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>
67 #include <asm/pci_x86.h>
70 #include <linux/acpi.h>
72 #include <acpi/pdc_intel.h>
73 #include <acpi/processor.h>
74 #include <xen/interface/platform.h>
80 #include "multicalls.h"
82 EXPORT_SYMBOL_GPL(hypercall_page
);
84 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
85 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
87 enum xen_domain_type xen_domain_type
= XEN_NATIVE
;
88 EXPORT_SYMBOL_GPL(xen_domain_type
);
90 unsigned long *machine_to_phys_mapping
= (void *)MACH2PHYS_VIRT_START
;
91 EXPORT_SYMBOL(machine_to_phys_mapping
);
92 unsigned long machine_to_phys_nr
;
93 EXPORT_SYMBOL(machine_to_phys_nr
);
95 struct start_info
*xen_start_info
;
96 EXPORT_SYMBOL_GPL(xen_start_info
);
98 struct shared_info xen_dummy_shared_info
;
100 void *xen_initial_gdt
;
102 RESERVE_BRK(shared_info_page_brk
, PAGE_SIZE
);
103 __read_mostly
int xen_have_vector_callback
;
104 EXPORT_SYMBOL_GPL(xen_have_vector_callback
);
107 * Point at some empty memory to start with. We map the real shared_info
108 * page as soon as fixmap is up and running.
110 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
113 * Flag to determine whether vcpu info placement is available on all
114 * VCPUs. We assume it is to start with, and then set it to zero on
115 * the first failure. This is because it can succeed on some VCPUs
116 * and not others, since it can involve hypervisor memory allocation,
117 * or because the guest failed to guarantee all the appropriate
118 * constraints on all VCPUs (ie buffer can't cross a page boundary).
120 * Note that any particular CPU may be using a placed vcpu structure,
121 * but we can only optimise if the all are.
123 * 0: not available, 1: available
125 static int have_vcpu_info_placement
= 1;
127 static void clamp_max_cpus(void)
130 if (setup_max_cpus
> MAX_VIRT_CPUS
)
131 setup_max_cpus
= MAX_VIRT_CPUS
;
135 static void xen_vcpu_setup(int cpu
)
137 struct vcpu_register_vcpu_info info
;
139 struct vcpu_info
*vcpup
;
141 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
143 if (cpu
< MAX_VIRT_CPUS
)
144 per_cpu(xen_vcpu
,cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
146 if (!have_vcpu_info_placement
) {
147 if (cpu
>= MAX_VIRT_CPUS
)
152 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
153 info
.mfn
= arbitrary_virt_to_mfn(vcpup
);
154 info
.offset
= offset_in_page(vcpup
);
156 /* Check to see if the hypervisor will put the vcpu_info
157 structure where we want it, which allows direct access via
158 a percpu-variable. */
159 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
162 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
163 have_vcpu_info_placement
= 0;
166 /* This cpu is using the registered vcpu info, even if
167 later ones fail to. */
168 per_cpu(xen_vcpu
, cpu
) = vcpup
;
173 * On restore, set the vcpu placement up again.
174 * If it fails, then we're in a bad state, since
175 * we can't back out from using it...
177 void xen_vcpu_restore(void)
181 for_each_online_cpu(cpu
) {
182 bool other_cpu
= (cpu
!= smp_processor_id());
185 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
188 xen_setup_runstate_info(cpu
);
190 if (have_vcpu_info_placement
)
194 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
199 static void __init
xen_banner(void)
201 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
202 struct xen_extraversion extra
;
203 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
205 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
207 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
208 version
>> 16, version
& 0xffff, extra
.extraversion
,
209 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
212 static __read_mostly
unsigned int cpuid_leaf1_edx_mask
= ~0;
213 static __read_mostly
unsigned int cpuid_leaf1_ecx_mask
= ~0;
215 static __read_mostly
unsigned int cpuid_leaf1_ecx_set_mask
;
216 static __read_mostly
unsigned int cpuid_leaf5_ecx_val
;
217 static __read_mostly
unsigned int cpuid_leaf5_edx_val
;
219 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
220 unsigned int *cx
, unsigned int *dx
)
222 unsigned maskebx
= ~0;
223 unsigned maskecx
= ~0;
224 unsigned maskedx
= ~0;
227 * Mask out inconvenient features, to try and disable as many
228 * unsupported kernel subsystems as possible.
