KVM: x86: Add KVM_GET/SET_VCPU_EVENTS
[linux-2.6.git] / arch / x86 / kvm / x86.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *   Amit Shah    <amit.shah@qumranet.com>
14  *   Ben-Ami Yassour <benami@il.ibm.com>
15  *
16  * This work is licensed under the terms of the GNU GPL, version 2.  See
17  * the COPYING file in the top-level directory.
18  *
19  */
20
21 #include <linux/kvm_host.h>
22 #include "irq.h"
23 #include "mmu.h"
24 #include "i8254.h"
25 #include "tss.h"
26 #include "kvm_cache_regs.h"
27 #include "x86.h"
28
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
32 #include <linux/fs.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
40 #include <linux/user-return-notifier.h>
41 #include <trace/events/kvm.h>
42 #undef TRACE_INCLUDE_FILE
43 #define CREATE_TRACE_POINTS
44 #include "trace.h"
45
46 #include <asm/uaccess.h>
47 #include <asm/msr.h>
48 #include <asm/desc.h>
49 #include <asm/mtrr.h>
50 #include <asm/mce.h>
51
52 #define MAX_IO_MSRS 256
53 #define CR0_RESERVED_BITS                                               \
54         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
55                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
56                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
57 #define CR4_RESERVED_BITS                                               \
58         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
59                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
60                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
61                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
62
63 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
64
65 #define KVM_MAX_MCE_BANKS 32
66 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
67
68 /* EFER defaults:
69  * - enable syscall per default because its emulated by KVM
70  * - enable LME and LMA per default on 64 bit KVM
71  */
72 #ifdef CONFIG_X86_64
73 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
74 #else
75 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
76 #endif
77
78 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
79 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
80
81 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
82 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
83                                     struct kvm_cpuid_entry2 __user *entries);
84
85 struct kvm_x86_ops *kvm_x86_ops;
86 EXPORT_SYMBOL_GPL(kvm_x86_ops);
87
88 int ignore_msrs = 0;
89 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
90
91 #define KVM_NR_SHARED_MSRS 16
92
93 struct kvm_shared_msrs_global {
94         int nr;
95         struct kvm_shared_msr {
96                 u32 msr;
97                 u64 value;
98         } msrs[KVM_NR_SHARED_MSRS];
99 };
100
101 struct kvm_shared_msrs {
102         struct user_return_notifier urn;
103         bool registered;
104         u64 current_value[KVM_NR_SHARED_MSRS];
105 };
106
107 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
108 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
109
110 struct kvm_stats_debugfs_item debugfs_entries[] = {
111         { "pf_fixed", VCPU_STAT(pf_fixed) },
112         { "pf_guest", VCPU_STAT(pf_guest) },
113         { "tlb_flush", VCPU_STAT(tlb_flush) },
114         { "invlpg", VCPU_STAT(invlpg) },
115         { "exits", VCPU_STAT(exits) },
116         { "io_exits", VCPU_STAT(io_exits) },
117         { "mmio_exits", VCPU_STAT(mmio_exits) },
118         { "signal_exits", VCPU_STAT(signal_exits) },
119         { "irq_window", VCPU_STAT(irq_window_exits) },
120         { "nmi_window", VCPU_STAT(nmi_window_exits) },
121         { "halt_exits", VCPU_STAT(halt_exits) },
122         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
123         { "hypercalls", VCPU_STAT(hypercalls) },
124         { "request_irq", VCPU_STAT(request_irq_exits) },
125         { "irq_exits", VCPU_STAT(irq_exits) },
126         { "host_state_reload", VCPU_STAT(host_state_reload) },
127         { "efer_reload", VCPU_STAT(efer_reload) },
128         { "fpu_reload", VCPU_STAT(fpu_reload) },
129         { "insn_emulation", VCPU_STAT(insn_emulation) },
130         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
131         { "irq_injections", VCPU_STAT(irq_injections) },
132         { "nmi_injections", VCPU_STAT(nmi_injections) },
133         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
134         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
135         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
136         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
137         { "mmu_flooded", VM_STAT(mmu_flooded) },
138         { "mmu_recycled", VM_STAT(mmu_recycled) },
139         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
140         { "mmu_unsync", VM_STAT(mmu_unsync) },
141         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
142         { "largepages", VM_STAT(lpages) },
143         { NULL }
144 };
145
146 static void kvm_on_user_return(struct user_return_notifier *urn)
147 {
148         unsigned slot;
149         struct kvm_shared_msr *global;
150         struct kvm_shared_msrs *locals
151                 = container_of(urn, struct kvm_shared_msrs, urn);
152
153         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
154                 global = &shared_msrs_global.msrs[slot];
155                 if (global->value != locals->current_value[slot]) {
156                         wrmsrl(global->msr, global->value);
157                         locals->current_value[slot] = global->value;
158                 }
159         }
160         locals->registered = false;
161         user_return_notifier_unregister(urn);
162 }
163
164 void kvm_define_shared_msr(unsigned slot, u32 msr)
165 {
166         int cpu;
167         u64 value;
168
169         if (slot >= shared_msrs_global.nr)
170                 shared_msrs_global.nr = slot + 1;
171         shared_msrs_global.msrs[slot].msr = msr;
172         rdmsrl_safe(msr, &value);
173         shared_msrs_global.msrs[slot].value = value;
174         for_each_online_cpu(cpu)
175                 per_cpu(shared_msrs, cpu).current_value[slot] = value;
176 }
177 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
178
179 static void kvm_shared_msr_cpu_online(void)
180 {
181         unsigned i;
182         struct kvm_shared_msrs *locals = &__get_cpu_var(shared_msrs);
183
184         for (i = 0; i < shared_msrs_global.nr; ++i)
185                 locals->current_value[i] = shared_msrs_global.msrs[i].value;
186 }
187
188 void kvm_set_shared_msr(unsigned slot, u64 value)
189 {
190         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
191
192         if (value == smsr->current_value[slot])
193                 return;
194         smsr->current_value[slot] = value;
195         wrmsrl(shared_msrs_global.msrs[slot].msr, value);
196         if (!smsr->registered) {
197                 smsr->urn.on_user_return = kvm_on_user_return;
198                 user_return_notifier_register(&smsr->urn);
199                 smsr->registered = true;
200         }
201 }
202 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
203
204 unsigned long segment_base(u16 selector)
205 {
206         struct descriptor_table gdt;
207         struct desc_struct *d;
208         unsigned long table_base;
209         unsigned long v;
210
211         if (selector == 0)
212                 return 0;
213
214         kvm_get_gdt(&gdt);
215         table_base = gdt.base;
216
217         if (selector & 4) {           /* from ldt */
218                 u16 ldt_selector = kvm_read_ldt();
219
220                 table_base = segment_base(ldt_selector);
221         }
222         d = (struct desc_struct *)(table_base + (selector & ~7));
223         v = get_desc_base(d);
224 #ifdef CONFIG_X86_64
225         if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
226                 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
227 #endif
228         return v;
229 }
230 EXPORT_SYMBOL_GPL(segment_base);
231
232 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
233 {
234         if (irqchip_in_kernel(vcpu->kvm))
235                 return vcpu->arch.apic_base;
236         else
237                 return vcpu->arch.apic_base;
238 }
239 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
240
241 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
242 {
243         /* TODO: reserve bits check */
244         if (irqchip_in_kernel(vcpu->kvm))
245                 kvm_lapic_set_base(vcpu, data);
246         else
247                 vcpu->arch.apic_base = data;
248 }
249 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
250
251 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
252 {
253         WARN_ON(vcpu->arch.exception.pending);
254         vcpu->arch.exception.pending = true;
255         vcpu->arch.exception.has_error_code = false;
256         vcpu->arch.exception.nr = nr;
257 }
258 EXPORT_SYMBOL_GPL(kvm_queue_exception);
259
260 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
261                            u32 error_code)
262 {
263         ++vcpu->stat.pf_guest;
264
265         if (vcpu->arch.exception.pending) {
266                 switch(vcpu->arch.exception.nr) {
267                 case DF_VECTOR:
268                         /* triple fault -> shutdown */
269                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
270                         return;
271                 case PF_VECTOR:
272                         vcpu->arch.exception.nr = DF_VECTOR;
273                         vcpu->arch.exception.error_code = 0;
274                         return;
275                 default:
276                         /* replace previous exception with a new one in a hope
277                            that instruction re-execution will regenerate lost
278                            exception */
279                         vcpu->arch.exception.pending = false;
280                         break;
281                 }
282         }
283         vcpu->arch.cr2 = addr;
284         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
285 }
286
287 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
288 {
289         vcpu->arch.nmi_pending = 1;
290 }
291 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
292
293 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
294 {
295         WARN_ON(vcpu->arch.exception.pending);
296         vcpu->arch.exception.pending = true;
297         vcpu->arch.exception.has_error_code = true;
298         vcpu->arch.exception.nr = nr;
299         vcpu->arch.exception.error_code = error_code;
300 }
301 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
302
303 /*
304  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
305  * a #GP and return false.
306  */
307 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
308 {
309         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
310                 return true;
311         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
312         return false;
313 }
314 EXPORT_SYMBOL_GPL(kvm_require_cpl);
315
316 /*
317  * Load the pae pdptrs.  Return true is they are all valid.
318  */
319 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
320 {
321         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
322         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
323         int i;
324         int ret;
325         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
326
327         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
328                                   offset * sizeof(u64), sizeof(pdpte));
329         if (ret < 0) {
330                 ret = 0;
331                 goto out;
332         }
333         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
334                 if (is_present_gpte(pdpte[i]) &&
335                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
336                         ret = 0;
337                         goto out;
338                 }
339         }
340         ret = 1;
341
342         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
343         __set_bit(VCPU_EXREG_PDPTR,
344                   (unsigned long *)&vcpu->arch.regs_avail);
345         __set_bit(VCPU_EXREG_PDPTR,
346                   (unsigned long *)&vcpu->arch.regs_dirty);
347 out:
348
349         return ret;
350 }
351 EXPORT_SYMBOL_GPL(load_pdptrs);
352
353 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
354 {
355         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
356         bool changed = true;
357         int r;
358
359         if (is_long_mode(vcpu) || !is_pae(vcpu))
360                 return false;
361
362         if (!test_bit(VCPU_EXREG_PDPTR,
363                       (unsigned long *)&vcpu->arch.regs_avail))
364                 return true;
365
366         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
367         if (r < 0)
368                 goto out;
369         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
370 out:
371
372         return changed;
373 }
374
375 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
376 {
377         if (cr0 & CR0_RESERVED_BITS) {
378                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
379                        cr0, vcpu->arch.cr0);
380                 kvm_inject_gp(vcpu, 0);
381                 return;
382         }
383
384         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
385                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
386                 kvm_inject_gp(vcpu, 0);
387                 return;
388         }
389
390         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
391                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
392                        "and a clear PE flag\n");
393                 kvm_inject_gp(vcpu, 0);
394                 return;
395         }
396
397         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
398 #ifdef CONFIG_X86_64
399                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
400                         int cs_db, cs_l;
401
402                         if (!is_pae(vcpu)) {
403                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
404                                        "in long mode while PAE is disabled\n");
405                                 kvm_inject_gp(vcpu, 0);
406                                 return;
407                         }
408                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
409                         if (cs_l) {
410                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
411                                        "in long mode while CS.L == 1\n");
412                                 kvm_inject_gp(vcpu, 0);
413                                 return;
414
415                         }
416                 } else
417 #endif
418                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
419                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
420                                "reserved bits\n");
421                         kvm_inject_gp(vcpu, 0);
422                         return;
423                 }
424
425         }
426
427         kvm_x86_ops->set_cr0(vcpu, cr0);
428         vcpu->arch.cr0 = cr0;
429
430         kvm_mmu_reset_context(vcpu);
431         return;
432 }
433 EXPORT_SYMBOL_GPL(kvm_set_cr0);
434
435 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
436 {
437         kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
438 }
439 EXPORT_SYMBOL_GPL(kvm_lmsw);
440
441 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
442 {
443         unsigned long old_cr4 = vcpu->arch.cr4;
444         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
445
446         if (cr4 & CR4_RESERVED_BITS) {
447                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
448                 kvm_inject_gp(vcpu, 0);
449                 return;
450         }
451
452         if (is_long_mode(vcpu)) {
453                 if (!(cr4 & X86_CR4_PAE)) {
454                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
455                                "in long mode\n");
456                         kvm_inject_gp(vcpu, 0);
457                         return;
458                 }
459         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
460                    && ((cr4 ^ old_cr4) & pdptr_bits)
461                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
462                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
463                 kvm_inject_gp(vcpu, 0);
464                 return;
465         }
466
467         if (cr4 & X86_CR4_VMXE) {
468                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
469                 kvm_inject_gp(vcpu, 0);
470                 return;
471         }
472         kvm_x86_ops->set_cr4(vcpu, cr4);
473         vcpu->arch.cr4 = cr4;
474         vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
475         kvm_mmu_reset_context(vcpu);
476 }
477 EXPORT_SYMBOL_GPL(kvm_set_cr4);
478
479 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
480 {
481         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
482                 kvm_mmu_sync_roots(vcpu);
483                 kvm_mmu_flush_tlb(vcpu);
484                 return;
485         }
486
487         if (is_long_mode(vcpu)) {
488                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
489                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
490                         kvm_inject_gp(vcpu, 0);
491                         return;
492                 }
493         } else {
494                 if (is_pae(vcpu)) {
495                         if (cr3 & CR3_PAE_RESERVED_BITS) {
496                                 printk(KERN_DEBUG
497                                        "set_cr3: #GP, reserved bits\n");
498                                 kvm_inject_gp(vcpu, 0);
499                                 return;
500                         }
501                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
502                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
503                                        "reserved bits\n");
504                                 kvm_inject_gp(vcpu, 0);
505                                 return;
506                         }
507                 }
508                 /*
509                  * We don't check reserved bits in nonpae mode, because
510                  * this isn't enforced, and VMware depends on this.
511                  */
512         }
513
514         /*
515          * Does the new cr3 value map to physical memory? (Note, we
516          * catch an invalid cr3 even in real-mode, because it would
517          * cause trouble later on when we turn on paging anyway.)
518          *
519          * A real CPU would silently accept an invalid cr3 and would
520          * attempt to use it - with largely undefined (and often hard
521          * to debug) behavior on the guest side.
522          */
523         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
524                 kvm_inject_gp(vcpu, 0);
525         else {
526                 vcpu->arch.cr3 = cr3;
527                 vcpu->arch.mmu.new_cr3(vcpu);
528         }
529 }
530 EXPORT_SYMBOL_GPL(kvm_set_cr3);
531
532 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
533 {
534         if (cr8 & CR8_RESERVED_BITS) {
535                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
536                 kvm_inject_gp(vcpu, 0);
537                 return;
538         }
539         if (irqchip_in_kernel(vcpu->kvm))
540                 kvm_lapic_set_tpr(vcpu, cr8);
541         else
542                 vcpu->arch.cr8 = cr8;
543 }
544 EXPORT_SYMBOL_GPL(kvm_set_cr8);
545
546 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
547 {
548         if (irqchip_in_kernel(vcpu->kvm))
549                 return kvm_lapic_get_cr8(vcpu);
550         else
551                 return vcpu->arch.cr8;
552 }
553 EXPORT_SYMBOL_GPL(kvm_get_cr8);
554
555 static inline u32 bit(int bitno)
556 {
557         return 1 << (bitno & 31);
558 }
559
560 /*
561  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
562  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
563  *
564  * This list is modified at module load time to reflect the
565  * capabilities of the host cpu. This capabilities test skips MSRs that are
566  * kvm-specific. Those are put in the beginning of the list.
567  */
568
569 #define KVM_SAVE_MSRS_BEGIN     2
570 static u32 msrs_to_save[] = {
571         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
572         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
573         MSR_K6_STAR,
574 #ifdef CONFIG_X86_64
575         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
576 #endif
577         MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
578 };
579
580 static unsigned num_msrs_to_save;
581
582 static u32 emulated_msrs[] = {
583         MSR_IA32_MISC_ENABLE,
584 };
585
586 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
587 {
588         if (efer & efer_reserved_bits) {
589                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
590                        efer);
591                 kvm_inject_gp(vcpu, 0);
592                 return;
593         }
594
595         if (is_paging(vcpu)
596             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
597                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
598                 kvm_inject_gp(vcpu, 0);
599                 return;
600         }
601
602         if (efer & EFER_FFXSR) {
603                 struct kvm_cpuid_entry2 *feat;
604
605                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
606                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
607                         printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
608                         kvm_inject_gp(vcpu, 0);
609                         return;
610                 }
611         }
612
613         if (efer & EFER_SVME) {
614                 struct kvm_cpuid_entry2 *feat;
615
616                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
617                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
618                         printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
619                         kvm_inject_gp(vcpu, 0);
620                         return;
621                 }
622         }
623
624         kvm_x86_ops->set_efer(vcpu, efer);
625
626         efer &= ~EFER_LMA;
627         efer |= vcpu->arch.shadow_efer & EFER_LMA;
628
629         vcpu->arch.shadow_efer = efer;
630
631         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
632         kvm_mmu_reset_context(vcpu);
633 }
634
635 void kvm_enable_efer_bits(u64 mask)
636 {
637        efer_reserved_bits &= ~mask;
638 }
639 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
640
641
642 /*
643  * Writes msr value into into the appropriate "register".
644  * Returns 0 on success, non-0 otherwise.
645  * Assumes vcpu_load() was already called.
646  */
647 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
648 {
649         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
650 }
651
652 /*
653  * Adapt set_msr() to msr_io()'s calling convention
654  */
655 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
656 {
657         return kvm_set_msr(vcpu, index, *data);
658 }
659
660 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
661 {
662         static int version;
663         struct pvclock_wall_clock wc;
664         struct timespec now, sys, boot;
665
666         if (!wall_clock)
667                 return;
668
669         version++;
670
671         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
672
673         /*
674          * The guest calculates current wall clock time by adding
675          * system time (updated by kvm_write_guest_time below) to the
676          * wall clock specified here.  guest system time equals host
677          * system time for us, thus we must fill in host boot time here.
