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>
29 #include <xen/interface/xen.h>
30 #include <xen/interface/physdev.h>
31 #include <xen/interface/vcpu.h>
32 #include <xen/interface/sched.h>
33 #include <xen/features.h>
36 #include <asm/paravirt.h>
38 #include <asm/xen/hypercall.h>
39 #include <asm/xen/hypervisor.h>
40 #include <asm/fixmap.h>
41 #include <asm/processor.h>
42 #include <asm/setup.h>
44 #include <asm/pgtable.h>
45 #include <asm/tlbflush.h>
46 #include <asm/reboot.h>
50 #include "multicalls.h"
52 EXPORT_SYMBOL_GPL(hypercall_page
);
54 DEFINE_PER_CPU(enum paravirt_lazy_mode
, xen_lazy_mode
);
56 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
57 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
58 DEFINE_PER_CPU(unsigned long, xen_cr3
);
60 struct start_info
*xen_start_info
;
61 EXPORT_SYMBOL_GPL(xen_start_info
);
63 static /* __initdata */ struct shared_info dummy_shared_info
;
66 * Point at some empty memory to start with. We map the real shared_info
67 * page as soon as fixmap is up and running.
69 struct shared_info
*HYPERVISOR_shared_info
= (void *)&dummy_shared_info
;
72 * Flag to determine whether vcpu info placement is available on all
73 * VCPUs. We assume it is to start with, and then set it to zero on
74 * the first failure. This is because it can succeed on some VCPUs
75 * and not others, since it can involve hypervisor memory allocation,
76 * or because the guest failed to guarantee all the appropriate
77 * constraints on all VCPUs (ie buffer can't cross a page boundary).
79 * Note that any particular CPU may be using a placed vcpu structure,
80 * but we can only optimise if the all are.
82 * 0: not available, 1: available
84 static int have_vcpu_info_placement
= 1;
86 static void __init
xen_vcpu_setup(int cpu
)
88 struct vcpu_register_vcpu_info info
;
90 struct vcpu_info
*vcpup
;
92 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
94 if (!have_vcpu_info_placement
)
95 return; /* already tested, not available */
97 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
99 info
.mfn
= virt_to_mfn(vcpup
);
100 info
.offset
= offset_in_page(vcpup
);
102 printk(KERN_DEBUG
"trying to map vcpu_info %d at %p, mfn %x, offset %d\n",
103 cpu
, vcpup
, info
.mfn
, info
.offset
);
105 /* Check to see if the hypervisor will put the vcpu_info
106 structure where we want it, which allows direct access via
107 a percpu-variable. */
108 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
111 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
112 have_vcpu_info_placement
= 0;
114 /* This cpu is using the registered vcpu info, even if
115 later ones fail to. */
116 per_cpu(xen_vcpu
, cpu
) = vcpup
;
118 printk(KERN_DEBUG
"cpu %d using vcpu_info at %p\n",
123 static void __init
xen_banner(void)
125 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
127 printk(KERN_INFO
"Hypervisor signature: %s\n", xen_start_info
->magic
);
130 static void xen_cpuid(unsigned int *eax
, unsigned int *ebx
,
131 unsigned int *ecx
, unsigned int *edx
)
133 unsigned maskedx
= ~0;
136 * Mask out inconvenient features, to try and disable as many
137 * unsupported kernel subsystems as possible.
