e7c920bfbc9dd742fc480010c0840809b55b3bde
[pandora-kernel.git] / arch / x86 / xen / enlighten.c
1 /*
2  * Core of Xen paravirt_ops implementation.
3  *
4  * This file contains the xen_paravirt_ops structure itself, and the
5  * implementations for:
6  * - privileged instructions
7  * - interrupt flags
8  * - segment operations
9  * - booting and setup
10  *
11  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
12  */
13
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>
27 #include <linux/mm.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
35 #include <xen/xen.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>
42 #include <xen/page.h>
43 #include <xen/hvm.h>
44 #include <xen/hvc-console.h>
45
46 #include <asm/paravirt.h>
47 #include <asm/apic.h>
48 #include <asm/page.h>
49 #include <asm/xen/pci.h>
50 #include <asm/xen/hypercall.h>
51 #include <asm/xen/hypervisor.h>
52 #include <asm/fixmap.h>
53 #include <asm/processor.h>
54 #include <asm/proto.h>
55 #include <asm/msr-index.h>
56 #include <asm/traps.h>
57 #include <asm/setup.h>
58 #include <asm/desc.h>
59 #include <asm/pgalloc.h>
60 #include <asm/pgtable.h>
61 #include <asm/tlbflush.h>
62 #include <asm/reboot.h>
63 #include <asm/stackprotector.h>
64 #include <asm/hypervisor.h>
65 #include <asm/pci_x86.h>
66
67 #include "xen-ops.h"
68 #include "mmu.h"
69 #include "multicalls.h"
70
71 EXPORT_SYMBOL_GPL(hypercall_page);
72
73 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
74 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
75
76 enum xen_domain_type xen_domain_type = XEN_NATIVE;
77 EXPORT_SYMBOL_GPL(xen_domain_type);
78
79 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
80 EXPORT_SYMBOL(machine_to_phys_mapping);
81 unsigned long  machine_to_phys_nr;
82 EXPORT_SYMBOL(machine_to_phys_nr);
83
84 struct start_info *xen_start_info;
85 EXPORT_SYMBOL_GPL(xen_start_info);
86
87 struct shared_info xen_dummy_shared_info;
88
89 void *xen_initial_gdt;
90
91 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
92 __read_mostly int xen_have_vector_callback;
93 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
94
95 /*
96  * Point at some empty memory to start with. We map the real shared_info
97  * page as soon as fixmap is up and running.
98  */
99 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
100
101 /*
102  * Flag to determine whether vcpu info placement is available on all
103  * VCPUs.  We assume it is to start with, and then set it to zero on
104  * the first failure.  This is because it can succeed on some VCPUs
105  * and not others, since it can involve hypervisor memory allocation,
106  * or because the guest failed to guarantee all the appropriate
107  * constraints on all VCPUs (ie buffer can't cross a page boundary).
108  *
109  * Note that any particular CPU may be using a placed vcpu structure,
110  * but we can only optimise if the all are.
111  *
112  * 0: not available, 1: available
113  */
114 static int have_vcpu_info_placement = 1;
115
116 static void clamp_max_cpus(void)
117 {
118 #ifdef CONFIG_SMP
119         if (setup_max_cpus > MAX_VIRT_CPUS)
120                 setup_max_cpus = MAX_VIRT_CPUS;
121 #endif
122 }
123
124 static void xen_vcpu_setup(int cpu)
125 {
126         struct vcpu_register_vcpu_info info;
127         int err;
128         struct vcpu_info *vcpup;
129
130         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
131
132         if (cpu < MAX_VIRT_CPUS)
133                 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
134
135         if (!have_vcpu_info_placement) {
136                 if (cpu >= MAX_VIRT_CPUS)
137                         clamp_max_cpus();
138                 return;
139         }
140
141         vcpup = &per_cpu(xen_vcpu_info, cpu);
142         info.mfn = arbitrary_virt_to_mfn(vcpup);
143         info.offset = offset_in_page(vcpup);
144
145         /* Check to see if the hypervisor will put the vcpu_info
146            structure where we want it, which allows direct access via
147            a percpu-variable. */
148         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
149
150         if (err) {
151                 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
152                 have_vcpu_info_placement = 0;
153                 clamp_max_cpus();
154         } else {
155                 /* This cpu is using the registered vcpu info, even if
156                    later ones fail to. */
157                 per_cpu(xen_vcpu, cpu) = vcpup;
158         }
159 }
160
161 /*
162  * On restore, set the vcpu placement up again.
163  * If it fails, then we're in a bad state, since
164  * we can't back out from using it...
165  */
166 void xen_vcpu_restore(void)
167 {
168         int cpu;
169
170         for_each_online_cpu(cpu) {
171                 bool other_cpu = (cpu != smp_processor_id());
172
173                 if (other_cpu &&
174                     HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
175                         BUG();
176
177                 xen_setup_runstate_info(cpu);
178
179                 if (have_vcpu_info_placement)
180                         xen_vcpu_setup(cpu);
181
182                 if (other_cpu &&
183                     HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
184                         BUG();
185         }
186 }
187
188 static void __init xen_banner(void)
189 {
190         unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
191         struct xen_extraversion extra;
192         HYPERVISOR_xen_version(XENVER_extraversion, &extra);
193
194         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
195                pv_info.name);
196         printk(KERN_INFO "Xen version: %d.%d%s%s\n",
197                version >> 16, version & 0xffff, extra.extraversion,
198                xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
199 }
200
201 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
202 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
203
204 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
205                       unsigned int *cx, unsigned int *dx)
206 {
207         unsigned maskebx = ~0;
208         unsigned maskecx = ~0;
209         unsigned maskedx = ~0;
210
211         /*
212          * Mask out inconvenient features, to try and disable as many
213          * unsupported kernel subsystems as possible.