678          */
679         now = current_kernel_time();
680         ktime_get_ts(&sys);
681         boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
682
683         wc.sec = boot.tv_sec;
684         wc.nsec = boot.tv_nsec;
685         wc.version = version;
686
687         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
688
689         version++;
690         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
691 }
692
693 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
694 {
695         uint32_t quotient, remainder;
696
697         /* Don't try to replace with do_div(), this one calculates
698          * "(dividend << 32) / divisor" */
699         __asm__ ( "divl %4"
700                   : "=a" (quotient), "=d" (remainder)
701                   : "0" (0), "1" (dividend), "r" (divisor) );
702         return quotient;
703 }
704
705 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
706 {
707         uint64_t nsecs = 1000000000LL;
708         int32_t  shift = 0;
709         uint64_t tps64;
710         uint32_t tps32;
711
712         tps64 = tsc_khz * 1000LL;
713         while (tps64 > nsecs*2) {
714                 tps64 >>= 1;
715                 shift--;
716         }
717
718         tps32 = (uint32_t)tps64;
719         while (tps32 <= (uint32_t)nsecs) {
720                 tps32 <<= 1;
721                 shift++;
722         }
723
724         hv_clock->tsc_shift = shift;
725         hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
726
727         pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
728                  __func__, tsc_khz, hv_clock->tsc_shift,
729                  hv_clock->tsc_to_system_mul);
730 }
731
732 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
733
734 static void kvm_write_guest_time(struct kvm_vcpu *v)
735 {
736         struct timespec ts;
737         unsigned long flags;
738         struct kvm_vcpu_arch *vcpu = &v->arch;
739         void *shared_kaddr;
740         unsigned long this_tsc_khz;
741
742         if ((!vcpu->time_page))
743                 return;
744
745         this_tsc_khz = get_cpu_var(cpu_tsc_khz);
746         if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
747                 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
748                 vcpu->hv_clock_tsc_khz = this_tsc_khz;
749         }
750         put_cpu_var(cpu_tsc_khz);
751
752         /* Keep irq disabled to prevent changes to the clock */
753         local_irq_save(flags);
754         kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
755         ktime_get_ts(&ts);
756         local_irq_restore(flags);
757
758         /* With all the info we got, fill in the values */
759
760         vcpu->hv_clock.system_time = ts.tv_nsec +
761                                      (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
762
763         /*
764          * The interface expects us to write an even number signaling that the
765          * update is finished. Since the guest won't see the intermediate
766          * state, we just increase by 2 at the end.
767          */
768         vcpu->hv_clock.version += 2;
769
770         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
771
772         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
773                sizeof(vcpu->hv_clock));
774
775         kunmap_atomic(shared_kaddr, KM_USER0);
776
777         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
778 }
779
780 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
781 {
782         struct kvm_vcpu_arch *vcpu = &v->arch;
783
784         if (!vcpu->time_page)
785                 return 0;
786         set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
787         return 1;
788 }
789
790 static bool msr_mtrr_valid(unsigned msr)
791 {
792         switch (msr) {
793         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
794         case MSR_MTRRfix64K_00000:
795         case MSR_MTRRfix16K_80000:
796         case MSR_MTRRfix16K_A0000:
797         case MSR_MTRRfix4K_C0000:
798         case MSR_MTRRfix4K_C8000:
799         case MSR_MTRRfix4K_D0000:
800         case MSR_MTRRfix4K_D8000:
801         case MSR_MTRRfix4K_E0000:
802         case MSR_MTRRfix4K_E8000:
803         case MSR_MTRRfix4K_F0000:
804         case MSR_MTRRfix4K_F8000:
805         case MSR_MTRRdefType:
806         case MSR_IA32_CR_PAT:
807                 return true;
808         case 0x2f8:
809                 return true;
810         }
811         return false;
812 }
813
814 static bool valid_pat_type(unsigned t)
815 {
816         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
817 }
818
819 static bool valid_mtrr_type(unsigned t)
820 {
821         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
822 }
823
824 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
825 {
826         int i;
827
828         if (!msr_mtrr_valid(msr))
829                 return false;
830
831         if (msr == MSR_IA32_CR_PAT) {
832                 for (i = 0; i < 8; i++)
833                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
834                                 return false;
835                 return true;
836         } else if (msr == MSR_MTRRdefType) {
837                 if (data & ~0xcff)
838                         return false;
839                 return valid_mtrr_type(data & 0xff);
840         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
841                 for (i = 0; i < 8 ; i++)
842                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
843                                 return false;
844                 return true;
845         }
846
847         /* variable MTRRs */
848         return valid_mtrr_type(data & 0xff);
849 }
850
851 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
852 {
853         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
854
855         if (!mtrr_valid(vcpu, msr, data))
856                 return 1;
857
858         if (msr == MSR_MTRRdefType) {
859                 vcpu->arch.mtrr_state.def_type = data;
860                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
861         } else if (msr == MSR_MTRRfix64K_00000)
862                 p[0] = data;
863         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
864                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
865         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
866                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
867         else if (msr == MSR_IA32_CR_PAT)
868                 vcpu->arch.pat = data;
869         else {  /* Variable MTRRs */
870                 int idx, is_mtrr_mask;
871                 u64 *pt;
872
873                 idx = (msr - 0x200) / 2;
874                 is_mtrr_mask = msr - 0x200 - 2 * idx;
875                 if (!is_mtrr_mask)
876                         pt =
877                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
878                 else
879                         pt =
880                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
881                 *pt = data;
882         }
883
884         kvm_mmu_reset_context(vcpu);
885         return 0;
886 }
887
888 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
889 {
890         u64 mcg_cap = vcpu->arch.mcg_cap;
891         unsigned bank_num = mcg_cap & 0xff;
892
893         switch (msr) {
894         case MSR_IA32_MCG_STATUS:
895                 vcpu->arch.mcg_status = data;
896                 break;
897         case MSR_IA32_MCG_CTL:
898                 if (!(mcg_cap & MCG_CTL_P))
899                         return 1;
900                 if (data != 0 && data != ~(u64)0)
901                         return -1;
902                 vcpu->arch.mcg_ctl = data;
903                 break;
904         default:
905                 if (msr >= MSR_IA32_MC0_CTL &&
906                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
907                         u32 offset = msr - MSR_IA32_MC0_CTL;
908                         /* only 0 or all 1s can be written to IA32_MCi_CTL */
909                         if ((offset & 0x3) == 0 &&
910                             data != 0 && data != ~(u64)0)
911                                 return -1;
912                         vcpu->arch.mce_banks[offset] = data;
913                         break;
914                 }
915                 return 1;
916         }
917         return 0;
918 }
919
920 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
921 {
922         struct kvm *kvm = vcpu->kvm;
923         int lm = is_long_mode(vcpu);
924         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
925                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
926         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
927                 : kvm->arch.xen_hvm_config.blob_size_32;
928         u32 page_num = data & ~PAGE_MASK;
929         u64 page_addr = data & PAGE_MASK;
930         u8 *page;
931         int r;
932
933         r = -E2BIG;
934         if (page_num >= blob_size)
935                 goto out;
936         r = -ENOMEM;
937         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
938         if (!page)
939                 goto out;
940         r = -EFAULT;
941         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
942                 goto out_free;
943         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
944                 goto out_free;
945         r = 0;
946 out_free:
947         kfree(page);
948 out:
949         return r;
950 }
951
952 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
953 {
954         switch (msr) {
955         case MSR_EFER:
956                 set_efer(vcpu, data);
957                 break;
958         case MSR_K7_HWCR:
959                 data &= ~(u64)0x40;     /* ignore flush filter disable */
960                 if (data != 0) {
961                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
962                                 data);
963                         return 1;
964                 }
965                 break;
966         case MSR_FAM10H_MMIO_CONF_BASE:
967                 if (data != 0) {
968                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
969                                 "0x%llx\n", data);
970                         return 1;
971                 }
972                 break;
973         case MSR_AMD64_NB_CFG:
974                 break;
975         case MSR_IA32_DEBUGCTLMSR:
976                 if (!data) {
977                         /* We support the non-activated case already */
978                         break;
979                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
980                         /* Values other than LBR and BTF are vendor-specific,
981                            thus reserved and should throw a #GP */
982                         return 1;
983                 }
984                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
985                         __func__, data);
986                 break;
987         case MSR_IA32_UCODE_REV:
988         case MSR_IA32_UCODE_WRITE:
989         case MSR_VM_HSAVE_PA:
990         case MSR_AMD64_PATCH_LOADER:
991                 break;
992         case 0x200 ... 0x2ff:
993                 return set_msr_mtrr(vcpu, msr, data);
994         case MSR_IA32_APICBASE:
995                 kvm_set_apic_base(vcpu, data);
996                 break;
997         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
998                 return kvm_x2apic_msr_write(vcpu, msr, data);
999         case MSR_IA32_MISC_ENABLE:
1000                 vcpu->arch.ia32_misc_enable_msr = data;
1001                 break;
1002         case MSR_KVM_WALL_CLOCK:
1003                 vcpu->kvm->arch.wall_clock = data;
1004                 kvm_write_wall_clock(vcpu->kvm, data);
1005                 break;
1006         case MSR_KVM_SYSTEM_TIME: {
1007                 if (vcpu->arch.time_page) {
1008                         kvm_release_page_dirty(vcpu->arch.time_page);
1009                         vcpu->arch.time_page = NULL;
1010                 }
1011
1012                 vcpu->arch.time = data;
1013
1014                 /* we verify if the enable bit is set... */
1015                 if (!(data & 1))
1016                         break;
1017
1018                 /* ...but clean it before doing the actual write */
1019                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1020
1021                 vcpu->arch.time_page =
1022                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1023
1024                 if (is_error_page(vcpu->arch.time_page)) {
1025                         kvm_release_page_clean(vcpu->arch.time_page);
1026                         vcpu->arch.time_page = NULL;
1027                 }
1028
1029                 kvm_request_guest_time_update(vcpu);
1030                 break;
1031         }
1032         case MSR_IA32_MCG_CTL:
1033         case MSR_IA32_MCG_STATUS:
1034         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1035                 return set_msr_mce(vcpu, msr, data);
1036
1037         /* Performance counters are not protected by a CPUID bit,
1038          * so we should check all of them in the generic path for the sake of
1039          * cross vendor migration.
1040          * Writing a zero into the event select MSRs disables them,
1041          * which we perfectly emulate ;-). Any other value should be at least
1042          * reported, some guests depend on them.
1043          */
1044         case MSR_P6_EVNTSEL0:
1045         case MSR_P6_EVNTSEL1:
1046         case MSR_K7_EVNTSEL0:
1047         case MSR_K7_EVNTSEL1:
1048         case MSR_K7_EVNTSEL2:
1049         case MSR_K7_EVNTSEL3:
1050                 if (data != 0)
1051                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1052                                 "0x%x data 0x%llx\n", msr, data);
1053                 break;
1054         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1055          * so we ignore writes to make it happy.
1056          */
1057         case MSR_P6_PERFCTR0:
1058         case MSR_P6_PERFCTR1:
1059         case MSR_K7_PERFCTR0:
1060         case MSR_K7_PERFCTR1:
1061         case MSR_K7_PERFCTR2:
1062         case MSR_K7_PERFCTR3:
1063                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1064                         "0x%x data 0x%llx\n", msr, data);
1065                 break;
1066         default:
1067                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1068                         return xen_hvm_config(vcpu, data);
1069                 if (!ignore_msrs) {
1070                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1071                                 msr, data);
1072                         return 1;
1073                 } else {
1074                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1075                                 msr, data);
1076                         break;
1077                 }
1078         }
1079         return 0;
1080 }
1081 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1082
1083
1084 /*
1085  * Reads an msr value (of 'msr_index') into 'pdata'.
1086  * Returns 0 on success, non-0 otherwise.
1087  * Assumes vcpu_load() was already called.
1088  */
1089 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1090 {
1091         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1092 }
1093
1094 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1095 {
1096         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1097
1098         if (!msr_mtrr_valid(msr))
1099                 return 1;
1100
1101         if (msr == MSR_MTRRdefType)
1102                 *pdata = vcpu->arch.mtrr_state.def_type +
1103                          (vcpu->arch.mtrr_state.enabled << 10);
1104         else if (msr == MSR_MTRRfix64K_00000)
1105                 *pdata = p[0];
1106         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1107                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1108         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1109                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1110         else if (msr == MSR_IA32_CR_PAT)
1111                 *pdata = vcpu->arch.pat;
1112         else {  /* Variable MTRRs */
1113                 int idx, is_mtrr_mask;
1114                 u64 *pt;
1115
1116                 idx = (msr - 0x200) / 2;
1117                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1118                 if (!is_mtrr_mask)
1119                         pt =
1120                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1121                 else
1122                         pt =
1123                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1124                 *pdata = *pt;
1125         }
1126
1127         return 0;
1128 }
1129
1130 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1131 {
1132         u64 data;
1133         u64 mcg_cap = vcpu->arch.mcg_cap;
1134         unsigned bank_num = mcg_cap & 0xff;
1135
1136         switch (msr) {
1137         case MSR_IA32_P5_MC_ADDR:
1138         case MSR_IA32_P5_MC_TYPE:
1139                 data = 0;
1140                 break;
1141         case MSR_IA32_MCG_CAP:
1142                 data = vcpu->arch.mcg_cap;
1143                 break;
1144         case MSR_IA32_MCG_CTL:
1145                 if (!(mcg_cap & MCG_CTL_P))
1146                         return 1;
1147                 data = vcpu->arch.mcg_ctl;
1148                 break;
1149         case MSR_IA32_MCG_STATUS:
1150                 data = vcpu->arch.mcg_status;
1151                 break;
1152         default:
1153                 if (msr >= MSR_IA32_MC0_CTL &&
1154                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1155                         u32 offset = msr - MSR_IA32_MC0_CTL;
1156                         data = vcpu->arch.mce_banks[offset];
1157                         break;
1158                 }
1159                 return 1;
1160         }
1161         *pdata = data;
1162         return 0;
1163 }
1164
1165 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1166 {
1167         u64 data;
1168
1169         switch (msr) {
1170         case MSR_IA32_PLATFORM_ID:
1171         case MSR_IA32_UCODE_REV:
1172         case MSR_IA32_EBL_CR_POWERON:
1173         case MSR_IA32_DEBUGCTLMSR:
1174         case MSR_IA32_LASTBRANCHFROMIP:
1175         case MSR_IA32_LASTBRANCHTOIP:
1176         case MSR_IA32_LASTINTFROMIP:
1177         case MSR_IA32_LASTINTTOIP:
1178         case MSR_K8_SYSCFG:
1179         case MSR_K7_HWCR:
1180         case MSR_VM_HSAVE_PA:
1181         case MSR_P6_PERFCTR0:
1182         case MSR_P6_PERFCTR1:
1183         case MSR_P6_EVNTSEL0:
1184         case MSR_P6_EVNTSEL1:
1185         case MSR_K7_EVNTSEL0:
1186         case MSR_K7_PERFCTR0:
1187         case MSR_K8_INT_PENDING_MSG:
1188         case MSR_AMD64_NB_CFG:
1189         case MSR_FAM10H_MMIO_CONF_BASE:
1190                 data = 0;
1191                 break;
1192         case MSR_MTRRcap:
1193                 data = 0x500 | KVM_NR_VAR_MTRR;
1194                 break;
1195         case 0x200 ... 0x2ff:
1196                 return get_msr_mtrr(vcpu, msr, pdata);
1197         case 0xcd: /* fsb frequency */
1198                 data = 3;
1199                 break;
1200         case MSR_IA32_APICBASE:
1201                 data = kvm_get_apic_base(vcpu);
1202                 break;
1203         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1204                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1205                 break;
1206         case MSR_IA32_MISC_ENABLE:
1207                 data = vcpu->arch.ia32_misc_enable_msr;
1208                 break;
1209         case MSR_IA32_PERF_STATUS:
1210                 /* TSC increment by tick */
1211                 data = 1000ULL;
1212                 /* CPU multiplier */
1213                 data |= (((uint64_t)4ULL) << 40);
1214                 break;
1215         case MSR_EFER:
1216                 data = vcpu->arch.shadow_efer;
1217                 break;
1218         case MSR_KVM_WALL_CLOCK:
1219                 data = vcpu->kvm->arch.wall_clock;
1220                 break;
1221         case MSR_KVM_SYSTEM_TIME:
1222                 data = vcpu->arch.time;
1223                 break;
1224         case MSR_IA32_P5_MC_ADDR:
1225         case MSR_IA32_P5_MC_TYPE:
1226         case MSR_IA32_MCG_CAP:
1227         case MSR_IA32_MCG_CTL:
1228         case MSR_IA32_MCG_STATUS:
1229         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1230                 return get_msr_mce(vcpu, msr, pdata);
1231         default:
1232                 if (!ignore_msrs) {
1233                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1234                         return 1;
1235                 } else {
1236                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1237                         data = 0;
1238                 }
1239                 break;
1240         }
1241         *pdata = data;
1242         return 0;
1243 }
1244 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1245
1246 /*
1247  * Read or write a bunch of msrs. All parameters are kernel addresses.
1248  *
1249  * @return number of msrs set successfully.
1250  */
1251 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1252                     struct kvm_msr_entry *entries,
1253                     int (*do_msr)(struct kvm_vcpu *vcpu,
1254                                   unsigned index, u64 *data))
1255 {
1256         int i;
1257
1258         vcpu_load(vcpu);
1259
1260         down_read(&vcpu->kvm->slots_lock);
1261         for (i = 0; i < msrs->nmsrs; ++i)
1262                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1263                         break;
1264         up_read(&vcpu->kvm->slots_lock);
1265
1266         vcpu_put(vcpu);
1267
1268         return i;
1269 }
1270
1271 /*
1272  * Read or write a bunch of msrs. Parameters are user addresses.
1273  *
1274  * @return number of msrs set successfully.