140 maskedx
= ~((1 << X86_FEATURE_APIC
) | /* disable APIC */
141 (1 << X86_FEATURE_ACPI
) | /* disable ACPI */
142 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
144 asm(XEN_EMULATE_PREFIX
"cpuid"
149 : "0" (*eax
), "2" (*ecx
));
153 static void xen_set_debugreg(int reg
, unsigned long val
)
155 HYPERVISOR_set_debugreg(reg
, val
);
158 static unsigned long xen_get_debugreg(int reg
)
160 return HYPERVISOR_get_debugreg(reg
);
163 static unsigned long xen_save_fl(void)
165 struct vcpu_info
*vcpu
;
168 vcpu
= x86_read_percpu(xen_vcpu
);
170 /* flag has opposite sense of mask */
171 flags
= !vcpu
->evtchn_upcall_mask
;
173 /* convert to IF type flag
177 return (-flags
) & X86_EFLAGS_IF
;
180 static void xen_restore_fl(unsigned long flags
)
182 struct vcpu_info
*vcpu
;
184 /* convert from IF type flag */
185 flags
= !(flags
& X86_EFLAGS_IF
);
187 /* There's a one instruction preempt window here. We need to
188 make sure we're don't switch CPUs between getting the vcpu
189 pointer and updating the mask. */
191 vcpu
= x86_read_percpu(xen_vcpu
);
192 vcpu
->evtchn_upcall_mask
= flags
;
193 preempt_enable_no_resched();
195 /* Doesn't matter if we get preempted here, because any
196 pending event will get dealt with anyway. */
199 preempt_check_resched();
200 barrier(); /* unmask then check (avoid races) */
201 if (unlikely(vcpu
->evtchn_upcall_pending
))
202 force_evtchn_callback();
206 static void xen_irq_disable(void)
208 /* There's a one instruction preempt window here. We need to
209 make sure we're don't switch CPUs between getting the vcpu
210 pointer and updating the mask. */
212 x86_read_percpu(xen_vcpu
)->evtchn_upcall_mask
= 1;
213 preempt_enable_no_resched();
216 static void xen_irq_enable(void)
218 struct vcpu_info
*vcpu
;
220 /* There's a one instruction preempt window here. We need to
221 make sure we're don't switch CPUs between getting the vcpu
222 pointer and updating the mask. */
224 vcpu
= x86_read_percpu(xen_vcpu
);
225 vcpu
->evtchn_upcall_mask
= 0;
226 preempt_enable_no_resched();
228 /* Doesn't matter if we get preempted here, because any
229 pending event will get dealt with anyway. */
231 barrier(); /* unmask then check (avoid races) */
232 if (unlikely(vcpu
->evtchn_upcall_pending
))
233 force_evtchn_callback();
236 static void xen_safe_halt(void)
238 /* Blocking includes an implicit local_irq_enable(). */
239 if (HYPERVISOR_sched_op(SCHEDOP_block
, 0) != 0)
243 static void xen_halt(void)
246 HYPERVISOR_vcpu_op(VCPUOP_down
, smp_processor_id(), NULL
);
251 static void xen_set_lazy_mode(enum paravirt_lazy_mode mode
)
253 BUG_ON(preemptible());
256 case PARAVIRT_LAZY_NONE
:
257 BUG_ON(x86_read_percpu(xen_lazy_mode
) == PARAVIRT_LAZY_NONE
);
260 case PARAVIRT_LAZY_MMU
:
261 case PARAVIRT_LAZY_CPU
:
262 BUG_ON(x86_read_percpu(xen_lazy_mode
) != PARAVIRT_LAZY_NONE
);
265 case PARAVIRT_LAZY_FLUSH
:
266 /* flush if necessary, but don't change state */
267 if (x86_read_percpu(xen_lazy_mode
) != PARAVIRT_LAZY_NONE
)
273 x86_write_percpu(xen_lazy_mode
, mode
);
276 static unsigned long xen_store_tr(void)
281 static void xen_set_ldt(const void *addr
, unsigned entries
)
283 unsigned long linear_addr
= (unsigned long)addr
;
284 struct mmuext_op
*op
;
285 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
288 op
->cmd
= MMUEXT_SET_LDT
;
290 /* ldt my be vmalloced, use arbitrary_virt_to_machine */
292 maddr
= arbitrary_virt_to_machine((unsigned long)addr
