xen: remove cast from HYPERVISOR_shared_info assignment
[pandora-kernel.git] / arch / x86 / xen / enlighten.c
1 /*
2  * Core of Xen paravirt_ops implementation.
3  *
4  * This file contains the xen_paravirt_ops structure itself, and the
5  * implementations for:
6  * - privileged instructions
7  * - interrupt flags
8  * - segment operations
9  * - booting and setup
10  *
11  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
12  */
13
14 #include <linux/cpu.h>
15 #include <linux/kernel.h>
16 #include <linux/init.h>
17 #include <linux/smp.h>
18 #include <linux/preempt.h>
19 #include <linux/hardirq.h>
20 #include <linux/percpu.h>
21 #include <linux/delay.h>
22 #include <linux/start_kernel.h>
23 #include <linux/sched.h>
24 #include <linux/kprobes.h>
25 #include <linux/bootmem.h>
26 #include <linux/module.h>
27 #include <linux/mm.h>
28 #include <linux/page-flags.h>
29 #include <linux/highmem.h>
30 #include <linux/console.h>
31 #include <linux/pci.h>
32 #include <linux/gfp.h>
33 #include <linux/memblock.h>
34
35 #include <xen/xen.h>
36 #include <xen/interface/xen.h>
37 #include <xen/interface/version.h>
38 #include <xen/interface/physdev.h>
39 #include <xen/interface/vcpu.h>
40 #include <xen/interface/memory.h>
41 #include <xen/interface/xen-mca.h>
42 #include <xen/features.h>
43 #include <xen/page.h>
44 #include <xen/hvm.h>
45 #include <xen/hvc-console.h>
46 #include <xen/acpi.h>
47
48 #include <asm/paravirt.h>
49 #include <asm/apic.h>
50 #include <asm/page.h>
51 #include <asm/xen/pci.h>
52 #include <asm/xen/hypercall.h>
53 #include <asm/xen/hypervisor.h>
54 #include <asm/fixmap.h>
55 #include <asm/processor.h>
56 #include <asm/proto.h>
57 #include <asm/msr-index.h>
58 #include <asm/traps.h>
59 #include <asm/setup.h>
60 #include <asm/desc.h>
61 #include <asm/pgalloc.h>
62 #include <asm/pgtable.h>
63 #include <asm/tlbflush.h>
64 #include <asm/reboot.h>
65 #include <asm/stackprotector.h>
66 #include <asm/hypervisor.h>
67 #include <asm/mwait.h>
68 #include <asm/pci_x86.h>
69
70 #ifdef CONFIG_ACPI
71 #include <linux/acpi.h>
72 #include <asm/acpi.h>
73 #include <acpi/pdc_intel.h>
74 #include <acpi/processor.h>
75 #include <xen/interface/platform.h>
76 #endif
77
78 #include "xen-ops.h"
79 #include "mmu.h"
80 #include "smp.h"
81 #include "multicalls.h"
82
83 EXPORT_SYMBOL_GPL(hypercall_page);
84
85 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
86 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
87
88 enum xen_domain_type xen_domain_type = XEN_NATIVE;
89 EXPORT_SYMBOL_GPL(xen_domain_type);
90
91 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
92 EXPORT_SYMBOL(machine_to_phys_mapping);
93 unsigned long  machine_to_phys_nr;
94 EXPORT_SYMBOL(machine_to_phys_nr);
95
96 struct start_info *xen_start_info;
97 EXPORT_SYMBOL_GPL(xen_start_info);
98
99 struct shared_info xen_dummy_shared_info;
100
101 void *xen_initial_gdt;
102
103 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
104 __read_mostly int xen_have_vector_callback;
105 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
106
107 /*
108  * Point at some empty memory to start with. We map the real shared_info
109  * page as soon as fixmap is up and running.
110  */
111 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
112
113 /*
114  * Flag to determine whether vcpu info placement is available on all
115  * VCPUs.  We assume it is to start with, and then set it to zero on
116  * the first failure.  This is because it can succeed on some VCPUs
117  * and not others, since it can involve hypervisor memory allocation,
118  * or because the guest failed to guarantee all the appropriate
119  * constraints on all VCPUs (ie buffer can't cross a page boundary).
120  *
121  * Note that any particular CPU may be using a placed vcpu structure,
122  * but we can only optimise if the all are.
123  *
124  * 0: not available, 1: available
125  */
126 static int have_vcpu_info_placement = 1;
127
128 struct tls_descs {
129         struct desc_struct desc[3];
130 };
131
132 /*
133  * Updating the 3 TLS descriptors in the GDT on every task switch is
134  * surprisingly expensive so we avoid updating them if they haven't
135  * changed.  Since Xen writes different descriptors than the one
136  * passed in the update_descriptor hypercall we keep shadow copies to
137  * compare against.
138  */
139 static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
140
141 static void clamp_max_cpus(void)
142 {
143 #ifdef CONFIG_SMP
144         if (setup_max_cpus > MAX_VIRT_CPUS)
145                 setup_max_cpus = MAX_VIRT_CPUS;
146 #endif
147 }
148
149 static void xen_vcpu_setup(int cpu)
150 {
151         struct vcpu_register_vcpu_info info;
152         int err;
153         struct vcpu_info *vcpup;
154
155         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
156
157         if (cpu < MAX_VIRT_CPUS)
158                 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
159
160         if (!have_vcpu_info_placement) {
161                 if (cpu >= MAX_VIRT_CPUS)
162                         clamp_max_cpus();
163                 return;
164         }
165
166         vcpup = &per_cpu(xen_vcpu_info, cpu);
167         info.mfn = arbitrary_virt_to_mfn(vcpup);
168         info.offset = offset_in_page(vcpup);
169
170         /* Check to see if the hypervisor will put the vcpu_info
171            structure where we want it, which allows direct access via
172            a percpu-variable. */
173         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
174
175         if (err) {
176                 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
177                 have_vcpu_info_placement = 0;
178                 clamp_max_cpus();
179         } else {
180                 /* This cpu is using the registered vcpu info, even if
181                    later ones fail to. */
182                 per_cpu(xen_vcpu, cpu) = vcpup;
183         }
184 }
185
186 /*
187  * On restore, set the vcpu placement up again.
