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/cpu.h>
15 #include <linux/kernel.h>
16 #include <linux/init.h>
17 #include <linux/smp.h>
18 #include <linux/preempt.h>
19 #include <linux/hardirq.h>
20 #include <linux/percpu.h>
21 #include <linux/delay.h>
22 #include <linux/start_kernel.h>
23 #include <linux/sched.h>
24 #include <linux/kprobes.h>
25 #include <linux/bootmem.h>
26 #include <linux/module.h>
28 #include <linux/page-flags.h>
29 #include <linux/highmem.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/memblock.h>
34 #include <linux/edd.h>
37 #include <xen/events.h>
38 #include <xen/interface/xen.h>
39 #include <xen/interface/version.h>
40 #include <xen/interface/physdev.h>
41 #include <xen/interface/vcpu.h>
42 #include <xen/interface/memory.h>
43 #include <xen/interface/xen-mca.h>
44 #include <xen/features.h>
47 #include <xen/hvc-console.h>
49 #include <xen/features.h>
51 #include <asm/paravirt.h>
54 #include <asm/xen/pci.h>
55 #include <asm/xen/hypercall.h>
56 #include <asm/xen/hypervisor.h>
57 #include <asm/fixmap.h>
58 #include <asm/processor.h>
59 #include <asm/proto.h>
60 #include <asm/msr-index.h>
61 #include <asm/traps.h>
62 #include <asm/setup.h>
64 #include <asm/pgalloc.h>
65 #include <asm/pgtable.h>
66 #include <asm/tlbflush.h>
67 #include <asm/reboot.h>
68 #include <asm/stackprotector.h>
69 #include <asm/hypervisor.h>
70 #include <asm/mwait.h>
71 #include <asm/pci_x86.h>
75 #include <linux/acpi.h>
77 #include <acpi/pdc_intel.h>
78 #include <acpi/processor.h>
79 #include <xen/interface/platform.h>
85 #include "multicalls.h"
87 EXPORT_SYMBOL_GPL(hypercall_page
);
90 * Pointer to the xen_vcpu_info structure or
91 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
92 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
93 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
94 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
95 * acknowledge pending events.
96 * Also more subtly it is used by the patched version of irq enable/disable
97 * e.g. xen_irq_enable_direct and xen_iret in PV mode.
99 * The desire to be able to do those mask/unmask operations as a single
100 * instruction by using the per-cpu offset held in %gs is the real reason
101 * vcpu info is in a per-cpu pointer and the original reason for this
105 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
108 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
109 * hypercall. This can be used both in PV and PVHVM mode. The structure
110 * overrides the default per_cpu(xen_vcpu, cpu) value.
112 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
114 enum xen_domain_type xen_domain_type
= XEN_NATIVE
;
115 EXPORT_SYMBOL_GPL(xen_domain_type
);
117 unsigned long *machine_to_phys_mapping
= (void *)MACH2PHYS_VIRT_START
;
118 EXPORT_SYMBOL(machine_to_phys_mapping
);
119 unsigned long machine_to_phys_nr
;
120 EXPORT_SYMBOL(machine_to_phys_nr
);
122 struct start_info
*xen_start_info
;
123 EXPORT_SYMBOL_GPL(xen_start_info
);
125 struct shared_info xen_dummy_shared_info
;
127 void *xen_initial_gdt
;
129 RESERVE_BRK(shared_info_page_brk
, PAGE_SIZE
);
130 __read_mostly
int xen_have_vector_callback
;
131 EXPORT_SYMBOL_GPL(xen_have_vector_callback
);
134 * Point at some empty memory to start with. We map the real shared_info
135 * page as soon as fixmap is up and running.
137 struct shared_info
*HYPERVISOR_shared_info
= &xen_dummy_shared_info
;
140 * Flag to determine whether vcpu info placement is available on all
141 * VCPUs. We assume it is to start with, and then set it to zero on
142 * the first failure. This is because it can succeed on some VCPUs
143 * and not others, since it can involve hypervisor memory allocation,
144 * or because the guest failed to guarantee all the appropriate
145 * constraints on all VCPUs (ie buffer can't cross a page boundary).
147 * Note that any particular CPU may be using a placed vcpu structure,
148 * but we can only optimise if the all are.
150 * 0: not available, 1: available
152 static int have_vcpu_info_placement
= 1;
155 struct desc_struct desc
[3];
159 * Updating the 3 TLS descriptors in the GDT on every task switch is
160 * surprisingly expensive so we avoid updating them if they haven't
161 * changed. Since Xen writes different descriptors than the one
162 * passed in the update_descriptor hypercall we keep shadow copies to
165 static DEFINE_PER_CPU(struct tls_descs
, shadow_tls_desc
);
167 static void clamp_max_cpus(void)
170 if (setup_max_cpus
> MAX_VIRT_CPUS
)
171 setup_max_cpus
= MAX_VIRT_CPUS
;
175 static void xen_vcpu_setup(int cpu
)
177 struct vcpu_register_vcpu_info info
;
179 struct vcpu_info
*vcpup
;
181 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
184 * This path is called twice on PVHVM - first during bootup via
185 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
186 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
187 * As we can only do the VCPUOP_register_vcpu_info once lets
188 * not over-write its result.
190 * For PV it is called during restore (xen_vcpu_restore) and bootup
191 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
194 if (xen_hvm_domain()) {
195 if (per_cpu(xen_vcpu
, cpu
) == &per_cpu(xen_vcpu_info
, cpu
))
198 if (cpu
< MAX_VIRT_CPUS
)
199 per_cpu(xen_vcpu
,cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
201 if (!have_vcpu_info_placement
) {
202 if (cpu
>= MAX_VIRT_CPUS
)
207 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
208 info
.mfn
= arbitrary_virt_to_mfn(vcpup
);
209 info
.offset
= offset_in_page(vcpup
);
211 /* Check to see if the hypervisor will put the vcpu_info
212 structure where we want it, which allows direct access via
214 N.B. This hypercall can _only_ be called once per CPU. Subsequent
215 calls will error out with -EINVAL. This is due to the fact that
216 hypervisor has no unregister variant and this hypercall does not
217 allow to over-write info.mfn and info.offset.
