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/export.h>
28 #include <linux/page-flags.h>
29 #include <linux/highmem.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/memblock.h>
34 #include <linux/edd.h>
35 #include <linux/frame.h>
38 #include <xen/events.h>
39 #include <xen/interface/xen.h>
40 #include <xen/interface/version.h>
41 #include <xen/interface/physdev.h>
42 #include <xen/interface/vcpu.h>
43 #include <xen/interface/memory.h>
44 #include <xen/interface/nmi.h>
45 #include <xen/interface/xen-mca.h>
46 #include <xen/features.h>
48 #include <xen/hvc-console.h>
51 #include <asm/paravirt.h>
54 #include <asm/xen/pci.h>
55 #include <asm/xen/hypercall.h>
56 #include <asm/xen/hypervisor.h>
57 #include <asm/xen/cpuid.h>
58 #include <asm/fixmap.h>
59 #include <asm/processor.h>
60 #include <asm/proto.h>
61 #include <asm/msr-index.h>
62 #include <asm/traps.h>
63 #include <asm/setup.h>
65 #include <asm/pgalloc.h>
66 #include <asm/pgtable.h>
67 #include <asm/tlbflush.h>
68 #include <asm/reboot.h>
69 #include <asm/stackprotector.h>
70 #include <asm/hypervisor.h>
71 #include <asm/mach_traps.h>
72 #include <asm/mwait.h>
73 #include <asm/pci_x86.h>
77 #include <linux/acpi.h>
79 #include <acpi/pdc_intel.h>
80 #include <acpi/processor.h>
81 #include <xen/interface/platform.h>
87 #include "multicalls.h"
90 void *xen_initial_gdt
;
92 RESERVE_BRK(shared_info_page_brk
, PAGE_SIZE
);
94 static int xen_cpu_up_prepare_pv(unsigned int cpu
);
95 static int xen_cpu_dead_pv(unsigned int cpu
);
98 struct desc_struct desc
[3];
102 * Updating the 3 TLS descriptors in the GDT on every task switch is
103 * surprisingly expensive so we avoid updating them if they haven't
104 * changed. Since Xen writes different descriptors than the one
105 * passed in the update_descriptor hypercall we keep shadow copies to
108 static DEFINE_PER_CPU(struct tls_descs
, shadow_tls_desc
);
111 * On restore, set the vcpu placement up again.
112 * If it fails, then we're in a bad state, since
113 * we can't back out from using it...
115 void xen_vcpu_restore(void)
119 for_each_possible_cpu(cpu
) {
120 bool other_cpu
= (cpu
!= smp_processor_id());
121 bool is_up
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, xen_vcpu_nr(cpu
),
124 if (other_cpu
&& is_up
&&
125 HYPERVISOR_vcpu_op(VCPUOP_down
, xen_vcpu_nr(cpu
), NULL
))
128 xen_setup_runstate_info(cpu
);
130 if (xen_have_vcpu_info_placement
)
133 if (other_cpu
&& is_up
&&
134 HYPERVISOR_vcpu_op(VCPUOP_up
, xen_vcpu_nr(cpu
), NULL
))
139 static void __init
xen_banner(void)
141 unsigned version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
142 struct xen_extraversion extra
;
143 HYPERVISOR_xen_version(XENVER_extraversion
, &extra
);
145 pr_info("Booting paravirtualized kernel %son %s\n",
146 xen_feature(XENFEAT_auto_translated_physmap
) ?
147 "with PVH extensions " : "", pv_info
.name
);
148 printk(KERN_INFO
"Xen version: %d.%d%s%s\n",
149 version
>> 16, version
& 0xffff, extra
.extraversion
,
150 xen_feature(XENFEAT_mmu_pt_update_preserve_ad
) ? " (preserve-AD)" : "");
152 /* Check if running on Xen version (major, minor) or later */
154 xen_running_on_version_or_later(unsigned int major
, unsigned int minor
)
156 unsigned int version
;
161 version
= HYPERVISOR_xen_version(XENVER_version
, NULL
);
162 if ((((version
>> 16) == major
) && ((version
& 0xffff) >= minor
)) ||
163 ((version
>> 16) > major
))
168 static __read_mostly
unsigned int cpuid_leaf5_ecx_val
;
169 static __read_mostly
unsigned int cpuid_leaf5_edx_val
;
171 static void xen_cpuid(unsigned int *ax
, unsigned int *bx
,
172 unsigned int *cx
, unsigned int *dx
)
174 unsigned maskebx
= ~0;
177 * Mask out inconvenient features, to try and disable as many
178 * unsupported kernel subsystems as possible.
