KVM: remove the usage of the mmap_sem for the protection of the memory slots.
[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  *
8  * Authors:
9  *   Avi Kivity   <avi@qumranet.com>
10  *   Yaniv Kamay  <yaniv@qumranet.com>
11  *
12  * This work is licensed under the terms of the GNU GPL, version 2.  See
13  * the COPYING file in the top-level directory.
14  *
15  */
16
17 #include <linux/kvm_host.h>
18 #include "segment_descriptor.h"
19 #include "irq.h"
20 #include "mmu.h"
21
22 #include <linux/kvm.h>
23 #include <linux/fs.h>
24 #include <linux/vmalloc.h>
25 #include <linux/module.h>
26 #include <linux/mman.h>
27 #include <linux/highmem.h>
28
29 #include <asm/uaccess.h>
30 #include <asm/msr.h>
31
32 #define MAX_IO_MSRS 256
33 #define CR0_RESERVED_BITS                                               \
34         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
35                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
36                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
37 #define CR4_RESERVED_BITS                                               \
38         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
39                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
40                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
41                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
42
43 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
44 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
45
46 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
47 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
48
49 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
50                                     struct kvm_cpuid_entry2 __user *entries);
51
52 struct kvm_x86_ops *kvm_x86_ops;
53
54 struct kvm_stats_debugfs_item debugfs_entries[] = {
55         { "pf_fixed", VCPU_STAT(pf_fixed) },
56         { "pf_guest", VCPU_STAT(pf_guest) },
57         { "tlb_flush", VCPU_STAT(tlb_flush) },
58         { "invlpg", VCPU_STAT(invlpg) },
59         { "exits", VCPU_STAT(exits) },
60         { "io_exits", VCPU_STAT(io_exits) },
61         { "mmio_exits", VCPU_STAT(mmio_exits) },
62         { "signal_exits", VCPU_STAT(signal_exits) },
63         { "irq_window", VCPU_STAT(irq_window_exits) },
64         { "halt_exits", VCPU_STAT(halt_exits) },
65         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
66         { "request_irq", VCPU_STAT(request_irq_exits) },
67         { "irq_exits", VCPU_STAT(irq_exits) },
68         { "host_state_reload", VCPU_STAT(host_state_reload) },
69         { "efer_reload", VCPU_STAT(efer_reload) },
70         { "fpu_reload", VCPU_STAT(fpu_reload) },
71         { "insn_emulation", VCPU_STAT(insn_emulation) },
72         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
73         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
74         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
75         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
76         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
77         { "mmu_flooded", VM_STAT(mmu_flooded) },
78         { "mmu_recycled", VM_STAT(mmu_recycled) },
79         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
80         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
81         { NULL }
82 };
83
84
85 unsigned long segment_base(u16 selector)
86 {
87         struct descriptor_table gdt;
88         struct segment_descriptor *d;
89         unsigned long table_base;
90         unsigned long v;
91
92         if (selector == 0)
93                 return 0;
94
95         asm("sgdt %0" : "=m"(gdt));
96         table_base = gdt.base;
97
98         if (selector & 4) {           /* from ldt */
99                 u16 ldt_selector;
100
101                 asm("sldt %0" : "=g"(ldt_selector));
102                 table_base = segment_base(ldt_selector);
103         }
104         d = (struct segment_descriptor *)(table_base + (selector & ~7));
105         v = d->base_low | ((unsigned long)d->base_mid << 16) |
106                 ((unsigned long)d->base_high << 24);
107 #ifdef CONFIG_X86_64
108         if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
109                 v |= ((unsigned long) \
110                       ((struct segment_descriptor_64 *)d)->base_higher) << 32;
111 #endif
112         return v;
113 }
114 EXPORT_SYMBOL_GPL(segment_base);
115
116 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
117 {
118         if (irqchip_in_kernel(vcpu->kvm))
119                 return vcpu->arch.apic_base;
120         else
121                 return vcpu->arch.apic_base;
122 }
123 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
124
125 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
126 {
127         /* TODO: reserve bits check */
128         if (irqchip_in_kernel(vcpu->kvm))
129                 kvm_lapic_set_base(vcpu, data);
130         else
131                 vcpu->arch.apic_base = data;
132 }
133 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
134
135 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
136 {
137         WARN_ON(vcpu->arch.exception.pending);
138         vcpu->arch.exception.pending = true;
139         vcpu->arch.exception.has_error_code = false;
140         vcpu->arch.exception.nr = nr;
141 }
142 EXPORT_SYMBOL_GPL(kvm_queue_exception);
143
144 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
145                            u32 error_code)
146 {
147         ++vcpu->stat.pf_guest;
148         if (vcpu->arch.exception.pending && vcpu->arch.exception.nr == PF_VECTOR) {
149                 printk(KERN_DEBUG "kvm: inject_page_fault:"
150                        " double fault 0x%lx\n", addr);
151                 vcpu->arch.exception.nr = DF_VECTOR;
152                 vcpu->arch.exception.error_code = 0;
153                 return;
154         }
155         vcpu->arch.cr2 = addr;
156         kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
157 }
158
159 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
160 {
161         WARN_ON(vcpu->arch.exception.pending);
162         vcpu->arch.exception.pending = true;
163         vcpu->arch.exception.has_error_code = true;
164         vcpu->arch.exception.nr = nr;
165         vcpu->arch.exception.error_code = error_code;
166 }
167 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
168
169 static void __queue_exception(struct kvm_vcpu *vcpu)
170 {
171         kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
172                                      vcpu->arch.exception.has_error_code,
173                                      vcpu->arch.exception.error_code);
174 }
175
176 /*
177  * Load the pae pdptrs.  Return true is they are all valid.
178  */
179 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
180 {
181         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
182         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
183         int i;
184         int ret;
185         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
186
187         down_read(&vcpu->kvm->slots_lock);
188         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
189                                   offset * sizeof(u64), sizeof(pdpte));
190         if (ret < 0) {
191                 ret = 0;
192                 goto out;
193         }
194         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
195                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
196                         ret = 0;
197                         goto out;
198                 }
199         }
200         ret = 1;
201
202         memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
203 out:
204         up_read(&vcpu->kvm->slots_lock);
205
206         return ret;
207 }
208
209 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
210 {
211         u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
212         bool changed = true;
213         int r;
214
215         if (is_long_mode(vcpu) || !is_pae(vcpu))
216                 return false;
217
218         down_read(&vcpu->kvm->slots_lock);
219         r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
220         if (r < 0)
221                 goto out;
222         changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
223 out:
224         up_read(&vcpu->kvm->slots_lock);
225
226         return changed;
227 }
228
229 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
230 {
231         if (cr0 & CR0_RESERVED_BITS) {
232                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
233                        cr0, vcpu->arch.cr0);
234                 kvm_inject_gp(vcpu, 0);
235                 return;
236         }
237
238         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
239                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
240                 kvm_inject_gp(vcpu, 0);
241                 return;
242         }
243
244         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
245                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
246                        "and a clear PE flag\n");
247                 kvm_inject_gp(vcpu, 0);
248                 return;
249         }
250
251         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
252 #ifdef CONFIG_X86_64
253                 if ((vcpu->arch.shadow_efer & EFER_LME)) {
254                         int cs_db, cs_l;
255
256                         if (!is_pae(vcpu)) {
257                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
258                                        "in long mode while PAE is disabled\n");
259                                 kvm_inject_gp(vcpu, 0);
260                                 return;
261                         }
262                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
263                         if (cs_l) {
264                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
265                                        "in long mode while CS.L == 1\n");
266                                 kvm_inject_gp(vcpu, 0);
267                                 return;
268
269                         }
270                 } else
271 #endif
272                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
273                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
274                                "reserved bits\n");
275                         kvm_inject_gp(vcpu, 0);
276                         return;
277                 }
278
279         }
280
281         kvm_x86_ops->set_cr0(vcpu, cr0);
282         vcpu->arch.cr0 = cr0;
283
284         kvm_mmu_reset_context(vcpu);
285         return;
286 }
287 EXPORT_SYMBOL_GPL(set_cr0);
288
289 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
290 {
291         set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
292 }
293 EXPORT_SYMBOL_GPL(lmsw);
294
295 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
296 {
297         if (cr4 & CR4_RESERVED_BITS) {
298                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
299                 kvm_inject_gp(vcpu, 0);
300                 return;
301         }
302
303         if (is_long_mode(vcpu)) {
304                 if (!(cr4 & X86_CR4_PAE)) {
305                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
306                                "in long mode\n");
307                         kvm_inject_gp(vcpu, 0);
308                         return;
309                 }
310         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
311                    && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
312                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
313                 kvm_inject_gp(vcpu, 0);
314                 return;
315         }
316
317         if (cr4 & X86_CR4_VMXE) {
318                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
319                 kvm_inject_gp(vcpu, 0);
320                 return;
321         }
322         kvm_x86_ops->set_cr4(vcpu, cr4);
323         vcpu->arch.cr4 = cr4;
324         kvm_mmu_reset_context(vcpu);
325 }
326 EXPORT_SYMBOL_GPL(set_cr4);
327
328 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
329 {
330         if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
331                 kvm_mmu_flush_tlb(vcpu);
332                 return;
333         }
334
335         if (is_long_mode(vcpu)) {
336                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
337                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
338                         kvm_inject_gp(vcpu, 0);
339                         return;
340                 }
341         } else {
342                 if (is_pae(vcpu)) {
343                         if (cr3 & CR3_PAE_RESERVED_BITS) {
344                                 printk(KERN_DEBUG
345                                        "set_cr3: #GP, reserved bits\n");
346                                 kvm_inject_gp(vcpu, 0);
347                                 return;
348                         }
349                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
350                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
351                                        "reserved bits\n");
352                                 kvm_inject_gp(vcpu, 0);
353                                 return;
354                         }
355                 }
356                 /*
357                  * We don't check reserved bits in nonpae mode, because
358                  * this isn't enforced, and VMware depends on this.
359                  */
360         }
361
362         down_read(&vcpu->kvm->slots_lock);
363         /*
364          * Does the new cr3 value map to physical memory? (Note, we
365          * catch an invalid cr3 even in real-mode, because it would
366          * cause trouble later on when we turn on paging anyway.)
367          *
368          * A real CPU would silently accept an invalid cr3 and would
369          * attempt to use it - with largely undefined (and often hard
370          * to debug) behavior on the guest side.
371          */
372         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
373                 kvm_inject_gp(vcpu, 0);
374         else {
375                 vcpu->arch.cr3 = cr3;
376                 vcpu->arch.mmu.new_cr3(vcpu);
377         }
378         up_read(&vcpu->kvm->slots_lock);
379 }
380 EXPORT_SYMBOL_GPL(set_cr3);
381
382 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
383 {
384         if (cr8 & CR8_RESERVED_BITS) {
385                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
386                 kvm_inject_gp(vcpu, 0);
387                 return;
388         }
389         if (irqchip_in_kernel(vcpu->kvm))
390                 kvm_lapic_set_tpr(vcpu, cr8);
391         else
392                 vcpu->arch.cr8 = cr8;
393 }
394 EXPORT_SYMBOL_GPL(set_cr8);
395
396 unsigned long get_cr8(struct kvm_vcpu *vcpu)
397 {
398         if (irqchip_in_kernel(vcpu->kvm))
399                 return kvm_lapic_get_cr8(vcpu);
400         else
401                 return vcpu->arch.cr8;
402 }
403 EXPORT_SYMBOL_GPL(get_cr8);
404
405 /*
406  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
407  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
408  *
409  * This list is modified at module load time to reflect the
410  * capabilities of the host cpu.
