30822e8e64aceff30d0a8a6810b62fcf1cd62d57
[muen/linux.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/export.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 #include <linux/edd.h>
35 #include <linux/frame.h>
36
37 #include <linux/kexec.h>
38
39 #include <xen/xen.h>
40 #include <xen/events.h>
41 #include <xen/interface/xen.h>
42 #include <xen/interface/version.h>
43 #include <xen/interface/physdev.h>
44 #include <xen/interface/vcpu.h>
45 #include <xen/interface/memory.h>
46 #include <xen/interface/nmi.h>
47 #include <xen/interface/xen-mca.h>
48 #include <xen/interface/hvm/start_info.h>
49 #include <xen/features.h>
50 #include <xen/page.h>
51 #include <xen/hvm.h>
52 #include <xen/hvc-console.h>
53 #include <xen/acpi.h>
54
55 #include <asm/paravirt.h>
56 #include <asm/apic.h>
57 #include <asm/page.h>
58 #include <asm/xen/pci.h>
59 #include <asm/xen/hypercall.h>
60 #include <asm/xen/hypervisor.h>
61 #include <asm/xen/cpuid.h>
62 #include <asm/fixmap.h>
63 #include <asm/processor.h>
64 #include <asm/proto.h>
65 #include <asm/msr-index.h>
66 #include <asm/traps.h>
67 #include <asm/setup.h>
68 #include <asm/desc.h>
69 #include <asm/pgalloc.h>
70 #include <asm/pgtable.h>
71 #include <asm/tlbflush.h>
72 #include <asm/reboot.h>
73 #include <asm/stackprotector.h>
74 #include <asm/hypervisor.h>
75 #include <asm/mach_traps.h>
76 #include <asm/mwait.h>
77 #include <asm/pci_x86.h>
78 #include <asm/cpu.h>
79 #include <asm/e820/api.h> 
80
81 #ifdef CONFIG_ACPI
82 #include <linux/acpi.h>
83 #include <asm/acpi.h>
84 #include <acpi/pdc_intel.h>
85 #include <acpi/processor.h>
86 #include <xen/interface/platform.h>
87 #endif
88
89 #include "xen-ops.h"
90 #include "mmu.h"
91 #include "smp.h"
92 #include "multicalls.h"
93 #include "pmu.h"
94
95 EXPORT_SYMBOL_GPL(hypercall_page);
96
97 /*
98  * Pointer to the xen_vcpu_info structure or
99  * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
100  * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
101  * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
102  * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
103  * acknowledge pending events.
104  * Also more subtly it is used by the patched version of irq enable/disable
105  * e.g. xen_irq_enable_direct and xen_iret in PV mode.
106  *
107  * The desire to be able to do those mask/unmask operations as a single
108  * instruction by using the per-cpu offset held in %gs is the real reason
109  * vcpu info is in a per-cpu pointer and the original reason for this
110  * hypercall.
111  *
112  */
113 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
114
115 /*
116  * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
117  * hypercall. This can be used both in PV and PVHVM mode. The structure
118  * overrides the default per_cpu(xen_vcpu, cpu) value.
119  */
120 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
121
122 /* Linux <-> Xen vCPU id mapping */
123 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
124 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
125
126 enum xen_domain_type xen_domain_type = XEN_NATIVE;
127 EXPORT_SYMBOL_GPL(xen_domain_type);
128
129 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
130 EXPORT_SYMBOL(machine_to_phys_mapping);
131 unsigned long  machine_to_phys_nr;
132 EXPORT_SYMBOL(machine_to_phys_nr);
133
134 struct start_info *xen_start_info;
135 EXPORT_SYMBOL_GPL(xen_start_info);
136
137 struct shared_info xen_dummy_shared_info;
138
139 void *xen_initial_gdt;
140
141 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
142
143 static int xen_cpu_up_prepare(unsigned int cpu);
144 static int xen_cpu_up_online(unsigned int cpu);
145 static int xen_cpu_dead(unsigned int cpu);
146
147 /*
148  * Point at some empty memory to start with. We map the real shared_info
149  * page as soon as fixmap is up and running.
150  */
151 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
152
153 /*
154  * Flag to determine whether vcpu info placement is available on all
155  * VCPUs.  We assume it is to start with, and then set it to zero on
156  * the first failure.  This is because it can succeed on some VCPUs
157  * and not others, since it can involve hypervisor memory allocation,
158  * or because the guest failed to guarantee all the appropriate
159  * constraints on all VCPUs (ie buffer can't cross a page boundary).
160  *
161  * Note that any particular CPU may be using a placed vcpu structure,
162  * but we can only optimise if the all are.
163  *
164  * 0: not available, 1: available
165  */
166 static int have_vcpu_info_placement = 1;
167
168 struct tls_descs {
169         struct desc_struct desc[3];
170 };
171
172 /*
173  * Updating the 3 TLS descriptors in the GDT on every task switch is
174  * surprisingly expensive so we avoid updating them if they haven't
175  * changed.  Since Xen writes different descriptors than the one
176  * passed in the update_descriptor hypercall we keep shadow copies to
177  * compare against.
178  */
179 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
180
181 #ifdef CONFIG_XEN_PVH
182 /*
183  * PVH variables.
184  *
185  * xen_pvh and pvh_bootparams need to live in data segment since they
186  * are used after startup_{32|64}, which clear .bss, are invoked.
187  */
188 bool xen_pvh __attribute__((section(".data"))) = 0;
189 struct boot_params pvh_bootparams __attribute__((section(".data")));
190
191 struct hvm_start_info pvh_start_info;
192 unsigned int pvh_start_info_sz = sizeof(pvh_start_info);
193 #endif
194
195 static void clamp_max_cpus(void)
196 {
197 #ifdef CONFIG_SMP
198         if (setup_max_cpus > MAX_VIRT_CPUS)
199                 setup_max_cpus = MAX_VIRT_CPUS;
200 #endif
201 }
202
203 void xen_vcpu_setup(int cpu)
204 {
205         struct vcpu_register_vcpu_info info;
206         int err;
207         struct vcpu_info *vcpup;
208
209         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
210
211         /*
212          * This path is called twice on PVHVM - first during bootup via
213          * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
214          * hotplugged: cpu_up -> xen_hvm_cpu_notify.
215          * As we can only do the VCPUOP_register_vcpu_info once lets
216          * not over-write its result.
217          *
218          * For PV it is called during restore (xen_vcpu_restore) and bootup
219          * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
220          * use this function.
221          */
222         if (xen_hvm_domain()) {
223                 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
224                         return;
225         }
226         if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS)
227                 per_cpu(xen_vcpu, cpu) =
228                         &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
229
230         if (!have_vcpu_info_placement) {
231                 if (cpu >= MAX_VIRT_CPUS)
232                         clamp_max_cpus();
233                 return;
234         }
235
236         vcpup = &per_cpu(xen_vcpu_info, cpu);
237         info.mfn = arbitrary_virt_to_mfn(vcpup);
238         info.offset = offset_in_page(vcpup);
239
240         /* Check to see if the hypervisor will put the vcpu_info
241            structure where we want it, which allows direct access via
242            a percpu-variable.
243            N.B. This hypercall can _only_ be called once per CPU. Subsequent
244            calls will error out with -EINVAL. This is due to the fact that
245            hypervisor has no unregister variant and this hypercall does not
246            allow to over-write info.mfn and info.offset.
247          */
248         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
249                                  &info);
250
251         if (err) {
252                 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
253                 have_vcpu_info_placement = 0;
254                 clamp_max_cpus();
255         } else {
256                 /* This cpu is using the registered vcpu info, even if
257                    later ones fail to. */
258                 per_cpu(xen_vcpu, cpu) = vcpup;
259         }
260 }
261
262 /*
263  * On restore, set the vcpu placement up again.
264  * If it fails, then we're in a bad state, since
265  * we can't back out from using it...
