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/interface/xen-mca.h>
42 #include <xen/features.h>
45 #include <xen/hvc-console.h>
48 #include <asm/paravirt.h>
51 #include <asm/xen/pci.h>
52 #include <asm/xen/hypercall.h>
53 #include <asm/xen/hypervisor.h>
54 #include <asm/fixmap.h>
55 #include <asm/processor.h>
56 #include <asm/proto.h>
57 #include <asm/msr-index.h>
58 #include <asm/traps.h>
59 #include <asm/setup.h>
61 #include <asm/pgalloc.h>
62 #include <asm/pgtable.h>
63 #include <asm/tlbflush.h>
64 #include <asm/reboot.h>
65 #include <asm/stackprotector.h>
66 #include <asm/hypervisor.h>
67 #include <asm/mwait.h>
68 #include <asm/pci_x86.h>
71 #include <linux/acpi.h>
73 #include <acpi/pdc_intel.h>
74 #include <acpi/processor.h>
75 #include <xen/interface/platform.h>
81 #include "multicalls.h"
83 EXPORT_SYMBOL_GPL(hypercall_page
);
85 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
86 DEFINE_PER_CPU(struct vcpu_info
, xen_vcpu_info
);
88 enum xen_domain_type xen_domain_type
= XEN_NATIVE
;
89 EXPORT_SYMBOL_GPL(xen_domain_type
);
91 unsigned long *machine_to_phys_mapping
= (void *)MACH2PHYS_VIRT_START
;
92 EXPORT_SYMBOL(machine_to_phys_mapping
);
93 unsigned long machine_to_phys_nr
;
94 EXPORT_SYMBOL(machine_to_phys_nr
);
96 struct start_info
*xen_start_info
;
97 EXPORT_SYMBOL_GPL(xen_start_info
);
99 struct shared_info xen_dummy_shared_info
;
101 void *xen_initial_gdt
;
103 RESERVE_BRK(shared_info_page_brk
, PAGE_SIZE
);
104 __read_mostly
int xen_have_vector_callback
;
105 EXPORT_SYMBOL_GPL(xen_have_vector_callback
);
108 * Point at some empty memory to start with. We map the real shared_info
109 * page as soon as fixmap is up and running.
111 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
114 * Flag to determine whether vcpu info placement is available on all
115 * VCPUs. We assume it is to start with, and then set it to zero on
116 * the first failure. This is because it can succeed on some VCPUs
117 * and not others, since it can involve hypervisor memory allocation,
118 * or because the guest failed to guarantee all the appropriate
119 * constraints on all VCPUs (ie buffer can't cross a page boundary).
121 * Note that any particular CPU may be using a placed vcpu structure,
122 * but we can only optimise if the all are.
124 * 0: not available, 1: available
126 static int have_vcpu_info_placement
= 1;
128 static void clamp_max_cpus(void)
131 if (setup_max_cpus
> MAX_VIRT_CPUS
)
132 setup_max_cpus
= MAX_VIRT_CPUS
;
136 static void xen_vcpu_setup(int cpu
)
138 struct vcpu_register_vcpu_info info
;
140 struct vcpu_info
*vcpup
;
142 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
144 if (cpu
< MAX_VIRT_CPUS
)
145 per_cpu(xen_vcpu
,cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
147 if (!have_vcpu_info_placement
) {
148 if (cpu
>= MAX_VIRT_CPUS
)
153 vcpup
= &per_cpu(xen_vcpu_info
, cpu
);
154 info
.mfn
= arbitrary_virt_to_mfn(vcpup
);
155 info
.offset
= offset_in_page(vcpup
);
157 /* Check to see if the hypervisor will put the vcpu_info
158 structure where we want it, which allows direct access via
159 a percpu-variable. */
160 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, cpu
, &info
);
163 printk(KERN_DEBUG
"register_vcpu_info failed: err=%d\n", err
);
164 have_vcpu_info_placement
= 0;
167 /* This cpu is using the registered vcpu info, even if
168 later ones fail to. */
169 per_cpu(xen_vcpu
, cpu
) = vcpup
;
174 * On restore, set the vcpu placement up again.
175 * If it fails, then we're in a bad state, since
176 * we can't back out from using it...
