(xen_cr3
);
697 static void set_current_cr3(void *v
)
699 x86_write_percpu(xen_current_cr3
, (unsigned long)v
);
702 static void xen_write_cr3(unsigned long cr3
)
704 struct mmuext_op
*op
;
705 struct multicall_space mcs
;
706 unsigned long mfn
= pfn_to_mfn(PFN_DOWN(cr3
));
708 BUG_ON(preemptible());
710 mcs
= xen_mc_entry(sizeof(*op
)); /* disables interrupts */
712 /* Update while interrupts are disabled, so its atomic with
714 x86_write_percpu(xen_cr3
, cr3
);
717 op
->cmd
= MMUEXT_NEW_BASEPTR
;
720 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
722 /* Update xen_update_cr3 once the batch has actually
724 xen_mc_callback(set_current_cr3
, (void *)cr3
);
726 xen_mc_issue(PARAVIRT_LAZY_CPU
); /* interrupts restored */
729 /* Early in boot, while setting up the initial pagetable, assume
730 everything is pinned. */
731 static __init
void xen_alloc_pte_init(struct mm_struct
*mm
, u32 pfn
)
733 #ifdef CONFIG_FLATMEM
734 BUG_ON(mem_map
); /* should only be used early */
736 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
739 /* Early release_pte assumes that all pts are pinned, since there's
740 only init_mm and anything attached to that is pinned. */
741 static void xen_release_pte_init(u32 pfn
)
743 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
746 static void pin_pagetable_pfn(unsigned cmd
, unsigned long pfn
)
750 op
.arg1
.mfn
= pfn_to_mfn(pfn
);
751 if (HYPERVISOR_mmuext_op(&op
, 1, NULL
, DOMID_SELF
))
755 /* This needs to make sure the new pte page is pinned iff its being
756 attached to a pinned pagetable. */
757 static void xen_alloc_ptpage(struct mm_struct
*mm
, u32 pfn
, unsigned level
)
759 struct page
*page
= pfn_to_page(pfn
);
761 if (PagePinned(virt_to_page(mm
->pgd
))) {
764 if (!PageHighMem(page
)) {
765 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
767 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE
, pfn
);
769 /* make sure there are no stray mappings of
775 static void xen_alloc_pte(struct mm_struct
*mm
, u32 pfn
)
777 xen_alloc_ptpage(mm
, pfn
, PT_PTE
);
780 static void xen_alloc_pmd(struct mm_struct
*mm
, u32 pfn
)
782 xen_alloc_ptpage(mm
, pfn
, PT_PMD
);
785 /* This should never happen until we're OK to use struct page */
786 static void xen_release_ptpage(u32 pfn
, unsigned level
)
788 struct page
*page
= pfn_to_page(pfn
);
790 if (PagePinned(page
)) {
791 if (!PageHighMem(page
)) {
793 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, pfn
);
794 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
796 ClearPagePinned(page
);
800 static void xen_release_pte(u32 pfn
)
802 xen_release_ptpage(pfn
, PT_PTE
);
805 static void xen_release_pmd(u32 pfn
)
807 xen_release_ptpage(pfn
, PT_PMD
);
810 #ifdef CONFIG_HIGHPTE
811 static void *xen_kmap_atomic_pte(struct page
*page
, enum km_type type
)
813 pgprot_t prot
= PAGE_KERNEL
;
815 if (PagePinned(page
))
816 prot
= PAGE_KERNEL_RO
;
818 if (0 && PageHighMem(page
))
819 printk("mapping highpte %lx type %d prot %s\n",
820 page_to_pfn(page
), type
,
821 (unsigned long)pgprot_val(prot
) & _PAGE_RW
? "WRITE" : "READ");
823 return kmap_atomic_prot(page
, type
, prot
);
827 static __init pte_t
mask_rw_pte(pte_t
*ptep
, pte_t pte
)
829 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
830 if (pte_val_ma(*ptep
) & _PAGE_PRESENT
)
831 pte
= __pte_ma(((pte_val_ma(*ptep
) & _PAGE_RW
) | ~_PAGE_RW
) &
837 /* Init-time set_pte while constructing initial pagetables, which
838 doesn't allow RO pagetable pages to be remapped RW */
839 static __init
void xen_set_pte_init(pte_t
*ptep
, pte_t pte
)
841 pte
= mask_rw_pte(ptep
, pte
);
843 xen_set_pte(ptep
, pte
);
846 static __init
void xen_pagetable_setup_start(pgd_t
*base
)
848 pgd_t
*xen_pgd
= (pgd_t
*)xen_start_info
->pt_base
;
851 /* special set_pte for pagetable initialization */
852 pv_mmu_ops
.set_pte
= xen_set_pte_init
;
856 * copy top-level of Xen-supplied pagetable into place. This
857 * is a stand-in while we copy the pmd pages.