266  */
267 void xen_vcpu_restore(void)
268 {
269         int cpu;
270
271         for_each_possible_cpu(cpu) {
272                 bool other_cpu = (cpu != smp_processor_id());
273                 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, xen_vcpu_nr(cpu),
274                                                 NULL);
275
276                 if (other_cpu && is_up &&
277                     HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
278                         BUG();
279
280                 xen_setup_runstate_info(cpu);
281
282                 if (have_vcpu_info_placement)
283                         xen_vcpu_setup(cpu);
284
285                 if (other_cpu && is_up &&
286                     HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
287                         BUG();
288         }
289 }
290
291 static void __init xen_banner(void)
292 {
293         unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
294         struct xen_extraversion extra;
295         HYPERVISOR_xen_version(XENVER_extraversion, &extra);
296
297         pr_info("Booting paravirtualized kernel %son %s\n",
298                 xen_feature(XENFEAT_auto_translated_physmap) ?
299                         "with PVH extensions " : "", pv_info.name);
300         printk(KERN_INFO "Xen version: %d.%d%s%s\n",
301                version >> 16, version & 0xffff, extra.extraversion,
302                xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
303 }
304 /* Check if running on Xen version (major, minor) or later */
305 bool
306 xen_running_on_version_or_later(unsigned int major, unsigned int minor)
307 {
308         unsigned int version;
309
310         if (!xen_domain())
311                 return false;
312
313         version = HYPERVISOR_xen_version(XENVER_version, NULL);
314         if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
315                 ((version >> 16) > major))
316                 return true;
317         return false;
318 }
319
320 #define CPUID_THERM_POWER_LEAF 6
321 #define APERFMPERF_PRESENT 0
322
323 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
324 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
325
326 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
327 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
328 static __read_mostly unsigned int cpuid_leaf5_edx_val;
329
330 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
331                       unsigned int *cx, unsigned int *dx)
332 {
333         unsigned maskebx = ~0;
334         unsigned maskecx = ~0;
335         unsigned maskedx = ~0;
336         unsigned setecx = 0;
337         /*
338          * Mask out inconvenient features, to try and disable as many
339          * unsupported kernel subsystems as possible.
340          */
341         switch (*ax) {
342         case 1:
343                 maskecx = cpuid_leaf1_ecx_mask;
344                 setecx = cpuid_leaf1_ecx_set_mask;
345                 maskedx = cpuid_leaf1_edx_mask;
346                 break;
347
348         case CPUID_MWAIT_LEAF:
349                 /* Synthesize the values.. */
350                 *ax = 0;
351                 *bx = 0;
352                 *cx = cpuid_leaf5_ecx_val;
353                 *dx = cpuid_leaf5_edx_val;
354                 return;
355
356         case CPUID_THERM_POWER_LEAF:
357                 /* Disabling APERFMPERF for kernel usage */
358                 maskecx = ~(1 << APERFMPERF_PRESENT);
359                 break;
360
361         case 0xb:
362                 /* Suppress extended topology stuff */
363                 maskebx = 0;
364                 break;
365         }
366
367         asm(XEN_EMULATE_PREFIX "cpuid"
368                 : "=a" (*ax),
369                   "=b" (*bx),
370                   "=c" (*cx),
371                   "=d" (*dx)
372                 : "0" (*ax), "2" (*cx));
373
374         *bx &= maskebx;
375         *cx &= maskecx;
376         *cx |= setecx;
377         *dx &= maskedx;
378 }
379 STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
380
381 static bool __init xen_check_mwait(void)
382 {
383 #ifdef CONFIG_ACPI
384         struct xen_platform_op op = {
385                 .cmd                    = XENPF_set_processor_pminfo,
386                 .u.set_pminfo.id        = -1,
387                 .u.set_pminfo.type      = XEN_PM_PDC,
388         };
389         uint32_t buf[3];
390         unsigned int ax, bx, cx, dx;
391         unsigned int mwait_mask;
392
393         /* We need to determine whether it is OK to expose the MWAIT
394          * capability to the kernel to harvest deeper than C3 states from ACPI
395          * _CST using the processor_harvest_xen.c module. For this to work, we
396          * need to gather the MWAIT_LEAF values (which the cstate.c code
397          * checks against). The hypervisor won't expose the MWAIT flag because
398          * it would break backwards compatibility; so we will find out directly
399          * from the hardware and hypercall.
400          */
401         if (!xen_initial_domain())
402                 return false;
403
404         /*
405          * When running under platform earlier than Xen4.2, do not expose
406          * mwait, to avoid the risk of loading native acpi pad driver
407          */
408         if (!xen_running_on_version_or_later(4, 2))
409                 return false;
410
411         ax = 1;
412         cx = 0;
413
414         native_cpuid(&ax, &bx, &cx, &dx);
415
416         mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
417                      (1 << (X86_FEATURE_MWAIT % 32));
418
419         if ((cx & mwait_mask) != mwait_mask)
420                 return false;
421
422         /* We need to emulate the MWAIT_LEAF and for that we need both
423          * ecx and edx. The hypercall provides only partial information.
424          */
425
426         ax = CPUID_MWAIT_LEAF;
427         bx = 0;
428         cx = 0;
429         dx = 0;
430
431         native_cpuid(&ax, &bx, &cx, &dx);
432
433         /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
434          * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
435          */
436         buf[0] = ACPI_PDC_REVISION_ID;
437         buf[1] = 1;
438         buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
439
440         set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
441
442         if ((HYPERVISOR_platform_op(&op) == 0) &&
443             (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
444                 cpuid_leaf5_ecx_val = cx;
445                 cpuid_leaf5_edx_val = dx;
446         }
447         return true;
448 #else
449         return false;
450 #endif
451 }
452 static void __init xen_init_cpuid_mask(void)
453 {
454         unsigned int ax, bx, cx, dx;
455         unsigned int xsave_mask;
456
457         cpuid_leaf1_edx_mask =
458                 ~((1 << X86_FEATURE_MTRR) |  /* disable MTRR */
459                   (1 << X86_FEATURE_ACC));   /* thermal monitoring */
460
461         if (!xen_initial_domain())
462                 cpuid_leaf1_edx_mask &=
463                         ~((1 << X86_FEATURE_ACPI));  /* disable ACPI */
464
465         cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32));
466
467         ax = 1;
468         cx = 0;
469         cpuid(1, &ax, &bx, &cx, &dx);
470
471         xsave_mask =
472                 (1 << (X86_FEATURE_XSAVE % 32)) |
473                 (1 << (X86_FEATURE_OSXSAVE % 32));
474
475         /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
476         if ((cx & xsave_mask) != xsave_mask)
477                 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
478         if (xen_check_mwait())
479                 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
480 }
481
482 static void xen_set_debugreg(int reg, unsigned long val)
483 {
484         HYPERVISOR_set_debugreg(reg, val);
485 }
486
487 static unsigned long xen_get_debugreg(int reg)
488 {
489         return HYPERVISOR_get_debugreg(reg);
490 }
491
492 static void xen_end_context_switch(struct task_struct *next)
493 {
494         xen_mc_flush();
495         paravirt_end_context_switch(next);
496 }
497
498 static unsigned long xen_store_tr(void)
499 {
500         return 0;
501 }
502
503 /*
504  * Set the page permissions for a particular virtual address.  If the
505  * address is a vmalloc mapping (or other non-linear mapping), then
506  * find the linear mapping of the page and also set its protections to
507  * match.