219 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
222 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
223 have_vcpu_info_placement
= 0;
226 /* This cpu is using the registered vcpu info, even if
227 later ones fail to. */
228 per_cpu(xen_vcpu
, cpu
) = vcpup
;
233 * On restore, set the vcpu placement up again.
234 * If it fails, then we're in a bad state, since
235 * we can't back out from using it...
237 void xen_vcpu_restore(void)
241 for_each_possible_cpu(cpu
) {
242 bool other_cpu
= (cpu
!= smp_processor_id());
243 bool is_up
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, cpu
, NULL
);
245 if (other_cpu
&& is_up
&&
246 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
249 xen_setup_runstate_info(cpu
);
251 if (have_vcpu_info_placement
)
254 if (other_cpu
&& is_up
&&
255 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
260 static void __init
xen_banner(void)
262 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
263 struct xen_extraversion extra
;
264 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
266 pr_info("Booting paravirtualized kernel %son %s\n",
267 xen_feature(XENFEAT_auto_translated_physmap
) ?
268 "with PVH extensions " : "", pv_info
.name
);
269 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
270 version
>> 16, version
& 0xffff, extra
.extraversion
,
271 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
273 /* Check if running on Xen version (major, minor) or later */
275 xen_running_on_version_or_later(unsigned int major
, unsigned int minor
)
277 unsigned int version
;
282 version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
283 if ((((version
>> 16) == major
) && ((version
& 0xffff) >= minor
)) ||
284 ((version
>> 16) > major
))
289 #define CPUID_THERM_POWER_LEAF 6
290 #define APERFMPERF_PRESENT 0
292 static __read_mostly
unsigned int cpuid_leaf1_edx_mask
= ~0;
293 static __read_mostly
unsigned int cpuid_leaf1_ecx_mask
= ~0;
295 static __read_mostly
unsigned int cpuid_leaf1_ecx_set_mask
;
296 static __read_mostly
unsigned int cpuid_leaf5_ecx_val
;
297 static __read_mostly
unsigned int cpuid_leaf5_edx_val
;
299 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
300 unsigned int *cx
, unsigned int *dx
)
302 unsigned maskebx
= ~0;
303 unsigned maskecx
= ~0;
304 unsigned maskedx
= ~0;
307 * Mask out inconvenient features, to try and disable as many
308 * unsupported kernel subsystems as possible.
312 maskecx
= cpuid_leaf1_ecx_mask
;
313 setecx
= cpuid_leaf1_ecx_set_mask
;
314 maskedx
= cpuid_leaf1_edx_mask
;
317 case CPUID_MWAIT_LEAF
:
318 /* Synthesize the values.. */
321 *cx
= cpuid_leaf5_ecx_val
;
322 *dx
= cpuid_leaf5_edx_val
;
325 case CPUID_THERM_POWER_LEAF
:
326 /* Disabling APERFMPERF for kernel usage */
327 maskecx
= ~(1 << APERFMPERF_PRESENT
);
331 /* Suppress extended topology stuff */
336 asm(XEN_EMULATE_PREFIX
"cpuid"
341 : "0" (*ax
), "2" (*cx
));
350 static bool __init
xen_check_mwait(void)
353 struct xen_platform_op op
= {
354 .cmd
= XENPF_set_processor_pminfo
,
355 .u
.set_pminfo
.id
= -1,
356 .u
.set_pminfo
.type
= XEN_PM_PDC
,
359 unsigned int ax
, bx
, cx
, dx
;
360 unsigned int mwait_mask
;
362 /* We need to determine whether it is OK to expose the MWAIT
363 * capability to the kernel to harvest deeper than C3 states from ACPI
364 * _CST using the processor_harvest_xen.c module. For this to work, we
365 * need to gather the MWAIT_LEAF values (which the cstate.c code
366 * checks against). The hypervisor won't expose the MWAIT flag because
367 * it would break backwards compatibility; so we will find out directly
368 * from the hardware and hypercall.
370 if (!xen_initial_domain())
374 * When running under platform earlier than Xen4.2, do not expose
375 * mwait, to avoid the risk of loading native acpi pad driver
377 if (!xen_running_on_version_or_later(4, 2))
383 native_cpuid(&ax
, &bx
, &cx
, &dx
);
385 mwait_mask
= (1 << (X86_FEATURE_EST
% 32)) |
386 (1 << (X86_FEATURE_MWAIT
% 32));
388 if ((cx
& mwait_mask
) != mwait_mask
)
391 /* We need to emulate the MWAIT_LEAF and for that we need both
392 * ecx and edx. The hypercall provides only partial information.