188  * If it fails, then we're in a bad state, since
189  * we can't back out from using it...
190  */
191 void xen_vcpu_restore(void)
192 {
193         int cpu;
194
195         for_each_online_cpu(cpu) {
196                 bool other_cpu = (cpu != smp_processor_id());
197
198                 if (other_cpu &&
199                     HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
200                         BUG();
201
202                 xen_setup_runstate_info(cpu);
203
204                 if (have_vcpu_info_placement)
205                         xen_vcpu_setup(cpu);
206
207                 if (other_cpu &&
208                     HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
209                         BUG();
210         }
211 }
212
213 static void __init xen_banner(void)
214 {
215         unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
216         struct xen_extraversion extra;
217         HYPERVISOR_xen_version(XENVER_extraversion, &extra);
218
219         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
220                pv_info.name);
221         printk(KERN_INFO "Xen version: %d.%d%s%s\n",
222                version >> 16, version & 0xffff, extra.extraversion,
223                xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
224 }
225
226 #define CPUID_THERM_POWER_LEAF 6
227 #define APERFMPERF_PRESENT 0
228
229 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
230 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
231
232 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
233 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
234 static __read_mostly unsigned int cpuid_leaf5_edx_val;
235
236 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
237                       unsigned int *cx, unsigned int *dx)
238 {
239         unsigned maskebx = ~0;
240         unsigned maskecx = ~0;
241         unsigned maskedx = ~0;
242         unsigned setecx = 0;
243         /*
244          * Mask out inconvenient features, to try and disable as many
245          * unsupported kernel subsystems as possible.
246          */
247         switch (*ax) {
248         case 1:
249                 maskecx = cpuid_leaf1_ecx_mask;
250                 setecx = cpuid_leaf1_ecx_set_mask;
251                 maskedx = cpuid_leaf1_edx_mask;
252                 break;
253
254         case CPUID_MWAIT_LEAF:
255                 /* Synthesize the values.. */
256                 *ax = 0;
257                 *bx = 0;
258                 *cx = cpuid_leaf5_ecx_val;
259                 *dx = cpuid_leaf5_edx_val;
260                 return;
261
262         case CPUID_THERM_POWER_LEAF:
263                 /* Disabling APERFMPERF for kernel usage */
264                 maskecx = ~(1 << APERFMPERF_PRESENT);
265                 break;
266
267         case 0xb:
268                 /* Suppress extended topology stuff */
269                 maskebx = 0;
270                 break;
271         }
272
273         asm(XEN_EMULATE_PREFIX "cpuid"
274                 : "=a" (*ax),
275                   "=b" (*bx),
276                   "=c" (*cx),
277                   "=d" (*dx)
278                 : "0" (*ax), "2" (*cx));
279
280         *bx &= maskebx;
281         *cx &= maskecx;
282         *cx |= setecx;
283         *dx &= maskedx;
284
285 }
286
287 static bool __init xen_check_mwait(void)
288 {
289 #if defined(CONFIG_ACPI) && !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR) && \
290         !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR_MODULE)
291         struct xen_platform_op op = {
292                 .cmd                    = XENPF_set_processor_pminfo,
293                 .u.set_pminfo.id        = -1,
294                 .u.set_pminfo.type      = XEN_PM_PDC,
295         };
296         uint32_t buf[3];
297         unsigned int ax, bx, cx, dx;
298         unsigned int mwait_mask;
299
300         /* We need to determine whether it is OK to expose the MWAIT
301          * capability to the kernel to harvest deeper than C3 states from ACPI
302          * _CST using the processor_harvest_xen.c module. For this to work, we
303          * need to gather the MWAIT_LEAF values (which the cstate.c code
304          * checks against). The hypervisor won't expose the MWAIT flag because
305          * it would break backwards compatibility; so we will find out directly
306          * from the hardware and hypercall.
307          */
308         if (!xen_initial_domain())
309                 return false;
310
311         ax = 1;
312         cx = 0;
313
314         native_cpuid(&ax, &bx, &cx, &dx);
315
316         mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
317                      (1 << (X86_FEATURE_MWAIT % 32));
318
319         if ((cx & mwait_mask) != mwait_mask)
320                 return false;
321
322         /* We need to emulate the MWAIT_LEAF and for that we need both
323          * ecx and edx. The hypercall provides only partial information.
324          */
325
326         ax = CPUID_MWAIT_LEAF;
327         bx = 0;
328         cx = 0;
329         dx = 0;
330
331         native_cpuid(&ax, &bx, &cx, &dx);
332
333         /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
334          * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
335          */
336         buf[0] = ACPI_PDC_REVISION_ID;
337         buf[1] = 1;
338         buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
339
340         set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
341
342         if ((HYPERVISOR_dom0_op(&op) == 0) &&
343             (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
344                 cpuid_leaf5_ecx_val = cx;
345                 cpuid_leaf5_edx_val = dx;
346         }
347         return true;
348 #else
349         return false;
350 #endif
351 }
352 static void __init xen_init_cpuid_mask(void)
353 {
354         unsigned int ax, bx, cx, dx;
355         unsigned int xsave_mask;
356
357         cpuid_leaf1_edx_mask =
358                 ~((1 << X86_FEATURE_MTRR) |  /* disable MTRR */
359                   (1 << X86_FEATURE_ACC));   /* thermal monitoring */
360
361         if (!xen_initial_domain())
362                 cpuid_leaf1_edx_mask &=
363                         ~((1 << X86_FEATURE_APIC) |  /* disable local APIC */
364                           (1 << X86_FEATURE_ACPI));  /* disable ACPI */
365         ax = 1;
366         cx = 0;
367         xen_cpuid(&ax, &bx, &cx, &dx);
368
369         xsave_mask =
370                 (1 << (X86_FEATURE_XSAVE % 32)) |
371                 (1 << (X86_FEATURE_OSXSAVE % 32));
372
373         /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
374         if ((cx & xsave_mask) != xsave_mask)
375                 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
376         if (xen_check_mwait())
377                 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
378 }
379
380 static void xen_set_debugreg(int reg, unsigned long val)
381 {
382         HYPERVISOR_set_debugreg(reg, val);
383 }
384
385 static unsigned long xen_get_debugreg(int reg)
386 {
387         return HYPERVISOR_get_debugreg(reg);
388 }
389
390 static void xen_end_context_switch(struct task_struct *next)
391 {
392         xen_mc_flush();
393         paravirt_end_context_switch(next);
394 }
395
396 static unsigned long xen_store_tr(void)
397 {
398         return 0;
399 }
400
401 /*
402  * Set the page permissions for a particular virtual address.  If the
403  * address is a vmalloc mapping (or other non-linear mapping), then
404  * find the linear mapping of the page and also set its protections to
405  * match.
