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/nmi.h>
44 #include <xen/interface/xen-mca.h>
45 #include <xen/features.h>
48 #include <xen/hvc-console.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/mach_traps.h>
71 #include <asm/mwait.h>
72 #include <asm/pci_x86.h>
76 #include <linux/acpi.h>
78 #include <acpi/pdc_intel.h>
79 #include <acpi/processor.h>
80 #include <xen/interface/platform.h>
86 #include "multicalls.h"
88 EXPORT_SYMBOL_GPL(hypercall_page
);
91 * Pointer to the xen_vcpu_info structure or
92 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
93 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
94 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
95 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
96 * acknowledge pending events.
97 * Also more subtly it is used by the patched version of irq enable/disable
98 * e.g. xen_irq_enable_direct and xen_iret in PV mode.
100 * The desire to be able to do those mask/unmask operations as a single
101 * instruction by using the per-cpu offset held in %gs is the real reason
102 * vcpu info is in a per-cpu pointer and the original reason for this
106 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
109 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
110 * hypercall. This can be used both in PV and PVHVM mode. The structure
111 * overrides the default per_cpu(xen_vcpu, cpu) value.
113 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
115 enum xen_domain_type xen_domain_type
= XEN_NATIVE
;
116 EXPORT_SYMBOL_GPL(xen_domain_type
);
118 unsigned long *machine_to_phys_mapping
= (void *)MACH2PHYS_VIRT_START
;
119 EXPORT_SYMBOL(machine_to_phys_mapping
);
120 unsigned long machine_to_phys_nr
;
121 EXPORT_SYMBOL(machine_to_phys_nr
);
123 struct start_info
*xen_start_info
;
124 EXPORT_SYMBOL_GPL(xen_start_info
);
126 struct shared_info xen_dummy_shared_info
;
128 void *xen_initial_gdt
;
130 RESERVE_BRK(shared_info_page_brk
, PAGE_SIZE
);
131 __read_mostly
int xen_have_vector_callback
;
132 EXPORT_SYMBOL_GPL(xen_have_vector_callback
);
135 * Point at some empty memory to start with. We map the real shared_info
136 * page as soon as fixmap is up and running.
138 struct shared_info
*HYPERVISOR_shared_info
= &xen_dummy_shared_info
;
141 * Flag to determine whether vcpu info placement is available on all
142 * VCPUs. We assume it is to start with, and then set it to zero on
143 * the first failure. This is because it can succeed on some VCPUs
144 * and not others, since it can involve hypervisor memory allocation,
145 * or because the guest failed to guarantee all the appropriate
146 * constraints on all VCPUs (ie buffer can't cross a page boundary).
148 * Note that any particular CPU may be using a placed vcpu structure,
149 * but we can only optimise if the all are.
151 * 0: not available, 1: available
153 static int have_vcpu_info_placement
= 1;
156 struct desc_struct desc
[3];
160 * Updating the 3 TLS descriptors in the GDT on every task switch is
161 * surprisingly expensive so we avoid updating them if they haven't
162 * changed. Since Xen writes different descriptors than the one
163 * passed in the update_descriptor hypercall we keep shadow copies to
166 static DEFINE_PER_CPU(struct tls_descs
, shadow_tls_desc
);
168 static void clamp_max_cpus(void)
171 if (setup_max_cpus
> MAX_VIRT_CPUS
)
172 setup_max_cpus
= MAX_VIRT_CPUS
;
176 static void xen_vcpu_setup(int cpu
)
178 struct vcpu_register_vcpu_info info
;
180 struct vcpu_info
*vcpup
;
182 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
185 * This path is called twice on PVHVM - first during bootup via
186 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
187 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
188 * As we can only do the VCPUOP_register_vcpu_info once lets
189 * not over-write its result.
191 * For PV it is called during restore (xen_vcpu_restore) and bootup
192 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
195 if (xen_hvm_domain()) {
196 if (per_cpu(xen_vcpu
, cpu
) == &per_cpu(xen_vcpu_info
, cpu
))
199 if (cpu
< MAX_VIRT_CPUS
)
200 per_cpu(xen_vcpu
,cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
202 if (!have_vcpu_info_placement
) {
203 if (cpu
>= MAX_VIRT_CPUS
)
208 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
209 info
.mfn
= arbitrary_virt_to_mfn(vcpup
);
210 info
.offset
= offset_in_page(vcpup
);
212 /* Check to see if the hypervisor will put the vcpu_info
213 structure where we want it, which allows direct access via
215 N.B. This hypercall can _only_ be called once per CPU. Subsequent
216 calls will error out with -EINVAL. This is due to the fact that
217 hypervisor has no unregister variant and this hypercall does not
218 allow to over-write info.mfn and info.offset.
220 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
223 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
224 have_vcpu_info_placement
= 0;
227 /* This cpu is using the registered vcpu info, even if
228 later ones fail to. */
229 per_cpu(xen_vcpu
, cpu
) = vcpup
;
234 * On restore, set the vcpu placement up again.
235 * If it fails, then we're in a bad state, since
236 * we can't back out from using it...
238 void xen_vcpu_restore(void)
242 for_each_possible_cpu(cpu
) {
243 bool other_cpu
= (cpu
!= smp_processor_id());
244 bool is_up
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, cpu
, NULL
);
246 if (other_cpu
&& is_up
&&
247 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
250 xen_setup_runstate_info(cpu
);
252 if (have_vcpu_info_placement
)
255 if (other_cpu
&& is_up
&&
256 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
261 static void __init
xen_banner(void)
263 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
264 struct xen_extraversion extra
;
265 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
267 pr_info("Booting paravirtualized kernel %son %s\n",
268 xen_feature(XENFEAT_auto_translated_physmap
) ?