395 ax
= CPUID_MWAIT_LEAF
;
400 native_cpuid(&ax
, &bx
, &cx
, &dx
);
402 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
403 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
405 buf
[0] = ACPI_PDC_REVISION_ID
;
407 buf
[2] = (ACPI_PDC_C_CAPABILITY_SMP
| ACPI_PDC_EST_CAPABILITY_SWSMP
);
409 set_xen_guest_handle(op
.u
.set_pminfo
.pdc
, buf
);
411 if ((HYPERVISOR_dom0_op(&op
) == 0) &&
412 (buf
[2] & (ACPI_PDC_C_C1_FFH
| ACPI_PDC_C_C2C3_FFH
))) {
413 cpuid_leaf5_ecx_val
= cx
;
414 cpuid_leaf5_edx_val
= dx
;
421 static void __init
xen_init_cpuid_mask(void)
423 unsigned int ax
, bx
, cx
, dx
;
424 unsigned int xsave_mask
;
426 cpuid_leaf1_edx_mask
=
427 ~((1 << X86_FEATURE_MTRR
) | /* disable MTRR */
428 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
430 if (!xen_initial_domain())
431 cpuid_leaf1_edx_mask
&=
432 ~((1 << X86_FEATURE_ACPI
)); /* disable ACPI */
434 cpuid_leaf1_ecx_mask
&= ~(1 << (X86_FEATURE_X2APIC
% 32));
438 cpuid(1, &ax
, &bx
, &cx
, &dx
);
441 (1 << (X86_FEATURE_XSAVE
% 32)) |
442 (1 << (X86_FEATURE_OSXSAVE
% 32));
444 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
445 if ((cx
& xsave_mask
) != xsave_mask
)
446 cpuid_leaf1_ecx_mask
&= ~xsave_mask
; /* disable XSAVE & OSXSAVE */
447 if (xen_check_mwait())
448 cpuid_leaf1_ecx_set_mask
= (1 << (X86_FEATURE_MWAIT
% 32));
451 static void xen_set_debugreg(int reg
, unsigned long val
)
453 HYPERVISOR_set_debugreg(reg
, val
);
456 static unsigned long xen_get_debugreg(int reg
)
458 return HYPERVISOR_get_debugreg(reg
);
461 static void xen_end_context_switch(struct task_struct
*next
)
464 paravirt_end_context_switch(next
);
467 static unsigned long xen_store_tr(void)
473 * Set the page permissions for a particular virtual address. If the
474 * address is a vmalloc mapping (or other non-linear mapping), then
475 * find the linear mapping of the page and also set its protections to
478 static void set_aliased_prot(void *v
, pgprot_t prot
)
486 ptep
= lookup_address((unsigned long)v
, &level
);
487 BUG_ON(ptep
== NULL
);
489 pfn
= pte_pfn(*ptep
);
490 page
= pfn_to_page(pfn
);
492 pte
= pfn_pte(pfn
, prot
);
494 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
497 if (!PageHighMem(page
)) {
498 void *av
= __va(PFN_PHYS(pfn
));
501 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
507 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
509 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
512 for(i
= 0; i
< entries
; i
+= entries_per_page
)
513 set_aliased_prot(ldt
+ i
, PAGE_KERNEL_RO
);
516 static void xen_free_ldt(struct desc_struct
*ldt
, unsigned entries
)
518 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
521 for(i
= 0; i
< entries
; i
+= entries_per_page
)
522 set_aliased_prot(ldt
+ i
, PAGE_KERNEL
);
525 static void xen_set_ldt(const void *addr
, unsigned entries
)
527 struct mmuext_op
*op
;
528 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
530 trace_xen_cpu_set_ldt(addr
, entries
);
533 op
->cmd
= MMUEXT_SET_LDT
;
534 op
->arg1
.linear_addr
= (unsigned long)addr
;
535 op
->arg2
.nr_ents
= entries
;
537 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
539 xen_mc_issue(PARAVIRT_LAZY_CPU
);
542 static void xen_load_gdt(const struct desc_ptr
*dtr
)
544 unsigned long va
= dtr
->address
;
545 unsigned int size
= dtr
->size
+ 1;
546 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
547 unsigned long frames
[pages
];
551 * A GDT can be up to 64k in size, which corresponds to 8192
552 * 8-byte entries, or 16 4k pages..
555 BUG_ON(size
> 65536);
556 BUG_ON(va
& ~PAGE_MASK
);
558 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
561 unsigned long pfn
, mfn
;
565 * The GDT is per-cpu and is in the percpu data area.
566 * That can be virtually mapped, so we need to do a
567 * page-walk to get the underlying MFN for the
568 * hypercall. The page can also be in the kernel's
569 * linear range, so we need to RO that mapping too.
571 ptep
= lookup_address(va
, &level
);
572 BUG_ON(ptep
== NULL
);
574 pfn
= pte_pfn(*ptep
);
575 mfn
= pfn_to_mfn(pfn
);
576 virt
= __va(PFN_PHYS(pfn
));
580 make_lowmem_page_readonly((void *)va
);
581 make_lowmem_page_readonly(virt
);
584 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
589 * load_gdt for early boot, when the gdt is only mapped once
591 static void __init
xen_load_gdt_boot(const struct desc_ptr
*dtr
)
593 unsigned long va
= dtr
->address
;
594 unsigned int size
= dtr
->size
+ 1;
595 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
596 unsigned long frames
[pages
];
600 * A GDT can be up to 64k in size, which corresponds to 8192
601 * 8-byte entries, or 16 4k pages..
604 BUG_ON(size
> 65536);
605 BUG_ON(va
& ~PAGE_MASK
);
607 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
609 unsigned long pfn
, mfn
;
611 pfn
= virt_to_pfn(va
);
612 mfn
= pfn_to_mfn(pfn
);
614 pte
= pfn_pte(pfn
, PAGE_KERNEL_RO
);
616 if (HYPERVISOR_update_va_mapping((unsigned long)va
, pte
, 0))
622 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
626 static inline bool desc_equal(const struct desc_struct
*d1
,
627 const struct desc_struct
*d2
)
629 return d1
->a
== d2
->a
&& d1
->b
== d2
->b
;
632 static void load_TLS_descriptor(struct thread_struct
*t
,
633 unsigned int cpu
, unsigned int i
)
635 struct desc_struct
*shadow
= &per_cpu(shadow_tls_desc
, cpu
).desc
[i
];
636 struct desc_struct
*gdt
;
638 struct multicall_space mc
;
640 if (desc_equal(shadow
, &t
->tls_array
[i
]))
643 *shadow
= t
->tls_array
[i
];
645 gdt
= get_cpu_gdt_table(cpu
);
646 maddr
= arbitrary_virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
647 mc
= __xen_mc_entry(0);
649 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
652 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
655 * XXX sleazy hack: If we're being called in a lazy-cpu zone
656 * and lazy gs handling is enabled, it means we're in a
657 * context switch, and %gs has just been saved. This means we
658 * can zero it out to prevent faults on exit from the
659 * hypervisor if the next process has no %gs. Either way, it
660 * has been saved, and the new value will get loaded properly.