178 void xen_vcpu_restore(void)
182 for_each_online_cpu(cpu
) {
183 bool other_cpu
= (cpu
!= smp_processor_id());
186 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
))
189 xen_setup_runstate_info(cpu
);
191 if (have_vcpu_info_placement
)
195 HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
))
200 static void __init
xen_banner(void)
202 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
203 struct xen_extraversion extra
;
204 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
206 printk(KERN_INFO
"Booting paravirtualized kernel on %s\n",
208 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
209 version
>> 16, version
& 0xffff, extra
.extraversion
,
210 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
213 #define CPUID_THERM_POWER_LEAF 6
214 #define APERFMPERF_PRESENT 0
216 static __read_mostly
unsigned int cpuid_leaf1_edx_mask
= ~0;
217 static __read_mostly
unsigned int cpuid_leaf1_ecx_mask
= ~0;
219 static __read_mostly
unsigned int cpuid_leaf1_ecx_set_mask
;
220 static __read_mostly
unsigned int cpuid_leaf5_ecx_val
;
221 static __read_mostly
unsigned int cpuid_leaf5_edx_val
;
223 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
224 unsigned int *cx
, unsigned int *dx
)
226 unsigned maskebx
= ~0;
227 unsigned maskecx
= ~0;
228 unsigned maskedx
= ~0;
231 * Mask out inconvenient features, to try and disable as many
232 * unsupported kernel subsystems as possible.
236 maskecx
= cpuid_leaf1_ecx_mask
;
237 setecx
= cpuid_leaf1_ecx_set_mask
;
238 maskedx
= cpuid_leaf1_edx_mask
;
241 case CPUID_MWAIT_LEAF
:
242 /* Synthesize the values.. */
245 *cx
= cpuid_leaf5_ecx_val
;
246 *dx
= cpuid_leaf5_edx_val
;
249 case CPUID_THERM_POWER_LEAF
:
250 /* Disabling APERFMPERF for kernel usage */
251 maskecx
= ~(1 << APERFMPERF_PRESENT
);
255 /* Suppress extended topology stuff */
260 asm(XEN_EMULATE_PREFIX
"cpuid"
265 : "0" (*ax
), "2" (*cx
));
274 static bool __init
xen_check_mwait(void)
276 #if defined(CONFIG_ACPI) && !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR) && \
277 !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR_MODULE)
278 struct xen_platform_op op
= {
279 .cmd
= XENPF_set_processor_pminfo
,
280 .u
.set_pminfo
.id
= -1,
281 .u
.set_pminfo
.type
= XEN_PM_PDC
,
284 unsigned int ax
, bx
, cx
, dx
;
285 unsigned int mwait_mask
;
287 /* We need to determine whether it is OK to expose the MWAIT
288 * capability to the kernel to harvest deeper than C3 states from ACPI
289 * _CST using the processor_harvest_xen.c module. For this to work, we
290 * need to gather the MWAIT_LEAF values (which the cstate.c code
291 * checks against). The hypervisor won't expose the MWAIT flag because
292 * it would break backwards compatibility; so we will find out directly
293 * from the hardware and hypercall.
295 if (!xen_initial_domain())
301 native_cpuid(&ax
, &bx
, &cx
, &dx
);
303 mwait_mask
= (1 << (X86_FEATURE_EST
% 32)) |
304 (1 << (X86_FEATURE_MWAIT
% 32));
306 if ((cx
& mwait_mask
) != mwait_mask
)
309 /* We need to emulate the MWAIT_LEAF and for that we need both
310 * ecx and edx. The hypercall provides only partial information.
313 ax
= CPUID_MWAIT_LEAF
;
318 native_cpuid(&ax
, &bx
, &cx
, &dx
);
320 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
321 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
323 buf
[0] = ACPI_PDC_REVISION_ID
;
325 buf
[2] = (ACPI_PDC_C_CAPABILITY_SMP
| ACPI_PDC_EST_CAPABILITY_SWSMP
);
327 set_xen_guest_handle(op
.u
.set_pminfo
.pdc
, buf
);
329 if ((HYPERVISOR_dom0_op(&op
) == 0) &&
330 (buf
[2] & (ACPI_PDC_C_C1_FFH
| ACPI_PDC_C_C2C3_FFH
))) {
331 cpuid_leaf5_ecx_val
= cx
;
332 cpuid_leaf5_edx_val
= dx
;
339 static void __init
xen_init_cpuid_mask(void)
341 unsigned int ax
, bx
, cx
, dx
;
342 unsigned int xsave_mask
;
344 cpuid_leaf1_edx_mask
=
345 ~((1 << X86_FEATURE_MTRR
) | /* disable MTRR */
346 (1 << X86_FEATURE_ACC
)); /* thermal monitoring */
348 if (!xen_initial_domain())
349 cpuid_leaf1_edx_mask
&=
350 ~((1 << X86_FEATURE_APIC
) | /* disable local APIC */
351 (1 << X86_FEATURE_ACPI
)); /* disable ACPI */
354 xen_cpuid(&ax
, &bx
, &cx
, &dx
);
357 (1 << (X86_FEATURE_XSAVE
% 32)) |
358 (1 << (X86_FEATURE_OSXSAVE
% 32));
360 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
361 if ((cx
& xsave_mask
) != xsave_mask
)
362 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 inline bool desc_equal(const struct desc_struct
*d1
,
543 const struct desc_struct
*d2
)
545 return d1
->a
== d2
->a
&& d1
->b
== d2
->b
;
548 static void load_TLS_descriptor(struct thread_struct
*t
,
549 unsigned int cpu
, unsigned int i
)
551 struct desc_struct
*gdt
= get_cpu_gdt_table(cpu
);
552 xmaddr_t maddr
= arbitrary_virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
553 struct multicall_space mc
= __xen_mc_entry(0);
555 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
558 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
561 * XXX sleazy hack: If we're being called in a lazy-cpu zone
562 * and lazy gs handling is enabled, it means we're in a
563 * context switch, and %gs has just been saved. This means we
564 * can zero it out to prevent faults on exit from the
565 * hypervisor if the next process has no %gs. Either way, it
566 * has been saved, and the new value will get loaded properly.