269 "with PVH extensions " : "", pv_info
.name
);
270 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
271 version
>> 16, version
& 0xffff, extra
.extraversion
,
272 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
274 /* Check if running on Xen version (major, minor) or later */
276 xen_running_on_version_or_later(unsigned int major
, unsigned int minor
)
278 unsigned int version
;
283 version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
284 if ((((version
>> 16) == major
) && ((version
& 0xffff) >= minor
)) ||
285 ((version
>> 16) > major
))
290 #define CPUID_THERM_POWER_LEAF 6
291 #define APERFMPERF_PRESENT 0
293 static __read_mostly
unsigned int cpuid_leaf1_edx_mask
= ~0;
294 static __read_mostly
unsigned int cpuid_leaf1_ecx_mask
= ~0;
296 static __read_mostly
unsigned int cpuid_leaf1_ecx_set_mask
;
297 static __read_mostly
unsigned int cpuid_leaf5_ecx_val
;
298 static __read_mostly
unsigned int cpuid_leaf5_edx_val
;
300 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
301 unsigned int *cx
, unsigned int *dx
)
303 unsigned maskebx
= ~0;
304 unsigned maskecx
= ~0;
305 unsigned maskedx
= ~0;
308 * Mask out inconvenient features, to try and disable as many
309 * unsupported kernel subsystems as possible.
313 maskecx
= cpuid_leaf1_ecx_mask
;
314 setecx
= cpuid_leaf1_ecx_set_mask
;
315 maskedx
= cpuid_leaf1_edx_mask
;
318 case CPUID_MWAIT_LEAF
:
319 /* Synthesize the values.. */
322 *cx
= cpuid_leaf5_ecx_val
;
323 *dx
= cpuid_leaf5_edx_val
;
326 case CPUID_THERM_POWER_LEAF
:
327 /* Disabling APERFMPERF for kernel usage */
328 maskecx
= ~(1 << APERFMPERF_PRESENT
);
332 /* Suppress extended topology stuff */
337 asm(XEN_EMULATE_PREFIX
"cpuid"
342 : "0" (*ax
), "2" (*cx
));
351 static bool __init
xen_check_mwait(void)
354 struct xen_platform_op op
= {
355 .cmd
= XENPF_set_processor_pminfo
,
356 .u
.set_pminfo
.id
= -1,
357 .u
.set_pminfo
.type
= XEN_PM_PDC
,
360 unsigned int ax
, bx
, cx
, dx
;
361 unsigned int mwait_mask
;
363 /* We need to determine whether it is OK to expose the MWAIT
364 * capability to the kernel to harvest deeper than C3 states from ACPI
365 * _CST using the processor_harvest_xen.c module. For this to work, we
366 * need to gather the MWAIT_LEAF values (which the cstate.c code
367 * checks against). The hypervisor won't expose the MWAIT flag because
368 * it would break backwards compatibility; so we will find out directly
369 * from the hardware and hypercall.
371 if (!xen_initial_domain())
375 * When running under platform earlier than Xen4.2, do not expose
376 * mwait, to avoid the risk of loading native acpi pad driver
378 if (!xen_running_on_version_or_later(4, 2))
384 native_cpuid(&ax
, &bx
, &cx
, &dx
);
386 mwait_mask
= (1 << (X86_FEATURE_EST
% 32)) |
387 (1 << (X86_FEATURE_MWAIT
% 32));
389 if ((cx
& mwait_mask
) != mwait_mask
)
392 /* We need to emulate the MWAIT_LEAF and for that we need both
393 * ecx and edx. The hypercall provides only partial information.
396 ax
= CPUID_MWAIT_LEAF
;
401 native_cpuid(&ax
, &bx
, &cx
, &dx
);
403 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
404 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
406 buf
[0] = ACPI_PDC_REVISION_ID
;
408 buf
[2] = (ACPI_PDC_C_CAPABILITY_SMP
| ACPI_PDC_EST_CAPABILITY_SWSMP
);
410 set_xen_guest_handle(op
.u
.set_pminfo
.pdc
, buf
);
412 if ((HYPERVISOR_dom0_op(&op
) == 0) &&
413 (buf
[2] & (ACPI_PDC_C_C1_FFH
| ACPI_PDC_C_C2C3_FFH
))) {
414 cpuid_leaf5_ecx_val
= cx
;
415 cpuid_leaf5_edx_val
= dx
;
422 static void __init
xen_init_cpuid_mask(void)
424 unsigned int ax
, bx
, cx
, dx
;
425 unsigned int xsave_mask
;
427 cpuid_leaf1_edx_mask
=
428 ~((1 << X86_FEATURE_MTRR
) | /* disable MTRR */
429 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
431 if (!xen_initial_domain())
432 cpuid_leaf1_edx_mask
&=
433 ~((1 << X86_FEATURE_ACPI
)); /* disable ACPI */
435 cpuid_leaf1_ecx_mask
&= ~(1 << (X86_FEATURE_X2APIC
% 32));
439 cpuid(1, &ax
, &bx
, &cx
, &dx
);
442 (1 << (X86_FEATURE_XSAVE
% 32)) |
443 (1 << (X86_FEATURE_OSXSAVE
% 32));
445 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
446 if ((cx
& xsave_mask
) != xsave_mask
)
447 cpuid_leaf1_ecx_mask
&= ~xsave_mask
; /* disable XSAVE & OSXSAVE */
448 if (xen_check_mwait())
449 cpuid_leaf1_ecx_set_mask
= (1 << (X86_FEATURE_MWAIT
% 32));
452 static void xen_set_debugreg(int reg
, unsigned long val
)
454 HYPERVISOR_set_debugreg(reg
, val
);
457 static unsigned long xen_get_debugreg(int reg
)
459 return HYPERVISOR_get_debugreg(reg
);
462 static void xen_end_context_switch(struct task_struct
*next
)
465 paravirt_end_context_switch(next
);
468 static unsigned long xen_store_tr(void)
474 * Set the page permissions for a particular virtual address. If the
475 * address is a vmalloc mapping (or other non-linear mapping), then
476 * find the linear mapping of the page and also set its protections to
479 static void set_aliased_prot(void *v
, pgprot_t prot
)
487 ptep
= lookup_address((unsigned long)v
, &level
);
488 BUG_ON(ptep
== NULL
);
490 pfn
= pte_pfn(*ptep
);
491 page
= pfn_to_page(pfn
);
493 pte
= pfn_pte(pfn
, prot
);
495 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
498 if (!PageHighMem(page
)) {
499 void *av
= __va(PFN_PHYS(pfn
));
502 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
508 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
510 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
513 for(i
= 0; i
< entries
; i
+= entries_per_page
)
514 set_aliased_prot(ldt
+ i
, PAGE_KERNEL_RO
);
517 static void xen_free_ldt(struct desc_struct
*ldt
, unsigned entries
)
519 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
522 for(i
= 0; i
< entries
; i
+= entries_per_page
)
523 set_aliased_prot(ldt
+ i
, PAGE_KERNEL
);
526 static void xen_set_ldt(const void *addr
, unsigned entries
)
528 struct mmuext_op
*op
;
529 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
531 trace_xen_cpu_set_ldt(addr
, entries
);
534 op
->cmd
= MMUEXT_SET_LDT
;
535 op
->arg1
.linear_addr
= (unsigned long)addr
;
536 op
->arg2
.nr_ents
= entries
;
538 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
540 xen_mc_issue(PARAVIRT_LAZY_CPU
);
543 static void xen_load_gdt(const struct desc_ptr
*dtr
)
545 unsigned long va
= dtr
->address
;
546 unsigned int size
= dtr
->size
+ 1;
547 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
548 unsigned long frames
[pages
];
552 * A GDT can be up to 64k in size, which corresponds to 8192
553 * 8-byte entries, or 16 4k pages..