);
293 linear_addr
= (unsigned long)maddr
.maddr
;
295 op
->arg1
.linear_addr
= linear_addr
;
296 op
->arg2
.nr_ents
= entries
;
298 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
300 xen_mc_issue(PARAVIRT_LAZY_CPU
);
303 static void xen_load_gdt(const struct Xgt_desc_struct
*dtr
)
305 unsigned long *frames
;
306 unsigned long va
= dtr
->address
;
307 unsigned int size
= dtr
->size
+ 1;
308 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
310 struct multicall_space mcs
;
312 /* A GDT can be up to 64k in size, which corresponds to 8192
313 8-byte entries, or 16 4k pages.. */
315 BUG_ON(size
> 65536);
316 BUG_ON(va
& ~PAGE_MASK
);
318 mcs
= xen_mc_entry(sizeof(*frames
) * pages
);
321 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
322 frames
[f
] = virt_to_mfn(va
);
323 make_lowmem_page_readonly((void *)va
);
326 MULTI_set_gdt(mcs
.mc
, frames
, size
/ sizeof(struct desc_struct
));
328 xen_mc_issue(PARAVIRT_LAZY_CPU
);
331 static void load_TLS_descriptor(struct thread_struct
*t
,
332 unsigned int cpu
, unsigned int i
)
334 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
335 xmaddr_t maddr
= virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
336 struct multicall_space mc
= __xen_mc_entry(0);
338 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
341 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
345 load_TLS_descriptor(t
, cpu
, 0);
346 load_TLS_descriptor(t
, cpu
, 1);
347 load_TLS_descriptor(t
, cpu
, 2);
349 xen_mc_issue(PARAVIRT_LAZY_CPU
);
352 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
353 * it means we're in a context switch, and %gs has just been
354 * saved. This means we can zero it out to prevent faults on
355 * exit from the hypervisor if the next process has no %gs.
356 * Either way, it has been saved, and the new value will get
357 * loaded properly. This will go away as soon as Xen has been
358 * modified to not save/restore %gs for normal hypercalls.
360 if (xen_get_lazy_mode() == PARAVIRT_LAZY_CPU
)
364 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
367 unsigned long lp
= (unsigned long)&dt
[entrynum
];
368 xmaddr_t mach_lp
= virt_to_machine(lp
);
369 u64 entry
= (u64
)high
<< 32 | low
;
374 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
380 static int cvt_gate_to_trap(int vector
, u32 low
, u32 high
,
381 struct trap_info
*info
)
385 type
= (high
>> 8) & 0x1f;
386 dpl
= (high
>> 13) & 3;
388 if (type
!= 0xf && type
!= 0xe)
391 info
->vector
= vector
;
392 info
->address
= (high
& 0xffff0000) | (low
& 0x0000ffff);
393 info
->cs
= low
>> 16;
395 /* interrupt gates clear IF */
402 /* Locations of each CPU's IDT */
403 static DEFINE_PER_CPU(struct Xgt_desc_struct
, idt_desc
);
405 /* Set an IDT entry. If the entry is part of the current IDT, then
407 static void xen_write_idt_entry(struct desc_struct
*dt
, int entrynum
,
410 unsigned long p
= (unsigned long)&dt
[entrynum
];
411 unsigned long start
, end
;
415 start
= __get_cpu_var(idt_desc
).address
;
416 end
= start
+ __get_cpu_var(idt_desc
).size
+ 1;
420 write_dt_entry(dt
, entrynum
, low
, high
);
422 if (p
>= start
&& (p
+ 8) <= end
) {
423 struct trap_info info
[2];
427 if (cvt_gate_to_trap(entrynum
, low
, high
, &info
[0]))
428 if (HYPERVISOR_set_trap_table(info
))
435 static void xen_convert_trap_info(const struct Xgt_desc_struct
*desc
,
436 struct trap_info
*traps
)
438 unsigned in
, out
, count