411  */
412 static u32 msrs_to_save[] = {
413         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
414         MSR_K6_STAR,
415 #ifdef CONFIG_X86_64
416         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
417 #endif
418         MSR_IA32_TIME_STAMP_COUNTER,
419 };
420
421 static unsigned num_msrs_to_save;
422
423 static u32 emulated_msrs[] = {
424         MSR_IA32_MISC_ENABLE,
425 };
426
427 #ifdef CONFIG_X86_64
428
429 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
430 {
431         if (efer & EFER_RESERVED_BITS) {
432                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
433                        efer);
434                 kvm_inject_gp(vcpu, 0);
435                 return;
436         }
437
438         if (is_paging(vcpu)
439             && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
440                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
441                 kvm_inject_gp(vcpu, 0);
442                 return;
443         }
444
445         kvm_x86_ops->set_efer(vcpu, efer);
446
447         efer &= ~EFER_LMA;
448         efer |= vcpu->arch.shadow_efer & EFER_LMA;
449
450         vcpu->arch.shadow_efer = efer;
451 }
452
453 #endif
454
455 /*
456  * Writes msr value into into the appropriate "register".
457  * Returns 0 on success, non-0 otherwise.
458  * Assumes vcpu_load() was already called.
459  */
460 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
461 {
462         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
463 }
464
465 /*
466  * Adapt set_msr() to msr_io()'s calling convention
467  */
468 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
469 {
470         return kvm_set_msr(vcpu, index, *data);
471 }
472
473
474 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
475 {
476         switch (msr) {
477 #ifdef CONFIG_X86_64
478         case MSR_EFER:
479                 set_efer(vcpu, data);
480                 break;
481 #endif
482         case MSR_IA32_MC0_STATUS:
483                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
484                        __FUNCTION__, data);
485                 break;
486         case MSR_IA32_MCG_STATUS:
487                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
488                         __FUNCTION__, data);
489                 break;
490         case MSR_IA32_MCG_CTL:
491                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
492                         __FUNCTION__, data);
493                 break;
494         case MSR_IA32_UCODE_REV:
495         case MSR_IA32_UCODE_WRITE:
496         case 0x200 ... 0x2ff: /* MTRRs */
497                 break;
498         case MSR_IA32_APICBASE:
499                 kvm_set_apic_base(vcpu, data);
500                 break;
501         case MSR_IA32_MISC_ENABLE:
502                 vcpu->arch.ia32_misc_enable_msr = data;
503                 break;
504         default:
505                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
506                 return 1;
507         }
508         return 0;
509 }
510 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
511
512
513 /*
514  * Reads an msr value (of 'msr_index') into 'pdata'.
515  * Returns 0 on success, non-0 otherwise.
516  * Assumes vcpu_load() was already called.
517  */
518 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
519 {
520         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
521 }
522
523 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
524 {
525         u64 data;
526
527         switch (msr) {
528         case 0xc0010010: /* SYSCFG */
529         case 0xc0010015: /* HWCR */
530         case MSR_IA32_PLATFORM_ID:
531         case MSR_IA32_P5_MC_ADDR:
532         case MSR_IA32_P5_MC_TYPE:
533         case MSR_IA32_MC0_CTL:
534         case MSR_IA32_MCG_STATUS:
535         case MSR_IA32_MCG_CAP:
536         case MSR_IA32_MCG_CTL:
537         case MSR_IA32_MC0_MISC:
538         case MSR_IA32_MC0_MISC+4:
539         case MSR_IA32_MC0_MISC+8:
540         case MSR_IA32_MC0_MISC+12:
541         case MSR_IA32_MC0_MISC+16:
542         case MSR_IA32_UCODE_REV:
543         case MSR_IA32_PERF_STATUS:
544         case MSR_IA32_EBL_CR_POWERON:
545                 /* MTRR registers */
546         case 0xfe:
547         case 0x200 ... 0x2ff:
548                 data = 0;
549                 break;
550         case 0xcd: /* fsb frequency */
551                 data = 3;
552                 break;
553         case MSR_IA32_APICBASE:
554                 data = kvm_get_apic_base(vcpu);
555                 break;
556         case MSR_IA32_MISC_ENABLE:
557                 data = vcpu->arch.ia32_misc_enable_msr;
558                 break;
559 #ifdef CONFIG_X86_64
560         case MSR_EFER:
561                 data = vcpu->arch.shadow_efer;
562                 break;
563 #endif
564         default:
565                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
566                 return 1;
567         }
568         *pdata = data;
569         return 0;
570 }
571 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
572
573 /*
574  * Read or write a bunch of msrs. All parameters are kernel addresses.
575  *
576  * @return number of msrs set successfully.
577  */
578 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
579                     struct kvm_msr_entry *entries,
580                     int (*do_msr)(struct kvm_vcpu *vcpu,
581                                   unsigned index, u64 *data))
582 {
583         int i;
584
585         vcpu_load(vcpu);
586
587         for (i = 0; i < msrs->nmsrs; ++i)
588                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
589                         break;
590
591         vcpu_put(vcpu);
592
593         return i;
594 }
595
596 /*
597  * Read or write a bunch of msrs. Parameters are user addresses.
598  *
599  * @return number of msrs set successfully.
600  */
601 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
602                   int (*do_msr)(struct kvm_vcpu *vcpu,
603                                 unsigned index, u64 *data),
604                   int writeback)
605 {
606         struct kvm_msrs msrs;
607         struct kvm_msr_entry *entries;
608         int r, n;
609         unsigned size;
610
611         r = -EFAULT;
612         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
613                 goto out;
614
615         r = -E2BIG;
616         if (msrs.nmsrs >= MAX_IO_MSRS)
617                 goto out;
618
619         r = -ENOMEM;
620         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
621         entries = vmalloc(size);
622         if (!entries)
623                 goto out;
624
625         r = -EFAULT;
626         if (copy_from_user(entries, user_msrs->entries, size))
627                 goto out_free;
628
629         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
630         if (r < 0)
631                 goto out_free;
632
633         r = -EFAULT;
634         if (writeback && copy_to_user(user_msrs->entries, entries, size))
635                 goto out_free;
636
637         r = n;
638
639 out_free:
640         vfree(entries);
641 out:
642         return r;
643 }
644
645 /*
646  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
647  * cached on it.
648  */
649 void decache_vcpus_on_cpu(int cpu)
650 {
651         struct kvm *vm;
652         struct kvm_vcpu *vcpu;
653         int i;
654
655         spin_lock(&kvm_lock);
656         list_for_each_entry(vm, &vm_list, vm_list)
657                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
658                         vcpu = vm->vcpus[i];
659                         if (!vcpu)
660                                 continue;
661                         /*
662                          * If the vcpu is locked, then it is running on some
663                          * other cpu and therefore it is not cached on the
664                          * cpu in question.
665                          *
666                          * If it's not locked, check the last cpu it executed
667                          * on.
668                          */
669                         if (mutex_trylock(&vcpu->mutex)) {
670                                 if (vcpu->cpu == cpu) {
671                                         kvm_x86_ops->vcpu_decache(vcpu);
672                                         vcpu->cpu = -1;
673                                 }
674                                 mutex_unlock(&vcpu->mutex);
675                         }
676                 }
677         spin_unlock(&kvm_lock);
678 }
679
680 int kvm_dev_ioctl_check_extension(long ext)
681 {
682         int r;
683
684         switch (ext) {
685         case KVM_CAP_IRQCHIP:
686         case KVM_CAP_HLT:
687         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
688         case KVM_CAP_USER_MEMORY:
689         case KVM_CAP_SET_TSS_ADDR:
690         case KVM_CAP_EXT_CPUID:
691                 r = 1;
692                 break;
693         case KVM_CAP_VAPIC:
694                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
695                 break;
696         default:
697                 r = 0;
698                 break;
699         }
700         return r;
701
702 }
703
704 long kvm_arch_dev_ioctl(struct file *filp,
705                         unsigned int ioctl, unsigned long arg)
706 {
707         void __user *argp = (void __user *)arg;
708         long r;
709
710         switch (ioctl) {
711         case KVM_GET_MSR_INDEX_LIST: {
712                 struct kvm_msr_list __user *user_msr_list = argp;
713                 struct kvm_msr_list msr_list;
714                 unsigned n;
715
716                 r = -EFAULT;
717                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
718                         goto out;
719                 n = msr_list.nmsrs;
720                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
721                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
722                         goto out;
723                 r = -E2BIG;
724                 if (n < num_msrs_to_save)
725                         goto out;
726                 r = -EFAULT;
727                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
728                                  num_msrs_to_save * sizeof(u32)))
729                         goto out;
730                 if (copy_to_user(user_msr_list->indices
731                                  + num_msrs_to_save * sizeof(u32),
732                                  &emulated_msrs,
733                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
734                         goto out;
735                 r = 0;
736                 break;
737         }
738         case KVM_GET_SUPPORTED_CPUID: {
739                 struct kvm_cpuid2 __user *cpuid_arg = argp;
740                 struct kvm_cpuid2 cpuid;
741
742                 r = -EFAULT;
743                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
744                         goto out;
745                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
746                         cpuid_arg->entries);
747                 if (r)
748                         goto out;
749
750                 r = -EFAULT;
751                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
752                         goto out;
753                 r = 0;
754                 break;
755         }
756         default:
757                 r = -EINVAL;
758         }
759 out:
760         return r;
761 }
762
763 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
764 {
765         kvm_x86_ops->vcpu_load(vcpu, cpu);
766 }
767
768 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
769 {
770         kvm_x86_ops->vcpu_put(vcpu);
771         kvm_put_guest_fpu(vcpu);
772 }
773
774 static int is_efer_nx(void)
775 {
776         u64 efer;
777
778         rdmsrl(MSR_EFER, efer);
779         return efer & EFER_NX;
780 }
781
782 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
783 {
784         int i;
785         struct kvm_cpuid_entry2 *e, *entry;
786
787         entry = NULL;
788         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
789                 e = &vcpu->arch.cpuid_entries[i];
790                 if (e->function == 0x80000001) {
791                         entry = e;
792                         break;
793                 }
794         }
795         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
796                 entry->edx &= ~(1 << 20);
797                 printk(KERN_INFO "kvm: guest NX capability removed\n");
798         }
799 }
800
801 /* when an old userspace process fills a new kernel module */
802 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
803                                     struct kvm_cpuid *cpuid,
804                                     struct kvm_cpuid_entry __user *entries)
805 {
806         int r, i;
807         struct kvm_cpuid_entry *cpuid_entries;
808
809         r = -E2BIG;
810         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
811                 goto out;
812         r = -ENOMEM;
813         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
814         if (!cpuid_entries)
815                 goto out;
816         r = -EFAULT;
817         if (copy_from_user(cpuid_entries, entries,
818                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
819                 goto out_free;
820         for (i = 0; i < cpuid->nent; i++) {
821                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
822                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
823                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
824                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
825                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
826                 vcpu->arch.cpuid_entries[i].index = 0;
827                 vcpu->arch.cpuid_entries[i].flags = 0;
828                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
829                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
830                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
831         }
832         vcpu->arch.cpuid_nent = cpuid->nent;
833         cpuid_fix_nx_cap(vcpu);
834         r = 0;
835
836 out_free:
837         vfree(cpuid_entries);
838 out:
839         return r;
840 }
841
842 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
843                                     struct kvm_cpuid2 *cpuid,
844                                     struct kvm_cpuid_entry2 __user *entries)
845 {
846         int r;
847
848         r = -E2BIG;
849         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
850                 goto out;
851         r = -EFAULT;
852         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
853                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
854                 goto out;
855         vcpu->arch.cpuid_nent = cpuid->nent;
856         return 0;
857
858 out:
859         return r;
860 }
861
862 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
863                                     struct kvm_cpuid2 *cpuid,
864                                     struct kvm_cpuid_entry2 __user *entries)
865 {
866         int r;
867
868         r = -E2BIG;
869         if (cpuid->nent < vcpu->arch.cpuid_nent)
870                 goto out;
871         r = -EFAULT;
