Merge branch 'x86/core' into x86/mce2
[linux-2.6] / 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/kernel.h>
15 #include <linux/init.h>
16 #include <linux/smp.h>
17 #include <linux/preempt.h>
18 #include <linux/hardirq.h>
19 #include <linux/percpu.h>
20 #include <linux/delay.h>
21 #include <linux/start_kernel.h>
22 #include <linux/sched.h>
23 #include <linux/bootmem.h>
24 #include <linux/module.h>
25 #include <linux/mm.h>
26 #include <linux/page-flags.h>
27 #include <linux/highmem.h>
28 #include <linux/console.h>
29
30 #include <xen/interface/xen.h>
31 #include <xen/interface/version.h>
32 #include <xen/interface/physdev.h>
33 #include <xen/interface/vcpu.h>
34 #include <xen/features.h>
35 #include <xen/page.h>
36 #include <xen/hvc-console.h>
37
38 #include <asm/paravirt.h>
39 #include <asm/apic.h>
40 #include <asm/page.h>
41 #include <asm/xen/hypercall.h>
42 #include <asm/xen/hypervisor.h>
43 #include <asm/fixmap.h>
44 #include <asm/processor.h>
45 #include <asm/msr-index.h>
46 #include <asm/setup.h>
47 #include <asm/desc.h>
48 #include <asm/pgtable.h>
49 #include <asm/tlbflush.h>
50 #include <asm/reboot.h>
51
52 #include "xen-ops.h"
53 #include "mmu.h"
54 #include "multicalls.h"
55
56 EXPORT_SYMBOL_GPL(hypercall_page);
57
58 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
59 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
60
61 enum xen_domain_type xen_domain_type = XEN_NATIVE;
62 EXPORT_SYMBOL_GPL(xen_domain_type);
63
64 struct start_info *xen_start_info;
65 EXPORT_SYMBOL_GPL(xen_start_info);
66
67 struct shared_info xen_dummy_shared_info;
68
69 void *xen_initial_gdt;
70
71 /*
72  * Point at some empty memory to start with. We map the real shared_info
73  * page as soon as fixmap is up and running.
74  */
75 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
76
77 /*
78  * Flag to determine whether vcpu info placement is available on all
79  * VCPUs.  We assume it is to start with, and then set it to zero on
80  * the first failure.  This is because it can succeed on some VCPUs
81  * and not others, since it can involve hypervisor memory allocation,
82  * or because the guest failed to guarantee all the appropriate
83  * constraints on all VCPUs (ie buffer can't cross a page boundary).
84  *
85  * Note that any particular CPU may be using a placed vcpu structure,
86  * but we can only optimise if the all are.
87  *
88  * 0: not available, 1: available
89  */
90 static int have_vcpu_info_placement = 1;
91
92 static void xen_vcpu_setup(int cpu)
93 {
94         struct vcpu_register_vcpu_info info;
95         int err;
96         struct vcpu_info *vcpup;
97
98         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
99         per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
100
101         if (!have_vcpu_info_placement)
102                 return;         /* already tested, not available */
103
104         vcpup = &per_cpu(xen_vcpu_info, cpu);
105
106         info.mfn = virt_to_mfn(vcpup);
107         info.offset = offset_in_page(vcpup);
108
109         printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
110                cpu, vcpup, info.mfn, info.offset);
111
112         /* Check to see if the hypervisor will put the vcpu_info
113            structure where we want it, which allows direct access via
114            a percpu-variable. */
115         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
116
117         if (err) {
118                 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
119                 have_vcpu_info_placement = 0;
120         } else {
121                 /* This cpu is using the registered vcpu info, even if
122                    later ones fail to. */
123                 per_cpu(xen_vcpu, cpu) = vcpup;
124
125                 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
126                        cpu, vcpup);
127         }
128 }
129
130 /*
131  * On restore, set the vcpu placement up again.
132  * If it fails, then we're in a bad state, since
133  * we can't back out from using it...