661 * This will go away as soon as Xen has been modified to not
662 * save/restore %gs for normal hypercalls.
664 * On x86_64, this hack is not used for %gs, because gs points
665 * to KERNEL_GS_BASE (and uses it for PDA references), so we
666 * must not zero %gs on x86_64
668 * For x86_64, we need to zero %fs, otherwise we may get an
669 * exception between the new %fs descriptor being loaded and
670 * %fs being effectively cleared at __switch_to().
672 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
682 load_TLS_descriptor(t
, cpu
, 0);
683 load_TLS_descriptor(t
, cpu
, 1);
684 load_TLS_descriptor(t
, cpu
, 2);
686 xen_mc_issue(PARAVIRT_LAZY_CPU
);
690 static void xen_load_gs_index(unsigned int idx
)
692 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
697 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
700 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
701 u64 entry
= *(u64
*)ptr
;
703 trace_xen_cpu_write_ldt_entry(dt
, entrynum
, entry
);
708 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
714 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
715 struct trap_info
*info
)
719 if (val
->type
!= GATE_TRAP
&& val
->type
!= GATE_INTERRUPT
)
722 info
->vector
= vector
;
724 addr
= gate_offset(*val
);
727 * Look for known traps using IST, and substitute them
728 * appropriately. The debugger ones are the only ones we care
729 * about. Xen will handle faults like double_fault,
730 * so we should never see them. Warn if
731 * there's an unexpected IST-using fault handler.
733 if (addr
== (unsigned long)debug
)
734 addr
= (unsigned long)xen_debug
;
735 else if (addr
== (unsigned long)int3
)
736 addr
= (unsigned long)xen_int3
;
737 else if (addr
== (unsigned long)stack_segment
)
738 addr
= (unsigned long)xen_stack_segment
;
739 else if (addr
== (unsigned long)double_fault
) {
740 /* Don't need to handle these */
742 #ifdef CONFIG_X86_MCE
743 } else if (addr
== (unsigned long)machine_check
) {
745 * when xen hypervisor inject vMCE to guest,
746 * use native mce handler to handle it
750 } else if (addr
== (unsigned long)nmi
)
752 * Use the native version as well.
756 /* Some other trap using IST? */
757 if (WARN_ON(val
->ist
!= 0))
760 #endif /* CONFIG_X86_64 */
761 info
->address
= addr
;
763 info
->cs
= gate_segment(*val
);
764 info
->flags
= val
->dpl
;
765 /* interrupt gates clear IF */
766 if (val
->type
== GATE_INTERRUPT
)
767 info
->flags
|= 1 << 2;
772 /* Locations of each CPU's IDT */
773 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
775 /* Set an IDT entry. If the entry is part of the current IDT, then
777 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
779 unsigned long p
= (unsigned long)&dt
[entrynum
];
780 unsigned long start
, end
;
782 trace_xen_cpu_write_idt_entry(dt
, entrynum
, g
);
786 start
= __this_cpu_read(idt_desc
.address
);
787 end
= start
+ __this_cpu_read(idt_desc
.size
) + 1;
791 native_write_idt_entry(dt
, entrynum
, g
);
793 if (p
>= start
&& (p
+ 8) <= end
) {
794 struct trap_info info
[2];
798 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
799 if (HYPERVISOR_set_trap_table(info
))
806 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
807 struct trap_info
*traps
)
809 unsigned in
, out
, count
;
811 count
= (desc
->size
+1) / sizeof(gate_desc
);
814 for (in
= out
= 0; in
< count
; in
++) {
815 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
817 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
820 traps
[out
].address
= 0;
823 void xen_copy_trap_info(struct trap_info
*traps
)
825 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
827 xen_convert_trap_info(desc
, traps
);
830 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
831 hold a spinlock to protect the static traps[] array (static because
832 it avoids allocation, and saves stack space). */
833 static void xen_load_idt(const struct desc_ptr
*desc
)
835 static DEFINE_SPINLOCK(lock
);
836 static struct trap_info traps
[257];
838 trace_xen_cpu_load_idt(desc
);
842 __get_cpu_var(idt_desc
) = *desc
;
844 xen_convert_trap_info(desc
, traps
);
847 if (HYPERVISOR_set_trap_table(traps
))
853 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
854 they're handled differently. */
855 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
856 const void *desc
, int type
)
858 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
869 xmaddr_t maddr
= arbitrary_virt_to_machine(&dt
[entry
]);
872 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
882 * Version of write_gdt_entry for use at early boot-time needed to
883 * update an entry as simply as possible.
