]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/kvm/x86.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: x86 emulator: unify four switch statements into two
[mirror_ubuntu-artful-kernel.git] / drivers / kvm / x86.c
CommitLineData
043405e1
CO		 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
313a3dc7 17#include "kvm.h"
043405e1 18#include "x86.h"
d825ed0a 19#include "x86_emulate.h"
5fb76f9b 20#include "segment_descriptor.h"
313a3dc7
CO		 21#include "irq.h"
22
23#include <linux/kvm.h>
24#include <linux/fs.h>
25#include <linux/vmalloc.h>
5fb76f9b 26#include <linux/module.h>
0de10343 27#include <linux/mman.h>
043405e1
CO		 28
29#include <asm/uaccess.h>
d825ed0a 30#include <asm/msr.h>
043405e1 31
313a3dc7 32#define MAX_IO_MSRS 256
a03490ed
CO		 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)
15c4a640 44#define EFER_RESERVED_BITS 0xfffffffffffff2fe
313a3dc7 45
ba1389b7
AK		 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
417bc304 48
97896d04
ZX		 49struct kvm_x86_ops *kvm_x86_ops;
50
417bc304 51struct kvm_stats_debugfs_item debugfs_entries[] = {
ba1389b7
AK		 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) },
4cee5764
AK		 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) },
0f74a24c 76 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
417bc304
HB		 77 { NULL }
78};
79
80
5fb76f9b
CO		 81unsigned long segment_base(u16 selector)
82{
83 struct descriptor_table gdt;
84 struct segment_descriptor *d;
85 unsigned long table_base;
86 unsigned long v;
87
88 if (selector == 0)
89 return 0;
90
91 asm("sgdt %0" : "=m"(gdt));
92 table_base = gdt.base;
93
94 if (selector & 4) { /* from ldt */
95 u16 ldt_selector;
96
97 asm("sldt %0" : "=g"(ldt_selector));
98 table_base = segment_base(ldt_selector);
99 }
100 d = (struct segment_descriptor *)(table_base + (selector & ~7));
101 v = d->base_low | ((unsigned long)d->base_mid << 16) |
102 ((unsigned long)d->base_high << 24);
103#ifdef CONFIG_X86_64
104 if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
105 v |= ((unsigned long) \
106 ((struct segment_descriptor_64 *)d)->base_higher) << 32;
107#endif
108 return v;
109}
110EXPORT_SYMBOL_GPL(segment_base);
111
6866b83e
CO		 112u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
113{
114 if (irqchip_in_kernel(vcpu->kvm))
115 return vcpu->apic_base;
116 else
117 return vcpu->apic_base;
118}
119EXPORT_SYMBOL_GPL(kvm_get_apic_base);
120
121void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
122{
123 /* TODO: reserve bits check */
124 if (irqchip_in_kernel(vcpu->kvm))
125 kvm_lapic_set_base(vcpu, data);
126 else
127 vcpu->apic_base = data;
128}
129EXPORT_SYMBOL_GPL(kvm_set_apic_base);
130
a03490ed
CO		 131static void inject_gp(struct kvm_vcpu *vcpu)
132{
133 kvm_x86_ops->inject_gp(vcpu, 0);
134}
135
136/*
137 * Load the pae pdptrs. Return true is they are all valid.
138 */
139int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
140{
141 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
142 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
143 int i;
144 int ret;
145 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
146
147 mutex_lock(&vcpu->kvm->lock);
148 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
149 offset * sizeof(u64), sizeof(pdpte));
150 if (ret < 0) {
151 ret = 0;
152 goto out;
153 }
154 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
155 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
156 ret = 0;
157 goto out;
158 }
159 }
160 ret = 1;
161
162 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
163out:
164 mutex_unlock(&vcpu->kvm->lock);
165
166 return ret;
167}
168
d835dfec
AK		 169static bool pdptrs_changed(struct kvm_vcpu *vcpu)
170{
171 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
172 bool changed = true;
173 int r;
174
175 if (is_long_mode(vcpu) || !is_pae(vcpu))
176 return false;
177
178 mutex_lock(&vcpu->kvm->lock);
179 r = kvm_read_guest(vcpu->kvm, vcpu->cr3 & ~31u, pdpte, sizeof(pdpte));
180 if (r < 0)
181 goto out;
182 changed = memcmp(pdpte, vcpu->pdptrs, sizeof(pdpte)) != 0;
183out:
184 mutex_unlock(&vcpu->kvm->lock);
185
186 return changed;
187}
188
a03490ed
CO		 189void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
190{
191 if (cr0 & CR0_RESERVED_BITS) {
192 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
193 cr0, vcpu->cr0);
194 inject_gp(vcpu);
195 return;
196 }
197
198 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
199 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
200 inject_gp(vcpu);
201 return;
202 }
203
204 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
205 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
206 "and a clear PE flag\n");
207 inject_gp(vcpu);
208 return;
209 }
210
211 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
212#ifdef CONFIG_X86_64
213 if ((vcpu->shadow_efer & EFER_LME)) {
214 int cs_db, cs_l;
215
216 if (!is_pae(vcpu)) {
217 printk(KERN_DEBUG "set_cr0: #GP, start paging "
218 "in long mode while PAE is disabled\n");
219 inject_gp(vcpu);
220 return;
221 }
222 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