872         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
873                            vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
874                 goto out;
875         return 0;
876
877 out:
878         cpuid->nent = vcpu->arch.cpuid_nent;
879         return r;
880 }
881
882 static inline u32 bit(int bitno)
883 {
884         return 1 << (bitno & 31);
885 }
886
887 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
888                           u32 index)
889 {
890         entry->function = function;
891         entry->index = index;
892         cpuid_count(entry->function, entry->index,
893                 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
894         entry->flags = 0;
895 }
896
897 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
898                          u32 index, int *nent, int maxnent)
899 {
900         const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
901                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
902                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
903                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
904                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
905                 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
906                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
907                 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
908                 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
909                 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
910         const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
911                 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
912                 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
913                 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
914                 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
915                 bit(X86_FEATURE_PGE) |
916                 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
917                 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
918                 bit(X86_FEATURE_SYSCALL) |
919                 (bit(X86_FEATURE_NX) && is_efer_nx()) |
920 #ifdef CONFIG_X86_64
921                 bit(X86_FEATURE_LM) |
922 #endif
923                 bit(X86_FEATURE_MMXEXT) |
924                 bit(X86_FEATURE_3DNOWEXT) |
925                 bit(X86_FEATURE_3DNOW);
926         const u32 kvm_supported_word3_x86_features =
927                 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
928         const u32 kvm_supported_word6_x86_features =
929                 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
930
931         /* all func 2 cpuid_count() should be called on the same cpu */
932         get_cpu();
933         do_cpuid_1_ent(entry, function, index);
934         ++*nent;
935
936         switch (function) {
937         case 0:
938                 entry->eax = min(entry->eax, (u32)0xb);
939                 break;
940         case 1:
941                 entry->edx &= kvm_supported_word0_x86_features;
942                 entry->ecx &= kvm_supported_word3_x86_features;
943                 break;
944         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
945          * may return different values. This forces us to get_cpu() before
946          * issuing the first command, and also to emulate this annoying behavior
947          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
948         case 2: {
949                 int t, times = entry->eax & 0xff;
950
951                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
952                 for (t = 1; t < times && *nent < maxnent; ++t) {
953                         do_cpuid_1_ent(&entry[t], function, 0);
954                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
955                         ++*nent;
956                 }
957                 break;
958         }
959         /* function 4 and 0xb have additional index. */
960         case 4: {
961                 int index, cache_type;
962
963                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
964                 /* read more entries until cache_type is zero */
965                 for (index = 1; *nent < maxnent; ++index) {
966                         cache_type = entry[index - 1].eax & 0x1f;
967                         if (!cache_type)
968                                 break;
969                         do_cpuid_1_ent(&entry[index], function, index);
970                         entry[index].flags |=
971                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
972                         ++*nent;
973                 }
974                 break;
975         }
976         case 0xb: {
977                 int index, level_type;
978
979                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
980                 /* read more entries until level_type is zero */
981                 for (index = 1; *nent < maxnent; ++index) {
982                         level_type = entry[index - 1].ecx & 0xff;
983                         if (!level_type)
984                                 break;
985                         do_cpuid_1_ent(&entry[index], function, index);
986                         entry[index].flags |=
987                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
988                         ++*nent;
989                 }
990                 break;
991         }
992         case 0x80000000:
993                 entry->eax = min(entry->eax, 0x8000001a);
994                 break;
995         case 0x80000001:
996                 entry->edx &= kvm_supported_word1_x86_features;
997                 entry->ecx &= kvm_supported_word6_x86_features;
998                 break;
999         }
1000         put_cpu();
1001 }
1002
1003 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1004                                     struct kvm_cpuid_entry2 __user *entries)
1005 {
1006         struct kvm_cpuid_entry2 *cpuid_entries;
1007         int limit, nent = 0, r = -E2BIG;
1008         u32 func;
1009
1010         if (cpuid->nent < 1)
1011                 goto out;
1012         r = -ENOMEM;
1013         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1014         if (!cpuid_entries)
1015                 goto out;
1016
1017         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1018         limit = cpuid_entries[0].eax;
1019         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1020                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1021                                 &nent, cpuid->nent);
1022         r = -E2BIG;
1023         if (nent >= cpuid->nent)
1024                 goto out_free;
1025
1026         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1027         limit = cpuid_entries[nent - 1].eax;
1028         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1029                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1030                                &nent, cpuid->nent);
1031         r = -EFAULT;
1032         if (copy_to_user(entries, cpuid_entries,
1033                         nent * sizeof(struct kvm_cpuid_entry2)))
1034                 goto out_free;
1035         cpuid->nent = nent;
1036         r = 0;
1037
1038 out_free:
1039         vfree(cpuid_entries);
1040 out:
1041         return r;
1042 }
1043
1044 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1045                                     struct kvm_lapic_state *s)
1046 {
1047         vcpu_load(vcpu);
1048         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1049         vcpu_put(vcpu);
1050
1051         return 0;
1052 }
1053
1054 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1055                                     struct kvm_lapic_state *s)
1056 {
1057         vcpu_load(vcpu);
1058         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1059         kvm_apic_post_state_restore(vcpu);
1060         vcpu_put(vcpu);
1061
1062         return 0;
1063 }
1064
1065 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1066                                     struct kvm_interrupt *irq)
1067 {
1068         if (irq->irq < 0 || irq->irq >= 256)
1069                 return -EINVAL;
1070         if (irqchip_in_kernel(vcpu->kvm))
1071                 return -ENXIO;
1072         vcpu_load(vcpu);
1073
1074         set_bit(irq->irq, vcpu->arch.irq_pending);
1075         set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1076
1077         vcpu_put(vcpu);
1078
1079         return 0;
1080 }
1081
1082 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1083                                            struct kvm_tpr_access_ctl *tac)
1084 {
1085         if (tac->flags)
1086                 return -EINVAL;
1087         vcpu->arch.tpr_access_reporting = !!tac->enabled;
1088         return 0;
1089 }
1090
1091 long kvm_arch_vcpu_ioctl(struct file *filp,
1092                          unsigned int ioctl, unsigned long arg)
1093 {
1094         struct kvm_vcpu *vcpu = filp->private_data;
1095         void __user *argp = (void __user *)arg;
1096         int r;
1097
1098         switch (ioctl) {
1099         case KVM_GET_LAPIC: {
1100                 struct kvm_lapic_state lapic;
1101
1102                 memset(&lapic, 0, sizeof lapic);
1103                 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1104                 if (r)
1105                         goto out;
1106                 r = -EFAULT;
1107                 if (copy_to_user(argp, &lapic, sizeof lapic))
1108                         goto out;
1109                 r = 0;
1110                 break;
1111         }
1112         case KVM_SET_LAPIC: {
1113                 struct kvm_lapic_state lapic;
1114
1115                 r = -EFAULT;
1116                 if (copy_from_user(&lapic, argp, sizeof lapic))
1117                         goto out;
1118                 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1119                 if (r)
1120                         goto out;
1121                 r = 0;
1122                 break;
1123         }
1124         case KVM_INTERRUPT: {
1125                 struct kvm_interrupt irq;
1126
1127                 r = -EFAULT;
1128                 if (copy_from_user(&irq, argp, sizeof irq))
1129                         goto out;
1130                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1131                 if (r)
1132                         goto out;
1133                 r = 0;
1134                 break;
1135         }
1136         case KVM_SET_CPUID: {
1137                 struct kvm_cpuid __user *cpuid_arg = argp;
1138                 struct kvm_cpuid cpuid;
1139
1140                 r = -EFAULT;
1141                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1142                         goto out;
1143                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1144                 if (r)
1145                         goto out;
1146                 break;
1147         }
1148         case KVM_SET_CPUID2: {
1149                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1150                 struct kvm_cpuid2 cpuid;
1151
1152                 r = -EFAULT;
1153                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1154                         goto out;
1155                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1156                                 cpuid_arg->entries);
1157                 if (r)
1158                         goto out;
1159                 break;
1160         }
1161         case KVM_GET_CPUID2: {
1162                 struct kvm_cpuid2 __user *cpuid_arg = argp;
1163                 struct kvm_cpuid2 cpuid;
1164
1165                 r = -EFAULT;
1166                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1167                         goto out;
1168                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1169                                 cpuid_arg->entries);
1170                 if (r)
1171                         goto out;
1172                 r = -EFAULT;
1173                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1174                         goto out;
1175                 r = 0;
1176                 break;
1177         }
1178         case KVM_GET_MSRS:
1179                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1180                 break;
1181         case KVM_SET_MSRS:
1182                 r = msr_io(vcpu, argp, do_set_msr, 0);
1183                 break;
1184         case KVM_TPR_ACCESS_REPORTING: {
1185                 struct kvm_tpr_access_ctl tac;
1186
1187                 r = -EFAULT;
1188                 if (copy_from_user(&tac, argp, sizeof tac))
1189                         goto out;
1190                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1191                 if (r)
1192                         goto out;
1193                 r = -EFAULT;
1194                 if (copy_to_user(argp, &tac, sizeof tac))
1195                         goto out;
1196                 r = 0;
1197                 break;
1198         };
1199         case KVM_SET_VAPIC_ADDR: {
1200                 struct kvm_vapic_addr va;
1201
1202                 r = -EINVAL;
1203                 if (!irqchip_in_kernel(vcpu->kvm))
1204                         goto out;
1205                 r = -EFAULT;
1206                 if (copy_from_user(&va, argp, sizeof va))
1207                         goto out;
1208                 r = 0;
1209                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1210                 break;
1211         }
1212         default:
1213                 r = -EINVAL;
1214         }
1215 out:
1216         return r;
1217 }
1218
1219 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1220 {
1221         int ret;
1222
1223         if (addr > (unsigned int)(-3 * PAGE_SIZE))
1224                 return -1;
1225         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1226         return ret;
1227 }
1228
1229 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1230                                           u32 kvm_nr_mmu_pages)
1231 {
1232         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1233                 return -EINVAL;
1234
1235         down_write(&kvm->slots_lock);
1236
1237         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1238         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1239
1240         up_write(&kvm->slots_lock);
1241         return 0;
1242 }
1243
1244 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1245 {
1246         return kvm->arch.n_alloc_mmu_pages;
1247 }
1248
1249 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1250 {
1251         int i;
1252         struct kvm_mem_alias *alias;
1253
1254         for (i = 0; i < kvm->arch.naliases; ++i) {
1255                 alias = &kvm->arch.aliases[i];
1256                 if (gfn >= alias->base_gfn
1257                     && gfn < alias->base_gfn + alias->npages)
1258                         return alias->target_gfn + gfn - alias->base_gfn;
1259         }
1260         return gfn;
1261 }
1262
1263 /*
1264  * Set a new alias region.  Aliases map a portion of physical memory into
1265  * another portion.  This is useful for memory windows, for example the PC
1266  * VGA region.