181 case CPUID_MWAIT_LEAF
:
182 /* Synthesize the values.. */
185 *cx
= cpuid_leaf5_ecx_val
;
186 *dx
= cpuid_leaf5_edx_val
;
190 /* Suppress extended topology stuff */
195 asm(XEN_EMULATE_PREFIX
"cpuid"
200 : "0" (*ax
), "2" (*cx
));
204 STACK_FRAME_NON_STANDARD(xen_cpuid
); /* XEN_EMULATE_PREFIX */
206 static bool __init
xen_check_mwait(void)
209 struct xen_platform_op op
= {
210 .cmd
= XENPF_set_processor_pminfo
,
211 .u
.set_pminfo
.id
= -1,
212 .u
.set_pminfo
.type
= XEN_PM_PDC
,
215 unsigned int ax
, bx
, cx
, dx
;
216 unsigned int mwait_mask
;
218 /* We need to determine whether it is OK to expose the MWAIT
219 * capability to the kernel to harvest deeper than C3 states from ACPI
220 * _CST using the processor_harvest_xen.c module. For this to work, we
221 * need to gather the MWAIT_LEAF values (which the cstate.c code
222 * checks against). The hypervisor won't expose the MWAIT flag because
223 * it would break backwards compatibility; so we will find out directly
224 * from the hardware and hypercall.
226 if (!xen_initial_domain())
230 * When running under platform earlier than Xen4.2, do not expose
231 * mwait, to avoid the risk of loading native acpi pad driver
233 if (!xen_running_on_version_or_later(4, 2))
239 native_cpuid(&ax
, &bx
, &cx
, &dx
);
241 mwait_mask
= (1 << (X86_FEATURE_EST
% 32)) |
242 (1 << (X86_FEATURE_MWAIT
% 32));
244 if ((cx
& mwait_mask
) != mwait_mask
)
247 /* We need to emulate the MWAIT_LEAF and for that we need both
248 * ecx and edx. The hypercall provides only partial information.
251 ax
= CPUID_MWAIT_LEAF
;
256 native_cpuid(&ax
, &bx
, &cx
, &dx
);
258 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
259 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
261 buf
[0] = ACPI_PDC_REVISION_ID
;
263 buf
[2] = (ACPI_PDC_C_CAPABILITY_SMP
| ACPI_PDC_EST_CAPABILITY_SWSMP
);
265 set_xen_guest_handle(op
.u
.set_pminfo
.pdc
, buf
);
267 if ((HYPERVISOR_platform_op(&op
) == 0) &&
268 (buf
[2] & (ACPI_PDC_C_C1_FFH
| ACPI_PDC_C_C2C3_FFH
))) {
269 cpuid_leaf5_ecx_val
= cx
;
270 cpuid_leaf5_edx_val
= dx
;
278 static bool __init
xen_check_xsave(void)
280 unsigned int err
, eax
, edx
;
283 * Xen 4.0 and older accidentally leaked the host XSAVE flag into guest
284 * view, despite not being able to support guests using the
285 * functionality. Probe for the actual availability of XSAVE by seeing
286 * whether xgetbv executes successfully or raises #UD.
288 asm volatile("1: .byte 0x0f,0x01,0xd0\n\t" /* xgetbv */
289 "xor %[err], %[err]\n"
291 ".pushsection .fixup,\"ax\"\n\t"
292 "3: movl $1,%[err]\n\t"
296 : [err
] "=r" (err
), "=a" (eax
), "=d" (edx
)
302 static void __init
xen_init_capabilities(void)
304 setup_clear_cpu_cap(X86_BUG_SYSRET_SS_ATTRS
);
305 setup_force_cpu_cap(X86_FEATURE_XENPV
);
306 setup_clear_cpu_cap(X86_FEATURE_DCA
);
307 setup_clear_cpu_cap(X86_FEATURE_APERFMPERF
);
308 setup_clear_cpu_cap(X86_FEATURE_MTRR
);
309 setup_clear_cpu_cap(X86_FEATURE_ACC
);
310 setup_clear_cpu_cap(X86_FEATURE_X2APIC
);
312 if (!xen_initial_domain())
313 setup_clear_cpu_cap(X86_FEATURE_ACPI
);
315 if (xen_check_mwait())
316 setup_force_cpu_cap(X86_FEATURE_MWAIT
);
318 setup_clear_cpu_cap(X86_FEATURE_MWAIT
);
320 if (xen_check_xsave()) {
321 setup_force_cpu_cap(X86_FEATURE_XSAVE
);
322 setup_force_cpu_cap(X86_FEATURE_OSXSAVE
);
324 setup_clear_cpu_cap(X86_FEATURE_XSAVE
);
325 setup_clear_cpu_cap(X86_FEATURE_OSXSAVE
);
329 static void xen_set_debugreg(int reg
, unsigned long val
)
331 HYPERVISOR_set_debugreg(reg
, val
);
334 static unsigned long xen_get_debugreg(int reg
)
336 return HYPERVISOR_get_debugreg(reg
);
339 static void xen_end_context_switch(struct task_struct
*next
)
342 paravirt_end_context_switch(next
);
345 static unsigned long xen_store_tr(void)
351 * Set the page permissions for a particular virtual address. If the
352 * address is a vmalloc mapping (or other non-linear mapping), then
353 * find the linear mapping of the page and also set its protections to
356 static void set_aliased_prot(void *v
, pgprot_t prot
)
365 ptep
= lookup_address((unsigned long)v
, &level
);
366 BUG_ON(ptep
== NULL
);
368 pfn
= pte_pfn(*ptep
);
369 page
= pfn_to_page(pfn
);
371 pte
= pfn_pte(pfn
, prot
);
374 * Careful: update_va_mapping() will fail if the virtual address
375 * we're poking isn't populated in the page tables. We don't
376 * need to worry about the direct map (that's always in the page
377 * tables), but we need to be careful about vmap space. In
378 * particular, the top level page table can lazily propagate
379 * entries between processes, so if we've switched mms since we
380 * vmapped the target in the first place, we might not have the
381 * top-level page table entry populated.