567 * This will go away as soon as Xen has been modified to not
568 * save/restore %gs for normal hypercalls.
570 * On x86_64, this hack is not used for %gs, because gs points
571 * to KERNEL_GS_BASE (and uses it for PDA references), so we
572 * must not zero %gs on x86_64
574 * For x86_64, we need to zero %fs, otherwise we may get an
575 * exception between the new %fs descriptor being loaded and
576 * %fs being effectively cleared at __switch_to().
578 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
588 load_TLS_descriptor(t
, cpu
, 0);
589 load_TLS_descriptor(t
, cpu
, 1);
590 load_TLS_descriptor(t
, cpu
, 2);
592 xen_mc_issue(PARAVIRT_LAZY_CPU
);
596 static void xen_load_gs_index(unsigned int idx
)
598 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
603 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
606 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
607 u64 entry
= *(u64
*)ptr
;
609 trace_xen_cpu_write_ldt_entry(dt
, entrynum
, entry
);
614 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
620 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
621 struct trap_info
*info
)
625 if (val
->type
!= GATE_TRAP
&& val
->type
!= GATE_INTERRUPT
)
628 info
->vector
= vector
;
630 addr
= gate_offset(*val
);
633 * Look for known traps using IST, and substitute them
634 * appropriately. The debugger ones are the only ones we care
635 * about. Xen will handle faults like double_fault,
636 * so we should never see them. Warn if
637 * there's an unexpected IST-using fault handler.
639 if (addr
== (unsigned long)debug
)
640 addr
= (unsigned long)xen_debug
;
641 else if (addr
== (unsigned long)int3
)
642 addr
= (unsigned long)xen_int3
;
643 else if (addr
== (unsigned long)stack_segment
)
644 addr
= (unsigned long)xen_stack_segment
;
645 else if (addr
== (unsigned long)double_fault
||
646 addr
== (unsigned long)nmi
) {
647 /* Don't need to handle these */
649 #ifdef CONFIG_X86_MCE
650 } else if (addr
== (unsigned long)machine_check
) {
652 * when xen hypervisor inject vMCE to guest,
653 * use native mce handler to handle it
658 /* Some other trap using IST? */
659 if (WARN_ON(val
->ist
!= 0))
662 #endif /* CONFIG_X86_64 */
663 info
->address
= addr
;
665 info
->cs
= gate_segment(*val
);
666 info
->flags
= val
->dpl
;
667 /* interrupt gates clear IF */
668 if (val
->type
== GATE_INTERRUPT
)
669 info
->flags
|= 1 << 2;
674 /* Locations of each CPU's IDT */
675 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
677 /* Set an IDT entry. If the entry is part of the current IDT, then
679 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
681 unsigned long p
= (unsigned long)&dt
[entrynum
];
682 unsigned long start
, end
;
684 trace_xen_cpu_write_idt_entry(dt
, entrynum
, g
);
688 start
= __this_cpu_read(idt_desc
.address
);
689 end
= start
+ __this_cpu_read(idt_desc
.size
) + 1;
693 native_write_idt_entry(dt
, entrynum
, g
);
695 if (p
>= start
&& (p
+ 8) <= end
) {
696 struct trap_info info
[2];
700 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
701 if (HYPERVISOR_set_trap_table(info
))
708 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
709 struct trap_info
*traps
)
711 unsigned in
, out
, count
;
713 count
= (desc
->size
+1) / sizeof(gate_desc
);
716 for (in
= out
= 0; in
< count
; in
++) {
717 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
719 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
722 traps
[out
].address
= 0;
725 void xen_copy_trap_info(struct trap_info
*traps
)
727 const struct desc_ptr
*desc
= &__get_cpu_var(idt_desc
);
729 xen_convert_trap_info(desc
, traps
);
732 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
733 hold a spinlock to protect the static traps[] array (static because
734 it avoids allocation, and saves stack space). */
735 static void xen_load_idt(const struct desc_ptr
*desc
)
737 static DEFINE_SPINLOCK(lock
);
738 static struct trap_info traps
[257];
740 trace_xen_cpu_load_idt(desc
);
744 __get_cpu_var(idt_desc
) = *desc
;
746 xen_convert_trap_info(desc
, traps
);
749 if (HYPERVISOR_set_trap_table(traps
))
755 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
756 they're handled differently. */
757 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
758 const void *desc
, int type
)
760 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
771 xmaddr_t maddr
= arbitrary_virt_to_machine(&dt
[entry
]);
774 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
784 * Version of write_gdt_entry for use at early boot-time needed to
785 * update an entry as simply as possible.