134  */
135 void xen_vcpu_restore(void)
136 {
137         if (have_vcpu_info_placement) {
138                 int cpu;
139
140                 for_each_online_cpu(cpu) {
141                         bool other_cpu = (cpu != smp_processor_id());
142
143                         if (other_cpu &&
144                             HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
145                                 BUG();
146
147                         xen_vcpu_setup(cpu);
148
149                         if (other_cpu &&
150                             HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
151                                 BUG();
152                 }
153
154                 BUG_ON(!have_vcpu_info_placement);
155         }
156 }
157
158 static void __init xen_banner(void)
159 {
160         unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
161         struct xen_extraversion extra;
162         HYPERVISOR_xen_version(XENVER_extraversion, &extra);
163
164         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
165                pv_info.name);
166         printk(KERN_INFO "Xen version: %d.%d%s%s\n",
167                version >> 16, version & 0xffff, extra.extraversion,
168                xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
169 }
170
171 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
172                       unsigned int *cx, unsigned int *dx)
173 {
174         unsigned maskedx = ~0;
175
176         /*
177          * Mask out inconvenient features, to try and disable as many
178          * unsupported kernel subsystems as possible.
179          */
180         if (*ax == 1)
181                 maskedx = ~((1 << X86_FEATURE_APIC) |  /* disable APIC */
182                             (1 << X86_FEATURE_ACPI) |  /* disable ACPI */
183                             (1 << X86_FEATURE_MCE)  |  /* disable MCE */
184                             (1 << X86_FEATURE_MCA)  |  /* disable MCA */
185                             (1 << X86_FEATURE_ACC));   /* thermal monitoring */
186
187         asm(XEN_EMULATE_PREFIX "cpuid"
188                 : "=a" (*ax),
189                   "=b" (*bx),
190                   "=c" (*cx),
191                   "=d" (*dx)
192                 : "0" (*ax), "2" (*cx));
193         *dx &= maskedx;
194 }
195
196 static void xen_set_debugreg(int reg, unsigned long val)
197 {
198         HYPERVISOR_set_debugreg(reg, val);
199 }
200
201 static unsigned long xen_get_debugreg(int reg)
202 {
203         return HYPERVISOR_get_debugreg(reg);
204 }
205
206 void xen_leave_lazy(void)
207 {
208         paravirt_leave_lazy(paravirt_get_lazy_mode());
209         xen_mc_flush();
210 }
211
212 static unsigned long xen_store_tr(void)
213 {
214         return 0;
215 }
216
217 /*
218  * Set the page permissions for a particular virtual address.  If the
219  * address is a vmalloc mapping (or other non-linear mapping), then
220  * find the linear mapping of the page and also set its protections to
221  * match.
222  */
223 static void set_aliased_prot(void *v, pgprot_t prot)
224 {
225         int level;
226         pte_t *ptep;
227         pte_t pte;
228         unsigned long pfn;
229         struct page *page;
230
231         ptep = lookup_address((unsigned long)v, &level);
232         BUG_ON(ptep == NULL);
233
234         pfn = pte_pfn(*ptep);
235         page = pfn_to_page(pfn);
236
237         pte = pfn_pte(pfn, prot);
238
239         if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
240                 BUG();
241
242         if (!PageHighMem(page)) {
243                 void *av = __va(PFN_PHYS(pfn));
244
245                 if (av != v)
246                         if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
247                                 BUG();
248         } else
249                 kmap_flush_unused();
250 }
251
252 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
253 {
254         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
255         int i;
256
257         for(i = 0; i < entries; i += entries_per_page)
258                 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
259 }
260
261 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
262 {
263         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
264         int i;
265
266         for(i = 0; i < entries; i += entries_per_page)
267                 set_aliased_prot(ldt + i, PAGE_KERNEL);
268 }
269
270 static void xen_set_ldt(const void *addr, unsigned entries)
271 {
272         struct mmuext_op *op;
273         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