885 static void __init
xen_write_gdt_entry_boot(struct desc_struct
*dt
, int entry
,
886 const void *desc
, int type
)
888 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
897 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
899 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
900 dt
[entry
] = *(struct desc_struct
*)desc
;
906 static void xen_load_sp0(struct tss_struct
*tss
,
907 struct thread_struct
*thread
)
909 struct multicall_space mcs
;
911 mcs
= xen_mc_entry(0);
912 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
913 xen_mc_issue(PARAVIRT_LAZY_CPU
);
916 static void xen_set_iopl_mask(unsigned mask
)
918 struct physdev_set_iopl set_iopl
;
920 /* Force the change at ring 0. */
921 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
922 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
925 static void xen_io_delay(void)
929 #ifdef CONFIG_X86_LOCAL_APIC
930 static unsigned long xen_set_apic_id(unsigned int x
)
935 static unsigned int xen_get_apic_id(unsigned long x
)
937 return ((x
)>>24) & 0xFFu
;
939 static u32
xen_apic_read(u32 reg
)
941 struct xen_platform_op op
= {
942 .cmd
= XENPF_get_cpuinfo
,
943 .interface_version
= XENPF_INTERFACE_VERSION
,
944 .u
.pcpu_info
.xen_cpuid
= 0,
948 /* Shouldn't need this as APIC is turned off for PV, and we only
949 * get called on the bootup processor. But just in case. */
950 if (!xen_initial_domain() || smp_processor_id())
959 ret
= HYPERVISOR_dom0_op(&op
);
963 return op
.u
.pcpu_info
.apic_id
<< 24;
966 static void xen_apic_write(u32 reg
, u32 val
)
968 /* Warn to see if there's any stray references */
972 static u64
xen_apic_icr_read(void)
977 static void xen_apic_icr_write(u32 low
, u32 id
)
979 /* Warn to see if there's any stray references */
983 static void xen_apic_wait_icr_idle(void)
988 static u32
xen_safe_apic_wait_icr_idle(void)
993 static void set_xen_basic_apic_ops(void)
995 apic
->read
= xen_apic_read
;
996 apic
->write
= xen_apic_write
;
997 apic
->icr_read
= xen_apic_icr_read
;
998 apic
->icr_write
= xen_apic_icr_write
;
999 apic
->wait_icr_idle
= xen_apic_wait_icr_idle
;
1000 apic
->safe_wait_icr_idle
= xen_safe_apic_wait_icr_idle
;
1001 apic
->set_apic_id
= xen_set_apic_id
;
1002 apic
->get_apic_id
= xen_get_apic_id
;
1005 apic
->send_IPI_allbutself
= xen_send_IPI_allbutself
;
1006 apic
->send_IPI_mask_allbutself
= xen_send_IPI_mask_allbutself
;
1007 apic
->send_IPI_mask
= xen_send_IPI_mask
;
1008 apic
->send_IPI_all
= xen_send_IPI_all
;
1009 apic
->send_IPI_self
= xen_send_IPI_self
;
1015 static void xen_clts(void)
1017 struct multicall_space mcs
;
1019 mcs
= xen_mc_entry(0);
1021 MULTI_fpu_taskswitch(mcs
.mc
, 0);
1023 xen_mc_issue(PARAVIRT_LAZY_CPU
);
1026 static DEFINE_PER_CPU(unsigned long, xen_cr0_value
);
1028 static unsigned long xen_read_cr0(void)
1030 unsigned long cr0
= this_cpu_read(xen_cr0_value
);
1032 if (unlikely(cr0
== 0)) {
1033 cr0
= native_read_cr0();
1034 this_cpu_write(xen_cr0_value
, cr0
);
1040 static void xen_write_cr0(unsigned long cr0
)
1042 struct multicall_space mcs
;
1044 this_cpu_write(xen_cr0_value
, cr0
);
1046 /* Only pay attention to cr0.TS; everything else is
1048 mcs
= xen_mc_entry(0);
1050 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
1052 xen_mc_issue(PARAVIRT_LAZY_CPU
);
1055 static void xen_write_cr4(unsigned long cr4
)
1057 cr4
&= ~X86_CR4_PGE
;
1058 cr4
&= ~X86_CR4_PSE
;
1060 native_write_cr4(cr4
);
1062 #ifdef CONFIG_X86_64
1063 static inline unsigned long xen_read_cr8(void)
1067 static inline void xen_write_cr8(unsigned long val
)
1072 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
1079 #ifdef CONFIG_X86_64
1083 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
1084 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
1085 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
1088 base
= ((u64
)high
<< 32) | low
;
1089 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
1097 case MSR_SYSCALL_MASK
:
1098 case MSR_IA32_SYSENTER_CS
:
1099 case MSR_IA32_SYSENTER_ESP
:
1100 case MSR_IA32_SYSENTER_EIP
:
1101 /* Fast syscall setup is all done in hypercalls, so
1102 these are all ignored. Stub them out here to stop
1103 Xen console noise. */
1106 case MSR_IA32_CR_PAT
:
1107 if (smp_processor_id() == 0)
1108 xen_set_pat(((u64
)high
<< 32) | low
);
1112 ret
= native_write_msr_safe(msr
, low
, high
);
1118 void xen_setup_shared_info(void)
1120 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1121 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
1122 xen_start_info
->shared_info
);
1124 HYPERVISOR_shared_info
=
1125 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
1127 HYPERVISOR_shared_info
=
1128 (struct shared_info
*)__va(xen_start_info
->shared_info
);
1131 /* In UP this is as good a place as any to set up shared info */
1132 xen_setup_vcpu_info_placement();
1135 xen_setup_mfn_list_list();
1138 /* This is called once we have the cpu_possible_mask */
1139 void xen_setup_vcpu_info_placement(void)
1143 for_each_possible_cpu(cpu
)
1144 xen_vcpu_setup(cpu
);
1146 /* xen_vcpu_setup managed to place the vcpu_info within the
1147 percpu area for all cpus, so make use of it */
1148 if (have_vcpu_info_placement
) {
1149 pv_irq_ops
.save_fl
= __PV_IS_CALLEE_SAVE(xen_save_fl_direct
);
1150 pv_irq_ops
.restore_fl
= __PV_IS_CALLEE_SAVE(xen_restore_fl_direct
);
1151 pv_irq_ops
.irq_disable
= __PV_IS_CALLEE_SAVE(xen_irq_disable_direct
);
1152 pv_irq_ops
.irq_enable
= __PV_IS_CALLEE_SAVE(xen_irq_enable_direct
);
1153 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1157 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1158 unsigned long addr