214          */
215         switch (*ax) {
216         case 1:
217                 maskecx = cpuid_leaf1_ecx_mask;
218                 maskedx = cpuid_leaf1_edx_mask;
219                 break;
220
221         case 0xb:
222                 /* Suppress extended topology stuff */
223                 maskebx = 0;
224                 break;
225         }
226
227         asm(XEN_EMULATE_PREFIX "cpuid"
228                 : "=a" (*ax),
229                   "=b" (*bx),
230                   "=c" (*cx),
231                   "=d" (*dx)
232                 : "0" (*ax), "2" (*cx));
233
234         *bx &= maskebx;
235         *cx &= maskecx;
236         *dx &= maskedx;
237 }
238
239 static void __init xen_init_cpuid_mask(void)
240 {
241         unsigned int ax, bx, cx, dx;
242         unsigned int xsave_mask;
243
244         cpuid_leaf1_edx_mask =
245                 ~((1 << X86_FEATURE_MCE)  |  /* disable MCE */
246                   (1 << X86_FEATURE_MCA)  |  /* disable MCA */
247                   (1 << X86_FEATURE_MTRR) |  /* disable MTRR */
248                   (1 << X86_FEATURE_ACC));   /* thermal monitoring */
249
250         if (!xen_initial_domain())
251                 cpuid_leaf1_edx_mask &=
252                         ~((1 << X86_FEATURE_APIC) |  /* disable local APIC */
253                           (1 << X86_FEATURE_ACPI));  /* disable ACPI */
254         ax = 1;
255         cx = 0;
256         xen_cpuid(&ax, &bx, &cx, &dx);
257
258         xsave_mask =
259                 (1 << (X86_FEATURE_XSAVE % 32)) |
260                 (1 << (X86_FEATURE_OSXSAVE % 32));
261
262         /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
263         if ((cx & xsave_mask) != xsave_mask)
264                 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
265 }
266
267 static void xen_set_debugreg(int reg, unsigned long val)
268 {
269         HYPERVISOR_set_debugreg(reg, val);
270 }
271
272 static unsigned long xen_get_debugreg(int reg)
273 {
274         return HYPERVISOR_get_debugreg(reg);
275 }
276
277 static void xen_end_context_switch(struct task_struct *next)
278 {
279         xen_mc_flush();
280         paravirt_end_context_switch(next);
281 }
282
283 static unsigned long xen_store_tr(void)
284 {
285         return 0;
286 }
287
288 /*
289  * Set the page permissions for a particular virtual address.  If the
290  * address is a vmalloc mapping (or other non-linear mapping), then
291  * find the linear mapping of the page and also set its protections to
292  * match.
293  */
294 static void set_aliased_prot(void *v, pgprot_t prot)
295 {
296         int level;
297         pte_t *ptep;
298         pte_t pte;
299         unsigned long pfn;
300         struct page *page;
301
302         ptep = lookup_address((unsigned long)v, &level);
303         BUG_ON(ptep == NULL);
304
305         pfn = pte_pfn(*ptep);
306         page = pfn_to_page(pfn);
307
308         pte = pfn_pte(pfn, prot);
309
310         if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
311                 BUG();
312
313         if (!PageHighMem(page)) {
314                 void *av = __va(PFN_PHYS(pfn));
315
316                 if (av != v)
317                         if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
318                                 BUG();
319         } else
320                 kmap_flush_unused();
321 }
322
323 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
324 {
325         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
326         int i;
327
328         for(i = 0; i < entries; i += entries_per_page)
329                 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
330 }
331
332 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
333 {
334         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
335         int i;
336
337         for(i = 0; i < entries; i += entries_per_page)
338                 set_aliased_prot(ldt + i, PAGE_KERNEL);
339 }
340
341 static void xen_set_ldt(const void *addr, unsigned entries)
342 {
343         struct mmuext_op *op;
344         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
345
346         trace_xen_cpu_set_ldt(addr, entries);
347
348         op = mcs.args;
349         op->cmd = MMUEXT_SET_LDT;
350         op->arg1.linear_addr = (unsigned long)addr;
351         op->arg2.nr_ents = entries;
352
353         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
354
355         xen_mc_issue(PARAVIRT_LAZY_CPU);
356 }
357
358 static void xen_load_gdt(const struct desc_ptr *dtr)
359 {
360         unsigned long va = dtr->address;
361         unsigned int size = dtr->size + 1;
362         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
363         unsigned long frames[pages];
364         int f;
365
366         /*
367          * A GDT can be up to 64k in size, which corresponds to 8192
368          * 8-byte entries, or 16 4k pages..