787 static void __init
xen_write_gdt_entry_boot(struct desc_struct
*dt
, int entry
,
788 const void *desc
, int type
)
790 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
799 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
801 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
802 dt
[entry
] = *(struct desc_struct
*)desc
;
808 static void xen_load_sp0(struct tss_struct
*tss
,
809 struct thread_struct
*thread
)
811 struct multicall_space mcs
;
813 mcs
= xen_mc_entry(0);
814 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
815 xen_mc_issue(PARAVIRT_LAZY_CPU
);
818 static void xen_set_iopl_mask(unsigned mask
)
820 struct physdev_set_iopl set_iopl
;
822 /* Force the change at ring 0. */
823 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
824 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
827 static void xen_io_delay(void)
831 #ifdef CONFIG_X86_LOCAL_APIC
832 static unsigned long xen_set_apic_id(unsigned int x
)
837 static unsigned int xen_get_apic_id(unsigned long x
)
839 return ((x
)>>24) & 0xFFu
;
841 static u32
xen_apic_read(u32 reg
)
843 struct xen_platform_op op
= {
844 .cmd
= XENPF_get_cpuinfo
,
845 .interface_version
= XENPF_INTERFACE_VERSION
,
846 .u
.pcpu_info
.xen_cpuid
= 0,
850 /* Shouldn't need this as APIC is turned off for PV, and we only
851 * get called on the bootup processor. But just in case. */
852 if (!xen_initial_domain() || smp_processor_id())
861 ret
= HYPERVISOR_dom0_op(&op
);
865 return op
.u
.pcpu_info
.apic_id
<< 24;
868 static void xen_apic_write(u32 reg
, u32 val
)
870 /* Warn to see if there's any stray references */
874 static u64
xen_apic_icr_read(void)
879 static void xen_apic_icr_write(u32 low
, u32 id
)
881 /* Warn to see if there's any stray references */
885 static void xen_apic_wait_icr_idle(void)
890 static u32
xen_safe_apic_wait_icr_idle(void)
895 static void set_xen_basic_apic_ops(void)
897 apic
->read
= xen_apic_read
;
898 apic
->write
= xen_apic_write
;
899 apic
->icr_read
= xen_apic_icr_read
;
900 apic
->icr_write
= xen_apic_icr_write
;
901 apic
->wait_icr_idle
= xen_apic_wait_icr_idle
;
902 apic
->safe_wait_icr_idle
= xen_safe_apic_wait_icr_idle
;
903 apic
->set_apic_id
= xen_set_apic_id
;
904 apic
->get_apic_id
= xen_get_apic_id
;
907 apic
->send_IPI_allbutself
= xen_send_IPI_allbutself
;
908 apic
->send_IPI_mask_allbutself
= xen_send_IPI_mask_allbutself
;
909 apic
->send_IPI_mask
= xen_send_IPI_mask
;
910 apic
->send_IPI_all
= xen_send_IPI_all
;
911 apic
->send_IPI_self
= xen_send_IPI_self
;
917 static void xen_clts(void)
919 struct multicall_space mcs
;
921 mcs
= xen_mc_entry(0);
923 MULTI_fpu_taskswitch(mcs
.mc
, 0);
925 xen_mc_issue(PARAVIRT_LAZY_CPU
);
928 static DEFINE_PER_CPU(unsigned long, xen_cr0_value
);
930 static unsigned long xen_read_cr0(void)
932 unsigned long cr0
= this_cpu_read(xen_cr0_value
);
934 if (unlikely(cr0
== 0)) {
935 cr0
= native_read_cr0();
936 this_cpu_write(xen_cr0_value
, cr0
);
942 static void xen_write_cr0(unsigned long cr0
)
944 struct multicall_space mcs
;
946 this_cpu_write(xen_cr0_value
, cr0
);
948 /* Only pay attention to cr0.TS; everything else is
950 mcs
= xen_mc_entry(0);
952 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
954 xen_mc_issue(PARAVIRT_LAZY_CPU