, unsigned len
)
1160 char *start
, *end
, *reloc
;
1163 start
= end
= reloc
= NULL
;
1165 #define SITE(op, x) \
1166 case PARAVIRT_PATCH(op.x): \
1167 if (have_vcpu_info_placement) { \
1168 start = (char *)xen_##x##_direct; \
1169 end = xen_##x##_direct_end; \
1170 reloc = xen_##x##_direct_reloc; \
1175 SITE(pv_irq_ops
, irq_enable
);
1176 SITE(pv_irq_ops
, irq_disable
);
1177 SITE(pv_irq_ops
, save_fl
);
1178 SITE(pv_irq_ops
, restore_fl
);
1182 if (start
== NULL
|| (end
-start
) > len
)
1185 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1187 /* Note: because reloc is assigned from something that
1188 appears to be an array, gcc assumes it's non-null,
1189 but doesn't know its relationship with start and
1191 if (reloc
> start
&& reloc
< end
) {
1192 int reloc_off
= reloc
- start
;
1193 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1194 long delta
= start
- (char *)addr
;
1202 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1210 static const struct pv_info xen_info __initconst
= {
1211 .paravirt_enabled
= 1,
1212 .shared_kernel_pmd
= 0,
1214 #ifdef CONFIG_X86_64
1215 .extra_user_64bit_cs
= FLAT_USER_CS64
,
1221 static const struct pv_init_ops xen_init_ops __initconst
= {
1225 static const struct pv_cpu_ops xen_cpu_ops __initconst
= {
1228 .set_debugreg
= xen_set_debugreg
,
1229 .get_debugreg
= xen_get_debugreg
,
1233 .read_cr0
= xen_read_cr0
,
1234 .write_cr0
= xen_write_cr0
,
1236 .read_cr4
= native_read_cr4
,
1237 .read_cr4_safe
= native_read_cr4_safe
,
1238 .write_cr4
= xen_write_cr4
,
1240 #ifdef CONFIG_X86_64
1241 .read_cr8
= xen_read_cr8
,
1242 .write_cr8
= xen_write_cr8
,
1245 .wbinvd
= native_wbinvd
,
1247 .read_msr
= native_read_msr_safe
,
1248 .write_msr
= xen_write_msr_safe
,
1250 .read_tsc
= native_read_tsc
,
1251 .read_pmc
= native_read_pmc
,
1253 .read_tscp
= native_read_tscp
,
1256 .irq_enable_sysexit
= xen_sysexit
,
1257 #ifdef CONFIG_X86_64
1258 .usergs_sysret32
= xen_sysret32
,
1259 .usergs_sysret64
= xen_sysret64
,
1262 .load_tr_desc
= paravirt_nop
,
1263 .set_ldt
= xen_set_ldt
,
1264 .load_gdt
= xen_load_gdt
,
1265 .load_idt
= xen_load_idt
,
1266 .load_tls
= xen_load_tls
,
1267 #ifdef CONFIG_X86_64
1268 .load_gs_index
= xen_load_gs_index
,
1271 .alloc_ldt
= xen_alloc_ldt
,
1272 .free_ldt
= xen_free_ldt
,
1274 .store_idt
= native_store_idt
,
1275 .store_tr
= xen_store_tr
,
1277 .write_ldt_entry
= xen_write_ldt_entry
,
1278 .write_gdt_entry
= xen_write_gdt_entry
,
1279 .write_idt_entry
= xen_write_idt_entry
,
1280 .load_sp0
= xen_load_sp0
,
1282 .set_iopl_mask
= xen_set_iopl_mask
,
1283 .io_delay
= xen_io_delay
,
1285 /* Xen takes care of %gs when switching to usermode for us */
1286 .swapgs
= paravirt_nop
,
1288 .start_context_switch
= paravirt_start_context_switch
,
1289 .end_context_switch
= xen_end_context_switch
,
1292 static const struct pv_apic_ops xen_apic_ops __initconst
= {
1293 #ifdef CONFIG_X86_LOCAL_APIC
1294 .startup_ipi_hook
= paravirt_nop
,
1298 static void xen_reboot(int reason
)
1300 struct sched_shutdown r
= { .reason
= reason
};
1302 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1306 static void xen_restart(char *msg
)
1308 xen_reboot(SHUTDOWN_reboot
);
1311 static void xen_emergency_restart(void)
1313 xen_reboot(SHUTDOWN_reboot
);
1316 static void xen_machine_halt(void)
1318 xen_reboot(SHUTDOWN_poweroff
);
1321 static void xen_machine_power_off(void)
1325 xen_reboot(SHUTDOWN_poweroff
);
1328 static void xen_crash_shutdown(struct pt_regs
*regs
)
1330 xen_reboot(SHUTDOWN_crash
);
1334 xen_panic_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
1336 xen_reboot(SHUTDOWN_crash
);
1340 static struct notifier_block xen_panic_block
= {
1341 .notifier_call
= xen_panic_event
,
1344 int xen_panic_handler_init(void)
1346 atomic_notifier_chain_register(&panic_notifier_list
, &xen_panic_block
);
1350 static const struct machine_ops xen_machine_ops __initconst
= {
1351 .restart
= xen_restart
,
1352 .halt
= xen_machine_halt
,
1353 .power_off
= xen_machine_power_off
,
1354 .shutdown
= xen_machine_halt
,
1355 .crash_shutdown
= xen_crash_shutdown
,
1356 .emergency_restart
= xen_emergency_restart
,
1359 static void __init
xen_boot_params_init_edd(void)
1361 #if IS_ENABLED(CONFIG_EDD)
1362 struct xen_platform_op op
;
1363 struct edd_info
*edd_info
;
1368 edd_info
= boot_params
.eddbuf
;
1369 mbr_signature
= boot_params
.edd_mbr_sig_buffer
;
1371 op
.cmd
= XENPF_firmware_info
;
1373 op
.u
.firmware_info
.type
= XEN_FW_DISK_INFO
;
1374 for (nr
= 0; nr
< EDDMAXNR
; nr
++) {
1375 struct edd_info
*info
= edd_info
+ nr
;
1377 op
.u
.firmware_info
.index
= nr
;
1378 info
->params
.length
= sizeof(info
->params
);
1379 set_xen_guest_handle(op
.u
.firmware_info
.u
.disk_info
.edd_params
,
1381 ret
= HYPERVISOR_dom0_op(&op
);
1385 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1388 C(interface_support
);
1389 C(legacy_max_cylinder
);
1391 C(legacy_sectors_per_track
);
1394 boot_params
.eddbuf_entries
= nr
;
1396 op
.u
.firmware_info
.type
= XEN_FW_DISK_MBR_SIGNATURE
;
1397 for (nr
= 0; nr
< EDD_MBR_SIG_MAX
; nr
++) {
1398 op
.u
.firmware_info
.index
= nr
;
1399 ret
= HYPERVISOR_dom0_op(&op
);
1402 mbr_signature
[nr
] = op
.u
.firmware_info
.u
.disk_mbr_signature
.mbr_signature
;
1404 boot_params
.edd_mbr_sig_buf_entries
= nr
;
1409 * Set up the GDT and segment registers for -fstack-protector. Until
1410 * we do this, we have to be careful not to call any stack-protected
1411 * function, which is most of the kernel.