232 maskecx
= cpuid_leaf1_ecx_mask
;
233 setecx
= cpuid_leaf1_ecx_set_mask
;
234 maskedx
= cpuid_leaf1_edx_mask
;
237 case CPUID_MWAIT_LEAF
:
238 /* Synthesize the values.. */
241 *cx
= cpuid_leaf5_ecx_val
;
242 *dx
= cpuid_leaf5_edx_val
;
246 /* Suppress extended topology stuff */
251 asm(XEN_EMULATE_PREFIX
"cpuid"
256 : "0" (*ax
), "2" (*cx
));
265 static bool __init
xen_check_mwait(void)
267 #if defined(CONFIG_ACPI) && !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR) && \
268 !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR_MODULE)
269 struct xen_platform_op op
= {
270 .cmd
= XENPF_set_processor_pminfo
,
271 .u
.set_pminfo
.id
= -1,
272 .u
.set_pminfo
.type
= XEN_PM_PDC
,
275 unsigned int ax
, bx
, cx
, dx
;
276 unsigned int mwait_mask
;
278 /* We need to determine whether it is OK to expose the MWAIT
279 * capability to the kernel to harvest deeper than C3 states from ACPI
280 * _CST using the processor_harvest_xen.c module. For this to work, we
281 * need to gather the MWAIT_LEAF values (which the cstate.c code
282 * checks against). The hypervisor won't expose the MWAIT flag because
283 * it would break backwards compatibility; so we will find out directly
284 * from the hardware and hypercall.
286 if (!xen_initial_domain())
292 native_cpuid(&ax
, &bx
, &cx
, &dx
);
294 mwait_mask
= (1 << (X86_FEATURE_EST
% 32)) |
295 (1 << (X86_FEATURE_MWAIT
% 32));
297 if ((cx
& mwait_mask
) != mwait_mask
)
300 /* We need to emulate the MWAIT_LEAF and for that we need both
301 * ecx and edx. The hypercall provides only partial information.
304 ax
= CPUID_MWAIT_LEAF
;
309 native_cpuid(&ax
, &bx
, &cx
, &dx
);
311 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
312 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
314 buf
[0] = ACPI_PDC_REVISION_ID
;
316 buf
[2] = (ACPI_PDC_C_CAPABILITY_SMP
| ACPI_PDC_EST_CAPABILITY_SWSMP
);
318 set_xen_guest_handle(op
.u
.set_pminfo
.pdc
, buf
);
320 if ((HYPERVISOR_dom0_op(&op
) == 0) &&
321 (buf
[2] & (ACPI_PDC_C_C1_FFH
| ACPI_PDC_C_C2C3_FFH
))) {
322 cpuid_leaf5_ecx_val
= cx
;
323 cpuid_leaf5_edx_val
= dx
;
330 static void __init
xen_init_cpuid_mask(void)
332 unsigned int ax
, bx
, cx
, dx
;
333 unsigned int xsave_mask
;
335 cpuid_leaf1_edx_mask
=
336 ~((1 << X86_FEATURE_MCE
) | /* disable MCE */
337 (1 << X86_FEATURE_MCA
) | /* disable MCA */
338 (1 << X86_FEATURE_MTRR
) | /* disable MTRR */
339 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
341 if (!xen_initial_domain())
342 cpuid_leaf1_edx_mask
&=
343 ~((1 << X86_FEATURE_APIC
) | /* disable local APIC */
344 (1 << X86_FEATURE_ACPI
)); /* disable ACPI */
347 xen_cpuid(&ax
, &bx
, &cx
, &dx
);
350 (1 << (X86_FEATURE_XSAVE
% 32)) |
351 (1 << (X86_FEATURE_OSXSAVE
% 32));
353 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
354 if ((cx
& xsave_mask
) != xsave_mask
)
355 cpuid_leaf1_ecx_mask
&= ~xsave_mask
; /* disable XSAVE & OSXSAVE */
356 if (xen_check_mwait())
357 cpuid_leaf1_ecx_set_mask
= (1 << (X86_FEATURE_MWAIT
% 32));
360 static void xen_set_debugreg(int reg
, unsigned long val
)
362 HYPERVISOR_set_debugreg(reg
, val
);
365 static unsigned long xen_get_debugreg(int reg
)
367 return HYPERVISOR_get_debugreg(reg
);
370 static void xen_end_context_switch(struct task_struct
*next
)
373 paravirt_end_context_switch(next
);
376 static unsigned long xen_store_tr(void)
382 * Set the page permissions for a particular virtual address. If the
383 * address is a vmalloc mapping (or other non-linear mapping), then
384 * find the linear mapping of the page and also set its protections to
387 static void set_aliased_prot(void *v
, pgprot_t prot
)
395 ptep
= lookup_address((unsigned long)v
, &level
);
396 BUG_ON(ptep
== NULL
);
398 pfn
= pte_pfn(*ptep
);
399 page
= pfn_to_page(pfn
);
401 pte
= pfn_pte(pfn
, prot
);
403 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
406 if (!PageHighMem(page
)) {
407 void *av
= __va(PFN_PHYS(pfn
));
410 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
416 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
418 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
421 for(i
= 0; i
< entries
; i
+= entries_per_page
)
422 set_aliased_prot(ldt
+ i
, PAGE_KERNEL_RO
);
425 static void xen_free_ldt(struct desc_struct
*ldt
, unsigned entries
)
427 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
430 for(i
= 0; i
< entries
; i
+= entries_per_page
)
431 set_aliased_prot(ldt
+ i
, PAGE_KERNEL
);
434 static void xen_set_ldt(const void *addr
, unsigned entries
)
436 struct mmuext_op
*op
;
437 struct multicall_space mcs
= xen_mc_entry(sizeof(*op
));
439 trace_xen_cpu_set_ldt(addr
, entries
);
442 op
->cmd
= MMUEXT_SET_LDT
;
443 op
->arg1
.linear_addr
= (unsigned long)addr
;
444 op
->arg2
.nr_ents
= entries
;
446 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
448 xen_mc_issue(PARAVIRT_LAZY_CPU
);
451 static void xen_load_gdt(const struct desc_ptr
*dtr
)
453 unsigned long va
= dtr
->address
;
454 unsigned int size
= dtr
->size
+ 1;
455 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
456 unsigned long frames
[pages
];
460 * A GDT can be up to 64k in size, which corresponds to 8192
461 * 8-byte entries, or 16 4k pages..