1267  */
1268 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1269                                          struct kvm_memory_alias *alias)
1270 {
1271         int r, n;
1272         struct kvm_mem_alias *p;
1273
1274         r = -EINVAL;
1275         /* General sanity checks */
1276         if (alias->memory_size & (PAGE_SIZE - 1))
1277                 goto out;
1278         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1279                 goto out;
1280         if (alias->slot >= KVM_ALIAS_SLOTS)
1281                 goto out;
1282         if (alias->guest_phys_addr + alias->memory_size
1283             < alias->guest_phys_addr)
1284                 goto out;
1285         if (alias->target_phys_addr + alias->memory_size
1286             < alias->target_phys_addr)
1287                 goto out;
1288
1289         down_write(&kvm->slots_lock);
1290
1291         p = &kvm->arch.aliases[alias->slot];
1292         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1293         p->npages = alias->memory_size >> PAGE_SHIFT;
1294         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1295
1296         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1297                 if (kvm->arch.aliases[n - 1].npages)
1298                         break;
1299         kvm->arch.naliases = n;
1300
1301         kvm_mmu_zap_all(kvm);
1302
1303         up_write(&kvm->slots_lock);
1304
1305         return 0;
1306
1307 out:
1308         return r;
1309 }
1310
1311 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1312 {
1313         int r;
1314
1315         r = 0;
1316         switch (chip->chip_id) {
1317         case KVM_IRQCHIP_PIC_MASTER:
1318                 memcpy(&chip->chip.pic,
1319                         &pic_irqchip(kvm)->pics[0],
1320                         sizeof(struct kvm_pic_state));
1321                 break;
1322         case KVM_IRQCHIP_PIC_SLAVE:
1323                 memcpy(&chip->chip.pic,
1324                         &pic_irqchip(kvm)->pics[1],
1325                         sizeof(struct kvm_pic_state));
1326                 break;
1327         case KVM_IRQCHIP_IOAPIC:
1328                 memcpy(&chip->chip.ioapic,
1329                         ioapic_irqchip(kvm),
1330                         sizeof(struct kvm_ioapic_state));
1331                 break;
1332         default:
1333                 r = -EINVAL;
1334                 break;
1335         }
1336         return r;
1337 }
1338
1339 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1340 {
1341         int r;
1342
1343         r = 0;
1344         switch (chip->chip_id) {
1345         case KVM_IRQCHIP_PIC_MASTER:
1346                 memcpy(&pic_irqchip(kvm)->pics[0],
1347                         &chip->chip.pic,
1348                         sizeof(struct kvm_pic_state));
1349                 break;
1350         case KVM_IRQCHIP_PIC_SLAVE:
1351                 memcpy(&pic_irqchip(kvm)->pics[1],
1352                         &chip->chip.pic,
1353                         sizeof(struct kvm_pic_state));
1354                 break;
1355         case KVM_IRQCHIP_IOAPIC:
1356                 memcpy(ioapic_irqchip(kvm),
1357                         &chip->chip.ioapic,
1358                         sizeof(struct kvm_ioapic_state));
1359                 break;
1360         default:
1361                 r = -EINVAL;
1362                 break;
1363         }
1364         kvm_pic_update_irq(pic_irqchip(kvm));
1365         return r;
1366 }
1367
1368 /*
1369  * Get (and clear) the dirty memory log for a memory slot.
1370  */
1371 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1372                                       struct kvm_dirty_log *log)
1373 {
1374         int r;
1375         int n;
1376         struct kvm_memory_slot *memslot;
1377         int is_dirty = 0;
1378
1379         down_write(&kvm->slots_lock);
1380
1381         r = kvm_get_dirty_log(kvm, log, &is_dirty);
1382         if (r)
1383                 goto out;
1384
1385         /* If nothing is dirty, don't bother messing with page tables. */
1386         if (is_dirty) {
1387                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1388                 kvm_flush_remote_tlbs(kvm);
1389                 memslot = &kvm->memslots[log->slot];
1390                 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1391                 memset(memslot->dirty_bitmap, 0, n);
1392         }
1393         r = 0;
1394 out:
1395         up_write(&kvm->slots_lock);
1396         return r;
1397 }
1398
1399 long kvm_arch_vm_ioctl(struct file *filp,
1400                        unsigned int ioctl, unsigned long arg)
1401 {
1402         struct kvm *kvm = filp->private_data;
1403         void __user *argp = (void __user *)arg;
1404         int r = -EINVAL;
1405
1406         switch (ioctl) {
1407         case KVM_SET_TSS_ADDR:
1408                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1409                 if (r < 0)
1410                         goto out;
1411                 break;
1412         case KVM_SET_MEMORY_REGION: {
1413                 struct kvm_memory_region kvm_mem;
1414                 struct kvm_userspace_memory_region kvm_userspace_mem;
1415
1416                 r = -EFAULT;
1417                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1418                         goto out;
1419                 kvm_userspace_mem.slot = kvm_mem.slot;
1420                 kvm_userspace_mem.flags = kvm_mem.flags;
1421                 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1422                 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1423                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1424                 if (r)
1425                         goto out;
1426                 break;
1427         }
1428         case KVM_SET_NR_MMU_PAGES:
1429                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1430                 if (r)
1431                         goto out;
1432                 break;
1433         case KVM_GET_NR_MMU_PAGES:
1434                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1435                 break;
1436         case KVM_SET_MEMORY_ALIAS: {
1437                 struct kvm_memory_alias alias;
1438
1439                 r = -EFAULT;
1440                 if (copy_from_user(&alias, argp, sizeof alias))
1441                         goto out;
1442                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1443                 if (r)
1444                         goto out;
1445                 break;
1446         }
1447         case KVM_CREATE_IRQCHIP:
1448                 r = -ENOMEM;
1449                 kvm->arch.vpic = kvm_create_pic(kvm);
1450                 if (kvm->arch.vpic) {
1451                         r = kvm_ioapic_init(kvm);
1452                         if (r) {
1453                                 kfree(kvm->arch.vpic);
1454                                 kvm->arch.vpic = NULL;
1455                                 goto out;
1456                         }
1457                 } else
1458                         goto out;
1459                 break;
1460         case KVM_IRQ_LINE: {
1461                 struct kvm_irq_level irq_event;
1462
1463                 r = -EFAULT;
1464                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1465                         goto out;
1466                 if (irqchip_in_kernel(kvm)) {
1467                         mutex_lock(&kvm->lock);
1468                         if (irq_event.irq < 16)
1469                                 kvm_pic_set_irq(pic_irqchip(kvm),
1470                                         irq_event.irq,
1471                                         irq_event.level);
1472                         kvm_ioapic_set_irq(kvm->arch.vioapic,
1473                                         irq_event.irq,
1474                                         irq_event.level);
1475                         mutex_unlock(&kvm->lock);
1476                         r = 0;
1477                 }
1478                 break;
1479         }
1480         case KVM_GET_IRQCHIP: {
1481                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1482                 struct kvm_irqchip chip;
1483
1484                 r = -EFAULT;
1485                 if (copy_from_user(&chip, argp, sizeof chip))
1486                         goto out;
1487                 r = -ENXIO;
1488                 if (!irqchip_in_kernel(kvm))
1489                         goto out;
1490                 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1491                 if (r)
1492                         goto out;
1493                 r = -EFAULT;
1494                 if (copy_to_user(argp, &chip, sizeof chip))
1495                         goto out;
1496                 r = 0;
1497                 break;
1498         }
1499         case KVM_SET_IRQCHIP: {
1500                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1501                 struct kvm_irqchip chip;
1502
1503                 r = -EFAULT;
1504                 if (copy_from_user(&chip, argp, sizeof chip))
1505                         goto out;
1506                 r = -ENXIO;
1507                 if (!irqchip_in_kernel(kvm))
1508                         goto out;
1509                 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1510                 if (r)
1511                         goto out;
1512                 r = 0;
1513                 break;
1514         }
1515         default:
1516                 ;
1517         }
1518 out:
1519         return r;
1520 }
1521
1522 static void kvm_init_msr_list(void)
1523 {
1524         u32 dummy[2];
1525         unsigned i, j;
1526
1527         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1528                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1529                         continue;
1530                 if (j < i)
1531                         msrs_to_save[j] = msrs_to_save[i];
1532                 j++;
1533         }
1534         num_msrs_to_save = j;
1535 }
1536
1537 /*
1538  * Only apic need an MMIO device hook, so shortcut now..
1539  */
1540 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1541                                                 gpa_t addr)
1542 {
1543         struct kvm_io_device *dev;
1544
1545         if (vcpu->arch.apic) {
1546                 dev = &vcpu->arch.apic->dev;
1547                 if (dev->in_range(dev, addr))
1548                         return dev;
1549         }
1550         return NULL;
1551 }
1552
1553
1554 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1555                                                 gpa_t addr)
1556 {
1557         struct kvm_io_device *dev;
1558
1559         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1560         if (dev == NULL)
1561                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1562         return dev;
1563 }
1564
1565 int emulator_read_std(unsigned long addr,
1566                              void *val,
1567                              unsigned int bytes,
1568                              struct kvm_vcpu *vcpu)
1569 {
1570         void *data = val;
1571         int r = X86EMUL_CONTINUE;
1572
1573         down_read(&vcpu->kvm->slots_lock);
1574         while (bytes) {
1575                 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1576                 unsigned offset = addr & (PAGE_SIZE-1);
1577                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1578                 int ret;
1579
1580                 if (gpa == UNMAPPED_GVA) {
1581                         r = X86EMUL_PROPAGATE_FAULT;
1582                         goto out;
1583                 }
1584                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1585                 if (ret < 0) {
1586                         r = X86EMUL_UNHANDLEABLE;
1587                         goto out;
1588                 }
1589
1590                 bytes -= tocopy;
1591                 data += tocopy;
1592                 addr += tocopy;
1593         }
1594 out:
1595         up_read(&vcpu->kvm->slots_lock);
1596         return r;
1597 }
1598 EXPORT_SYMBOL_GPL(emulator_read_std);
1599
1600 static int emulator_read_emulated(unsigned long addr,
1601                                   void *val,
1602                                   unsigned int bytes,
1603                                   struct kvm_vcpu *vcpu)
1604 {
1605         struct kvm_io_device *mmio_dev;
1606         gpa_t                 gpa;
1607
1608         if (vcpu->mmio_read_completed) {
1609                 memcpy(val, vcpu->mmio_data, bytes);
1610                 vcpu->mmio_read_completed = 0;
1611                 return X86EMUL_CONTINUE;
1612         }
1613
1614         down_read(&vcpu->kvm->slots_lock);
1615         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1616         up_read(&vcpu->kvm->slots_lock);
1617
1618         /* For APIC access vmexit */
1619         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1620                 goto mmio;
1621
1622         if (emulator_read_std(addr, val, bytes, vcpu)
1623                         == X86EMUL_CONTINUE)
1624                 return X86EMUL_CONTINUE;
1625         if (gpa == UNMAPPED_GVA)
1626                 return X86EMUL_PROPAGATE_FAULT;
1627
1628 mmio:
1629         /*
1630          * Is this MMIO handled locally?
1631          */
1632         mutex_lock(&vcpu->kvm->lock);
1633         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1634         if (mmio_dev) {
1635                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1636                 mutex_unlock(&vcpu->kvm->lock);
1637                 return X86EMUL_CONTINUE;
1638         }
1639         mutex_unlock(&vcpu->kvm->lock);
1640
1641         vcpu->mmio_needed = 1;
1642         vcpu->mmio_phys_addr = gpa;
1643         vcpu->mmio_size = bytes;
1644         vcpu->mmio_is_write = 0;
1645
1646         return X86EMUL_UNHANDLEABLE;
1647 }
1648
1649 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1650                                const void *val, int bytes)
1651 {
1652         int ret;
1653
1654         down_read(&vcpu->kvm->slots_lock);
1655         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1656         if (ret < 0) {
1657                 up_read(&vcpu->kvm->slots_lock);
1658                 return 0;
1659         }
1660         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1661         up_read(&vcpu->kvm->slots_lock);
1662         return 1;
1663 }
1664
1665 static int emulator_write_emulated_onepage(unsigned long addr,
1666                                            const void *val,
1667                                            unsigned int bytes,
1668                                            struct kvm_vcpu *vcpu)
1669 {
1670         struct kvm_io_device *mmio_dev;
1671         gpa_t                 gpa;
1672
1673         down_read(&vcpu->kvm->slots_lock);
1674         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1675         up_read(&vcpu->kvm->slots_lock);
1676
1677         if (gpa == UNMAPPED_GVA) {
1678                 kvm_inject_page_fault(vcpu, addr, 2);
1679                 return X86EMUL_PROPAGATE_FAULT;
1680         }
1681
1682         /* For APIC access vmexit */
1683         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1684                 goto mmio;
1685
1686         if (emulator_write_phys(vcpu, gpa, val, bytes))
1687                 return X86EMUL_CONTINUE;
1688
1689 mmio:
1690         /*
1691          * Is this MMIO handled locally?