406  */
407 static void set_aliased_prot(void *v, pgprot_t prot)
408 {
409         int level;
410         pte_t *ptep;
411         pte_t pte;
412         unsigned long pfn;
413         struct page *page;
414
415         ptep = lookup_address((unsigned long)v, &level);
416         BUG_ON(ptep == NULL);
417
418         pfn = pte_pfn(*ptep);
419         page = pfn_to_page(pfn);
420
421         pte = pfn_pte(pfn, prot);
422
423         if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
424                 BUG();
425
426         if (!PageHighMem(page)) {
427                 void *av = __va(PFN_PHYS(pfn));
428
429                 if (av != v)
430                         if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
431                                 BUG();
432         } else
433                 kmap_flush_unused();
434 }
435
436 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
437 {
438         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
439         int i;
440
441         for(i = 0; i < entries; i += entries_per_page)
442                 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
443 }
444
445 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
446 {
447         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
448         int i;
449
450         for(i = 0; i < entries; i += entries_per_page)
451                 set_aliased_prot(ldt + i, PAGE_KERNEL);
452 }
453
454 static void xen_set_ldt(const void *addr, unsigned entries)
455 {
456         struct mmuext_op *op;
457         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
458
459         trace_xen_cpu_set_ldt(addr, entries);
460
461         op = mcs.args;
462         op->cmd = MMUEXT_SET_LDT;
463         op->arg1.linear_addr = (unsigned long)addr;
464         op->arg2.nr_ents = entries;
465
466         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
467
468         xen_mc_issue(PARAVIRT_LAZY_CPU);
469 }
470
471 static void xen_load_gdt(const struct desc_ptr *dtr)
472 {
473         unsigned long va = dtr->address;
474         unsigned int size = dtr->size + 1;
475         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
476         unsigned long frames[pages];
477         int f;
478
479         /*
480          * A GDT can be up to 64k in size, which corresponds to 8192
481          * 8-byte entries, or 16 4k pages..
482          */
483
484         BUG_ON(size > 65536);
485         BUG_ON(va & ~PAGE_MASK);
486
487         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
488                 int level;
489                 pte_t *ptep;
490                 unsigned long pfn, mfn;
491                 void *virt;
492
493                 /*
494                  * The GDT is per-cpu and is in the percpu data area.
495                  * That can be virtually mapped, so we need to do a
496                  * page-walk to get the underlying MFN for the
497                  * hypercall.  The page can also be in the kernel's
498                  * linear range, so we need to RO that mapping too.
499                  */
500                 ptep = lookup_address(va, &level);
501                 BUG_ON(ptep == NULL);
502
503                 pfn = pte_pfn(*ptep);
504                 mfn = pfn_to_mfn(pfn);
505                 virt = __va(PFN_PHYS(pfn));
506
507                 frames[f] = mfn;
508
509                 make_lowmem_page_readonly((void *)va);
510                 make_lowmem_page_readonly(virt);
511         }
512
513         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
514                 BUG();
515 }
516
517 /*
518  * load_gdt for early boot, when the gdt is only mapped once
519  */
520 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
521 {
522         unsigned long va = dtr->address;
523         unsigned int size = dtr->size + 1;
524         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
525         unsigned long frames[pages];
526         int f;
527
528         /*
529          * A GDT can be up to 64k in size, which corresponds to 8192
530          * 8-byte entries, or 16 4k pages..
531          */
532
533         BUG_ON(size > 65536);
534         BUG_ON(va & ~PAGE_MASK);
535
536         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
537                 pte_t pte;
538                 unsigned long pfn, mfn;
539
540                 pfn = virt_to_pfn(va);
541                 mfn = pfn_to_mfn(pfn);
542
543                 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
544
545                 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
546                         BUG();
547
548                 frames[f] = mfn;
549         }
550
551         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
552                 BUG();
553 }
554
555 static inline bool desc_equal(const struct desc_struct *d1,
556                               const struct desc_struct *d2)
557 {
558         return d1->a == d2->a && d1->b == d2->b;
559 }
560
561 static void load_TLS_descriptor(struct thread_struct *t,
562                                 unsigned int cpu, unsigned int i)
563 {
564         struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
565         struct desc_struct *gdt;
566         xmaddr_t maddr;
567         struct multicall_space mc;
568
569         if (desc_equal(shadow, &t->tls_array[i]))
570                 return;
571
572         *shadow = t->tls_array[i];
573
574         gdt = get_cpu_gdt_table(cpu);
575         maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
576         mc = __xen_mc_entry(0);
577
578         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
579 }
580
581 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
582 {
583         /*
584          * XXX sleazy hack: If we're being called in a lazy-cpu zone
585          * and lazy gs handling is enabled, it means we're in a
586          * context switch, and %gs has just been saved.  This means we
587          * can zero it out to prevent faults on exit from the
588          * hypervisor if the next process has no %gs.  Either way, it
589          * has been saved, and the new value will get loaded properly.
590          * This will go away as soon as Xen has been modified to not
591          * save/restore %gs for normal hypercalls.
592          *
593          * On x86_64, this hack is not used for %gs, because gs points
594          * to KERNEL_GS_BASE (and uses it for PDA references), so we
595          * must not zero %gs on x86_64
596          *
597          * For x86_64, we need to zero %fs, otherwise we may get an
598          * exception between the new %fs descriptor being loaded and
599          * %fs being effectively cleared at __switch_to().