556 BUG_ON(size
> 65536);
557 BUG_ON(va
& ~PAGE_MASK
);
559 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
562 unsigned long pfn
, mfn
;
566 * The GDT is per-cpu and is in the percpu data area.
567 * That can be virtually mapped, so we need to do a
568 * page-walk to get the underlying MFN for the
569 * hypercall. The page can also be in the kernel's
570 * linear range, so we need to RO that mapping too.
572 ptep
= lookup_address(va
, &level
);
573 BUG_ON(ptep
== NULL
);
575 pfn
= pte_pfn(*ptep
);
576 mfn
= pfn_to_mfn(pfn
);
577 virt
= __va(PFN_PHYS(pfn
));
581 make_lowmem_page_readonly((void *)va
);
582 make_lowmem_page_readonly(virt
);
585 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
590 * load_gdt for early boot, when the gdt is only mapped once
592 static void __init
xen_load_gdt_boot(const struct desc_ptr
*dtr
)
594 unsigned long va
= dtr
->address
;
595 unsigned int size
= dtr
->size
+ 1;
596 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
597 unsigned long frames
[pages
];
601 * A GDT can be up to 64k in size, which corresponds to 8192
602 * 8-byte entries, or 16 4k pages..
605 BUG_ON(size
> 65536);
606 BUG_ON(va
& ~PAGE_MASK
);
608 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
610 unsigned long pfn
, mfn
;
612 pfn
= virt_to_pfn(va
);
613 mfn
= pfn_to_mfn(pfn
);
615 pte
= pfn_pte(pfn
, PAGE_KERNEL_RO
);
617 if (HYPERVISOR_update_va_mapping((unsigned long)va
, pte
, 0))
623 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
627 static inline bool desc_equal(const struct desc_struct
*d1
,
628 const struct desc_struct
*d2
)
630 return d1
->a
== d2
->a
&& d1
->b
== d2
->b
;
633 static void load_TLS_descriptor(struct thread_struct
*t
,
634 unsigned int cpu
, unsigned int i
)
636 struct desc_struct
*shadow
= &per_cpu(shadow_tls_desc
, cpu
).desc
[i
];
637 struct desc_struct
*gdt
;
639 struct multicall_space mc
;
641 if (desc_equal(shadow
, &t
->tls_array
[i
]))
644 *shadow
= t
->tls_array
[i
];
646 gdt
= get_cpu_gdt_table(cpu
);
647 maddr
= arbitrary_virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
648 mc
= __xen_mc_entry(0);
650 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
653 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
656 * XXX sleazy hack: If we're being called in a lazy-cpu zone
657 * and lazy gs handling is enabled, it means we're in a
658 * context switch, and %gs has just been saved. This means we
659 * can zero it out to prevent faults on exit from the
660 * hypervisor if the next process has no %gs. Either way, it
661 * has been saved, and the new value will get loaded properly.
662 * This will go away as soon as Xen has been modified to not
663 * save/restore %gs for normal hypercalls.
665 * On x86_64, this hack is not used for %gs, because gs points
666 * to KERNEL_GS_BASE (and uses it for PDA references), so we
667 * must not zero %gs on x86_64
669 * For x86_64, we need to zero %fs, otherwise we may get an
670 * exception between the new %fs descriptor being loaded and
671 * %fs being effectively cleared at __switch_to().
673 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
683 load_TLS_descriptor(t
, cpu
, 0);
684 load_TLS_descriptor(t
, cpu
, 1);
685 load_TLS_descriptor(t
, cpu
, 2);
687 xen_mc_issue(PARAVIRT_LAZY_CPU
);
691 static void xen_load_gs_index(unsigned int idx
)
693 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
698 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
701 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
702 u64 entry
= *(u64
*)ptr
;
704 trace_xen_cpu_write_ldt_entry(dt
, entrynum
, entry
);
709 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
715 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
716 struct trap_info
*info
)
720 if (val
->type
!= GATE_TRAP
&& val
->type
!= GATE_INTERRUPT
)
723 info
->vector
= vector
;
725 addr
= gate_offset(*val
);
728 * Look for known traps using IST, and substitute them
729 * appropriately. The debugger ones are the only ones we care
730 * about. Xen will handle faults like double_fault,
731 * so we should never see them. Warn if
732 * there's an unexpected IST-using fault handler.
734 if (addr
== (unsigned long)debug
)
735 addr
= (unsigned long)xen_debug
;
736 else if (addr
== (unsigned long)int3
)
737 addr
= (unsigned long)xen_int3
;
738 else if (addr
== (unsigned long)stack_segment
)
739 addr
= (unsigned long)xen_stack_segment
;
740 else if (addr
== (unsigned long)double_fault
) {
741 /* Don't need to handle these */
743 #ifdef CONFIG_X86_MCE
744 } else if (addr
== (unsigned long)machine_check
) {
746 * when xen hypervisor inject vMCE to guest,
747 * use native mce handler to handle it
751 } else if (addr
== (unsigned long)nmi
)
753 * Use the native version as well.
