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>
36 #include <xen/interface/xen.h>
37 #include <xen/interface/version.h>
38 #include <xen/interface/physdev.h>
39 #include <xen/interface/vcpu.h>
40 #include <xen/interface/memory.h>
41 #include <xen/features.h>
44 #include <xen/hvc-console.h>
47 #include <asm/paravirt.h>
50 #include <asm/xen/pci.h>
51 #include <asm/xen/hypercall.h>
52 #include <asm/xen/hypervisor.h>
53 #include <asm/fixmap.h>
54 #include <asm/processor.h>
55 #include <asm/proto.h>
56 #include <asm/msr-index.h>
57 #include <asm/traps.h>
58 #include <asm/setup.h>
60 #include <asm/pgalloc.h>
61 #include <asm/pgtable.h>
62 #include <asm/tlbflush.h>
63 #include <asm/reboot.h>
64 #include <asm/stackprotector.h>
65 #include <asm/hypervisor.h>
66 #include <asm/mwait.h>
69 #include <linux/acpi.h>
71 #include <acpi/pdc_intel.h>
72 #include <acpi/processor.h>
73 #include <xen/interface/platform.h>
79 #include "multicalls.h"
81 EXPORT_SYMBOL_GPL(hypercall_page
);
83 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
84 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
86 enum xen_domain_type xen_domain_type
= XEN_NATIVE
;
87 EXPORT_SYMBOL_GPL(xen_domain_type
);
89 unsigned long *machine_to_phys_mapping
= (void *)MACH2PHYS_VIRT_START
;
90 EXPORT_SYMBOL(machine_to_phys_mapping
);
91 unsigned long machine_to_phys_nr
;
92 EXPORT_SYMBOL(machine_to_phys_nr
);
94 struct start_info
*xen_start_info
;
95 EXPORT_SYMBOL_GPL(xen_start_info
);
97 struct shared_info xen_dummy_shared_info
;
99 void *xen_initial_gdt
;
101 RESERVE_BRK(shared_info_page_brk
, PAGE_SIZE
);
102 __read_mostly
int xen_have_vector_callback
;
103 EXPORT_SYMBOL_GPL(xen_have_vector_callback
);
106 * Point at some empty memory to start with. We map the real shared_info
107 * page as soon as fixmap is up and running.
109 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
112 * Flag to determine whether vcpu info placement is available on all
113 * VCPUs. We assume it is to start with, and then set it to zero on
114 * the first failure. This is because it can succeed on some VCPUs
115 * and not others, since it can involve hypervisor memory allocation,
116 * or because the guest failed to guarantee all the appropriate
117 * constraints on all VCPUs (ie buffer can't cross a page boundary).
119 * Note that any particular CPU may be using a placed vcpu structure,
120 * but we can only optimise if the all are.
122 * 0: not available, 1: available
124 static int have_vcpu_info_placement
= 1;
126 static void clamp_max_cpus(void)
129 if (setup_max_cpus
> MAX_VIRT_CPUS
)
130 setup_max_cpus
= MAX_VIRT_CPUS
;
134 static void xen_vcpu_setup(int cpu
)
136 struct vcpu_register_vcpu_info info
;
138 struct vcpu_info
*vcpup
;
140 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
142 if (cpu
< MAX_VIRT_CPUS
)
143 per_cpu(xen_vcpu
,cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
145 if (!have_vcpu_info_placement
) {
146 if (cpu
>= MAX_VIRT_CPUS
)
151 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
152 info
.mfn
= arbitrary_virt_to_mfn(vcpup
);
153 info
.offset
= offset_in_page(vcpup
);
155 /* Check to see if the hypervisor will put the vcpu_info
156 structure where we want it, which allows direct access via
157 a percpu-variable. */
158 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
161 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
162 have_vcpu_info_placement
= 0;
165 /* This cpu is using the registered vcpu info, even if
166 later ones fail to. */
167 per_cpu(xen_vcpu
, cpu
) = vcpup
;
172 * On restore, set the vcpu placement up again.
173 * If it fails, then we're in a bad state, since
174 * we can't back out from using it...
176 void xen_vcpu_restore(void)
180 for_each_online_cpu(cpu
) {
181 bool other_cpu
= (cpu
!= smp_processor_id());
184 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
187 xen_setup_runstate_info(cpu
);
189 if (have_vcpu_info_placement
)
193 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
198 static void __init
xen_banner(void)
200 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
201 struct xen_extraversion extra
;
202 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
204 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
206 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
207 version
>> 16, version
& 0xffff, extra
.extraversion
,
208 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
211 #define CPUID_THERM_POWER_LEAF 6
212 #define APERFMPERF_PRESENT 0
214 static __read_mostly
unsigned int cpuid_leaf1_edx_mask
= ~0;
215 static __read_mostly
unsigned int cpuid_leaf1_ecx_mask
= ~0;
217 static __read_mostly
unsigned int cpuid_leaf1_ecx_set_mask
;
218 static __read_mostly
unsigned int cpuid_leaf5_ecx_val
;
219 static __read_mostly
unsigned int cpuid_leaf5_edx_val
;
221 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
222 unsigned int *cx
, unsigned int *dx
)
224 unsigned maskebx
= ~0;
225 unsigned maskecx
= ~0;
226 unsigned maskedx
= ~0;
229 * Mask out inconvenient features, to try and disable as many
230 * unsupported kernel subsystems as possible.