;
440 count
= (desc
->size
+1) / 8;
443 for (in
= out
= 0; in
< count
; in
++) {
444 const u32
*entry
= (u32
*)(desc
->address
+ in
* 8);
446 if (cvt_gate_to_trap(in
, entry
[0], entry
[1], &traps
[out
]))
449 traps
[out
].address
= 0;
452 void xen_copy_trap_info(struct trap_info
*traps
)
454 const struct Xgt_desc_struct
*desc
= &__get_cpu_var(idt_desc
);
456 xen_convert_trap_info(desc
, traps
);
459 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
460 hold a spinlock to protect the static traps[] array (static because
461 it avoids allocation, and saves stack space). */
462 static void xen_load_idt(const struct Xgt_desc_struct
*desc
)
464 static DEFINE_SPINLOCK(lock
);
465 static struct trap_info traps
[257];
469 __get_cpu_var(idt_desc
) = *desc
;
471 xen_convert_trap_info(desc
, traps
);
474 if (HYPERVISOR_set_trap_table(traps
))
480 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
481 they're handled differently. */
482 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
487 switch ((high
>> 8) & 0xff) {
494 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
495 u64 desc
= (u64
)high
<< 32 | low
;
498 if (HYPERVISOR_update_descriptor(maddr
.maddr
, desc
))
507 static void xen_load_esp0(struct tss_struct
*tss
,
508 struct thread_struct
*thread
)
510 struct multicall_space mcs
= xen_mc_entry(0);
511 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->esp0
);
512 xen_mc_issue(PARAVIRT_LAZY_CPU
);
515 static void xen_set_iopl_mask(unsigned mask
)
517 struct physdev_set_iopl set_iopl
;
519 /* Force the change at ring 0. */
520 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
521 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
524 static void xen_io_delay(void)
528 #ifdef CONFIG_X86_LOCAL_APIC
529 static unsigned long xen_apic_read(unsigned long reg
)
534 static void xen_apic_write(unsigned long reg
, unsigned long val
)
536 /* Warn to see if there's any stray references */
541 static void xen_flush_tlb(void)
543 struct mmuext_op
*op
;
544 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
547 op
->cmd
= MMUEXT_TLB_FLUSH_LOCAL
;
548 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
550 xen_mc_issue(PARAVIRT_LAZY_MMU
);
553 static void xen_flush_tlb_single(unsigned long addr
)
555 struct mmuext_op
*op
;
556 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
559 op
->cmd
= MMUEXT_INVLPG_LOCAL
;
560 op
->arg1
.linear_addr
= addr
& PAGE_MASK
;
561 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
563 xen_mc_issue(PARAVIRT_LAZY_MMU
);
566 static void xen_flush_tlb_others(const cpumask_t
*cpus
, struct mm_struct
*mm
,
573 cpumask_t cpumask
= *cpus
;
574 struct multicall_space mcs
;
577 * A couple of (to be removed) sanity checks:
579 * - current CPU must not be in mask
580 * - mask must exist :)
582 BUG_ON(cpus_empty(cpumask
));
583 BUG_ON(cpu_isset(smp_processor_id(), cpumask
));
586 /* If a CPU which we ran on has gone down, OK. */
587 cpus_and(cpumask
, cpumask
, cpu_online_map
);
588 if (cpus_empty(cpumask
))
591 mcs
= xen_mc_entry(sizeof(*args
));
593 args
->mask
= cpumask
;
594 args
->op
.arg2
.vcpumask
= &args
->mask
;
596 if (va
== TLB_FLUSH_ALL
) {
597 args
->op
.cmd
= MMUEXT_TLB_FLUSH_MULTI
;
599 args
->op
.cmd
= MMUEXT_INVLPG_MULTI
;
600 args
->op
.arg1
.linear_addr
= va
;
603 MULTI_mmuext_op(mcs
.mc
, &args
->op
, 1, NULL
, DOMID_SELF
);
605 xen_mc_issue(PARAVIRT_LAZY_MMU
);
608 static void xen_write_cr2(unsigned long cr2