383 * We disable preemption because we want the same mm active when
384 * we probe the target and when we issue the hypercall. We'll
385 * have the same nominal mm, but if we're a kernel thread, lazy
386 * mm dropping could change our pgd.
388 * Out of an abundance of caution, this uses __get_user() to fault
389 * in the target address just in case there's some obscure case
390 * in which the target address isn't readable.
395 probe_kernel_read(&dummy
, v
, 1);
397 if (HYPERVISOR_update_va_mapping((unsigned long)v
, pte
, 0))
400 if (!PageHighMem(page
)) {
401 void *av
= __va(PFN_PHYS(pfn
));
404 if (HYPERVISOR_update_va_mapping((unsigned long)av
, pte
, 0))
412 static void xen_alloc_ldt(struct desc_struct
*ldt
, unsigned entries
)
414 const unsigned entries_per_page
= PAGE_SIZE
/ LDT_ENTRY_SIZE
;
418 * We need to mark the all aliases of the LDT pages RO. We
419 * don't need to call vm_flush_aliases(), though, since that's
420 * only responsible for flushing aliases out the TLBs, not the
421 * page tables, and Xen will flush the TLB for us if needed.
423 * To avoid confusing future readers: none of this is necessary
424 * to load the LDT. The hypervisor only checks this when the
425 * LDT is faulted in due to subsequent descriptor access.
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
= DIV_ROUND_UP(size
, 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
= DIV_ROUND_UP(size
, 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
*shadow
= &per_cpu(shadow_tls_desc
, cpu
).desc
[i
];
552 struct desc_struct
*gdt
;
554 struct multicall_space mc
;
556 if (desc_equal(shadow
, &t
->tls_array
[i
]))
559 *shadow
= t
->tls_array
[i
];
561 gdt
= get_cpu_gdt_rw(cpu
);
562 maddr
= arbitrary_virt_to_machine(&gdt
[GDT_ENTRY_TLS_MIN
+i
]);
563 mc
= __xen_mc_entry(0);
565 MULTI_update_descriptor(mc
.mc
, maddr
.maddr
, t
->tls_array
[i
]);
568 static void xen_load_tls(struct thread_struct
*t
, unsigned int cpu
)
571 * XXX sleazy hack: If we're being called in a lazy-cpu zone
572 * and lazy gs handling is enabled, it means we're in a
573 * context switch, and %gs has just been saved. This means we
574 * can zero it out to prevent faults on exit from the
575 * hypervisor if the next process has no %gs. Either way, it
576 * has been saved, and the new value will get loaded properly.
577 * This will go away as soon as Xen has been modified to not
578 * save/restore %gs for normal hypercalls.
580 * On x86_64, this hack is not used for %gs, because gs points
581 * to KERNEL_GS_BASE (and uses it for PDA references), so we
582 * must not zero %gs on x86_64
584 * For x86_64, we need to zero %fs, otherwise we may get an
585 * exception between the new %fs descriptor being loaded and
586 * %fs being effectively cleared at __switch_to().
588 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU
) {
598 load_TLS_descriptor(t
, cpu
, 0);
599 load_TLS_descriptor(t
, cpu
, 1);
600 load_TLS_descriptor(t
, cpu
, 2);
602 xen_mc_issue(PARAVIRT_LAZY_CPU
);
606 static void xen_load_gs_index(unsigned int idx
)
608 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL
, idx
))
613 static void xen_write_ldt_entry(struct desc_struct
*dt
, int entrynum
,
616 xmaddr_t mach_lp
= arbitrary_virt_to_machine(&dt
[entrynum
]);
617 u64 entry
= *(u64
*)ptr
;
619 trace_xen_cpu_write_ldt_entry(dt
, entrynum
, entry
);
624 if (HYPERVISOR_update_descriptor(mach_lp
.maddr
, entry
))
630 static int cvt_gate_to_trap(int vector
, const gate_desc
*val
,
631 struct trap_info
*info
)
635 if (val
->type
!= GATE_TRAP
&& val
->type
!= GATE_INTERRUPT
)
638 info
->vector
= vector
;
640 addr
= gate_offset(*val
);
643 * Look for known traps using IST, and substitute them
644 * appropriately. The debugger ones are the only ones we care
645 * about. Xen will handle faults like double_fault,
646 * so we should never see them. Warn if
647 * there's an unexpected IST-using fault handler.