1692          */
1693         mutex_lock(&vcpu->kvm->lock);
1694         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1695         if (mmio_dev) {
1696                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1697                 mutex_unlock(&vcpu->kvm->lock);
1698                 return X86EMUL_CONTINUE;
1699         }
1700         mutex_unlock(&vcpu->kvm->lock);
1701
1702         vcpu->mmio_needed = 1;
1703         vcpu->mmio_phys_addr = gpa;
1704         vcpu->mmio_size = bytes;
1705         vcpu->mmio_is_write = 1;
1706         memcpy(vcpu->mmio_data, val, bytes);
1707
1708         return X86EMUL_CONTINUE;
1709 }
1710
1711 int emulator_write_emulated(unsigned long addr,
1712                                    const void *val,
1713                                    unsigned int bytes,
1714                                    struct kvm_vcpu *vcpu)
1715 {
1716         /* Crossing a page boundary? */
1717         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1718                 int rc, now;
1719
1720                 now = -addr & ~PAGE_MASK;
1721                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1722                 if (rc != X86EMUL_CONTINUE)
1723                         return rc;
1724                 addr += now;
1725                 val += now;
1726                 bytes -= now;
1727         }
1728         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1729 }
1730 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1731
1732 static int emulator_cmpxchg_emulated(unsigned long addr,
1733                                      const void *old,
1734                                      const void *new,
1735                                      unsigned int bytes,
1736                                      struct kvm_vcpu *vcpu)
1737 {
1738         static int reported;
1739
1740         if (!reported) {
1741                 reported = 1;
1742                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1743         }
1744 #ifndef CONFIG_X86_64
1745         /* guests cmpxchg8b have to be emulated atomically */
1746         if (bytes == 8) {
1747                 gpa_t gpa;
1748                 struct page *page;
1749                 char *kaddr;
1750                 u64 val;
1751
1752                 down_read(&vcpu->kvm->slots_lock);
1753                 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1754
1755                 if (gpa == UNMAPPED_GVA ||
1756                    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1757                         goto emul_write;
1758
1759                 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1760                         goto emul_write;
1761
1762                 val = *(u64 *)new;
1763
1764                 down_read(&current->mm->mmap_sem);
1765                 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1766                 up_read(&current->mm->mmap_sem);
1767
1768                 kaddr = kmap_atomic(page, KM_USER0);
1769                 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
1770                 kunmap_atomic(kaddr, KM_USER0);
1771                 kvm_release_page_dirty(page);
1772         emul_write:
1773                 up_read(&vcpu->kvm->slots_lock);
1774         }
1775 #endif
1776
1777         return emulator_write_emulated(addr, new, bytes, vcpu);
1778 }
1779
1780 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1781 {
1782         return kvm_x86_ops->get_segment_base(vcpu, seg);
1783 }
1784
1785 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1786 {
1787         return X86EMUL_CONTINUE;
1788 }
1789
1790 int emulate_clts(struct kvm_vcpu *vcpu)
1791 {
1792         kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1793         return X86EMUL_CONTINUE;
1794 }
1795
1796 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1797 {
1798         struct kvm_vcpu *vcpu = ctxt->vcpu;
1799
1800         switch (dr) {
1801         case 0 ... 3:
1802                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1803                 return X86EMUL_CONTINUE;
1804         default:
1805                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1806                 return X86EMUL_UNHANDLEABLE;
1807         }
1808 }
1809
1810 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1811 {
1812         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1813         int exception;
1814
1815         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1816         if (exception) {
1817                 /* FIXME: better handling */
1818                 return X86EMUL_UNHANDLEABLE;
1819         }
1820         return X86EMUL_CONTINUE;
1821 }
1822
1823 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1824 {
1825         static int reported;
1826         u8 opcodes[4];
1827         unsigned long rip = vcpu->arch.rip;
1828         unsigned long rip_linear;
1829
1830         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1831
1832         if (reported)
1833                 return;
1834
1835         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1836
1837         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1838                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1839         reported = 1;
1840 }
1841 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1842
1843 struct x86_emulate_ops emulate_ops = {
1844         .read_std            = emulator_read_std,
1845         .read_emulated       = emulator_read_emulated,
1846         .write_emulated      = emulator_write_emulated,
1847         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1848 };
1849
1850 int emulate_instruction(struct kvm_vcpu *vcpu,
1851                         struct kvm_run *run,
1852                         unsigned long cr2,
1853                         u16 error_code,
1854                         int emulation_type)
1855 {
1856         int r;
1857         struct decode_cache *c;
1858
1859         vcpu->arch.mmio_fault_cr2 = cr2;
1860         kvm_x86_ops->cache_regs(vcpu);
1861
1862         vcpu->mmio_is_write = 0;
1863         vcpu->arch.pio.string = 0;
1864
1865         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
1866                 int cs_db, cs_l;
1867                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1868
1869                 vcpu->arch.emulate_ctxt.vcpu = vcpu;
1870                 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1871                 vcpu->arch.emulate_ctxt.mode =
1872                         (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
1873                         ? X86EMUL_MODE_REAL : cs_l
1874                         ? X86EMUL_MODE_PROT64 : cs_db
1875                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1876
1877                 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1878                         vcpu->arch.emulate_ctxt.cs_base = 0;
1879                         vcpu->arch.emulate_ctxt.ds_base = 0;
1880                         vcpu->arch.emulate_ctxt.es_base = 0;
1881                         vcpu->arch.emulate_ctxt.ss_base = 0;
1882                 } else {
1883                         vcpu->arch.emulate_ctxt.cs_base =
1884                                         get_segment_base(vcpu, VCPU_SREG_CS);
1885                         vcpu->arch.emulate_ctxt.ds_base =
1886                                         get_segment_base(vcpu, VCPU_SREG_DS);
1887                         vcpu->arch.emulate_ctxt.es_base =
1888                                         get_segment_base(vcpu, VCPU_SREG_ES);
1889                         vcpu->arch.emulate_ctxt.ss_base =
1890                                         get_segment_base(vcpu, VCPU_SREG_SS);
1891                 }
1892
1893                 vcpu->arch.emulate_ctxt.gs_base =
1894                                         get_segment_base(vcpu, VCPU_SREG_GS);
1895                 vcpu->arch.emulate_ctxt.fs_base =
1896                                         get_segment_base(vcpu, VCPU_SREG_FS);
1897
1898                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
1899
1900                 /* Reject the instructions other than VMCALL/VMMCALL when
1901                  * try to emulate invalid opcode */
1902                 c = &vcpu->arch.emulate_ctxt.decode;
1903                 if ((emulation_type & EMULTYPE_TRAP_UD) &&
1904                     (!(c->twobyte && c->b == 0x01 &&
1905                       (c->modrm_reg == 0 || c->modrm_reg == 3) &&
1906                        c->modrm_mod == 3 && c->modrm_rm == 1)))
1907                         return EMULATE_FAIL;
1908
1909                 ++vcpu->stat.insn_emulation;
1910                 if (r)  {
1911                         ++vcpu->stat.insn_emulation_fail;
1912                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1913                                 return EMULATE_DONE;
1914                         return EMULATE_FAIL;
1915                 }
1916         }
1917
1918         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
1919
1920         if (vcpu->arch.pio.string)
1921                 return EMULATE_DO_MMIO;
1922
1923         if ((r || vcpu->mmio_is_write) && run) {
1924                 run->exit_reason = KVM_EXIT_MMIO;
1925                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1926                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1927                 run->mmio.len = vcpu->mmio_size;
1928                 run->mmio.is_write = vcpu->mmio_is_write;
1929         }
1930
1931         if (r) {
1932                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1933                         return EMULATE_DONE;
1934                 if (!vcpu->mmio_needed) {
1935                         kvm_report_emulation_failure(vcpu, "mmio");
1936                         return EMULATE_FAIL;
1937                 }
1938                 return EMULATE_DO_MMIO;
1939         }
1940
1941         kvm_x86_ops->decache_regs(vcpu);
1942         kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
1943
1944         if (vcpu->mmio_is_write) {
1945                 vcpu->mmio_needed = 0;
1946                 return EMULATE_DO_MMIO;
1947         }
1948
1949         return EMULATE_DONE;
1950 }
1951 EXPORT_SYMBOL_GPL(emulate_instruction);
1952
1953 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1954 {
1955         int i;
1956
1957         for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
1958                 if (vcpu->arch.pio.guest_pages[i]) {
1959                         kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
1960                         vcpu->arch.pio.guest_pages[i] = NULL;
1961                 }
1962 }
1963
1964 static int pio_copy_data(struct kvm_vcpu *vcpu)
1965 {
1966         void *p = vcpu->arch.pio_data;
1967         void *q;
1968         unsigned bytes;
1969         int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
1970
1971         q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1972                  PAGE_KERNEL);
1973         if (!q) {
1974                 free_pio_guest_pages(vcpu);
1975                 return -ENOMEM;
1976         }
1977         q += vcpu->arch.pio.guest_page_offset;
1978         bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
1979         if (vcpu->arch.pio.in)
1980                 memcpy(q, p, bytes);
1981         else
1982                 memcpy(p, q, bytes);
1983         q -= vcpu->arch.pio.guest_page_offset;
1984         vunmap(q);
1985         free_pio_guest_pages(vcpu);
1986         return 0;
1987 }
1988
1989 int complete_pio(struct kvm_vcpu *vcpu)
1990 {
1991         struct kvm_pio_request *io = &vcpu->arch.pio;
1992         long delta;
1993         int r;
1994
1995         kvm_x86_ops->cache_regs(vcpu);
1996
1997         if (!io->string) {
1998                 if (io->in)
1999                         memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
2000                                io->size);
2001         } else {
2002                 if (io->in) {
2003                         r = pio_copy_data(vcpu);
2004                         if (r) {
2005                                 kvm_x86_ops->cache_regs(vcpu);
2006                                 return r;
2007                         }
2008                 }
2009
2010                 delta = 1;
2011                 if (io->rep) {
2012                         delta *= io->cur_count;
2013                         /*
2014                          * The size of the register should really depend on
2015                          * current address size.