1413 static void __init
xen_setup_stackprotector(void)
1415 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry_boot
;
1416 pv_cpu_ops
.load_gdt
= xen_load_gdt_boot
;
1418 setup_stack_canary_segment(0);
1419 switch_to_new_gdt(0);
1421 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry
;
1422 pv_cpu_ops
.load_gdt
= xen_load_gdt
;
1425 static void __init
xen_pvh_early_guest_init(void)
1427 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1430 if (xen_feature(XENFEAT_hvm_callback_vector
))
1431 xen_have_vector_callback
= 1;
1433 #ifdef CONFIG_X86_32
1434 BUG(); /* PVH: Implement proper support. */
1438 /* First C function to be called on Xen boot */
1439 asmlinkage
void __init
xen_start_kernel(void)
1441 struct physdev_set_iopl set_iopl
;
1444 if (!xen_start_info
)
1447 xen_domain_type
= XEN_PV_DOMAIN
;
1449 xen_setup_features();
1450 xen_pvh_early_guest_init();
1451 xen_setup_machphys_mapping();
1453 /* Install Xen paravirt ops */
1455 pv_init_ops
= xen_init_ops
;
1456 pv_apic_ops
= xen_apic_ops
;
1457 if (!xen_pvh_domain())
1458 pv_cpu_ops
= xen_cpu_ops
;
1460 x86_init
.resources
.memory_setup
= xen_memory_setup
;
1461 x86_init
.oem
.arch_setup
= xen_arch_setup
;
1462 x86_init
.oem
.banner
= xen_banner
;
1464 xen_init_time_ops();
1467 * Set up some pagetable state before starting to set any ptes.
1472 /* Prevent unwanted bits from being set in PTEs. */
1473 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1475 if (!xen_initial_domain())
1477 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1479 __supported_pte_mask
|= _PAGE_IOMAP
;
1482 * Prevent page tables from being allocated in highmem, even
1483 * if CONFIG_HIGHPTE is enabled.
1485 __userpte_alloc_gfp
&= ~__GFP_HIGHMEM
;
1487 /* Work out if we support NX */
1491 xen_build_dynamic_phys_to_machine();
1494 * Set up kernel GDT and segment registers, mainly so that
1495 * -fstack-protector code can be executed.
1497 xen_setup_stackprotector();
1500 xen_init_cpuid_mask();
1502 #ifdef CONFIG_X86_LOCAL_APIC
1504 * set up the basic apic ops.
1506 set_xen_basic_apic_ops();
1509 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1510 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1511 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1514 machine_ops
= xen_machine_ops
;
1517 * The only reliable way to retain the initial address of the
1518 * percpu gdt_page is to remember it here, so we can go and
1519 * mark it RW later, when the initial percpu area is freed.
1521 xen_initial_gdt
= &per_cpu(gdt_page
, 0);
1525 #ifdef CONFIG_ACPI_NUMA
1527 * The pages we from Xen are not related to machine pages, so
1528 * any NUMA information the kernel tries to get from ACPI will
1529 * be meaningless. Prevent it from trying.
1533 #ifdef CONFIG_X86_PAT
1535 * For right now disable the PAT. We should remove this once
1536 * git commit 8eaffa67b43e99ae581622c5133e20b0f48bcef1
1537 * (xen/pat: Disable PAT support for now) is reverted.
1541 /* Don't do the full vcpu_info placement stuff until we have a
1542 possible map and a non-dummy shared_info. */
1543 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1545 local_irq_disable();
1546 early_boot_irqs_disabled
= true;
1548 xen_raw_console_write("mapping kernel into physical memory\n");
1549 xen_setup_kernel_pagetable((pgd_t
*)xen_start_info
->pt_base
, xen_start_info
->nr_pages
);
1551 /* Allocate and initialize top and mid mfn levels for p2m structure */
1552 xen_build_mfn_list_list();
1554 /* keep using Xen gdt for now; no urgent need to change it */
1556 #ifdef CONFIG_X86_32
1557 pv_info
.kernel_rpl
= 1;
1558 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1559 pv_info
.kernel_rpl
= 0;
1561 pv_info
.kernel_rpl
= 0;
1563 /* set the limit of our address space */
1566 /* PVH: runs at default kernel iopl of 0 */
1567 if (!xen_pvh_domain()) {
1569 * We used to do this in xen_arch_setup, but that is too late
1570 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1571 * early_amd_init which pokes 0xcf8 port.