649 if (addr
== (unsigned long)debug
)
650 addr
= (unsigned long)xen_debug
;
651 else if (addr
== (unsigned long)int3
)
652 addr
= (unsigned long)xen_int3
;
653 else if (addr
== (unsigned long)stack_segment
)
654 addr
= (unsigned long)xen_stack_segment
;
655 else if (addr
== (unsigned long)double_fault
) {
656 /* Don't need to handle these */
658 #ifdef CONFIG_X86_MCE
659 } else if (addr
== (unsigned long)machine_check
) {
661 * when xen hypervisor inject vMCE to guest,
662 * use native mce handler to handle it
666 } else if (addr
== (unsigned long)nmi
)
668 * Use the native version as well.
672 /* Some other trap using IST? */
673 if (WARN_ON(val
->ist
!= 0))
676 #endif /* CONFIG_X86_64 */
677 info
->address
= addr
;
679 info
->cs
= gate_segment(*val
);
680 info
->flags
= val
->dpl
;
681 /* interrupt gates clear IF */
682 if (val
->type
== GATE_INTERRUPT
)
683 info
->flags
|= 1 << 2;
688 /* Locations of each CPU's IDT */
689 static DEFINE_PER_CPU(struct desc_ptr
, idt_desc
);
691 /* Set an IDT entry. If the entry is part of the current IDT, then
693 static void xen_write_idt_entry(gate_desc
*dt
, int entrynum
, const gate_desc
*g
)
695 unsigned long p
= (unsigned long)&dt
[entrynum
];
696 unsigned long start
, end
;
698 trace_xen_cpu_write_idt_entry(dt
, entrynum
, g
);
702 start
= __this_cpu_read(idt_desc
.address
);
703 end
= start
+ __this_cpu_read(idt_desc
.size
) + 1;
707 native_write_idt_entry(dt
, entrynum
, g
);
709 if (p
>= start
&& (p
+ 8) <= end
) {
710 struct trap_info info
[2];
714 if (cvt_gate_to_trap(entrynum
, g
, &info
[0]))
715 if (HYPERVISOR_set_trap_table(info
))
722 static void xen_convert_trap_info(const struct desc_ptr
*desc
,
723 struct trap_info
*traps
)
725 unsigned in
, out
, count
;
727 count
= (desc
->size
+1) / sizeof(gate_desc
);
730 for (in
= out
= 0; in
< count
; in
++) {
731 gate_desc
*entry
= (gate_desc
*)(desc
->address
) + in
;
733 if (cvt_gate_to_trap(in
, entry
, &traps
[out
]))
736 traps
[out
].address
= 0;
739 void xen_copy_trap_info(struct trap_info
*traps
)
741 const struct desc_ptr
*desc
= this_cpu_ptr(&idt_desc
);
743 xen_convert_trap_info(desc
, traps
);
746 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
747 hold a spinlock to protect the static traps[] array (static because
748 it avoids allocation, and saves stack space). */
749 static void xen_load_idt(const struct desc_ptr
*desc
)
751 static DEFINE_SPINLOCK(lock
);
752 static struct trap_info traps
[257];
754 trace_xen_cpu_load_idt(desc
);
758 memcpy(this_cpu_ptr(&idt_desc
), desc
, sizeof(idt_desc
));
760 xen_convert_trap_info(desc
, traps
);
763 if (HYPERVISOR_set_trap_table(traps
))
769 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
770 they're handled differently. */
771 static void xen_write_gdt_entry(struct desc_struct
*dt
, int entry
,
772 const void *desc
, int type
)
774 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
785 xmaddr_t maddr
= arbitrary_virt_to_machine(&dt
[entry
]);
788 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
798 * Version of write_gdt_entry for use at early boot-time needed to
799 * update an entry as simply as possible.