)
610 x86_read_percpu(xen_vcpu
)->arch
.cr2
= cr2
;
613 static unsigned long xen_read_cr2(void)
615 return x86_read_percpu(xen_vcpu
)->arch
.cr2
;
618 static unsigned long xen_read_cr2_direct(void)
620 return x86_read_percpu(xen_vcpu_info
.arch
.cr2
);
623 static void xen_write_cr4(unsigned long cr4
)
625 /* Just ignore cr4 changes; Xen doesn't allow us to do
629 static unsigned long xen_read_cr3(void)
631 return x86_read_percpu(xen_cr3
);
634 static void xen_write_cr3(unsigned long cr3
)
636 BUG_ON(preemptible());
638 if (cr3
== x86_read_percpu(xen_cr3
)) {
639 /* just a simple tlb flush */
644 x86_write_percpu(xen_cr3
, cr3
);
648 struct mmuext_op
*op
;
649 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
650 unsigned long mfn
= pfn_to_mfn(PFN_DOWN(cr3
));
653 op
->cmd
= MMUEXT_NEW_BASEPTR
;
656 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
658 xen_mc_issue(PARAVIRT_LAZY_CPU
);
662 /* Early in boot, while setting up the initial pagetable, assume
663 everything is pinned. */
664 static __init
void xen_alloc_pt_init(struct mm_struct
*mm
, u32 pfn
)
666 BUG_ON(mem_map
); /* should only be used early */
667 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
670 /* This needs to make sure the new pte page is pinned iff its being
671 attached to a pinned pagetable. */
672 static void xen_alloc_pt(struct mm_struct
*mm
, u32 pfn
)
674 struct page
*page
= pfn_to_page(pfn
);
676 if (PagePinned(virt_to_page(mm
->pgd
))) {
679 if (!PageHighMem(page
))
680 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
682 /* make sure there are no stray mappings of
688 /* This should never happen until we're OK to use struct page */
689 static void xen_release_pt(u32 pfn
)
691 struct page
*page
= pfn_to_page(pfn
);
693 if (PagePinned(page
)) {
694 if (!PageHighMem(page
))
695 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
699 #ifdef CONFIG_HIGHPTE
700 static void *xen_kmap_atomic_pte(struct page
*page
, enum km_type type
)
702 pgprot_t prot
= PAGE_KERNEL
;
704 if (PagePinned(page
))
705 prot
= PAGE_KERNEL_RO
;
707 if (0 && PageHighMem(page
))
708 printk("mapping highpte %lx type %d prot %s\n",
709 page_to_pfn(page
), type
,
710 (unsigned long)pgprot_val(prot
) & _PAGE_RW
? "WRITE" : "READ");
712 return kmap_atomic_prot(page
, type
, prot
);
716 static __init pte_t
mask_rw_pte(pte_t
*ptep
, pte_t pte
)
718 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
719 if (pte_val_ma(*ptep
) & _PAGE_PRESENT
)
720 pte
= __pte_ma(((pte_val_ma(*ptep
) & _PAGE_RW
) | ~_PAGE_RW
) &
726 /* Init-time set_pte while constructing initial pagetables, which
727 doesn't allow RO pagetable pages to be remapped RW */
728 static __init
void xen_set_pte_init(pte_t
*ptep
, pte_t pte
)
730 pte
= mask_rw_pte(ptep
, pte
);
732 xen_set_pte(ptep
, pte
);
735 static __init
void xen_pagetable_setup_start(pgd_t
*base
)
737 pgd_t
*xen_pgd
= (pgd_t
*)xen_start_info
->pt_base
;
739 /* special set_pte for pagetable initialization */
740 paravirt_ops
.set_pte
= xen_set_pte_init
;
744 * copy top-level of Xen-supplied pagetable into place. For
745 * !PAE we can use this as-is, but for PAE it is a stand-in
746 * while we copy the pmd pages.
748 memcpy(base
, xen_pgd
, PTRS_PER_PGD
* sizeof(pgd_t
));
750 if (PTRS_PER_PMD
> 1) {
753 * For PAE, need to allocate new pmds, rather than
754 * share Xen's, since Xen doesn't like pmd's being
755 * shared between address spaces.