757 /* Some other trap using IST? */
758 if (WARN_ON(val
->ist
!= 0))
761 #endif /* CONFIG_X86_64 */
762 info
->address
= addr
;
764 info
->cs
= gate_segment(*val
);
765 info
->flags
= val
->dpl
;
766 /* interrupt gates clear IF */
767 if (val
->type
== GATE_INTERRUPT
)
768 info
->flags
|= 1 << 2;
773 /* Locations of each CPU's IDT */
774 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
776 /* Set an IDT entry. If the entry is part of the current IDT, then
778 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
780 unsigned long p
= (unsigned long)&dt
[entrynum
];
781 unsigned long start
, end
;
783 trace_xen_cpu_write_idt_entry(dt
, entrynum
, g
);
787 start
= __this_cpu_read(idt_desc
.address
);
788 end
= start
+ __this_cpu_read(idt_desc
.size
) + 1;
792 native_write_idt_entry(dt
, entrynum
, g
);
794 if (p
>= start
&& (p
+ 8) <= end
) {
795 struct trap_info info
[2];
799 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
800 if (HYPERVISOR_set_trap_table(info
))
807 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
808 struct trap_info
*traps
)
810 unsigned in
, out
, count
;
812 count
= (desc
->size
+1) / sizeof(gate_desc
);
815 for (in
= out
= 0; in
< count
; in
++) {
816 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
818 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
821 traps
[out
].address
= 0;
824 void xen_copy_trap_info(struct trap_info
*traps
)
826 const struct desc_ptr
*desc
= this_cpu_ptr(&idt_desc
);
828 xen_convert_trap_info(desc
, traps
);
831 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
832 hold a spinlock to protect the static traps[] array (static because
833 it avoids allocation, and saves stack space). */
834 static void xen_load_idt(const struct desc_ptr
*desc
)
836 static DEFINE_SPINLOCK(lock
);
837 static struct trap_info traps
[257];
839 trace_xen_cpu_load_idt(desc
);
843 memcpy(this_cpu_ptr(&idt_desc
), desc
, sizeof(idt_desc
));
845 xen_convert_trap_info(desc
, traps
);
848 if (HYPERVISOR_set_trap_table(traps
))
854 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
855 they're handled differently. */
856 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
857 const void *desc
, int type
)
859 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
870 xmaddr_t maddr
= arbitrary_virt_to_machine(&dt
[entry
]);
873 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
883 * Version of write_gdt_entry for use at early boot-time needed to
884 * update an entry as simply as possible.
886 static void __init
xen_write_gdt_entry_boot(struct desc_struct
*dt
, int entry
,
887 const void *desc
, int type
)
889 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
898 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
900 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
901 dt
[entry
] = *(struct desc_struct
*)desc
;
907 static void xen_load_sp0(struct tss_struct
*tss
,
908 struct thread_struct
*thread
)
910 struct multicall_space mcs
;
912 mcs
= xen_mc_entry(0);
913 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
914 xen_mc_issue(PARAVIRT_LAZY_CPU
);
917 static void xen_set_iopl_mask(unsigned mask
)
919 struct physdev_set_iopl set_iopl
;
921 /* Force the change at ring 0. */
922 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
923 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
926 static void xen_io_delay(void)
930 #ifdef CONFIG_X86_LOCAL_APIC
931 static unsigned long xen_set_apic_id(unsigned int x
)
936 static unsigned int xen_get_apic_id(unsigned long x
)
938 return ((x
)>>24) & 0xFFu
;
940 static u32
xen_apic_read(u32 reg
)
942 struct xen_platform_op op
= {
943 .cmd
= XENPF_get_cpuinfo
,
944 .interface_version
= XENPF_INTERFACE_VERSION
,
945 .u
.pcpu_info
.xen_cpuid
= 0,
949 /* Shouldn't need this as APIC is turned off for PV, and we only
950 * get called on the bootup processor. But just in case. */
951 if (!xen_initial_domain() || smp_processor_id())
960 ret
= HYPERVISOR_dom0_op(&op
);
964 return op
.u
.pcpu_info
.apic_id
<< 24;
967 static void xen_apic_write(u32 reg
, u32 val
)
969 /* Warn to see if there's any stray references */
973 static u64
xen_apic_icr_read(void)
978 static void xen_apic_icr_write(u32 low
, u32 id
)
980 /* Warn to see if there's any stray references */
984 static void xen_apic_wait_icr_idle(void)
989 static u32
xen_safe_apic_wait_icr_idle(void)
994 static void set_xen_basic_apic_ops(void)
996 apic
->read
= xen_apic_read
;
997 apic
->write
= xen_apic_write
;
998 apic
->icr_read
= xen_apic_icr_read
;
999 apic
->icr_write
= xen_apic_icr_write
;
1000 apic
->wait_icr_idle
= xen_apic_wait_icr_idle
;
1001 apic
->safe_wait_icr_idle
= xen_safe_apic_wait_icr_idle
;
1002 apic
->set_apic_id
= xen_set_apic_id
;
1003 apic
->get_apic_id
= xen_get_apic_id
;
1006 apic
->send_IPI_allbutself
= xen_send_IPI_allbutself
;
1007 apic
->send_IPI_mask_allbutself
= xen_send_IPI_mask_allbutself
;
1008 apic
->send_IPI_mask
= xen_send_IPI_mask