369          */
370
371         BUG_ON(size > 65536);
372         BUG_ON(va & ~PAGE_MASK);
373
374         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
375                 int level;
376                 pte_t *ptep;
377                 unsigned long pfn, mfn;
378                 void *virt;
379
380                 /*
381                  * The GDT is per-cpu and is in the percpu data area.
382                  * That can be virtually mapped, so we need to do a
383                  * page-walk to get the underlying MFN for the
384                  * hypercall.  The page can also be in the kernel's
385                  * linear range, so we need to RO that mapping too.
386                  */
387                 ptep = lookup_address(va, &level);
388                 BUG_ON(ptep == NULL);
389
390                 pfn = pte_pfn(*ptep);
391                 mfn = pfn_to_mfn(pfn);
392                 virt = __va(PFN_PHYS(pfn));
393
394                 frames[f] = mfn;
395
396                 make_lowmem_page_readonly((void *)va);
397                 make_lowmem_page_readonly(virt);
398         }
399
400         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
401                 BUG();
402 }
403
404 /*
405  * load_gdt for early boot, when the gdt is only mapped once
406  */
407 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
408 {
409         unsigned long va = dtr->address;
410         unsigned int size = dtr->size + 1;
411         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
412         unsigned long frames[pages];
413         int f;
414
415         /*
416          * A GDT can be up to 64k in size, which corresponds to 8192
417          * 8-byte entries, or 16 4k pages..
418          */
419
420         BUG_ON(size > 65536);
421         BUG_ON(va & ~PAGE_MASK);
422
423         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
424                 pte_t pte;
425                 unsigned long pfn, mfn;
426
427                 pfn = virt_to_pfn(va);
428                 mfn = pfn_to_mfn(pfn);
429
430                 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
431
432                 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
433                         BUG();
434
435                 frames[f] = mfn;
436         }
437
438         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
439                 BUG();
440 }
441
442 static void load_TLS_descriptor(struct thread_struct *t,
443                                 unsigned int cpu, unsigned int i)
444 {
445         struct desc_struct *gdt = get_cpu_gdt_table(cpu);
446         xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
447         struct multicall_space mc = __xen_mc_entry(0);
448
449         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
450 }
451
452 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
453 {
454         /*
455          * XXX sleazy hack: If we're being called in a lazy-cpu zone
456          * and lazy gs handling is enabled, it means we're in a
457          * context switch, and %gs has just been saved.  This means we
458          * can zero it out to prevent faults on exit from the
459          * hypervisor if the next process has no %gs.  Either way, it
460          * has been saved, and the new value will get loaded properly.
461          * This will go away as soon as Xen has been modified to not
462          * save/restore %gs for normal hypercalls.
463          *
464          * On x86_64, this hack is not used for %gs, because gs points
465          * to KERNEL_GS_BASE (and uses it for PDA references), so we
466          * must not zero %gs on x86_64
467          *
468          * For x86_64, we need to zero %fs, otherwise we may get an
469          * exception between the new %fs descriptor being loaded and
470          * %fs being effectively cleared at __switch_to().
471          */
472         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
473 #ifdef CONFIG_X86_32
474                 lazy_load_gs(0);
475 #else
476                 loadsegment(fs, 0);
477 #endif
478         }
479
480         xen_mc_batch();
481
482         load_TLS_descriptor(t, cpu, 0);
483         load_TLS_descriptor(t, cpu, 1);
484         load_TLS_descriptor(t, cpu, 2);
485
486         xen_mc_issue(PARAVIRT_LAZY_CPU);
487 }
488
489 #ifdef CONFIG_X86_64
490 static void xen_load_gs_index(unsigned int idx)
491 {
492         if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
493                 BUG();
494 }
495 #endif
496
497 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
498                                 const void *ptr)
499 {
500         xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
501         u64 entry = *(u64 *)ptr;
502
503         trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
504
505         preempt_disable();
506
507         xen_mc_flush();
508         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
509                 BUG();
510
511         preempt_enable();
512 }
513
514 static int cvt_gate_to_trap(int vector, const gate_desc *val,
515                             struct trap_info *info)
516 {
517         unsigned long addr;
518
519         if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
520                 return 0;
521
522         info->vector = vector;
523
524         addr = gate_offset(*val);
525 #ifdef CONFIG_X86_64
526         /*
527          * Look for known traps using IST, and substitute them
528          * appropriately.  The debugger ones are the only ones we care
529          * about.  Xen will handle faults like double_fault and
530          * machine_check, so we should never see them.  Warn if
531          * there's an unexpected IST-using fault handler.