223 if (cs_l) {
224 printk(KERN_DEBUG "set_cr0: #GP, start paging "
225 "in long mode while CS.L == 1\n");
226 inject_gp(vcpu);
227 return;
228
229 }
230 } else
231#endif
232 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
233 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
234 "reserved bits\n");
235 inject_gp(vcpu);
236 return;
237 }
238
239 }
240
241 kvm_x86_ops->set_cr0(vcpu, cr0);
242 vcpu->cr0 = cr0;
243
244 mutex_lock(&vcpu->kvm->lock);
245 kvm_mmu_reset_context(vcpu);
246 mutex_unlock(&vcpu->kvm->lock);
247 return;
248}
249EXPORT_SYMBOL_GPL(set_cr0);
250
251void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
252{
253 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
254}
255EXPORT_SYMBOL_GPL(lmsw);
256
257void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
258{
259 if (cr4 & CR4_RESERVED_BITS) {
260 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
261 inject_gp(vcpu);
262 return;
263 }
264
265 if (is_long_mode(vcpu)) {
266 if (!(cr4 & X86_CR4_PAE)) {
267 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
268 "in long mode\n");
269 inject_gp(vcpu);
270 return;
271 }
272 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
273 && !load_pdptrs(vcpu, vcpu->cr3)) {
274 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
275 inject_gp(vcpu);
276 return;
277 }
278
279 if (cr4 & X86_CR4_VMXE) {
280 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
281 inject_gp(vcpu);
282 return;
283 }
284 kvm_x86_ops->set_cr4(vcpu, cr4);
285 vcpu->cr4 = cr4;
286 mutex_lock(&vcpu->kvm->lock);
287 kvm_mmu_reset_context(vcpu);
288 mutex_unlock(&vcpu->kvm->lock);
289}
290EXPORT_SYMBOL_GPL(set_cr4);
291
292void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
293{
d835dfec
AK		 294 if (cr3 == vcpu->cr3 && !pdptrs_changed(vcpu)) {
295 kvm_mmu_flush_tlb(vcpu);
296 return;
297 }
298
a03490ed
CO		 299 if (is_long_mode(vcpu)) {
300 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
301 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
302 inject_gp(vcpu);
303 return;
304 }
305 } else {
306 if (is_pae(vcpu)) {
307 if (cr3 & CR3_PAE_RESERVED_BITS) {
308 printk(KERN_DEBUG
309 "set_cr3: #GP, reserved bits\n");
310 inject_gp(vcpu);
311 return;
312 }
313 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
314 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
315 "reserved bits\n");
316 inject_gp(vcpu);
317 return;
318 }
319 }
320 /*
321 * We don't check reserved bits in nonpae mode, because
322 * this isn't enforced, and VMware depends on this.
323 */
324 }
325
326 mutex_lock(&vcpu->kvm->lock);
327 /*
328 * Does the new cr3 value map to physical memory? (Note, we
329 * catch an invalid cr3 even in real-mode, because it would
330 * cause trouble later on when we turn on paging anyway.)
331 *
332 * A real CPU would silently accept an invalid cr3 and would
333 * attempt to use it - with largely undefined (and often hard
334 * to debug) behavior on the guest side.
335 */
336 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
337 inject_gp(vcpu);
338 else {
339 vcpu->cr3 = cr3;
340 vcpu->mmu.new_cr3(vcpu);
341 }
342 mutex_unlock(&vcpu->kvm->lock);
343}
344EXPORT_SYMBOL_GPL(set_cr3);
345
346void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
347{
348 if (cr8 & CR8_RESERVED_BITS) {
349 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
350 inject_gp(vcpu);
351 return;
352 }
353 if (irqchip_in_kernel(vcpu->kvm))
354 kvm_lapic_set_tpr(vcpu, cr8);
355 else
356 vcpu->cr8 = cr8;
357}
358EXPORT_SYMBOL_GPL(set_cr8);
359
360unsigned long get_cr8(struct kvm_vcpu *vcpu)
361{
362 if (irqchip_in_kernel(vcpu->kvm))
363 return kvm_lapic_get_cr8(vcpu);
364 else
365 return vcpu->cr8;
366}
367EXPORT_SYMBOL_GPL(get_cr8);
368
043405e1
CO		 369/*
370 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
371 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
372 *
373 * This list is modified at module load time to reflect the
374 * capabilities of the host cpu.
375 */
376static u32 msrs_to_save[] = {
377 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
378 MSR_K6_STAR,
379#ifdef CONFIG_X86_64
380 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
381#endif
382 MSR_IA32_TIME_STAMP_COUNTER,
383};
384
385static unsigned num_msrs_to_save;
386
387static u32 emulated_msrs[] = {
388 MSR_IA32_MISC_ENABLE,
389};
390
15c4a640
CO		 391#ifdef CONFIG_X86_64
392
393static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
394{
395 if (efer & EFER_RESERVED_BITS) {
396 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
397 efer);
398 inject_gp(vcpu);
399 return;
400 }
401
402 if (is_paging(vcpu)
403 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
404 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
405 inject_gp(vcpu);
406 return;
407 }
408
409 kvm_x86_ops->set_efer(vcpu, efer);
410
411 efer &= ~EFER_LMA;
412 efer |= vcpu->shadow_efer & EFER_LMA;
413
414 vcpu->shadow_efer = efer;
415}
416
417#endif
418
419/*
420 * Writes msr value into into the appropriate "register".
421 * Returns 0 on success, non-0 otherwise.