757 for (i
= 0; i
< PTRS_PER_PGD
; i
++) {
758 if (pgd_val_ma(xen_pgd
[i
]) & _PAGE_PRESENT
) {
759 pmd_t
*pmd
= (pmd_t
*)alloc_bootmem_low_pages(PAGE_SIZE
);
761 memcpy(pmd
, (void *)pgd_page_vaddr(xen_pgd
[i
]),
764 make_lowmem_page_readonly(pmd
);
766 set_pgd(&base
[i
], __pgd(1 + __pa(pmd
)));
772 /* make sure zero_page is mapped RO so we can use it in pagetables */
773 make_lowmem_page_readonly(empty_zero_page
);
774 make_lowmem_page_readonly(base
);
776 * Switch to new pagetable. This is done before
777 * pagetable_init has done anything so that the new pages
778 * added to the table can be prepared properly for Xen.
780 xen_write_cr3(__pa(base
));
783 static __init
void xen_pagetable_setup_done(pgd_t
*base
)
785 /* This will work as long as patching hasn't happened yet
787 paravirt_ops
.alloc_pt
= xen_alloc_pt
;
788 paravirt_ops
.set_pte
= xen_set_pte
;
790 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
792 * Create a mapping for the shared info page.
793 * Should be set_fixmap(), but shared_info is a machine
794 * address with no corresponding pseudo-phys address.
796 set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP
),
797 PFN_DOWN(xen_start_info
->shared_info
),
800 HYPERVISOR_shared_info
=
801 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
804 HYPERVISOR_shared_info
=
805 (struct shared_info
*)__va(xen_start_info
->shared_info
);
807 /* Actually pin the pagetable down, but we can't set PG_pinned
808 yet because the page structures don't exist yet. */
811 #ifdef CONFIG_X86_PAE
812 op
.cmd
= MMUEXT_PIN_L3_TABLE
;
814 op
.cmd
= MMUEXT_PIN_L3_TABLE
;
816 op
.arg1
.mfn
= pfn_to_mfn(PFN_DOWN(__pa(base
)));
817 if (HYPERVISOR_mmuext_op(&op
, 1, NULL
, DOMID_SELF
))
822 /* This is called once we have the cpu_possible_map */
823 void __init
xen_setup_vcpu_info_placement(void)
827 for_each_possible_cpu(cpu
)
830 /* xen_vcpu_setup managed to place the vcpu_info within the
831 percpu area for all cpus, so make use of it */
832 if (have_vcpu_info_placement
) {
833 printk(KERN_INFO
"Xen: using vcpu_info placement\n");
835 paravirt_ops
.save_fl
= xen_save_fl_direct
;
836 paravirt_ops
.restore_fl
= xen_restore_fl_direct
;
837 paravirt_ops
.irq_disable
= xen_irq_disable_direct
;
838 paravirt_ops
.irq_enable
= xen_irq_enable_direct
;
839 paravirt_ops
.read_cr2
= xen_read_cr2_direct
;
840 paravirt_ops
.iret
= xen_iret_direct
;
844 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
845 unsigned long addr
, unsigned len
)
847 char *start
, *end
, *reloc
;
850 start
= end
= reloc
= NULL
;
853 case PARAVIRT_PATCH(x): \
854 if (have_vcpu_info_placement) { \
855 start = (char *)xen_##x##_direct; \
856 end = xen_##x##_direct_end; \
857 reloc = xen_##x##_direct_reloc; \
869 if (start
== NULL
|| (end
-start
) > len
)
872 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
874 /* Note: because reloc is assigned from something that
875 appears to be an array, gcc assumes it's non-null,
876 but doesn't know its relationship with start and
878 if (reloc
> start
&& reloc
< end
) {
879 int reloc_off
= reloc
- start
;
880 long *relocp
= (long *)(insnbuf
+ reloc_off
);
881 long delta
= start
- (char *)addr