532          */
533         if (addr == (unsigned long)debug)
534                 addr = (unsigned long)xen_debug;
535         else if (addr == (unsigned long)int3)
536                 addr = (unsigned long)xen_int3;
537         else if (addr == (unsigned long)stack_segment)
538                 addr = (unsigned long)xen_stack_segment;
539         else if (addr == (unsigned long)double_fault ||
540                  addr == (unsigned long)nmi) {
541                 /* Don't need to handle these */
542                 return 0;
543 #ifdef CONFIG_X86_MCE
544         } else if (addr == (unsigned long)machine_check) {
545                 return 0;
546 #endif
547         } else {
548                 /* Some other trap using IST? */
549                 if (WARN_ON(val->ist != 0))
550                         return 0;
551         }
552 #endif  /* CONFIG_X86_64 */
553         info->address = addr;
554
555         info->cs = gate_segment(*val);
556         info->flags = val->dpl;
557         /* interrupt gates clear IF */
558         if (val->type == GATE_INTERRUPT)
559                 info->flags |= 1 << 2;
560
561         return 1;
562 }
563
564 /* Locations of each CPU's IDT */
565 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
566
567 /* Set an IDT entry.  If the entry is part of the current IDT, then
568    also update Xen. */
569 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
570 {
571         unsigned long p = (unsigned long)&dt[entrynum];
572         unsigned long start, end;
573
574         trace_xen_cpu_write_idt_entry(dt, entrynum, g);
575
576         preempt_disable();
577
578         start = __this_cpu_read(idt_desc.address);
579         end = start + __this_cpu_read(idt_desc.size) + 1;
580
581         xen_mc_flush();
582
583         native_write_idt_entry(dt, entrynum, g);
584
585         if (p >= start && (p + 8) <= end) {
586                 struct trap_info info[2];
587
588                 info[1].address = 0;
589
590                 if (cvt_gate_to_trap(entrynum, g, &info[0]))
591                         if (HYPERVISOR_set_trap_table(info))
592                                 BUG();
593         }
594
595         preempt_enable();
596 }
597
598 static void xen_convert_trap_info(const struct desc_ptr *desc,
599                                   struct trap_info *traps)
600 {
601         unsigned in, out, count;
602
603         count = (desc->size+1) / sizeof(gate_desc);
604         BUG_ON(count > 256);
605
606         for (in = out = 0; in < count; in++) {
607                 gate_desc *entry = (gate_desc*)(desc->address) + in;
608
609                 if (cvt_gate_to_trap(in, entry, &traps[out]))
610                         out++;
611         }
612         traps[out].address = 0;
613 }
614
615 void xen_copy_trap_info(struct trap_info *traps)
616 {
617         const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
618
619         xen_convert_trap_info(desc, traps);
620 }
621
622 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
623    hold a spinlock to protect the static traps[] array (static because
624    it avoids allocation, and saves stack space). */
625 static void xen_load_idt(const struct desc_ptr *desc)
626 {
627         static DEFINE_SPINLOCK(lock);
628         static struct trap_info traps[257];
629
630         trace_xen_cpu_load_idt(desc);
631
632         spin_lock(&lock);
633
634         __get_cpu_var(idt_desc) = *desc;
635
636         xen_convert_trap_info(desc, traps);
637
638         xen_mc_flush();
639         if (HYPERVISOR_set_trap_table(traps))
640                 BUG();
641
642         spin_unlock(&lock);
643 }
644
645 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
646    they're handled differently. */
647 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
648                                 const void *desc, int type)
649 {
650         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
651
652         preempt_disable();
653
654         switch (type) {
655         case DESC_LDT:
656         case DESC_TSS:
657                 /* ignore */
658                 break;
659
660         default: {
661                 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
662
663                 xen_mc_flush();
664                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
665                         BUG();
666         }
667
668         }
669
670         preempt_enable();
671 }
672
673 /*
674  * Version of write_gdt_entry for use at early boot-time needed to
675  * update an entry as simply as possible.
676  */
677 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
678                                             const void *desc, int type)
679 {
680         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
681
682         switch (type) {
683         case DESC_LDT:
684         case DESC_TSS:
685                 /* ignore */
686                 break;
687
688         default: {
689                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
690
691                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
692                         dt[entry] = *(struct desc_struct *)desc;
693         }
694
695         }
696 }
697
698 static void xen_load_sp0(struct tss_struct *tss,
699                          struct thread_struct *thread)
700 {
701         struct multicall_space mcs;
702
703         mcs = xen_mc_entry(0);
704         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
705         xen_mc_issue(PARAVIRT_LAZY_CPU);
706 }
707
708 static void xen_set_iopl_mask(unsigned mask)
709 {
710         struct physdev_set_iopl set_iopl;
711
712         /* Force the change at ring 0. */
713         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
714         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
715 }
716