234 maskecx
= cpuid_leaf1_ecx_mask
;
235 setecx
= cpuid_leaf1_ecx_set_mask
;
236 maskedx
= cpuid_leaf1_edx_mask
;
239 case CPUID_MWAIT_LEAF
:
240 /* Synthesize the values.. */
243 *cx
= cpuid_leaf5_ecx_val
;
244 *dx
= cpuid_leaf5_edx_val
;
247 case CPUID_THERM_POWER_LEAF
:
248 /* Disabling APERFMPERF for kernel usage */
249 maskecx
= ~(1 << APERFMPERF_PRESENT
);
253 /* Suppress extended topology stuff */
258 asm(XEN_EMULATE_PREFIX
"cpuid"
263 : "0" (*ax
), "2" (*cx
));
272 static bool __init
xen_check_mwait(void)
275 struct xen_platform_op op
= {
276 .cmd
= XENPF_set_processor_pminfo
,
277 .u
.set_pminfo
.id
= -1,
278 .u
.set_pminfo
.type
= XEN_PM_PDC
,
281 unsigned int ax
, bx
, cx
, dx
;
282 unsigned int mwait_mask
;
284 /* We need to determine whether it is OK to expose the MWAIT
285 * capability to the kernel to harvest deeper than C3 states from ACPI
286 * _CST using the processor_harvest_xen.c module. For this to work, we
287 * need to gather the MWAIT_LEAF values (which the cstate.c code
288 * checks against). The hypervisor won't expose the MWAIT flag because
289 * it would break backwards compatibility; so we will find out directly
290 * from the hardware and hypercall.
292 if (!xen_initial_domain())
298 native_cpuid(&ax
, &bx
, &cx
, &dx
);
300 mwait_mask
= (1 << (X86_FEATURE_EST
% 32)) |
301 (1 << (X86_FEATURE_MWAIT
% 32));
303 if ((cx
& mwait_mask
) != mwait_mask
)
306 /* We need to emulate the MWAIT_LEAF and for that we need both
307 * ecx and edx. The hypercall provides only partial information.
310 ax
= CPUID_MWAIT_LEAF
;
315 native_cpuid(&ax
, &bx
, &cx
, &dx
);
317 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
318 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
320 buf
[0] = ACPI_PDC_REVISION_ID
;
322 buf
[2] = (ACPI_PDC_C_CAPABILITY_SMP
| ACPI_PDC_EST_CAPABILITY_SWSMP
);
324 set_xen_guest_handle(op
.u
.set_pminfo
.pdc
, buf
);
326 if ((HYPERVISOR_dom0_op(&op
) == 0) &&
327 (buf
[2] & (ACPI_PDC_C_C1_FFH
| ACPI_PDC_C_C2C3_FFH
))) {
328 cpuid_leaf5_ecx_val
= cx
;
329 cpuid_leaf5_edx_val
= dx
;
336 static void __init
xen_init_cpuid_mask(void)
338 unsigned int ax
, bx
, cx
, dx
;
339 unsigned int xsave_mask
;
341 cpuid_leaf1_edx_mask
=
342 ~((1 << X86_FEATURE_MCE
) | /* disable MCE */
343 (1 << X86_FEATURE_MCA
) | /* disable MCA */
344 (1 << X86_FEATURE_MTRR
) | /* disable MTRR */
345 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
347 if (!xen_initial_domain())
348 cpuid_leaf1_edx_mask
&=
349 ~((1 << X86_FEATURE_APIC
) | /* disable local APIC */
350 (1 << X86_FEATURE_ACPI
)); /* disable ACPI */
353 xen_cpuid(&ax
, &bx
, &cx
, &dx
);
356 (1 << (X86_FEATURE_XSAVE
% 32)) |
357 (1 << (X86_FEATURE_OSXSAVE
% 32));
359 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
360 if ((cx
& xsave_mask
) != xsave_mask
)
361 cpuid_leaf1_ecx_mask
&= ~xsave_mask
; /* disable XSAVE & OSXSAVE */
363 if (xen_check_mwait())
364 cpuid_leaf1_ecx_set_mask
= (1 << (X86_FEATURE_MWAIT
% 32));
367 static void xen_set_debugreg(int reg
, unsigned long val
)
369 HYPERVISOR_set_debugreg(reg
, val
);
372 static unsigned long xen_get_debugreg(int reg
)
374 return HYPERVISOR_get_debugreg(reg
);
377 static void xen_end_context_switch(struct task_struct
*next
)
380 paravirt_end_context_switch(next
);
383 static unsigned long xen_store_tr(void)
389 * Set the page permissions for a particular virtual address. If the
390 * address is a vmalloc mapping (or other non-linear mapping), then
391 * find the linear mapping of the page and also set its protections to
394 static void set_aliased_prot(void *v
, pgprot_t prot
)
402 ptep
= lookup_address((unsigned long)v
, &level
);
403 BUG_ON(ptep
== NULL
);
405 pfn
= pte_pfn(*ptep
);
406 page
= pfn_to_page(pfn
);
408 pte
= pfn_pte(pfn
, prot
);