464 BUG_ON(size
> 65536);
465 BUG_ON(va
& ~PAGE_MASK
);
467 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
470 unsigned long pfn
, mfn
;
474 * The GDT is per-cpu and is in the percpu data area.
475 * That can be virtually mapped, so we need to do a
476 * page-walk to get the underlying MFN for the
477 * hypercall. The page can also be in the kernel's
478 * linear range, so we need to RO that mapping too.
480 ptep
= lookup_address(va
, &level
);
481 BUG_ON(ptep
== NULL
);
483 pfn
= pte_pfn(*ptep
);
484 mfn
= pfn_to_mfn(pfn
);
485 virt
= __va(PFN_PHYS(pfn
));
489 make_lowmem_page_readonly((void *)va
);
490 make_lowmem_page_readonly(virt
);
493 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
498 * load_gdt for early boot, when the gdt is only mapped once
500 static void __init
xen_load_gdt_boot(const struct desc_ptr
*dtr
)
502 unsigned long va
= dtr
->address
;
503 unsigned int size
= dtr
->size
+ 1;
504 unsigned pages
= (size
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
505 unsigned long frames
[pages
];
509 * A GDT can be up to 64k in size, which corresponds to 8192
510 * 8-byte entries, or 16 4k pages..
513 BUG_ON(size
> 65536);
514 BUG_ON(va
& ~PAGE_MASK
);
516 for (f
= 0; va
< dtr
->address
+ size
; va
+= PAGE_SIZE
, f
++) {
518 unsigned long pfn
, mfn
;
520 pfn
= virt_to_pfn(va
);
521 mfn
= pfn_to_mfn(pfn
);
523 pte
= pfn_pte(pfn
, PAGE_KERNEL_RO
);
525 if (HYPERVISOR_update_va_mapping((unsigned long)va
, pte
, 0))
531 if (HYPERVISOR_set_gdt(frames
, size
/ sizeof(struct desc_struct
)))
535 static void load_TLS_descriptor(struct thread_struct
*t
,
536 unsigned int cpu
, unsigned int i
)
538 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
539 xmaddr_t maddr
= arbitrary_virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
540 struct multicall_space mc
= __xen_mc_entry(0);
542 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
545 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
548 * XXX sleazy hack: If we're being called in a lazy-cpu zone
549 * and lazy gs handling is enabled, it means we're in a
550 * context switch, and %gs has just been saved. This means we
551 * can zero it out to prevent faults on exit from the
552 * hypervisor if the next process has no %gs. Either way, it
553 * has been saved, and the new value will get loaded properly.
554 * This will go away as soon as Xen has been modified to not
555 * save/restore %gs for normal hypercalls.
557 * On x86_64, this hack is not used for %gs, because gs points
558 * to KERNEL_GS_BASE (and uses it for PDA references), so we
559 * must not zero %gs on x86_64
561 * For x86_64, we need to zero %fs, otherwise we may get an
562 * exception between the new %fs descriptor being loaded and
563 * %fs being effectively cleared at __switch_to().
565 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
575 load_TLS_descriptor(t
, cpu
, 0);
576 load_TLS_descriptor(t
, cpu
, 1);
577 load_TLS_descriptor(t
, cpu
, 2);
579 xen_mc_issue(PARAVIRT_LAZY_CPU
);
583 static void xen_load_gs_index(unsigned int idx
)
585 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
590 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
593 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
594 u64 entry
= *(u64
*)ptr
;
596 trace_xen_cpu_write_ldt_entry(dt
, entrynum
, entry
);
601 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
607 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
608 struct trap_info
*info
)
612 if (val
->type
!= GATE_TRAP
&& val
->type
!= GATE_INTERRUPT
)
615 info
->vector
= vector
;
617 addr
= gate_offset(*val
);
620 * Look for known traps using IST, and substitute them
621 * appropriately. The debugger ones are the only ones we care
622 * about. Xen will handle faults like double_fault and
623 * machine_check, so we should never see them. Warn if
624 * there's an unexpected IST-using fault handler.