1574 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
1576 xen_raw_printk("physdev_op failed %d\n", rc
);
1579 #ifdef CONFIG_X86_32
1580 /* set up basic CPUID stuff */
1581 cpu_detect(&new_cpu_data
);
1582 set_cpu_cap(&new_cpu_data
, X86_FEATURE_FPU
);
1583 new_cpu_data
.wp_works_ok
= 1;
1584 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1587 /* Poke various useful things into boot_params */
1588 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1589 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1590 ? __pa(xen_start_info
->mod_start
) : 0;
1591 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1592 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1594 if (!xen_initial_domain()) {
1595 add_preferred_console("xenboot", 0, NULL
);
1596 add_preferred_console("tty", 0, NULL
);
1597 add_preferred_console("hvc", 0, NULL
);
1599 x86_init
.pci
.arch_init
= pci_xen_init
;
1601 const struct dom0_vga_console_info
*info
=
1602 (void *)((char *)xen_start_info
+
1603 xen_start_info
->console
.dom0
.info_off
);
1604 struct xen_platform_op op
= {
1605 .cmd
= XENPF_firmware_info
,
1606 .interface_version
= XENPF_INTERFACE_VERSION
,
1607 .u
.firmware_info
.type
= XEN_FW_KBD_SHIFT_FLAGS
,
1610 xen_init_vga(info
, xen_start_info
->console
.dom0
.info_size
);
1611 xen_start_info
->console
.domU
.mfn
= 0;
1612 xen_start_info
->console
.domU
.evtchn
= 0;
1614 if (HYPERVISOR_dom0_op(&op
) == 0)
1615 boot_params
.kbd_status
= op
.u
.firmware_info
.u
.kbd_shift_flags
;
1619 /* Make sure ACS will be enabled */
1622 xen_acpi_sleep_register();
1624 /* Avoid searching for BIOS MP tables */
1625 x86_init
.mpparse
.find_smp_config
= x86_init_noop
;
1626 x86_init
.mpparse
.get_smp_config
= x86_init_uint_noop
;
1628 xen_boot_params_init_edd();
1631 /* PCI BIOS service won't work from a PV guest. */
1632 pci_probe
&= ~PCI_PROBE_BIOS
;
1634 xen_raw_console_write("about to get started...\n");
1636 xen_setup_runstate_info(0);
1638 /* Start the world */
1639 #ifdef CONFIG_X86_32
1640 i386_start_kernel();
1642 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));
1646 void __ref
xen_hvm_init_shared_info(void)
1649 struct xen_add_to_physmap xatp
;
1650 static struct shared_info
*shared_info_page
= 0;
1652 if (!shared_info_page
)
1653 shared_info_page
= (struct shared_info
*)
1654 extend_brk(PAGE_SIZE
, PAGE_SIZE
);
1655 xatp
.domid
= DOMID_SELF
;
1657 xatp
.space
= XENMAPSPACE_shared_info
;
1658 xatp
.gpfn
= __pa(shared_info_page
) >> PAGE_SHIFT
;
1659 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap
, &xatp
))
1662 HYPERVISOR_shared_info
= (struct shared_info
*)shared_info_page
;
1664 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1665 * page, we use it in the event channel upcall and in some pvclock
1666 * related functions. We don't need the vcpu_info placement
1667 * optimizations because we don't use any pv_mmu or pv_irq op on
1669 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1670 * online but xen_hvm_init_shared_info is run at resume time too and
1671 * in that case multiple vcpus might be online. */
1672 for_each_online_cpu(cpu
) {
1673 /* Leave it to be NULL. */
1674 if (cpu
>= MAX_VIRT_CPUS
)
1676 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
1680 #ifdef CONFIG_XEN_PVHVM
1681 static void __init
init_hvm_pv_info(void)
1684 uint32_t eax
, ebx
, ecx
, edx
, pages
, msr
, base
;
1687 base
= xen_cpuid_base();
1688 cpuid(base
+ 1, &eax
, &ebx
, &ecx
, &edx
);
1691 minor
= eax
& 0xffff;
1692 printk(KERN_INFO
"Xen version %d.%d.\n", major
, minor
);
1694 cpuid(base
+ 2, &pages
, &msr
, &ecx
, &edx
);
1696 pfn
= __pa(hypercall_page
);
1697 wrmsr_safe(msr
, (u32
)pfn
, (u32
)(pfn
>> 32));
1699 xen_setup_features();
1701 pv_info
.name
= "Xen HVM";
1703 xen_domain_type
= XEN_HVM_DOMAIN
;
1706 static int xen_hvm_cpu_notify(struct notifier_block
*self
, unsigned long action
,
1709 int cpu
= (long)hcpu
;
1711 case CPU_UP_PREPARE
:
1712 xen_vcpu_setup(cpu
);
1713 if (xen_have_vector_callback
) {
1714 if (xen_feature(XENFEAT_hvm_safe_pvclock
))
1715 xen_setup_timer(cpu
);
1724 static struct notifier_block xen_hvm_cpu_notifier
= {
1725 .notifier_call
= xen_hvm_cpu_notify
,
1728 static void __init
xen_hvm_guest_init(void)
1732 xen_hvm_init_shared_info();
1734 xen_panic_handler_init();
1736 if (xen_feature(XENFEAT_hvm_callback_vector
))
1737 xen_have_vector_callback
= 1;
1739 register_cpu_notifier(&xen_hvm_cpu_notifier
);
1740 xen_unplug_emulated_devices();
1741 x86_init
.irqs
.intr_init
= xen_init_IRQ
;
1742 xen_hvm_init_time_ops();
1743 xen_hvm_init_mmu_ops();
1746 static uint32_t __init
xen_hvm_platform(void)
1748 if (xen_pv_domain())
1751 return xen_cpuid_base();
1754 bool xen_hvm_need_lapic(void)
1756 if (xen_pv_domain())
1758 if (!xen_hvm_domain())
1760 if (xen_feature(XENFEAT_hvm_pirqs
) && xen_have_vector_callback
)
1764 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic
);
1766 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst
= {
1768 .detect
= xen_hvm_platform
,
1769 .init_platform
= xen_hvm_guest_init
,
1770 .x2apic_available
= xen_x2apic_para_available
,
1772 EXPORT_SYMBOL(x86_hyper_xen_hvm
);