410 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
413 if (!PageHighMem(page
)) {
414 void *av
= __va(PFN_PHYS(pfn
));
417 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
423 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
425 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
428 for(i
= 0; i
< entries
; i
+= entries_per_page
)
429 set_aliased_prot(ldt
+ i
, PAGE_KERNEL_RO
);
432 static void xen_free_ldt(struct desc_struct
*ldt
, unsigned entries
)
434 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
437 for(i
= 0; i
< entries
; i
+= entries_per_page
)
438 set_aliased_prot(ldt
+ i
, PAGE_KERNEL
);
441 static void xen_set_ldt(const void *addr
, unsigned entries
)
443 struct mmuext_op
*op
;
444 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
446 trace_xen_cpu_set_ldt(addr
, entries
);
449 op
->cmd
= MMUEXT_SET_LDT
;
450 op
->arg1
.linear_addr
= (unsigned long)addr
;
451 op
->arg2
.nr_ents
= entries
;
453 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
455 xen_mc_issue(PARAVIRT_LAZY_CPU
);
458 static void xen_load_gdt(const struct desc_ptr
*dtr
)
460 unsigned long va
= dtr
->address
;
461 unsigned int size
= dtr
->size
+ 1;
462 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
463 unsigned long frames
[pages
];
467 * A GDT can be up to 64k in size, which corresponds to 8192
468 * 8-byte entries, or 16 4k pages..
471 BUG_ON(size
> 65536);
472 BUG_ON(va
& ~PAGE_MASK
);
474 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
477 unsigned long pfn
, mfn
;
481 * The GDT is per-cpu and is in the percpu data area.
482 * That can be virtually mapped, so we need to do a
483 * page-walk to get the underlying MFN for the
484 * hypercall. The page can also be in the kernel's
485 * linear range, so we need to RO that mapping too.
487 ptep
= lookup_address(va
, &level
);
488 BUG_ON(ptep
== NULL
);
490 pfn
= pte_pfn(*ptep
);
491 mfn
= pfn_to_mfn(pfn
);
492 virt
= __va(PFN_PHYS(pfn
));
496 make_lowmem_page_readonly((void *)va
);
497 make_lowmem_page_readonly(virt
);
500 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
505 * load_gdt for early boot, when the gdt is only mapped once
507 static void __init
xen_load_gdt_boot(const struct desc_ptr
*dtr
)
509 unsigned long va
= dtr
->address
;
510 unsigned int size
= dtr
->size
+ 1;
511 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
512 unsigned long frames
[pages
];
516 * A GDT can be up to 64k in size, which corresponds to 8192
517 * 8-byte entries, or 16 4k pages..
520 BUG_ON(size
> 65536);
521 BUG_ON(va
& ~PAGE_MASK
);
523 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
525 unsigned long pfn
, mfn
;
527 pfn
= virt_to_pfn(va
);
528 mfn
= pfn_to_mfn(pfn
);
530 pte
= pfn_pte(pfn
, PAGE_KERNEL_RO
);
532 if (HYPERVISOR_update_va_mapping((unsigned long)va
, pte
, 0))
538 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
542 static void load_TLS_descriptor(struct thread_struct
*t
,
543 unsigned int cpu
, unsigned int i
)
545 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
546 xmaddr_t maddr
= arbitrary_virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
547 struct multicall_space mc
= __xen_mc_entry(0);
549 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
552 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
555 * XXX sleazy hack: If we're being called in a lazy-cpu zone
556 * and lazy gs handling is enabled, it means we're in a
557 * context switch, and %gs has just been saved. This means we
558 * can zero it out to prevent faults on exit from the
559 * hypervisor if the next process has no %gs. Either way, it
560 * has been saved, and the new value will get loaded properly.
561 * This will go away as soon as Xen has been modified to not
562 * save/restore %gs for normal hypercalls.