422 * Assumes vcpu_load() was already called.
423 */
424int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
425{
426 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
427}
428
313a3dc7
CO		 429/*
430 * Adapt set_msr() to msr_io()'s calling convention
431 */
432static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
433{
434 return kvm_set_msr(vcpu, index, *data);
435}
436
15c4a640
CO		 437
438int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
439{
440 switch (msr) {
441#ifdef CONFIG_X86_64
442 case MSR_EFER:
443 set_efer(vcpu, data);
444 break;
445#endif
446 case MSR_IA32_MC0_STATUS:
447 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
448 __FUNCTION__, data);
449 break;
450 case MSR_IA32_MCG_STATUS:
451 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
452 __FUNCTION__, data);
453 break;
454 case MSR_IA32_UCODE_REV:
455 case MSR_IA32_UCODE_WRITE:
456 case 0x200 ... 0x2ff: /* MTRRs */
457 break;
458 case MSR_IA32_APICBASE:
459 kvm_set_apic_base(vcpu, data);
460 break;
461 case MSR_IA32_MISC_ENABLE:
462 vcpu->ia32_misc_enable_msr = data;
463 break;
464 default:
465 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
466 return 1;
467 }
468 return 0;
469}
470EXPORT_SYMBOL_GPL(kvm_set_msr_common);
471
472
473/*
474 * Reads an msr value (of 'msr_index') into 'pdata'.
475 * Returns 0 on success, non-0 otherwise.
476 * Assumes vcpu_load() was already called.
477 */
478int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
479{
480 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
481}
482
483int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
484{
485 u64 data;
486
487 switch (msr) {
488 case 0xc0010010: /* SYSCFG */
489 case 0xc0010015: /* HWCR */
490 case MSR_IA32_PLATFORM_ID:
491 case MSR_IA32_P5_MC_ADDR:
492 case MSR_IA32_P5_MC_TYPE:
493 case MSR_IA32_MC0_CTL:
494 case MSR_IA32_MCG_STATUS:
495 case MSR_IA32_MCG_CAP:
496 case MSR_IA32_MC0_MISC:
497 case MSR_IA32_MC0_MISC+4:
498 case MSR_IA32_MC0_MISC+8:
499 case MSR_IA32_MC0_MISC+12:
500 case MSR_IA32_MC0_MISC+16:
501 case MSR_IA32_UCODE_REV:
502 case MSR_IA32_PERF_STATUS:
503 case MSR_IA32_EBL_CR_POWERON:
504 /* MTRR registers */
505 case 0xfe:
506 case 0x200 ... 0x2ff:
507 data = 0;
508 break;
509 case 0xcd: /* fsb frequency */
510 data = 3;
511 break;
512 case MSR_IA32_APICBASE:
513 data = kvm_get_apic_base(vcpu);
514 break;
515 case MSR_IA32_MISC_ENABLE:
516 data = vcpu->ia32_misc_enable_msr;
517 break;
518#ifdef CONFIG_X86_64
519 case MSR_EFER:
520 data = vcpu->shadow_efer;
521 break;
522#endif
523 default:
524 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
525 return 1;
526 }
527 *pdata = data;
528 return 0;
529}
530EXPORT_SYMBOL_GPL(kvm_get_msr_common);
531
313a3dc7
CO		 532/*
533 * Read or write a bunch of msrs. All parameters are kernel addresses.
534 *
535 * @return number of msrs set successfully.
536 */
537static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
538 struct kvm_msr_entry *entries,
539 int (*do_msr)(struct kvm_vcpu *vcpu,
540 unsigned index, u64 *data))
541{
542 int i;
543
544 vcpu_load(vcpu);
545
546 for (i = 0; i < msrs->nmsrs; ++i)
547 if (do_msr(vcpu, entries[i].index, &entries[i].data))
548 break;
549
550 vcpu_put(vcpu);
551
552 return i;
553}
554
555/*
556 * Read or write a bunch of msrs. Parameters are user addresses.
557 *
558 * @return number of msrs set successfully.
559 */
560static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
561 int (*do_msr)(struct kvm_vcpu *vcpu,
562 unsigned index, u64 *data),
563 int writeback)
564{
565 struct kvm_msrs msrs;
566 struct kvm_msr_entry *entries;
567 int r, n;
568 unsigned size;
569
570 r = -EFAULT;
571 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
572 goto out;
573
574 r = -E2BIG;
575 if (msrs.nmsrs >= MAX_IO_MSRS)
576 goto out;
577
578 r = -ENOMEM;
579 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
580 entries = vmalloc(size);
581 if (!entries)
582 goto out;
583
584 r = -EFAULT;
585 if (copy_from_user(entries, user_msrs->entries, size))
586 goto out_free;
587
588 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
589 if (r < 0)
590 goto out_free;
591
592 r = -EFAULT;
593 if (writeback && copy_to_user(user_msrs->entries, entries, size))
594 goto out_free;
595
596 r = n;
597
598out_free:
599 vfree(entries);
600out:
601 return r;
602}
603
e9b11c17
ZX		 604/*
605 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
606 * cached on it.
607 */
608void decache_vcpus_on_cpu(int cpu)
609{
610 struct kvm *vm;
611 struct kvm_vcpu *vcpu;
612 int i;
613
614 spin_lock(&kvm_lock);
615 list_for_each_entry(vm, &vm_list, vm_list)
616 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
617 vcpu = vm->vcpus[i];
618 if (!vcpu)
619 continue;
620 /*
621 * If the vcpu is locked, then it is running on some
622 * other cpu and therefore it is not cached on the
623 * cpu in question.