1275  */
1276 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1277                   int (*do_msr)(struct kvm_vcpu *vcpu,
1278                                 unsigned index, u64 *data),
1279                   int writeback)
1280 {
1281         struct kvm_msrs msrs;
1282         struct kvm_msr_entry *entries;
1283         int r, n;
1284         unsigned size;
1285
1286         r = -EFAULT;
1287         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1288                 goto out;
1289
1290         r = -E2BIG;
1291         if (msrs.nmsrs >= MAX_IO_MSRS)
1292                 goto out;
1293
1294         r = -ENOMEM;
1295         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1296         entries = vmalloc(size);
1297         if (!entries)
1298                 goto out;
1299
1300         r = -EFAULT;
1301         if (copy_from_user(entries, user_msrs->entries, size))
1302                 goto out_free;
1303
1304         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1305         if (r < 0)
1306                 goto out_free;
1307
1308         r = -EFAULT;
1309         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1310                 goto out_free;
1311
1312         r = n;
1313
1314 out_free:
1315         vfree(entries);
1316 out:
1317         return r;
1318 }
1319
1320 int kvm_dev_ioctl_check_extension(long ext)
1321 {
1322         int r;
1323
1324         switch (ext) {
1325         case KVM_CAP_IRQCHIP:
1326         case KVM_CAP_HLT:
1327         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1328         case KVM_CAP_SET_TSS_ADDR:
1329         case KVM_CAP_EXT_CPUID:
1330         case KVM_CAP_CLOCKSOURCE:
1331         case KVM_CAP_PIT:
1332         case KVM_CAP_NOP_IO_DELAY:
1333         case KVM_CAP_MP_STATE:
1334         case KVM_CAP_SYNC_MMU:
1335         case KVM_CAP_REINJECT_CONTROL:
1336         case KVM_CAP_IRQ_INJECT_STATUS:
1337         case KVM_CAP_ASSIGN_DEV_IRQ:
1338         case KVM_CAP_IRQFD:
1339         case KVM_CAP_IOEVENTFD:
1340         case KVM_CAP_PIT2:
1341         case KVM_CAP_PIT_STATE2:
1342         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1343         case KVM_CAP_XEN_HVM:
1344         case KVM_CAP_ADJUST_CLOCK:
1345         case KVM_CAP_VCPU_EVENTS:
1346                 r = 1;
1347                 break;
1348         case KVM_CAP_COALESCED_MMIO:
1349                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1350                 break;
1351         case KVM_CAP_VAPIC:
1352                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1353                 break;
1354         case KVM_CAP_NR_VCPUS:
1355                 r = KVM_MAX_VCPUS;
1356                 break;
1357         case KVM_CAP_NR_MEMSLOTS:
1358                 r = KVM_MEMORY_SLOTS;
1359                 break;
1360         case KVM_CAP_PV_MMU:    /* obsolete */
1361                 r = 0;
1362                 break;
1363         case KVM_CAP_IOMMU:
1364                 r = iommu_found();
1365                 break;
1366         case KVM_CAP_MCE:
1367                 r = KVM_MAX_MCE_BANKS;
1368                 break;
1369         default:
1370                 r = 0;
1371                 break;
1372         }
1373         return r;
1374
1375 }
1376
1377 long kvm_arch_dev_ioctl(struct file *filp,
1378                         unsigned int ioctl, unsigned long arg)
1379 {
1380         void __user *argp = (void __user *)arg;
1381         long r;
1382
1383         switch (ioctl) {
1384         case KVM_GET_MSR_INDEX_LIST: {
1385                 struct kvm_msr_list __user *user_msr_list = argp;
1386                 struct kvm_msr_list msr_list;
1387                 unsigned n;
1388
1389                 r = -EFAULT;
1390                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1391                         goto out;
1392                 n = msr_list.nmsrs;
1393                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1394                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1395                         goto out;
1396                 r = -E2BIG;
1397                 if (n < msr_list.nmsrs)
1398                         goto out;
1399                 r = -EFAULT;
1400                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1401                                  num_msrs_to_save * sizeof(u32)))
1402                         goto out;
1403                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1404                                  &emulated_msrs,
1405                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1406                         goto out;
1407                 r = 0;
1408                 break;
1409         }
1410         case KVM_GET_SUPPORTED_CPUID: {
1411                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1412                 struct kvm_cpuid2 cpuid;
1413
1414                 r = -EFAULT;
1415                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1416                         goto out;
1417                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1418                                                       cpuid_arg->entries);
1419                 if (r)
1420                         goto out;
1421
1422                 r = -EFAULT;
1423                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1424                         goto out;
1425                 r = 0;
1426                 break;
1427         }
1428         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1429                 u64 mce_cap;
1430
1431                 mce_cap = KVM_MCE_CAP_SUPPORTED;
1432                 r = -EFAULT;
1433                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1434                         goto out;
1435                 r = 0;
1436                 break;
1437         }
1438         default:
1439                 r = -EINVAL;
1440         }
1441 out:
1442         return r;
1443 }
1444
1445 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1446 {
1447         kvm_x86_ops->vcpu_load(vcpu, cpu);
1448         if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1449                 unsigned long khz = cpufreq_quick_get(cpu);
1450                 if (!khz)
1451                         khz = tsc_khz;
1452                 per_cpu(cpu_tsc_khz, cpu) = khz;
1453         }
1454         kvm_request_guest_time_update(vcpu);
1455 }
1456
1457 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1458 {
1459         kvm_x86_ops->vcpu_put(vcpu);
1460         kvm_put_guest_fpu(vcpu);
1461 }
1462
1463 static int is_efer_nx(void)
1464 {
1465         unsigned long long efer = 0;
1466
1467         rdmsrl_safe(MSR_EFER, &efer);
1468         return efer & EFER_NX;
1469 }
1470
1471 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1472 {
1473         int i;
1474         struct kvm_cpuid_entry2 *e, *entry;
1475
1476         entry = NULL;
1477         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1478                 e = &vcpu->arch.cpuid_entries[i];
1479                 if (e->function == 0x80000001) {
1480                         entry = e;
1481                         break;
1482                 }
1483         }
1484         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1485                 entry->edx &= ~(1 << 20);
1486                 printk(KERN_INFO "kvm: guest NX capability removed\n");
1487         }
1488 }
1489
1490 /* when an old userspace process fills a new kernel module */
1491 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1492                                     struct kvm_cpuid *cpuid,
1493                                     struct kvm_cpuid_entry __user *entries)
1494 {
1495         int r, i;
1496         struct kvm_cpuid_entry *cpuid_entries;
1497
1498         r = -E2BIG;
1499         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1500                 goto out;
1501         r = -ENOMEM;
1502         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1503         if (!cpuid_entries)
1504                 goto out;
1505         r = -EFAULT;
1506         if (copy_from_user(cpuid_entries, entries,
1507                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1508                 goto out_free;
1509         for (i = 0; i < cpuid->nent; i++) {
1510                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1511                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1512                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1513                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1514                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1515                 vcpu->arch.cpuid_entries[i].index = 0;
1516                 vcpu->arch.cpuid_entries[i].flags = 0;
1517                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1518                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1519                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1520         }
1521         vcpu->arch.cpuid_nent = cpuid->nent;
1522         cpuid_fix_nx_cap(vcpu);
1523         r = 0;
1524         kvm_apic_set_version(vcpu);
1525
1526 out_free:
1527         vfree(cpuid_entries);
1528 out:
1529         return r;
1530 }
1531
1532 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1533                                      struct kvm_cpuid2 *cpuid,
1534                                      struct kvm_cpuid_entry2 __user *entries)
1535 {
1536         int r;
1537
1538         r = -E2BIG;
1539         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1540                 goto out;
1541         r = -EFAULT;
1542         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1543                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1544                 goto out;
1545         vcpu->arch.cpuid_nent = cpuid->nent;
1546         kvm_apic_set_version(vcpu);
1547         return 0;
1548
1549 out:
1550         return r;
1551 }
1552
1553 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1554                                      struct kvm_cpuid2 *cpuid,
1555                                      struct kvm_cpuid_entry2 __user *entries)
1556 {
1557         int r;
1558
1559         r = -E2BIG;
1560         if (cpuid->nent < vcpu->arch.cpuid_nent)
1561                 goto out;
1562         r = -EFAULT;
1563         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1564                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1565                 goto out;
1566         return 0;
1567
1568 out:
1569         cpuid->nent = vcpu->arch.cpuid_nent;
1570         return r;
1571 }
1572
1573 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1574                            u32 index)
1575 {
1576         entry->function = function;
1577         entry->index = index;
1578         cpuid_count(entry->function, entry->index,
1579                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1580         entry->flags = 0;
1581 }
1582
1583 #define F(x) bit(X86_FEATURE_##x)
1584
1585 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1586                          u32 index, int *nent, int maxnent)
1587 {
1588         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1589         unsigned f_gbpages = kvm_x86_ops->gb_page_enable() ? F(GBPAGES) : 0;
1590 #ifdef CONFIG_X86_64
1591         unsigned f_lm = F(LM);
1592 #else
1593         unsigned f_lm = 0;
1594 #endif
1595
1596         /* cpuid 1.edx */
1597         const u32 kvm_supported_word0_x86_features =
1598                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1599                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1600                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1601                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1602                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1603                 0 /* Reserved, DS, ACPI */ | F(MMX) |
1604                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1605                 0 /* HTT, TM, Reserved, PBE */;
1606         /* cpuid 0x80000001.edx */
1607         const u32 kvm_supported_word1_x86_features =
1608                 F(FPU) | F(VME) | F(DE) | F(PSE) |
1609                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1610                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1611                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1612                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1613                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1614                 F(FXSR) | F(FXSR_OPT) | f_gbpages | 0 /* RDTSCP */ |
1615                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1616         /* cpuid 1.ecx */
1617         const u32 kvm_supported_word4_x86_features =
1618                 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1619                 0 /* DS-CPL, VMX, SMX, EST */ |
1620                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1621                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1622                 0 /* Reserved, DCA */ | F(XMM4_1) |
1623                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1624                 0 /* Reserved, XSAVE, OSXSAVE */;
1625         /* cpuid 0x80000001.ecx */
1626         const u32 kvm_supported_word6_x86_features =
1627                 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1628                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1629                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1630                 0 /* SKINIT */ | 0 /* WDT */;
1631
1632         /* all calls to cpuid_count() should be made on the same cpu */
1633         get_cpu();
1634         do_cpuid_1_ent(entry, function, index);
1635         ++*nent;
1636
1637         switch (function) {
1638         case 0:
1639                 entry->eax = min(entry->eax, (u32)0xb);
1640                 break;
1641         case 1:
1642                 entry->edx &= kvm_supported_word0_x86_features;
1643                 entry->ecx &= kvm_supported_word4_x86_features;
1644                 /* we support x2apic emulation even if host does not support
1645                  * it since we emulate x2apic in software */
1646                 entry->ecx |= F(X2APIC);
1647                 break;
1648         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1649          * may return different values. This forces us to get_cpu() before
1650          * issuing the first command, and also to emulate this annoying behavior
1651          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1652         case 2: {
1653                 int t, times = entry->eax & 0xff;
1654
1655                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1656                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1657                 for (t = 1; t < times && *nent < maxnent; ++t) {
1658                         do_cpuid_1_ent(&entry[t], function, 0);
1659                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1660                         ++*nent;
1661                 }
1662                 break;
1663         }
1664         /* function 4 and 0xb have additional index. */
1665         case 4: {
1666                 int i, cache_type;
1667
1668                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1669                 /* read more entries until cache_type is zero */
1670                 for (i = 1; *nent < maxnent; ++i) {
1671                         cache_type = entry[i - 1].eax & 0x1f;
1672                         if (!cache_type)
1673                                 break;
1674                         do_cpuid_1_ent(&entry[i], function, i);
1675                         entry[i].flags |=
1676                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1677                         ++*nent;
1678                 }
1679                 break;
1680         }
1681         case 0xb: {
1682                 int i, level_type;
1683
1684                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1685                 /* read more entries until level_type is zero */
1686                 for (i = 1; *nent < maxnent; ++i) {
1687                         level_type = entry[i - 1].ecx & 0xff00;
1688                         if (!level_type)
1689                                 break;
1690                         do_cpuid_1_ent(&entry[i], function, i);
1691                         entry[i].flags |=
1692                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1693                         ++*nent;
1694                 }
1695                 break;
1696         }
1697         case 0x80000000:
1698                 entry->eax = min(entry->eax, 0x8000001a);
1699                 break;
1700         case 0x80000001:
1701                 entry->edx &= kvm_supported_word1_x86_features;
1702                 entry->ecx &= kvm_supported_word6_x86_features;
1703                 break;
1704         }
1705         put_cpu();
1706 }
1707
1708 #undef F
1709
1710 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1711                                      struct kvm_cpuid_entry2 __user *entries)
1712 {
1713         struct kvm_cpuid_entry2 *cpuid_entries;
1714         int limit, nent = 0, r = -E2BIG;
1715         u32 func;
1716
1717         if (cpuid->nent < 1)
1718                 goto out;
1719         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1720                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1721         r = -ENOMEM;
1722         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1723         if (!cpuid_entries)
1724                 goto out;
1725
1726         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1727         limit = cpuid_entries[0].eax;
1728         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1729                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1730                              &nent, cpuid->nent);
1731         r = -E2BIG;
1732         if (nent >= cpuid->nent)
1733                 goto out_free;
1734
1735         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1736         limit = cpuid_entries[nent - 1].eax;
1737         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1738                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1739                              &nent, cpuid->nent);
1740         r = -E2BIG;
1741         if (nent >= cpuid->nent)
1742                 goto out_free;
1743
1744         r = -EFAULT;
1745         if (copy_to_user(entries, cpuid_entries,
1746                          nent * sizeof(struct kvm_cpuid_entry2)))
1747                 goto out_free;
1748         cpuid->nent = nent;
1749         r = 0;
1750
1751 out_free:
1752         vfree(cpuid_entries);
1753 out:
1754         return r;
1755 }
1756
1757 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1758                                     struct kvm_lapic_state *s)
1759 {
1760         vcpu_load(vcpu);
1761         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1762         vcpu_put(vcpu);
1763
1764         return 0;
1765 }
1766
1767 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1768                                     struct kvm_lapic_state *s)
1769 {
1770         vcpu_load(vcpu);
1771         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1772         kvm_apic_post_state_restore(vcpu);
1773         update_cr8_intercept(vcpu);
1774         vcpu_put(vcpu);
1775
1776         return 0;
1777 }
1778
1779 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1780                                     struct kvm_interrupt *irq)
1781 {
1782         if (irq->irq < 0 || irq->irq >= 256)
1783                 return -EINVAL;
1784         if (irqchip_in_kernel(vcpu->kvm))
1785                 return -ENXIO;
1786         vcpu_load(vcpu);
1787
1788         kvm_queue_interrupt(vcpu, irq->irq, false);
1789
1790         vcpu_put(vcpu);
1791
1792         return 0;
1793 }
1794
1795 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1796 {
1797         vcpu_load(vcpu);
1798         kvm_inject_nmi(vcpu);
1799         vcpu_put(vcpu);
1800
1801         return 0;
1802 }
1803
1804 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1805                                            struct kvm_tpr_access_ctl *tac)
1806 {
1807         if (tac->flags)
1808                 return -EINVAL;
1809         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1810         return 0;
1811 }
1812
1813 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1814                                         u64 mcg_cap)
1815 {
1816         int r;
1817         unsigned bank_num = mcg_cap & 0xff, bank;
1818
1819         r = -EINVAL;
1820         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1821                 goto out;
1822         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1823                 goto out;
1824         r = 0;
1825         vcpu->arch.mcg_cap = mcg_cap;
1826         /* Init IA32_MCG_CTL to all 1s */
1827         if (mcg_cap & MCG_CTL_P)
1828                 vcpu->arch.mcg_ctl = ~(u64)0;
1829         /* Init IA32_MCi_CTL to all 1s */
1830         for (bank = 0; bank < bank_num; bank++)
1831                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1832 out:
1833         return r;
1834 }
1835
1836 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1837                                       struct kvm_x86_mce *mce)
1838 {
1839         u64 mcg_cap = vcpu->arch.mcg_cap;
1840         unsigned bank_num = mcg_cap & 0xff;
1841         u64 *banks = vcpu->arch.mce_banks;
1842
1843         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1844                 return -EINVAL;
1845         /*
1846          * if IA32_MCG_CTL is not all 1s, the uncorrected error
1847          * reporting is disabled
1848          */
1849         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1850             vcpu->arch.mcg_ctl != ~(u64)0)
1851                 return 0;
1852         banks += 4 * mce->bank;
1853         /*
1854          * if IA32_MCi_CTL is not all 1s, the uncorrected error
1855          * reporting is disabled for the bank
1856          */
1857         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1858                 return 0;
1859         if (mce->status & MCI_STATUS_UC) {
1860                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1861                     !(vcpu->arch.cr4 & X86_CR4_MCE)) {
1862                         printk(KERN_DEBUG "kvm: set_mce: "
1863                                "injects mce exception while "
1864                                "previous one is in progress!\n");
1865                         set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1866                         return 0;
1867                 }
1868                 if (banks[1] & MCI_STATUS_VAL)
1869                         mce->status |= MCI_STATUS_OVER;
1870                 banks[2] = mce->addr;
1871                 banks[3] = mce->misc;
1872                 vcpu->arch.mcg_status = mce->mcg_status;
1873                 banks[1] = mce->status;
1874                 kvm_queue_exception(vcpu, MC_VECTOR);
1875         } else if (!(banks[1] & MCI_STATUS_VAL)
1876                    || !(banks[1] & MCI_STATUS_UC)) {
1877                 if (banks[1] & MCI_STATUS_VAL)
1878                         mce->status |= MCI_STATUS_OVER;
1879                 banks[2] = mce->addr;
1880                 banks[3] = mce->misc;
1881                 banks[1] = mce->status;
1882         } else
1883                 banks[1] |= MCI_STATUS_OVER;
1884         return 0;
1885 }
1886
1887 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
1888                                                struct kvm_vcpu_events *events)
1889 {
1890         vcpu_load(vcpu);
1891
1892         events->exception.injected = vcpu->arch.exception.pending;
1893         events->exception.nr = vcpu->arch.exception.nr;
1894         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
1895         events->exception.error_code = vcpu->arch.exception.error_code;
1896
1897         events->interrupt.injected = vcpu->arch.interrupt.pending;