859 memcpy(base
, xen_pgd
, PTRS_PER_PGD
* sizeof(pgd_t
));
862 * For PAE, need to allocate new pmds, rather than
863 * share Xen's, since Xen doesn't like pmd's being
864 * shared between address spaces.
866 for (i
= 0; i
< PTRS_PER_PGD
; i
++) {
867 if (pgd_val_ma(xen_pgd
[i
]) & _PAGE_PRESENT
) {
868 pmd_t
*pmd
= (pmd_t
*)alloc_bootmem_low_pages(PAGE_SIZE
);
870 memcpy(pmd
, (void *)pgd_page_vaddr(xen_pgd
[i
]),
873 make_lowmem_page_readonly(pmd
);
875 set_pgd(&base
[i
], __pgd(1 + __pa(pmd
)));
880 /* make sure zero_page is mapped RO so we can use it in pagetables */
881 make_lowmem_page_readonly(empty_zero_page
);
882 make_lowmem_page_readonly(base
);
884 * Switch to new pagetable. This is done before
885 * pagetable_init has done anything so that the new pages
886 * added to the table can be prepared properly for Xen.
888 xen_write_cr3(__pa(base
));
890 /* Unpin initial Xen pagetable */
891 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
,
892 PFN_DOWN(__pa(xen_start_info
->pt_base
)));
895 void xen_setup_shared_info(void)
897 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
898 unsigned long addr
= fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
901 * Create a mapping for the shared info page.
902 * Should be set_fixmap(), but shared_info is a machine
903 * address with no corresponding pseudo-phys address.
906 PFN_DOWN(xen_start_info
->shared_info
),
909 HYPERVISOR_shared_info
= (struct shared_info
*)addr
;
911 HYPERVISOR_shared_info
=
912 (struct shared_info
*)__va(xen_start_info
->shared_info
);
915 /* In UP this is as good a place as any to set up shared info */
916 xen_setup_vcpu_info_placement();
919 xen_setup_mfn_list_list();
922 static __init
void xen_pagetable_setup_done(pgd_t
*base
)
924 /* This will work as long as patching hasn't happened yet
926 pv_mmu_ops
.alloc_pte
= xen_alloc_pte
;
927 pv_mmu_ops
.alloc_pmd
= xen_alloc_pmd
;
928 pv_mmu_ops
.release_pte
= xen_release_pte
;
929 pv_mmu_ops
.release_pmd
= xen_release_pmd
;
930 pv_mmu_ops
.set_pte
= xen_set_pte
;
932 xen_setup_shared_info();
934 /* Actually pin the pagetable down, but we can't set PG_pinned
935 yet because the page structures don't exist yet. */
936 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE
, PFN_DOWN(__pa(base
)));
939 static __init
void xen_post_allocator_init(void)
941 pv_mmu_ops
.set_pmd
= xen_set_pmd
;
942 pv_mmu_ops
.set_pud
= xen_set_pud
;
944 xen_mark_init_mm_pinned();
947 /* This is called once we have the cpu_possible_map */
948 void xen_setup_vcpu_info_placement(void)
952 for_each_possible_cpu(cpu
)
955 /* xen_vcpu_setup managed to place the vcpu_info within the
956 percpu area for all cpus, so make use of it */
957 if (have_vcpu_info_placement
) {
958 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
960 pv_irq_ops
.save_fl
= xen_save_fl_direct
;
961 pv_irq_ops
.restore_fl
= xen_restore_fl_direct
;
962 pv_irq_ops
.irq_disable
= xen_irq_disable_direct
;
963 pv_irq_ops
.irq_enable
= xen_irq_enable_direct
;
964 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
968 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
969 unsigned long addr
, unsigned len
)
971 char *start
, *end