801 static void __init
xen_write_gdt_entry_boot(struct desc_struct
*dt
, int entry
,
802 const void *desc
, int type
)
804 trace_xen_cpu_write_gdt_entry(dt
, entry
, desc
, type
);
813 xmaddr_t maddr
= virt_to_machine(&dt
[entry
]);
815 if (HYPERVISOR_update_descriptor(maddr
.maddr
, *(u64
*)desc
))
816 dt
[entry
] = *(struct desc_struct
*)desc
;
822 static void xen_load_sp0(struct tss_struct
*tss
,
823 struct thread_struct
*thread
)
825 struct multicall_space mcs
;
827 mcs
= xen_mc_entry(0);
828 MULTI_stack_switch(mcs
.mc
, __KERNEL_DS
, thread
->sp0
);
829 xen_mc_issue(PARAVIRT_LAZY_CPU
);
830 tss
->x86_tss
.sp0
= thread
->sp0
;
833 void xen_set_iopl_mask(unsigned mask
)
835 struct physdev_set_iopl set_iopl
;
837 /* Force the change at ring 0. */
838 set_iopl
.iopl
= (mask
== 0) ? 1 : (mask
>> 12) & 3;
839 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
842 static void xen_io_delay(void)
846 static DEFINE_PER_CPU(unsigned long, xen_cr0_value
);
848 static unsigned long xen_read_cr0(void)
850 unsigned long cr0
= this_cpu_read(xen_cr0_value
);
852 if (unlikely(cr0
== 0)) {
853 cr0
= native_read_cr0();
854 this_cpu_write(xen_cr0_value
, cr0
);
860 static void xen_write_cr0(unsigned long cr0
)
862 struct multicall_space mcs
;
864 this_cpu_write(xen_cr0_value
, cr0
);
866 /* Only pay attention to cr0.TS; everything else is
868 mcs
= xen_mc_entry(0);
870 MULTI_fpu_taskswitch(mcs
.mc
, (cr0
& X86_CR0_TS
) != 0);
872 xen_mc_issue(PARAVIRT_LAZY_CPU
);
875 static void xen_write_cr4(unsigned long cr4
)
877 cr4
&= ~(X86_CR4_PGE
| X86_CR4_PSE
| X86_CR4_PCE
);
879 native_write_cr4(cr4
);
882 static inline unsigned long xen_read_cr8(void)
886 static inline void xen_write_cr8(unsigned long val
)
892 static u64
xen_read_msr_safe(unsigned int msr
, int *err
)
896 if (pmu_msr_read(msr
, &val
, err
))
899 val
= native_read_msr_safe(msr
, err
);
901 case MSR_IA32_APICBASE
:
902 #ifdef CONFIG_X86_X2APIC
903 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC
& 31))))
905 val
&= ~X2APIC_ENABLE
;
911 static int xen_write_msr_safe(unsigned int msr
, unsigned low
, unsigned high
)
922 case MSR_FS_BASE
: which
= SEGBASE_FS
; goto set
;
923 case MSR_KERNEL_GS_BASE
: which
= SEGBASE_GS_USER
; goto set
;
924 case MSR_GS_BASE
: which
= SEGBASE_GS_KERNEL
; goto set
;
927 base
= ((u64
)high
<< 32) | low
;
928 if (HYPERVISOR_set_segment_base(which
, base
) != 0)
936 case MSR_SYSCALL_MASK
:
937 case MSR_IA32_SYSENTER_CS
:
938 case MSR_IA32_SYSENTER_ESP
:
939 case MSR_IA32_SYSENTER_EIP
:
940 /* Fast syscall setup is all done in hypercalls, so
941 these are all ignored. Stub them out here to stop
942 Xen console noise. */
946 if (!pmu_msr_write(msr
, low
, high
, &ret
))
947 ret
= native_write_msr_safe(msr
, low
, high
);
953 static u64
xen_read_msr(unsigned int msr
)
956 * This will silently swallow a #GP from RDMSR. It may be worth
961 return xen_read_msr_safe(msr
, &err
);
964 static void xen_write_msr(unsigned int msr
, unsigned low
, unsigned high
)
967 * This will silently swallow a #GP from WRMSR. It may be worth
970 xen_write_msr_safe(msr
, low
, high
);
973 void xen_setup_shared_info(void)
975 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
976 set_fixmap(FIX_PARAVIRT_BOOTMAP
,
977 xen_start_info
->shared_info
);
979 HYPERVISOR_shared_info
=
980 (struct shared_info
*)fix_to_virt(FIX_PARAVIRT_BOOTMAP
);
982 HYPERVISOR_shared_info
=
983 (struct shared_info
*)__va(xen_start_info
->shared_info
);
986 /* In UP this is as good a place as any to set up shared info */
987 xen_setup_vcpu_info_placement();
990 xen_setup_mfn_list_list();
993 /* This is called once we have the cpu_possible_mask */
994 void xen_setup_vcpu_info_placement(void)
998 for_each_possible_cpu(cpu
) {
999 /* Set up direct vCPU id mapping for PV guests. */
1000 per_cpu(xen_vcpu_id
, cpu
) = cpu
;
1001 xen_vcpu_setup(cpu
);
1005 * xen_vcpu_setup managed to place the vcpu_info within the
1006 * percpu area for all cpus, so make use of it.