274
275         op = mcs.args;
276         op->cmd = MMUEXT_SET_LDT;
277         op->arg1.linear_addr = (unsigned long)addr;
278         op->arg2.nr_ents = entries;
279
280         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
281
282         xen_mc_issue(PARAVIRT_LAZY_CPU);
283 }
284
285 static void xen_load_gdt(const struct desc_ptr *dtr)
286 {
287         unsigned long *frames;
288         unsigned long va = dtr->address;
289         unsigned int size = dtr->size + 1;
290         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
291         int f;
292         struct multicall_space mcs;
293
294         /* A GDT can be up to 64k in size, which corresponds to 8192
295            8-byte entries, or 16 4k pages.. */
296
297         BUG_ON(size > 65536);
298         BUG_ON(va & ~PAGE_MASK);
299
300         mcs = xen_mc_entry(sizeof(*frames) * pages);
301         frames = mcs.args;
302
303         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
304                 frames[f] = virt_to_mfn(va);
305                 make_lowmem_page_readonly((void *)va);
306         }
307
308         MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
309
310         xen_mc_issue(PARAVIRT_LAZY_CPU);
311 }
312
313 static void load_TLS_descriptor(struct thread_struct *t,
314                                 unsigned int cpu, unsigned int i)
315 {
316         struct desc_struct *gdt = get_cpu_gdt_table(cpu);
317         xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
318         struct multicall_space mc = __xen_mc_entry(0);
319
320         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
321 }
322
323 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
324 {
325         /*
326          * XXX sleazy hack: If we're being called in a lazy-cpu zone
327          * and lazy gs handling is enabled, it means we're in a
328          * context switch, and %gs has just been saved.  This means we
329          * can zero it out to prevent faults on exit from the
330          * hypervisor if the next process has no %gs.  Either way, it
331          * has been saved, and the new value will get loaded properly.
332          * This will go away as soon as Xen has been modified to not
333          * save/restore %gs for normal hypercalls.
334          *
335          * On x86_64, this hack is not used for %gs, because gs points
336          * to KERNEL_GS_BASE (and uses it for PDA references), so we
337          * must not zero %gs on x86_64
338          *
339          * For x86_64, we need to zero %fs, otherwise we may get an
340          * exception between the new %fs descriptor being loaded and
341          * %fs being effectively cleared at __switch_to().
342          */
343         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
344 #ifdef CONFIG_X86_32
345                 lazy_load_gs(0);
346 #else
347                 loadsegment(fs, 0);
348 #endif
349         }
350
351         xen_mc_batch();
352
353         load_TLS_descriptor(t, cpu, 0);
354         load_TLS_descriptor(t, cpu, 1);
355         load_TLS_descriptor(t, cpu, 2);
356
357         xen_mc_issue(PARAVIRT_LAZY_CPU);
358 }
359
360 #ifdef CONFIG_X86_64
361 static void xen_load_gs_index(unsigned int idx)
362 {
363         if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
364                 BUG();
365 }
366 #endif
367
368 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
369                                 const void *ptr)
370 {
371         xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
372         u64 entry = *(u64 *)ptr;
373
374         preempt_disable();
375
376         xen_mc_flush();
377         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
378                 BUG();
379
380         preempt_enable();
381 }
382
383 static int cvt_gate_to_trap(int vector, const gate_desc *val,
384                             struct trap_info *info)
385 {
386         if (val->type != 0xf && val->type != 0xe)
387                 return 0;
388
389         info->vector = vector;
390         info->address = gate_offset(*val);
391         info->cs = gate_segment(*val);
392         info->flags = val->dpl;
393         /* interrupt gates clear IF */
394         if (val->type == 0xe)
395                 info->flags |= 4;
396
397         return 1;
398 }
399
400 /* Locations of each CPU's IDT */
401 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
402
403 /* Set an IDT entry.  If the entry is part of the current IDT, then
404    also update Xen. */
405 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