1898         events->interrupt.nr = vcpu->arch.interrupt.nr;
1899         events->interrupt.soft = vcpu->arch.interrupt.soft;
1900
1901         events->nmi.injected = vcpu->arch.nmi_injected;
1902         events->nmi.pending = vcpu->arch.nmi_pending;
1903         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
1904
1905         events->sipi_vector = vcpu->arch.sipi_vector;
1906
1907         events->flags = 0;
1908
1909         vcpu_put(vcpu);
1910 }
1911
1912 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
1913                                               struct kvm_vcpu_events *events)
1914 {
1915         if (events->flags)
1916                 return -EINVAL;
1917
1918         vcpu_load(vcpu);
1919
1920         vcpu->arch.exception.pending = events->exception.injected;
1921         vcpu->arch.exception.nr = events->exception.nr;
1922         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
1923         vcpu->arch.exception.error_code = events->exception.error_code;
1924
1925         vcpu->arch.interrupt.pending = events->interrupt.injected;
1926         vcpu->arch.interrupt.nr = events->interrupt.nr;
1927         vcpu->arch.interrupt.soft = events->interrupt.soft;
1928         if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
1929                 kvm_pic_clear_isr_ack(vcpu->kvm);
1930
1931         vcpu->arch.nmi_injected = events->nmi.injected;
1932         vcpu->arch.nmi_pending = events->nmi.pending;
1933         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
1934
1935         vcpu->arch.sipi_vector = events->sipi_vector;
1936
1937         vcpu_put(vcpu);
1938
1939         return 0;
1940 }
1941
1942 long kvm_arch_vcpu_ioctl(struct file *filp,
1943                          unsigned int ioctl, unsigned long arg)
1944 {
1945         struct kvm_vcpu *vcpu = filp->private_data;
1946         void __user *argp = (void __user *)arg;
1947         int r;
1948         struct kvm_lapic_state *lapic = NULL;
1949
1950         switch (ioctl) {
1951         case KVM_GET_LAPIC: {
1952                 r = -EINVAL;
1953                 if (!vcpu->arch.apic)
1954                         goto out;
1955                 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1956
1957                 r = -ENOMEM;
1958                 if (!lapic)
1959                         goto out;
1960                 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1961                 if (r)
1962                         goto out;
1963                 r = -EFAULT;
1964                 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1965                         goto out;
1966                 r = 0;
1967                 break;
1968         }
1969         case KVM_SET_LAPIC: {
1970                 r = -EINVAL;
1971                 if (!vcpu->arch.apic)
1972                         goto out;
1973                 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1974                 r = -ENOMEM;
1975                 if (!lapic)
1976                         goto out;
1977                 r = -EFAULT;
1978                 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1979                         goto out;
1980                 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1981                 if (r)
1982                         goto out;
1983                 r = 0;
1984                 break;
1985         }
1986         case KVM_INTERRUPT: {
1987                 struct kvm_interrupt irq;
1988
1989                 r = -EFAULT;
1990                 if (copy_from_user(&irq, argp, sizeof irq))
1991                         goto out;
1992                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1993                 if (r)
1994                         goto out;
1995                 r = 0;
1996                 break;
1997         }
1998         case KVM_NMI: {
1999                 r = kvm_vcpu_ioctl_nmi(vcpu);
2000                 if (r)
2001                         goto out;
2002                 r = 0;
2003                 break;
2004         }
2005         case KVM_SET_CPUID: {
2006                 struct kvm_cpuid __user *cpuid_arg = argp;
2007                 struct kvm_cpuid cpuid;
2008
2009                 r = -EFAULT;
2010                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2011                         goto out;
2012                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2013                 if (r)
2014                         goto out;
2015                 break;
2016         }
2017         case KVM_SET_CPUID2: {
2018                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2019                 struct kvm_cpuid2 cpuid;
2020
2021                 r = -EFAULT;
2022                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2023                         goto out;
2024                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2025                                               cpuid_arg->entries);
2026                 if (r)
2027                         goto out;
2028                 break;
2029         }
2030         case KVM_GET_CPUID2: {
2031                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2032                 struct kvm_cpuid2 cpuid;
2033
2034                 r = -EFAULT;
2035                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2036                         goto out;
2037                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2038                                               cpuid_arg->entries);
2039                 if (r)
2040                         goto out;
2041                 r = -EFAULT;
2042                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2043                         goto out;
2044                 r = 0;
2045                 break;
2046         }
2047         case KVM_GET_MSRS:
2048                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2049                 break;
2050         case KVM_SET_MSRS:
2051                 r = msr_io(vcpu, argp, do_set_msr, 0);
2052                 break;
2053         case KVM_TPR_ACCESS_REPORTING: {
2054                 struct kvm_tpr_access_ctl tac;
2055
2056                 r = -EFAULT;
2057                 if (copy_from_user(&tac, argp, sizeof tac))
2058                         goto out;
2059                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2060                 if (r)
2061                         goto out;
2062                 r = -EFAULT;
2063                 if (copy_to_user(argp, &tac, sizeof tac))
2064                         goto out;
2065                 r = 0;
2066                 break;
2067         };
2068         case KVM_SET_VAPIC_ADDR: {
2069                 struct kvm_vapic_addr va;
2070
2071                 r = -EINVAL;
2072                 if (!irqchip_in_kernel(vcpu->kvm))
2073                         goto out;
2074                 r = -EFAULT;
2075                 if (copy_from_user(&va, argp, sizeof va))
2076                         goto out;
2077                 r = 0;
2078                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2079                 break;
2080         }
2081         case KVM_X86_SETUP_MCE: {
2082                 u64 mcg_cap;
2083
2084                 r = -EFAULT;
2085                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2086                         goto out;
2087                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2088                 break;
2089         }
2090         case KVM_X86_SET_MCE: {
2091                 struct kvm_x86_mce mce;
2092
2093                 r = -EFAULT;
2094                 if (copy_from_user(&mce, argp, sizeof mce))
2095                         goto out;
2096                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2097                 break;
2098         }
2099         case KVM_GET_VCPU_EVENTS: {
2100                 struct kvm_vcpu_events events;
2101
2102                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2103
2104                 r = -EFAULT;
2105                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2106                         break;
2107                 r = 0;
2108                 break;
2109         }
2110         case KVM_SET_VCPU_EVENTS: {
2111                 struct kvm_vcpu_events events;
2112
2113                 r = -EFAULT;
2114                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2115                         break;
2116
2117                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2118                 break;
2119         }
2120         default:
2121                 r = -EINVAL;
2122         }
2123 out:
2124         kfree(lapic);
2125         return r;
2126 }
2127
2128 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2129 {
2130         int ret;
2131
2132         if (addr > (unsigned int)(-3 * PAGE_SIZE))
2133                 return -1;
2134         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2135         return ret;
2136 }
2137
2138 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2139                                               u64 ident_addr)
2140 {
2141         kvm->arch.ept_identity_map_addr = ident_addr;
2142         return 0;
2143 }
2144
2145 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2146                                           u32 kvm_nr_mmu_pages)
2147 {
2148         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2149                 return -EINVAL;
2150
2151         down_write(&kvm->slots_lock);
2152         spin_lock(&kvm->mmu_lock);
2153
2154         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2155         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2156
2157         spin_unlock(&kvm->mmu_lock);
2158         up_write(&kvm->slots_lock);
2159         return 0;
2160 }
2161
2162 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2163 {
2164         return kvm->arch.n_alloc_mmu_pages;
2165 }
2166
2167 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2168 {
2169         int i;
2170         struct kvm_mem_alias *alias;
2171
2172         for (i = 0; i < kvm->arch.naliases; ++i) {
2173                 alias = &kvm->arch.aliases[i];
2174                 if (gfn >= alias->base_gfn
2175                     && gfn < alias->base_gfn + alias->npages)
2176                         return alias->target_gfn + gfn - alias->base_gfn;
2177         }
2178         return gfn;
2179 }
2180
2181 /*
2182  * Set a new alias region.  Aliases map a portion of physical memory into
2183  * another portion.  This is useful for memory windows, for example the PC
2184  * VGA region.
2185  */
2186 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2187                                          struct kvm_memory_alias *alias)
2188 {
2189         int r, n;
2190         struct kvm_mem_alias *p;
2191
2192         r = -EINVAL;
2193         /* General sanity checks */
2194         if (alias->memory_size & (PAGE_SIZE - 1))
2195                 goto out;
2196         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2197                 goto out;
2198         if (alias->slot >= KVM_ALIAS_SLOTS)
2199                 goto out;
2200         if (alias->guest_phys_addr + alias->memory_size
2201             < alias->guest_phys_addr)
2202                 goto out;
2203         if (alias->target_phys_addr + alias->memory_size
2204             < alias->target_phys_addr)
2205                 goto out;
2206
2207         down_write(&kvm->slots_lock);
2208         spin_lock(&kvm->mmu_lock);
2209
2210         p = &kvm->arch.aliases[alias->slot];
2211         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2212         p->npages = alias->memory_size >> PAGE_SHIFT;
2213         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2214
2215         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2216                 if (kvm->arch.aliases[n - 1].npages)
2217                         break;
2218         kvm->arch.naliases = n;
2219
2220         spin_unlock(&kvm->mmu_lock);
2221         kvm_mmu_zap_all(kvm);
2222
2223         up_write(&kvm->slots_lock);
2224
2225         return 0;
2226
2227 out:
2228         return r;
2229 }
2230
2231 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2232 {
2233         int r;
2234
2235         r = 0;
2236         switch (chip->chip_id) {
2237         case KVM_IRQCHIP_PIC_MASTER:
2238                 memcpy(&chip->chip.pic,
2239                         &pic_irqchip(kvm)->pics[0],
2240                         sizeof(struct kvm_pic_state));
2241                 break;
2242         case KVM_IRQCHIP_PIC_SLAVE:
2243                 memcpy(&chip->chip.pic,
2244                         &pic_irqchip(kvm)->pics[1],
2245                         sizeof(struct kvm_pic_state));
2246                 break;
2247         case KVM_IRQCHIP_IOAPIC:
2248                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2249                 break;
2250         default:
2251                 r = -EINVAL;
2252                 break;
2253         }
2254         return r;
2255 }
2256
2257 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2258 {
2259         int r;
2260
2261         r = 0;
2262         switch (chip->chip_id) {
2263         case KVM_IRQCHIP_PIC_MASTER:
2264                 spin_lock(&pic_irqchip(kvm)->lock);
2265                 memcpy(&pic_irqchip(kvm)->pics[0],
2266                         &chip->chip.pic,
2267                         sizeof(struct kvm_pic_state));
2268                 spin_unlock(&pic_irqchip(kvm)->lock);
2269                 break;
2270         case KVM_IRQCHIP_PIC_SLAVE:
2271                 spin_lock(&pic_irqchip(kvm)->lock);
2272                 memcpy(&pic_irqchip(kvm)->pics[1],
2273                         &chip->chip.pic,
2274                         sizeof(struct kvm_pic_state));
2275                 spin_unlock(&pic_irqchip(kvm)->lock);
2276                 break;
2277         case KVM_IRQCHIP_IOAPIC:
2278                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2279                 break;
2280         default:
2281                 r = -EINVAL;
2282                 break;
2283         }
2284         kvm_pic_update_irq(pic_irqchip(kvm));
2285         return r;
2286 }
2287
2288 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2289 {
2290         int r = 0;
2291
2292         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2293         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2294         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2295         return r;
2296 }
2297
2298 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2299 {
2300         int r = 0;
2301
2302         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2303         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2304         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2305         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2306         return r;
2307 }
2308
2309 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2310 {
2311         int r = 0;
2312
2313         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2314         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2315                 sizeof(ps->channels));
2316         ps->flags = kvm->arch.vpit->pit_state.flags;
2317         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2318         return r;
2319 }
2320
2321 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2322 {
2323         int r = 0, start = 0;
2324         u32 prev_legacy, cur_legacy;
2325         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2326         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2327         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2328         if (!prev_legacy && cur_legacy)
2329                 start = 1;
2330         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2331                sizeof(kvm->arch.vpit->pit_state.channels));
2332         kvm->arch.vpit->pit_state.flags = ps->flags;
2333         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2334         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2335         return r;
2336 }
2337
2338 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2339                                  struct kvm_reinject_control *control)
2340 {
2341         if (!kvm->arch.vpit)
2342                 return -ENXIO;
2343         mutex_lock(&kvm->arch.vpit->pit_state.lock);
2344         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2345         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2346         return 0;
2347 }
2348
2349 /*
2350  * Get (and clear) the dirty memory log for a memory slot.
2351  */
2352 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2353                                       struct kvm_dirty_log *log)
2354 {
2355         int r;
2356         int n;
2357         struct kvm_memory_slot *memslot;
2358         int is_dirty = 0;
2359
2360         down_write(&kvm->slots_lock);
2361
2362         r = kvm_get_dirty_log(kvm, log, &is_dirty);
2363         if (r)
2364                 goto out;
2365
2366         /* If nothing is dirty, don't bother messing with page tables. */
2367         if (is_dirty) {
2368                 spin_lock(&kvm->mmu_lock);
2369                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2370                 spin_unlock(&kvm->mmu_lock);
2371                 memslot = &kvm->memslots[log->slot];
2372                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2373                 memset(memslot->dirty_bitmap, 0, n);
2374         }
2375         r = 0;
2376 out:
2377         up_write(&kvm->slots_lock);
2378         return r;
2379 }
2380
2381 long kvm_arch_vm_ioctl(struct file *filp,
2382                        unsigned int ioctl, unsigned long arg)
2383 {
2384         struct kvm *kvm = filp->private_data;
2385         void __user *argp = (void __user *)arg;
2386         int r = -ENOTTY;
2387         /*
2388          * This union makes it completely explicit to gcc-3.x
2389          * that these two variables' stack usage should be
2390          * combined, not added together.
2391          */
2392         union {
2393                 struct kvm_pit_state ps;
2394                 struct kvm_pit_state2 ps2;
2395                 struct kvm_memory_alias alias;
2396                 struct kvm_pit_config pit_config;
2397         } u;
2398
2399         switch (ioctl) {
2400         case KVM_SET_TSS_ADDR:
2401                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2402                 if (r < 0)
2403                         goto out;
2404                 break;
2405         case KVM_SET_IDENTITY_MAP_ADDR: {
2406                 u64 ident_addr;
2407
2408                 r = -EFAULT;
2409                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2410                         goto out;
2411                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2412                 if (r < 0)
2413                         goto out;
2414                 break;
2415         }
2416         case KVM_SET_MEMORY_REGION: {
2417                 struct kvm_memory_region kvm_mem;
2418                 struct kvm_userspace_memory_region kvm_userspace_mem;
2419
2420                 r = -EFAULT;
2421                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2422                         goto out;
2423                 kvm_userspace_mem.slot = kvm_mem.slot;
2424                 kvm_userspace_mem.flags = kvm_mem.flags;
2425                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2426                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2427                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2428                 if (r)
2429                         goto out;
2430                 break;
2431         }
2432         case KVM_SET_NR_MMU_PAGES:
2433                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2434                 if (r)
2435                         goto out;
2436                 break;
2437         case KVM_GET_NR_MMU_PAGES:
2438                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2439                 break;
2440         case KVM_SET_MEMORY_ALIAS:
2441                 r = -EFAULT;
2442                 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2443                         goto out;
2444                 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2445                 if (r)
2446                         goto out;
2447                 break;
2448         case KVM_CREATE_IRQCHIP: {
2449                 struct kvm_pic *vpic;
2450
2451                 mutex_lock(&kvm->lock);
2452                 r = -EEXIST;
2453                 if (kvm->arch.vpic)
2454                         goto create_irqchip_unlock;
2455                 r = -ENOMEM;
2456                 vpic = kvm_create_pic(kvm);
2457                 if (vpic) {
2458                         r = kvm_ioapic_init(kvm);
2459                         if (r) {
2460                                 kfree(vpic);
2461                                 goto create_irqchip_unlock;
2462                         }
2463                 } else
2464                         goto create_irqchip_unlock;
2465                 smp_wmb();
2466                 kvm->arch.vpic = vpic;
2467                 smp_wmb();
2468                 r = kvm_setup_default_irq_routing(kvm);
2469                 if (r) {
2470                         mutex_lock(&kvm->irq_lock);
2471                         kfree(kvm->arch.vpic);
2472                         kfree(kvm->arch.vioapic);
2473                         kvm->arch.vpic = NULL;
2474                         kvm->arch.vioapic = NULL;
2475                         mutex_unlock(&kvm->irq_lock);
2476                 }
2477         create_irqchip_unlock:
2478                 mutex_unlock(&kvm->lock);
2479                 break;
2480         }
2481         case KVM_CREATE_PIT:
2482                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2483                 goto create_pit;
2484         case KVM_CREATE_PIT2:
2485                 r = -EFAULT;
2486                 if (copy_from_user(&u.pit_config, argp,
2487                                    sizeof(struct kvm_pit_config)))
2488                         goto out;
2489         create_pit:
2490                 down_write(&kvm->slots_lock);
2491                 r = -EEXIST;
2492                 if (kvm->arch.vpit)
2493                         goto create_pit_unlock;
2494                 r = -ENOMEM;
2495                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2496                 if (kvm->arch.vpit)
2497                         r = 0;
2498         create_pit_unlock:
2499                 up_write(&kvm->slots_lock);
2500                 break;
2501         case KVM_IRQ_LINE_STATUS:
2502         case KVM_IRQ_LINE: {
2503                 struct kvm_irq_level irq_event;
2504
2505                 r = -EFAULT;
2506                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2507                         goto out;
2508                 if (irqchip_in_kernel(kvm)) {
2509                         __s32 status;
2510                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2511                                         irq_event.irq, irq_event.level);
2512                         if (ioctl == KVM_IRQ_LINE_STATUS) {
2513                                 irq_event.status = status;
2514                                 if (copy_to_user(argp, &irq_event,
2515                                                         sizeof irq_event))
2516                                         goto out;
2517                         }
2518                         r = 0;
2519                 }
2520                 break;
2521         }
2522         case KVM_GET_IRQCHIP: {
2523                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2524                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2525
2526                 r = -ENOMEM;
2527                 if (!chip)
2528                         goto out;
2529                 r = -EFAULT;
2530                 if (copy_from_user(chip, argp, sizeof *chip))
2531                         goto get_irqchip_out;
2532                 r = -ENXIO;