508  */
509 static void set_aliased_prot(void *v, pgprot_t prot)
510 {
511         int level;
512         pte_t *ptep;
513         pte_t pte;
514         unsigned long pfn;
515         struct page *page;
516         unsigned char dummy;
517
518         ptep = lookup_address((unsigned long)v, &level);
519         BUG_ON(ptep == NULL);
520
521         pfn = pte_pfn(*ptep);
522         page = pfn_to_page(pfn);
523
524         pte = pfn_pte(pfn, prot);
525
526         /*
527          * Careful: update_va_mapping() will fail if the virtual address
528          * we're poking isn't populated in the page tables.  We don't
529          * need to worry about the direct map (that's always in the page
530          * tables), but we need to be careful about vmap space.  In
531          * particular, the top level page table can lazily propagate
532          * entries between processes, so if we've switched mms since we
533          * vmapped the target in the first place, we might not have the
534          * top-level page table entry populated.
535          *
536          * We disable preemption because we want the same mm active when
537          * we probe the target and when we issue the hypercall.  We'll
538          * have the same nominal mm, but if we're a kernel thread, lazy
539          * mm dropping could change our pgd.
540          *
541          * Out of an abundance of caution, this uses __get_user() to fault
542          * in the target address just in case there's some obscure case
543          * in which the target address isn't readable.
544          */
545
546         preempt_disable();
547
548         probe_kernel_read(&dummy, v, 1);
549
550         if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
551                 BUG();
552
553         if (!PageHighMem(page)) {
554                 void *av = __va(PFN_PHYS(pfn));
555
556                 if (av != v)
557                         if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
558                                 BUG();
559         } else
560                 kmap_flush_unused();
561
562         preempt_enable();
563 }
564
565 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
566 {
567         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
568         int i;
569
570         /*
571          * We need to mark the all aliases of the LDT pages RO.  We
572          * don't need to call vm_flush_aliases(), though, since that's
573          * only responsible for flushing aliases out the TLBs, not the
574          * page tables, and Xen will flush the TLB for us if needed.
575          *
576          * To avoid confusing future readers: none of this is necessary
577          * to load the LDT.  The hypervisor only checks this when the
578          * LDT is faulted in due to subsequent descriptor access.
579          */
580
581         for(i = 0; i < entries; i += entries_per_page)
582                 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
583 }
584
585 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
586 {
587         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
588         int i;
589
590         for(i = 0; i < entries; i += entries_per_page)
591                 set_aliased_prot(ldt + i, PAGE_KERNEL);
592 }
593
594 static void xen_set_ldt(const void *addr, unsigned entries)
595 {
596         struct mmuext_op *op;
597         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
598
599         trace_xen_cpu_set_ldt(addr, entries);
600
601         op = mcs.args;
602         op->cmd = MMUEXT_SET_LDT;
603         op->arg1.linear_addr = (unsigned long)addr;
604         op->arg2.nr_ents = entries;
605
606         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
607
608         xen_mc_issue(PARAVIRT_LAZY_CPU);
609 }
610
611 static void xen_load_gdt(const struct desc_ptr *dtr)
612 {
613         unsigned long va = dtr->address;
614         unsigned int size = dtr->size + 1;
615         unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
616         unsigned long frames[pages];
617         int f;
618
619         /*
620          * A GDT can be up to 64k in size, which corresponds to 8192
621          * 8-byte entries, or 16 4k pages..
622          */
623
624         BUG_ON(size > 65536);
625         BUG_ON(va & ~PAGE_MASK);
626
627         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
628                 int level;
629                 pte_t *ptep;
630                 unsigned long pfn, mfn;
631                 void *virt;
632
633                 /*
634                  * The GDT is per-cpu and is in the percpu data area.
635                  * That can be virtually mapped, so we need to do a
636                  * page-walk to get the underlying MFN for the
637                  * hypercall.  The page can also be in the kernel's
638                  * linear range, so we need to RO that mapping too.
639                  */
640                 ptep = lookup_address(va, &level);
641                 BUG_ON(ptep == NULL);
642
643                 pfn = pte_pfn(*ptep);
644                 mfn = pfn_to_mfn(pfn);
645                 virt = __va(PFN_PHYS(pfn));
646
647                 frames[f] = mfn;
648
649                 make_lowmem_page_readonly((void *)va);
650                 make_lowmem_page_readonly(virt);
651         }
652
653         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
654                 BUG();
655 }
656
657 /*
658  * load_gdt for early boot, when the gdt is only mapped once
659  */
660 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
661 {
662         unsigned long va = dtr->address;
663         unsigned int size = dtr->size + 1;
664         unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
665         unsigned long frames[pages];
666         int f;
667
668         /*
669          * A GDT can be up to 64k in size, which corresponds to 8192
670          * 8-byte entries, or 16 4k pages..
671          */
672
673         BUG_ON(size > 65536);
674         BUG_ON(va & ~PAGE_MASK);
675
676         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
677                 pte_t pte;
678                 unsigned long pfn, mfn;
679
680                 pfn = virt_to_pfn(va);
681                 mfn = pfn_to_mfn(pfn);
682
683                 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
684
685                 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
686                         BUG();
687
688                 frames[f] = mfn;
689         }
690
691         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
692                 BUG();
693 }
694
695 static inline bool desc_equal(const struct desc_struct *d1,
696                               const struct desc_struct *d2)
697 {
698         return d1->a == d2->a && d1->b == d2->b;
699 }
700
701 static void load_TLS_descriptor(struct thread_struct *t,
702                                 unsigned int cpu, unsigned int i)
703 {
704         struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
705         struct desc_struct *gdt;
706         xmaddr_t maddr;
707         struct multicall_space mc;
708
709         if (desc_equal(shadow, &t->tls_array[i]))
710                 return;
711
712         *shadow = t->tls_array[i];
713
714         gdt = get_cpu_gdt_rw(cpu);
715         maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
716         mc = __xen_mc_entry(0);
717
718         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
719 }
720
721 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
722 {
723         /*
724          * XXX sleazy hack: If we're being called in a lazy-cpu zone
725          * and lazy gs handling is enabled, it means we're in a
726          * context switch, and %gs has just been saved.  This means we
727          * can zero it out to prevent faults on exit from the
728          * hypervisor if the next process has no %gs.  Either way, it
729          * has been saved, and the new value will get loaded properly.
730          * This will go away as soon as Xen has been modified to not
731          * save/restore %gs for normal hypercalls.
732          *
733          * On x86_64, this hack is not used for %gs, because gs points
734          * to KERNEL_GS_BASE (and uses it for PDA references), so we
735          * must not zero %gs on x86_64
736          *
737          * For x86_64, we need to zero %fs, otherwise we may get an
738          * exception between the new %fs descriptor being loaded and
739          * %fs being effectively cleared at __switch_to().
740          */
741         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
742 #ifdef CONFIG_X86_32
743                 lazy_load_gs(0);
744 #else
745                 loadsegment(fs, 0);
746 #endif
747         }
748
749         xen_mc_batch();
750
751         load_TLS_descriptor(t, cpu, 0);
752         load_TLS_descriptor(t, cpu, 1);
753         load_TLS_descriptor(t, cpu, 2);
754
755         xen_mc_issue(PARAVIRT_LAZY_CPU);
756 }
757
758 #ifdef CONFIG_X86_64
759 static void xen_load_gs_index(unsigned int idx)
760 {
761         if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
762                 BUG();
763 }
764 #endif
765
766 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
767                                 const void *ptr)
768 {
769         xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
770         u64 entry = *(u64 *)ptr;
771
772         trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
773
774         preempt_disable();
775
776         xen_mc_flush();
777         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
778                 BUG();
779
780         preempt_enable();
781 }
782
783 static int cvt_gate_to_trap(int vector, const gate_desc *val,
784                             struct trap_info *info)
785 {
786         unsigned long addr;
787
788         if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
789                 return 0;
790
791         info->vector = vector;
792
793         addr = gate_offset(*val);
794 #ifdef CONFIG_X86_64
795         /*
796          * Look for known traps using IST, and substitute them
797          * appropriately.  The debugger ones are the only ones we care
798          * about.  Xen will handle faults like double_fault,
799          * so we should never see them.  Warn if
800          * there's an unexpected IST-using fault handler.