564 * On x86_64, this hack is not used for %gs, because gs points
565 * to KERNEL_GS_BASE (and uses it for PDA references), so we
566 * must not zero %gs on x86_64
568 * For x86_64, we need to zero %fs, otherwise we may get an
569 * exception between the new %fs descriptor being loaded and
570 * %fs being effectively cleared at __switch_to().
572 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
582 load_TLS_descriptor(t
, cpu
, 0);
583 load_TLS_descriptor(t
, cpu
, 1);
584 load_TLS_descriptor(t
, cpu
, 2);
586 xen_mc_issue(PARAVIRT_LAZY_CPU
);
590 static void xen_load_gs_index(unsigned int idx
)
592 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
597 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
600 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
601 u64 entry
= *(u64
*)ptr
;
603 trace_xen_cpu_write_ldt_entry(dt
, entrynum
, entry
);
608 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
614 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
615 struct trap_info
*info
)
619 if (val
->type
!= GATE_TRAP
&& val
->type
!= GATE_INTERRUPT
)
622 info
->vector
= vector
;
624 addr
= gate_offset(*val
);
627 * Look for known traps using IST, and substitute them
628 * appropriately. The debugger ones are the only ones we care
629 * about. Xen will handle faults like double_fault and
630 * machine_check, so we should never see them. Warn if
631 * there's an unexpected IST-using fault handler.
633 if (addr
== (unsigned long)debug
)
634 addr
= (unsigned long)xen_debug
;
635 else if (addr
== (unsigned long)int3
)
636 addr
= (unsigned long)xen_int3
;
637 else if (addr
== (unsigned long)stack_segment
)
638 addr
= (unsigned long)xen_stack_segment
;
639 else if (addr
== (unsigned long)double_fault
||
640 addr
== (unsigned long)nmi
) {
641 /* Don't need to handle these */
643 #ifdef CONFIG_X86_MCE
644 } else if (addr
== (unsigned long)machine_check
) {
648 /* Some other trap using IST? */
649 if (WARN_ON(val
->ist
!= 0))
652 #endif /* CONFIG_X86_64 */
653 info
->address
= addr
;
655 info
->cs
= gate_segment(*val
);
656 info
->flags
= val
->dpl
;
657 /* interrupt gates clear IF */
658 if (val
->type
== GATE_INTERRUPT
)
659 info
->flags
|= 1 << 2;
664 /* Locations of each CPU's IDT */
665 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
667 /* Set an IDT entry. If the entry is part of the current IDT, then
669 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
671 unsigned long p
= (unsigned long)&dt
[entrynum
];
672 unsigned long start
, end
;
674 trace_xen_cpu_write_idt_entry(dt
, entrynum
, g
);
678 start
= __this_cpu_read(idt_desc
.address
);
679 end
= start
+ __this_cpu_read(idt_desc
.size
) + 1;
683 native_write_idt_entry(dt
, entrynum
, g
);
685 if (p
>= start
&& (p
+ 8) <= end
) {
686 struct trap_info info
[2];
690 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
691 if (HYPERVISOR_set_trap_table(info
))
698 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
699 struct trap_info
*traps
)
701 unsigned in
, out
, count
;
703 count
= (desc
->size
+1) / sizeof(gate_desc
);
706 for (in
= out
= 0; in
< count
; in
++) {
707 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
709 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
712 traps
[out
].address
= 0;
715 void xen_copy_trap_info(struct trap_info
*traps
)
717 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
719 xen_convert_trap_info(desc
, traps
);
722 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
723 hold a spinlock to protect the static traps[] array (static because
724 it avoids allocation, and saves stack space). */
725 static void xen_load_idt(const struct desc_ptr
*desc
)
727 static DEFINE_SPINLOCK(lock
);
728 static struct trap_info traps
[257];
730 trace_xen_cpu_load_idt(desc
);
734 __get_cpu_var(idt_desc
) = *desc
;
736 xen_convert_trap_info(desc
, traps
);
739 if (HYPERVISOR_set_trap_table(traps
))
745 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
746 they're handled differently. */
747 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
748 const void *desc
, int type
)
750 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
761 xmaddr_t maddr
= arbitrary_virt_to_machine(&dt
[entry
]);
764 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
774 * Version of write_gdt_entry for use at early boot-time needed to
775 * update an entry as simply as possible.