;
889 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
897 static const struct paravirt_ops xen_paravirt_ops __initdata
= {
898 .paravirt_enabled
= 1,
899 .shared_kernel_pmd
= 0,
902 .banner
= xen_banner
,
906 .memory_setup
= xen_memory_setup
,
907 .arch_setup
= xen_arch_setup
,
908 .init_IRQ
= xen_init_IRQ
,
909 .post_allocator_init
= xen_mark_init_mm_pinned
,
911 .time_init
= xen_time_init
,
912 .set_wallclock
= xen_set_wallclock
,
913 .get_wallclock
= xen_get_wallclock
,
914 .get_cpu_khz
= xen_cpu_khz
,
915 .sched_clock
= xen_sched_clock
,
919 .set_debugreg
= xen_set_debugreg
,
920 .get_debugreg
= xen_get_debugreg
,
924 .read_cr0
= native_read_cr0
,
925 .write_cr0
= native_write_cr0
,
927 .read_cr2
= xen_read_cr2
,
928 .write_cr2
= xen_write_cr2
,
930 .read_cr3
= xen_read_cr3
,
931 .write_cr3
= xen_write_cr3
,
933 .read_cr4
= native_read_cr4
,
934 .read_cr4_safe
= native_read_cr4_safe
,
935 .write_cr4
= xen_write_cr4
,
937 .save_fl
= xen_save_fl
,
938 .restore_fl
= xen_restore_fl
,
939 .irq_disable
= xen_irq_disable
,
940 .irq_enable
= xen_irq_enable
,
941 .safe_halt
= xen_safe_halt
,
943 .wbinvd
= native_wbinvd
,
945 .read_msr
= native_read_msr_safe
,
946 .write_msr
= native_write_msr_safe
,
947 .read_tsc
= native_read_tsc
,
948 .read_pmc
= native_read_pmc
,
950 .iret
= (void *)&hypercall_page
[__HYPERVISOR_iret
],
951 .irq_enable_sysexit
= NULL
, /* never called */
953 .load_tr_desc
= paravirt_nop
,
954 .set_ldt
= xen_set_ldt
,
955 .load_gdt
= xen_load_gdt
,
956 .load_idt
= xen_load_idt
,
957 .load_tls
= xen_load_tls
,
959 .store_gdt
= native_store_gdt
,
960 .store_idt
= native_store_idt
,
961 .store_tr
= xen_store_tr
,
963 .write_ldt_entry
= xen_write_ldt_entry
,
964 .write_gdt_entry
= xen_write_gdt_entry
,
965 .write_idt_entry
= xen_write_idt_entry
,
966 .load_esp0
= xen_load_esp0
,
968 .set_iopl_mask
= xen_set_iopl_mask
,
969 .io_delay
= xen_io_delay
,
971 #ifdef CONFIG_X86_LOCAL_APIC
972 .apic_write
= xen_apic_write
,
973 .apic_write_atomic
= xen_apic_write
,
974 .apic_read
= xen_apic_read
,
975 .setup_boot_clock
= paravirt_nop
,
976 .setup_secondary_clock
= paravirt_nop
,
977 .startup_ipi_hook
= paravirt_nop
,
980 .flush_tlb_user
= xen_flush_tlb
,
981 .flush_tlb_kernel
= xen_flush_tlb
,
982 .flush_tlb_single
= xen_flush_tlb_single
,
983 .flush_tlb_others
= xen_flush_tlb_others
,
985 .pte_update
= paravirt_nop
,
986 .pte_update_defer
= paravirt_nop
,
988 .pagetable_setup_start
= xen_pagetable_setup_start
,
989 .pagetable_setup_done
= xen_pagetable_setup_done
,
991 .alloc_pt
= xen_alloc_pt_init
,
992 .release_pt
= xen_release_pt
,
993 .alloc_pd
= paravirt_nop
,
994 .alloc_pd_clone
= paravirt_nop
,
995 .release_pd
= paravirt_nop
,
997 #ifdef CONFIG_HIGHPTE
998 .kmap_atomic_pte
= xen_kmap_atomic_pte
,
1001 .set_pte
= NULL
, /* see xen_pagetable_setup_* */
1002 .set_pte_at
= xen_set_pte_at
,
1003 .set_pmd
= xen_set_pmd
,
1005 .pte_val
= xen_pte_val
,
1006 .pgd_val
= xen_pgd_val
,
1008 .make_pte
= xen_make_pte
,
1009 .make_pgd
= xen_make_pgd
,
1011 #ifdef CONFIG_X86_PAE
1012 .set_pte_atomic
= xen_set_pte_atomic
,
1013 .set_pte_present
= xen_set_pte_at
,
1014 .set_pud
= xen_set_pud
,
1015 .pte_clear