1008 if (xen_have_vcpu_info_placement
) {
1009 pv_irq_ops
.save_fl
= __PV_IS_CALLEE_SAVE(xen_save_fl_direct
);
1010 pv_irq_ops
.restore_fl
= __PV_IS_CALLEE_SAVE(xen_restore_fl_direct
);
1011 pv_irq_ops
.irq_disable
= __PV_IS_CALLEE_SAVE(xen_irq_disable_direct
);
1012 pv_irq_ops
.irq_enable
= __PV_IS_CALLEE_SAVE(xen_irq_enable_direct
);
1013 pv_mmu_ops
.read_cr2
= xen_read_cr2_direct
;
1017 static unsigned xen_patch(u8 type
, u16 clobbers
, void *insnbuf
,
1018 unsigned long addr
, unsigned len
)
1020 char *start
, *end
, *reloc
;
1023 start
= end
= reloc
= NULL
;
1025 #define SITE(op, x) \
1026 case PARAVIRT_PATCH(op.x): \
1027 if (xen_have_vcpu_info_placement) { \
1028 start = (char *)xen_##x##_direct; \
1029 end = xen_##x##_direct_end; \
1030 reloc = xen_##x##_direct_reloc; \
1035 SITE(pv_irq_ops
, irq_enable
);
1036 SITE(pv_irq_ops
, irq_disable
);
1037 SITE(pv_irq_ops
, save_fl
);
1038 SITE(pv_irq_ops
, restore_fl
);
1042 if (start
== NULL
|| (end
-start
) > len
)
1045 ret
= paravirt_patch_insns(insnbuf
, len
, start
, end
);
1047 /* Note: because reloc is assigned from something that
1048 appears to be an array, gcc assumes it's non-null,
1049 but doesn't know its relationship with start and
1051 if (reloc
> start
&& reloc
< end
) {
1052 int reloc_off
= reloc
- start
;
1053 long *relocp
= (long *)(insnbuf
+ reloc_off
);
1054 long delta
= start
- (char *)addr
;
1062 ret
= paravirt_patch_default(type
, clobbers
, insnbuf
,
1070 static const struct pv_info xen_info __initconst
= {
1071 .shared_kernel_pmd
= 0,
1073 #ifdef CONFIG_X86_64
1074 .extra_user_64bit_cs
= FLAT_USER_CS64
,
1079 static const struct pv_init_ops xen_init_ops __initconst
= {
1083 static const struct pv_cpu_ops xen_cpu_ops __initconst
= {
1086 .set_debugreg
= xen_set_debugreg
,
1087 .get_debugreg
= xen_get_debugreg
,
1089 .read_cr0
= xen_read_cr0
,
1090 .write_cr0
= xen_write_cr0
,
1092 .read_cr4
= native_read_cr4
,
1093 .write_cr4
= xen_write_cr4
,
1095 #ifdef CONFIG_X86_64
1096 .read_cr8
= xen_read_cr8
,
1097 .write_cr8
= xen_write_cr8
,
1100 .wbinvd
= native_wbinvd
,
1102 .read_msr
= xen_read_msr
,
1103 .write_msr
= xen_write_msr
,
1105 .read_msr_safe
= xen_read_msr_safe
,
1106 .write_msr_safe
= xen_write_msr_safe
,
1108 .read_pmc
= xen_read_pmc
,
1111 #ifdef CONFIG_X86_64
1112 .usergs_sysret64
= xen_sysret64
,
1115 .load_tr_desc
= paravirt_nop
,
1116 .set_ldt
= xen_set_ldt
,
1117 .load_gdt
= xen_load_gdt
,
1118 .load_idt
= xen_load_idt
,
1119 .load_tls
= xen_load_tls
,
1120 #ifdef CONFIG_X86_64
1121 .load_gs_index
= xen_load_gs_index
,
1124 .alloc_ldt
= xen_alloc_ldt
,
1125 .free_ldt
= xen_free_ldt
,
1127 .store_idt
= native_store_idt
,
1128 .store_tr
= xen_store_tr
,
1130 .write_ldt_entry
= xen_write_ldt_entry
,
1131 .write_gdt_entry
= xen_write_gdt_entry
,
1132 .write_idt_entry
= xen_write_idt_entry
,
1133 .load_sp0
= xen_load_sp0
,
1135 .set_iopl_mask
= xen_set_iopl_mask
,
1136 .io_delay
= xen_io_delay
,
1138 /* Xen takes care of %gs when switching to usermode for us */
1139 .swapgs
= paravirt_nop
,
1141 .start_context_switch
= paravirt_start_context_switch
,
1142 .end_context_switch
= xen_end_context_switch
,
1145 static void xen_restart(char *msg
)
1147 xen_reboot(SHUTDOWN_reboot
);
1150 static void xen_machine_halt(void)
1152 xen_reboot(SHUTDOWN_poweroff
);
1155 static void xen_machine_power_off(void)
1159 xen_reboot(SHUTDOWN_poweroff
);
1162 static void xen_crash_shutdown(struct pt_regs
*regs
)
1164 xen_reboot(SHUTDOWN_crash
);
1167 static const struct machine_ops xen_machine_ops __initconst
= {
1168 .restart
= xen_restart
,
1169 .halt
= xen_machine_halt
,
1170 .power_off
= xen_machine_power_off
,
1171 .shutdown
= xen_machine_halt
,
1172 .crash_shutdown
= xen_crash_shutdown
,
1173 .emergency_restart
= xen_emergency_restart
,
1176 static unsigned char xen_get_nmi_reason(void)