626 if (addr
== (unsigned long)debug
)
627 addr
= (unsigned long)xen_debug
;
628 else if (addr
== (unsigned long)int3
)
629 addr
= (unsigned long)xen_int3
;
630 else if (addr
== (unsigned long)stack_segment
)
631 addr
= (unsigned long)xen_stack_segment
;
632 else if (addr
== (unsigned long)double_fault
||
633 addr
== (unsigned long)nmi
) {
634 /* Don't need to handle these */
636 #ifdef CONFIG_X86_MCE
637 } else if (addr
== (unsigned long)machine_check
) {
641 /* Some other trap using IST? */
642 if (WARN_ON(val
->ist
!= 0))
645 #endif /* CONFIG_X86_64 */
646 info
->address
= addr
;
648 info
->cs
= gate_segment(*val
);
649 info
->flags
= val
->dpl
;
650 /* interrupt gates clear IF */
651 if (val
->type
== GATE_INTERRUPT
)
652 info
->flags
|= 1 << 2;
657 /* Locations of each CPU's IDT */
658 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
660 /* Set an IDT entry. If the entry is part of the current IDT, then
662 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
664 unsigned long p
= (unsigned long)&dt
[entrynum
];
665 unsigned long start
, end
;
667 trace_xen_cpu_write_idt_entry(dt
, entrynum
, g
);
671 start
= __this_cpu_read(idt_desc
.address
);
672 end
= start
+ __this_cpu_read(idt_desc
.size
) + 1;
676 native_write_idt_entry(dt
, entrynum
, g
);
678 if (p
>= start
&& (p
+ 8) <= end
) {
679 struct trap_info info
[2];
683 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
684 if (HYPERVISOR_set_trap_table(info
))
691 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
692 struct trap_info
*traps
)
694 unsigned in
, out
, count
;
696 count
= (desc
->size
+1) / sizeof(gate_desc
);
699 for (in
= out
= 0; in
< count
; in
++) {
700 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
702 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
705 traps
[out
].address
= 0;
708 void xen_copy_trap_info(struct trap_info
*traps
)
710 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
712 xen_convert_trap_info(desc
, traps
);
715 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
716 hold a spinlock to protect the static traps[] array (static because
717 it avoids allocation, and saves stack space). */
718 static void xen_load_idt(const struct desc_ptr
*desc
)
720 static DEFINE_SPINLOCK(lock
);
721 static struct trap_info traps
[257];
723 trace_xen_cpu_load_idt(desc
);
727 __get_cpu_var(idt_desc
) = *desc
;
729 xen_convert_trap_info(desc
, traps
);
732 if (HYPERVISOR_set_trap_table(traps
))
738 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
739 they're handled differently. */
740 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
741 const void *desc
, int type
)
743 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
754 xmaddr_t maddr
= arbitrary_virt_to_machine(&dt
[entry
]);
757 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
767 * Version of write_gdt_entry for use at early boot-time needed to
768 * update an entry as simply as possible.