600          */
601         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
602 #ifdef CONFIG_X86_32
603                 lazy_load_gs(0);
604 #else
605                 loadsegment(fs, 0);
606 #endif
607         }
608
609         xen_mc_batch();
610
611         load_TLS_descriptor(t, cpu, 0);
612         load_TLS_descriptor(t, cpu, 1);
613         load_TLS_descriptor(t, cpu, 2);
614
615         xen_mc_issue(PARAVIRT_LAZY_CPU);
616 }
617
618 #ifdef CONFIG_X86_64
619 static void xen_load_gs_index(unsigned int idx)
620 {
621         if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
622                 BUG();
623 }
624 #endif
625
626 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
627                                 const void *ptr)
628 {
629         xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
630         u64 entry = *(u64 *)ptr;
631
632         trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
633
634         preempt_disable();
635
636         xen_mc_flush();
637         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
638                 BUG();
639
640         preempt_enable();
641 }
642
643 static int cvt_gate_to_trap(int vector, const gate_desc *val,
644                             struct trap_info *info)
645 {
646         unsigned long addr;
647
648         if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
649                 return 0;
650
651         info->vector = vector;
652
653         addr = gate_offset(*val);
654 #ifdef CONFIG_X86_64
655         /*
656          * Look for known traps using IST, and substitute them
657          * appropriately.  The debugger ones are the only ones we care
658          * about.  Xen will handle faults like double_fault,
659          * so we should never see them.  Warn if
660          * there's an unexpected IST-using fault handler.
661          */
662         if (addr == (unsigned long)debug)
663                 addr = (unsigned long)xen_debug;
664         else if (addr == (unsigned long)int3)
665                 addr = (unsigned long)xen_int3;
666         else if (addr == (unsigned long)stack_segment)
667                 addr = (unsigned long)xen_stack_segment;
668         else if (addr == (unsigned long)double_fault ||
669                  addr == (unsigned long)nmi) {
670                 /* Don't need to handle these */
671                 return 0;
672 #ifdef CONFIG_X86_MCE
673         } else if (addr == (unsigned long)machine_check) {
674                 /*
675                  * when xen hypervisor inject vMCE to guest,
676                  * use native mce handler to handle it
677                  */
678                 ;
679 #endif
680         } else {
681                 /* Some other trap using IST? */
682                 if (WARN_ON(val->ist != 0))
683                         return 0;
684         }
685 #endif  /* CONFIG_X86_64 */
686         info->address = addr;
687
688         info->cs = gate_segment(*val);
689         info->flags = val->dpl;
690         /* interrupt gates clear IF */
691         if (val->type == GATE_INTERRUPT)
692                 info->flags |= 1 << 2;
693
694         return 1;
695 }
696
697 /* Locations of each CPU's IDT */
698 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
699
700 /* Set an IDT entry.  If the entry is part of the current IDT, then
701    also update Xen. */
702 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
703 {
704         unsigned long p = (unsigned long)&dt[entrynum];
705         unsigned long start, end;
706
707         trace_xen_cpu_write_idt_entry(dt, entrynum, g);
708
709         preempt_disable();
710
711         start = __this_cpu_read(idt_desc.address);
712         end = start + __this_cpu_read(idt_desc.size) + 1;
713
714         xen_mc_flush();
715
716         native_write_idt_entry(dt, entrynum, g);
717
718         if (p >= start && (p + 8) <= end) {
719                 struct trap_info info[2];
720
721                 info[1].address = 0;
722
723                 if (cvt_gate_to_trap(entrynum, g, &info[0]))
724                         if (HYPERVISOR_set_trap_table(info))
725                                 BUG();
726         }
727
728         preempt_enable();
729 }
730
731 static void xen_convert_trap_info(const struct desc_ptr *desc,
732                                   struct trap_info *traps)
733 {
734         unsigned in, out, count;
735
736         count = (desc->size+1) / sizeof(gate_desc);
737         BUG_ON(count > 256);
738
739         for (in = out = 0; in < count; in++) {
740                 gate_desc *entry = (gate_desc*)(desc->address) + in;
741
742                 if (cvt_gate_to_trap(in, entry, &traps[out]))
743                         out++;
744         }
745         traps[out].address = 0;
746 }
747
748 void xen_copy_trap_info(struct trap_info *traps)
749 {
750         const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
751
752         xen_convert_trap_info(desc, traps);
753 }
754
755 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
756    hold a spinlock to protect the static traps[] array (static because
757    it avoids allocation, and saves stack space). */
758 static void xen_load_idt(const struct desc_ptr *desc)
759 {
760         static DEFINE_SPINLOCK(lock);
761         static struct trap_info traps[257];
762
763         trace_xen_cpu_load_idt(desc);
764
765         spin_lock(&lock);
766
767         __get_cpu_var(idt_desc) = *desc;
768
769         xen_convert_trap_info(desc, traps);
770
771         xen_mc_flush();
772         if (HYPERVISOR_set_trap_table(traps))
773                 BUG();
774
775         spin_unlock(&lock);
776 }
777
778 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
779    they're handled differently. */
780 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
781                                 const void *desc, int type)
782 {
783         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
784
785         preempt_disable();
786
787         switch (type) {
788         case DESC_LDT:
789         case DESC_TSS:
790                 /* ignore */
791                 break;
792
793         default: {
794                 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
795
796                 xen_mc_flush();
797                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
798                         BUG();
799         }
800
801         }
802
803         preempt_enable();
804 }
805
806 /*
807  * Version of write_gdt_entry for use at early boot-time needed to
808  * update an entry as simply as possible.