;
1009 apic
->send_IPI_all
= xen_send_IPI_all
;
1010 apic
->send_IPI_self
= xen_send_IPI_self
;
1016 static void xen_clts(void)
1018 struct multicall_space mcs
;
1020 mcs
= xen_mc_entry(0);
1022 MULTI_fpu_taskswitch(mcs
.mc
, 0);
1024 xen_mc_issue(PARAVIRT_LAZY_CPU
);
1027 static DEFINE_PER_CPU(unsigned long, xen_cr0_value
);
1029 static unsigned long xen_read_cr0(void)
1031 unsigned long cr0
= this_cpu_read(xen_cr0_value
);
1033 if (unlikely(cr0
== 0)) {
1034 cr0
= native_read_cr0();
1035 this_cpu_write(xen_cr0_value
, cr0
);
1041 static void xen_write_cr0(unsigned long cr0
)
1043 struct multicall_space mcs
;
1045 this_cpu_write(xen_cr0_value
, cr0
);
1047 /* Only pay attention to cr0.TS; everything else is
1049 mcs
= xen_mc_entry(0);
1051 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
1053 xen_mc_issue(PARAVIRT_LAZY_CPU
);
1056 static void xen_write_cr4(unsigned long cr4
)
1058 cr4
&= ~X86_CR4_PGE
;
1059 cr4
&= ~X86_CR4_PSE
;
1061 native_write_cr4(cr4
);
1063 #ifdef CONFIG_X86_64
1064 static inline unsigned long xen_read_cr8(void)
1068 static inline void xen_write_cr8(unsigned long val
)
1073 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
1080 #ifdef CONFIG_X86_64
1084 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
1085 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
1086 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
1089 base
= ((u64
)high
<< 32) | low
;
1090 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
1098 case MSR_SYSCALL_MASK
:
1099 case MSR_IA32_SYSENTER_CS
:
1100 case MSR_IA32_SYSENTER_ESP
:
1101 case MSR_IA32_SYSENTER_EIP
:
1102 /* Fast syscall setup is all done in hypercalls, so
1103 these are all ignored. Stub them out here to stop
1104 Xen console noise. */
1107 ret
= native_write_msr_safe(msr
, low
, high
);
1113 void xen_setup_shared_info(void)
1115 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1116 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
1117 xen_start_info
->shared_info
);
1119 HYPERVISOR_shared_info
=
1120 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
1122 HYPERVISOR_shared_info
=
1123 (struct shared_info
*)__va(xen_start_info
->shared_info
);
1126 /* In UP this is as good a place as any to set up shared info */
1127 xen_setup_vcpu_info_placement();
1130 xen_setup_mfn_list_list();
1133 /* This is called once we have the cpu_possible_mask */
1134 void xen_setup_vcpu_info_placement(void)
1138 for_each_possible_cpu(cpu
)
1139 xen_vcpu_setup(cpu
);
1141 /* xen_vcpu_setup managed to place the vcpu_info within the
1142 * percpu area for all cpus, so make use of it. Note that for
1143 * PVH we want to use native IRQ mechanism. */
1144 if (have_vcpu_info_placement
&& !xen_pvh_domain()) {
1145 pv_irq_ops
.save_fl
= __PV_IS_CALLEE_SAVE(xen_save_fl_direct
);
1146 pv_irq_ops
.restore_fl
= __PV_IS_CALLEE_SAVE(xen_restore_fl_direct
);
1147 pv_irq_ops
.irq_disable
= __PV_IS_CALLEE_SAVE(xen_irq_disable_direct
);
1148 pv_irq_ops
.irq_enable
= __PV_IS_CALLEE_SAVE(xen_irq_enable_direct
);
1149 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1153 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1154 unsigned long addr
, unsigned len
)
1156 char *start
, *end
, *reloc
;
1159 start
= end
= reloc
= NULL
;
1161 #define SITE(op, x) \
1162 case PARAVIRT_PATCH(op.x): \
1163 if (have_vcpu_info_placement) { \
1164 start = (char *)xen_##x##_direct; \
1165 end = xen_##x##_direct_end; \
1166 reloc = xen_##x##_direct_reloc; \
1171 SITE(pv_irq_ops
, irq_enable
);
1172 SITE(pv_irq_ops
, irq_disable
);
1173 SITE(pv_irq_ops
, save_fl
);
1174 SITE(pv_irq_ops
, restore_fl
);
1178 if (start
== NULL
|| (end
-start
) > len
)
1181 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1183 /* Note: because reloc is assigned from something that
1184 appears to be an array, gcc assumes it's non-null,
1185 but doesn't know its relationship with start and
1187 if (reloc
> start
&& reloc
< end
) {
1188 int reloc_off
= reloc
- start
;
1189 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1190 long delta
= start
- (char *)addr
;
1198 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1206 static const struct pv_info xen_info __initconst
= {
1207 .paravirt_enabled
= 1,
1208 .shared_kernel_pmd
= 0,
1210 #ifdef CONFIG_X86_64
1211 .extra_user_64bit_cs
= FLAT_USER_CS64
,
1217 static const struct pv_init_ops xen_init_ops __initconst
= {
1221 static const struct pv_cpu_ops xen_cpu_ops __initconst
= {
1224 .set_debugreg
= xen_set_debugreg
,
1225 .get_debugreg
= xen_get_debugreg
,
1229 .read_cr0
= xen_read_cr0
,
1230 .write_cr0
= xen_write_cr0
,
1232 .read_cr4
= native_read_cr4
,
1233 .read_cr4_safe
= native_read_cr4_safe
,
1234 .write_cr4
= xen_write_cr4
,
1236 #ifdef CONFIG_X86_64
1237 .read_cr8
= xen_read_cr8
,
1238 .write_cr8
= xen_write_cr8
,
1241 .wbinvd
= native_wbinvd
,
1243 .read_msr
= native_read_msr_safe
,
1244 .write_msr
= xen_write_msr_safe
,
1246 .read_tsc
