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