717 static void xen_io_delay(void)
718 {
719 }
720
721 #ifdef CONFIG_X86_LOCAL_APIC
722 static u32 xen_apic_read(u32 reg)
723 {
724         return 0;
725 }
726
727 static void xen_apic_write(u32 reg, u32 val)
728 {
729         /* Warn to see if there's any stray references */
730         WARN_ON(1);
731 }
732
733 static u64 xen_apic_icr_read(void)
734 {
735         return 0;
736 }
737
738 static void xen_apic_icr_write(u32 low, u32 id)
739 {
740         /* Warn to see if there's any stray references */
741         WARN_ON(1);
742 }
743
744 static void xen_apic_wait_icr_idle(void)
745 {
746         return;
747 }
748
749 static u32 xen_safe_apic_wait_icr_idle(void)
750 {
751         return 0;
752 }
753
754 static void set_xen_basic_apic_ops(void)
755 {
756         apic->read = xen_apic_read;
757         apic->write = xen_apic_write;
758         apic->icr_read = xen_apic_icr_read;
759         apic->icr_write = xen_apic_icr_write;
760         apic->wait_icr_idle = xen_apic_wait_icr_idle;
761         apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
762 }
763
764 #endif
765
766 static void xen_clts(void)
767 {
768         struct multicall_space mcs;
769
770         mcs = xen_mc_entry(0);
771
772         MULTI_fpu_taskswitch(mcs.mc, 0);
773
774         xen_mc_issue(PARAVIRT_LAZY_CPU);
775 }
776
777 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
778
779 static unsigned long xen_read_cr0(void)
780 {
781         unsigned long cr0 = percpu_read(xen_cr0_value);
782
783         if (unlikely(cr0 == 0)) {
784                 cr0 = native_read_cr0();
785                 percpu_write(xen_cr0_value, cr0);
786         }
787
788         return cr0;
789 }
790
791 static void xen_write_cr0(unsigned long cr0)
792 {
793         struct multicall_space mcs;
794
795         percpu_write(xen_cr0_value, cr0);
796
797         /* Only pay attention to cr0.TS; everything else is
798            ignored. */
799         mcs = xen_mc_entry(0);
800
801         MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
802
803         xen_mc_issue(PARAVIRT_LAZY_CPU);
804 }
805
806 static void xen_write_cr4(unsigned long cr4)
807 {
808         cr4 &= ~X86_CR4_PGE;
809         cr4 &= ~X86_CR4_PSE;
810
811         native_write_cr4(cr4);
812 }
813
814 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
815 {
816         int ret;
817
818         ret = 0;
819
820         switch (msr) {
821 #ifdef CONFIG_X86_64
822                 unsigned which;
823                 u64 base;
824
825         case MSR_FS_BASE:               which = SEGBASE_FS; goto set;
826         case MSR_KERNEL_GS_BASE:        which = SEGBASE_GS_USER; goto set;
827         case MSR_GS_BASE:               which = SEGBASE_GS_KERNEL; goto set;
828
829         set:
830                 base = ((u64)high << 32) | low;
831                 if (HYPERVISOR_set_segment_base(which, base) != 0)
832                         ret = -EIO;
833                 break;
834 #endif
835
836         case MSR_STAR:
837         case MSR_CSTAR:
838         case MSR_LSTAR:
839         case MSR_SYSCALL_MASK:
840         case MSR_IA32_SYSENTER_CS:
841         case MSR_IA32_SYSENTER_ESP:
842         case MSR_IA32_SYSENTER_EIP:
843                 /* Fast syscall setup is all done in hypercalls, so
844                    these are all ignored.  Stub them out here to stop
845                    Xen console noise. */
846                 break;
847
848         case MSR_IA32_CR_PAT:
849                 if (smp_processor_id() == 0)
850                         xen_set_pat(((u64)high << 32) | low);
851                 break;
852
853         default:
854                 ret = native_write_msr_safe(msr, low, high);
855         }
856
857         return ret;
858 }
859
860 void xen_setup_shared_info(void)
861 {
862         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
863                 set_fixmap(FIX_PARAVIRT_BOOTMAP,
864                            xen_start_info->shared_info);
865
866                 HYPERVISOR_shared_info =
867                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
868         } else
869                 HYPERVISOR_shared_info =
870                         (struct shared_info *)__va(xen_start_info->shared_info);
871
872 #ifndef CONFIG_SMP
873         /* In UP this is as good a place as any to set up shared info */
874         xen_setup_vcpu_info_placement();
875 #endif
876
877         xen_setup_mfn_list_list();
878 }
879
880 /* This is called once we have the cpu_possible_map */
881 void xen_setup_vcpu_info_placement(void)
882 {
883         int cpu;
884
885         for_each_possible_cpu(cpu)
886                 xen_vcpu_setup(cpu);
887
888         /* xen_vcpu_setup managed to place the vcpu_info within the
889            percpu area for all cpus, so make use of it */
890         if (have_vcpu_info_placement) {
891                 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
892                 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
893                 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
894                 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
895                 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
896         }
897 }
898
899 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
900                           unsigned long addr, unsigned len)
901 {
902         char *start, *end, *reloc;
903         unsigned ret;
904
905         start = end = reloc = NULL;
906
907 #define SITE(op, x)                                                     \
908         case PARAVIRT_PATCH(op.x):                                      \
909         if (have_vcpu_info_placement) {                                 \
910                 start = (char *)xen_##x##_direct;                       \
911                 end = xen_##x##_direct_end;                             \