= native_read_tsc
,
1247 .read_pmc
= native_read_pmc
,
1249 .read_tscp
= native_read_tscp
,
1252 .irq_enable_sysexit
= xen_sysexit
,
1253 #ifdef CONFIG_X86_64
1254 .usergs_sysret32
= xen_sysret32
,
1255 .usergs_sysret64
= xen_sysret64
,
1258 .load_tr_desc
= paravirt_nop
,
1259 .set_ldt
= xen_set_ldt
,
1260 .load_gdt
= xen_load_gdt
,
1261 .load_idt
= xen_load_idt
,
1262 .load_tls
= xen_load_tls
,
1263 #ifdef CONFIG_X86_64
1264 .load_gs_index
= xen_load_gs_index
,
1267 .alloc_ldt
= xen_alloc_ldt
,
1268 .free_ldt
= xen_free_ldt
,
1270 .store_idt
= native_store_idt
,
1271 .store_tr
= xen_store_tr
,
1273 .write_ldt_entry
= xen_write_ldt_entry
,
1274 .write_gdt_entry
= xen_write_gdt_entry
,
1275 .write_idt_entry
= xen_write_idt_entry
,
1276 .load_sp0
= xen_load_sp0
,
1278 .set_iopl_mask
= xen_set_iopl_mask
,
1279 .io_delay
= xen_io_delay
,
1281 /* Xen takes care of %gs when switching to usermode for us */
1282 .swapgs
= paravirt_nop
,
1284 .start_context_switch
= paravirt_start_context_switch
,
1285 .end_context_switch
= xen_end_context_switch
,
1288 static const struct pv_apic_ops xen_apic_ops __initconst
= {
1289 #ifdef CONFIG_X86_LOCAL_APIC
1290 .startup_ipi_hook
= paravirt_nop
,
1294 static void xen_reboot(int reason
)
1296 struct sched_shutdown r
= { .reason
= reason
};
1298 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1302 static void xen_restart(char *msg
)
1304 xen_reboot(SHUTDOWN_reboot
);
1307 static void xen_emergency_restart(void)
1309 xen_reboot(SHUTDOWN_reboot
);
1312 static void xen_machine_halt(void)
1314 xen_reboot(SHUTDOWN_poweroff
);
1317 static void xen_machine_power_off(void)
1321 xen_reboot(SHUTDOWN_poweroff
);
1324 static void xen_crash_shutdown(struct pt_regs
*regs
)
1326 xen_reboot(SHUTDOWN_crash
);
1330 xen_panic_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
1332 xen_reboot(SHUTDOWN_crash
);
1336 static struct notifier_block xen_panic_block
= {
1337 .notifier_call
= xen_panic_event
,
1341 int xen_panic_handler_init(void)
1343 atomic_notifier_chain_register(&panic_notifier_list
, &xen_panic_block
);
1347 static const struct machine_ops xen_machine_ops __initconst
= {
1348 .restart
= xen_restart
,
1349 .halt
= xen_machine_halt
,
1350 .power_off
= xen_machine_power_off
,
1351 .shutdown
= xen_machine_halt
,
1352 .crash_shutdown
= xen_crash_shutdown
,
1353 .emergency_restart
= xen_emergency_restart
,
1356 static unsigned char xen_get_nmi_reason(void)
1358 unsigned char reason
= 0;
1360 /* Construct a value which looks like it came from port 0x61. */
1361 if (test_bit(_XEN_NMIREASON_io_error
,
1362 &HYPERVISOR_shared_info
->arch
.nmi_reason
))
1363 reason
|= NMI_REASON_IOCHK
;
1364 if (test_bit(_XEN_NMIREASON_pci_serr
,
1365 &HYPERVISOR_shared_info
->arch
.nmi_reason
))
1366 reason
|= NMI_REASON_SERR
;
1371 static void __init
xen_boot_params_init_edd(void)
1373 #if IS_ENABLED(CONFIG_EDD)
1374 struct xen_platform_op op
;
1375 struct edd_info
*edd_info
;
1380 edd_info
= boot_params
.eddbuf
;
1381 mbr_signature
= boot_params
.edd_mbr_sig_buffer
;
1383 op
.cmd
= XENPF_firmware_info
;
1385 op
.u
.firmware_info
.type
= XEN_FW_DISK_INFO
;
1386 for (nr
= 0; nr
< EDDMAXNR
; nr
++) {
1387 struct edd_info
*info
= edd_info
+ nr
;
1389 op
.u
.firmware_info
.index
= nr
;
1390 info
->params
.length
= sizeof(info
->params
);
1391 set_xen_guest_handle(op
.u
.firmware_info
.u
.disk_info
.edd_params
,
1393 ret
= HYPERVISOR_dom0_op(&op
);
1397 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1400 C(interface_support
);
1401 C(legacy_max_cylinder
);
1403 C(legacy_sectors_per_track
);
1406 boot_params
.eddbuf_entries
= nr
;
1408 op
.u
.firmware_info
.type
= XEN_FW_DISK_MBR_SIGNATURE
;
1409 for (nr
= 0; nr
< EDD_MBR_SIG_MAX
; nr
++) {
1410 op
.u
.firmware_info
.index
= nr
;
1411 ret
= HYPERVISOR_dom0_op(&op
);
1414 mbr_signature
[nr
] = op
.u
.firmware_info
.u
.disk_mbr_signature
.mbr_signature
;
1416 boot_params
.edd_mbr_sig_buf_entries
= nr
;
1421 * Set up the GDT and segment registers for -fstack-protector. Until
1422 * we do this, we have to be careful not to call any stack-protected
1423 * function, which is most of the kernel.
1425 * Note, that it is __ref because the only caller of this after init
1426 * is PVH which is not going to use xen_load_gdt_boot or other
1429 static void __ref
xen_setup_gdt(int cpu
)
1431 if (xen_feature(XENFEAT_auto_translated_physmap
)) {
1432 #ifdef CONFIG_X86_64
1433 unsigned long dummy
;
1435 load_percpu_segment(cpu
); /* We need to access per-cpu area */
1436 switch_to_new_gdt(cpu
); /* GDT and GS set */
1438 /* We are switching of the Xen provided GDT to our HVM mode
1439 * GDT. The new GDT has __KERNEL_CS with CS.L = 1
1440 * and we are jumping to reload it.
1442 asm volatile ("pushq %0\n"
1443 "leaq 1f(%%rip),%0\n"
1447 : "=&r" (dummy
) : "0" (__KERNEL_CS
));
1450 * While not needed, we also set the %es, %ds, and %fs
1451 * to zero. We don't care about %ss as it is NULL.