624 *
625 * If it's not locked, check the last cpu it executed
626 * on.
627 */
628 if (mutex_trylock(&vcpu->mutex)) {
629 if (vcpu->cpu == cpu) {
630 kvm_x86_ops->vcpu_decache(vcpu);
631 vcpu->cpu = -1;
632 }
633 mutex_unlock(&vcpu->mutex);
634 }
635 }
636 spin_unlock(&kvm_lock);
637}
638
018d00d2
ZX		 639int kvm_dev_ioctl_check_extension(long ext)
640{
641 int r;
642
643 switch (ext) {
644 case KVM_CAP_IRQCHIP:
645 case KVM_CAP_HLT:
646 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
647 case KVM_CAP_USER_MEMORY:
648 case KVM_CAP_SET_TSS_ADDR:
07716717 649 case KVM_CAP_EXT_CPUID:
018d00d2
ZX		 650 r = 1;
651 break;
652 default:
653 r = 0;
654 break;
655 }
656 return r;
657
658}
659
043405e1
CO		 660long kvm_arch_dev_ioctl(struct file *filp,
661 unsigned int ioctl, unsigned long arg)
662{
663 void __user *argp = (void __user *)arg;
664 long r;
665
666 switch (ioctl) {
667 case KVM_GET_MSR_INDEX_LIST: {
668 struct kvm_msr_list __user *user_msr_list = argp;
669 struct kvm_msr_list msr_list;
670 unsigned n;
671
672 r = -EFAULT;
673 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
674 goto out;
675 n = msr_list.nmsrs;
676 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
677 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
678 goto out;
679 r = -E2BIG;
680 if (n < num_msrs_to_save)
681 goto out;
682 r = -EFAULT;
683 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
684 num_msrs_to_save * sizeof(u32)))
685 goto out;
686 if (copy_to_user(user_msr_list->indices
687 + num_msrs_to_save * sizeof(u32),
688 &emulated_msrs,
689 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
690 goto out;
691 r = 0;
692 break;
693 }
694 default:
695 r = -EINVAL;
696 }
697out:
698 return r;
699}
700
313a3dc7
CO		 701void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
702{
703 kvm_x86_ops->vcpu_load(vcpu, cpu);
704}
705
706void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
707{
708 kvm_x86_ops->vcpu_put(vcpu);
9327fd11 709 kvm_put_guest_fpu(vcpu);
313a3dc7
CO		 710}
711
07716717 712static int is_efer_nx(void)
313a3dc7
CO		 713{
714 u64 efer;
313a3dc7
CO		 715
716 rdmsrl(MSR_EFER, efer);
07716717
DK		 717 return efer & EFER_NX;
718}
719
720static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
721{
722 int i;
723 struct kvm_cpuid_entry2 *e, *entry;
724
313a3dc7
CO		 725 entry = NULL;
726 for (i = 0; i < vcpu->cpuid_nent; ++i) {
727 e = &vcpu->cpuid_entries[i];
728 if (e->function == 0x80000001) {
729 entry = e;
730 break;
731 }
732 }
07716717 733 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
313a3dc7
CO		 734 entry->edx &= ~(1 << 20);
735 printk(KERN_INFO "kvm: guest NX capability removed\n");
736 }
737}
738
07716717 739/* when an old userspace process fills a new kernel module */
313a3dc7
CO		 740static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
741 struct kvm_cpuid *cpuid,
742 struct kvm_cpuid_entry __user *entries)
07716717
DK		 743{
744 int r, i;
745 struct kvm_cpuid_entry *cpuid_entries;
746
747 r = -E2BIG;
748 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
749 goto out;
750 r = -ENOMEM;
751 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
752 if (!cpuid_entries)
753 goto out;
754 r = -EFAULT;
755 if (copy_from_user(cpuid_entries, entries,
756 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
757 goto out_free;
758 for (i = 0; i < cpuid->nent; i++) {
759 vcpu->cpuid_entries[i].function = cpuid_entries[i].function;
760 vcpu->cpuid_entries[i].eax = cpuid_entries[i].eax;
761 vcpu->cpuid_entries[i].ebx = cpuid_entries[i].ebx;
762 vcpu->cpuid_entries[i].ecx = cpuid_entries[i].ecx;
763 vcpu->cpuid_entries[i].edx = cpuid_entries[i].edx;
764 vcpu->cpuid_entries[i].index = 0;
765 vcpu->cpuid_entries[i].flags = 0;
766 vcpu->cpuid_entries[i].padding[0] = 0;
767 vcpu->cpuid_entries[i].padding[1] = 0;
768 vcpu->cpuid_entries[i].padding[2] = 0;
769 }
770 vcpu->cpuid_nent = cpuid->nent;
771 cpuid_fix_nx_cap(vcpu);
772 r = 0;
773
774out_free:
775 vfree(cpuid_entries);
776out:
777 return r;
778}
779
780static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
781 struct kvm_cpuid2 *cpuid,
782 struct kvm_cpuid_entry2 __user *entries)
313a3dc7
CO		 783{
784 int r;
785
786 r = -E2BIG;
787 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
788 goto out;
789 r = -EFAULT;
790 if (copy_from_user(&vcpu->cpuid_entries, entries,
07716717 791 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
313a3dc7
CO		 792 goto out;
793 vcpu->cpuid_nent = cpuid->nent;
313a3dc7
CO		 794 return 0;
795
796out:
797 return r;
798}
799
07716717
DK		 800static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
801 struct kvm_cpuid2 *cpuid,
802 struct kvm_cpuid_entry2 __user *entries)
803{
804 int r;
805
806 r = -E2BIG;