912                 reloc = xen_##x##_direct_reloc;                         \
913         }                                                               \
914         goto patch_site
915
916         switch (type) {
917                 SITE(pv_irq_ops, irq_enable);
918                 SITE(pv_irq_ops, irq_disable);
919                 SITE(pv_irq_ops, save_fl);
920                 SITE(pv_irq_ops, restore_fl);
921 #undef SITE
922
923         patch_site:
924                 if (start == NULL || (end-start) > len)
925                         goto default_patch;
926
927                 ret = paravirt_patch_insns(insnbuf, len, start, end);
928
929                 /* Note: because reloc is assigned from something that
930                    appears to be an array, gcc assumes it's non-null,
931                    but doesn't know its relationship with start and
932                    end. */
933                 if (reloc > start && reloc < end) {
934                         int reloc_off = reloc - start;
935                         long *relocp = (long *)(insnbuf + reloc_off);
936                         long delta = start - (char *)addr;
937
938                         *relocp += delta;
939                 }
940                 break;
941
942         default_patch:
943         default:
944                 ret = paravirt_patch_default(type, clobbers, insnbuf,
945                                              addr, len);
946                 break;
947         }
948
949         return ret;
950 }
951
952 static const struct pv_info xen_info __initconst = {
953         .paravirt_enabled = 1,
954         .shared_kernel_pmd = 0,
955
956 #ifdef CONFIG_X86_64
957         .extra_user_64bit_cs = FLAT_USER_CS64,
958 #endif
959
960         .name = "Xen",
961 };
962
963 static const struct pv_init_ops xen_init_ops __initconst = {
964         .patch = xen_patch,
965 };
966
967 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
968         .cpuid = xen_cpuid,
969
970         .set_debugreg = xen_set_debugreg,
971         .get_debugreg = xen_get_debugreg,
972
973         .clts = xen_clts,
974
975         .read_cr0 = xen_read_cr0,
976         .write_cr0 = xen_write_cr0,
977
978         .read_cr4 = native_read_cr4,
979         .read_cr4_safe = native_read_cr4_safe,
980         .write_cr4 = xen_write_cr4,
981
982         .wbinvd = native_wbinvd,
983
984         .read_msr = native_read_msr_safe,
985         .write_msr = xen_write_msr_safe,
986         .read_tsc = native_read_tsc,
987         .read_pmc = native_read_pmc,
988
989         .iret = xen_iret,
990         .irq_enable_sysexit = xen_sysexit,
991 #ifdef CONFIG_X86_64
992         .usergs_sysret32 = xen_sysret32,
993         .usergs_sysret64 = xen_sysret64,
994 #endif
995
996         .load_tr_desc = paravirt_nop,
997         .set_ldt = xen_set_ldt,
998         .load_gdt = xen_load_gdt,
999         .load_idt = xen_load_idt,
1000         .load_tls = xen_load_tls,
1001 #ifdef CONFIG_X86_64
1002         .load_gs_index = xen_load_gs_index,
1003 #endif
1004
1005         .alloc_ldt = xen_alloc_ldt,
1006         .free_ldt = xen_free_ldt,
1007
1008         .store_gdt = native_store_gdt,
1009         .store_idt = native_store_idt,
1010         .store_tr = xen_store_tr,
1011
1012         .write_ldt_entry = xen_write_ldt_entry,
1013         .write_gdt_entry = xen_write_gdt_entry,
1014         .write_idt_entry = xen_write_idt_entry,
1015         .load_sp0 = xen_load_sp0,
1016
1017         .set_iopl_mask = xen_set_iopl_mask,
1018         .io_delay = xen_io_delay,
1019
1020         /* Xen takes care of %gs when switching to usermode for us */
1021         .swapgs = paravirt_nop,
1022
1023         .start_context_switch = paravirt_start_context_switch,
1024         .end_context_switch = xen_end_context_switch,
1025 };
1026
1027 static const struct pv_apic_ops xen_apic_ops __initconst = {
1028 #ifdef CONFIG_X86_LOCAL_APIC
1029         .startup_ipi_hook = paravirt_nop,
1030 #endif
1031 };
1032
1033 static void xen_reboot(int reason)
1034 {
1035         struct sched_shutdown r = { .reason = reason };
1036
1037         if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1038                 BUG();
1039 }
1040
1041 static void xen_restart(char *msg)
1042 {
1043         xen_reboot(SHUTDOWN_reboot);
1044 }
1045
1046 static void xen_emergency_restart(void)
1047 {
1048         xen_reboot(SHUTDOWN_reboot);
1049 }
1050
1051 static void xen_machine_halt(void)
1052 {
1053         xen_reboot(SHUTDOWN_poweroff);
1054 }
1055
1056 static void xen_machine_power_off(void)
1057 {
1058         if (pm_power_off)
1059                 pm_power_off();
1060         xen_reboot(SHUTDOWN_poweroff);
1061 }
1062
1063 static void xen_crash_shutdown(struct pt_regs *regs)
1064 {
1065         xen_reboot(SHUTDOWN_crash);
1066 }
1067
1068 static int
1069 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1070 {
1071         xen_reboot(SHUTDOWN_crash);
1072         return NOTIFY_DONE;
1073 }
1074
1075 static struct notifier_block xen_panic_block = {
1076         .notifier_call= xen_panic_event,
1077 };
1078
1079 int xen_panic_handler_init(void)
1080 {
1081         atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1082         return 0;
1083 }
1084
1085 static const struct machine_ops xen_machine_ops __initconst = {
1086         .restart = xen_restart,
1087         .halt = xen_machine_halt,
1088         .power_off = xen_machine_power_off,
1089         .shutdown = xen_machine_halt,
1090         .crash_shutdown = xen_crash_shutdown,
1091         .emergency_restart = xen_emergency_restart,
1092 };
1093
1094 /*
1095  * Set up the GDT and segment registers for -fstack-protector.  Until
1096  * we do this, we have to be careful not to call any stack-protected
1097  * function, which is most of the kernel.