770 static void __init
xen_write_gdt_entry_boot(struct desc_struct
*dt
, int entry
,
771 const void *desc
, int type
)
773 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
782 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
784 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
785 dt
[entry
] = *(struct desc_struct
*)desc
;
791 static void xen_load_sp0(struct tss_struct
*tss
,
792 struct thread_struct
*thread
)
794 struct multicall_space mcs
;
796 mcs
= xen_mc_entry(0);
797 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
798 xen_mc_issue(PARAVIRT_LAZY_CPU
);
801 static void xen_set_iopl_mask(unsigned mask
)
803 struct physdev_set_iopl set_iopl
;
805 /* Force the change at ring 0. */
806 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
807 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
810 static void xen_io_delay(void)
814 #ifdef CONFIG_X86_LOCAL_APIC
815 static unsigned long xen_set_apic_id(unsigned int x
)
820 static unsigned int xen_get_apic_id(unsigned long x
)
822 return ((x
)>>24) & 0xFFu
;
824 static u32
xen_apic_read(u32 reg
)
826 struct xen_platform_op op
= {
827 .cmd
= XENPF_get_cpuinfo
,
828 .interface_version
= XENPF_INTERFACE_VERSION
,
829 .u
.pcpu_info
.xen_cpuid
= 0,
833 /* Shouldn't need this as APIC is turned off for PV, and we only
834 * get called on the bootup processor. But just in case. */
835 if (!xen_initial_domain() || smp_processor_id())
844 ret
= HYPERVISOR_dom0_op(&op
);
848 return op
.u
.pcpu_info
.apic_id
<< 24;
851 static void xen_apic_write(u32 reg
, u32 val
)
853 /* Warn to see if there's any stray references */
857 static u64
xen_apic_icr_read(void)
862 static void xen_apic_icr_write(u32 low
, u32 id
)
864 /* Warn to see if there's any stray references */
868 static void xen_apic_wait_icr_idle(void)
873 static u32
xen_safe_apic_wait_icr_idle(void)
878 static void set_xen_basic_apic_ops(void)
880 apic
->read
= xen_apic_read
;
881 apic
->write
= xen_apic_write
;
882 apic
->icr_read
= xen_apic_icr_read
;
883 apic
->icr_write
= xen_apic_icr_write
;
884 apic
->wait_icr_idle
= xen_apic_wait_icr_idle
;
885 apic
->safe_wait_icr_idle
= xen_safe_apic_wait_icr_idle
;
886 apic
->set_apic_id
= xen_set_apic_id
;
887 apic
->get_apic_id
= xen_get_apic_id
;
890 apic
->send_IPI_allbutself
= xen_send_IPI_allbutself
;
891 apic
->send_IPI_mask_allbutself
= xen_send_IPI_mask_allbutself
;
892 apic
->send_IPI_mask
= xen_send_IPI_mask
;
893 apic
->send_IPI_all
= xen_send_IPI_all
;
894 apic
->send_IPI_self
= xen_send_IPI_self
;
900 static void xen_clts(void)
902 struct multicall_space mcs
;
904 mcs
= xen_mc_entry(0);
906 MULTI_fpu_taskswitch(mcs
.mc
, 0);
908 xen_mc_issue(PARAVIRT_LAZY_CPU
);
911 static DEFINE_PER_CPU(unsigned long, xen_cr0_value
);
913 static unsigned long xen_read_cr0(void)
915 unsigned long cr0
= this_cpu_read(xen_cr0_value
);
917 if (unlikely(cr0
== 0)) {
918 cr0
= native_read_cr0();
919 this_cpu_write(xen_cr0_value
, cr0
);
925 static void xen_write_cr0(unsigned long cr0
)
927 struct multicall_space mcs
;
929 this_cpu_write(xen_cr0_value
, cr0
);
931 /* Only pay attention to cr0.TS; everything else is
933 mcs
= xen_mc_entry(0);
935 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
937 xen_mc_issue(PARAVIRT_LAZY_CPU
);
940 static void xen_write_cr4(unsigned long cr4
)
945 native_write_cr4(cr4
);
948 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
959 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
960 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
961 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
964 base
= ((u64
)high
<< 32) | low
;
965 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
973 case MSR_SYSCALL_MASK
:
974 case MSR_IA32_SYSENTER_CS
:
975 case MSR_IA32_SYSENTER_ESP
:
976 case MSR_IA32_SYSENTER_EIP
:
977 /* Fast syscall setup is all done in hypercalls, so
978 these are all ignored. Stub them out here to stop
979 Xen console noise. */
982 case MSR_IA32_CR_PAT
:
983 if (smp_processor_id() == 0)
984 xen_set_pat(((u64
)high
<< 32) | low
);
988 ret
= native_write_msr_safe(msr
, low
, high
);
994 void xen_setup_shared_info(void)
996 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
997 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
998 xen_start_info
->shared_info
);
1000 HYPERVISOR_shared_info
=
1001 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
1003 HYPERVISOR_shared_info
=
1004 (struct shared_info
*)__va(xen_start_info
->shared_info
);
1007 /* In UP this is as good a place as any to set up shared info */
1008 xen_setup_vcpu_info_placement();
1011 xen_setup_mfn_list_list();
1014 /* This is called once we have the cpu_possible_mask */
1015 void xen_setup_vcpu_info_placement(void)
1019 for_each_possible_cpu(cpu
)
1020 xen_vcpu_setup(cpu
);
1022 /* xen_vcpu_setup managed to place the vcpu_info within the