777 static void __init
xen_write_gdt_entry_boot(struct desc_struct
*dt
, int entry
,
778 const void *desc
, int type
)
780 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
789 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
791 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
792 dt
[entry
] = *(struct desc_struct
*)desc
;
798 static void xen_load_sp0(struct tss_struct
*tss
,
799 struct thread_struct
*thread
)
801 struct multicall_space mcs
;
803 mcs
= xen_mc_entry(0);
804 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
805 xen_mc_issue(PARAVIRT_LAZY_CPU
);
808 static void xen_set_iopl_mask(unsigned mask
)
810 struct physdev_set_iopl set_iopl
;
812 /* Force the change at ring 0. */
813 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
814 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
817 static void xen_io_delay(void)
821 #ifdef CONFIG_X86_LOCAL_APIC
822 static u32
xen_apic_read(u32 reg
)
827 static void xen_apic_write(u32 reg
, u32 val
)
829 /* Warn to see if there's any stray references */
833 static u64
xen_apic_icr_read(void)
838 static void xen_apic_icr_write(u32 low
, u32 id
)
840 /* Warn to see if there's any stray references */
844 static void xen_apic_wait_icr_idle(void)
849 static u32
xen_safe_apic_wait_icr_idle(void)
854 static void set_xen_basic_apic_ops(void)
856 apic
->read
= xen_apic_read
;
857 apic
->write
= xen_apic_write
;
858 apic
->icr_read
= xen_apic_icr_read
;
859 apic
->icr_write
= xen_apic_icr_write
;
860 apic
->wait_icr_idle
= xen_apic_wait_icr_idle
;
861 apic
->safe_wait_icr_idle
= xen_safe_apic_wait_icr_idle
;
864 apic
->send_IPI_allbutself
= xen_send_IPI_allbutself
;
865 apic
->send_IPI_mask_allbutself
= xen_send_IPI_mask_allbutself
;
866 apic
->send_IPI_mask
= xen_send_IPI_mask
;
867 apic
->send_IPI_all
= xen_send_IPI_all
;
868 apic
->send_IPI_self
= xen_send_IPI_self
;
874 static void xen_clts(void)
876 struct multicall_space mcs
;
878 mcs
= xen_mc_entry(0);
880 MULTI_fpu_taskswitch(mcs
.mc
, 0);
882 xen_mc_issue(PARAVIRT_LAZY_CPU
);
885 static DEFINE_PER_CPU(unsigned long, xen_cr0_value
);
887 static unsigned long xen_read_cr0(void)
889 unsigned long cr0
= this_cpu_read(xen_cr0_value
);
891 if (unlikely(cr0
== 0)) {
892 cr0
= native_read_cr0();
893 this_cpu_write(xen_cr0_value
, cr0
);
899 static void xen_write_cr0(unsigned long cr0
)
901 struct multicall_space mcs
;
903 this_cpu_write(xen_cr0_value
, cr0
);
905 /* Only pay attention to cr0.TS; everything else is
907 mcs
= xen_mc_entry(0);
909 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
911 xen_mc_issue(PARAVIRT_LAZY_CPU
);
914 static void xen_write_cr4(unsigned long cr4
)
919 native_write_cr4(cr4
);
922 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
933 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
934 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
935 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
938 base
= ((u64
)high
<< 32) | low
;
939 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
947 case MSR_SYSCALL_MASK
:
948 case MSR_IA32_SYSENTER_CS
:
949 case MSR_IA32_SYSENTER_ESP
:
950 case MSR_IA32_SYSENTER_EIP
:
951 /* Fast syscall setup is all done in hypercalls, so
952 these are all ignored. Stub them out here to stop
953 Xen console noise. */
956 case MSR_IA32_CR_PAT
:
957 if (smp_processor_id() == 0)
958 xen_set_pat(((u64
)high
<< 32) | low
);
962 ret
= native_write_msr_safe(msr
, low
, high
);
968 void xen_setup_shared_info(void)
970 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
971 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
972 xen_start_info
->shared_info
);
974 HYPERVISOR_shared_info
=
975 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
977 HYPERVISOR_shared_info
=
978 (struct shared_info
*)__va(xen_start_info
->shared_info
);
981 /* In UP this is as good a place as any to set up shared info */
982 xen_setup_vcpu_info_placement();
985 xen_setup_mfn_list_list();
988 /* This is called once we have the cpu_possible_mask */
989 void xen_setup_vcpu_info_placement(void)
993 for_each_possible_cpu(cpu
)
996 /* xen_vcpu_setup managed to place the vcpu_info within the
997 percpu area for all cpus, so make use of it */
998 if (have_vcpu_info_placement
) {
999 pv_irq_ops
.save_fl
= __PV_IS_CALLEE_SAVE(xen_save_fl_direct
);
1000 pv_irq_ops
.restore_fl
= __PV_IS_CALLEE_SAVE(xen_restore_fl_direct
);
1001 pv_irq_ops
.irq_disable
= __PV_IS_CALLEE_SAVE(xen_irq_disable_direct
);