1098  */
1099 static void __init xen_setup_stackprotector(void)
1100 {
1101         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1102         pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1103
1104         setup_stack_canary_segment(0);
1105         switch_to_new_gdt(0);
1106
1107         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1108         pv_cpu_ops.load_gdt = xen_load_gdt;
1109 }
1110
1111 /* First C function to be called on Xen boot */
1112 asmlinkage void __init xen_start_kernel(void)
1113 {
1114         struct physdev_set_iopl set_iopl;
1115         int rc;
1116         pgd_t *pgd;
1117
1118         if (!xen_start_info)
1119                 return;
1120
1121         xen_domain_type = XEN_PV_DOMAIN;
1122
1123         xen_setup_machphys_mapping();
1124
1125         /* Install Xen paravirt ops */
1126         pv_info = xen_info;
1127         pv_init_ops = xen_init_ops;
1128         pv_cpu_ops = xen_cpu_ops;
1129         pv_apic_ops = xen_apic_ops;
1130
1131         x86_init.resources.memory_setup = xen_memory_setup;
1132         x86_init.oem.arch_setup = xen_arch_setup;
1133         x86_init.oem.banner = xen_banner;
1134
1135         xen_init_time_ops();
1136
1137         /*
1138          * Set up some pagetable state before starting to set any ptes.
1139          */
1140
1141         xen_init_mmu_ops();
1142
1143         /* Prevent unwanted bits from being set in PTEs. */
1144         __supported_pte_mask &= ~_PAGE_GLOBAL;
1145         if (!xen_initial_domain())
1146                 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1147
1148         __supported_pte_mask |= _PAGE_IOMAP;
1149
1150         /*
1151          * Prevent page tables from being allocated in highmem, even
1152          * if CONFIG_HIGHPTE is enabled.
1153          */
1154         __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1155
1156         /* Work out if we support NX */
1157         x86_configure_nx();
1158
1159         xen_setup_features();
1160
1161         /* Get mfn list */
1162         if (!xen_feature(XENFEAT_auto_translated_physmap))
1163                 xen_build_dynamic_phys_to_machine();
1164
1165         /*
1166          * Set up kernel GDT and segment registers, mainly so that
1167          * -fstack-protector code can be executed.
1168          */
1169         xen_setup_stackprotector();
1170
1171         xen_init_irq_ops();
1172         xen_init_cpuid_mask();
1173
1174 #ifdef CONFIG_X86_LOCAL_APIC
1175         /*
1176          * set up the basic apic ops.
1177          */
1178         set_xen_basic_apic_ops();
1179 #endif
1180
1181         if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1182                 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1183                 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1184         }
1185
1186         machine_ops = xen_machine_ops;
1187
1188         /*
1189          * The only reliable way to retain the initial address of the
1190          * percpu gdt_page is to remember it here, so we can go and
1191          * mark it RW later, when the initial percpu area is freed.
1192          */
1193         xen_initial_gdt = &per_cpu(gdt_page, 0);
1194
1195         xen_smp_init();
1196
1197 #ifdef CONFIG_ACPI_NUMA
1198         /*
1199          * The pages we from Xen are not related to machine pages, so
1200          * any NUMA information the kernel tries to get from ACPI will
1201          * be meaningless.  Prevent it from trying.
1202          */
1203         acpi_numa = -1;
1204 #endif
1205
1206         pgd = (pgd_t *)xen_start_info->pt_base;
1207
1208         if (!xen_initial_domain())
1209                 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1210
1211         __supported_pte_mask |= _PAGE_IOMAP;
1212         /* Don't do the full vcpu_info placement stuff until we have a
1213            possible map and a non-dummy shared_info. */
1214         per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1215
1216         local_irq_disable();
1217         early_boot_irqs_disabled = true;
1218
1219         memblock_init();
1220
1221         xen_raw_console_write("mapping kernel into physical memory\n");
1222         pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1223         xen_ident_map_ISA();
1224
1225         /* Allocate and initialize top and mid mfn levels for p2m structure */
1226         xen_build_mfn_list_list();
1227
1228         /* keep using Xen gdt for now; no urgent need to change it */
1229
1230 #ifdef CONFIG_X86_32
1231         pv_info.kernel_rpl = 1;
1232         if (xen_feature(XENFEAT_supervisor_mode_kernel))
1233                 pv_info.kernel_rpl = 0;
1234 #else
1235         pv_info.kernel_rpl = 0;
1236 #endif
1237         /* set the limit of our address space */
1238         xen_reserve_top();
1239
1240         /* We used to do this in xen_arch_setup, but that is too late on AMD
1241          * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1242          * which pokes 0xcf8 port.