1023 percpu area for all cpus, so make use of it */
1024 if (have_vcpu_info_placement
) {
1025 pv_irq_ops
.save_fl
= __PV_IS_CALLEE_SAVE(xen_save_fl_direct
);
1026 pv_irq_ops
.restore_fl
= __PV_IS_CALLEE_SAVE(xen_restore_fl_direct
);
1027 pv_irq_ops
.irq_disable
= __PV_IS_CALLEE_SAVE(xen_irq_disable_direct
);
1028 pv_irq_ops
.irq_enable
= __PV_IS_CALLEE_SAVE(xen_irq_enable_direct
);
1029 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1033 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1034 unsigned long addr
, unsigned len
)
1036 char *start
, *end
, *reloc
;
1039 start
= end
= reloc
= NULL
;
1041 #define SITE(op, x) \
1042 case PARAVIRT_PATCH(op.x): \
1043 if (have_vcpu_info_placement) { \
1044 start = (char *)xen_##x##_direct; \
1045 end = xen_##x##_direct_end; \
1046 reloc = xen_##x##_direct_reloc; \
1051 SITE(pv_irq_ops
, irq_enable
);
1052 SITE(pv_irq_ops
, irq_disable
);
1053 SITE(pv_irq_ops
, save_fl
);
1054 SITE(pv_irq_ops
, restore_fl
);
1058 if (start
== NULL
|| (end
-start
) > len
)
1061 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1063 /* Note: because reloc is assigned from something that
1064 appears to be an array, gcc assumes it's non-null,
1065 but doesn't know its relationship with start and
1067 if (reloc
> start
&& reloc
< end
) {
1068 int reloc_off
= reloc
- start
;
1069 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1070 long delta
= start
- (char *)addr
;
1078 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1086 static const struct pv_info xen_info __initconst
= {
1087 .paravirt_enabled
= 1,
1088 .shared_kernel_pmd
= 0,
1090 #ifdef CONFIG_X86_64
1091 .extra_user_64bit_cs
= FLAT_USER_CS64
,
1097 static const struct pv_init_ops xen_init_ops __initconst
= {
1101 static const struct pv_cpu_ops xen_cpu_ops __initconst
= {
1104 .set_debugreg
= xen_set_debugreg
,
1105 .get_debugreg
= xen_get_debugreg
,
1109 .read_cr0
= xen_read_cr0
,
1110 .write_cr0
= xen_write_cr0
,
1112 .read_cr4
= native_read_cr4
,
1113 .read_cr4_safe
= native_read_cr4_safe
,
1114 .write_cr4
= xen_write_cr4
,
1116 .wbinvd
= native_wbinvd
,
1118 .read_msr
= native_read_msr_safe
,
1119 .write_msr
= xen_write_msr_safe
,
1121 .read_tsc
= native_read_tsc
,
1122 .read_pmc
= native_read_pmc
,
1125 .irq_enable_sysexit
= xen_sysexit
,
1126 #ifdef CONFIG_X86_64
1127 .usergs_sysret32
= xen_sysret32
,
1128 .usergs_sysret64
= xen_sysret64
,
1131 .load_tr_desc
= paravirt_nop
,
1132 .set_ldt
= xen_set_ldt
,
1133 .load_gdt
= xen_load_gdt
,
1134 .load_idt
= xen_load_idt
,
1135 .load_tls
= xen_load_tls
,
1136 #ifdef CONFIG_X86_64
1137 .load_gs_index
= xen_load_gs_index
,
1140 .alloc_ldt
= xen_alloc_ldt
,
1141 .free_ldt
= xen_free_ldt
,
1143 .store_gdt
= native_store_gdt
,
1144 .store_idt
= native_store_idt
,
1145 .store_tr
= xen_store_tr
,
1147 .write_ldt_entry
= xen_write_ldt_entry
,
1148 .write_gdt_entry
= xen_write_gdt_entry
,
1149 .write_idt_entry
= xen_write_idt_entry
,
1150 .load_sp0
= xen_load_sp0
,
1152 .set_iopl_mask
= xen_set_iopl_mask
,
1153 .io_delay
= xen_io_delay
,
1155 /* Xen takes care of %gs when switching to usermode for us */
1156 .swapgs
= paravirt_nop
,
1158 .start_context_switch
= paravirt_start_context_switch
,
1159 .end_context_switch
= xen_end_context_switch
,
1162 static const struct pv_apic_ops xen_apic_ops __initconst
= {
1163 #ifdef CONFIG_X86_LOCAL_APIC
1164 .startup_ipi_hook
= paravirt_nop
,
1168 static void xen_reboot(int reason
)
1170 struct sched_shutdown r
= { .reason
= reason
};
1172 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1176 static void xen_restart(char *msg
)
1178 xen_reboot(SHUTDOWN_reboot
);
1181 static void xen_emergency_restart(void)
1183 xen_reboot(SHUTDOWN_reboot
);
1186 static void xen_machine_halt(void)
1188 xen_reboot(SHUTDOWN_poweroff
);
1191 static void xen_machine_power_off(void)
1195 xen_reboot(SHUTDOWN_poweroff
);
1198 static void xen_crash_shutdown(struct pt_regs
*regs
)
1200 xen_reboot(SHUTDOWN_crash
);
1204 xen_panic_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
1206 xen_reboot(SHUTDOWN_crash
);
1210 static struct notifier_block xen_panic_block
= {
1211 .notifier_call
= xen_panic_event
,
1214 int xen_panic_handler_init(void)
1216 atomic_notifier_chain_register(&panic_notifier_list
, &xen_panic_block
);
1220 static const struct machine_ops xen_machine_ops __initconst
= {
1221 .restart
= xen_restart
,
1222 .halt
= xen_machine_halt
,
1223 .power_off
= xen_machine_power_off
,
1224 .shutdown
= xen_machine_halt
,
1225 .crash_shutdown
= xen_crash_shutdown
,
1226 .emergency_restart
= xen_emergency_restart
,
1230 * Set up the GDT and segment registers for -fstack-protector. Until
1231 * we do this, we have to be careful not to call any stack-protected
1232 * function, which is most of the kernel.