);
957 static void xen_write_cr4(unsigned long cr4
)
962 native_write_cr4(cr4
);
965 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
976 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
977 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
978 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
981 base
= ((u64
)high
<< 32) | low
;
982 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
990 case MSR_SYSCALL_MASK
:
991 case MSR_IA32_SYSENTER_CS
:
992 case MSR_IA32_SYSENTER_ESP
:
993 case MSR_IA32_SYSENTER_EIP
:
994 /* Fast syscall setup is all done in hypercalls, so
995 these are all ignored. Stub them out here to stop
996 Xen console noise. */
999 case MSR_IA32_CR_PAT
:
1000 if (smp_processor_id() == 0)
1001 xen_set_pat(((u64
)high
<< 32) | low
);
1005 ret
= native_write_msr_safe(msr
, low
, high
);
1011 void xen_setup_shared_info(void)
1013 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1014 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
1015 xen_start_info
->shared_info
);
1017 HYPERVISOR_shared_info
=
1018 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
1020 HYPERVISOR_shared_info
=
1021 (struct shared_info
*)__va(xen_start_info
->shared_info
);
1024 /* In UP this is as good a place as any to set up shared info */
1025 xen_setup_vcpu_info_placement();
1028 xen_setup_mfn_list_list();
1031 /* This is called once we have the cpu_possible_mask */
1032 void xen_setup_vcpu_info_placement(void)
1036 for_each_possible_cpu(cpu
)
1037 xen_vcpu_setup(cpu
);
1039 /* xen_vcpu_setup managed to place the vcpu_info within the
1040 percpu area for all cpus, so make use of it */
1041 if (have_vcpu_info_placement
) {
1042 pv_irq_ops
.save_fl
= __PV_IS_CALLEE_SAVE(xen_save_fl_direct
);
1043 pv_irq_ops
.restore_fl
= __PV_IS_CALLEE_SAVE(xen_restore_fl_direct
);
1044 pv_irq_ops
.irq_disable
= __PV_IS_CALLEE_SAVE(xen_irq_disable_direct
);
1045 pv_irq_ops
.irq_enable
= __PV_IS_CALLEE_SAVE(xen_irq_enable_direct
);
1046 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1050 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1051 unsigned long addr
, unsigned len
)
1053 char *start
, *end
, *reloc
;
1056 start
= end
= reloc
= NULL
;
1058 #define SITE(op, x) \
1059 case PARAVIRT_PATCH(op.x): \
1060 if (have_vcpu_info_placement) { \
1061 start = (char *)xen_##x##_direct; \
1062 end = xen_##x##_direct_end; \
1063 reloc = xen_##x##_direct_reloc; \
1068 SITE(pv_irq_ops
, irq_enable
);
1069 SITE(pv_irq_ops
, irq_disable
);
1070 SITE(pv_irq_ops
, save_fl
);
1071 SITE(pv_irq_ops
, restore_fl
);
1075 if (start
== NULL
|| (end
-start
) > len
)
1078 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1080 /* Note: because reloc is assigned from something that
1081 appears to be an array, gcc assumes it's non-null,
1082 but doesn't know its relationship with start and
1084 if (reloc
> start
&& reloc
< end
) {
1085 int reloc_off
= reloc
- start
;
1086 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1087 long delta
= start
- (char *)addr
;
1095 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1103 static const struct pv_info xen_info __initconst
= {
1104 .paravirt_enabled
= 1,
1105 .shared_kernel_pmd
= 0,
1107 #ifdef CONFIG_X86_64