2533                 if (!irqchip_in_kernel(kvm))
2534                         goto get_irqchip_out;
2535                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2536                 if (r)
2537                         goto get_irqchip_out;
2538                 r = -EFAULT;
2539                 if (copy_to_user(argp, chip, sizeof *chip))
2540                         goto get_irqchip_out;
2541                 r = 0;
2542         get_irqchip_out:
2543                 kfree(chip);
2544                 if (r)
2545                         goto out;
2546                 break;
2547         }
2548         case KVM_SET_IRQCHIP: {
2549                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2550                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2551
2552                 r = -ENOMEM;
2553                 if (!chip)
2554                         goto out;
2555                 r = -EFAULT;
2556                 if (copy_from_user(chip, argp, sizeof *chip))
2557                         goto set_irqchip_out;
2558                 r = -ENXIO;
2559                 if (!irqchip_in_kernel(kvm))
2560                         goto set_irqchip_out;
2561                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2562                 if (r)
2563                         goto set_irqchip_out;
2564                 r = 0;
2565         set_irqchip_out:
2566                 kfree(chip);
2567                 if (r)
2568                         goto out;
2569                 break;
2570         }
2571         case KVM_GET_PIT: {
2572                 r = -EFAULT;
2573                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2574                         goto out;
2575                 r = -ENXIO;
2576                 if (!kvm->arch.vpit)
2577                         goto out;
2578                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2579                 if (r)
2580                         goto out;
2581                 r = -EFAULT;
2582                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2583                         goto out;
2584                 r = 0;
2585                 break;
2586         }
2587         case KVM_SET_PIT: {
2588                 r = -EFAULT;
2589                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2590                         goto out;
2591                 r = -ENXIO;
2592                 if (!kvm->arch.vpit)
2593                         goto out;
2594                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2595                 if (r)
2596                         goto out;
2597                 r = 0;
2598                 break;
2599         }
2600         case KVM_GET_PIT2: {
2601                 r = -ENXIO;
2602                 if (!kvm->arch.vpit)
2603                         goto out;
2604                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2605                 if (r)
2606                         goto out;
2607                 r = -EFAULT;
2608                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2609                         goto out;
2610                 r = 0;
2611                 break;
2612         }
2613         case KVM_SET_PIT2: {
2614                 r = -EFAULT;
2615                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2616                         goto out;
2617                 r = -ENXIO;
2618                 if (!kvm->arch.vpit)
2619                         goto out;
2620                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2621                 if (r)
2622                         goto out;
2623                 r = 0;
2624                 break;
2625         }
2626         case KVM_REINJECT_CONTROL: {
2627                 struct kvm_reinject_control control;
2628                 r =  -EFAULT;
2629                 if (copy_from_user(&control, argp, sizeof(control)))
2630                         goto out;
2631                 r = kvm_vm_ioctl_reinject(kvm, &control);
2632                 if (r)
2633                         goto out;
2634                 r = 0;
2635                 break;
2636         }
2637         case KVM_XEN_HVM_CONFIG: {
2638                 r = -EFAULT;
2639                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2640                                    sizeof(struct kvm_xen_hvm_config)))
2641                         goto out;
2642                 r = -EINVAL;
2643                 if (kvm->arch.xen_hvm_config.flags)
2644                         goto out;
2645                 r = 0;
2646                 break;
2647         }
2648         case KVM_SET_CLOCK: {
2649                 struct timespec now;
2650                 struct kvm_clock_data user_ns;
2651                 u64 now_ns;
2652                 s64 delta;
2653
2654                 r = -EFAULT;
2655                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2656                         goto out;
2657
2658                 r = -EINVAL;
2659                 if (user_ns.flags)
2660                         goto out;
2661
2662                 r = 0;
2663                 ktime_get_ts(&now);
2664                 now_ns = timespec_to_ns(&now);
2665                 delta = user_ns.clock - now_ns;
2666                 kvm->arch.kvmclock_offset = delta;
2667                 break;
2668         }
2669         case KVM_GET_CLOCK: {
2670                 struct timespec now;
2671                 struct kvm_clock_data user_ns;
2672                 u64 now_ns;
2673
2674                 ktime_get_ts(&now);
2675                 now_ns = timespec_to_ns(&now);
2676                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2677                 user_ns.flags = 0;
2678
2679                 r = -EFAULT;
2680                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2681                         goto out;
2682                 r = 0;
2683                 break;
2684         }
2685
2686         default:
2687                 ;
2688         }
2689 out:
2690         return r;
2691 }
2692
2693 static void kvm_init_msr_list(void)
2694 {
2695         u32 dummy[2];
2696         unsigned i, j;
2697
2698         /* skip the first msrs in the list. KVM-specific */
2699         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
2700                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2701                         continue;
2702                 if (j < i)
2703                         msrs_to_save[j] = msrs_to_save[i];
2704                 j++;
2705         }
2706         num_msrs_to_save = j;
2707 }
2708
2709 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2710                            const void *v)
2711 {
2712         if (vcpu->arch.apic &&
2713             !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2714                 return 0;
2715
2716         return kvm_io_bus_write(&vcpu->kvm->mmio_bus, addr, len, v);
2717 }
2718
2719 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2720 {
2721         if (vcpu->arch.apic &&
2722             !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2723                 return 0;
2724
2725         return kvm_io_bus_read(&vcpu->kvm->mmio_bus, addr, len, v);
2726 }
2727
2728 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2729                                struct kvm_vcpu *vcpu)
2730 {
2731         void *data = val;
2732         int r = X86EMUL_CONTINUE;
2733
2734         while (bytes) {
2735                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2736                 unsigned offset = addr & (PAGE_SIZE-1);
2737                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2738                 int ret;
2739
2740                 if (gpa == UNMAPPED_GVA) {
2741                         r = X86EMUL_PROPAGATE_FAULT;
2742                         goto out;
2743                 }
2744                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2745                 if (ret < 0) {
2746                         r = X86EMUL_UNHANDLEABLE;
2747                         goto out;
2748                 }
2749
2750                 bytes -= toread;
2751                 data += toread;
2752                 addr += toread;
2753         }
2754 out:
2755         return r;
2756 }
2757
2758 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2759                                 struct kvm_vcpu *vcpu)
2760 {
2761         void *data = val;
2762         int r = X86EMUL_CONTINUE;
2763
2764         while (bytes) {
2765                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2766                 unsigned offset = addr & (PAGE_SIZE-1);
2767                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2768                 int ret;
2769
2770                 if (gpa == UNMAPPED_GVA) {
2771                         r = X86EMUL_PROPAGATE_FAULT;
2772                         goto out;
2773                 }
2774                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2775                 if (ret < 0) {
2776                         r = X86EMUL_UNHANDLEABLE;
2777                         goto out;
2778                 }
2779
2780                 bytes -= towrite;
2781                 data += towrite;
2782                 addr += towrite;
2783         }
2784 out:
2785         return r;
2786 }
2787
2788
2789 static int emulator_read_emulated(unsigned long addr,
2790                                   void *val,
2791                                   unsigned int bytes,
2792                                   struct kvm_vcpu *vcpu)
2793 {
2794         gpa_t                 gpa;
2795
2796         if (vcpu->mmio_read_completed) {
2797                 memcpy(val, vcpu->mmio_data, bytes);
2798                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
2799                                vcpu->mmio_phys_addr, *(u64 *)val);
2800                 vcpu->mmio_read_completed = 0;
2801                 return X86EMUL_CONTINUE;
2802         }
2803
2804         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2805
2806         /* For APIC access vmexit */
2807         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2808                 goto mmio;
2809
2810         if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2811                                 == X86EMUL_CONTINUE)
2812                 return X86EMUL_CONTINUE;
2813         if (gpa == UNMAPPED_GVA)
2814                 return X86EMUL_PROPAGATE_FAULT;
2815
2816 mmio:
2817         /*
2818          * Is this MMIO handled locally?
2819          */
2820         if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
2821                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
2822                 return X86EMUL_CONTINUE;
2823         }
2824
2825         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
2826
2827         vcpu->mmio_needed = 1;
2828         vcpu->mmio_phys_addr = gpa;
2829         vcpu->mmio_size = bytes;
2830         vcpu->mmio_is_write = 0;
2831
2832         return X86EMUL_UNHANDLEABLE;
2833 }
2834
2835 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2836                           const void *val, int bytes)
2837 {
2838         int ret;
2839
2840         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2841         if (ret < 0)
2842                 return 0;
2843         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2844         return 1;
2845 }
2846
2847 static int emulator_write_emulated_onepage(unsigned long addr,
2848                                            const void *val,
2849                                            unsigned int bytes,
2850                                            struct kvm_vcpu *vcpu)
2851 {
2852         gpa_t                 gpa;
2853
2854         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2855
2856         if (gpa == UNMAPPED_GVA) {
2857                 kvm_inject_page_fault(vcpu, addr, 2);
2858                 return X86EMUL_PROPAGATE_FAULT;
2859         }
2860
2861         /* For APIC access vmexit */
2862         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2863                 goto mmio;
2864
2865         if (emulator_write_phys(vcpu, gpa, val, bytes))
2866                 return X86EMUL_CONTINUE;
2867
2868 mmio:
2869         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
2870         /*
2871          * Is this MMIO handled locally?
2872          */
2873         if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
2874                 return X86EMUL_CONTINUE;
2875
2876         vcpu->mmio_needed = 1;
2877         vcpu->mmio_phys_addr = gpa;
2878         vcpu->mmio_size = bytes;
2879         vcpu->mmio_is_write = 1;
2880         memcpy(vcpu->mmio_data, val, bytes);
2881
2882         return X86EMUL_CONTINUE;
2883 }
2884
2885 int emulator_write_emulated(unsigned long addr,
2886                                    const void *val,
2887                                    unsigned int bytes,
2888                                    struct kvm_vcpu *vcpu)
2889 {
2890         /* Crossing a page boundary? */
2891         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2892                 int rc, now;
2893
2894                 now = -addr & ~PAGE_MASK;
2895                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2896                 if (rc != X86EMUL_CONTINUE)
2897                         return rc;
2898                 addr += now;
2899                 val += now;
2900                 bytes -= now;
2901         }
2902         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2903 }
2904 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2905
2906 static int emulator_cmpxchg_emulated(unsigned long addr,
2907                                      const void *old,
2908                                      const void *new,
2909                                      unsigned int bytes,
2910                                      struct kvm_vcpu *vcpu)
2911 {
2912         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2913 #ifndef CONFIG_X86_64
2914         /* guests cmpxchg8b have to be emulated atomically */
2915         if (bytes == 8) {
2916                 gpa_t gpa;
2917                 struct page *page;
2918                 char *kaddr;
2919                 u64 val;
2920
2921                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2922
2923                 if (gpa == UNMAPPED_GVA ||
2924                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2925                         goto emul_write;
2926
2927                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2928                         goto emul_write;
2929
2930                 val = *(u64 *)new;
2931
2932                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2933
2934                 kaddr = kmap_atomic(page, KM_USER0);
2935                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2936                 kunmap_atomic(kaddr, KM_USER0);
2937                 kvm_release_page_dirty(page);
2938         }
2939 emul_write:
2940 #endif
2941
2942         return emulator_write_emulated(addr, new, bytes, vcpu);
2943 }
2944
2945 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2946 {
2947         return kvm_x86_ops->get_segment_base(vcpu, seg);
2948 }
2949
2950 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2951 {
2952         kvm_mmu_invlpg(vcpu, address);
2953         return X86EMUL_CONTINUE;
2954 }
2955
2956 int emulate_clts(struct kvm_vcpu *vcpu)
2957 {
2958         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2959         return X86EMUL_CONTINUE;
2960 }
2961
2962 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2963 {
2964         struct kvm_vcpu *vcpu = ctxt->vcpu;
2965
2966         switch (dr) {
2967         case 0 ... 3:
2968                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2969                 return X86EMUL_CONTINUE;
2970         default:
2971                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2972                 return X86EMUL_UNHANDLEABLE;
2973         }
2974 }
2975
2976 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2977 {
2978         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2979         int exception;
2980
2981         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2982         if (exception) {
2983                 /* FIXME: better handling */
2984                 return X86EMUL_UNHANDLEABLE;
2985         }
2986         return X86EMUL_CONTINUE;
2987 }
2988
2989 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2990 {
2991         u8 opcodes[4];
2992         unsigned long rip = kvm_rip_read(vcpu);
2993         unsigned long rip_linear;
2994
2995         if (!printk_ratelimit())
2996                 return;
2997
2998         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2999
3000         kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
3001
3002         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3003                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3004 }
3005 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3006
3007 static struct x86_emulate_ops emulate_ops = {
3008         .read_std            = kvm_read_guest_virt,
3009         .read_emulated       = emulator_read_emulated,
3010         .write_emulated      = emulator_write_emulated,
3011         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
3012 };
3013
3014 static void cache_all_regs(struct kvm_vcpu *vcpu)
3015 {
3016         kvm_register_read(vcpu, VCPU_REGS_RAX);
3017         kvm_register_read(vcpu, VCPU_REGS_RSP);
3018         kvm_register_read(vcpu, VCPU_REGS_RIP);
3019         vcpu->arch.regs_dirty = ~0;
3020 }
3021
3022 int emulate_instruction(struct kvm_vcpu *vcpu,
3023                         unsigned long cr2,
3024                         u16 error_code,
3025                         int emulation_type)
3026 {
3027         int r, shadow_mask;
3028         struct decode_cache *c;
3029         struct kvm_run *run = vcpu->run;
3030
3031         kvm_clear_exception_queue(vcpu);
3032         vcpu->arch.mmio_fault_cr2 = cr2;
3033         /*
3034          * TODO: fix emulate.c to use guest_read/write_register
3035          * instead of direct ->regs accesses, can save hundred cycles
3036          * on Intel for instructions that don't read/change RSP, for
3037          * for example.
3038          */
3039         cache_all_regs(vcpu);
3040
3041         vcpu->mmio_is_write = 0;
3042         vcpu->arch.pio.string = 0;
3043
3044         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3045                 int cs_db, cs_l;
3046                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3047
3048                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3049                 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3050                 vcpu->arch.emulate_ctxt.mode =
3051                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3052                         ? X86EMUL_MODE_REAL : cs_l
3053                         ? X86EMUL_MODE_PROT64 : cs_db
3054                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3055
3056                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3057
3058                 /* Only allow emulation of specific instructions on #UD
3059                  * (namely VMMCALL, sysenter, sysexit, syscall)*/
3060                 c = &vcpu->arch.emulate_ctxt.decode;
3061                 if (emulation_type & EMULTYPE_TRAP_UD) {
3062                         if (!c->twobyte)
3063                                 return EMULATE_FAIL;
3064                         switch (c->b) {
3065                         case 0x01: /* VMMCALL */
3066                                 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3067                                         return EMULATE_FAIL;
3068                                 break;
3069                         case 0x34: /* sysenter */
3070                         case 0x35: /* sysexit */
3071                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3072                                         return EMULATE_FAIL;
3073                                 break;
3074                         case 0x05: /* syscall */
3075                                 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3076                                         return EMULATE_FAIL;
3077                                 break;
3078                         default:
3079                                 return EMULATE_FAIL;
3080                         }
3081
3082                         if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3083                                 return EMULATE_FAIL;
3084                 }
3085
3086                 ++vcpu->stat.insn_emulation;
3087                 if (r)  {
3088                         ++vcpu->stat.insn_emulation_fail;
3089                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3090                                 return EMULATE_DONE;
3091                         return EMULATE_FAIL;
3092                 }
3093         }
3094
3095         if (emulation_type & EMULTYPE_SKIP) {
3096                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3097                 return EMULATE_DONE;
3098         }
3099
3100         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3101         shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3102
3103         if (r == 0)
3104                 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3105
3106         if (vcpu->arch.pio.string)
3107                 return EMULATE_DO_MMIO;
3108
3109         if ((r || vcpu->mmio_is_write) && run) {
3110                 run->exit_reason = KVM_EXIT_MMIO;
3111                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3112                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3113                 run->mmio.len = vcpu->mmio_size;
3114                 run->mmio.is_write = vcpu->mmio_is_write;
3115         }
3116
3117         if (r) {
3118                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3119                         return EMULATE_DONE;
3120                 if (!vcpu->mmio_needed) {
3121                         kvm_report_emulation_failure(vcpu, "mmio");
3122                         return EMULATE_FAIL;
3123                 }
3124                 return EMULATE_DO_MMIO;
3125         }
3126
3127         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3128
3129         if (vcpu->mmio_is_write) {
3130                 vcpu->mmio_needed = 0;
3131                 return EMULATE_DO_MMIO;
3132         }
3133
3134         return EMULATE_DONE;
3135 }
3136 EXPORT_SYMBOL_GPL(emulate_instruction);
3137
3138 static int pio_copy_data(struct kvm_vcpu *vcpu)
3139 {
3140         void *p = vcpu->arch.pio_data;
3141         gva_t q = vcpu->arch.pio.guest_gva;
3142         unsigned bytes;
3143         int ret;
3144
3145         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3146         if (vcpu->arch.pio.in)
3147                 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
3148         else
3149                 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
3150         return ret;
3151 }
3152
3153 int complete_pio(struct kvm_vcpu *vcpu)
3154 {
3155         struct kvm_pio_request *io = &vcpu->arch.pio;
3156         long delta;
3157         int r;
3158         unsigned long val;
3159
3160         if (!io->string) {
3161                 if (io->in) {
3162                         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3163                         memcpy(&val, vcpu->arch.pio_data, io->size);
3164                         kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3165                 }
3166         } else {
3167                 if (io->in) {
3168                         r = pio_copy_data(vcpu);
3169                         if (r)
3170                                 return r;
3171                 }
3172
3173                 delta = 1;
3174                 if (io->rep) {
3175                         delta *= io->cur_count;
3176                         /*
3177                          * The size of the register should really depend on
3178                          * current address size.