1234 static void __init
xen_setup_stackprotector(void)
1236 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry_boot
;
1237 pv_cpu_ops
.load_gdt
= xen_load_gdt_boot
;
1239 setup_stack_canary_segment(0);
1240 switch_to_new_gdt(0);
1242 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry
;
1243 pv_cpu_ops
.load_gdt
= xen_load_gdt
;
1246 /* First C function to be called on Xen boot */
1247 asmlinkage
void __init
xen_start_kernel(void)
1249 struct physdev_set_iopl set_iopl
;
1253 if (!xen_start_info
)
1256 xen_domain_type
= XEN_PV_DOMAIN
;
1258 xen_setup_machphys_mapping();
1260 /* Install Xen paravirt ops */
1262 pv_init_ops
= xen_init_ops
;
1263 pv_cpu_ops
= xen_cpu_ops
;
1264 pv_apic_ops
= xen_apic_ops
;
1266 x86_init
.resources
.memory_setup
= xen_memory_setup
;
1267 x86_init
.oem
.arch_setup
= xen_arch_setup
;
1268 x86_init
.oem
.banner
= xen_banner
;
1270 xen_init_time_ops();
1273 * Set up some pagetable state before starting to set any ptes.
1278 /* Prevent unwanted bits from being set in PTEs. */
1279 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1281 if (!xen_initial_domain())
1283 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1285 __supported_pte_mask
|= _PAGE_IOMAP
;
1288 * Prevent page tables from being allocated in highmem, even
1289 * if CONFIG_HIGHPTE is enabled.
1291 __userpte_alloc_gfp
&= ~__GFP_HIGHMEM
;
1293 /* Work out if we support NX */
1296 xen_setup_features();
1299 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1300 xen_build_dynamic_phys_to_machine();
1303 * Set up kernel GDT and segment registers, mainly so that
1304 * -fstack-protector code can be executed.
1306 xen_setup_stackprotector();
1309 xen_init_cpuid_mask();
1311 #ifdef CONFIG_X86_LOCAL_APIC
1313 * set up the basic apic ops.
1315 set_xen_basic_apic_ops();
1318 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1319 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1320 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1323 machine_ops
= xen_machine_ops
;
1326 * The only reliable way to retain the initial address of the
1327 * percpu gdt_page is to remember it here, so we can go and
1328 * mark it RW later, when the initial percpu area is freed.
1330 xen_initial_gdt
= &per_cpu(gdt_page
, 0);
1334 #ifdef CONFIG_ACPI_NUMA
1336 * The pages we from Xen are not related to machine pages, so
1337 * any NUMA information the kernel tries to get from ACPI will
1338 * be meaningless. Prevent it from trying.
1343 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1345 /* Don't do the full vcpu_info placement stuff until we have a
1346 possible map and a non-dummy shared_info. */
1347 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1349 local_irq_disable();
1350 early_boot_irqs_disabled
= true;
1352 xen_raw_console_write("mapping kernel into physical memory\n");
1353 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
1355 /* Allocate and initialize top and mid mfn levels for p2m structure */
1356 xen_build_mfn_list_list();
1358 /* keep using Xen gdt for now; no urgent need to change it */
1360 #ifdef CONFIG_X86_32
1361 pv_info
.kernel_rpl
= 1;
1362 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1363 pv_info
.kernel_rpl
= 0;
1365 pv_info
.kernel_rpl
= 0;
1367 /* set the limit of our address space */
1370 /* We used to do this in xen_arch_setup, but that is too late on AMD
1371 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1372 * which pokes 0xcf8 port.