1452 * Strictly speaking this is not needed as Xen zeros those
1453 * out (and also MSR_FS_BASE, MSR_GS_BASE, MSR_KERNEL_GS_BASE)
1455 * Linux zeros them in cpu_init() and in secondary_startup_64
1462 /* PVH: TODO Implement. */
1465 return; /* PVH does not need any PV GDT ops. */
1467 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry_boot
;
1468 pv_cpu_ops
.load_gdt
= xen_load_gdt_boot
;
1470 setup_stack_canary_segment(0);
1471 switch_to_new_gdt(0);
1473 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry
;
1474 pv_cpu_ops
.load_gdt
= xen_load_gdt
;
1477 #ifdef CONFIG_XEN_PVH
1479 * A PV guest starts with default flags that are not set for PVH, set them
1482 static void xen_pvh_set_cr_flags(int cpu
)
1485 /* Some of these are setup in 'secondary_startup_64'. The others:
1486 * X86_CR0_TS, X86_CR0_PE, X86_CR0_ET are set by Xen for HVM guests
1487 * (which PVH shared codepaths), while X86_CR0_PG is for PVH. */
1488 write_cr0(read_cr0() | X86_CR0_MP
| X86_CR0_NE
| X86_CR0_WP
| X86_CR0_AM
);
1493 * For BSP, PSE PGE are set in probe_page_size_mask(), for APs
1494 * set them here. For all, OSFXSR OSXMMEXCPT are set in fpu_init.
1497 set_in_cr4(X86_CR4_PSE
);
1500 set_in_cr4(X86_CR4_PGE
);
1504 * Note, that it is ref - because the only caller of this after init
1505 * is PVH which is not going to use xen_load_gdt_boot or other
1508 void __ref
xen_pvh_secondary_vcpu_init(int cpu
)
1511 xen_pvh_set_cr_flags(cpu
);
1514 static void __init
xen_pvh_early_guest_init(void)
1516 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1519 if (!xen_feature(XENFEAT_hvm_callback_vector
))
1522 xen_have_vector_callback
= 1;
1524 xen_pvh_early_cpu_init(0, false);
1525 xen_pvh_set_cr_flags(0);
1527 #ifdef CONFIG_X86_32
1528 BUG(); /* PVH: Implement proper support. */
1531 #endif /* CONFIG_XEN_PVH */
1533 /* First C function to be called on Xen boot */
1534 asmlinkage __visible
void __init
xen_start_kernel(void)
1536 struct physdev_set_iopl set_iopl
;
1537 unsigned long initrd_start
= 0;
1540 if (!xen_start_info
)
1543 xen_domain_type
= XEN_PV_DOMAIN
;
1545 xen_setup_features();
1546 #ifdef CONFIG_XEN_PVH
1547 xen_pvh_early_guest_init();
1549 xen_setup_machphys_mapping();
1551 /* Install Xen paravirt ops */
1553 pv_init_ops
= xen_init_ops
;
1554 pv_apic_ops
= xen_apic_ops
;
1555 if (!xen_pvh_domain()) {
1556 pv_cpu_ops
= xen_cpu_ops
;
1558 x86_platform
.get_nmi_reason
= xen_get_nmi_reason
;
1561 if (xen_feature(XENFEAT_auto_translated_physmap
))
1562 x86_init
.resources
.memory_setup
= xen_auto_xlated_memory_setup
;
1564 x86_init
.resources
.memory_setup
= xen_memory_setup
;
1565 x86_init
.oem
.arch_setup
= xen_arch_setup
;
1566 x86_init
.oem
.banner
= xen_banner
;
1568 xen_init_time_ops();
1571 * Set up some pagetable state before starting to set any ptes.
1576 /* Prevent unwanted bits from being set in PTEs. */
1577 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1580 * Prevent page tables from being allocated in highmem, even
1581 * if CONFIG_HIGHPTE is enabled.
1583 __userpte_alloc_gfp
&= ~__GFP_HIGHMEM
;
1585 /* Work out if we support NX */
1589 xen_build_dynamic_phys_to_machine();
1592 * Set up kernel GDT and segment registers, mainly so that
1593 * -fstack-protector code can be executed.
1598 xen_init_cpuid_mask();
1600 #ifdef CONFIG_X86_LOCAL_APIC
1602 * set up the basic apic ops.
1604 set_xen_basic_apic_ops();
1607 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1608 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1609 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1612 machine_ops
= xen_machine_ops
;
1615 * The only reliable way to retain the initial address of the
1616 * percpu gdt_page is to remember it here, so we can go and
1617 * mark it RW later, when the initial percpu area is freed.
1619 xen_initial_gdt
= &per_cpu(gdt_page
, 0);
1623 #ifdef CONFIG_ACPI_NUMA
1625 * The pages we from Xen are not related to machine pages, so
1626 * any NUMA information the kernel tries to get from ACPI will
1627 * be meaningless. Prevent it from trying.
1631 /* Don't do the full vcpu_info placement stuff until we have a
1632 possible map and a non-dummy shared_info. */
1633 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1635 local_irq_disable();
1636 early_boot_irqs_disabled
= true;
1638 xen_raw_console_write("mapping kernel into physical memory\n");
1639 xen_setup_kernel_pagetable((pgd_t
*)xen_start_info
->pt_base
, xen_start_info
->nr_pages
);
1642 * Modify the cache mode translation tables to match Xen's PAT
1646 pat_init_cache_modes();
1648 /* keep using Xen gdt for now; no urgent need to change it */
1650 #ifdef CONFIG_X86_32
1651 pv_info
.kernel_rpl
= 1;
1652 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1653 pv_info
.kernel_rpl
= 0;
1655 pv_info
.kernel_rpl
= 0;
1657 /* set the limit of our address space */
1660 /* PVH: runs at default kernel iopl of 0 */
1661 if (!xen_pvh_domain()) {
1663 * We used to do this in xen_arch_setup, but that is too late
1664 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1665 * early_amd_init which pokes 0xcf8 port.