406 {
407         unsigned long p = (unsigned long)&dt[entrynum];
408         unsigned long start, end;
409
410         preempt_disable();
411
412         start = __get_cpu_var(idt_desc).address;
413         end = start + __get_cpu_var(idt_desc).size + 1;
414
415         xen_mc_flush();
416
417         native_write_idt_entry(dt, entrynum, g);
418
419         if (p >= start && (p + 8) <= end) {
420                 struct trap_info info[2];
421
422                 info[1].address = 0;
423
424                 if (cvt_gate_to_trap(entrynum, g, &info[0]))
425                         if (HYPERVISOR_set_trap_table(info))
426                                 BUG();
427         }
428
429         preempt_enable();
430 }
431
432 static void xen_convert_trap_info(const struct desc_ptr *desc,
433                                   struct trap_info *traps)
434 {
435         unsigned in, out, count;
436
437         count = (desc->size+1) / sizeof(gate_desc);
438         BUG_ON(count > 256);
439
440         for (in = out = 0; in < count; in++) {
441                 gate_desc *entry = (gate_desc*)(desc->address) + in;
442
443                 if (cvt_gate_to_trap(in, entry, &traps[out]))
444                         out++;
445         }
446         traps[out].address = 0;
447 }
448
449 void xen_copy_trap_info(struct trap_info *traps)
450 {
451         const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
452
453         xen_convert_trap_info(desc, traps);
454 }
455
456 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
457    hold a spinlock to protect the static traps[] array (static because
458    it avoids allocation, and saves stack space). */
459 static void xen_load_idt(const struct desc_ptr *desc)
460 {
461         static DEFINE_SPINLOCK(lock);
462         static struct trap_info traps[257];
463
464         spin_lock(&lock);
465
466         __get_cpu_var(idt_desc) = *desc;
467
468         xen_convert_trap_info(desc, traps);
469
470         xen_mc_flush();
471         if (HYPERVISOR_set_trap_table(traps))
472                 BUG();
473
474         spin_unlock(&lock);
475 }
476
477 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
478    they're handled differently. */
479 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
480                                 const void *desc, int type)
481 {
482         preempt_disable();
483
484         switch (type) {
485         case DESC_LDT:
486         case DESC_TSS:
487                 /* ignore */
488                 break;
489
490         default: {
491                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
492
493                 xen_mc_flush();
494                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
495                         BUG();
496         }
497
498         }
499
500         preempt_enable();
501 }
502
503 static void xen_load_sp0(struct tss_struct *tss,
504                          struct thread_struct *thread)
505 {
506         struct multicall_space mcs = xen_mc_entry(0);
507         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
508         xen_mc_issue(PARAVIRT_LAZY_CPU);
509 }
510
511 static void xen_set_iopl_mask(unsigned mask)
512 {
513         struct physdev_set_iopl set_iopl;
514
515         /* Force the change at ring 0. */
516         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
517         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
518 }
519
520 static void xen_io_delay(void)
521 {
522 }
523
524 #ifdef CONFIG_X86_LOCAL_APIC
525 static u32 xen_apic_read(u32 reg)
526 {
527         return 0;
528 }
529
530 static void xen_apic_write(u32 reg, u32 val)
531 {
532         /* Warn to see if there's any stray references */
533         WARN_ON(1);
534 }
535
536 static u64 xen_apic_icr_read(void)
537 {
538         return 0;
539 }
540
541 static void xen_apic_icr_write(u32 low, u32 id)
542 {
543         /* Warn to see if there's any stray references */
544         WARN_ON(1);
545 }
546
547 static void xen_apic_wait_icr_idle(void)
548 {
549         return;
550 }
551
552 static u32 xen_safe_apic_wait_icr_idle(void)
553 {
554         return 0;
555 }
556
557 static void set_xen_basic_apic_ops(void)
558 {
559         apic->read = xen_apic_read;
560         apic->write = xen_apic_write;
561         apic->icr_read = xen_apic_icr_read;
562         apic->icr_write = xen_apic_icr_write;
563         apic->wait_icr_idle = xen_apic_wait_icr_idle;
564         apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
565 }
566
567 #endif
568
569
570 static void xen_clts(void)
571 {