3179                          */
3180                         val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3181                         val -= delta;
3182                         kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3183                 }
3184                 if (io->down)
3185                         delta = -delta;
3186                 delta *= io->size;
3187                 if (io->in) {
3188                         val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3189                         val += delta;
3190                         kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3191                 } else {
3192                         val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3193                         val += delta;
3194                         kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3195                 }
3196         }
3197
3198         io->count -= io->cur_count;
3199         io->cur_count = 0;
3200
3201         return 0;
3202 }
3203
3204 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3205 {
3206         /* TODO: String I/O for in kernel device */
3207         int r;
3208
3209         if (vcpu->arch.pio.in)
3210                 r = kvm_io_bus_read(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
3211                                     vcpu->arch.pio.size, pd);
3212         else
3213                 r = kvm_io_bus_write(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
3214                                      vcpu->arch.pio.size, pd);
3215         return r;
3216 }
3217
3218 static int pio_string_write(struct kvm_vcpu *vcpu)
3219 {
3220         struct kvm_pio_request *io = &vcpu->arch.pio;
3221         void *pd = vcpu->arch.pio_data;
3222         int i, r = 0;
3223
3224         for (i = 0; i < io->cur_count; i++) {
3225                 if (kvm_io_bus_write(&vcpu->kvm->pio_bus,
3226                                      io->port, io->size, pd)) {
3227                         r = -EOPNOTSUPP;
3228                         break;
3229                 }
3230                 pd += io->size;
3231         }
3232         return r;
3233 }
3234
3235 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3236 {
3237         unsigned long val;
3238
3239         vcpu->run->exit_reason = KVM_EXIT_IO;
3240         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3241         vcpu->run->io.size = vcpu->arch.pio.size = size;
3242         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3243         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3244         vcpu->run->io.port = vcpu->arch.pio.port = port;
3245         vcpu->arch.pio.in = in;
3246         vcpu->arch.pio.string = 0;
3247         vcpu->arch.pio.down = 0;
3248         vcpu->arch.pio.rep = 0;
3249
3250         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3251                       size, 1);
3252
3253         val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3254         memcpy(vcpu->arch.pio_data, &val, 4);
3255
3256         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3257                 complete_pio(vcpu);
3258                 return 1;
3259         }
3260         return 0;
3261 }
3262 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3263
3264 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3265                   int size, unsigned long count, int down,
3266                   gva_t address, int rep, unsigned port)
3267 {
3268         unsigned now, in_page;
3269         int ret = 0;
3270
3271         vcpu->run->exit_reason = KVM_EXIT_IO;
3272         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3273         vcpu->run->io.size = vcpu->arch.pio.size = size;
3274         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3275         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3276         vcpu->run->io.port = vcpu->arch.pio.port = port;
3277         vcpu->arch.pio.in = in;
3278         vcpu->arch.pio.string = 1;
3279         vcpu->arch.pio.down = down;
3280         vcpu->arch.pio.rep = rep;
3281
3282         trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3283                       size, count);
3284
3285         if (!count) {
3286                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3287                 return 1;
3288         }
3289
3290         if (!down)
3291                 in_page = PAGE_SIZE - offset_in_page(address);
3292         else
3293                 in_page = offset_in_page(address) + size;
3294         now = min(count, (unsigned long)in_page / size);
3295         if (!now)
3296                 now = 1;
3297         if (down) {
3298                 /*
3299                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
3300                  */
3301                 pr_unimpl(vcpu, "guest string pio down\n");
3302                 kvm_inject_gp(vcpu, 0);
3303                 return 1;
3304         }
3305         vcpu->run->io.count = now;
3306         vcpu->arch.pio.cur_count = now;
3307
3308         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3309                 kvm_x86_ops->skip_emulated_instruction(vcpu);
3310
3311         vcpu->arch.pio.guest_gva = address;
3312
3313         if (!vcpu->arch.pio.in) {
3314                 /* string PIO write */
3315                 ret = pio_copy_data(vcpu);
3316                 if (ret == X86EMUL_PROPAGATE_FAULT) {
3317                         kvm_inject_gp(vcpu, 0);
3318                         return 1;
3319                 }
3320                 if (ret == 0 && !pio_string_write(vcpu)) {
3321                         complete_pio(vcpu);
3322                         if (vcpu->arch.pio.count == 0)
3323                                 ret = 1;
3324                 }
3325         }
3326         /* no string PIO read support yet */
3327
3328         return ret;
3329 }
3330 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3331
3332 static void bounce_off(void *info)
3333 {
3334         /* nothing */
3335 }
3336
3337 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3338                                      void *data)
3339 {
3340         struct cpufreq_freqs *freq = data;
3341         struct kvm *kvm;
3342         struct kvm_vcpu *vcpu;
3343         int i, send_ipi = 0;
3344
3345         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3346                 return 0;
3347         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3348                 return 0;
3349         per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3350
3351         spin_lock(&kvm_lock);
3352         list_for_each_entry(kvm, &vm_list, vm_list) {
3353                 kvm_for_each_vcpu(i, vcpu, kvm) {
3354                         if (vcpu->cpu != freq->cpu)
3355                                 continue;
3356                         if (!kvm_request_guest_time_update(vcpu))
3357                                 continue;
3358                         if (vcpu->cpu != smp_processor_id())
3359                                 send_ipi++;
3360                 }
3361         }
3362         spin_unlock(&kvm_lock);
3363
3364         if (freq->old < freq->new && send_ipi) {
3365                 /*
3366                  * We upscale the frequency.  Must make the guest
3367                  * doesn't see old kvmclock values while running with
3368                  * the new frequency, otherwise we risk the guest sees
3369                  * time go backwards.
3370                  *
3371                  * In case we update the frequency for another cpu
3372                  * (which might be in guest context) send an interrupt
3373                  * to kick the cpu out of guest context.  Next time
3374                  * guest context is entered kvmclock will be updated,
3375                  * so the guest will not see stale values.
3376                  */
3377                 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3378         }
3379         return 0;
3380 }
3381
3382 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3383         .notifier_call  = kvmclock_cpufreq_notifier
3384 };
3385
3386 static void kvm_timer_init(void)
3387 {
3388         int cpu;
3389
3390         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3391                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3392                                           CPUFREQ_TRANSITION_NOTIFIER);
3393                 for_each_online_cpu(cpu) {
3394                         unsigned long khz = cpufreq_get(cpu);
3395                         if (!khz)
3396                                 khz = tsc_khz;
3397                         per_cpu(cpu_tsc_khz, cpu) = khz;
3398                 }
3399         } else {
3400                 for_each_possible_cpu(cpu)
3401                         per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3402         }
3403 }
3404
3405 int kvm_arch_init(void *opaque)
3406 {
3407         int r;
3408         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3409
3410         if (kvm_x86_ops) {
3411                 printk(KERN_ERR "kvm: already loaded the other module\n");
3412                 r = -EEXIST;
3413                 goto out;
3414         }
3415
3416         if (!ops->cpu_has_kvm_support()) {
3417                 printk(KERN_ERR "kvm: no hardware support\n");
3418                 r = -EOPNOTSUPP;
3419                 goto out;
3420         }
3421         if (ops->disabled_by_bios()) {
3422                 printk(KERN_ERR "kvm: disabled by bios\n");
3423                 r = -EOPNOTSUPP;
3424                 goto out;
3425         }
3426
3427         r = kvm_mmu_module_init();
3428         if (r)
3429                 goto out;
3430
3431         kvm_init_msr_list();
3432
3433         kvm_x86_ops = ops;
3434         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3435         kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3436         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3437                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
3438
3439         kvm_timer_init();
3440
3441         return 0;
3442
3443 out:
3444         return r;
3445 }
3446
3447 void kvm_arch_exit(void)
3448 {
3449         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3450                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3451                                             CPUFREQ_TRANSITION_NOTIFIER);
3452         kvm_x86_ops = NULL;
3453         kvm_mmu_module_exit();
3454 }
3455
3456 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3457 {
3458         ++vcpu->stat.halt_exits;
3459         if (irqchip_in_kernel(vcpu->kvm)) {
3460                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3461                 return 1;
3462         } else {
3463                 vcpu->run->exit_reason = KVM_EXIT_HLT;
3464                 return 0;
3465         }
3466 }
3467 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3468
3469 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3470                            unsigned long a1)
3471 {
3472         if (is_long_mode(vcpu))
3473                 return a0;
3474         else
3475                 return a0 | ((gpa_t)a1 << 32);
3476 }
3477
3478 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3479 {
3480         unsigned long nr, a0, a1, a2, a3, ret;
3481         int r = 1;
3482
3483         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3484         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3485         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3486         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3487         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3488
3489         trace_kvm_hypercall(nr, a0, a1, a2, a3);
3490
3491         if (!is_long_mode(vcpu)) {
3492                 nr &= 0xFFFFFFFF;
3493                 a0 &= 0xFFFFFFFF;
3494                 a1 &= 0xFFFFFFFF;
3495                 a2 &= 0xFFFFFFFF;
3496                 a3 &= 0xFFFFFFFF;
3497         }
3498
3499         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3500                 ret = -KVM_EPERM;
3501                 goto out;
3502         }
3503
3504         switch (nr) {
3505         case KVM_HC_VAPIC_POLL_IRQ:
3506                 ret = 0;
3507                 break;
3508         case KVM_HC_MMU_OP:
3509                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3510                 break;
3511         default:
3512                 ret = -KVM_ENOSYS;
3513                 break;
3514         }
3515 out:
3516         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3517         ++vcpu->stat.hypercalls;
3518         return r;
3519 }
3520 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3521
3522 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3523 {
3524         char instruction[3];
3525         int ret = 0;
3526         unsigned long rip = kvm_rip_read(vcpu);
3527
3528
3529         /*
3530          * Blow out the MMU to ensure that no other VCPU has an active mapping
3531          * to ensure that the updated hypercall appears atomically across all
3532          * VCPUs.
3533          */
3534         kvm_mmu_zap_all(vcpu->kvm);
3535
3536         kvm_x86_ops->patch_hypercall(vcpu, instruction);
3537         if (emulator_write_emulated(rip, instruction, 3, vcpu)
3538             != X86EMUL_CONTINUE)
3539                 ret = -EFAULT;
3540
3541         return ret;
3542 }
3543
3544 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3545 {
3546         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3547 }
3548
3549 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3550 {
3551         struct descriptor_table dt = { limit, base };
3552
3553         kvm_x86_ops->set_gdt(vcpu, &dt);
3554 }
3555
3556 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3557 {
3558         struct descriptor_table dt = { limit, base };
3559
3560         kvm_x86_ops->set_idt(vcpu, &dt);
3561 }
3562
3563 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3564                    unsigned long *rflags)
3565 {
3566         kvm_lmsw(vcpu, msw);
3567         *rflags = kvm_get_rflags(vcpu);
3568 }
3569
3570 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3571 {
3572         unsigned long value;
3573
3574         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3575         switch (cr) {
3576         case 0:
3577                 value = vcpu->arch.cr0;
3578                 break;
3579         case 2:
3580                 value = vcpu->arch.cr2;
3581                 break;
3582         case 3:
3583                 value = vcpu->arch.cr3;
3584                 break;
3585         case 4:
3586                 value = vcpu->arch.cr4;
3587                 break;
3588         case 8:
3589                 value = kvm_get_cr8(vcpu);
3590                 break;
3591         default:
3592                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3593                 return 0;
3594         }
3595
3596         return value;
3597 }
3598
3599 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3600                      unsigned long *rflags)
3601 {
3602         switch (cr) {
3603         case 0:
3604                 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3605                 *rflags = kvm_get_rflags(vcpu);
3606                 break;
3607         case 2:
3608                 vcpu->arch.cr2 = val;
3609                 break;
3610         case 3:
3611                 kvm_set_cr3(vcpu, val);
3612                 break;
3613         case 4:
3614                 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3615                 break;
3616         case 8:
3617                 kvm_set_cr8(vcpu, val & 0xfUL);
3618                 break;
3619         default:
3620                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3621         }
3622 }
3623
3624 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3625 {
3626         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3627         int j, nent = vcpu->arch.cpuid_nent;
3628
3629         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3630         /* when no next entry is found, the current entry[i] is reselected */
3631         for (j = i + 1; ; j = (j + 1) % nent) {
3632                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3633                 if (ej->function == e->function) {
3634                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3635                         return j;
3636                 }
3637         }
3638         return 0; /* silence gcc, even though control never reaches here */
3639 }
3640
3641 /* find an entry with matching function, matching index (if needed), and that
3642  * should be read next (if it's stateful) */
3643 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3644         u32 function, u32 index)
3645 {
3646         if (e->function != function)
3647                 return 0;
3648         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3649                 return 0;
3650         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3651             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3652                 return 0;
3653         return 1;
3654 }
3655
3656 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3657                                               u32 function, u32 index)
3658 {
3659         int i;
3660         struct kvm_cpuid_entry2 *best = NULL;
3661
3662         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3663                 struct kvm_cpuid_entry2 *e;
3664
3665                 e = &vcpu->arch.cpuid_entries[i];
3666                 if (is_matching_cpuid_entry(e, function, index)) {
3667                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3668                                 move_to_next_stateful_cpuid_entry(vcpu, i);
3669                         best = e;
3670                         break;
3671                 }
3672                 /*
3673                  * Both basic or both extended?
3674                  */
3675                 if (((e->function ^ function) & 0x80000000) == 0)
3676                         if (!best || e->function > best->function)
3677                                 best = e;
3678         }
3679         return best;
3680 }
3681
3682 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3683 {
3684         struct kvm_cpuid_entry2 *best;
3685
3686         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3687         if (best)
3688                 return best->eax & 0xff;
3689         return 36;
3690 }
3691
3692 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3693 {
3694         u32 function, index;
3695         struct kvm_cpuid_entry2 *best;
3696
3697         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3698         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3699         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3700         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3701         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3702         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3703         best = kvm_find_cpuid_entry(vcpu, function, index);
3704         if (best) {
3705                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3706                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3707                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3708                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3709         }
3710         kvm_x86_ops->skip_emulated_instruction(vcpu);
3711         trace_kvm_cpuid(function,
3712                         kvm_register_read(vcpu, VCPU_REGS_RAX),
3713                         kvm_register_read(vcpu, VCPU_REGS_RBX),
3714                         kvm_register_read(vcpu, VCPU_REGS_RCX),
3715                         kvm_register_read(vcpu, VCPU_REGS_RDX));
3716 }
3717 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3718
3719 /*
3720  * Check if userspace requested an interrupt window, and that the
3721  * interrupt window is open.
3722  *
3723  * No need to exit to userspace if we already have an interrupt queued.
3724  */
3725 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
3726 {
3727         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3728                 vcpu->run->request_interrupt_window &&
3729                 kvm_arch_interrupt_allowed(vcpu));
3730 }
3731
3732 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
3733 {
3734         struct kvm_run *kvm_run = vcpu->run;
3735
3736         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3737         kvm_run->cr8 = kvm_get_cr8(vcpu);
3738         kvm_run->apic_base = kvm_get_apic_base(vcpu);
3739         if (irqchip_in_kernel(vcpu->kvm))
3740                 kvm_run->ready_for_interrupt_injection = 1;
3741         else
3742                 kvm_run->ready_for_interrupt_injection =
3743                         kvm_arch_interrupt_allowed(vcpu) &&
3744                         !kvm_cpu_has_interrupt(vcpu) &&
3745                         !kvm_event_needs_reinjection(vcpu);
3746 }
3747
3748 static void vapic_enter(struct kvm_vcpu *vcpu)
3749 {
3750         struct kvm_lapic *apic = vcpu->arch.apic;
3751         struct page *page;
3752
3753         if (!apic || !apic->vapic_addr)
3754                 return;
3755
3756         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3757
3758         vcpu->arch.apic->vapic_page = page;
3759 }
3760
3761 static void vapic_exit(struct kvm_vcpu *vcpu)
3762 {
3763         struct kvm_lapic *apic = vcpu->arch.apic;
3764
3765         if (!apic || !apic->vapic_addr)
3766                 return;
3767
3768         down_read(&vcpu->kvm->slots_lock);
3769         kvm_release_page_dirty(apic->vapic_page);
3770         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3771         up_read(&vcpu->kvm->slots_lock);
3772 }
3773
3774 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3775 {
3776         int max_irr, tpr;
3777
3778         if (!kvm_x86_ops->update_cr8_intercept)
3779                 return;
3780
3781         if (!vcpu->arch.apic)
3782                 return;
3783
3784         if (!vcpu->arch.apic->vapic_addr)
3785                 max_irr = kvm_lapic_find_highest_irr(vcpu);
3786         else
3787                 max_irr = -1;
3788
3789         if (max_irr != -1)
3790                 max_irr >>= 4;
3791
3792         tpr = kvm_lapic_get_cr8(vcpu);
3793
3794         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3795 }
3796
3797 static void inject_pending_event(struct kvm_vcpu *vcpu)
3798 {
3799         /* try to reinject previous events if any */
3800         if (vcpu->arch.exception.pending) {
3801                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3802                                           vcpu->arch.exception.has_error_code,
3803                                           vcpu->arch.exception.error_code);
3804                 return;
3805         }
3806
3807         if (vcpu->arch.nmi_injected) {
3808                 kvm_x86_ops->set_nmi(vcpu);
3809                 return;
3810         }
3811
3812         if (vcpu->arch.interrupt.pending) {
3813                 kvm_x86_ops->set_irq(vcpu);
3814                 return;
3815         }
3816
3817         /* try to inject new event if pending */
3818         if (vcpu->arch.nmi_pending) {
3819                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3820                         vcpu->arch.nmi_pending = false;
3821                         vcpu->arch.nmi_injected = true;
3822                         kvm_x86_ops->set_nmi(vcpu);
3823                 }
3824         } else if (kvm_cpu_has_interrupt(vcpu)) {
3825                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3826                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3827                                             false);
3828                         kvm_x86_ops->set_irq(vcpu);
3829                 }
3830         }
3831 }
3832
3833 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
3834 {
3835         int r;
3836         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3837                 vcpu->run->request_interrupt_window;
3838
3839         if (vcpu->requests)
3840                 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3841                         kvm_mmu_unload(vcpu);
3842
3843         r = kvm_mmu_reload(vcpu);
3844         if (unlikely(r))
3845                 goto out;
3846
3847         if (vcpu->requests) {
3848                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3849                         __kvm_migrate_timers(vcpu);
3850                 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3851                         kvm_write_guest_time(vcpu);
3852                 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3853                         kvm_mmu_sync_roots(vcpu);
3854                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3855                         kvm_x86_ops->tlb_flush(vcpu);
3856                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3857                                        &vcpu->requests)) {
3858                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
3859                         r = 0;
3860                         goto out;
3861                 }
3862                 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3863                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3864                         r = 0;
3865                         goto out;
3866                 }
3867         }
3868
3869         preempt_disable();
3870
3871         kvm_x86_ops->prepare_guest_switch(vcpu);
3872         kvm_load_guest_fpu(vcpu);
3873
3874         local_irq_disable();
3875
3876         clear_bit(KVM_REQ_KICK, &vcpu->requests);
3877         smp_mb__after_clear_bit();
3878
3879         if (vcpu->requests || need_resched() || signal_pending(current)) {
3880                 set_bit(KVM_REQ_KICK, &vcpu->requests);
3881                 local_irq_enable();
3882                 preempt_enable();
3883                 r = 1;
3884                 goto out;
3885         }
3886
3887         inject_pending_event(vcpu);
3888
3889         /* enable NMI/IRQ window open exits if needed */
3890         if (vcpu->arch.nmi_pending)
3891                 kvm_x86_ops->enable_nmi_window(vcpu);
3892         else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3893                 kvm_x86_ops->enable_irq_window(vcpu);
3894
3895         if (kvm_lapic_enabled(vcpu)) {
3896                 update_cr8_intercept(vcpu);
3897                 kvm_lapic_sync_to_vapic(vcpu);
3898         }
3899
3900         up_read(&vcpu->kvm->slots_lock);
3901
3902         kvm_guest_enter();
3903
3904         if (unlikely(vcpu->arch.switch_db_regs)) {
3905                 set_debugreg(0, 7);
3906                 set_debugreg(vcpu->arch.eff_db[0], 0);
3907                 set_debugreg(vcpu->arch.eff_db[1], 1);
3908                 set_debugreg(vcpu->arch.eff_db[2], 2);
3909                 set_debugreg(vcpu->arch.eff_db[3], 3);
3910         }
3911
3912         trace_kvm_entry(vcpu->vcpu_id);
3913         kvm_x86_ops->run(vcpu);
3914
3915         if (unlikely(vcpu->arch.switch_db_regs || test_thread_flag(TIF_DEBUG))) {
3916                 set_debugreg(current->thread.debugreg0, 0);
3917                 set_debugreg(current->thread.debugreg1, 1);
3918                 set_debugreg(current->thread.debugreg2, 2);
3919                 set_debugreg(current->thread.debugreg3, 3);
3920                 set_debugreg(current->thread.debugreg6, 6);
3921                 set_debugreg(current->thread.debugreg7, 7);
3922         }
3923
3924         set_bit(KVM_REQ_KICK, &vcpu->requests);
3925         local_irq_enable();
3926
3927         ++vcpu->stat.exits;
3928
3929         /*
3930          * We must have an instruction between local_irq_enable() and
3931          * kvm_guest_exit(), so the timer interrupt isn't delayed by
3932          * the interrupt shadow.  The stat.exits increment will do nicely.