1002 pv_irq_ops
.irq_enable
= __PV_IS_CALLEE_SAVE(xen_irq_enable_direct
);
1003 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1007 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1008 unsigned long addr
, unsigned len
)
1010 char *start
, *end
, *reloc
;
1013 start
= end
= reloc
= NULL
;
1015 #define SITE(op, x) \
1016 case PARAVIRT_PATCH(op.x): \
1017 if (have_vcpu_info_placement) { \
1018 start = (char *)xen_##x##_direct; \
1019 end = xen_##x##_direct_end; \
1020 reloc = xen_##x##_direct_reloc; \
1025 SITE(pv_irq_ops
, irq_enable
);
1026 SITE(pv_irq_ops
, irq_disable
);
1027 SITE(pv_irq_ops
, save_fl
);
1028 SITE(pv_irq_ops
, restore_fl
);
1032 if (start
== NULL
|| (end
-start
) > len
)
1035 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1037 /* Note: because reloc is assigned from something that
1038 appears to be an array, gcc assumes it's non-null,
1039 but doesn't know its relationship with start and
1041 if (reloc
> start
&& reloc
< end
) {
1042 int reloc_off
= reloc
- start
;
1043 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1044 long delta
= start
- (char *)addr
;
1052 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1060 static const struct pv_info xen_info __initconst
= {
1061 .paravirt_enabled
= 1,
1062 .shared_kernel_pmd
= 0,
1064 #ifdef CONFIG_X86_64
1065 .extra_user_64bit_cs
= FLAT_USER_CS64
,
1071 static const struct pv_init_ops xen_init_ops __initconst
= {
1075 static const struct pv_cpu_ops xen_cpu_ops __initconst
= {
1078 .set_debugreg
= xen_set_debugreg
,
1079 .get_debugreg
= xen_get_debugreg
,
1083 .read_cr0
= xen_read_cr0
,
1084 .write_cr0
= xen_write_cr0
,
1086 .read_cr4
= native_read_cr4
,
1087 .read_cr4_safe
= native_read_cr4_safe
,
1088 .write_cr4
= xen_write_cr4
,
1090 .wbinvd
= native_wbinvd
,
1092 .read_msr
= native_read_msr_safe
,
1093 .write_msr
= xen_write_msr_safe
,
1094 .read_tsc
= native_read_tsc
,
1095 .read_pmc
= native_read_pmc
,
1098 .irq_enable_sysexit
= xen_sysexit
,
1099 #ifdef CONFIG_X86_64
1100 .usergs_sysret32
= xen_sysret32
,
1101 .usergs_sysret64
= xen_sysret64
,
1104 .load_tr_desc
= paravirt_nop
,
1105 .set_ldt
= xen_set_ldt
,
1106 .load_gdt
= xen_load_gdt
,
1107 .load_idt
= xen_load_idt
,
1108 .load_tls
= xen_load_tls
,
1109 #ifdef CONFIG_X86_64
1110 .load_gs_index
= xen_load_gs_index
,
1113 .alloc_ldt
= xen_alloc_ldt
,
1114 .free_ldt
= xen_free_ldt
,
1116 .store_gdt
= native_store_gdt
,
1117 .store_idt
= native_store_idt
,
1118 .store_tr
= xen_store_tr
,
1120 .write_ldt_entry
= xen_write_ldt_entry
,
1121 .write_gdt_entry
= xen_write_gdt_entry
,
1122 .write_idt_entry
= xen_write_idt_entry
,
1123 .load_sp0
= xen_load_sp0
,
1125 .set_iopl_mask
= xen_set_iopl_mask
,
1126 .io_delay
= xen_io_delay
,
1128 /* Xen takes care of %gs when switching to usermode for us */
1129 .swapgs
= paravirt_nop
,
1131 .start_context_switch
= paravirt_start_context_switch
,
1132 .end_context_switch
= xen_end_context_switch
,
1135 static const struct pv_apic_ops xen_apic_ops __initconst
= {
1136 #ifdef CONFIG_X86_LOCAL_APIC
1137 .startup_ipi_hook
= paravirt_nop
,
1141 static void xen_reboot(int reason
)
1143 struct sched_shutdown r
= { .reason
= reason
};
1145 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1149 static void xen_restart(char *msg
)
1151 xen_reboot(SHUTDOWN_reboot
);
1154 static void xen_emergency_restart(void)
1156 xen_reboot(SHUTDOWN_reboot
);
1159 static void xen_machine_halt(void)
1161 xen_reboot(SHUTDOWN_poweroff
);
1164 static void xen_machine_power_off(void)
1168 xen_reboot(SHUTDOWN_poweroff
);
1171 static void xen_crash_shutdown(struct pt_regs
*regs
)
1173 xen_reboot(SHUTDOWN_crash
);
1177 xen_panic_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
1179 xen_reboot(SHUTDOWN_crash
);
1183 static struct notifier_block xen_panic_block
= {
1184 .notifier_call
= xen_panic_event
,
1187 int xen_panic_handler_init(void)
1189 atomic_notifier_chain_register(&panic_notifier_list
, &xen_panic_block
);
1193 static const struct machine_ops xen_machine_ops __initconst
= {
1194 .restart
= xen_restart
,
1195 .halt
= xen_machine_halt
,
1196 .power_off
= xen_machine_power_off
,
1197 .shutdown
= xen_machine_halt
,
1198 .crash_shutdown
= xen_crash_shutdown
,
1199 .emergency_restart
= xen_emergency_restart
,
1203 * Set up the GDT and segment registers for -fstack-protector. Until
1204 * we do this, we have to be careful not to call any stack-protected
1205 * function, which is most of the kernel.