= xen_pte_clear
,
1016 .pmd_clear
= xen_pmd_clear
,
1018 .make_pmd
= xen_make_pmd
,
1019 .pmd_val
= xen_pmd_val
,
1022 .activate_mm
= xen_activate_mm
,
1023 .dup_mmap
= xen_dup_mmap
,
1024 .exit_mmap
= xen_exit_mmap
,
1026 .set_lazy_mode
= xen_set_lazy_mode
,
1030 static const struct smp_ops xen_smp_ops __initdata
= {
1031 .smp_prepare_boot_cpu
= xen_smp_prepare_boot_cpu
,
1032 .smp_prepare_cpus
= xen_smp_prepare_cpus
,
1033 .cpu_up
= xen_cpu_up
,
1034 .smp_cpus_done
= xen_smp_cpus_done
,
1036 .smp_send_stop
= xen_smp_send_stop
,
1037 .smp_send_reschedule
= xen_smp_send_reschedule
,
1038 .smp_call_function_mask
= xen_smp_call_function_mask
,
1040 #endif /* CONFIG_SMP */
1042 static void xen_reboot(int reason
)
1048 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, reason
))
1052 static void xen_restart(char *msg
)
1054 xen_reboot(SHUTDOWN_reboot
);
1057 static void xen_emergency_restart(void)
1059 xen_reboot(SHUTDOWN_reboot
);
1062 static void xen_machine_halt(void)
1064 xen_reboot(SHUTDOWN_poweroff
);
1067 static void xen_crash_shutdown(struct pt_regs
*regs
)
1069 xen_reboot(SHUTDOWN_crash
);
1072 static const struct machine_ops __initdata xen_machine_ops
= {
1073 .restart
= xen_restart
,
1074 .halt
= xen_machine_halt
,
1075 .power_off
= xen_machine_halt
,
1076 .shutdown
= xen_machine_halt
,
1077 .crash_shutdown
= xen_crash_shutdown
,
1078 .emergency_restart
= xen_emergency_restart
,
1082 /* First C function to be called on Xen boot */
1083 asmlinkage
void __init
xen_start_kernel(void)
1087 if (!xen_start_info
)
1090 BUG_ON(memcmp(xen_start_info
->magic
, "xen-3.0", 7) != 0);
1092 /* Install Xen paravirt ops */
1093 paravirt_ops
= xen_paravirt_ops
;
1094 machine_ops
= xen_machine_ops
;
1097 smp_ops
= xen_smp_ops
;
1100 xen_setup_features();
1103 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1104 phys_to_machine_mapping
= (unsigned long *)xen_start_info
->mfn_list
;
1106 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1108 init_pg_tables_end
= __pa(pgd
) + xen_start_info
->nr_pt_frames
*PAGE_SIZE
;
1110 init_mm
.pgd
= pgd
; /* use the Xen pagetables to start */
1112 /* keep using Xen gdt for now; no urgent need to change it */
1114 x86_write_percpu(xen_cr3
, __pa(pgd
));
1117 /* Don't do the full vcpu_info placement stuff until we have a
1119 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1121 /* May as well do it now, since there's no good time to call
1123 xen_setup_vcpu_info_placement();
1126 paravirt_ops
.kernel_rpl
= 1;
1127 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1128 paravirt_ops
.kernel_rpl
= 0;
1130 /* set the limit of our address space */
1131 reserve_top_address(-HYPERVISOR_VIRT_START
+ 2 * PAGE_SIZE
);
1133 /* set up basic CPUID stuff */
1134 cpu_detect(&new_cpu_data
);
1135 new_cpu_data
.hard_math
= 1;
1136 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1138 /* Poke various useful things into boot_params */
1139 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1140 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1141 ? __pa(xen_start_info
->mod_start
) : 0;
1142 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1144 /* Start the world */