2016                          */
2017                         vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
2018                 }
2019                 if (io->down)
2020                         delta = -delta;
2021                 delta *= io->size;
2022                 if (io->in)
2023                         vcpu->arch.regs[VCPU_REGS_RDI] += delta;
2024                 else
2025                         vcpu->arch.regs[VCPU_REGS_RSI] += delta;
2026         }
2027
2028         kvm_x86_ops->decache_regs(vcpu);
2029
2030         io->count -= io->cur_count;
2031         io->cur_count = 0;
2032
2033         return 0;
2034 }
2035
2036 static void kernel_pio(struct kvm_io_device *pio_dev,
2037                        struct kvm_vcpu *vcpu,
2038                        void *pd)
2039 {
2040         /* TODO: String I/O for in kernel device */
2041
2042         mutex_lock(&vcpu->kvm->lock);
2043         if (vcpu->arch.pio.in)
2044                 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2045                                   vcpu->arch.pio.size,
2046                                   pd);
2047         else
2048                 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2049                                    vcpu->arch.pio.size,
2050                                    pd);
2051         mutex_unlock(&vcpu->kvm->lock);
2052 }
2053
2054 static void pio_string_write(struct kvm_io_device *pio_dev,
2055                              struct kvm_vcpu *vcpu)
2056 {
2057         struct kvm_pio_request *io = &vcpu->arch.pio;
2058         void *pd = vcpu->arch.pio_data;
2059         int i;
2060
2061         mutex_lock(&vcpu->kvm->lock);
2062         for (i = 0; i < io->cur_count; i++) {
2063                 kvm_iodevice_write(pio_dev, io->port,
2064                                    io->size,
2065                                    pd);
2066                 pd += io->size;
2067         }
2068         mutex_unlock(&vcpu->kvm->lock);
2069 }
2070
2071 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2072                                                gpa_t addr)
2073 {
2074         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
2075 }
2076
2077 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2078                   int size, unsigned port)
2079 {
2080         struct kvm_io_device *pio_dev;
2081
2082         vcpu->run->exit_reason = KVM_EXIT_IO;
2083         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2084         vcpu->run->io.size = vcpu->arch.pio.size = size;
2085         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2086         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2087         vcpu->run->io.port = vcpu->arch.pio.port = port;
2088         vcpu->arch.pio.in = in;
2089         vcpu->arch.pio.string = 0;
2090         vcpu->arch.pio.down = 0;
2091         vcpu->arch.pio.guest_page_offset = 0;
2092         vcpu->arch.pio.rep = 0;
2093
2094         kvm_x86_ops->cache_regs(vcpu);
2095         memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2096         kvm_x86_ops->decache_regs(vcpu);
2097
2098         kvm_x86_ops->skip_emulated_instruction(vcpu);
2099
2100         pio_dev = vcpu_find_pio_dev(vcpu, port);
2101         if (pio_dev) {
2102                 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2103                 complete_pio(vcpu);
2104                 return 1;
2105         }
2106         return 0;
2107 }
2108 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2109
2110 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2111                   int size, unsigned long count, int down,
2112                   gva_t address, int rep, unsigned port)
2113 {
2114         unsigned now, in_page;
2115         int i, ret = 0;
2116         int nr_pages = 1;
2117         struct page *page;
2118         struct kvm_io_device *pio_dev;
2119
2120         vcpu->run->exit_reason = KVM_EXIT_IO;
2121         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2122         vcpu->run->io.size = vcpu->arch.pio.size = size;
2123         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2124         vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2125         vcpu->run->io.port = vcpu->arch.pio.port = port;
2126         vcpu->arch.pio.in = in;
2127         vcpu->arch.pio.string = 1;
2128         vcpu->arch.pio.down = down;
2129         vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2130         vcpu->arch.pio.rep = rep;
2131
2132         if (!count) {
2133                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2134                 return 1;
2135         }
2136
2137         if (!down)
2138                 in_page = PAGE_SIZE - offset_in_page(address);
2139         else
2140                 in_page = offset_in_page(address) + size;
2141         now = min(count, (unsigned long)in_page / size);
2142         if (!now) {
2143                 /*
2144                  * String I/O straddles page boundary.  Pin two guest pages
2145                  * so that we satisfy atomicity constraints.  Do just one
2146                  * transaction to avoid complexity.
2147                  */
2148                 nr_pages = 2;
2149                 now = 1;
2150         }
2151         if (down) {
2152                 /*
2153                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
2154                  */
2155                 pr_unimpl(vcpu, "guest string pio down\n");
2156                 kvm_inject_gp(vcpu, 0);
2157                 return 1;
2158         }
2159         vcpu->run->io.count = now;
2160         vcpu->arch.pio.cur_count = now;
2161
2162         if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2163                 kvm_x86_ops->skip_emulated_instruction(vcpu);
2164
2165         for (i = 0; i < nr_pages; ++i) {
2166                 down_read(&vcpu->kvm->slots_lock);
2167                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2168                 vcpu->arch.pio.guest_pages[i] = page;
2169                 up_read(&vcpu->kvm->slots_lock);
2170                 if (!page) {
2171                         kvm_inject_gp(vcpu, 0);
2172                         free_pio_guest_pages(vcpu);
2173                         return 1;
2174                 }
2175         }
2176
2177         pio_dev = vcpu_find_pio_dev(vcpu, port);
2178         if (!vcpu->arch.pio.in) {
2179                 /* string PIO write */
2180                 ret = pio_copy_data(vcpu);
2181                 if (ret >= 0 && pio_dev) {
2182                         pio_string_write(pio_dev, vcpu);
2183                         complete_pio(vcpu);
2184                         if (vcpu->arch.pio.count == 0)
2185                                 ret = 1;
2186                 }
2187         } else if (pio_dev)
2188                 pr_unimpl(vcpu, "no string pio read support yet, "
2189                        "port %x size %d count %ld\n",
2190                         port, size, count);
2191
2192         return ret;
2193 }
2194 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2195
2196 int kvm_arch_init(void *opaque)
2197 {
2198         int r;
2199         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2200
2201         if (kvm_x86_ops) {
2202                 printk(KERN_ERR "kvm: already loaded the other module\n");
2203                 r = -EEXIST;
2204                 goto out;
2205         }
2206
2207         if (!ops->cpu_has_kvm_support()) {
2208                 printk(KERN_ERR "kvm: no hardware support\n");
2209                 r = -EOPNOTSUPP;
2210                 goto out;
2211         }
2212         if (ops->disabled_by_bios()) {
2213                 printk(KERN_ERR "kvm: disabled by bios\n");
2214                 r = -EOPNOTSUPP;
2215                 goto out;
2216         }
2217
2218         r = kvm_mmu_module_init();
2219         if (r)
2220                 goto out;
2221
2222         kvm_init_msr_list();
2223
2224         kvm_x86_ops = ops;
2225         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2226         return 0;
2227
2228 out:
2229         return r;
2230 }
2231
2232 void kvm_arch_exit(void)
2233 {
2234         kvm_x86_ops = NULL;
2235         kvm_mmu_module_exit();
2236 }
2237
2238 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2239 {
2240         ++vcpu->stat.halt_exits;
2241         if (irqchip_in_kernel(vcpu->kvm)) {
2242                 vcpu->arch.mp_state = VCPU_MP_STATE_HALTED;
2243                 kvm_vcpu_block(vcpu);
2244                 if (vcpu->arch.mp_state != VCPU_MP_STATE_RUNNABLE)
2245                         return -EINTR;
2246                 return 1;
2247         } else {
2248                 vcpu->run->exit_reason = KVM_EXIT_HLT;
2249                 return 0;
2250         }
2251 }
2252 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2253
2254 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2255 {
2256         unsigned long nr, a0, a1, a2, a3, ret;
2257
2258         kvm_x86_ops->cache_regs(vcpu);
2259
2260         nr = vcpu->arch.regs[VCPU_REGS_RAX];
2261         a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2262         a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2263         a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2264         a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2265
2266         if (!is_long_mode(vcpu)) {
2267                 nr &= 0xFFFFFFFF;
2268                 a0 &= 0xFFFFFFFF;
2269                 a1 &= 0xFFFFFFFF;
2270                 a2 &= 0xFFFFFFFF;
2271                 a3 &= 0xFFFFFFFF;
2272         }
2273
2274         switch (nr) {
2275         case KVM_HC_VAPIC_POLL_IRQ:
2276                 ret = 0;
2277                 break;
2278         default:
2279                 ret = -KVM_ENOSYS;
2280                 break;
2281         }
2282         vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2283         kvm_x86_ops->decache_regs(vcpu);
2284         return 0;
2285 }
2286 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2287
2288 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2289 {
2290         char instruction[3];
2291         int ret = 0;
2292
2293
2294         /*
2295          * Blow out the MMU to ensure that no other VCPU has an active mapping
2296          * to ensure that the updated hypercall appears atomically across all
2297          * VCPUs.
2298          */
2299         kvm_mmu_zap_all(vcpu->kvm);
2300
2301         kvm_x86_ops->cache_regs(vcpu);
2302         kvm_x86_ops->patch_hypercall(vcpu, instruction);
2303         if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2304             != X86EMUL_CONTINUE)
2305                 ret = -EFAULT;
2306
2307         return ret;
2308 }
2309
2310 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2311 {
2312         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2313 }
2314
2315 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2316 {
2317         struct descriptor_table dt = { limit, base };
2318
2319         kvm_x86_ops->set_gdt(vcpu, &dt);
2320 }
2321
2322 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2323 {
2324         struct descriptor_table dt = { limit, base };
2325
2326         kvm_x86_ops->set_idt(vcpu, &dt);
2327 }
2328
2329 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2330                    unsigned long *rflags)
2331 {
2332         lmsw(vcpu, msw);
2333         *rflags = kvm_x86_ops->get_rflags(vcpu);
2334 }
2335
2336 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2337 {
2338         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2339         switch (cr) {
2340         case 0:
2341                 return vcpu->arch.cr0;
2342         case 2:
2343                 return vcpu->arch.cr2;
2344         case 3:
2345                 return vcpu->arch.cr3;
2346         case 4:
2347                 return vcpu->arch.cr4;
2348         case 8:
2349                 return get_cr8(vcpu);
2350         default:
2351                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2352                 return 0;
2353         }
2354 }
2355
2356 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2357                      unsigned long *rflags)
2358 {
2359         switch (cr) {
2360         case 0:
2361                 set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2362                 *rflags = kvm_x86_ops->get_rflags(vcpu);
2363                 break;
2364         case 2:
2365                 vcpu->arch.cr2 = val;
2366                 break;
2367         case 3:
2368                 set_cr3(vcpu, val);
2369                 break;
2370         case 4:
2371                 set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2372                 break;
2373         case 8:
2374                 set_cr8(vcpu, val & 0xfUL);
2375                 break;
2376         default:
2377                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2378         }
2379 }
2380
2381 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2382 {
2383         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2384         int j, nent = vcpu->arch.cpuid_nent;
2385
2386         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2387         /* when no next entry is found, the current entry[i] is reselected */
2388         for (j = i + 1; j == i; j = (j + 1) % nent) {
2389                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2390                 if (ej->function == e->function) {
2391                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2392                         return j;
2393                 }
2394         }
2395         return 0; /* silence gcc, even though control never reaches here */
2396 }
2397
2398 /* find an entry with matching function, matching index (if needed), and that
2399  * should be read next (if it's stateful) */
2400 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2401         u32 function, u32 index)
2402 {
2403         if (e->function != function)
2404                 return 0;
2405         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2406                 return 0;
2407         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2408                 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2409                 return 0;
2410         return 1;
2411 }
2412
2413 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2414 {
2415         int i;
2416         u32 function, index;
2417         struct kvm_cpuid_entry2 *e, *best;
2418
2419         kvm_x86_ops->cache_regs(vcpu);
2420         function = vcpu->arch.regs[VCPU_REGS_RAX];
2421         index = vcpu->arch.regs[VCPU_REGS_RCX];
2422         vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2423         vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2424         vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2425         vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2426         best = NULL;
2427         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2428                 e = &vcpu->arch.cpuid_entries[i];
2429                 if (is_matching_cpuid_entry(e, function, index)) {
2430                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2431                                 move_to_next_stateful_cpuid_entry(vcpu, i);
2432                         best = e;
2433                         break;
2434                 }
2435                 /*
2436                  * Both basic or both extended?
2437                  */
2438                 if (((e->function ^ function) & 0x80000000) == 0)
2439                         if (!best || e->function > best->function)
2440                                 best = e;
2441         }
2442         if (best) {
2443                 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2444                 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2445                 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2446                 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2447         }
2448         kvm_x86_ops->decache_regs(vcpu);
2449         kvm_x86_ops->skip_emulated_instruction(vcpu);
2450 }
2451 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2452
2453 /*
2454  * Check if userspace requested an interrupt window, and that the
2455  * interrupt window is open.
2456  *
2457  * No need to exit to userspace if we already have an interrupt queued.