1207 static void __init
xen_setup_stackprotector(void)
1209 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry_boot
;
1210 pv_cpu_ops
.load_gdt
= xen_load_gdt_boot
;
1212 setup_stack_canary_segment(0);
1213 switch_to_new_gdt(0);
1215 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry
;
1216 pv_cpu_ops
.load_gdt
= xen_load_gdt
;
1219 /* First C function to be called on Xen boot */
1220 asmlinkage
void __init
xen_start_kernel(void)
1222 struct physdev_set_iopl set_iopl
;
1226 if (!xen_start_info
)
1229 xen_domain_type
= XEN_PV_DOMAIN
;
1231 xen_setup_machphys_mapping();
1233 /* Install Xen paravirt ops */
1235 pv_init_ops
= xen_init_ops
;
1236 pv_cpu_ops
= xen_cpu_ops
;
1237 pv_apic_ops
= xen_apic_ops
;
1239 x86_init
.resources
.memory_setup
= xen_memory_setup
;
1240 x86_init
.oem
.arch_setup
= xen_arch_setup
;
1241 x86_init
.oem
.banner
= xen_banner
;
1243 xen_init_time_ops();
1246 * Set up some pagetable state before starting to set any ptes.
1251 /* Prevent unwanted bits from being set in PTEs. */
1252 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1254 if (!xen_initial_domain())
1256 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1258 __supported_pte_mask
|= _PAGE_IOMAP
;
1261 * Prevent page tables from being allocated in highmem, even
1262 * if CONFIG_HIGHPTE is enabled.
1264 __userpte_alloc_gfp
&= ~__GFP_HIGHMEM
;
1266 /* Work out if we support NX */
1269 xen_setup_features();
1272 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1273 xen_build_dynamic_phys_to_machine();
1276 * Set up kernel GDT and segment registers, mainly so that
1277 * -fstack-protector code can be executed.
1279 xen_setup_stackprotector();
1282 xen_init_cpuid_mask();
1284 #ifdef CONFIG_X86_LOCAL_APIC
1286 * set up the basic apic ops.
1288 set_xen_basic_apic_ops();
1291 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1292 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1293 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1296 machine_ops
= xen_machine_ops
;
1299 * The only reliable way to retain the initial address of the
1300 * percpu gdt_page is to remember it here, so we can go and
1301 * mark it RW later, when the initial percpu area is freed.
1303 xen_initial_gdt
= &per_cpu(gdt_page
, 0);
1307 #ifdef CONFIG_ACPI_NUMA
1309 * The pages we from Xen are not related to machine pages, so
1310 * any NUMA information the kernel tries to get from ACPI will
1311 * be meaningless. Prevent it from trying.
1316 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1318 /* Don't do the full vcpu_info placement stuff until we have a
1319 possible map and a non-dummy shared_info. */
1320 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1322 local_irq_disable();
1323 early_boot_irqs_disabled
= true;
1325 xen_raw_console_write("mapping kernel into physical memory\n");
1326 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
1328 /* Allocate and initialize top and mid mfn levels for p2m structure */
1329 xen_build_mfn_list_list();
1331 /* keep using Xen gdt for now; no urgent need to change it */
1333 #ifdef CONFIG_X86_32
1334 pv_info
.kernel_rpl
= 1;
1335 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1336 pv_info
.kernel_rpl
= 0;
1338 pv_info
.kernel_rpl
= 0;
1340 /* set the limit of our address space */
1343 /* We used to do this in xen_arch_setup, but that is too late on AMD
1344 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1345 * which pokes 0xcf8 port.