801          */
802         if (addr == (unsigned long)debug)
803                 addr = (unsigned long)xen_debug;
804         else if (addr == (unsigned long)int3)
805                 addr = (unsigned long)xen_int3;
806         else if (addr == (unsigned long)stack_segment)
807                 addr = (unsigned long)xen_stack_segment;
808         else if (addr == (unsigned long)double_fault) {
809                 /* Don't need to handle these */
810                 return 0;
811 #ifdef CONFIG_X86_MCE
812         } else if (addr == (unsigned long)machine_check) {
813                 /*
814                  * when xen hypervisor inject vMCE to guest,
815                  * use native mce handler to handle it
816                  */
817                 ;
818 #endif
819         } else if (addr == (unsigned long)nmi)
820                 /*
821                  * Use the native version as well.
822                  */
823                 ;
824         else {
825                 /* Some other trap using IST? */
826                 if (WARN_ON(val->ist != 0))
827                         return 0;
828         }
829 #endif  /* CONFIG_X86_64 */
830         info->address = addr;
831
832         info->cs = gate_segment(*val);
833         info->flags = val->dpl;
834         /* interrupt gates clear IF */
835         if (val->type == GATE_INTERRUPT)
836                 info->flags |= 1 << 2;
837
838         return 1;
839 }
840
841 /* Locations of each CPU's IDT */
842 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
843
844 /* Set an IDT entry.  If the entry is part of the current IDT, then
845    also update Xen. */
846 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
847 {
848         unsigned long p = (unsigned long)&dt[entrynum];
849         unsigned long start, end;
850
851         trace_xen_cpu_write_idt_entry(dt, entrynum, g);
852
853         preempt_disable();
854
855         start = __this_cpu_read(idt_desc.address);
856         end = start + __this_cpu_read(idt_desc.size) + 1;
857
858         xen_mc_flush();
859
860         native_write_idt_entry(dt, entrynum, g);
861
862         if (p >= start && (p + 8) <= end) {
863                 struct trap_info info[2];
864
865                 info[1].address = 0;
866
867                 if (cvt_gate_to_trap(entrynum, g, &info[0]))
868                         if (HYPERVISOR_set_trap_table(info))
869                                 BUG();
870         }
871
872         preempt_enable();
873 }
874
875 static void xen_convert_trap_info(const struct desc_ptr *desc,
876                                   struct trap_info *traps)
877 {
878         unsigned in, out, count;
879
880         count = (desc->size+1) / sizeof(gate_desc);
881         BUG_ON(count > 256);
882
883         for (in = out = 0; in < count; in++) {
884                 gate_desc *entry = (gate_desc*)(desc->address) + in;
885
886                 if (cvt_gate_to_trap(in, entry, &traps[out]))
887                         out++;
888         }
889         traps[out].address = 0;
890 }
891
892 void xen_copy_trap_info(struct trap_info *traps)
893 {
894         const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
895
896         xen_convert_trap_info(desc, traps);
897 }
898
899 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
900    hold a spinlock to protect the static traps[] array (static because
901    it avoids allocation, and saves stack space). */
902 static void xen_load_idt(const struct desc_ptr *desc)
903 {
904         static DEFINE_SPINLOCK(lock);
905         static struct trap_info traps[257];
906
907         trace_xen_cpu_load_idt(desc);
908
909         spin_lock(&lock);
910
911         memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
912
913         xen_convert_trap_info(desc, traps);
914
915         xen_mc_flush();
916         if (HYPERVISOR_set_trap_table(traps))
917                 BUG();
918
919         spin_unlock(&lock);
920 }
921
922 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
923    they're handled differently. */
924 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
925                                 const void *desc, int type)
926 {
927         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
928
929         preempt_disable();
930
931         switch (type) {
932         case DESC_LDT:
933         case DESC_TSS:
934                 /* ignore */
935                 break;
936
937         default: {
938                 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
939
940                 xen_mc_flush();
941                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
942                         BUG();
943         }
944
945         }
946
947         preempt_enable();
948 }
949
950 /*
951  * Version of write_gdt_entry for use at early boot-time needed to
952  * update an entry as simply as possible.
953  */
954 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
955                                             const void *desc, int type)
956 {
957         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
958
959         switch (type) {
960         case DESC_LDT:
961         case DESC_TSS:
962                 /* ignore */
963                 break;
964
965         default: {
966                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
967
968                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
969                         dt[entry] = *(struct desc_struct *)desc;
970         }
971
972         }
973 }
974
975 static void xen_load_sp0(struct tss_struct *tss,
976                          struct thread_struct *thread)
977 {
978         struct multicall_space mcs;
979
980         mcs = xen_mc_entry(0);
981         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
982         xen_mc_issue(PARAVIRT_LAZY_CPU);
983         tss->x86_tss.sp0 = thread->sp0;
984 }
985
986 void xen_set_iopl_mask(unsigned mask)
987 {
988         struct physdev_set_iopl set_iopl;
989
990         /* Force the change at ring 0. */
991         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
992         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
993 }
994
995 static void xen_io_delay(void)
996 {
997 }
998
999 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
1000
1001 static unsigned long xen_read_cr0(void)
1002 {
1003         unsigned long cr0 = this_cpu_read(xen_cr0_value);
1004
1005         if (unlikely(cr0 == 0)) {
1006                 cr0 = native_read_cr0();
1007                 this_cpu_write(xen_cr0_value, cr0);
1008         }
1009
1010         return cr0;
1011 }
1012
1013 static void xen_write_cr0(unsigned long cr0)
1014 {
1015         struct multicall_space mcs;
1016
1017         this_cpu_write(xen_cr0_value, cr0);
1018
1019         /* Only pay attention to cr0.TS; everything else is
1020            ignored. */
1021         mcs = xen_mc_entry(0);
1022
1023         MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1024
1025         xen_mc_issue(PARAVIRT_LAZY_CPU);
1026 }
1027
1028 static void xen_write_cr4(unsigned long cr4)
1029 {
1030         cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1031
1032         native_write_cr4(cr4);
1033 }
1034 #ifdef CONFIG_X86_64
1035 static inline unsigned long xen_read_cr8(void)
1036 {
1037         return 0;
1038 }
1039 static inline void xen_write_cr8(unsigned long val)
1040 {
1041         BUG_ON(val);
1042 }
1043 #endif
1044
1045 static u64 xen_read_msr_safe(unsigned int msr, int *err)
1046 {
1047         u64 val;
1048
1049         if (pmu_msr_read(msr, &val, err))
1050                 return val;
1051
1052         val = native_read_msr_safe(msr, err);
1053         switch (msr) {
1054         case MSR_IA32_APICBASE:
1055 #ifdef CONFIG_X86_X2APIC
1056                 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
1057 #endif
1058                         val &= ~X2APIC_ENABLE;
1059                 break;
1060         }
1061         return val;
1062 }
1063
1064 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1065 {
1066         int ret;
1067
1068         ret = 0;
1069
1070         switch (msr) {
1071 #ifdef CONFIG_X86_64
1072                 unsigned which;
1073                 u64 base;
1074
1075         case MSR_FS_BASE:               which = SEGBASE_FS; goto set;
1076         case MSR_KERNEL_GS_BASE:        which = SEGBASE_GS_USER; goto set;
1077         case MSR_GS_BASE:               which = SEGBASE_GS_KERNEL; goto set;
1078
1079         set:
1080                 base = ((u64)high << 32) | low;
1081                 if (HYPERVISOR_set_segment_base(which, base) != 0)
1082                         ret = -EIO;
1083                 break;
1084 #endif
1085
1086         case MSR_STAR:
1087         case MSR_CSTAR:
1088         case MSR_LSTAR:
1089         case MSR_SYSCALL_MASK:
1090         case MSR_IA32_SYSENTER_CS:
1091         case MSR_IA32_SYSENTER_ESP:
1092         case MSR_IA32_SYSENTER_EIP:
1093                 /* Fast syscall setup is all done in hypercalls, so
1094                    these are all ignored.  Stub them out here to stop
1095                    Xen console noise. */
1096                 break;
1097
1098         default:
1099                 if (!pmu_msr_write(msr, low, high, &ret))
1100                         ret = native_write_msr_safe(msr, low, high);
1101         }
1102
1103         return ret;
1104 }
1105
1106 static u64 xen_read_msr(unsigned int msr)
1107 {
1108         /*
1109          * This will silently swallow a #GP from RDMSR.  It may be worth
1110          * changing that.