809  */
810 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
811                                             const void *desc, int type)
812 {
813         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
814
815         switch (type) {
816         case DESC_LDT:
817         case DESC_TSS:
818                 /* ignore */
819                 break;
820
821         default: {
822                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
823
824                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
825                         dt[entry] = *(struct desc_struct *)desc;
826         }
827
828         }
829 }
830
831 static void xen_load_sp0(struct tss_struct *tss,
832                          struct thread_struct *thread)
833 {
834         struct multicall_space mcs;
835
836         mcs = xen_mc_entry(0);
837         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
838         xen_mc_issue(PARAVIRT_LAZY_CPU);
839 }
840
841 static void xen_set_iopl_mask(unsigned mask)
842 {
843         struct physdev_set_iopl set_iopl;
844
845         /* Force the change at ring 0. */
846         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
847         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
848 }
849
850 static void xen_io_delay(void)
851 {
852 }
853
854 #ifdef CONFIG_X86_LOCAL_APIC
855 static unsigned long xen_set_apic_id(unsigned int x)
856 {
857         WARN_ON(1);
858         return x;
859 }
860 static unsigned int xen_get_apic_id(unsigned long x)
861 {
862         return ((x)>>24) & 0xFFu;
863 }
864 static u32 xen_apic_read(u32 reg)
865 {
866         struct xen_platform_op op = {
867                 .cmd = XENPF_get_cpuinfo,
868                 .interface_version = XENPF_INTERFACE_VERSION,
869                 .u.pcpu_info.xen_cpuid = 0,
870         };
871         int ret = 0;
872
873         /* Shouldn't need this as APIC is turned off for PV, and we only
874          * get called on the bootup processor. But just in case. */
875         if (!xen_initial_domain() || smp_processor_id())
876                 return 0;
877
878         if (reg == APIC_LVR)
879                 return 0x10;
880
881         if (reg != APIC_ID)
882                 return 0;
883
884         ret = HYPERVISOR_dom0_op(&op);
885         if (ret)
886                 return 0;
887
888         return op.u.pcpu_info.apic_id << 24;
889 }
890
891 static void xen_apic_write(u32 reg, u32 val)
892 {
893         /* Warn to see if there's any stray references */
894         WARN_ON(1);
895 }
896
897 static u64 xen_apic_icr_read(void)
898 {
899         return 0;
900 }
901
902 static void xen_apic_icr_write(u32 low, u32 id)
903 {
904         /* Warn to see if there's any stray references */
905         WARN_ON(1);
906 }
907
908 static void xen_apic_wait_icr_idle(void)
909 {
910         return;
911 }
912
913 static u32 xen_safe_apic_wait_icr_idle(void)
914 {
915         return 0;
916 }
917
918 static void set_xen_basic_apic_ops(void)
919 {
920         apic->read = xen_apic_read;
921         apic->write = xen_apic_write;
922         apic->icr_read = xen_apic_icr_read;
923         apic->icr_write = xen_apic_icr_write;
924         apic->wait_icr_idle = xen_apic_wait_icr_idle;
925         apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
926         apic->set_apic_id = xen_set_apic_id;
927         apic->get_apic_id = xen_get_apic_id;
928
929 #ifdef CONFIG_SMP
930         apic->send_IPI_allbutself = xen_send_IPI_allbutself;
931         apic->send_IPI_mask_allbutself = xen_send_IPI_mask_allbutself;
932         apic->send_IPI_mask = xen_send_IPI_mask;
933         apic->send_IPI_all = xen_send_IPI_all;
934         apic->send_IPI_self = xen_send_IPI_self;
935 #endif
936 }
937
938 #endif
939
940 static void xen_clts(void)
941 {
942         struct multicall_space mcs;
943
944         mcs = xen_mc_entry(0);
945
946         MULTI_fpu_taskswitch(mcs.mc, 0);
947
948         xen_mc_issue(PARAVIRT_LAZY_CPU);
949 }
950
951 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
952
953 static unsigned long xen_read_cr0(void)
954 {
955         unsigned long cr0 = this_cpu_read(xen_cr0_value);
956
957         if (unlikely(cr0 == 0)) {
958                 cr0 = native_read_cr0();
959                 this_cpu_write(xen_cr0_value, cr0);
960         }
961
962         return cr0;
963 }
964
965 static void xen_write_cr0(unsigned long cr0)
966 {
967         struct multicall_space mcs;
968
969         this_cpu_write(xen_cr0_value, cr0);
970
971         /* Only pay attention to cr0.TS; everything else is
972            ignored. */
973         mcs = xen_mc_entry(0);
974
975         MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
976
977         xen_mc_issue(PARAVIRT_LAZY_CPU);
978 }
979
980 static void xen_write_cr4(unsigned long cr4)
981 {
982         cr4 &= ~X86_CR4_PGE;
983         cr4 &= ~X86_CR4_PSE;
984
985         native_write_cr4(cr4);
986 }
987
988 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
989 {
990         int ret;
991
992         ret = 0;
993
994         switch (msr) {
995 #ifdef CONFIG_X86_64
996                 unsigned which;
997                 u64 base;
998
999         case MSR_FS_BASE:               which = SEGBASE_FS; goto set;
1000         case MSR_KERNEL_GS_BASE:        which = SEGBASE_GS_USER; goto set;
1001         case MSR_GS_BASE:               which = SEGBASE_GS_KERNEL; goto set;
1002
1003         set:
1004                 base = ((u64)high << 32) | low;
1005                 if (HYPERVISOR_set_segment_base(which, base) != 0)
1006                         ret = -EIO;
1007                 break;
1008 #endif
1009
1010         case MSR_STAR:
1011         case MSR_CSTAR:
1012         case MSR_LSTAR:
1013         case MSR_SYSCALL_MASK:
1014         case MSR_IA32_SYSENTER_CS:
1015         case MSR_IA32_SYSENTER_ESP:
1016         case MSR_IA32_SYSENTER_EIP:
1017                 /* Fast syscall setup is all done in hypercalls, so
1018                    these are all ignored.  Stub them out here to stop
1019                    Xen console noise. */
1020                 break;
1021
1022         case MSR_IA32_CR_PAT:
1023                 if (smp_processor_id() == 0)
1024                         xen_set_pat(((u64)high << 32) | low);
1025                 break;
1026
1027         default:
1028                 ret = native_write_msr_safe(msr, low, high);
1029         }
1030
1031         return ret;
1032 }
1033
1034 void xen_setup_shared_info(void)
1035 {
1036         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1037                 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1038                            xen_start_info->shared_info);
1039
1040                 HYPERVISOR_shared_info =
1041                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1042         } else
1043                 HYPERVISOR_shared_info =
1044                         (struct shared_info *)__va(xen_start_info->shared_info);
1045
1046 #ifndef CONFIG_SMP
1047         /* In UP this is as good a place as any to set up shared info */
1048         xen_setup_vcpu_info_placement();
1049 #endif
1050
1051         xen_setup_mfn_list_list();
1052 }
1053
1054 /* This is called once we have the cpu_possible_mask */
1055 void xen_setup_vcpu_info_placement(void)
1056 {
1057         int cpu;