1375 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
1377 xen_raw_printk("physdev_op failed %d\n", rc
);
1379 #ifdef CONFIG_X86_32
1380 /* set up basic CPUID stuff */
1381 cpu_detect(&new_cpu_data
);
1382 new_cpu_data
.hard_math
= 1;
1383 new_cpu_data
.wp_works_ok
= 1;
1384 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1387 /* Poke various useful things into boot_params */
1388 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1389 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1390 ? __pa(xen_start_info
->mod_start
) : 0;
1391 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1392 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1394 if (!xen_initial_domain()) {
1395 add_preferred_console("xenboot", 0, NULL
);
1396 add_preferred_console("tty", 0, NULL
);
1397 add_preferred_console("hvc", 0, NULL
);
1399 x86_init
.pci
.arch_init
= pci_xen_init
;
1401 const struct dom0_vga_console_info
*info
=
1402 (void *)((char *)xen_start_info
+
1403 xen_start_info
->console
.dom0
.info_off
);
1405 xen_init_vga(info
, xen_start_info
->console
.dom0
.info_size
);
1406 xen_start_info
->console
.domU
.mfn
= 0;
1407 xen_start_info
->console
.domU
.evtchn
= 0;
1411 /* Make sure ACS will be enabled */
1414 xen_acpi_sleep_register();
1417 /* PCI BIOS service won't work from a PV guest. */
1418 pci_probe
&= ~PCI_PROBE_BIOS
;
1420 xen_raw_console_write("about to get started...\n");
1422 xen_setup_runstate_info(0);
1424 /* Start the world */
1425 #ifdef CONFIG_X86_32
1426 i386_start_kernel();
1428 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));
1432 static int init_hvm_pv_info(int *major
, int *minor
)
1434 uint32_t eax
, ebx
, ecx
, edx
, pages
, msr
, base
;
1437 base
= xen_cpuid_base();
1438 cpuid(base
+ 1, &eax
, &ebx
, &ecx
, &edx
);
1441 *minor
= eax
& 0xffff;
1442 printk(KERN_INFO
"Xen version %d.%d.\n", *major
, *minor
);
1444 cpuid(base
+ 2, &pages
, &msr
, &ecx
, &edx
);
1446 pfn
= __pa(hypercall_page
);
1447 wrmsr_safe(msr
, (u32
)pfn
, (u32
)(pfn
>> 32));
1449 xen_setup_features();
1451 pv_info
.name
= "Xen HVM";
1453 xen_domain_type
= XEN_HVM_DOMAIN
;
1458 void __ref
xen_hvm_init_shared_info(void)
1461 struct xen_add_to_physmap xatp
;
1462 static struct shared_info
*shared_info_page
= 0;
1464 if (!shared_info_page
)
1465 shared_info_page
= (struct shared_info
*)
1466 extend_brk(PAGE_SIZE
, PAGE_SIZE
);
1467 xatp
.domid
= DOMID_SELF
;
1469 xatp
.space
= XENMAPSPACE_shared_info
;
1470 xatp
.gpfn
= __pa(shared_info_page
) >> PAGE_SHIFT
;
1471 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap
, &xatp
))
1474 HYPERVISOR_shared_info
= (struct shared_info
*)shared_info_page
;
1476 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1477 * page, we use it in the event channel upcall and in some pvclock
1478 * related functions. We don't need the vcpu_info placement
1479 * optimizations because we don't use any pv_mmu or pv_irq op on
1481 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1482 * online but xen_hvm_init_shared_info is run at resume time too and
1483 * in that case multiple vcpus might be online. */
1484 for_each_online_cpu(cpu
) {
1485 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
1489 #ifdef CONFIG_XEN_PVHVM
1490 static int __cpuinit
xen_hvm_cpu_notify(struct notifier_block
*self
,
1491 unsigned long action
, void *hcpu
)
1493 int cpu
= (long)hcpu
;
1495 case CPU_UP_PREPARE
:
1496 xen_vcpu_setup(cpu
);
1497 if (xen_have_vector_callback
)
1498 xen_init_lock_cpu(cpu
);
1506 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata
= {
1507 .notifier_call
= xen_hvm_cpu_notify
,
1510 static void __init
xen_hvm_guest_init(void)
1515 r
= init_hvm_pv_info(&major
, &minor
);
1519 xen_hvm_init_shared_info();
1521 if (xen_feature(XENFEAT_hvm_callback_vector
))
1522 xen_have_vector_callback
= 1;
1524 register_cpu_notifier(&xen_hvm_cpu_notifier
);
1525 xen_unplug_emulated_devices();
1526 x86_init
.irqs
.intr_init
= xen_init_IRQ
;
1527 xen_hvm_init_time_ops();
1528 xen_hvm_init_mmu_ops();
1531 static bool __init
xen_hvm_platform(void)
1533 if (xen_pv_domain())
1536 if (!xen_cpuid_base())
1542 bool xen_hvm_need_lapic(void)
1544 if (xen_pv_domain())
1546 if (!xen_hvm_domain())
1548 if (xen_feature(XENFEAT_hvm_pirqs
) && xen_have_vector_callback
)
1552 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic
);
1554 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst
= {
1556 .detect
= xen_hvm_platform
,
1557 .init_platform
= xen_hvm_guest_init
,
1559 EXPORT_SYMBOL(x86_hyper_xen_hvm
);