807 if (cpuid->nent < vcpu->cpuid_nent)
808 goto out;
809 r = -EFAULT;
810 if (copy_to_user(entries, &vcpu->cpuid_entries,
811 vcpu->cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
812 goto out;
813 return 0;
814
815out:
816 cpuid->nent = vcpu->cpuid_nent;
817 return r;
818}
819
820static inline u32 bit(int bitno)
821{
822 return 1 << (bitno & 31);
823}
824
825static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
826 u32 index)
827{
828 entry->function = function;
829 entry->index = index;
830 cpuid_count(entry->function, entry->index,
831 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
832 entry->flags = 0;
833}
834
835static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
836 u32 index, int *nent, int maxnent)
837{
838 const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
839 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
840 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
841 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
842 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
843 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
844 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
845 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
846 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
847 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
848 const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
849 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
850 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
851 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
852 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
853 bit(X86_FEATURE_PGE) |
854 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
855 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
856 bit(X86_FEATURE_SYSCALL) |
857 (bit(X86_FEATURE_NX) && is_efer_nx()) |
858#ifdef CONFIG_X86_64
859 bit(X86_FEATURE_LM) |
860#endif
861 bit(X86_FEATURE_MMXEXT) |
862 bit(X86_FEATURE_3DNOWEXT) |
863 bit(X86_FEATURE_3DNOW);
864 const u32 kvm_supported_word3_x86_features =
865 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
866 const u32 kvm_supported_word6_x86_features =
867 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
868
869 /* all func 2 cpuid_count() should be called on the same cpu */
870 get_cpu();
871 do_cpuid_1_ent(entry, function, index);
872 ++*nent;
873
874 switch (function) {
875 case 0:
876 entry->eax = min(entry->eax, (u32)0xb);
877 break;
878 case 1:
879 entry->edx &= kvm_supported_word0_x86_features;
880 entry->ecx &= kvm_supported_word3_x86_features;
881 break;
882 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
883 * may return different values. This forces us to get_cpu() before
884 * issuing the first command, and also to emulate this annoying behavior
885 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
886 case 2: {
887 int t, times = entry->eax & 0xff;
888
889 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
890 for (t = 1; t < times && *nent < maxnent; ++t) {
891 do_cpuid_1_ent(&entry[t], function, 0);
892 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
893 ++*nent;
894 }
895 break;
896 }
897 /* function 4 and 0xb have additional index. */
898 case 4: {
899 int index, cache_type;
900
901 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
902 /* read more entries until cache_type is zero */
903 for (index = 1; *nent < maxnent; ++index) {
904 cache_type = entry[index - 1].eax & 0x1f;
905 if (!cache_type)
906 break;
907 do_cpuid_1_ent(&entry[index], function, index);
908 entry[index].flags |=
909 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
910 ++*nent;
911 }
912 break;
913 }
914 case 0xb: {
915 int index, level_type;
916
917 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
918 /* read more entries until level_type is zero */
919 for (index = 1; *nent < maxnent; ++index) {
920 level_type = entry[index - 1].ecx & 0xff;
921 if (!level_type)
922 break;
923 do_cpuid_1_ent(&entry[index], function, index);
924 entry[index].flags |=
925 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
926 ++*nent;
927 }
928 break;
929 }
930 case 0x80000000:
931 entry->eax = min(entry->eax, 0x8000001a);
932 break;
933 case 0x80000001:
934 entry->edx &= kvm_supported_word1_x86_features;
935 entry->ecx &= kvm_supported_word6_x86_features;
936 break;
937 }
938 put_cpu();
939}
940
941static int kvm_vm_ioctl_get_supported_cpuid(struct kvm *kvm,
942 struct kvm_cpuid2 *cpuid,
943 struct kvm_cpuid_entry2 __user *entries)
944{
945 struct kvm_cpuid_entry2 *cpuid_entries;
946 int limit, nent = 0, r = -E2BIG;
947 u32 func;
948
949 if (cpuid->nent < 1)
950 goto out;
951 r = -ENOMEM;
952 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
953 if (!cpuid_entries)
954 goto out;
955