2458  */
2459 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2460                                           struct kvm_run *kvm_run)
2461 {
2462         return (!vcpu->arch.irq_summary &&
2463                 kvm_run->request_interrupt_window &&
2464                 vcpu->arch.interrupt_window_open &&
2465                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2466 }
2467
2468 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2469                               struct kvm_run *kvm_run)
2470 {
2471         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2472         kvm_run->cr8 = get_cr8(vcpu);
2473         kvm_run->apic_base = kvm_get_apic_base(vcpu);
2474         if (irqchip_in_kernel(vcpu->kvm))
2475                 kvm_run->ready_for_interrupt_injection = 1;
2476         else
2477                 kvm_run->ready_for_interrupt_injection =
2478                                         (vcpu->arch.interrupt_window_open &&
2479                                          vcpu->arch.irq_summary == 0);
2480 }
2481
2482 static void vapic_enter(struct kvm_vcpu *vcpu)
2483 {
2484         struct kvm_lapic *apic = vcpu->arch.apic;
2485         struct page *page;
2486
2487         if (!apic || !apic->vapic_addr)
2488                 return;
2489
2490         down_read(&current->mm->mmap_sem);
2491         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2492         up_read(&current->mm->mmap_sem);
2493
2494         vcpu->arch.apic->vapic_page = page;
2495 }
2496
2497 static void vapic_exit(struct kvm_vcpu *vcpu)
2498 {
2499         struct kvm_lapic *apic = vcpu->arch.apic;
2500
2501         if (!apic || !apic->vapic_addr)
2502                 return;
2503
2504         kvm_release_page_dirty(apic->vapic_page);
2505         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
2506 }
2507
2508 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2509 {
2510         int r;
2511
2512         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2513                 pr_debug("vcpu %d received sipi with vector # %x\n",
2514                        vcpu->vcpu_id, vcpu->arch.sipi_vector);
2515                 kvm_lapic_reset(vcpu);
2516                 r = kvm_x86_ops->vcpu_reset(vcpu);
2517                 if (r)
2518                         return r;
2519                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
2520         }
2521
2522         vapic_enter(vcpu);
2523
2524 preempted:
2525         if (vcpu->guest_debug.enabled)
2526                 kvm_x86_ops->guest_debug_pre(vcpu);
2527
2528 again:
2529         r = kvm_mmu_reload(vcpu);
2530         if (unlikely(r))
2531                 goto out;
2532
2533         if (vcpu->requests) {
2534                 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2535                         __kvm_migrate_apic_timer(vcpu);
2536                 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
2537                                        &vcpu->requests)) {
2538                         kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
2539                         r = 0;
2540                         goto out;
2541                 }
2542         }
2543
2544         kvm_inject_pending_timer_irqs(vcpu);
2545
2546         preempt_disable();
2547
2548         kvm_x86_ops->prepare_guest_switch(vcpu);
2549         kvm_load_guest_fpu(vcpu);
2550
2551         local_irq_disable();
2552
2553         if (need_resched()) {
2554                 local_irq_enable();
2555                 preempt_enable();
2556                 r = 1;
2557                 goto out;
2558         }
2559
2560         if (signal_pending(current)) {
2561                 local_irq_enable();
2562                 preempt_enable();
2563                 r = -EINTR;
2564                 kvm_run->exit_reason = KVM_EXIT_INTR;
2565                 ++vcpu->stat.signal_exits;
2566                 goto out;
2567         }
2568
2569         if (vcpu->arch.exception.pending)
2570                 __queue_exception(vcpu);
2571         else if (irqchip_in_kernel(vcpu->kvm))
2572                 kvm_x86_ops->inject_pending_irq(vcpu);
2573         else
2574                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2575
2576         kvm_lapic_sync_to_vapic(vcpu);
2577
2578         vcpu->guest_mode = 1;
2579         kvm_guest_enter();
2580
2581         if (vcpu->requests)
2582                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2583                         kvm_x86_ops->tlb_flush(vcpu);
2584
2585         kvm_x86_ops->run(vcpu, kvm_run);
2586
2587         vcpu->guest_mode = 0;
2588         local_irq_enable();
2589
2590         ++vcpu->stat.exits;
2591
2592         /*
2593          * We must have an instruction between local_irq_enable() and
2594          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2595          * the interrupt shadow.  The stat.exits increment will do nicely.
2596          * But we need to prevent reordering, hence this barrier():
2597          */
2598         barrier();
2599
2600         kvm_guest_exit();
2601
2602         preempt_enable();
2603
2604         /*
2605          * Profile KVM exit RIPs:
2606          */
2607         if (unlikely(prof_on == KVM_PROFILING)) {
2608                 kvm_x86_ops->cache_regs(vcpu);
2609                 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2610         }
2611
2612         if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2613                 vcpu->arch.exception.pending = false;
2614
2615         kvm_lapic_sync_from_vapic(vcpu);
2616
2617         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2618
2619         if (r > 0) {
2620                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2621                         r = -EINTR;
2622                         kvm_run->exit_reason = KVM_EXIT_INTR;
2623                         ++vcpu->stat.request_irq_exits;
2624                         goto out;
2625                 }
2626                 if (!need_resched())
2627                         goto again;
2628         }
2629
2630 out:
2631         if (r > 0) {
2632                 kvm_resched(vcpu);
2633                 goto preempted;
2634         }
2635
2636         post_kvm_run_save(vcpu, kvm_run);
2637
2638         vapic_exit(vcpu);
2639
2640         return r;
2641 }
2642
2643 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2644 {
2645         int r;
2646         sigset_t sigsaved;
2647
2648         vcpu_load(vcpu);
2649
2650         if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2651                 kvm_vcpu_block(vcpu);
2652                 vcpu_put(vcpu);
2653                 return -EAGAIN;
2654         }
2655
2656         if (vcpu->sigset_active)
2657                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2658
2659         /* re-sync apic's tpr */
2660         if (!irqchip_in_kernel(vcpu->kvm))
2661                 set_cr8(vcpu, kvm_run->cr8);
2662
2663         if (vcpu->arch.pio.cur_count) {
2664                 r = complete_pio(vcpu);
2665                 if (r)
2666                         goto out;
2667         }
2668 #if CONFIG_HAS_IOMEM
2669         if (vcpu->mmio_needed) {
2670                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2671                 vcpu->mmio_read_completed = 1;
2672                 vcpu->mmio_needed = 0;
2673                 r = emulate_instruction(vcpu, kvm_run,
2674                                         vcpu->arch.mmio_fault_cr2, 0,
2675                                         EMULTYPE_NO_DECODE);
2676                 if (r == EMULATE_DO_MMIO) {
2677                         /*
2678                          * Read-modify-write.  Back to userspace.
2679                          */
2680                         r = 0;
2681                         goto out;
2682                 }
2683         }
2684 #endif
2685         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2686                 kvm_x86_ops->cache_regs(vcpu);
2687                 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2688                 kvm_x86_ops->decache_regs(vcpu);
2689         }
2690
2691         r = __vcpu_run(vcpu, kvm_run);
2692
2693 out:
2694         if (vcpu->sigset_active)
2695                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2696
2697         vcpu_put(vcpu);
2698         return r;
2699 }
2700
2701 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2702 {
2703         vcpu_load(vcpu);
2704
2705         kvm_x86_ops->cache_regs(vcpu);
2706
2707         regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2708         regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2709         regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2710         regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2711         regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2712         regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2713         regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2714         regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2715 #ifdef CONFIG_X86_64
2716         regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
2717         regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
2718         regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
2719         regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
2720         regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
2721         regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
2722         regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
2723         regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
2724 #endif
2725
2726         regs->rip = vcpu->arch.rip;
2727         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2728
2729         /*
2730          * Don't leak debug flags in case they were set for guest debugging
2731          */
2732         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2733                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2734
2735         vcpu_put(vcpu);
2736
2737         return 0;
2738 }
2739
2740 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2741 {
2742         vcpu_load(vcpu);
2743
2744         vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
2745         vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
2746         vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
2747         vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
2748         vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
2749         vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
2750         vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
2751         vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
2752 #ifdef CONFIG_X86_64
2753         vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
2754         vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
2755         vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
2756         vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
2757         vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
2758         vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
2759         vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
2760         vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
2761 #endif
2762
2763         vcpu->arch.rip = regs->rip;
2764         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2765
2766         kvm_x86_ops->decache_regs(vcpu);
2767
2768         vcpu_put(vcpu);
2769
2770         return 0;
2771 }
2772
2773 static void get_segment(struct kvm_vcpu *vcpu,
2774                         struct kvm_segment *var, int seg)
2775 {
2776         return kvm_x86_ops->get_segment(vcpu, var, seg);
2777 }
2778
2779 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2780 {
2781         struct kvm_segment cs;
2782
2783         get_segment(vcpu, &cs, VCPU_SREG_CS);
2784         *db = cs.db;
2785         *l = cs.l;
2786 }
2787 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2788
2789 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2790                                   struct kvm_sregs *sregs)
2791 {
2792         struct descriptor_table dt;
2793         int pending_vec;
2794
2795         vcpu_load(vcpu);
2796
2797         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2798         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2799         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2800         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2801         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2802         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2803
2804         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2805         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2806
2807         kvm_x86_ops->get_idt(vcpu, &dt);
2808         sregs->idt.limit = dt.limit;
2809         sregs->idt.base = dt.base;
2810         kvm_x86_ops->get_gdt(vcpu, &dt);
2811         sregs->gdt.limit = dt.limit;
2812         sregs->gdt.base = dt.base;
2813
2814         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2815         sregs->cr0 = vcpu->arch.cr0;
2816         sregs->cr2 = vcpu->arch.cr2;
2817         sregs->cr3 = vcpu->arch.cr3;
2818         sregs->cr4 = vcpu->arch.cr4;
2819         sregs->cr8 = get_cr8(vcpu);
2820         sregs->efer = vcpu->arch.shadow_efer;
2821         sregs->apic_base = kvm_get_apic_base(vcpu);
2822
2823         if (irqchip_in_kernel(vcpu->kvm)) {
2824                 memset(sregs->interrupt_bitmap, 0,
2825                        sizeof sregs->interrupt_bitmap);
2826                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2827                 if (pending_vec >= 0)
2828                         set_bit(pending_vec,
2829                                 (unsigned long *)sregs->interrupt_bitmap);
2830         } else
2831                 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
2832                        sizeof sregs->interrupt_bitmap);
2833
2834         vcpu_put(vcpu);
2835
2836         return 0;
2837 }
2838
2839 static void set_segment(struct kvm_vcpu *vcpu,
2840                         struct kvm_segment *var, int seg)
2841 {
2842         return kvm_x86_ops->set_segment(vcpu, var, seg);
2843 }
2844
2845 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2846                                   struct kvm_sregs *sregs)
2847 {
2848         int mmu_reset_needed = 0;
2849         int i, pending_vec, max_bits;
2850         struct descriptor_table dt;
2851
2852         vcpu_load(vcpu);
2853
2854         dt.limit = sregs->idt.limit;
2855         dt.base = sregs->idt.base;
2856         kvm_x86_ops->set_idt(vcpu, &dt);
2857         dt.limit = sregs->gdt.limit;
2858         dt.base = sregs->gdt.base;
2859         kvm_x86_ops->set_gdt(vcpu, &dt);
2860
2861         vcpu->arch.cr2 = sregs->cr2;
2862         mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
2863         vcpu->arch.cr3 = sregs->cr3;
2864
2865         set_cr8(vcpu, sregs->cr8);
2866
2867         mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
2868 #ifdef CONFIG_X86_64
2869         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2870 #endif
2871         kvm_set_apic_base(vcpu, sregs->apic_base);
2872
2873         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2874
2875         mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
2876         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2877         vcpu->arch.cr0 = sregs->cr0;
2878
2879         mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
2880         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2881         if (!is_long_mode(vcpu) && is_pae(vcpu))
2882                 load_pdptrs(vcpu, vcpu->arch.cr3);
2883
2884         if (mmu_reset_needed)
2885                 kvm_mmu_reset_context(vcpu);
2886
2887         if (!irqchip_in_kernel(vcpu->kvm)) {
2888                 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
2889                        sizeof vcpu->arch.irq_pending);
2890                 vcpu->arch.irq_summary = 0;
2891                 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
2892                         if (vcpu->arch.irq_pending[i])
2893                                 __set_bit(i, &vcpu->arch.irq_summary);
2894         } else {
2895                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2896                 pending_vec = find_first_bit(
2897                         (const unsigned long *)sregs->interrupt_bitmap,
2898                         max_bits);
2899                 /* Only pending external irq is handled here */
2900                 if (pending_vec < max_bits) {
2901                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2902                         pr_debug("Set back pending irq %d\n",
2903                                  pending_vec);
2904                 }
2905         }
2906
2907         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2908         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2909         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2910         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2911         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2912         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2913
2914         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2915         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2916
2917         vcpu_put(vcpu);
2918
2919         return 0;
2920 }
2921
2922 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2923                                     struct kvm_debug_guest *dbg)
2924 {
2925         int r;
2926
2927         vcpu_load(vcpu);
2928
2929         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2930
2931         vcpu_put(vcpu);
2932
2933         return r;
2934 }
2935
2936 /*
2937  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2938  * we have asm/x86/processor.h
2939  */
2940 struct fxsave {
2941         u16     cwd;
2942         u16     swd;
2943         u16     twd;
2944         u16     fop;
2945         u64     rip;
2946         u64     rdp;
2947         u32     mxcsr;
2948         u32     mxcsr_mask;
2949         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2950 #ifdef CONFIG_X86_64
2951         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2952 #else
2953         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2954 #endif
2955 };
2956
2957 /*
2958  * Translate a guest virtual address to a guest physical address.