1178 unsigned char reason
= 0;
1180 /* Construct a value which looks like it came from port 0x61. */
1181 if (test_bit(_XEN_NMIREASON_io_error
,
1182 &HYPERVISOR_shared_info
->arch
.nmi_reason
))
1183 reason
|= NMI_REASON_IOCHK
;
1184 if (test_bit(_XEN_NMIREASON_pci_serr
,
1185 &HYPERVISOR_shared_info
->arch
.nmi_reason
))
1186 reason
|= NMI_REASON_SERR
;
1191 static void __init
xen_boot_params_init_edd(void)
1193 #if IS_ENABLED(CONFIG_EDD)
1194 struct xen_platform_op op
;
1195 struct edd_info
*edd_info
;
1200 edd_info
= boot_params
.eddbuf
;
1201 mbr_signature
= boot_params
.edd_mbr_sig_buffer
;
1203 op
.cmd
= XENPF_firmware_info
;
1205 op
.u
.firmware_info
.type
= XEN_FW_DISK_INFO
;
1206 for (nr
= 0; nr
< EDDMAXNR
; nr
++) {
1207 struct edd_info
*info
= edd_info
+ nr
;
1209 op
.u
.firmware_info
.index
= nr
;
1210 info
->params
.length
= sizeof(info
->params
);
1211 set_xen_guest_handle(op
.u
.firmware_info
.u
.disk_info
.edd_params
,
1213 ret
= HYPERVISOR_platform_op(&op
);
1217 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1220 C(interface_support
);
1221 C(legacy_max_cylinder
);
1223 C(legacy_sectors_per_track
);
1226 boot_params
.eddbuf_entries
= nr
;
1228 op
.u
.firmware_info
.type
= XEN_FW_DISK_MBR_SIGNATURE
;
1229 for (nr
= 0; nr
< EDD_MBR_SIG_MAX
; nr
++) {
1230 op
.u
.firmware_info
.index
= nr
;
1231 ret
= HYPERVISOR_platform_op(&op
);
1234 mbr_signature
[nr
] = op
.u
.firmware_info
.u
.disk_mbr_signature
.mbr_signature
;
1236 boot_params
.edd_mbr_sig_buf_entries
= nr
;
1241 * Set up the GDT and segment registers for -fstack-protector. Until
1242 * we do this, we have to be careful not to call any stack-protected
1243 * function, which is most of the kernel.
1245 static void xen_setup_gdt(int cpu
)
1247 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry_boot
;
1248 pv_cpu_ops
.load_gdt
= xen_load_gdt_boot
;
1250 setup_stack_canary_segment(0);
1251 switch_to_new_gdt(0);
1253 pv_cpu_ops
.write_gdt_entry
= xen_write_gdt_entry
;
1254 pv_cpu_ops
.load_gdt
= xen_load_gdt
;
1257 static void __init
xen_dom0_set_legacy_features(void)
1259 x86_platform
.legacy
.rtc
= 1;
1262 /* First C function to be called on Xen boot */
1263 asmlinkage __visible
void __init
xen_start_kernel(void)
1265 struct physdev_set_iopl set_iopl
;
1266 unsigned long initrd_start
= 0;
1269 if (!xen_start_info
)
1272 xen_domain_type
= XEN_PV_DOMAIN
;
1274 xen_setup_features();
1276 xen_setup_machphys_mapping();
1278 /* Install Xen paravirt ops */
1280 pv_init_ops
= xen_init_ops
;
1281 pv_cpu_ops
= xen_cpu_ops
;
1283 x86_platform
.get_nmi_reason
= xen_get_nmi_reason
;
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
;
1301 * Prevent page tables from being allocated in highmem, even
1302 * if CONFIG_HIGHPTE is enabled.
1304 __userpte_alloc_gfp
&= ~__GFP_HIGHMEM
;
1306 /* Work out if we support NX */
1310 xen_build_dynamic_phys_to_machine();
1313 * Set up kernel GDT and segment registers, mainly so that
1314 * -fstack-protector code can be executed.
1319 xen_init_capabilities();
1321 #ifdef CONFIG_X86_LOCAL_APIC
1323 * set up the basic apic ops.
1328 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad
)) {
1329 pv_mmu_ops
.ptep_modify_prot_start
= xen_ptep_modify_prot_start
;
1330 pv_mmu_ops
.ptep_modify_prot_commit
= xen_ptep_modify_prot_commit
;
1333 machine_ops
= xen_machine_ops
;
1336 * The only reliable way to retain the initial address of the
1337 * percpu gdt_page is to remember it here, so we can go and
1338 * mark it RW later, when the initial percpu area is freed.
1340 xen_initial_gdt
= &per_cpu(gdt_page
, 0);
1344 #ifdef CONFIG_ACPI_NUMA
1346 * The pages we from Xen are not related to machine pages, so
1347 * any NUMA information the kernel tries to get from ACPI will
1348 * be meaningless. Prevent it from trying.