1348 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
1350 xen_raw_printk("physdev_op failed %d\n", rc
);
1352 #ifdef CONFIG_X86_32
1353 /* set up basic CPUID stuff */
1354 cpu_detect(&new_cpu_data
);
1355 new_cpu_data
.hard_math
= 1;
1356 new_cpu_data
.wp_works_ok
= 1;
1357 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1360 /* Poke various useful things into boot_params */
1361 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1362 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1363 ? __pa(xen_start_info
->mod_start
) : 0;
1364 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1365 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1367 if (!xen_initial_domain()) {
1368 add_preferred_console("xenboot", 0, NULL
);
1369 add_preferred_console("tty", 0, NULL
);
1370 add_preferred_console("hvc", 0, NULL
);
1372 x86_init
.pci
.arch_init
= pci_xen_init
;
1374 const struct dom0_vga_console_info
*info
=
1375 (void *)((char *)xen_start_info
+
1376 xen_start_info
->console
.dom0
.info_off
);
1378 xen_init_vga(info
, xen_start_info
->console
.dom0
.info_size
);
1379 xen_start_info
->console
.domU
.mfn
= 0;
1380 xen_start_info
->console
.domU
.evtchn
= 0;
1382 /* Make sure ACS will be enabled */
1385 xen_acpi_sleep_register();
1389 xen_raw_console_write("about to get started...\n");
1391 xen_setup_runstate_info(0);
1393 /* Start the world */
1394 #ifdef CONFIG_X86_32
1395 i386_start_kernel();
1397 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));
1401 static int init_hvm_pv_info(int *major
, int *minor
)
1403 uint32_t eax
, ebx
, ecx
, edx
, pages
, msr
, base
;
1406 base
= xen_cpuid_base();
1407 cpuid(base
+ 1, &eax
, &ebx
, &ecx
, &edx
);
1410 *minor
= eax
& 0xffff;
1411 printk(KERN_INFO
"Xen version %d.%d.\n", *major
, *minor
);
1413 cpuid(base
+ 2, &pages
, &msr
, &ecx
, &edx
);
1415 pfn
= __pa(hypercall_page
);
1416 wrmsr_safe(msr
, (u32
)pfn
, (u32
)(pfn
>> 32));
1418 xen_setup_features();
1420 pv_info
.name
= "Xen HVM";
1422 xen_domain_type
= XEN_HVM_DOMAIN
;
1427 void __ref
xen_hvm_init_shared_info(void)
1430 struct xen_add_to_physmap xatp
;
1431 static struct shared_info
*shared_info_page
= 0;
1433 if (!shared_info_page
)
1434 shared_info_page
= (struct shared_info
*)
1435 extend_brk(PAGE_SIZE
, PAGE_SIZE
);
1436 xatp
.domid
= DOMID_SELF
;
1438 xatp
.space
= XENMAPSPACE_shared_info
;
1439 xatp
.gpfn
= __pa(shared_info_page
) >> PAGE_SHIFT
;
1440 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap
, &xatp
))
1443 HYPERVISOR_shared_info
= (struct shared_info
*)shared_info_page
;
1445 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1446 * page, we use it in the event channel upcall and in some pvclock
1447 * related functions. We don't need the vcpu_info placement
1448 * optimizations because we don't use any pv_mmu or pv_irq op on
1450 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1451 * online but xen_hvm_init_shared_info is run at resume time too and
1452 * in that case multiple vcpus might be online. */
1453 for_each_online_cpu(cpu
) {
1454 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
1458 #ifdef CONFIG_XEN_PVHVM
1459 static int __cpuinit
xen_hvm_cpu_notify(struct notifier_block
*self
,
1460 unsigned long action
, void *hcpu
)
1462 int cpu
= (long)hcpu
;
1464 case CPU_UP_PREPARE
:
1465 xen_vcpu_setup(cpu
);
1466 if (xen_have_vector_callback
)
1467 xen_init_lock_cpu(cpu
);
1475 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata
= {
1476 .notifier_call
= xen_hvm_cpu_notify
,
1479 static void __init
xen_hvm_guest_init(void)
1484 r
= init_hvm_pv_info(&major
, &minor
);
1488 xen_hvm_init_shared_info();
1490 if (xen_feature(XENFEAT_hvm_callback_vector
))
1491 xen_have_vector_callback
= 1;
1493 register_cpu_notifier(&xen_hvm_cpu_notifier
);
1494 xen_unplug_emulated_devices();
1495 x86_init
.irqs
.intr_init
= xen_init_IRQ
;
1496 xen_hvm_init_time_ops();
1497 xen_hvm_init_mmu_ops();
1500 static bool __init
xen_hvm_platform(void)
1502 if (xen_pv_domain())
1505 if (!xen_cpuid_base())
1511 bool xen_hvm_need_lapic(void)
1513 if (xen_pv_domain())
1515 if (!xen_hvm_domain())
1517 if (xen_feature(XENFEAT_hvm_pirqs
) && xen_have_vector_callback
)
1521 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic
);
1523 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst
= {
1525 .detect
= xen_hvm_platform
,
1526 .init_platform
= xen_hvm_guest_init
,
1528 EXPORT_SYMBOL(x86_hyper_xen_hvm
);