1111          */
1112         int err;
1113
1114         return xen_read_msr_safe(msr, &err);
1115 }
1116
1117 static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1118 {
1119         /*
1120          * This will silently swallow a #GP from WRMSR.  It may be worth
1121          * changing that.
1122          */
1123         xen_write_msr_safe(msr, low, high);
1124 }
1125
1126 void xen_setup_shared_info(void)
1127 {
1128         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1129                 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1130                            xen_start_info->shared_info);
1131
1132                 HYPERVISOR_shared_info =
1133                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1134         } else
1135                 HYPERVISOR_shared_info =
1136                         (struct shared_info *)__va(xen_start_info->shared_info);
1137
1138 #ifndef CONFIG_SMP
1139         /* In UP this is as good a place as any to set up shared info */
1140         xen_setup_vcpu_info_placement();
1141 #endif
1142
1143         xen_setup_mfn_list_list();
1144 }
1145
1146 /* This is called once we have the cpu_possible_mask */
1147 void xen_setup_vcpu_info_placement(void)
1148 {
1149         int cpu;
1150
1151         for_each_possible_cpu(cpu) {
1152                 /* Set up direct vCPU id mapping for PV guests. */
1153                 per_cpu(xen_vcpu_id, cpu) = cpu;
1154                 xen_vcpu_setup(cpu);
1155         }
1156
1157         /*
1158          * xen_vcpu_setup managed to place the vcpu_info within the
1159          * percpu area for all cpus, so make use of it.
1160          */
1161         if (have_vcpu_info_placement) {
1162                 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1163                 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1164                 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1165                 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1166                 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1167         }
1168 }
1169
1170 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1171                           unsigned long addr, unsigned len)
1172 {
1173         char *start, *end, *reloc;
1174         unsigned ret;
1175
1176         start = end = reloc = NULL;
1177
1178 #define SITE(op, x)                                                     \
1179         case PARAVIRT_PATCH(op.x):                                      \
1180         if (have_vcpu_info_placement) {                                 \
1181                 start = (char *)xen_##x##_direct;                       \
1182                 end = xen_##x##_direct_end;                             \
1183                 reloc = xen_##x##_direct_reloc;                         \
1184         }                                                               \
1185         goto patch_site
1186
1187         switch (type) {
1188                 SITE(pv_irq_ops, irq_enable);
1189                 SITE(pv_irq_ops, irq_disable);
1190                 SITE(pv_irq_ops, save_fl);
1191                 SITE(pv_irq_ops, restore_fl);
1192 #undef SITE
1193
1194         patch_site:
1195                 if (start == NULL || (end-start) > len)
1196                         goto default_patch;
1197
1198                 ret = paravirt_patch_insns(insnbuf, len, start, end);
1199
1200                 /* Note: because reloc is assigned from something that
1201                    appears to be an array, gcc assumes it's non-null,
1202                    but doesn't know its relationship with start and
1203                    end. */
1204                 if (reloc > start && reloc < end) {
1205                         int reloc_off = reloc - start;
1206                         long *relocp = (long *)(insnbuf + reloc_off);
1207                         long delta = start - (char *)addr;
1208
1209                         *relocp += delta;
1210                 }
1211                 break;
1212
1213         default_patch:
1214         default:
1215                 ret = paravirt_patch_default(type, clobbers, insnbuf,
1216                                              addr, len);
1217                 break;
1218         }
1219
1220         return ret;
1221 }
1222
1223 static const struct pv_info xen_info __initconst = {
1224         .shared_kernel_pmd = 0,
1225
1226 #ifdef CONFIG_X86_64
1227         .extra_user_64bit_cs = FLAT_USER_CS64,
1228 #endif
1229         .name = "Xen",
1230 };
1231
1232 static const struct pv_init_ops xen_init_ops __initconst = {
1233         .patch = xen_patch,
1234 };
1235
1236 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1237         .cpuid = xen_cpuid,
1238
1239         .set_debugreg = xen_set_debugreg,
1240         .get_debugreg = xen_get_debugreg,
1241
1242         .read_cr0 = xen_read_cr0,
1243         .write_cr0 = xen_write_cr0,
1244
1245         .read_cr4 = native_read_cr4,
1246         .write_cr4 = xen_write_cr4,
1247
1248 #ifdef CONFIG_X86_64
1249         .read_cr8 = xen_read_cr8,
1250         .write_cr8 = xen_write_cr8,
1251 #endif
1252
1253         .wbinvd = native_wbinvd,
1254
1255         .read_msr = xen_read_msr,
1256         .write_msr = xen_write_msr,
1257
1258         .read_msr_safe = xen_read_msr_safe,
1259         .write_msr_safe = xen_write_msr_safe,
1260
1261         .read_pmc = xen_read_pmc,
1262
1263         .iret = xen_iret,
1264 #ifdef CONFIG_X86_64
1265         .usergs_sysret64 = xen_sysret64,
1266 #endif
1267
1268         .load_tr_desc = paravirt_nop,
1269         .set_ldt = xen_set_ldt,
1270         .load_gdt = xen_load_gdt,
1271         .load_idt = xen_load_idt,
1272         .load_tls = xen_load_tls,
1273 #ifdef CONFIG_X86_64
1274         .load_gs_index = xen_load_gs_index,
1275 #endif
1276
1277         .alloc_ldt = xen_alloc_ldt,
1278         .free_ldt = xen_free_ldt,
1279
1280         .store_idt = native_store_idt,
1281         .store_tr = xen_store_tr,
1282
1283         .write_ldt_entry = xen_write_ldt_entry,
1284         .write_gdt_entry = xen_write_gdt_entry,
1285         .write_idt_entry = xen_write_idt_entry,
1286         .load_sp0 = xen_load_sp0,
1287
1288         .set_iopl_mask = xen_set_iopl_mask,
1289         .io_delay = xen_io_delay,
1290
1291         /* Xen takes care of %gs when switching to usermode for us */
1292         .swapgs = paravirt_nop,
1293
1294         .start_context_switch = paravirt_start_context_switch,
1295         .end_context_switch = xen_end_context_switch,
1296 };
1297
1298 static void xen_reboot(int reason)
1299 {
1300         struct sched_shutdown r = { .reason = reason };
1301         int cpu;
1302
1303         for_each_online_cpu(cpu)
1304                 xen_pmu_finish(cpu);
1305
1306         if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1307                 BUG();
1308 }
1309
1310 static void xen_restart(char *msg)
1311 {
1312         xen_reboot(SHUTDOWN_reboot);
1313 }
1314
1315 static void xen_emergency_restart(void)
1316 {
1317         xen_reboot(SHUTDOWN_reboot);
1318 }
1319
1320 static void xen_machine_halt(void)
1321 {
1322         xen_reboot(SHUTDOWN_poweroff);
1323 }
1324
1325 static void xen_machine_power_off(void)
1326 {
1327         if (pm_power_off)
1328                 pm_power_off();
1329         xen_reboot(SHUTDOWN_poweroff);
1330 }
1331
1332 static void xen_crash_shutdown(struct pt_regs *regs)
1333 {
1334         xen_reboot(SHUTDOWN_crash);
1335 }
1336
1337 static int
1338 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1339 {
1340         if (!kexec_crash_loaded())
1341                 xen_reboot(SHUTDOWN_crash);
1342         return NOTIFY_DONE;
1343 }
1344
1345 static struct notifier_block xen_panic_block = {
1346         .notifier_call= xen_panic_event,
1347         .priority = INT_MIN
1348 };
1349
1350 int xen_panic_handler_init(void)
1351 {
1352         atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1353         return 0;
1354 }
1355
1356 static const struct machine_ops xen_machine_ops __initconst = {
1357         .restart = xen_restart,
1358         .halt = xen_machine_halt,
1359         .power_off = xen_machine_power_off,