956 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
957 limit = cpuid_entries[0].eax;
958 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
959 do_cpuid_ent(&cpuid_entries[nent], func, 0,
960 &nent, cpuid->nent);
961 r = -E2BIG;
962 if (nent >= cpuid->nent)
963 goto out_free;
964
965 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
966 limit = cpuid_entries[nent - 1].eax;
967 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
968 do_cpuid_ent(&cpuid_entries[nent], func, 0,
969 &nent, cpuid->nent);
970 r = -EFAULT;
971 if (copy_to_user(entries, cpuid_entries,
972 nent * sizeof(struct kvm_cpuid_entry2)))
973 goto out_free;
974 cpuid->nent = nent;
975 r = 0;
976
977out_free:
978 vfree(cpuid_entries);
979out:
980 return r;
981}
982
313a3dc7
CO		 983static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
984 struct kvm_lapic_state *s)
985{
986 vcpu_load(vcpu);
987 memcpy(s->regs, vcpu->apic->regs, sizeof *s);
988 vcpu_put(vcpu);
989
990 return 0;
991}
992
993static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
994 struct kvm_lapic_state *s)
995{
996 vcpu_load(vcpu);
997 memcpy(vcpu->apic->regs, s->regs, sizeof *s);
998 kvm_apic_post_state_restore(vcpu);
999 vcpu_put(vcpu);
1000
1001 return 0;
1002}
1003
1004long kvm_arch_vcpu_ioctl(struct file *filp,
1005 unsigned int ioctl, unsigned long arg)
1006{
1007 struct kvm_vcpu *vcpu = filp->private_data;
1008 void __user *argp = (void __user *)arg;
1009 int r;
1010
1011 switch (ioctl) {
1012 case KVM_GET_LAPIC: {
1013 struct kvm_lapic_state lapic;
1014
1015 memset(&lapic, 0, sizeof lapic);
1016 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1017 if (r)
1018 goto out;
1019 r = -EFAULT;
1020 if (copy_to_user(argp, &lapic, sizeof lapic))
1021 goto out;
1022 r = 0;
1023 break;
1024 }
1025 case KVM_SET_LAPIC: {
1026 struct kvm_lapic_state lapic;
1027
1028 r = -EFAULT;
1029 if (copy_from_user(&lapic, argp, sizeof lapic))
1030 goto out;
1031 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1032 if (r)
1033 goto out;
1034 r = 0;
1035 break;
1036 }
1037 case KVM_SET_CPUID: {
1038 struct kvm_cpuid __user *cpuid_arg = argp;
1039 struct kvm_cpuid cpuid;
1040
1041 r = -EFAULT;
1042 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1043 goto out;
1044 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1045 if (r)
1046 goto out;
1047 break;
1048 }
07716717
DK		 1049 case KVM_SET_CPUID2: {
1050 struct kvm_cpuid2 __user *cpuid_arg = argp;
1051 struct kvm_cpuid2 cpuid;
1052
1053 r = -EFAULT;
1054 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1055 goto out;
1056 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1057 cpuid_arg->entries);
1058 if (r)
1059 goto out;
1060 break;
1061 }
1062 case KVM_GET_CPUID2: {
1063 struct kvm_cpuid2 __user *cpuid_arg = argp;
1064 struct kvm_cpuid2 cpuid;
1065
1066 r = -EFAULT;
1067 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1068 goto out;
1069 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1070 cpuid_arg->entries);
1071 if (r)
1072 goto out;
1073 r = -EFAULT;
1074 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1075 goto out;
1076 r = 0;
1077 break;
1078 }
313a3dc7
CO		 1079 case KVM_GET_MSRS:
1080 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1081 break;
1082 case KVM_SET_MSRS:
1083 r = msr_io(vcpu, argp, do_set_msr, 0);
1084 break;
1085 default:
1086 r = -EINVAL;
1087 }
1088out:
1089 return r;
1090}
1091
1fe779f8
CO		 1092static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1093{
1094 int ret;
1095
1096 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1097 return -1;
1098 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1099 return ret;
1100}
1101
1102static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1103 u32 kvm_nr_mmu_pages)
1104{
1105 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1106 return -EINVAL;
1107
1108 mutex_lock(&kvm->lock);
1109
1110 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1111 kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
1112
1113 mutex_unlock(&kvm->lock);
1114 return 0;
1115}
1116
1117static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1118{
1119 return kvm->n_alloc_mmu_pages;
1120}
1121
e9f85cde
ZX		 1122gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1123{
1124 int i;
1125 struct kvm_mem_alias *alias;
1126
1127 for (i = 0; i < kvm->naliases; ++i) {
1128 alias = &kvm->aliases[i];
1129 if (gfn >= alias->base_gfn
1130 && gfn < alias->base_gfn + alias->npages)
1131 return alias->target_gfn + gfn - alias->base_gfn;
1132 }
1133 return gfn;
1134}
1135
1fe779f8
CO		 1136/*
1137 * Set a new alias region. Aliases map a portion of physical memory into
1138 * another portion. This is useful for memory windows, for example the PC
1139 * VGA region.