1668 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
1670 xen_raw_printk("physdev_op failed %d\n", rc
);
1673 #ifdef CONFIG_X86_32
1674 /* set up basic CPUID stuff */
1675 cpu_detect(&new_cpu_data
);
1676 set_cpu_cap(&new_cpu_data
, X86_FEATURE_FPU
);
1677 new_cpu_data
.wp_works_ok
= 1;
1678 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1681 if (xen_start_info
->mod_start
) {
1682 if (xen_start_info
->flags
& SIF_MOD_START_PFN
)
1683 initrd_start
= PFN_PHYS(xen_start_info
->mod_start
);
1685 initrd_start
= __pa(xen_start_info
->mod_start
);
1688 /* Poke various useful things into boot_params */
1689 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1690 boot_params
.hdr
.ramdisk_image
= initrd_start
;
1691 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1692 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1694 if (!xen_initial_domain()) {
1695 add_preferred_console("xenboot", 0, NULL
);
1696 add_preferred_console("tty", 0, NULL
);
1697 add_preferred_console("hvc", 0, NULL
);
1699 x86_init
.pci
.arch_init
= pci_xen_init
;
1701 const struct dom0_vga_console_info
*info
=
1702 (void *)((char *)xen_start_info
+
1703 xen_start_info
->console
.dom0
.info_off
);
1704 struct xen_platform_op op
= {
1705 .cmd
= XENPF_firmware_info
,
1706 .interface_version
= XENPF_INTERFACE_VERSION
,
1707 .u
.firmware_info
.type
= XEN_FW_KBD_SHIFT_FLAGS
,
1710 xen_init_vga(info
, xen_start_info
->console
.dom0
.info_size
);
1711 xen_start_info
->console
.domU
.mfn
= 0;
1712 xen_start_info
->console
.domU
.evtchn
= 0;
1714 if (HYPERVISOR_dom0_op(&op
) == 0)
1715 boot_params
.kbd_status
= op
.u
.firmware_info
.u
.kbd_shift_flags
;
1719 /* Make sure ACS will be enabled */
1722 xen_acpi_sleep_register();
1724 /* Avoid searching for BIOS MP tables */
1725 x86_init
.mpparse
.find_smp_config
= x86_init_noop
;
1726 x86_init
.mpparse
.get_smp_config
= x86_init_uint_noop
;
1728 xen_boot_params_init_edd();
1731 /* PCI BIOS service won't work from a PV guest. */
1732 pci_probe
&= ~PCI_PROBE_BIOS
;
1734 xen_raw_console_write("about to get started...\n");
1736 xen_setup_runstate_info(0);
1740 /* Start the world */
1741 #ifdef CONFIG_X86_32
1742 i386_start_kernel();
1744 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));
1748 void __ref
xen_hvm_init_shared_info(void)
1751 struct xen_add_to_physmap xatp
;
1752 static struct shared_info
*shared_info_page
= 0;
1754 if (!shared_info_page
)
1755 shared_info_page
= (struct shared_info
*)
1756 extend_brk(PAGE_SIZE
, PAGE_SIZE
);
1757 xatp
.domid
= DOMID_SELF
;
1759 xatp
.space
= XENMAPSPACE_shared_info
;
1760 xatp
.gpfn
= __pa(shared_info_page
) >> PAGE_SHIFT
;
1761 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap
, &xatp
))
1764 HYPERVISOR_shared_info
= (struct shared_info
*)shared_info_page
;
1766 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1767 * page, we use it in the event channel upcall and in some pvclock
1768 * related functions. We don't need the vcpu_info placement
1769 * optimizations because we don't use any pv_mmu or pv_irq op on
1771 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1772 * online but xen_hvm_init_shared_info is run at resume time too and
1773 * in that case multiple vcpus might be online. */
1774 for_each_online_cpu(cpu
) {
1775 /* Leave it to be NULL. */
1776 if (cpu
>= MAX_VIRT_CPUS
)
1778 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
1782 #ifdef CONFIG_XEN_PVHVM
1783 static void __init
init_hvm_pv_info(void)
1786 uint32_t eax
, ebx
, ecx
, edx
, pages
, msr
, base
;
1789 base
= xen_cpuid_base();
1790 cpuid(base
+ 1, &eax
, &ebx
, &ecx
, &edx
);
1793 minor
= eax
& 0xffff;
1794 printk(KERN_INFO
"Xen version %d.%d.\n", major
, minor
);
1796 cpuid(base
+ 2, &pages
, &msr
, &ecx
, &edx
);
1798 pfn
= __pa(hypercall_page
);
1799 wrmsr_safe(msr
, (u32
)pfn
, (u32
)(pfn
>> 32));
1801 xen_setup_features();
1803 pv_info
.name
= "Xen HVM";
1805 xen_domain_type
= XEN_HVM_DOMAIN
;
1808 static int xen_hvm_cpu_notify(struct notifier_block
*self
, unsigned long action
,
1811 int cpu
= (long)hcpu
;
1813 case CPU_UP_PREPARE
:
1814 xen_vcpu_setup(cpu
);
1815 if (xen_have_vector_callback
) {
1816 if (xen_feature(XENFEAT_hvm_safe_pvclock
))
1817 xen_setup_timer(cpu
);
1826 static struct notifier_block xen_hvm_cpu_notifier
= {
1827 .notifier_call
= xen_hvm_cpu_notify
,
1830 static void __init
xen_hvm_guest_init(void)
1834 xen_hvm_init_shared_info();
1836 xen_panic_handler_init();
1838 if (xen_feature(XENFEAT_hvm_callback_vector
))
1839 xen_have_vector_callback
= 1;
1841 register_cpu_notifier(&xen_hvm_cpu_notifier
);
1842 xen_unplug_emulated_devices();
1843 x86_init
.irqs
.intr_init
= xen_init_IRQ
;
1844 xen_hvm_init_time_ops();
1845 xen_hvm_init_mmu_ops();
1848 static bool xen_nopv
= false;
1849 static __init
int xen_parse_nopv(char *arg
)
1854 early_param("xen_nopv", xen_parse_nopv
);
1856 static uint32_t __init
xen_hvm_platform(void)
1861 if (xen_pv_domain())
1864 return xen_cpuid_base();
1867 bool xen_hvm_need_lapic(void)
1871 if (xen_pv_domain())
1873 if (!xen_hvm_domain())
1875 if (xen_feature(XENFEAT_hvm_pirqs
) && xen_have_vector_callback
)
1879 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic
);
1881 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst
= {
1883 .detect
= xen_hvm_platform
,
1884 .init_platform
= xen_hvm_guest_init
,
1885 .x2apic_available
= xen_x2apic_para_available
,
1887 EXPORT_SYMBOL(x86_hyper_xen_hvm
);