2959  */
2960 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2961                                     struct kvm_translation *tr)
2962 {
2963         unsigned long vaddr = tr->linear_address;
2964         gpa_t gpa;
2965
2966         vcpu_load(vcpu);
2967         down_read(&vcpu->kvm->slots_lock);
2968         gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
2969         up_read(&vcpu->kvm->slots_lock);
2970         tr->physical_address = gpa;
2971         tr->valid = gpa != UNMAPPED_GVA;
2972         tr->writeable = 1;
2973         tr->usermode = 0;
2974         vcpu_put(vcpu);
2975
2976         return 0;
2977 }
2978
2979 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2980 {
2981         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
2982
2983         vcpu_load(vcpu);
2984
2985         memcpy(fpu->fpr, fxsave->st_space, 128);
2986         fpu->fcw = fxsave->cwd;
2987         fpu->fsw = fxsave->swd;
2988         fpu->ftwx = fxsave->twd;
2989         fpu->last_opcode = fxsave->fop;
2990         fpu->last_ip = fxsave->rip;
2991         fpu->last_dp = fxsave->rdp;
2992         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2993
2994         vcpu_put(vcpu);
2995
2996         return 0;
2997 }
2998
2999 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
3000 {
3001         struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
3002
3003         vcpu_load(vcpu);
3004
3005         memcpy(fxsave->st_space, fpu->fpr, 128);
3006         fxsave->cwd = fpu->fcw;
3007         fxsave->swd = fpu->fsw;
3008         fxsave->twd = fpu->ftwx;
3009         fxsave->fop = fpu->last_opcode;
3010         fxsave->rip = fpu->last_ip;
3011         fxsave->rdp = fpu->last_dp;
3012         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
3013
3014         vcpu_put(vcpu);
3015
3016         return 0;
3017 }
3018
3019 void fx_init(struct kvm_vcpu *vcpu)
3020 {
3021         unsigned after_mxcsr_mask;
3022
3023         /* Initialize guest FPU by resetting ours and saving into guest's */
3024         preempt_disable();
3025         fx_save(&vcpu->arch.host_fx_image);
3026         fpu_init();
3027         fx_save(&vcpu->arch.guest_fx_image);
3028         fx_restore(&vcpu->arch.host_fx_image);
3029         preempt_enable();
3030
3031         vcpu->arch.cr0 |= X86_CR0_ET;
3032         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
3033         vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
3034         memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
3035                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
3036 }
3037 EXPORT_SYMBOL_GPL(fx_init);
3038
3039 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
3040 {
3041         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
3042                 return;
3043
3044         vcpu->guest_fpu_loaded = 1;
3045         fx_save(&vcpu->arch.host_fx_image);
3046         fx_restore(&vcpu->arch.guest_fx_image);
3047 }
3048 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
3049
3050 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
3051 {
3052         if (!vcpu->guest_fpu_loaded)
3053                 return;
3054
3055         vcpu->guest_fpu_loaded = 0;
3056         fx_save(&vcpu->arch.guest_fx_image);
3057         fx_restore(&vcpu->arch.host_fx_image);
3058         ++vcpu->stat.fpu_reload;
3059 }
3060 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
3061
3062 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
3063 {
3064         kvm_x86_ops->vcpu_free(vcpu);
3065 }
3066
3067 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
3068                                                 unsigned int id)
3069 {
3070         return kvm_x86_ops->vcpu_create(kvm, id);
3071 }
3072
3073 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
3074 {
3075         int r;
3076
3077         /* We do fxsave: this must be aligned. */
3078         BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
3079
3080         vcpu_load(vcpu);
3081         r = kvm_arch_vcpu_reset(vcpu);
3082         if (r == 0)
3083                 r = kvm_mmu_setup(vcpu);
3084         vcpu_put(vcpu);
3085         if (r < 0)
3086                 goto free_vcpu;
3087
3088         return 0;
3089 free_vcpu:
3090         kvm_x86_ops->vcpu_free(vcpu);
3091         return r;
3092 }
3093
3094 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3095 {
3096         vcpu_load(vcpu);
3097         kvm_mmu_unload(vcpu);
3098         vcpu_put(vcpu);
3099
3100         kvm_x86_ops->vcpu_free(vcpu);
3101 }
3102
3103 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
3104 {
3105         return kvm_x86_ops->vcpu_reset(vcpu);
3106 }
3107
3108 void kvm_arch_hardware_enable(void *garbage)
3109 {
3110         kvm_x86_ops->hardware_enable(garbage);
3111 }
3112
3113 void kvm_arch_hardware_disable(void *garbage)
3114 {
3115         kvm_x86_ops->hardware_disable(garbage);
3116 }
3117
3118 int kvm_arch_hardware_setup(void)
3119 {
3120         return kvm_x86_ops->hardware_setup();
3121 }
3122
3123 void kvm_arch_hardware_unsetup(void)
3124 {
3125         kvm_x86_ops->hardware_unsetup();
3126 }
3127
3128 void kvm_arch_check_processor_compat(void *rtn)
3129 {
3130         kvm_x86_ops->check_processor_compatibility(rtn);
3131 }
3132
3133 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3134 {
3135         struct page *page;
3136         struct kvm *kvm;
3137         int r;
3138
3139         BUG_ON(vcpu->kvm == NULL);
3140         kvm = vcpu->kvm;
3141
3142         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3143         if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3144                 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
3145         else
3146                 vcpu->arch.mp_state = VCPU_MP_STATE_UNINITIALIZED;
3147
3148         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3149         if (!page) {
3150                 r = -ENOMEM;
3151                 goto fail;
3152         }
3153         vcpu->arch.pio_data = page_address(page);
3154
3155         r = kvm_mmu_create(vcpu);
3156         if (r < 0)
3157                 goto fail_free_pio_data;
3158
3159         if (irqchip_in_kernel(kvm)) {
3160                 r = kvm_create_lapic(vcpu);
3161                 if (r < 0)
3162                         goto fail_mmu_destroy;
3163         }
3164
3165         return 0;
3166
3167 fail_mmu_destroy:
3168         kvm_mmu_destroy(vcpu);
3169 fail_free_pio_data:
3170         free_page((unsigned long)vcpu->arch.pio_data);
3171 fail:
3172         return r;
3173 }
3174
3175 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3176 {
3177         kvm_free_lapic(vcpu);
3178         kvm_mmu_destroy(vcpu);
3179         free_page((unsigned long)vcpu->arch.pio_data);
3180 }
3181
3182 struct  kvm *kvm_arch_create_vm(void)
3183 {
3184         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3185
3186         if (!kvm)
3187                 return ERR_PTR(-ENOMEM);
3188
3189         INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3190
3191         return kvm;
3192 }
3193
3194 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3195 {
3196         vcpu_load(vcpu);
3197         kvm_mmu_unload(vcpu);
3198         vcpu_put(vcpu);
3199 }
3200
3201 static void kvm_free_vcpus(struct kvm *kvm)
3202 {
3203         unsigned int i;
3204
3205         /*
3206          * Unpin any mmu pages first.
3207          */
3208         for (i = 0; i < KVM_MAX_VCPUS; ++i)
3209                 if (kvm->vcpus[i])
3210                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3211         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3212                 if (kvm->vcpus[i]) {
3213                         kvm_arch_vcpu_free(kvm->vcpus[i]);
3214                         kvm->vcpus[i] = NULL;
3215                 }
3216         }
3217
3218 }
3219
3220 void kvm_arch_destroy_vm(struct kvm *kvm)
3221 {
3222         kfree(kvm->arch.vpic);
3223         kfree(kvm->arch.vioapic);
3224         kvm_free_vcpus(kvm);
3225         kvm_free_physmem(kvm);
3226         kfree(kvm);
3227 }
3228
3229 int kvm_arch_set_memory_region(struct kvm *kvm,
3230                                 struct kvm_userspace_memory_region *mem,
3231                                 struct kvm_memory_slot old,
3232                                 int user_alloc)
3233 {
3234         int npages = mem->memory_size >> PAGE_SHIFT;
3235         struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3236
3237         /*To keep backward compatibility with older userspace,
3238          *x86 needs to hanlde !user_alloc case.
3239          */
3240         if (!user_alloc) {
3241                 if (npages && !old.rmap) {
3242                         down_write(&current->mm->mmap_sem);
3243                         memslot->userspace_addr = do_mmap(NULL, 0,
3244                                                      npages * PAGE_SIZE,
3245                                                      PROT_READ | PROT_WRITE,
3246                                                      MAP_SHARED | MAP_ANONYMOUS,
3247                                                      0);
3248                         up_write(&current->mm->mmap_sem);
3249
3250                         if (IS_ERR((void *)memslot->userspace_addr))
3251                                 return PTR_ERR((void *)memslot->userspace_addr);
3252                 } else {
3253                         if (!old.user_alloc && old.rmap) {
3254                                 int ret;
3255
3256                                 down_write(&current->mm->mmap_sem);
3257                                 ret = do_munmap(current->mm, old.userspace_addr,
3258                                                 old.npages * PAGE_SIZE);
3259                                 up_write(&current->mm->mmap_sem);
3260                                 if (ret < 0)
3261                                         printk(KERN_WARNING
3262                                        "kvm_vm_ioctl_set_memory_region: "
3263                                        "failed to munmap memory\n");
3264                         }
3265                 }
3266         }
3267
3268         if (!kvm->arch.n_requested_mmu_pages) {
3269                 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3270                 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3271         }
3272
3273         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3274         kvm_flush_remote_tlbs(kvm);
3275
3276         return 0;
3277 }
3278
3279 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3280 {
3281         return vcpu->arch.mp_state == VCPU_MP_STATE_RUNNABLE
3282                || vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED;
3283 }
3284
3285 static void vcpu_kick_intr(void *info)
3286 {
3287 #ifdef DEBUG
3288         struct kvm_vcpu *vcpu = (struct kvm_vcpu *)info;
3289         printk(KERN_DEBUG "vcpu_kick_intr %p \n", vcpu);
3290 #endif
3291 }
3292
3293 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3294 {
3295         int ipi_pcpu = vcpu->cpu;
3296
3297         if (waitqueue_active(&vcpu->wq)) {
3298                 wake_up_interruptible(&vcpu->wq);
3299                 ++vcpu->stat.halt_wakeup;
3300         }
3301         if (vcpu->guest_mode)
3302                 smp_call_function_single(ipi_pcpu, vcpu_kick_intr, vcpu, 0, 0);
3303 }