, *reloc
;
974 start
= end
= reloc
= NULL
;
976 #define SITE(op, x) \
977 case PARAVIRT_PATCH(op.x): \
978 if (have_vcpu_info_placement) { \
979 start = (char *)xen_##x##_direct; \
980 end = xen_##x##_direct_end; \
981 reloc = xen_##x##_direct_reloc; \
986 SITE(pv_irq_ops
, irq_enable
);
987 SITE(pv_irq_ops
, irq_disable
);
988 SITE(pv_irq_ops
, save_fl
);
989 SITE(pv_irq_ops
, restore_fl
);
993 if (start
== NULL
|| (end
-start
) > len
)
996 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
998 /* Note: because reloc is assigned from something that
999 appears to be an array, gcc assumes it's non-null,
1000 but doesn't know its relationship with start and
1002 if (reloc
> start
&& reloc
< end
) {
1003 int reloc_off
= reloc
- start
;
1004 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1005 long delta
= start
- (char *)addr
;
1013 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1021 static void xen_set_fixmap(unsigned idx
, unsigned long phys
, pgprot_t prot
)
1025 phys
>>= PAGE_SHIFT
;
1028 case FIX_BTMAP_END
... FIX_BTMAP_BEGIN
:
1029 #ifdef CONFIG_X86_F00F_BUG
1034 #ifdef CONFIG_X86_LOCAL_APIC
1035 case FIX_APIC_BASE
: /* maps dummy local APIC */
1037 pte
= pfn_pte(phys
, prot
);
1041 pte
= mfn_pte(phys
, prot
);
1045 __native_set_fixmap(idx
, pte
);
1048 static const struct pv_info xen_info __initdata
= {
1049 .paravirt_enabled
= 1,
1050 .shared_kernel_pmd
= 0,
1055 static const struct pv_init_ops xen_init_ops __initdata
= {
1058 .banner
= xen_banner
,
1059 .memory_setup
= xen_memory_setup
,
1060 .arch_setup
= xen_arch_setup
,
1061 .post_allocator_init
= xen_post_allocator_init
,
1064 static const struct pv_time_ops xen_time_ops __initdata
= {
1065 .time_init
= xen_time_init
,
1067 .set_wallclock
= xen_set_wallclock
,
1068 .get_wallclock
= xen_get_wallclock
,
1069 .get_tsc_khz
= xen_tsc_khz
,
1070 .sched_clock
= xen_sched_clock
,
1073 static const struct pv_cpu_ops xen_cpu_ops __initdata
= {
1076 .set_debugreg
= xen_set_debugreg
,
1077 .get_debugreg
= xen_get_debugreg
,
1081 .read_cr0
= native_read_cr0
,
1082 .write_cr0
= xen_write_cr0
,
1084 .read_cr4
= native_read_cr4
,
1085 .read_cr4_safe
= native_read_cr4_safe
,
1086 .write_cr4
= xen_write_cr4
,
1088 .wbinvd
= native_wbinvd
,
1090 .read_msr
= native_read_msr_safe
,
1091 .write_msr
= native_write_msr_safe
,
1092 .read_tsc
= native_read_tsc
,
1093 .read_pmc
= native_read_pmc
,
1096 .irq_enable_sysexit
= xen_sysexit
,
1098 .load_tr_desc
= paravirt_nop
,
1099 .set_ldt
= xen_set_ldt
,
1100 .load_gdt
= xen_load_gdt
,
1101 .load_idt
= xen_load_idt
,
1102 .load_tls
= xen_load_tls
,
1104 .store_gdt
= native_store_gdt
,
1105 .store_idt
= native_store_idt
,
1106 .store_tr
= xen_store_tr
,
1108 .write_ldt_entry
= xen_write_ldt_entry
,
1109 .write_gdt_entry
= xen_write_gdt_entry
,
1110 .write_idt_entry
= xen_write_idt_entry
,
1111 .load_sp0
= xen_load_sp0
,
1113 .set_iopl_mask
= xen_set_iopl_mask
,
1114 .io_delay
= xen_io_delay
,
1117 .enter
= paravirt_enter_lazy_cpu
,
1118 .leave
= xen_leave_lazy
,
1122 static const struct pv_irq_ops xen_irq_ops __initdata
= {
1123 .init_IRQ
= xen_init_IRQ
,
1124 .save_fl
= xen_save_fl
,
1125 .restore_fl
= xen_restore_fl