1243          */
1244         set_iopl.iopl = 1;
1245         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1246         if (rc != 0)
1247                 xen_raw_printk("physdev_op failed %d\n", rc);
1248
1249 #ifdef CONFIG_X86_32
1250         /* set up basic CPUID stuff */
1251         cpu_detect(&new_cpu_data);
1252         new_cpu_data.hard_math = 1;
1253         new_cpu_data.wp_works_ok = 1;
1254         new_cpu_data.x86_capability[0] = cpuid_edx(1);
1255 #endif
1256
1257         /* Poke various useful things into boot_params */
1258         boot_params.hdr.type_of_loader = (9 << 4) | 0;
1259         boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1260                 ? __pa(xen_start_info->mod_start) : 0;
1261         boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1262         boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1263
1264         if (!xen_initial_domain()) {
1265                 add_preferred_console("xenboot", 0, NULL);
1266                 add_preferred_console("tty", 0, NULL);
1267                 add_preferred_console("hvc", 0, NULL);
1268                 if (pci_xen)
1269                         x86_init.pci.arch_init = pci_xen_init;
1270         } else {
1271                 const struct dom0_vga_console_info *info =
1272                         (void *)((char *)xen_start_info +
1273                                  xen_start_info->console.dom0.info_off);
1274
1275                 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1276                 xen_start_info->console.domU.mfn = 0;
1277                 xen_start_info->console.domU.evtchn = 0;
1278
1279                 /* Make sure ACS will be enabled */
1280                 pci_request_acs();
1281         }
1282 #ifdef CONFIG_PCI
1283         /* PCI BIOS service won't work from a PV guest. */
1284         pci_probe &= ~PCI_PROBE_BIOS;
1285 #endif
1286         xen_raw_console_write("about to get started...\n");
1287
1288         xen_setup_runstate_info(0);
1289
1290         /* Start the world */
1291 #ifdef CONFIG_X86_32
1292         i386_start_kernel();
1293 #else
1294         x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1295 #endif
1296 }
1297
1298 static int init_hvm_pv_info(int *major, int *minor)
1299 {
1300         uint32_t eax, ebx, ecx, edx, pages, msr, base;
1301         u64 pfn;
1302
1303         base = xen_cpuid_base();
1304         cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1305
1306         *major = eax >> 16;
1307         *minor = eax & 0xffff;
1308         printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1309
1310         cpuid(base + 2, &pages, &msr, &ecx, &edx);
1311
1312         pfn = __pa(hypercall_page);
1313         wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1314
1315         xen_setup_features();
1316
1317         pv_info.name = "Xen HVM";
1318
1319         xen_domain_type = XEN_HVM_DOMAIN;
1320
1321         return 0;
1322 }
1323
1324 void __ref xen_hvm_init_shared_info(void)
1325 {
1326         int cpu;
1327         struct xen_add_to_physmap xatp;
1328         static struct shared_info *shared_info_page = 0;
1329
1330         if (!shared_info_page)
1331                 shared_info_page = (struct shared_info *)
1332                         extend_brk(PAGE_SIZE, PAGE_SIZE);
1333         xatp.domid = DOMID_SELF;
1334         xatp.idx = 0;
1335         xatp.space = XENMAPSPACE_shared_info;
1336         xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1337         if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1338                 BUG();
1339
1340         HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1341
1342         /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1343          * page, we use it in the event channel upcall and in some pvclock
1344          * related functions. We don't need the vcpu_info placement
1345          * optimizations because we don't use any pv_mmu or pv_irq op on
1346          * HVM.
1347          * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1348          * online but xen_hvm_init_shared_info is run at resume time too and
1349          * in that case multiple vcpus might be online. */
1350         for_each_online_cpu(cpu) {
1351                 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1352         }
1353 }
1354
1355 #ifdef CONFIG_XEN_PVHVM
1356 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1357                                     unsigned long action, void *hcpu)
1358 {
1359         int cpu = (long)hcpu;
1360         switch (action) {
1361         case CPU_UP_PREPARE:
1362                 xen_vcpu_setup(cpu);
1363                 if (xen_have_vector_callback)
1364                         xen_init_lock_cpu(cpu);
1365                 break;
1366         default:
1367                 break;
1368         }
1369         return NOTIFY_OK;
1370 }
1371
1372 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1373         .notifier_call  = xen_hvm_cpu_notify,
1374 };
1375
1376 static void __init xen_hvm_guest_init(void)
1377 {
1378         int r;
1379         int major, minor;
1380
1381         r = init_hvm_pv_info(&major, &minor);
1382         if (r < 0)
1383                 return;
1384
1385         xen_hvm_init_shared_info();
1386
1387         if (xen_feature(XENFEAT_hvm_callback_vector))
1388                 xen_have_vector_callback = 1;
1389         xen_hvm_smp_init();
1390         register_cpu_notifier(&xen_hvm_cpu_notifier);
1391         xen_unplug_emulated_devices();
1392         x86_init.irqs.intr_init = xen_init_IRQ;
1393         xen_hvm_init_time_ops();
1394         xen_hvm_init_mmu_ops();
1395 }
1396
1397 static bool __init xen_hvm_platform(void)
1398 {
1399         if (xen_pv_domain())
1400                 return false;
1401
1402         if (!xen_cpuid_base())
1403                 return false;
1404
1405         return true;
1406 }
1407
1408 bool xen_hvm_need_lapic(void)
1409 {
1410         if (xen_pv_domain())
1411                 return false;
1412         if (!xen_hvm_domain())
1413                 return false;
1414         if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1415                 return false;
1416         return true;
1417 }
1418 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1419
1420 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1421         .name                   = "Xen HVM",
1422         .detect                 = xen_hvm_platform,
1423         .init_platform          = xen_hvm_guest_init,
1424 };
1425 EXPORT_SYMBOL(x86_hyper_xen_hvm);
1426 #endif