1352 /* Don't do the full vcpu_info placement stuff until we have a
1353 possible map and a non-dummy shared_info. */
1354 per_cpu(xen_vcpu
, 0) = &HYPERVISOR_shared_info
->vcpu_info
[0];
1356 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv
, xen_cpu_dead_pv
));
1358 local_irq_disable();
1359 early_boot_irqs_disabled
= true;
1361 xen_raw_console_write("mapping kernel into physical memory\n");
1362 xen_setup_kernel_pagetable((pgd_t
*)xen_start_info
->pt_base
,
1363 xen_start_info
->nr_pages
);
1364 xen_reserve_special_pages();
1366 /* keep using Xen gdt for now; no urgent need to change it */
1368 #ifdef CONFIG_X86_32
1369 pv_info
.kernel_rpl
= 1;
1370 if (xen_feature(XENFEAT_supervisor_mode_kernel
))
1371 pv_info
.kernel_rpl
= 0;
1373 pv_info
.kernel_rpl
= 0;
1375 /* set the limit of our address space */
1379 * We used to do this in xen_arch_setup, but that is too late
1380 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1381 * early_amd_init which pokes 0xcf8 port.
1384 rc
= HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl
, &set_iopl
);
1386 xen_raw_printk("physdev_op failed %d\n", rc
);
1388 #ifdef CONFIG_X86_32
1389 /* set up basic CPUID stuff */
1390 cpu_detect(&new_cpu_data
);
1391 set_cpu_cap(&new_cpu_data
, X86_FEATURE_FPU
);
1392 new_cpu_data
.x86_capability
[CPUID_1_EDX
] = cpuid_edx(1);
1395 if (xen_start_info
->mod_start
) {
1396 if (xen_start_info
->flags
& SIF_MOD_START_PFN
)
1397 initrd_start
= PFN_PHYS(xen_start_info
->mod_start
);
1399 initrd_start
= __pa(xen_start_info
->mod_start
);
1402 /* Poke various useful things into boot_params */
1403 boot_params
.hdr
.type_of_loader
= (9 << 4) | 0;
1404 boot_params
.hdr
.ramdisk_image
= initrd_start
;
1405 boot_params
.hdr
.ramdisk_size
= xen_start_info
->mod_len
;
1406 boot_params
.hdr
.cmd_line_ptr
= __pa(xen_start_info
->cmd_line
);
1407 boot_params
.hdr
.hardware_subarch
= X86_SUBARCH_XEN
;
1409 if (!xen_initial_domain()) {
1410 add_preferred_console("xenboot", 0, NULL
);
1411 add_preferred_console("tty", 0, NULL
);
1412 add_preferred_console("hvc", 0, NULL
);
1414 x86_init
.pci
.arch_init
= pci_xen_init
;
1416 const struct dom0_vga_console_info
*info
=
1417 (void *)((char *)xen_start_info
+
1418 xen_start_info
->console
.dom0
.info_off
);
1419 struct xen_platform_op op
= {
1420 .cmd
= XENPF_firmware_info
,
1421 .interface_version
= XENPF_INTERFACE_VERSION
,
1422 .u
.firmware_info
.type
= XEN_FW_KBD_SHIFT_FLAGS
,
1425 x86_platform
.set_legacy_features
=
1426 xen_dom0_set_legacy_features
;
1427 xen_init_vga(info
, xen_start_info
->console
.dom0
.info_size
);
1428 xen_start_info
->console
.domU
.mfn
= 0;
1429 xen_start_info
->console
.domU
.evtchn
= 0;
1431 if (HYPERVISOR_platform_op(&op
) == 0)
1432 boot_params
.kbd_status
= op
.u
.firmware_info
.u
.kbd_shift_flags
;
1434 /* Make sure ACS will be enabled */
1437 xen_acpi_sleep_register();
1439 /* Avoid searching for BIOS MP tables */
1440 x86_init
.mpparse
.find_smp_config
= x86_init_noop
;
1441 x86_init
.mpparse
.get_smp_config
= x86_init_uint_noop
;
1443 xen_boot_params_init_edd();
1446 /* PCI BIOS service won't work from a PV guest. */
1447 pci_probe
&= ~PCI_PROBE_BIOS
;
1449 xen_raw_console_write("about to get started...\n");
1451 /* Let's presume PV guests always boot on vCPU with id 0. */
1452 per_cpu(xen_vcpu_id
, 0) = 0;
1454 xen_setup_runstate_info(0);
1458 /* Start the world */
1459 #ifdef CONFIG_X86_32
1460 i386_start_kernel();
1462 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1463 x86_64_start_reservations((char *)__pa_symbol(&boot_params
));
1467 static int xen_cpu_up_prepare_pv(unsigned int cpu
)
1471 xen_setup_timer(cpu
);
1473 rc
= xen_smp_intr_init(cpu
);
1475 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1480 rc
= xen_smp_intr_init_pv(cpu
);
1482 WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n",
1490 static int xen_cpu_dead_pv(unsigned int cpu
)
1492 xen_smp_intr_free(cpu
);
1493 xen_smp_intr_free_pv(cpu
);
1495 xen_teardown_timer(cpu
);
1500 static uint32_t __init
xen_platform_pv(void)
1502 if (xen_pv_domain())
1503 return xen_cpuid_base();
1508 const struct hypervisor_x86 x86_hyper_xen_pv
= {
1510 .detect
= xen_platform_pv
,
1511 .pin_vcpu
= xen_pin_vcpu
,
1513 EXPORT_SYMBOL(x86_hyper_xen_pv
);