1140 */
1141static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1142 struct kvm_memory_alias *alias)
1143{
1144 int r, n;
1145 struct kvm_mem_alias *p;
1146
1147 r = -EINVAL;
1148 /* General sanity checks */
1149 if (alias->memory_size & (PAGE_SIZE - 1))
1150 goto out;
1151 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1152 goto out;
1153 if (alias->slot >= KVM_ALIAS_SLOTS)
1154 goto out;
1155 if (alias->guest_phys_addr + alias->memory_size
1156 < alias->guest_phys_addr)
1157 goto out;
1158 if (alias->target_phys_addr + alias->memory_size
1159 < alias->target_phys_addr)
1160 goto out;
1161
1162 mutex_lock(&kvm->lock);
1163
1164 p = &kvm->aliases[alias->slot];
1165 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1166 p->npages = alias->memory_size >> PAGE_SHIFT;
1167 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1168
1169 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1170 if (kvm->aliases[n - 1].npages)
1171 break;
1172 kvm->naliases = n;
1173
1174 kvm_mmu_zap_all(kvm);
1175
1176 mutex_unlock(&kvm->lock);
1177
1178 return 0;
1179
1180out:
1181 return r;
1182}
1183
1184static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1185{
1186 int r;
1187
1188 r = 0;
1189 switch (chip->chip_id) {
1190 case KVM_IRQCHIP_PIC_MASTER:
1191 memcpy(&chip->chip.pic,
1192 &pic_irqchip(kvm)->pics[0],
1193 sizeof(struct kvm_pic_state));
1194 break;
1195 case KVM_IRQCHIP_PIC_SLAVE:
1196 memcpy(&chip->chip.pic,
1197 &pic_irqchip(kvm)->pics[1],
1198 sizeof(struct kvm_pic_state));
1199 break;
1200 case KVM_IRQCHIP_IOAPIC:
1201 memcpy(&chip->chip.ioapic,
1202 ioapic_irqchip(kvm),
1203 sizeof(struct kvm_ioapic_state));
1204 break;
1205 default:
1206 r = -EINVAL;
1207 break;
1208 }
1209 return r;
1210}
1211
1212static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1213{
1214 int r;
1215
1216 r = 0;
1217 switch (chip->chip_id) {
1218 case KVM_IRQCHIP_PIC_MASTER:
1219 memcpy(&pic_irqchip(kvm)->pics[0],
1220 &chip->chip.pic,
1221 sizeof(struct kvm_pic_state));
1222 break;
1223 case KVM_IRQCHIP_PIC_SLAVE:
1224 memcpy(&pic_irqchip(kvm)->pics[1],
1225 &chip->chip.pic,
1226 sizeof(struct kvm_pic_state));
1227 break;
1228 case KVM_IRQCHIP_IOAPIC:
1229 memcpy(ioapic_irqchip(kvm),
1230 &chip->chip.ioapic,
1231 sizeof(struct kvm_ioapic_state));
1232 break;
1233 default:
1234 r = -EINVAL;
1235 break;
1236 }
1237 kvm_pic_update_irq(pic_irqchip(kvm));
1238 return r;
1239}
1240
5bb064dc
ZX		 1241/*
1242 * Get (and clear) the dirty memory log for a memory slot.
1243 */
1244int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1245 struct kvm_dirty_log *log)
1246{
1247 int r;
1248 int n;
1249 struct kvm_memory_slot *memslot;
1250 int is_dirty = 0;
1251
1252 mutex_lock(&kvm->lock);
1253
1254 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1255 if (r)
1256 goto out;
1257
1258 /* If nothing is dirty, don't bother messing with page tables. */
1259 if (is_dirty) {
1260 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1261 kvm_flush_remote_tlbs(kvm);
1262 memslot = &kvm->memslots[log->slot];
1263 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1264 memset(memslot->dirty_bitmap, 0, n);
1265 }
1266 r = 0;
1267out:
1268 mutex_unlock(&kvm->lock);
1269 return r;
1270}
1271
1fe779f8
CO		 1272long kvm_arch_vm_ioctl(struct file *filp,
1273 unsigned int ioctl, unsigned long arg)
1274{
1275 struct kvm *kvm = filp->private_data;
1276 void __user *argp = (void __user *)arg;
1277 int r = -EINVAL;
1278
1279 switch (ioctl) {
1280 case KVM_SET_TSS_ADDR:
1281 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1282 if (r < 0)
1283 goto out;
1284 break;
1285 case KVM_SET_MEMORY_REGION: {
1286 struct kvm_memory_region kvm_mem;
1287 struct kvm_userspace_memory_region kvm_userspace_mem;
1288
1289 r = -EFAULT;
1290 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1291 goto out;
1292 kvm_userspace_mem.slot = kvm_mem.slot;
1293 kvm_userspace_mem.flags = kvm_mem.flags;
1294 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1295 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1296 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1297 if (r)
1298 goto out;
1299 break;
1300 }
1301 case KVM_SET_NR_MMU_PAGES:
1302 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1303 if (r)
1304 goto out;
1305 break;
1306 case KVM_GET_NR_MMU_PAGES:
1307 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1308 break;
1309 case KVM_SET_MEMORY_ALIAS: {
1310 struct kvm_memory_alias alias;
1311
1312 r = -EFAULT;
1313 if (copy_from_user(&alias, argp, sizeof alias))
1314 goto out;
1315 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1316 if (r)
1317 goto out;
1318 break;
1319 }
1320 case KVM_CREATE_IRQCHIP:
1321 r = -ENOMEM;
1322 kvm->vpic = kvm_create_pic(kvm);
1323 if (kvm->vpic) {
1324 r = kvm_ioapic_init(kvm);
1325 if (r) {
1326 kfree(kvm->vpic);
1327 kvm->vpic = NULL;
1328 goto out;
1329 }
1330 } else
1331 goto out;
1332 break;
1333 case KVM_IRQ_LINE: {
1334 struct kvm_irq_level irq_event;
1335
1336 r = -EFAULT;
1337 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1338 goto out;
1339 if (irqchip_in_kernel(kvm)) {