572         struct multicall_space mcs;
573
574         mcs = xen_mc_entry(0);
575
576         MULTI_fpu_taskswitch(mcs.mc, 0);
577
578         xen_mc_issue(PARAVIRT_LAZY_CPU);
579 }
580
581 static void xen_write_cr0(unsigned long cr0)
582 {
583         struct multicall_space mcs;
584
585         /* Only pay attention to cr0.TS; everything else is
586            ignored. */
587         mcs = xen_mc_entry(0);
588
589         MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
590
591         xen_mc_issue(PARAVIRT_LAZY_CPU);
592 }
593
594 static void xen_write_cr4(unsigned long cr4)
595 {
596         cr4 &= ~X86_CR4_PGE;
597         cr4 &= ~X86_CR4_PSE;
598
599         native_write_cr4(cr4);
600 }
601
602 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
603 {
604         int ret;
605
606         ret = 0;
607
608         switch (msr) {
609 #ifdef CONFIG_X86_64
610                 unsigned which;
611                 u64 base;
612
613         case MSR_FS_BASE:               which = SEGBASE_FS; goto set;
614         case MSR_KERNEL_GS_BASE:        which = SEGBASE_GS_USER; goto set;
615         case MSR_GS_BASE:               which = SEGBASE_GS_KERNEL; goto set;
616
617         set:
618                 base = ((u64)high << 32) | low;
619                 if (HYPERVISOR_set_segment_base(which, base) != 0)
620                         ret = -EFAULT;
621                 break;
622 #endif
623
624         case MSR_STAR:
625         case MSR_CSTAR:
626         case MSR_LSTAR:
627         case MSR_SYSCALL_MASK:
628         case MSR_IA32_SYSENTER_CS:
629         case MSR_IA32_SYSENTER_ESP:
630         case MSR_IA32_SYSENTER_EIP:
631                 /* Fast syscall setup is all done in hypercalls, so
632                    these are all ignored.  Stub them out here to stop
633                    Xen console noise. */
634                 break;
635
636         default:
637                 ret = native_write_msr_safe(msr, low, high);
638         }
639
640         return ret;
641 }
642
643 void xen_setup_shared_info(void)
644 {
645         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
646                 set_fixmap(FIX_PARAVIRT_BOOTMAP,
647                            xen_start_info->shared_info);
648
649                 HYPERVISOR_shared_info =
650                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
651         } else
652                 HYPERVISOR_shared_info =
653                         (struct shared_info *)__va(xen_start_info->shared_info);
654
655 #ifndef CONFIG_SMP
656         /* In UP this is as good a place as any to set up shared info */
657         xen_setup_vcpu_info_placement();
658 #endif
659
660         xen_setup_mfn_list_list();
661 }
662
663 /* This is called once we have the cpu_possible_map */
664 void xen_setup_vcpu_info_placement(void)
665 {
666         int cpu;
667
668         for_each_possible_cpu(cpu)
669                 xen_vcpu_setup(cpu);
670
671         /* xen_vcpu_setup managed to place the vcpu_info within the
672            percpu area for all cpus, so make use of it */
673         if (have_vcpu_info_placement) {
674                 printk(KERN_INFO "Xen: using vcpu_info placement\n");
675
676                 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
677                 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
678                 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
679                 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
680                 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
681         }
682 }
683
684 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
685                           unsigned long addr, unsigned len)
686 {
687         char *start, *end, *reloc;
688         unsigned ret;
689
690         start = end = reloc = NULL;
691
692 #define SITE(op, x)                                                     \
693         case PARAVIRT_PATCH(op.x):                                      \
694         if (have_vcpu_info_placement) {                                 \
695                 start = (char *)xen_##x##_direct;                       \
696                 end = xen_##x##_direct_end;                             \
697                 reloc = xen_##x##_direct_reloc;                         \
698         }                                                               \
699         goto patch_site
700
701         switch (type) {
702                 SITE(pv_irq_ops, irq_enable);
703                 SITE(pv_irq_ops, irq_disable);
704                 SITE(pv_irq_ops, save_fl);