1108 .extra_user_64bit_cs
= FLAT_USER_CS64
,
1114 static const struct pv_init_ops xen_init_ops __initconst
= {
1118 static const struct pv_cpu_ops xen_cpu_ops __initconst
= {
1121 .set_debugreg
= xen_set_debugreg
,
1122 .get_debugreg
= xen_get_debugreg
,
1126 .read_cr0
= xen_read_cr0
,
1127 .write_cr0
= xen_write_cr0
,
1129 .read_cr4
= native_read_cr4
,
1130 .read_cr4_safe
= native_read_cr4_safe
,
1131 .write_cr4
= xen_write_cr4
,
1133 .wbinvd
= native_wbinvd
,
1135 .read_msr
= native_read_msr_safe
,
1136 .rdmsr_regs
= native_rdmsr_safe_regs
,
1137 .write_msr
= xen_write_msr_safe
,
1138 .wrmsr_regs
= native_wrmsr_safe_regs
,
1140 .read_tsc
= native_read_tsc
,
1141 .read_pmc
= native_read_pmc
,
1144 .irq_enable_sysexit
= xen_sysexit
,
1145 #ifdef CONFIG_X86_64
1146 .usergs_sysret32
= xen_sysret32
,
1147 .usergs_sysret64
= xen_sysret64
,
1150 .load_tr_desc
= paravirt_nop
,
1151 .set_ldt
= xen_set_ldt
,
1152 .load_gdt
= xen_load_gdt
,
1153 .load_idt
= xen_load_idt
,
1154 .load_tls
= xen_load_tls
,
1155 #ifdef CONFIG_X86_64
1156 .load_gs_index
= xen_load_gs_index
,
1159 .alloc_ldt
= xen_alloc_ldt
,
1160 .free_ldt
= xen_free_ldt
,
1162 .store_gdt
= native_store_gdt
,
1163 .store_idt
= native_store_idt
,
1164 .store_tr
= xen_store_tr
,
1166 .write_ldt_entry
= xen_write_ldt_entry
,
1167 .write_gdt_entry
= xen_write_gdt_entry
,
1168 .write_idt_entry
= xen_write_idt_entry
,
1169 .load_sp0
= xen_load_sp0
,
1171 .set_iopl_mask
= xen_set_iopl_mask
,
1172 .io_delay
= xen_io_delay
,
1174 /* Xen takes care of %gs when switching to usermode for us */
1175 .swapgs
= paravirt_nop
,
1177 .start_context_switch
= paravirt_start_context_switch
,
1178 .end_context_switch
= xen_end_context_switch
,
1181 static const struct pv_apic_ops xen_apic_ops __initconst
= {
1182 #ifdef CONFIG_X86_LOCAL_APIC
1183 .startup_ipi_hook
= paravirt_nop
,
1187 static void xen_reboot(int reason
)
1189 struct sched_shutdown r
= { .reason
= reason
};
1191 if (HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
))
1195 static void xen_restart(char *msg
)
1197 xen_reboot(SHUTDOWN_reboot
);
1200 static void xen_emergency_restart(void)
1202 xen_reboot(SHUTDOWN_reboot
);
1205 static void xen_machine_halt(void)
1207 xen_reboot(SHUTDOWN_poweroff
);
1210 static void xen_machine_power_off(void)
1214 xen_reboot(SHUTDOWN_poweroff
);
1217 static void xen_crash_shutdown(struct pt_regs
*regs
)
1219 xen_reboot(SHUTDOWN_crash
);
1223 xen_panic_event(struct notifier_block
*this, unsigned long event
, void *ptr
)
1225 xen_reboot(SHUTDOWN_crash
);
1229 static struct notifier_block xen_panic_block
= {
1230 .notifier_call
= xen_panic_event
,
1233 int xen_panic_handler_init(void)
1235 atomic_notifier_chain_register(&panic_notifier_list
, &xen_panic_block
);
1239 static const struct machine_ops xen_machine_ops __initconst
= {
1240 .restart
= xen_restart
,
1241 .halt
= xen_machine_halt
,
1242 .power_off
= xen_machine_power_off
,
1243 .shutdown
= xen_machine_halt
,
1244 .crash_shutdown
= xen_crash_shutdown
,
1245 .emergency_restart
= xen_emergency_restart
,
1249 * Set up the GDT and segment registers for -fstack-protector. Until
1250 * we do this, we have to be careful not to call any stack-protected
1251 * function, which is most of the kernel.