1058
1059         for_each_possible_cpu(cpu)
1060                 xen_vcpu_setup(cpu);
1061
1062         /* xen_vcpu_setup managed to place the vcpu_info within the
1063            percpu area for all cpus, so make use of it */
1064         if (have_vcpu_info_placement) {
1065                 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1066                 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1067                 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1068                 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1069                 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1070         }
1071 }
1072
1073 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1074                           unsigned long addr, unsigned len)
1075 {
1076         char *start, *end, *reloc;
1077         unsigned ret;
1078
1079         start = end = reloc = NULL;
1080
1081 #define SITE(op, x)                                                     \
1082         case PARAVIRT_PATCH(op.x):                                      \
1083         if (have_vcpu_info_placement) {                                 \
1084                 start = (char *)xen_##x##_direct;                       \
1085                 end = xen_##x##_direct_end;                             \
1086                 reloc = xen_##x##_direct_reloc;                         \
1087         }                                                               \
1088         goto patch_site
1089
1090         switch (type) {
1091                 SITE(pv_irq_ops, irq_enable);
1092                 SITE(pv_irq_ops, irq_disable);
1093                 SITE(pv_irq_ops, save_fl);
1094                 SITE(pv_irq_ops, restore_fl);
1095 #undef SITE
1096
1097         patch_site:
1098                 if (start == NULL || (end-start) > len)
1099                         goto default_patch;
1100
1101                 ret = paravirt_patch_insns(insnbuf, len, start, end);
1102
1103                 /* Note: because reloc is assigned from something that
1104                    appears to be an array, gcc assumes it's non-null,
1105                    but doesn't know its relationship with start and
1106                    end. */
1107                 if (reloc > start && reloc < end) {
1108                         int reloc_off = reloc - start;
1109                         long *relocp = (long *)(insnbuf + reloc_off);
1110                         long delta = start - (char *)addr;
1111
1112                         *relocp += delta;
1113                 }
1114                 break;
1115
1116         default_patch:
1117         default:
1118                 ret = paravirt_patch_default(type, clobbers, insnbuf,
1119                                              addr, len);
1120                 break;
1121         }
1122
1123         return ret;
1124 }
1125
1126 static const struct pv_info xen_info __initconst = {
1127         .paravirt_enabled = 1,
1128         .shared_kernel_pmd = 0,
1129
1130 #ifdef CONFIG_X86_64
1131         .extra_user_64bit_cs = FLAT_USER_CS64,
1132 #endif
1133
1134         .name = "Xen",
1135 };
1136
1137 static const struct pv_init_ops xen_init_ops __initconst = {
1138         .patch = xen_patch,
1139 };
1140
1141 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1142         .cpuid = xen_cpuid,
1143
1144         .set_debugreg = xen_set_debugreg,
1145         .get_debugreg = xen_get_debugreg,
1146
1147         .clts = xen_clts,
1148
1149         .read_cr0 = xen_read_cr0,
1150         .write_cr0 = xen_write_cr0,
1151
1152         .read_cr4 = native_read_cr4,
1153         .read_cr4_safe = native_read_cr4_safe,
1154         .write_cr4 = xen_write_cr4,
1155
1156         .wbinvd = native_wbinvd,
1157
1158         .read_msr = native_read_msr_safe,
1159         .rdmsr_regs = native_rdmsr_safe_regs,
1160         .write_msr = xen_write_msr_safe,
1161         .wrmsr_regs = native_wrmsr_safe_regs,
1162
1163         .read_tsc = native_read_tsc,
1164         .read_pmc = native_read_pmc,
1165
1166         .iret = xen_iret,
1167         .irq_enable_sysexit = xen_sysexit,
1168 #ifdef CONFIG_X86_64
1169         .usergs_sysret32 = xen_sysret32,
1170         .usergs_sysret64 = xen_sysret64,
1171 #endif
1172
1173         .load_tr_desc = paravirt_nop,
1174         .set_ldt = xen_set_ldt,
1175         .load_gdt = xen_load_gdt,
1176         .load_idt = xen_load_idt,
1177         .load_tls = xen_load_tls,
1178 #ifdef CONFIG_X86_64
1179         .load_gs_index = xen_load_gs_index,
1180 #endif
1181
1182         .alloc_ldt = xen_alloc_ldt,
1183         .free_ldt = xen_free_ldt,
1184
1185         .store_gdt = native_store_gdt,
1186         .store_idt = native_store_idt,
1187         .store_tr = xen_store_tr,
1188
1189         .write_ldt_entry = xen_write_ldt_entry,
1190         .write_gdt_entry = xen_write_gdt_entry,
1191         .write_idt_entry = xen_write_idt_entry,
1192         .load_sp0 = xen_load_sp0,
1193
1194         .set_iopl_mask = xen_set_iopl_mask,
1195         .io_delay = xen_io_delay,
1196
1197         /* Xen takes care of %gs when switching to usermode for us */
1198         .swapgs = paravirt_nop,
1199
1200         .start_context_switch = paravirt_start_context_switch,
1201         .end_context_switch = xen_end_context_switch,
1202 };
1203
1204 static const struct pv_apic_ops xen_apic_ops __initconst = {
1205 #ifdef CONFIG_X86_LOCAL_APIC
1206         .startup_ipi_hook = paravirt_nop,
1207 #endif
1208 };
1209
1210 static void xen_reboot(int reason)
1211 {
1212         struct sched_shutdown r = { .reason = reason };
1213
1214         if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1215                 BUG();
1216 }
1217
1218 static void xen_restart(char *msg)
1219 {
1220         xen_reboot(SHUTDOWN_reboot);
1221 }
1222
1223 static void xen_emergency_restart(void)
1224 {
1225         xen_reboot(SHUTDOWN_reboot);
1226 }
1227
1228 static void xen_machine_halt(void)
1229 {
1230         xen_reboot(SHUTDOWN_poweroff);
1231 }
1232
1233 static void xen_machine_power_off(void)
1234 {
1235         if (pm_power_off)
1236                 pm_power_off();
1237         xen_reboot(SHUTDOWN_poweroff);
1238 }
1239
1240 static void xen_crash_shutdown(struct pt_regs *regs)
1241 {
1242         xen_reboot(SHUTDOWN_crash);
1243 }
1244
1245 static int
1246 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1247 {
1248         xen_reboot(SHUTDOWN_crash);
1249         return NOTIFY_DONE;
1250 }
1251
1252 static struct notifier_block xen_panic_block = {
1253         .notifier_call= xen_panic_event,
1254 };
1255
1256 int xen_panic_handler_init(void)
1257 {
1258         atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1259         return 0;
1260 }
1261
1262 static const struct machine_ops xen_machine_ops __initconst = {
1263         .restart = xen_restart,
1264         .halt = xen_machine_halt,
1265         .power_off = xen_machine_power_off,
1266         .shutdown = xen_machine_halt,
1267         .crash_shutdown = xen_crash_shutdown,
1268         .emergency_restart = xen_emergency_restart,
1269 };
1270
1271 /*
1272  * Set up the GDT and segment registers for -fstack-protector.  Until
1273  * we do this, we have to be careful not to call any stack-protected
1274  * function, which is most of the kernel.