3933          * But we need to prevent reordering, hence this barrier():
3934          */
3935         barrier();
3936
3937         kvm_guest_exit();
3938
3939         preempt_enable();
3940
3941         down_read(&vcpu->kvm->slots_lock);
3942
3943         /*
3944          * Profile KVM exit RIPs:
3945          */
3946         if (unlikely(prof_on == KVM_PROFILING)) {
3947                 unsigned long rip = kvm_rip_read(vcpu);
3948                 profile_hit(KVM_PROFILING, (void *)rip);
3949         }
3950
3951
3952         kvm_lapic_sync_from_vapic(vcpu);
3953
3954         r = kvm_x86_ops->handle_exit(vcpu);
3955 out:
3956         return r;
3957 }
3958
3959
3960 static int __vcpu_run(struct kvm_vcpu *vcpu)
3961 {
3962         int r;
3963
3964         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3965                 pr_debug("vcpu %d received sipi with vector # %x\n",
3966                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
3967                 kvm_lapic_reset(vcpu);
3968                 r = kvm_arch_vcpu_reset(vcpu);
3969                 if (r)
3970                         return r;
3971                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3972         }
3973
3974         down_read(&vcpu->kvm->slots_lock);
3975         vapic_enter(vcpu);
3976
3977         r = 1;
3978         while (r > 0) {
3979                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3980                         r = vcpu_enter_guest(vcpu);
3981                 else {
3982                         up_read(&vcpu->kvm->slots_lock);
3983                         kvm_vcpu_block(vcpu);
3984                         down_read(&vcpu->kvm->slots_lock);
3985                         if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3986                         {
3987                                 switch(vcpu->arch.mp_state) {
3988                                 case KVM_MP_STATE_HALTED:
3989                                         vcpu->arch.mp_state =
3990                                                 KVM_MP_STATE_RUNNABLE;
3991                                 case KVM_MP_STATE_RUNNABLE:
3992                                         break;
3993                                 case KVM_MP_STATE_SIPI_RECEIVED:
3994                                 default:
3995                                         r = -EINTR;
3996                                         break;
3997                                 }
3998                         }
3999                 }
4000
4001                 if (r <= 0)
4002                         break;
4003
4004                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4005                 if (kvm_cpu_has_pending_timer(vcpu))
4006                         kvm_inject_pending_timer_irqs(vcpu);
4007
4008                 if (dm_request_for_irq_injection(vcpu)) {
4009                         r = -EINTR;
4010                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4011                         ++vcpu->stat.request_irq_exits;
4012                 }
4013                 if (signal_pending(current)) {
4014                         r = -EINTR;
4015                         vcpu->run->exit_reason = KVM_EXIT_INTR;
4016                         ++vcpu->stat.signal_exits;
4017                 }
4018                 if (need_resched()) {
4019                         up_read(&vcpu->kvm->slots_lock);
4020                         kvm_resched(vcpu);
4021                         down_read(&vcpu->kvm->slots_lock);
4022                 }
4023         }
4024
4025         up_read(&vcpu->kvm->slots_lock);
4026         post_kvm_run_save(vcpu);
4027
4028         vapic_exit(vcpu);
4029
4030         return r;
4031 }
4032
4033 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4034 {
4035         int r;
4036         sigset_t sigsaved;
4037
4038         vcpu_load(vcpu);
4039
4040         if (vcpu->sigset_active)
4041                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4042
4043         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4044                 kvm_vcpu_block(vcpu);
4045                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4046                 r = -EAGAIN;
4047                 goto out;
4048         }
4049
4050         /* re-sync apic's tpr */
4051         if (!irqchip_in_kernel(vcpu->kvm))
4052                 kvm_set_cr8(vcpu, kvm_run->cr8);
4053
4054         if (vcpu->arch.pio.cur_count) {
4055                 r = complete_pio(vcpu);
4056                 if (r)
4057                         goto out;
4058         }
4059         if (vcpu->mmio_needed) {
4060                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4061                 vcpu->mmio_read_completed = 1;
4062                 vcpu->mmio_needed = 0;
4063
4064                 down_read(&vcpu->kvm->slots_lock);
4065                 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4066                                         EMULTYPE_NO_DECODE);
4067                 up_read(&vcpu->kvm->slots_lock);
4068                 if (r == EMULATE_DO_MMIO) {
4069                         /*
4070                          * Read-modify-write.  Back to userspace.
4071                          */
4072                         r = 0;
4073                         goto out;
4074                 }
4075         }
4076         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4077                 kvm_register_write(vcpu, VCPU_REGS_RAX,
4078                                      kvm_run->hypercall.ret);
4079
4080         r = __vcpu_run(vcpu);
4081
4082 out:
4083         if (vcpu->sigset_active)
4084                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4085
4086         vcpu_put(vcpu);
4087         return r;
4088 }
4089
4090 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4091 {
4092         vcpu_load(vcpu);
4093
4094         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4095         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4096         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4097         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4098         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4099         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4100         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4101         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4102 #ifdef CONFIG_X86_64
4103         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4104         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4105         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4106         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4107         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4108         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4109         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4110         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4111 #endif
4112
4113         regs->rip = kvm_rip_read(vcpu);
4114         regs->rflags = kvm_get_rflags(vcpu);
4115
4116         vcpu_put(vcpu);
4117
4118         return 0;
4119 }
4120
4121 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4122 {
4123         vcpu_load(vcpu);
4124
4125         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4126         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4127         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4128         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4129         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4130         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4131         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4132         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4133 #ifdef CONFIG_X86_64
4134         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4135         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4136         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4137         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4138         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4139         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4140         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4141         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4142 #endif
4143
4144         kvm_rip_write(vcpu, regs->rip);
4145         kvm_set_rflags(vcpu, regs->rflags);
4146
4147         vcpu->arch.exception.pending = false;
4148
4149         vcpu_put(vcpu);
4150
4151         return 0;
4152 }
4153
4154 void kvm_get_segment(struct kvm_vcpu *vcpu,
4155                      struct kvm_segment *var, int seg)
4156 {
4157         kvm_x86_ops->get_segment(vcpu, var, seg);
4158 }
4159
4160 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4161 {
4162         struct kvm_segment cs;
4163
4164         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4165         *db = cs.db;
4166         *l = cs.l;
4167 }
4168 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4169
4170 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4171                                   struct kvm_sregs *sregs)
4172 {
4173         struct descriptor_table dt;
4174
4175         vcpu_load(vcpu);
4176
4177         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4178         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4179         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4180         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4181         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4182         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4183
4184         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4185         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4186
4187         kvm_x86_ops->get_idt(vcpu, &dt);
4188         sregs->idt.limit = dt.limit;
4189         sregs->idt.base = dt.base;
4190         kvm_x86_ops->get_gdt(vcpu, &dt);
4191         sregs->gdt.limit = dt.limit;
4192         sregs->gdt.base = dt.base;
4193
4194         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4195         sregs->cr0 = vcpu->arch.cr0;
4196         sregs->cr2 = vcpu->arch.cr2;
4197         sregs->cr3 = vcpu->arch.cr3;
4198         sregs->cr4 = vcpu->arch.cr4;
4199         sregs->cr8 = kvm_get_cr8(vcpu);
4200         sregs->efer = vcpu->arch.shadow_efer;
4201         sregs->apic_base = kvm_get_apic_base(vcpu);
4202
4203         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4204
4205         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4206                 set_bit(vcpu->arch.interrupt.nr,
4207                         (unsigned long *)sregs->interrupt_bitmap);
4208
4209         vcpu_put(vcpu);
4210
4211         return 0;
4212 }
4213
4214 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4215                                     struct kvm_mp_state *mp_state)
4216 {
4217         vcpu_load(vcpu);
4218         mp_state->mp_state = vcpu->arch.mp_state;
4219         vcpu_put(vcpu);
4220         return 0;
4221 }
4222
4223 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4224                                     struct kvm_mp_state *mp_state)
4225 {
4226         vcpu_load(vcpu);
4227         vcpu->arch.mp_state = mp_state->mp_state;
4228         vcpu_put(vcpu);
4229         return 0;
4230 }
4231
4232 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4233                         struct kvm_segment *var, int seg)
4234 {
4235         kvm_x86_ops->set_segment(vcpu, var, seg);
4236 }
4237
4238 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4239                                    struct kvm_segment *kvm_desct)
4240 {
4241         kvm_desct->base = get_desc_base(seg_desc);
4242         kvm_desct->limit = get_desc_limit(seg_desc);
4243         if (seg_desc->g) {
4244                 kvm_desct->limit <<= 12;
4245                 kvm_desct->limit |= 0xfff;
4246         }
4247         kvm_desct->selector = selector;
4248         kvm_desct->type = seg_desc->type;
4249         kvm_desct->present = seg_desc->p;
4250         kvm_desct->dpl = seg_desc->dpl;
4251         kvm_desct->db = seg_desc->d;
4252         kvm_desct->s = seg_desc->s;
4253         kvm_desct->l = seg_desc->l;
4254         kvm_desct->g = seg_desc->g;
4255         kvm_desct->avl = seg_desc->avl;
4256         if (!selector)
4257                 kvm_desct->unusable = 1;
4258         else
4259                 kvm_desct->unusable = 0;
4260         kvm_desct->padding = 0;
4261 }
4262
4263 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4264                                           u16 selector,
4265                                           struct descriptor_table *dtable)
4266 {
4267         if (selector & 1 << 2) {
4268                 struct kvm_segment kvm_seg;
4269
4270                 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4271
4272                 if (kvm_seg.unusable)
4273                         dtable->limit = 0;
4274                 else
4275                         dtable->limit = kvm_seg.limit;
4276                 dtable->base = kvm_seg.base;
4277         }
4278         else
4279                 kvm_x86_ops->get_gdt(vcpu, dtable);
4280 }
4281
4282 /* allowed just for 8 bytes segments */
4283 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4284                                          struct desc_struct *seg_desc)
4285 {
4286         struct descriptor_table dtable;
4287         u16 index = selector >> 3;
4288
4289         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4290
4291         if (dtable.limit < index * 8 + 7) {
4292                 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4293                 return 1;
4294         }
4295         return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4296 }
4297
4298 /* allowed just for 8 bytes segments */
4299 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4300                                          struct desc_struct *seg_desc)
4301 {
4302         struct descriptor_table dtable;
4303         u16 index = selector >> 3;
4304
4305         get_segment_descriptor_dtable(vcpu, selector, &dtable);
4306
4307         if (dtable.limit < index * 8 + 7)
4308                 return 1;
4309         return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4310 }
4311
4312 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4313                              struct desc_struct *seg_desc)
4314 {
4315         u32 base_addr = get_desc_base(seg_desc);
4316
4317         return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4318 }
4319
4320 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4321 {
4322         struct kvm_segment kvm_seg;
4323
4324         kvm_get_segment(vcpu, &kvm_seg, seg);
4325         return kvm_seg.selector;
4326 }
4327
4328 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4329                                                 u16 selector,
4330                                                 struct kvm_segment *kvm_seg)
4331 {
4332         struct desc_struct seg_desc;
4333
4334         if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4335                 return 1;
4336         seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4337         return 0;
4338 }
4339
4340 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4341 {
4342         struct kvm_segment segvar = {
4343                 .base = selector << 4,
4344                 .limit = 0xffff,
4345                 .selector = selector,
4346                 .type = 3,
4347                 .present = 1,
4348                 .dpl = 3,
4349                 .db = 0,
4350                 .s = 1,
4351                 .l = 0,
4352                 .g = 0,
4353                 .avl = 0,
4354                 .unusable = 0,
4355         };
4356         kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4357         return 0;
4358 }
4359
4360 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4361 {
4362         return (seg != VCPU_SREG_LDTR) &&
4363                 (seg != VCPU_SREG_TR) &&
4364                 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4365 }
4366
4367 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4368                                 int type_bits, int seg)
4369 {
4370         struct kvm_segment kvm_seg;
4371
4372         if (is_vm86_segment(vcpu, seg) || !(vcpu->arch.cr0 & X86_CR0_PE))
4373                 return kvm_load_realmode_segment(vcpu, selector, seg);
4374         if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4375                 return 1;
4376         kvm_seg.type |= type_bits;
4377
4378         if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4379             seg != VCPU_SREG_LDTR)
4380                 if (!kvm_seg.s)
4381                         kvm_seg.unusable = 1;
4382
4383         kvm_set_segment(vcpu, &kvm_seg, seg);
4384         return 0;
4385 }
4386
4387 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4388                                 struct tss_segment_32 *tss)
4389 {
4390         tss->cr3 = vcpu->arch.cr3;
4391         tss->eip = kvm_rip_read(vcpu);
4392         tss->eflags = kvm_get_rflags(vcpu);
4393         tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4394         tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4395         tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4396         tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4397         tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4398         tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4399         tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4400         tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4401         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4402         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4403         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4404         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4405         tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4406         tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4407         tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4408 }
4409
4410 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4411                                   struct tss_segment_32 *tss)
4412 {
4413         kvm_set_cr3(vcpu, tss->cr3);
4414
4415         kvm_rip_write(vcpu, tss->eip);
4416         kvm_set_rflags(vcpu, tss->eflags | 2);
4417
4418         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4419         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4420         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4421         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4422         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4423         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4424         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4425         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4426
4427         if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4428                 return 1;
4429
4430         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4431                 return 1;
4432
4433         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4434                 return 1;
4435
4436         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4437                 return 1;
4438
4439         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4440                 return 1;
4441
4442         if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4443                 return 1;
4444
4445         if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4446                 return 1;
4447         return 0;
4448 }
4449
4450 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4451                                 struct tss_segment_16 *tss)
4452 {
4453         tss->ip = kvm_rip_read(vcpu);
4454         tss->flag = kvm_get_rflags(vcpu);
4455         tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4456         tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4457         tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4458         tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4459         tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4460         tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4461         tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4462         tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4463
4464         tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4465         tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4466         tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4467         tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4468         tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4469 }
4470
4471 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4472                                  struct tss_segment_16 *tss)
4473 {
4474         kvm_rip_write(vcpu, tss->ip);
4475         kvm_set_rflags(vcpu, tss->flag | 2);
4476         kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4477         kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4478         kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4479         kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4480         kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4481         kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4482         kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4483         kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4484
4485         if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4486                 return 1;
4487
4488         if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4489                 return 1;
4490
4491         if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4492                 return 1;
4493
4494         if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4495                 return 1;
4496
4497         if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4498                 return 1;
4499         return 0;
4500 }
4501
4502 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4503                               u16 old_tss_sel, u32 old_tss_base,
4504                               struct desc_struct *nseg_desc)
4505 {
4506         struct tss_segment_16 tss_segment_16;
4507         int ret = 0;
4508
4509         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4510                            sizeof tss_segment_16))
4511                 goto out;
4512
4513         save_state_to_tss16(vcpu, &tss_segment_16);
4514
4515         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4516                             sizeof tss_segment_16))
4517                 goto out;
4518
4519         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4520                            &tss_segment_16, sizeof tss_segment_16))
4521                 goto out;
4522
4523         if (old_tss_sel != 0xffff) {
4524                 tss_segment_16.prev_task_link = old_tss_sel;
4525
4526                 if (kvm_write_guest(vcpu->kvm,
4527                                     get_tss_base_addr(vcpu, nseg_desc),
4528                                     &tss_segment_16.prev_task_link,
4529                                     sizeof tss_segment_16.prev_task_link))
4530                         goto out;
4531         }
4532
4533         if (load_state_from_tss16(vcpu, &tss_segment_16))
4534                 goto out;
4535
4536         ret = 1;
4537 out:
4538         return ret;
4539 }
4540
4541 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4542                        u16 old_tss_sel, u32 old_tss_base,
4543                        struct desc_struct *nseg_desc)
4544 {
4545         struct tss_segment_32 tss_segment_32;
4546         int ret = 0;
4547
4548         if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4549                            sizeof tss_segment_32))
4550                 goto out;
4551
4552         save_state_to_tss32(vcpu, &tss_segment_32);
4553
4554         if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4555                             sizeof tss_segment_32))
4556                 goto out;
4557
4558         if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4559                            &tss_segment_32, sizeof tss_segment_32))
4560                 goto out;
4561
4562         if (old_tss_sel != 0xffff) {
4563                 tss_segment_32.prev_task_link = old_tss_sel;
4564
4565                 if (kvm_write_guest(vcpu->kvm,
4566                                     get_tss_base_addr(vcpu, nseg_desc),
4567                                     &tss_segment_32.prev_task_link,
4568                                     sizeof tss_segment_32.prev_task_link))
4569                         goto out;
4570         }
4571
4572         if (load_state_from_tss32(vcpu, &tss_segment_32))
4573                 goto out;
4574
4575         ret = 1;
4576 out:
4577         return ret;
4578 }
4579
4580 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4581 {
4582         struct kvm_segment tr_seg;
4583         struct desc_struct cseg_desc;
4584         struct desc_struct nseg_desc;
4585         int ret = 0;
4586         u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4587         u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4588
4589         old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4590
4591         /* FIXME: Handle errors. Failure to read either TSS or their
4592          * descriptors should generate a pagefault.