,
1126 .irq_disable
= xen_irq_disable
,
1127 .irq_enable
= xen_irq_enable
,
1128 .safe_halt
= xen_safe_halt
,
1130 #ifdef CONFIG_X86_64
1131 .adjust_exception_frame
= paravirt_nop
,
1135 static const struct pv_apic_ops xen_apic_ops __initdata
= {
1136 #ifdef CONFIG_X86_LOCAL_APIC
1137 .apic_write
= xen_apic_write
,
1138 .apic_write_atomic
= xen_apic_write
,
1139 .apic_read
= xen_apic_read
,
1140 .setup_boot_clock
= paravirt_nop
,
1141 .setup_secondary_clock
= paravirt_nop
,
1142 .startup_ipi_hook
= paravirt_nop
,
1146 static const struct pv_mmu_ops xen_mmu_ops __initdata
= {
1147 .pagetable_setup_start
= xen_pagetable_setup_start
,
1148 .pagetable_setup_done
= xen_pagetable_setup_done
,
1150 .read_cr2
= xen_read_cr2
,
1151 .write_cr2
= xen_write_cr2
,
1153 .read_cr3
= xen_read_cr3
,
1154 .write_cr3
= xen_write_cr3
,
1156 .flush_tlb_user
= xen_flush_tlb
,
1157 .flush_tlb_kernel
= xen_flush_tlb
,
1158 .flush_tlb_single
= xen_flush_tlb_single
,
1159 .flush_tlb_others
= xen_flush_tlb_others
,
1161 .pte_update
= paravirt_nop
,
1162 .pte_update_defer
= paravirt_nop
,
1164 .pgd_alloc
= __paravirt_pgd_alloc
,
1165 .pgd_free
= paravirt_nop
,
1167 .alloc_pte
= xen_alloc_pte_init
,
1168 .release_pte
= xen_release_pte_init
,
1169 .alloc_pmd
= xen_alloc_pte_init
,
1170 .alloc_pmd_clone
= paravirt_nop
,
1171 .release_pmd
= xen_release_pte_init
,
1173 #ifdef CONFIG_HIGHPTE
1174 .kmap_atomic_pte
= xen_kmap_atomic_pte
,
1177 .set_pte
= NULL
, /* see xen_pagetable_setup_* */
1178 .set_pte_at
= xen_set_pte_at
,
1179 .set_pmd
= xen_set_pmd_hyper
,
1181 .ptep_modify_prot_start
= __ptep_modify_prot_start
,
1182 .ptep_modify_prot_commit
= __ptep_modify_prot_commit
,
1184 .pte_val
= xen_pte_val
,
1185 .pte_flags
= native_pte_val
,
1186 .pgd_val
= xen_pgd_val
,
1188 .make_pte
= xen_make_pte
,
1189 .make_pgd
= xen_make_pgd
,
1191 .set_pte_atomic
= xen_set_pte_atomic
,
1192 .set_pte_present
= xen_set_pte_at
,
1193 .set_pud
= xen_set_pud_hyper
,
1194 .pte_clear
= xen_pte_clear
,
1195 .pmd_clear
= xen_pmd_clear
,
1197 .make_pmd
= xen_make_pmd
,
1198 .pmd_val
= xen_pmd_val
,
1200 .activate_mm
= xen_activate_mm
,
1201 .dup_mmap
= xen_dup_mmap
,
1202 .exit_mmap
= xen_exit_mmap
,
1205 .enter
= paravirt_enter_lazy_mmu
,
1206 .leave
= xen_leave_lazy
,
1209 .set_fixmap
= xen_set_fixmap
,
1213 static const struct smp_ops xen_smp_ops __initdata
= {
1214 .smp_prepare_boot_cpu
= xen_smp_prepare_boot_cpu
,
1215 .smp_prepare_cpus
= xen_smp_prepare_cpus
,
1216 .cpu_up
= xen_cpu_up
,
1217 .smp_cpus_done
= xen_smp_cpus_done
,
1219 .smp_send_stop
= xen_smp_send_stop
,
1220 .smp_send_reschedule
= xen_smp_send_reschedule
,
1222 .send_call_func_ipi
= xen_smp_send_call_function_ipi
,
1223 .send_call_func_single_ipi
= xen_smp_send_call_function_single_ipi
,
1225 #endif /* CONFIG_SMP */
1227 static void xen_reboot(int reason
)
1229 struct sched_shutdown r
= { .reason
= reason
};
1235 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1239 static void xen_restart(char *msg
)
1241 xen_reboot(SHUTDOWN_reboot
);
1244 static void xen_emergency_restart(void)
1246 xen_reboot(SHUTDOWN_reboot
);
1249 static void xen_machine_halt(void)
1251 xen_reboot(SHUTDOWN_poweroff
);