1253 static void __init
xen_setup_stackprotector(void)
1255 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry_boot
;
1256 pv_cpu_ops
.load_gdt
= xen_load_gdt_boot
;
1258 setup_stack_canary_segment(0);
1259 switch_to_new_gdt(0);
1261 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry
;
1262 pv_cpu_ops
.load_gdt
= xen_load_gdt
;
1265 /* First C function to be called on Xen boot */
1266 asmlinkage
void __init
xen_start_kernel(void)
1268 struct physdev_set_iopl set_iopl
;
1272 if (!xen_start_info
)
1275 xen_domain_type
= XEN_PV_DOMAIN
;
1277 xen_setup_machphys_mapping();
1279 /* Install Xen paravirt ops */
1281 pv_init_ops
= xen_init_ops
;
1282 pv_cpu_ops
= xen_cpu_ops
;
1283 pv_apic_ops
= xen_apic_ops
;
1285 x86_init
.resources
.memory_setup
= xen_memory_setup
;
1286 x86_init
.oem
.arch_setup
= xen_arch_setup
;
1287 x86_init
.oem
.banner
= xen_banner
;
1289 xen_init_time_ops();
1292 * Set up some pagetable state before starting to set any ptes.
1297 /* Prevent unwanted bits from being set in PTEs. */
1298 __supported_pte_mask
&= ~_PAGE_GLOBAL
;
1300 if (!xen_initial_domain())
1302 __supported_pte_mask
&= ~(_PAGE_PWT
| _PAGE_PCD
);
1304 __supported_pte_mask
|= _PAGE_IOMAP
;
1307 * Prevent page tables from being allocated in highmem, even
1308 * if CONFIG_HIGHPTE is enabled.
1310 __userpte_alloc_gfp
&= ~__GFP_HIGHMEM
;
1312 /* Work out if we support NX */
1315 xen_setup_features();
1318 if (!xen_feature(XENFEAT_auto_translated_physmap
))
1319 xen_build_dynamic_phys_to_machine();
1322 * Set up kernel GDT and segment registers, mainly so that
1323 * -fstack-protector code can be executed.
1325 xen_setup_stackprotector();
1328 xen_init_cpuid_mask();
1330 #ifdef CONFIG_X86_LOCAL_APIC
1332 * set up the basic apic ops.
1334 set_xen_basic_apic_ops();
1337 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1338 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1339 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1342 machine_ops
= xen_machine_ops
;
1345 * The only reliable way to retain the initial address of the
1346 * percpu gdt_page is to remember it here, so we can go and
1347 * mark it RW later, when the initial percpu area is freed.
1349 xen_initial_gdt
= &per_cpu(gdt_page
, 0);
1353 #ifdef CONFIG_ACPI_NUMA
1355 * The pages we from Xen are not related to machine pages, so
1356 * any NUMA information the kernel tries to get from ACPI will
1357 * be meaningless. Prevent it from trying.
1362 pgd
= (pgd_t
*)xen_start_info
->pt_base
;
1364 /* Don't do the full vcpu_info placement stuff until we have a
1365 possible map and a non-dummy shared_info. */
1366 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1368 local_irq_disable();
1369 early_boot_irqs_disabled
= true;
1371 xen_raw_console_write("mapping kernel into physical memory\n");
1372 pgd
= xen_setup_kernel_pagetable(pgd
, xen_start_info
->nr_pages
);
1374 /* Allocate and initialize top and mid mfn levels for p2m structure */
1375 xen_build_mfn_list_list();
1377 /* keep using Xen gdt for now; no urgent need to change it */
1379 #ifdef CONFIG_X86_32
1380 pv_info
.kernel_rpl
= 1;
1381 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1382 pv_info
.kernel_rpl
= 0;
1384 pv_info
.kernel_rpl
= 0;
1386 /* set the limit of our address space */
1389 /* We used to do this in xen_arch_setup, but that is too late on AMD
1390 * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1391 * which pokes 0xcf8 port.