1360         .shutdown = xen_machine_halt,
1361         .crash_shutdown = xen_crash_shutdown,
1362         .emergency_restart = xen_emergency_restart,
1363 };
1364
1365 static unsigned char xen_get_nmi_reason(void)
1366 {
1367         unsigned char reason = 0;
1368
1369         /* Construct a value which looks like it came from port 0x61. */
1370         if (test_bit(_XEN_NMIREASON_io_error,
1371                      &HYPERVISOR_shared_info->arch.nmi_reason))
1372                 reason |= NMI_REASON_IOCHK;
1373         if (test_bit(_XEN_NMIREASON_pci_serr,
1374                      &HYPERVISOR_shared_info->arch.nmi_reason))
1375                 reason |= NMI_REASON_SERR;
1376
1377         return reason;
1378 }
1379
1380 static void __init xen_boot_params_init_edd(void)
1381 {
1382 #if IS_ENABLED(CONFIG_EDD)
1383         struct xen_platform_op op;
1384         struct edd_info *edd_info;
1385         u32 *mbr_signature;
1386         unsigned nr;
1387         int ret;
1388
1389         edd_info = boot_params.eddbuf;
1390         mbr_signature = boot_params.edd_mbr_sig_buffer;
1391
1392         op.cmd = XENPF_firmware_info;
1393
1394         op.u.firmware_info.type = XEN_FW_DISK_INFO;
1395         for (nr = 0; nr < EDDMAXNR; nr++) {
1396                 struct edd_info *info = edd_info + nr;
1397
1398                 op.u.firmware_info.index = nr;
1399                 info->params.length = sizeof(info->params);
1400                 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1401                                      &info->params);
1402                 ret = HYPERVISOR_platform_op(&op);
1403                 if (ret)
1404                         break;
1405
1406 #define C(x) info->x = op.u.firmware_info.u.disk_info.x
1407                 C(device);
1408                 C(version);
1409                 C(interface_support);
1410                 C(legacy_max_cylinder);
1411                 C(legacy_max_head);
1412                 C(legacy_sectors_per_track);
1413 #undef C
1414         }
1415         boot_params.eddbuf_entries = nr;
1416
1417         op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1418         for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1419                 op.u.firmware_info.index = nr;
1420                 ret = HYPERVISOR_platform_op(&op);
1421                 if (ret)
1422                         break;
1423                 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1424         }
1425         boot_params.edd_mbr_sig_buf_entries = nr;
1426 #endif
1427 }
1428
1429 /*
1430  * Set up the GDT and segment registers for -fstack-protector.  Until
1431  * we do this, we have to be careful not to call any stack-protected
1432  * function, which is most of the kernel.
1433  */
1434 static void xen_setup_gdt(int cpu)
1435 {
1436         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1437         pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1438
1439         setup_stack_canary_segment(0);
1440         switch_to_new_gdt(0);
1441
1442         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1443         pv_cpu_ops.load_gdt = xen_load_gdt;
1444 }
1445
1446 static void __init xen_dom0_set_legacy_features(void)
1447 {
1448         x86_platform.legacy.rtc = 1;
1449 }
1450
1451 static int xen_cpuhp_setup(void)
1452 {
1453         int rc;
1454
1455         rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
1456                                        "x86/xen/hvm_guest:prepare",
1457                                        xen_cpu_up_prepare, xen_cpu_dead);
1458         if (rc >= 0) {
1459                 rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
1460                                                "x86/xen/hvm_guest:online",
1461                                                xen_cpu_up_online, NULL);
1462                 if (rc < 0)
1463                         cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
1464         }
1465
1466         return rc >= 0 ? 0 : rc;
1467 }
1468
1469 /* First C function to be called on Xen boot */
1470 asmlinkage __visible void __init xen_start_kernel(void)
1471 {
1472         struct physdev_set_iopl set_iopl;
1473         unsigned long initrd_start = 0;
1474         int rc;
1475
1476         if (!xen_start_info)
1477                 return;
1478
1479         xen_domain_type = XEN_PV_DOMAIN;
1480
1481         xen_setup_features();
1482
1483         xen_setup_machphys_mapping();
1484
1485         /* Install Xen paravirt ops */
1486         pv_info = xen_info;
1487         pv_init_ops = xen_init_ops;
1488         pv_cpu_ops = xen_cpu_ops;
1489
1490         x86_platform.get_nmi_reason = xen_get_nmi_reason;
1491
1492         x86_init.resources.memory_setup = xen_memory_setup;
1493         x86_init.oem.arch_setup = xen_arch_setup;
1494         x86_init.oem.banner = xen_banner;
1495
1496         xen_init_time_ops();
1497
1498         /*
1499          * Set up some pagetable state before starting to set any ptes.
1500          */
1501
1502         xen_init_mmu_ops();
1503
1504         /* Prevent unwanted bits from being set in PTEs. */
1505         __supported_pte_mask &= ~_PAGE_GLOBAL;
1506
1507         /*
1508          * Prevent page tables from being allocated in highmem, even
1509          * if CONFIG_HIGHPTE is enabled.
1510          */
1511         __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1512
1513         /* Work out if we support NX */
1514         x86_configure_nx();
1515
1516         /* Get mfn list */
1517         xen_build_dynamic_phys_to_machine();
1518
1519         /*
1520          * Set up kernel GDT and segment registers, mainly so that
1521          * -fstack-protector code can be executed.
1522          */
1523         xen_setup_gdt(0);
1524
1525         xen_init_irq_ops();
1526         xen_init_cpuid_mask();
1527
1528 #ifdef CONFIG_X86_LOCAL_APIC
1529         /*
1530          * set up the basic apic ops.
1531          */
1532         xen_init_apic();
1533 #endif
1534
1535         if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1536                 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1537                 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1538         }
1539
1540         machine_ops = xen_machine_ops;
1541
1542         /*
1543          * The only reliable way to retain the initial address of the
1544          * percpu gdt_page is to remember it here, so we can go and
1545          * mark it RW later, when the initial percpu area is freed.
1546          */
1547         xen_initial_gdt = &per_cpu(gdt_page, 0);
1548
1549         xen_smp_init();
1550
1551 #ifdef CONFIG_ACPI_NUMA
1552         /*
1553          * The pages we from Xen are not related to machine pages, so
1554          * any NUMA information the kernel tries to get from ACPI will
1555          * be meaningless.  Prevent it from trying.
1556          */
1557         acpi_numa = -1;
1558 #endif
1559         /* Don't do the full vcpu_info placement stuff until we have a
1560            possible map and a non-dummy shared_info. */
1561         per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1562
1563         WARN_ON(xen_cpuhp_setup());
1564
1565         local_irq_disable();
1566         early_boot_irqs_disabled = true;
1567
1568         xen_raw_console_write("mapping kernel into physical memory\n");
1569         xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1570                                    xen_start_info->nr_pages);
1571         xen_reserve_special_pages();
1572
1573         /* keep using Xen gdt for now; no urgent need to change it */
1574
1575 #ifdef CONFIG_X86_32
1576         pv_info.kernel_rpl = 1;
1577         if (xen_feature(XENFEAT_supervisor_mode_kernel))
1578                 pv_info.kernel_rpl = 0;
1579 #else
1580         pv_info.kernel_rpl = 0;
1581 #endif
1582         /* set the limit of our address space */
1583         xen_reserve_top();
1584
1585         /*
1586          * We used to do this in xen_arch_setup, but that is too late
1587          * on AMD were early_cpu_init (run before ->arch_setup()) calls
1588          * early_amd_init which pokes 0xcf8 port.