705                 SITE(pv_irq_ops, restore_fl);
706 #undef SITE
707
708         patch_site:
709                 if (start == NULL || (end-start) > len)
710                         goto default_patch;
711
712                 ret = paravirt_patch_insns(insnbuf, len, start, end);
713
714                 /* Note: because reloc is assigned from something that
715                    appears to be an array, gcc assumes it's non-null,
716                    but doesn't know its relationship with start and
717                    end. */
718                 if (reloc > start && reloc < end) {
719                         int reloc_off = reloc - start;
720                         long *relocp = (long *)(insnbuf + reloc_off);
721                         long delta = start - (char *)addr;
722
723                         *relocp += delta;
724                 }
725                 break;
726
727         default_patch:
728         default:
729                 ret = paravirt_patch_default(type, clobbers, insnbuf,
730                                              addr, len);
731                 break;
732         }
733
734         return ret;
735 }
736
737 static const struct pv_info xen_info __initdata = {
738         .paravirt_enabled = 1,
739         .shared_kernel_pmd = 0,
740
741         .name = "Xen",
742 };
743
744 static const struct pv_init_ops xen_init_ops __initdata = {
745         .patch = xen_patch,
746
747         .banner = xen_banner,
748         .memory_setup = xen_memory_setup,
749         .arch_setup = xen_arch_setup,
750         .post_allocator_init = xen_post_allocator_init,
751 };
752
753 static const struct pv_time_ops xen_time_ops __initdata = {
754         .time_init = xen_time_init,
755
756         .set_wallclock = xen_set_wallclock,
757         .get_wallclock = xen_get_wallclock,
758         .get_tsc_khz = xen_tsc_khz,
759         .sched_clock = xen_sched_clock,
760 };
761
762 static const struct pv_cpu_ops xen_cpu_ops __initdata = {
763         .cpuid = xen_cpuid,
764
765         .set_debugreg = xen_set_debugreg,
766         .get_debugreg = xen_get_debugreg,
767
768         .clts = xen_clts,
769
770         .read_cr0 = native_read_cr0,
771         .write_cr0 = xen_write_cr0,
772
773         .read_cr4 = native_read_cr4,
774         .read_cr4_safe = native_read_cr4_safe,
775         .write_cr4 = xen_write_cr4,
776
777         .wbinvd = native_wbinvd,
778
779         .read_msr = native_read_msr_safe,
780         .write_msr = xen_write_msr_safe,
781         .read_tsc = native_read_tsc,
782         .read_pmc = native_read_pmc,
783
784         .iret = xen_iret,
785         .irq_enable_sysexit = xen_sysexit,
786 #ifdef CONFIG_X86_64
787         .usergs_sysret32 = xen_sysret32,
788         .usergs_sysret64 = xen_sysret64,
789 #endif
790
791         .load_tr_desc = paravirt_nop,
792         .set_ldt = xen_set_ldt,
793         .load_gdt = xen_load_gdt,
794         .load_idt = xen_load_idt,
795         .load_tls = xen_load_tls,
796 #ifdef CONFIG_X86_64
797         .load_gs_index = xen_load_gs_index,
798 #endif
799
800         .alloc_ldt = xen_alloc_ldt,
801         .free_ldt = xen_free_ldt,
802
803         .store_gdt = native_store_gdt,
804         .store_idt = native_store_idt,
805         .store_tr = xen_store_tr,
806
807         .write_ldt_entry = xen_write_ldt_entry,
808         .write_gdt_entry = xen_write_gdt_entry,
809         .write_idt_entry = xen_write_idt_entry,
810         .load_sp0 = xen_load_sp0,
811
812         .set_iopl_mask = xen_set_iopl_mask,
813         .io_delay = xen_io_delay,
814
815         /* Xen takes care of %gs when switching to usermode for us */
816         .swapgs = paravirt_nop,
817
818         .lazy_mode = {
819                 .enter = paravirt_enter_lazy_cpu,
820                 .leave = xen_leave_lazy,
821         },
822 };
823
824 static const struct pv_apic_ops xen_apic_ops __initdata = {
825 #ifdef CONFIG_X86_LOCAL_APIC
826         .setup_boot_clock = paravirt_nop,
827         .setup_secondary_clock = paravirt_nop,
828         .startup_ipi_hook = paravirt_nop,
829 #endif
830 };
831
832 static void xen_reboot(int reason)
833 {
834         struct sched_shutdown r = { .reason = reason };
835
836 #ifdef CONFIG_SMP
837         smp_send_stop();
838 #endif
839
840         if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
841                 BUG();
842 }
843
844 static void xen_restart(char *msg)
845 {
846         xen_reboot(SHUTDOWN_reboot);
847 }
848
849 static void xen_emergency_restart(void)
850 {
851         xen_reboot(SHUTDOWN_reboot);
852 }