1394 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
1396 xen_raw_printk("physdev_op failed %d\n", rc
);
1398 #ifdef CONFIG_X86_32
1399 /* set up basic CPUID stuff */
1400 cpu_detect(&new_cpu_data
);
1401 new_cpu_data
.hard_math
= 1;
1402 new_cpu_data
.wp_works_ok
= 1;
1403 new_cpu_data
.x86_capability
[0] = cpuid_edx(1);
1406 /* Poke various useful things into boot_params */
1407 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1408 boot_params
.hdr
.ramdisk_image
= xen_start_info
->mod_start
1409 ? __pa(xen_start_info
->mod_start
) : 0;
1410 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1411 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1413 if (!xen_initial_domain()) {
1414 add_preferred_console("xenboot", 0, NULL
);
1415 add_preferred_console("tty", 0, NULL
);
1416 add_preferred_console("hvc", 0, NULL
);
1418 x86_init
.pci
.arch_init
= pci_xen_init
;
1420 const struct dom0_vga_console_info
*info
=
1421 (void *)((char *)xen_start_info
+
1422 xen_start_info
->console
.dom0
.info_off
);
1424 xen_init_vga(info
, xen_start_info
->console
.dom0
.info_size
);
1425 xen_start_info
->console
.domU
.mfn
= 0;
1426 xen_start_info
->console
.domU
.evtchn
= 0;
1430 /* Make sure ACS will be enabled */
1433 xen_acpi_sleep_register();
1436 /* PCI BIOS service won't work from a PV guest. */
1437 pci_probe
&= ~PCI_PROBE_BIOS
;
1439 xen_raw_console_write("about to get started...\n");
1441 xen_setup_runstate_info(0);
1443 /* Start the world */
1444 #ifdef CONFIG_X86_32
1445 i386_start_kernel();
1447 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));
1451 static int init_hvm_pv_info(int *major
, int *minor
)
1453 uint32_t eax
, ebx
, ecx
, edx
, pages
, msr
, base
;
1456 base
= xen_cpuid_base();
1457 cpuid(base
+ 1, &eax
, &ebx
, &ecx
, &edx
);
1460 *minor
= eax
& 0xffff;
1461 printk(KERN_INFO
"Xen version %d.%d.\n", *major
, *minor
);
1463 cpuid(base
+ 2, &pages
, &msr
, &ecx
, &edx
);
1465 pfn
= __pa(hypercall_page
);
1466 wrmsr_safe(msr
, (u32
)pfn
, (u32
)(pfn
>> 32));
1468 xen_setup_features();
1470 pv_info
.name
= "Xen HVM";
1472 xen_domain_type
= XEN_HVM_DOMAIN
;
1477 void __ref
xen_hvm_init_shared_info(void)
1480 struct xen_add_to_physmap xatp
;
1481 static struct shared_info
*shared_info_page
= 0;
1483 if (!shared_info_page
)
1484 shared_info_page
= (struct shared_info
*)
1485 extend_brk(PAGE_SIZE
, PAGE_SIZE
);
1486 xatp
.domid
= DOMID_SELF
;
1488 xatp
.space
= XENMAPSPACE_shared_info
;
1489 xatp
.gpfn
= __pa(shared_info_page
) >> PAGE_SHIFT
;
1490 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap
, &xatp
))
1493 HYPERVISOR_shared_info
= (struct shared_info
*)shared_info_page
;
1495 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1496 * page, we use it in the event channel upcall and in some pvclock
1497 * related functions. We don't need the vcpu_info placement
1498 * optimizations because we don't use any pv_mmu or pv_irq op on
1500 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1501 * online but xen_hvm_init_shared_info is run at resume time too and
1502 * in that case multiple vcpus might be online. */
1503 for_each_online_cpu(cpu
) {
1504 per_cpu(xen_vcpu
, cpu
) = &HYPERVISOR_shared_info
->vcpu_info
[cpu
];
1508 #ifdef CONFIG_XEN_PVHVM
1509 static int __cpuinit
xen_hvm_cpu_notify(struct notifier_block
*self
,
1510 unsigned long action
, void *hcpu
)
1512 int cpu
= (long)hcpu
;
1514 case CPU_UP_PREPARE
:
1515 xen_vcpu_setup(cpu
);
1516 if (xen_have_vector_callback
)
1517 xen_init_lock_cpu(cpu
);
1525 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata
= {
1526 .notifier_call
= xen_hvm_cpu_notify
,
1529 static void __init
xen_hvm_guest_init(void)
1534 r
= init_hvm_pv_info(&major
, &minor
);
1538 xen_hvm_init_shared_info();
1540 if (xen_feature(XENFEAT_hvm_callback_vector
))
1541 xen_have_vector_callback
= 1;
1543 register_cpu_notifier(&xen_hvm_cpu_notifier
);
1544 xen_unplug_emulated_devices();
1545 x86_init
.irqs
.intr_init
= xen_init_IRQ
;
1546 xen_hvm_init_time_ops();
1547 xen_hvm_init_mmu_ops();
1550 static bool __init
xen_hvm_platform(void)
1552 if (xen_pv_domain())
1555 if (!xen_cpuid_base())
1561 bool xen_hvm_need_lapic(void)
1563 if (xen_pv_domain())
1565 if (!xen_hvm_domain())
1567 if (xen_feature(XENFEAT_hvm_pirqs
) && xen_have_vector_callback
)
1571 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic
);
1573 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst
= {
1575 .detect
= xen_hvm_platform
,
1576 .init_platform
= xen_hvm_guest_init
,
1578 EXPORT_SYMBOL(x86_hyper_xen_hvm
);