4593          */
4594         if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4595                 goto out;
4596
4597         if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4598                 goto out;
4599
4600         if (reason != TASK_SWITCH_IRET) {
4601                 int cpl;
4602
4603                 cpl = kvm_x86_ops->get_cpl(vcpu);
4604                 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4605                         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4606                         return 1;
4607                 }
4608         }
4609
4610         if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4611                 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4612                 return 1;
4613         }
4614
4615         if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4616                 cseg_desc.type &= ~(1 << 1); //clear the B flag
4617                 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4618         }
4619
4620         if (reason == TASK_SWITCH_IRET) {
4621                 u32 eflags = kvm_get_rflags(vcpu);
4622                 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4623         }
4624
4625         /* set back link to prev task only if NT bit is set in eflags
4626            note that old_tss_sel is not used afetr this point */
4627         if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4628                 old_tss_sel = 0xffff;
4629
4630         if (nseg_desc.type & 8)
4631                 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4632                                          old_tss_base, &nseg_desc);
4633         else
4634                 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4635                                          old_tss_base, &nseg_desc);
4636
4637         if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4638                 u32 eflags = kvm_get_rflags(vcpu);
4639                 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4640         }
4641
4642         if (reason != TASK_SWITCH_IRET) {
4643                 nseg_desc.type |= (1 << 1);
4644                 save_guest_segment_descriptor(vcpu, tss_selector,
4645                                               &nseg_desc);
4646         }
4647
4648         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4649         seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4650         tr_seg.type = 11;
4651         kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4652 out:
4653         return ret;
4654 }
4655 EXPORT_SYMBOL_GPL(kvm_task_switch);
4656
4657 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4658                                   struct kvm_sregs *sregs)
4659 {
4660         int mmu_reset_needed = 0;
4661         int pending_vec, max_bits;
4662         struct descriptor_table dt;
4663
4664         vcpu_load(vcpu);
4665
4666         dt.limit = sregs->idt.limit;
4667         dt.base = sregs->idt.base;
4668         kvm_x86_ops->set_idt(vcpu, &dt);
4669         dt.limit = sregs->gdt.limit;
4670         dt.base = sregs->gdt.base;
4671         kvm_x86_ops->set_gdt(vcpu, &dt);
4672
4673         vcpu->arch.cr2 = sregs->cr2;
4674         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4675         vcpu->arch.cr3 = sregs->cr3;
4676
4677         kvm_set_cr8(vcpu, sregs->cr8);
4678
4679         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4680         kvm_x86_ops->set_efer(vcpu, sregs->efer);
4681         kvm_set_apic_base(vcpu, sregs->apic_base);
4682
4683         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4684
4685         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4686         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4687         vcpu->arch.cr0 = sregs->cr0;
4688
4689         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4690         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4691         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
4692                 load_pdptrs(vcpu, vcpu->arch.cr3);
4693                 mmu_reset_needed = 1;
4694         }
4695
4696         if (mmu_reset_needed)
4697                 kvm_mmu_reset_context(vcpu);
4698
4699         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4700         pending_vec = find_first_bit(
4701                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4702         if (pending_vec < max_bits) {
4703                 kvm_queue_interrupt(vcpu, pending_vec, false);
4704                 pr_debug("Set back pending irq %d\n", pending_vec);
4705                 if (irqchip_in_kernel(vcpu->kvm))
4706                         kvm_pic_clear_isr_ack(vcpu->kvm);
4707         }
4708
4709         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4710         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4711         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4712         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4713         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4714         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4715
4716         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4717         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4718
4719         update_cr8_intercept(vcpu);
4720
4721         /* Older userspace won't unhalt the vcpu on reset. */
4722         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4723             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4724             !(vcpu->arch.cr0 & X86_CR0_PE))
4725                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4726
4727         vcpu_put(vcpu);
4728
4729         return 0;
4730 }
4731
4732 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4733                                         struct kvm_guest_debug *dbg)
4734 {
4735         unsigned long rflags;
4736         int i, r;
4737
4738         vcpu_load(vcpu);
4739
4740         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
4741                 r = -EBUSY;
4742                 if (vcpu->arch.exception.pending)
4743                         goto unlock_out;
4744                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4745                         kvm_queue_exception(vcpu, DB_VECTOR);
4746                 else
4747                         kvm_queue_exception(vcpu, BP_VECTOR);
4748         }
4749
4750         /*
4751          * Read rflags as long as potentially injected trace flags are still
4752          * filtered out.
4753          */
4754         rflags = kvm_get_rflags(vcpu);
4755
4756         vcpu->guest_debug = dbg->control;
4757         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
4758                 vcpu->guest_debug = 0;
4759
4760         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4761                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4762                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4763                 vcpu->arch.switch_db_regs =
4764                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4765         } else {
4766                 for (i = 0; i < KVM_NR_DB_REGS; i++)
4767                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4768                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4769         }
4770
4771         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
4772                 vcpu->arch.singlestep_cs =
4773                         get_segment_selector(vcpu, VCPU_SREG_CS);
4774                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
4775         }
4776
4777         /*
4778          * Trigger an rflags update that will inject or remove the trace
4779          * flags.
4780          */
4781         kvm_set_rflags(vcpu, rflags);
4782
4783         kvm_x86_ops->set_guest_debug(vcpu, dbg);
4784
4785         r = 0;
4786
4787 unlock_out:
4788         vcpu_put(vcpu);
4789
4790         return r;
4791 }
4792
4793 /*
4794  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
4795  * we have asm/x86/processor.h
4796  */
4797 struct fxsave {
4798         u16     cwd;
4799         u16     swd;
4800         u16     twd;
4801         u16     fop;
4802         u64     rip;
4803         u64     rdp;
4804         u32     mxcsr;
4805         u32     mxcsr_mask;
4806         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
4807 #ifdef CONFIG_X86_64
4808         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
4809 #else
4810         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
4811 #endif
4812 };
4813
4814 /*
4815  * Translate a guest virtual address to a guest physical address.
4816  */
4817 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4818                                     struct kvm_translation *tr)
4819 {
4820         unsigned long vaddr = tr->linear_address;
4821         gpa_t gpa;
4822
4823         vcpu_load(vcpu);
4824         down_read(&vcpu->kvm->slots_lock);
4825         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4826         up_read(&vcpu->kvm->slots_lock);
4827         tr->physical_address = gpa;
4828         tr->valid = gpa != UNMAPPED_GVA;
4829         tr->writeable = 1;
4830         tr->usermode = 0;
4831         vcpu_put(vcpu);
4832
4833         return 0;
4834 }
4835
4836 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4837 {
4838         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4839
4840         vcpu_load(vcpu);
4841
4842         memcpy(fpu->fpr, fxsave->st_space, 128);
4843         fpu->fcw = fxsave->cwd;
4844         fpu->fsw = fxsave->swd;
4845         fpu->ftwx = fxsave->twd;
4846         fpu->last_opcode = fxsave->fop;
4847         fpu->last_ip = fxsave->rip;
4848         fpu->last_dp = fxsave->rdp;
4849         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4850
4851         vcpu_put(vcpu);
4852
4853         return 0;
4854 }
4855
4856 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4857 {
4858         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4859
4860         vcpu_load(vcpu);
4861
4862         memcpy(fxsave->st_space, fpu->fpr, 128);
4863         fxsave->cwd = fpu->fcw;
4864         fxsave->swd = fpu->fsw;
4865         fxsave->twd = fpu->ftwx;
4866         fxsave->fop = fpu->last_opcode;
4867         fxsave->rip = fpu->last_ip;
4868         fxsave->rdp = fpu->last_dp;
4869         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4870
4871         vcpu_put(vcpu);
4872
4873         return 0;
4874 }
4875
4876 void fx_init(struct kvm_vcpu *vcpu)
4877 {
4878         unsigned after_mxcsr_mask;
4879
4880         /*
4881          * Touch the fpu the first time in non atomic context as if
4882          * this is the first fpu instruction the exception handler
4883          * will fire before the instruction returns and it'll have to
4884          * allocate ram with GFP_KERNEL.
4885          */
4886         if (!used_math())
4887                 kvm_fx_save(&vcpu->arch.host_fx_image);
4888
4889         /* Initialize guest FPU by resetting ours and saving into guest's */
4890         preempt_disable();
4891         kvm_fx_save(&vcpu->arch.host_fx_image);
4892         kvm_fx_finit();
4893         kvm_fx_save(&vcpu->arch.guest_fx_image);
4894         kvm_fx_restore(&vcpu->arch.host_fx_image);
4895         preempt_enable();
4896
4897         vcpu->arch.cr0 |= X86_CR0_ET;
4898         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4899         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4900         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4901                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4902 }
4903 EXPORT_SYMBOL_GPL(fx_init);
4904
4905 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4906 {
4907         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4908                 return;
4909
4910         vcpu->guest_fpu_loaded = 1;
4911         kvm_fx_save(&vcpu->arch.host_fx_image);
4912         kvm_fx_restore(&vcpu->arch.guest_fx_image);
4913 }
4914 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4915
4916 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4917 {
4918         if (!vcpu->guest_fpu_loaded)
4919                 return;
4920
4921         vcpu->guest_fpu_loaded = 0;
4922         kvm_fx_save(&vcpu->arch.guest_fx_image);
4923         kvm_fx_restore(&vcpu->arch.host_fx_image);
4924         ++vcpu->stat.fpu_reload;
4925 }
4926 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4927
4928 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4929 {
4930         if (vcpu->arch.time_page) {
4931                 kvm_release_page_dirty(vcpu->arch.time_page);
4932                 vcpu->arch.time_page = NULL;
4933         }
4934
4935         kvm_x86_ops->vcpu_free(vcpu);
4936 }
4937
4938 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4939                                                 unsigned int id)
4940 {
4941         return kvm_x86_ops->vcpu_create(kvm, id);
4942 }
4943
4944 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4945 {
4946         int r;
4947
4948         /* We do fxsave: this must be aligned. */
4949         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4950
4951         vcpu->arch.mtrr_state.have_fixed = 1;
4952         vcpu_load(vcpu);
4953         r = kvm_arch_vcpu_reset(vcpu);
4954         if (r == 0)
4955                 r = kvm_mmu_setup(vcpu);
4956         vcpu_put(vcpu);
4957         if (r < 0)
4958                 goto free_vcpu;
4959
4960         return 0;
4961 free_vcpu:
4962         kvm_x86_ops->vcpu_free(vcpu);
4963         return r;
4964 }
4965
4966 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4967 {
4968         vcpu_load(vcpu);
4969         kvm_mmu_unload(vcpu);
4970         vcpu_put(vcpu);
4971
4972         kvm_x86_ops->vcpu_free(vcpu);
4973 }
4974
4975 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4976 {
4977         vcpu->arch.nmi_pending = false;
4978         vcpu->arch.nmi_injected = false;
4979
4980         vcpu->arch.switch_db_regs = 0;
4981         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4982         vcpu->arch.dr6 = DR6_FIXED_1;
4983         vcpu->arch.dr7 = DR7_FIXED_1;
4984
4985         return kvm_x86_ops->vcpu_reset(vcpu);
4986 }
4987
4988 int kvm_arch_hardware_enable(void *garbage)
4989 {
4990         /*
4991          * Since this may be called from a hotplug notifcation,
4992          * we can't get the CPU frequency directly.
4993          */
4994         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4995                 int cpu = raw_smp_processor_id();
4996                 per_cpu(cpu_tsc_khz, cpu) = 0;
4997         }
4998
4999         kvm_shared_msr_cpu_online();
5000
5001         return kvm_x86_ops->hardware_enable(garbage);
5002 }
5003
5004 void kvm_arch_hardware_disable(void *garbage)
5005 {
5006         kvm_x86_ops->hardware_disable(garbage);
5007 }
5008
5009 int kvm_arch_hardware_setup(void)
5010 {
5011         return kvm_x86_ops->hardware_setup();
5012 }
5013
5014 void kvm_arch_hardware_unsetup(void)
5015 {
5016         kvm_x86_ops->hardware_unsetup();
5017 }
5018
5019 void kvm_arch_check_processor_compat(void *rtn)
5020 {
5021         kvm_x86_ops->check_processor_compatibility(rtn);
5022 }
5023
5024 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5025 {
5026         struct page *page;
5027         struct kvm *kvm;
5028         int r;
5029
5030         BUG_ON(vcpu->kvm == NULL);
5031         kvm = vcpu->kvm;
5032
5033         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5034         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5035                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5036         else
5037                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5038
5039         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5040         if (!page) {
5041                 r = -ENOMEM;
5042                 goto fail;
5043         }
5044         vcpu->arch.pio_data = page_address(page);
5045
5046         r = kvm_mmu_create(vcpu);
5047         if (r < 0)
5048                 goto fail_free_pio_data;
5049
5050         if (irqchip_in_kernel(kvm)) {
5051                 r = kvm_create_lapic(vcpu);
5052                 if (r < 0)
5053                         goto fail_mmu_destroy;
5054         }
5055
5056         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5057                                        GFP_KERNEL);
5058         if (!vcpu->arch.mce_banks) {
5059                 r = -ENOMEM;
5060                 goto fail_mmu_destroy;
5061         }
5062         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5063
5064         return 0;
5065
5066 fail_mmu_destroy:
5067         kvm_mmu_destroy(vcpu);
5068 fail_free_pio_data:
5069         free_page((unsigned long)vcpu->arch.pio_data);
5070 fail:
5071         return r;
5072 }
5073
5074 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5075 {
5076         kvm_free_lapic(vcpu);
5077         down_read(&vcpu->kvm->slots_lock);
5078         kvm_mmu_destroy(vcpu);
5079         up_read(&vcpu->kvm->slots_lock);
5080         free_page((unsigned long)vcpu->arch.pio_data);
5081 }
5082
5083 struct  kvm *kvm_arch_create_vm(void)
5084 {
5085         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5086
5087         if (!kvm)
5088                 return ERR_PTR(-ENOMEM);
5089
5090         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5091         INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5092
5093         /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5094         set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5095
5096         rdtscll(kvm->arch.vm_init_tsc);
5097
5098         return kvm;
5099 }
5100
5101 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5102 {
5103         vcpu_load(vcpu);
5104         kvm_mmu_unload(vcpu);
5105         vcpu_put(vcpu);
5106 }
5107
5108 static void kvm_free_vcpus(struct kvm *kvm)
5109 {
5110         unsigned int i;
5111         struct kvm_vcpu *vcpu;
5112
5113         /*
5114          * Unpin any mmu pages first.
5115          */
5116         kvm_for_each_vcpu(i, vcpu, kvm)
5117                 kvm_unload_vcpu_mmu(vcpu);
5118         kvm_for_each_vcpu(i, vcpu, kvm)
5119                 kvm_arch_vcpu_free(vcpu);
5120
5121         mutex_lock(&kvm->lock);
5122         for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5123                 kvm->vcpus[i] = NULL;
5124
5125         atomic_set(&kvm->online_vcpus, 0);
5126         mutex_unlock(&kvm->lock);
5127 }
5128
5129 void kvm_arch_sync_events(struct kvm *kvm)
5130 {
5131         kvm_free_all_assigned_devices(kvm);
5132 }
5133
5134 void kvm_arch_destroy_vm(struct kvm *kvm)
5135 {
5136         kvm_iommu_unmap_guest(kvm);
5137         kvm_free_pit(kvm);
5138         kfree(kvm->arch.vpic);
5139         kfree(kvm->arch.vioapic);
5140         kvm_free_vcpus(kvm);
5141         kvm_free_physmem(kvm);
5142         if (kvm->arch.apic_access_page)
5143                 put_page(kvm->arch.apic_access_page);
5144         if (kvm->arch.ept_identity_pagetable)
5145                 put_page(kvm->arch.ept_identity_pagetable);
5146         kfree(kvm);
5147 }
5148
5149 int kvm_arch_set_memory_region(struct kvm *kvm,
5150                                 struct kvm_userspace_memory_region *mem,
5151                                 struct kvm_memory_slot old,
5152                                 int user_alloc)
5153 {
5154         int npages = mem->memory_size >> PAGE_SHIFT;
5155         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
5156
5157         /*To keep backward compatibility with older userspace,
5158          *x86 needs to hanlde !user_alloc case.
5159          */
5160         if (!user_alloc) {
5161                 if (npages && !old.rmap) {
5162                         unsigned long userspace_addr;
5163
5164                         down_write(&current->mm->mmap_sem);
5165                         userspace_addr = do_mmap(NULL, 0,
5166                                                  npages * PAGE_SIZE,
5167                                                  PROT_READ | PROT_WRITE,
5168                                                  MAP_PRIVATE | MAP_ANONYMOUS,
5169                                                  0);
5170                         up_write(&current->mm->mmap_sem);
5171
5172                         if (IS_ERR((void *)userspace_addr))
5173                                 return PTR_ERR((void *)userspace_addr);
5174
5175                         /* set userspace_addr atomically for kvm_hva_to_rmapp */
5176                         spin_lock(&kvm->mmu_lock);
5177                         memslot->userspace_addr = userspace_addr;
5178                         spin_unlock(&kvm->mmu_lock);
5179                 } else {
5180                         if (!old.user_alloc && old.rmap) {
5181                                 int ret;
5182
5183                                 down_write(&current->mm->mmap_sem);
5184                                 ret = do_munmap(current->mm, old.userspace_addr,
5185                                                 old.npages * PAGE_SIZE);
5186                                 up_write(&current->mm->mmap_sem);
5187                                 if (ret < 0)
5188                                         printk(KERN_WARNING
5189                                        "kvm_vm_ioctl_set_memory_region: "
5190                                        "failed to munmap memory\n");
5191                         }
5192                 }
5193         }
5194
5195         spin_lock(&kvm->mmu_lock);
5196         if (!kvm->arch.n_requested_mmu_pages) {
5197                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5198                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5199         }
5200
5201         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5202         spin_unlock(&kvm->mmu_lock);
5203
5204         return 0;
5205 }
5206
5207 void kvm_arch_flush_shadow(struct kvm *kvm)
5208 {
5209         kvm_mmu_zap_all(kvm);
5210         kvm_reload_remote_mmus(kvm);
5211 }
5212
5213 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5214 {
5215         return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5216                 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5217                 || vcpu->arch.nmi_pending ||
5218                 (kvm_arch_interrupt_allowed(vcpu) &&
5219                  kvm_cpu_has_interrupt(vcpu));
5220 }
5221
5222 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5223 {
5224         int me;
5225         int cpu = vcpu->cpu;
5226
5227         if (waitqueue_active(&vcpu->wq)) {
5228                 wake_up_interruptible(&vcpu->wq);
5229                 ++vcpu->stat.halt_wakeup;
5230         }
5231
5232         me = get_cpu();
5233         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5234                 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5235                         smp_send_reschedule(cpu);
5236         put_cpu();
5237 }
5238
5239 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5240 {
5241         return kvm_x86_ops->interrupt_allowed(vcpu);
5242 }
5243
5244 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5245 {
5246         unsigned long rflags;
5247
5248         rflags = kvm_x86_ops->get_rflags(vcpu);
5249         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5250                 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5251         return rflags;
5252 }
5253 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5254
5255 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5256 {
5257         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5258             vcpu->arch.singlestep_cs ==
5259                         get_segment_selector(vcpu, VCPU_SREG_CS) &&
5260             vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5261                 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5262         kvm_x86_ops->set_rflags(vcpu, rflags);
5263 }
5264 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5265
5266 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5267 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5268 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5269 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5270 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5271 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5272 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5273 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5274 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5275 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5276 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);