853
854 static void xen_machine_halt(void)
855 {
856         xen_reboot(SHUTDOWN_poweroff);
857 }
858
859 static void xen_crash_shutdown(struct pt_regs *regs)
860 {
861         xen_reboot(SHUTDOWN_crash);
862 }
863
864 static const struct machine_ops __initdata xen_machine_ops = {
865         .restart = xen_restart,
866         .halt = xen_machine_halt,
867         .power_off = xen_machine_halt,
868         .shutdown = xen_machine_halt,
869         .crash_shutdown = xen_crash_shutdown,
870         .emergency_restart = xen_emergency_restart,
871 };
872
873
874 /* First C function to be called on Xen boot */
875 asmlinkage void __init xen_start_kernel(void)
876 {
877         pgd_t *pgd;
878
879         if (!xen_start_info)
880                 return;
881
882         xen_domain_type = XEN_PV_DOMAIN;
883
884         BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
885
886         xen_setup_features();
887
888         /* Install Xen paravirt ops */
889         pv_info = xen_info;
890         pv_init_ops = xen_init_ops;
891         pv_time_ops = xen_time_ops;
892         pv_cpu_ops = xen_cpu_ops;
893         pv_apic_ops = xen_apic_ops;
894         pv_mmu_ops = xen_mmu_ops;
895
896         xen_init_irq_ops();
897
898 #ifdef CONFIG_X86_LOCAL_APIC
899         /*
900          * set up the basic apic ops.
901          */
902         set_xen_basic_apic_ops();
903 #endif
904
905         if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
906                 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
907                 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
908         }
909
910         machine_ops = xen_machine_ops;
911
912 #ifdef CONFIG_X86_64
913         /*
914          * Setup percpu state.  We only need to do this for 64-bit
915          * because 32-bit already has %fs set properly.
916          */
917         load_percpu_segment(0);
918 #endif
919         /*
920          * The only reliable way to retain the initial address of the
921          * percpu gdt_page is to remember it here, so we can go and
922          * mark it RW later, when the initial percpu area is freed.
923          */
924         xen_initial_gdt = &per_cpu(gdt_page, 0);
925
926         xen_smp_init();
927
928         /* Get mfn list */
929         if (!xen_feature(XENFEAT_auto_translated_physmap))
930                 xen_build_dynamic_phys_to_machine();
931
932         pgd = (pgd_t *)xen_start_info->pt_base;
933
934         /* Prevent unwanted bits from being set in PTEs. */
935         __supported_pte_mask &= ~_PAGE_GLOBAL;
936         if (!xen_initial_domain())
937                 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
938
939         /* Don't do the full vcpu_info placement stuff until we have a
940            possible map and a non-dummy shared_info. */
941         per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
942
943         local_irq_disable();
944         early_boot_irqs_off();
945
946         xen_raw_console_write("mapping kernel into physical memory\n");
947         pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
948
949         init_mm.pgd = pgd;
950
951         /* keep using Xen gdt for now; no urgent need to change it */
952
953         pv_info.kernel_rpl = 1;
954         if (xen_feature(XENFEAT_supervisor_mode_kernel))
955                 pv_info.kernel_rpl = 0;
956
957         /* set the limit of our address space */
958         xen_reserve_top();
959
960 #ifdef CONFIG_X86_32
961         /* set up basic CPUID stuff */
962         cpu_detect(&new_cpu_data);
963         new_cpu_data.hard_math = 1;
964         new_cpu_data.x86_capability[0] = cpuid_edx(1);
965 #endif
966
967         /* Poke various useful things into boot_params */
968         boot_params.hdr.type_of_loader = (9 << 4) | 0;
969         boot_params.hdr.ramdisk_image = xen_start_info->mod_start
970                 ? __pa(xen_start_info->mod_start) : 0;
971         boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
972         boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
973
974         if (!xen_initial_domain()) {
975                 add_preferred_console("xenboot", 0, NULL);
976                 add_preferred_console("tty", 0, NULL);
977                 add_preferred_console("hvc", 0, NULL);
978         }
979
980         xen_raw_console_write("about to get started...\n");
981
982         /* Start the world */
983 #ifdef CONFIG_X86_32
984         i386_start_kernel();
985 #else
986         x86_64_start_reservations((char *)__pa_symbol(&boot_params));
987 #endif
988 }