1275  */
1276 static void __init xen_setup_stackprotector(void)
1277 {
1278         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1279         pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1280
1281         setup_stack_canary_segment(0);
1282         switch_to_new_gdt(0);
1283
1284         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1285         pv_cpu_ops.load_gdt = xen_load_gdt;
1286 }
1287
1288 /* First C function to be called on Xen boot */
1289 asmlinkage void __init xen_start_kernel(void)
1290 {
1291         struct physdev_set_iopl set_iopl;
1292         int rc;
1293         pgd_t *pgd;
1294
1295         if (!xen_start_info)
1296                 return;
1297
1298         xen_domain_type = XEN_PV_DOMAIN;
1299
1300         xen_setup_machphys_mapping();
1301
1302         /* Install Xen paravirt ops */
1303         pv_info = xen_info;
1304         pv_init_ops = xen_init_ops;
1305         pv_cpu_ops = xen_cpu_ops;
1306         pv_apic_ops = xen_apic_ops;
1307
1308         x86_init.resources.memory_setup = xen_memory_setup;
1309         x86_init.oem.arch_setup = xen_arch_setup;
1310         x86_init.oem.banner = xen_banner;
1311
1312         xen_init_time_ops();
1313
1314         /*
1315          * Set up some pagetable state before starting to set any ptes.
1316          */
1317
1318         xen_init_mmu_ops();
1319
1320         /* Prevent unwanted bits from being set in PTEs. */
1321         __supported_pte_mask &= ~_PAGE_GLOBAL;
1322 #if 0
1323         if (!xen_initial_domain())
1324 #endif
1325                 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1326
1327         __supported_pte_mask |= _PAGE_IOMAP;
1328
1329         /*
1330          * Prevent page tables from being allocated in highmem, even
1331          * if CONFIG_HIGHPTE is enabled.
1332          */
1333         __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1334
1335         /* Work out if we support NX */
1336         x86_configure_nx();
1337
1338         xen_setup_features();
1339
1340         /* Get mfn list */
1341         if (!xen_feature(XENFEAT_auto_translated_physmap))
1342                 xen_build_dynamic_phys_to_machine();
1343
1344         /*
1345          * Set up kernel GDT and segment registers, mainly so that
1346          * -fstack-protector code can be executed.
1347          */
1348         xen_setup_stackprotector();
1349
1350         xen_init_irq_ops();
1351         xen_init_cpuid_mask();
1352
1353 #ifdef CONFIG_X86_LOCAL_APIC
1354         /*
1355          * set up the basic apic ops.
1356          */
1357         set_xen_basic_apic_ops();
1358 #endif
1359
1360         if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1361                 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1362                 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1363         }
1364
1365         machine_ops = xen_machine_ops;
1366
1367         /*
1368          * The only reliable way to retain the initial address of the
1369          * percpu gdt_page is to remember it here, so we can go and
1370          * mark it RW later, when the initial percpu area is freed.
1371          */
1372         xen_initial_gdt = &per_cpu(gdt_page, 0);
1373
1374         xen_smp_init();
1375
1376 #ifdef CONFIG_ACPI_NUMA
1377         /*
1378          * The pages we from Xen are not related to machine pages, so
1379          * any NUMA information the kernel tries to get from ACPI will
1380          * be meaningless.  Prevent it from trying.
1381          */
1382         acpi_numa = -1;
1383 #endif
1384
1385         pgd = (pgd_t *)xen_start_info->pt_base;
1386
1387         /* Don't do the full vcpu_info placement stuff until we have a
1388            possible map and a non-dummy shared_info. */
1389         per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1390
1391         local_irq_disable();
1392         early_boot_irqs_disabled = true;
1393
1394         xen_raw_console_write("mapping kernel into physical memory\n");
1395         pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1396
1397         /* Allocate and initialize top and mid mfn levels for p2m structure */
1398         xen_build_mfn_list_list();
1399
1400         /* keep using Xen gdt for now; no urgent need to change it */
1401
1402 #ifdef CONFIG_X86_32
1403         pv_info.kernel_rpl = 1;
1404         if (xen_feature(XENFEAT_supervisor_mode_kernel))
1405                 pv_info.kernel_rpl = 0;
1406 #else
1407         pv_info.kernel_rpl = 0;
1408 #endif
1409         /* set the limit of our address space */
1410         xen_reserve_top();
1411
1412         /* We used to do this in xen_arch_setup, but that is too late on AMD
1413          * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1414          * which pokes 0xcf8 port.