1589          */
1590         set_iopl.iopl = 1;
1591         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1592         if (rc != 0)
1593                 xen_raw_printk("physdev_op failed %d\n", rc);
1594
1595 #ifdef CONFIG_X86_32
1596         /* set up basic CPUID stuff */
1597         cpu_detect(&new_cpu_data);
1598         set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1599         new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1600 #endif
1601
1602         if (xen_start_info->mod_start) {
1603             if (xen_start_info->flags & SIF_MOD_START_PFN)
1604                 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1605             else
1606                 initrd_start = __pa(xen_start_info->mod_start);
1607         }
1608
1609         /* Poke various useful things into boot_params */
1610         boot_params.hdr.type_of_loader = (9 << 4) | 0;
1611         boot_params.hdr.ramdisk_image = initrd_start;
1612         boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1613         boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1614         boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1615
1616         if (!xen_initial_domain()) {
1617                 add_preferred_console("xenboot", 0, NULL);
1618                 add_preferred_console("tty", 0, NULL);
1619                 add_preferred_console("hvc", 0, NULL);
1620                 if (pci_xen)
1621                         x86_init.pci.arch_init = pci_xen_init;
1622         } else {
1623                 const struct dom0_vga_console_info *info =
1624                         (void *)((char *)xen_start_info +
1625                                  xen_start_info->console.dom0.info_off);
1626                 struct xen_platform_op op = {
1627                         .cmd = XENPF_firmware_info,
1628                         .interface_version = XENPF_INTERFACE_VERSION,
1629                         .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1630                 };
1631
1632                 x86_platform.set_legacy_features =
1633                                 xen_dom0_set_legacy_features;
1634                 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1635                 xen_start_info->console.domU.mfn = 0;
1636                 xen_start_info->console.domU.evtchn = 0;
1637
1638                 if (HYPERVISOR_platform_op(&op) == 0)
1639                         boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1640
1641                 /* Make sure ACS will be enabled */
1642                 pci_request_acs();
1643
1644                 xen_acpi_sleep_register();
1645
1646                 /* Avoid searching for BIOS MP tables */
1647                 x86_init.mpparse.find_smp_config = x86_init_noop;
1648                 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1649
1650                 xen_boot_params_init_edd();
1651         }
1652 #ifdef CONFIG_PCI
1653         /* PCI BIOS service won't work from a PV guest. */
1654         pci_probe &= ~PCI_PROBE_BIOS;
1655 #endif
1656         xen_raw_console_write("about to get started...\n");
1657
1658         /* Let's presume PV guests always boot on vCPU with id 0. */
1659         per_cpu(xen_vcpu_id, 0) = 0;
1660
1661         xen_setup_runstate_info(0);
1662
1663         xen_efi_init();
1664
1665         /* Start the world */
1666 #ifdef CONFIG_X86_32
1667         i386_start_kernel();
1668 #else
1669         cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1670         x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1671 #endif
1672 }
1673
1674 #ifdef CONFIG_XEN_PVH
1675
1676 static void xen_pvh_arch_setup(void)
1677 {
1678 #ifdef CONFIG_ACPI
1679         /* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
1680         if (nr_ioapics == 0)
1681                 acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
1682 #endif
1683 }
1684
1685 static void __init init_pvh_bootparams(void)
1686 {
1687         struct xen_memory_map memmap;
1688         unsigned int i;
1689         int rc;
1690
1691         memset(&pvh_bootparams, 0, sizeof(pvh_bootparams));
1692
1693         memmap.nr_entries = ARRAY_SIZE(pvh_bootparams.e820_table);
1694         set_xen_guest_handle(memmap.buffer, pvh_bootparams.e820_table);
1695         rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
1696         if (rc) {
1697                 xen_raw_printk("XENMEM_memory_map failed (%d)\n", rc);
1698                 BUG();
1699         }
1700
1701         if (memmap.nr_entries < E820_MAX_ENTRIES_ZEROPAGE - 1) {
1702                 pvh_bootparams.e820_table[memmap.nr_entries].addr =
1703                         ISA_START_ADDRESS;
1704                 pvh_bootparams.e820_table[memmap.nr_entries].size =
1705                         ISA_END_ADDRESS - ISA_START_ADDRESS;
1706                 pvh_bootparams.e820_table[memmap.nr_entries].type =
1707                         E820_TYPE_RESERVED;
1708                 memmap.nr_entries++;
1709         } else
1710                 xen_raw_printk("Warning: Can fit ISA range into e820\n");
1711
1712         pvh_bootparams.e820_entries = memmap.nr_entries;
1713         for (i = 0; i < pvh_bootparams.e820_entries; i++)
1714                 e820__range_add(pvh_bootparams.e820_table[i].addr,
1715                                 pvh_bootparams.e820_table[i].size,
1716                                 pvh_bootparams.e820_table[i].type);
1717
1718         e820__update_table(e820_table);
1719
1720         pvh_bootparams.hdr.cmd_line_ptr =
1721                 pvh_start_info.cmdline_paddr;
1722
1723         /* The first module is always ramdisk. */
1724         if (pvh_start_info.nr_modules) {
1725                 struct hvm_modlist_entry *modaddr =
1726                         __va(pvh_start_info.modlist_paddr);
1727                 pvh_bootparams.hdr.ramdisk_image = modaddr->paddr;
1728                 pvh_bootparams.hdr.ramdisk_size = modaddr->size;
1729         }
1730
1731         /*
1732          * See Documentation/x86/boot.txt.
1733          *
1734          * Version 2.12 supports Xen entry point but we will use default x86/PC
1735          * environment (i.e. hardware_subarch 0).
1736          */
1737         pvh_bootparams.hdr.version = 0x212;
1738         pvh_bootparams.hdr.type_of_loader = (9 << 4) | 0; /* Xen loader */
1739 }
1740
1741 /*
1742  * This routine (and those that it might call) should not use
1743  * anything that lives in .bss since that segment will be cleared later.
1744  */
1745 void __init xen_prepare_pvh(void)
1746 {
1747         u32 msr;
1748         u64 pfn;
1749
1750         if (pvh_start_info.magic != XEN_HVM_START_MAGIC_VALUE) {
1751                 xen_raw_printk("Error: Unexpected magic value (0x%08x)\n",
1752                                 pvh_start_info.magic);
1753                 BUG();
1754         }
1755
1756         xen_pvh = 1;
1757
1758         msr = cpuid_ebx(xen_cpuid_base() + 2);
1759         pfn = __pa(hypercall_page);
1760         wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1761
1762         init_pvh_bootparams();
1763
1764         x86_init.oem.arch_setup = xen_pvh_arch_setup;
1765 }
1766 #endif
1767
1768 void __ref xen_hvm_init_shared_info(void)
1769 {
1770         int cpu;
1771         struct xen_add_to_physmap xatp;
1772         static struct shared_info *shared_info_page = 0;
1773
1774         if (!shared_info_page)
1775                 shared_info_page = (struct shared_info *)
1776                         extend_brk(PAGE_SIZE, PAGE_SIZE);
1777         xatp.domid = DOMID_SELF;
1778         xatp.idx = 0;
1779         xatp.space = XENMAPSPACE_shared_info;
1780         xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1781         if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1782                 BUG();
1783
1784         HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1785
1786         /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1787          * page, we use it in the event channel upcall and in some pvclock
1788          * related functions. We don't need the vcpu_info placement
1789          * optimizations because we don't use any pv_mmu or pv_irq op on
1790          * HVM.