1254 static void xen_crash_shutdown(struct pt_regs
*regs
)
1256 xen_reboot(SHUTDOWN_crash
);
1259 static const struct machine_ops __initdata xen_machine_ops
= {
1260 .restart
= xen_restart
,
1261 .halt
= xen_machine_halt
,
1262 .power_off
= xen_machine_halt
,
1263 .shutdown
= xen_machine_halt
,
1264 .crash_shutdown
= xen_crash_shutdown
,
1265 .emergency_restart
= xen_emergency_restart
,
1269 static void __init
xen_reserve_top(void)
1271 unsigned long top
= HYPERVISOR_VIRT_START
;
1272 struct xen_platform_parameters pp
;
1274 if (HYPERVISOR_xen_version(XENVER_platform_parameters
, &pp
) == 0)
1275 top
= pp
.virt_start
;
1277 reserve_top_address(-top
+ 2 * PAGE_SIZE
);
1280 /* First C function to be called on Xen boot */
1281 asmlinkage
void __init
xen_start_kernel(void)
1285 if (!xen_start_info
)
1288 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3", 5) != 0);
1290 xen_setup_features();
1292 /* Install Xen paravirt ops */
1294 pv_init_ops
= xen_init_ops
;
1295 pv_time_ops
= xen_time_ops
;
1296 pv_cpu_ops
= xen_cpu_ops
;
1297 pv_irq_ops
= xen_irq_ops
;
1298 pv_apic_ops
= xen_apic_ops
;
1299 pv_mmu_ops
= xen_mmu_ops
;
1301 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1302 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1303 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1306 machine_ops
= xen_machine_ops
;
1309 smp_ops
= xen_smp_ops
;
1313 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1314 xen_build_dynamic_phys_to_machine();
1316 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1318 init_pg_tables_start
= __pa(pgd
);
1319 init_pg_tables_end
= __pa(pgd
) + xen_start_info
->nr_pt_frames
*PAGE_SIZE
;
1320 max_pfn_mapped
= (init_pg_tables_end
+ 512*1024) >> PAGE_SHIFT
;
1322 init_mm
.pgd
= pgd
; /* use the Xen pagetables to start */
1324 /* keep using Xen gdt for now; no urgent need to change it */
1326 x86_write_percpu(xen_cr3
, __pa(pgd
));
1327 x86_write_percpu(xen_current_cr3
, __pa(pgd
));
1329 /* Don't do the full vcpu_info placement stuff until we have a
1330 possible map and a non-dummy shared_info. */
1331 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1333 pv_info
.kernel_rpl
= 1;
1334 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1335 pv_info
.kernel_rpl
= 0;
1337 /* Prevent unwanted bits from being set in PTEs. */
1338 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1339 if (!is_initial_xendomain())
1340 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1342 /* set the limit of our address space */
1345 /* set up basic CPUID stuff */
1346 cpu_detect(&new_cpu_data
);
1347 new_cpu_data
.hard_math
= 1;
1348 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1350 /* Poke various useful things into boot_params */
1351 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1352 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1353 ? __pa(xen_start_info
->mod_start
) : 0;
1354 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1356 if (!is_initial_xendomain()) {
1357 add_preferred_console("xenboot", 0, NULL
);
1358 add_preferred_console("tty", 0, NULL
);
1359 add_preferred_console("hvc", 0, NULL
);
1362 /* Start the world */
1363 i386_start_kernel();