1415          */
1416         set_iopl.iopl = 1;
1417         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1418         if (rc != 0)
1419                 xen_raw_printk("physdev_op failed %d\n", rc);
1420
1421 #ifdef CONFIG_X86_32
1422         /* set up basic CPUID stuff */
1423         cpu_detect(&new_cpu_data);
1424         new_cpu_data.hard_math = 1;
1425         new_cpu_data.wp_works_ok = 1;
1426         new_cpu_data.x86_capability[0] = cpuid_edx(1);
1427 #endif
1428
1429         /* Poke various useful things into boot_params */
1430         boot_params.hdr.type_of_loader = (9 << 4) | 0;
1431         boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1432                 ? __pa(xen_start_info->mod_start) : 0;
1433         boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1434         boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1435
1436         if (!xen_initial_domain()) {
1437                 add_preferred_console("xenboot", 0, NULL);
1438                 add_preferred_console("tty", 0, NULL);
1439                 add_preferred_console("hvc", 0, NULL);
1440                 if (pci_xen)
1441                         x86_init.pci.arch_init = pci_xen_init;
1442         } else {
1443                 const struct dom0_vga_console_info *info =
1444                         (void *)((char *)xen_start_info +
1445                                  xen_start_info->console.dom0.info_off);
1446
1447                 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1448                 xen_start_info->console.domU.mfn = 0;
1449                 xen_start_info->console.domU.evtchn = 0;
1450
1451                 xen_init_apic();
1452
1453                 /* Make sure ACS will be enabled */
1454                 pci_request_acs();
1455
1456                 xen_acpi_sleep_register();
1457         }
1458 #ifdef CONFIG_PCI
1459         /* PCI BIOS service won't work from a PV guest. */
1460         pci_probe &= ~PCI_PROBE_BIOS;
1461 #endif
1462         xen_raw_console_write("about to get started...\n");
1463
1464         xen_setup_runstate_info(0);
1465
1466         /* Start the world */
1467 #ifdef CONFIG_X86_32
1468         i386_start_kernel();
1469 #else
1470         x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1471 #endif
1472 }
1473
1474 static int init_hvm_pv_info(int *major, int *minor)
1475 {
1476         uint32_t eax, ebx, ecx, edx, pages, msr, base;
1477         u64 pfn;
1478
1479         base = xen_cpuid_base();
1480         cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1481
1482         *major = eax >> 16;
1483         *minor = eax & 0xffff;
1484         printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1485
1486         cpuid(base + 2, &pages, &msr, &ecx, &edx);
1487
1488         pfn = __pa(hypercall_page);
1489         wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1490
1491         xen_setup_features();
1492
1493         pv_info.name = "Xen HVM";
1494
1495         xen_domain_type = XEN_HVM_DOMAIN;
1496
1497         return 0;
1498 }
1499
1500 void __ref xen_hvm_init_shared_info(void)
1501 {
1502         int cpu;
1503         struct xen_add_to_physmap xatp;
1504         static struct shared_info *shared_info_page = 0;
1505
1506         if (!shared_info_page)
1507                 shared_info_page = (struct shared_info *)
1508                         extend_brk(PAGE_SIZE, PAGE_SIZE);
1509         xatp.domid = DOMID_SELF;
1510         xatp.idx = 0;
1511         xatp.space = XENMAPSPACE_shared_info;
1512         xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1513         if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1514                 BUG();
1515
1516         HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1517
1518         /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1519          * page, we use it in the event channel upcall and in some pvclock
1520          * related functions. We don't need the vcpu_info placement
1521          * optimizations because we don't use any pv_mmu or pv_irq op on
1522          * HVM.
1523          * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1524          * online but xen_hvm_init_shared_info is run at resume time too and
1525          * in that case multiple vcpus might be online. */
1526         for_each_online_cpu(cpu) {
1527                 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1528         }
1529 }
1530
1531 #ifdef CONFIG_XEN_PVHVM
1532 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1533                                     unsigned long action, void *hcpu)
1534 {
1535         int cpu = (long)hcpu;
1536         switch (action) {
1537         case CPU_UP_PREPARE:
1538                 xen_vcpu_setup(cpu);
1539                 if (xen_have_vector_callback)
1540                         xen_init_lock_cpu(cpu);
1541                 break;
1542         default:
1543                 break;
1544         }
1545         return NOTIFY_OK;
1546 }
1547
1548 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1549         .notifier_call  = xen_hvm_cpu_notify,
1550 };
1551
1552 static void __init xen_hvm_guest_init(void)
1553 {
1554         int r;
1555         int major, minor;
1556
1557         r = init_hvm_pv_info(&major, &minor);
1558         if (r < 0)
1559                 return;
1560
1561         xen_hvm_init_shared_info();
1562
1563         if (xen_feature(XENFEAT_hvm_callback_vector))
1564                 xen_have_vector_callback = 1;
1565         xen_hvm_smp_init();
1566         register_cpu_notifier(&xen_hvm_cpu_notifier);
1567         xen_unplug_emulated_devices();
1568         x86_init.irqs.intr_init = xen_init_IRQ;
1569         xen_hvm_init_time_ops();
1570         xen_hvm_init_mmu_ops();
1571 }
1572
1573 static bool __init xen_hvm_platform(void)
1574 {
1575         if (xen_pv_domain())
1576                 return false;
1577
1578         if (!xen_cpuid_base())
1579                 return false;
1580
1581         return true;
1582 }
1583
1584 bool xen_hvm_need_lapic(void)
1585 {
1586         if (xen_pv_domain())
1587                 return false;
1588         if (!xen_hvm_domain())
1589                 return false;
1590         if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1591                 return false;
1592         return true;
1593 }
1594 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1595
1596 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1597         .name                   = "Xen HVM",
1598         .detect                 = xen_hvm_platform,
1599         .init_platform          = xen_hvm_guest_init,
1600 };
1601 EXPORT_SYMBOL(x86_hyper_xen_hvm);
1602 #endif