1791          * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1792          * online but xen_hvm_init_shared_info is run at resume time too and
1793          * in that case multiple vcpus might be online. */
1794         for_each_online_cpu(cpu) {
1795                 /* Leave it to be NULL. */
1796                 if (xen_vcpu_nr(cpu) >= MAX_VIRT_CPUS)
1797                         continue;
1798                 per_cpu(xen_vcpu, cpu) =
1799                         &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
1800         }
1801 }
1802
1803 #ifdef CONFIG_XEN_PVHVM
1804 static void __init init_hvm_pv_info(void)
1805 {
1806         int major, minor;
1807         uint32_t eax, ebx, ecx, edx, base;
1808
1809         base = xen_cpuid_base();
1810         eax = cpuid_eax(base + 1);
1811
1812         major = eax >> 16;
1813         minor = eax & 0xffff;
1814         printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1815
1816         xen_domain_type = XEN_HVM_DOMAIN;
1817
1818         /* PVH set up hypercall page in xen_prepare_pvh(). */
1819         if (xen_pvh_domain())
1820                 pv_info.name = "Xen PVH";
1821         else {
1822                 u64 pfn;
1823                 uint32_t msr;
1824
1825                 pv_info.name = "Xen HVM";
1826                 msr = cpuid_ebx(base + 2);
1827                 pfn = __pa(hypercall_page);
1828                 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1829         }
1830
1831         xen_setup_features();
1832
1833         cpuid(base + 4, &eax, &ebx, &ecx, &edx);
1834         if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
1835                 this_cpu_write(xen_vcpu_id, ebx);
1836         else
1837                 this_cpu_write(xen_vcpu_id, smp_processor_id());
1838 }
1839 #endif
1840
1841 static int xen_cpu_up_prepare(unsigned int cpu)
1842 {
1843         int rc;
1844
1845         if (xen_hvm_domain()) {
1846                 /*
1847                  * This can happen if CPU was offlined earlier and
1848                  * offlining timed out in common_cpu_die().
1849                  */
1850                 if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
1851                         xen_smp_intr_free(cpu);
1852                         xen_uninit_lock_cpu(cpu);
1853                 }
1854
1855                 if (cpu_acpi_id(cpu) != U32_MAX)
1856                         per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);
1857                 else
1858                         per_cpu(xen_vcpu_id, cpu) = cpu;
1859                 xen_vcpu_setup(cpu);
1860         }
1861
1862         if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
1863                 xen_setup_timer(cpu);
1864
1865         rc = xen_smp_intr_init(cpu);
1866         if (rc) {
1867                 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1868                      cpu, rc);
1869                 return rc;
1870         }
1871         return 0;
1872 }
1873
1874 static int xen_cpu_dead(unsigned int cpu)
1875 {
1876         xen_smp_intr_free(cpu);
1877
1878         if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
1879                 xen_teardown_timer(cpu);
1880
1881         return 0;
1882 }
1883
1884 static int xen_cpu_up_online(unsigned int cpu)
1885 {
1886         xen_init_lock_cpu(cpu);
1887         return 0;
1888 }
1889
1890 #ifdef CONFIG_XEN_PVHVM
1891 #ifdef CONFIG_KEXEC_CORE
1892 static void xen_hvm_shutdown(void)
1893 {
1894         native_machine_shutdown();
1895         if (kexec_in_progress)
1896                 xen_reboot(SHUTDOWN_soft_reset);
1897 }
1898
1899 static void xen_hvm_crash_shutdown(struct pt_regs *regs)
1900 {
1901         native_machine_crash_shutdown(regs);
1902         xen_reboot(SHUTDOWN_soft_reset);
1903 }
1904 #endif
1905
1906 static void __init xen_hvm_guest_init(void)
1907 {
1908         if (xen_pv_domain())
1909                 return;
1910
1911         init_hvm_pv_info();
1912
1913         xen_hvm_init_shared_info();
1914
1915         xen_panic_handler_init();
1916
1917         BUG_ON(!xen_feature(XENFEAT_hvm_callback_vector));
1918
1919         xen_hvm_smp_init();
1920         WARN_ON(xen_cpuhp_setup());
1921         xen_unplug_emulated_devices();
1922         x86_init.irqs.intr_init = xen_init_IRQ;
1923         xen_hvm_init_time_ops();
1924         xen_hvm_init_mmu_ops();
1925
1926         if (xen_pvh_domain())
1927                 machine_ops.emergency_restart = xen_emergency_restart;
1928 #ifdef CONFIG_KEXEC_CORE
1929         machine_ops.shutdown = xen_hvm_shutdown;
1930         machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
1931 #endif
1932 }
1933 #endif
1934
1935 static bool xen_nopv = false;
1936 static __init int xen_parse_nopv(char *arg)
1937 {
1938        xen_nopv = true;
1939        return 0;
1940 }
1941 early_param("xen_nopv", xen_parse_nopv);
1942
1943 static uint32_t __init xen_platform(void)
1944 {
1945         if (xen_nopv)
1946                 return 0;
1947
1948         return xen_cpuid_base();
1949 }
1950
1951 bool xen_hvm_need_lapic(void)
1952 {
1953         if (xen_nopv)
1954                 return false;
1955         if (xen_pv_domain())
1956                 return false;
1957         if (!xen_hvm_domain())
1958                 return false;
1959         if (xen_feature(XENFEAT_hvm_pirqs))
1960                 return false;
1961         return true;
1962 }
1963 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1964
1965 static void xen_set_cpu_features(struct cpuinfo_x86 *c)
1966 {
1967         if (xen_pv_domain()) {
1968                 clear_cpu_bug(c, X86_BUG_SYSRET_SS_ATTRS);
1969                 set_cpu_cap(c, X86_FEATURE_XENPV);
1970         }
1971 }
1972
1973 static void xen_pin_vcpu(int cpu)
1974 {
1975         static bool disable_pinning;
1976         struct sched_pin_override pin_override;
1977         int ret;
1978
1979         if (disable_pinning)
1980                 return;
1981
1982         pin_override.pcpu = cpu;
1983         ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
1984
1985         /* Ignore errors when removing override. */
1986         if (cpu < 0)
1987                 return;
1988
1989         switch (ret) {
1990         case -ENOSYS:
1991                 pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
1992                         cpu);
1993                 disable_pinning = true;
1994                 break;
1995         case -EPERM:
1996                 WARN(1, "Trying to pin vcpu without having privilege to do so\n");
1997                 disable_pinning = true;
1998                 break;
1999         case -EINVAL:
2000         case -EBUSY:
2001                 pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
2002                         cpu);
2003                 break;
2004         case 0:
2005                 break;
2006         default:
2007                 WARN(1, "rc %d while trying to pin vcpu\n", ret);
2008                 disable_pinning = true;
2009         }
2010 }
2011
2012 const struct hypervisor_x86 x86_hyper_xen = {
2013         .name                   = "Xen",
2014         .detect                 = xen_platform,
2015 #ifdef CONFIG_XEN_PVHVM
2016         .init_platform          = xen_hvm_guest_init,
2017 #endif
2018         .x2apic_available       = xen_x2apic_para_available,
2019         .set_cpu_features       = xen_set_cpu_features,
2020         .pin_vcpu               = xen_pin_vcpu,
2021 };
2022 EXPORT_SYMBOL(x86_hyper_xen);
2023
2024 #ifdef CONFIG_HOTPLUG_CPU
2025 void xen_arch_register_cpu(int num)
2026 {
2027         arch_register_cpu(num);
2028 }
2029 EXPORT_SYMBOL(xen_arch_register_cpu);
2030
2031 void xen_arch_unregister_cpu(int num)
2032 {
2033         arch_unregister_cpu(num);
2034 }
2035 EXPORT_SYMBOL(xen_arch_unregister_cpu);
2036 #endif