1340 mutex_lock(&kvm->lock);
1341 if (irq_event.irq < 16)
1342 kvm_pic_set_irq(pic_irqchip(kvm),
1343 irq_event.irq,
1344 irq_event.level);
1345 kvm_ioapic_set_irq(kvm->vioapic,
1346 irq_event.irq,
1347 irq_event.level);
1348 mutex_unlock(&kvm->lock);
1349 r = 0;
1350 }
1351 break;
1352 }
1353 case KVM_GET_IRQCHIP: {
1354 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1355 struct kvm_irqchip chip;
1356
1357 r = -EFAULT;
1358 if (copy_from_user(&chip, argp, sizeof chip))
1359 goto out;
1360 r = -ENXIO;
1361 if (!irqchip_in_kernel(kvm))
1362 goto out;
1363 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1364 if (r)
1365 goto out;
1366 r = -EFAULT;
1367 if (copy_to_user(argp, &chip, sizeof chip))
1368 goto out;
1369 r = 0;
1370 break;
1371 }
1372 case KVM_SET_IRQCHIP: {
1373 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1374 struct kvm_irqchip chip;
1375
1376 r = -EFAULT;
1377 if (copy_from_user(&chip, argp, sizeof chip))
1378 goto out;
1379 r = -ENXIO;
1380 if (!irqchip_in_kernel(kvm))
1381 goto out;
1382 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1383 if (r)
1384 goto out;
1385 r = 0;
1386 break;
1387 }
07716717
DK		 1388 case KVM_GET_SUPPORTED_CPUID: {
1389 struct kvm_cpuid2 __user *cpuid_arg = argp;
1390 struct kvm_cpuid2 cpuid;
1391
1392 r = -EFAULT;
1393 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1394 goto out;
1395 r = kvm_vm_ioctl_get_supported_cpuid(kvm, &cpuid,
1396 cpuid_arg->entries);
1397 if (r)
1398 goto out;
1399
1400 r = -EFAULT;
1401 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1402 goto out;
1403 r = 0;
1404 break;
1405 }
1fe779f8
CO		 1406 default:
1407 ;
1408 }
1409out:
1410 return r;
1411}
1412
a16b043c 1413static void kvm_init_msr_list(void)
043405e1
CO		 1414{
1415 u32 dummy[2];
1416 unsigned i, j;
1417
1418 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1419 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1420 continue;
1421 if (j < i)
1422 msrs_to_save[j] = msrs_to_save[i];
1423 j++;
1424 }
1425 num_msrs_to_save = j;
1426}
1427
bbd9b64e
CO		 1428/*
1429 * Only apic need an MMIO device hook, so shortcut now..
1430 */
1431static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1432 gpa_t addr)
1433{
1434 struct kvm_io_device *dev;
1435
1436 if (vcpu->apic) {
1437 dev = &vcpu->apic->dev;
1438 if (dev->in_range(dev, addr))
1439 return dev;
1440 }
1441 return NULL;
1442}
1443
1444
1445static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1446 gpa_t addr)
1447{
1448 struct kvm_io_device *dev;
1449
1450 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1451 if (dev == NULL)
1452 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1453 return dev;
1454}
1455
1456int emulator_read_std(unsigned long addr,
1457 void *val,
1458 unsigned int bytes,
1459 struct kvm_vcpu *vcpu)
1460{
1461 void *data = val;
1462
1463 while (bytes) {
1464 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1465 unsigned offset = addr & (PAGE_SIZE-1);
1466 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1467 int ret;
1468
1469 if (gpa == UNMAPPED_GVA)
1470 return X86EMUL_PROPAGATE_FAULT;
1471 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1472 if (ret < 0)
1473 return X86EMUL_UNHANDLEABLE;
1474
1475 bytes -= tocopy;
1476 data += tocopy;
1477 addr += tocopy;
1478 }
1479
1480 return X86EMUL_CONTINUE;
1481}
1482EXPORT_SYMBOL_GPL(emulator_read_std);
1483
bbd9b64e
CO		 1484static int emulator_read_emulated(unsigned long addr,
1485 void *val,
1486 unsigned int bytes,
1487 struct kvm_vcpu *vcpu)
1488{
1489 struct kvm_io_device *mmio_dev;
1490 gpa_t gpa;
1491
1492 if (vcpu->mmio_read_completed) {
1493 memcpy(val, vcpu->mmio_data, bytes);
1494 vcpu->mmio_read_completed = 0;
1495 return X86EMUL_CONTINUE;
1496 }
1497
1498 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1499
1500 /* For APIC access vmexit */
1501 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1502 goto mmio;
1503
1504 if (emulator_read_std(addr, val, bytes, vcpu)
1505 == X86EMUL_CONTINUE)
1506 return X86EMUL_CONTINUE;
1507 if (gpa == UNMAPPED_GVA)
1508 return X86EMUL_PROPAGATE_FAULT;
1509
1510mmio:
1511 /*
1512 * Is this MMIO handled locally?
1513 */
1514 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1515 if (mmio_dev) {
1516 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1517 return X86EMUL_CONTINUE;
1518 }
1519
1520 vcpu->mmio_needed = 1;
1521 vcpu->mmio_phys_addr = gpa;
1522 vcpu->mmio_size = bytes;
1523 vcpu->mmio_is_write = 0;
1524
1525 return X86EMUL_UNHANDLEABLE;
1526}
1527
1528static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1529 const void *val, int bytes)
1530{
1531 int ret;
1532
1533 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1534 if (ret < 0)
1535 return 0;
1536 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1537 return 1;
1538}
1539
1540static int emulator_write_emulated_onepage(unsigned long addr,
1541 const void *val,
1542 unsigned int bytes,
1543 struct kvm_vcpu *vcpu)
1544{
1545 struct kvm_io_device *mmio_dev;
1546 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1547
1548 if (gpa == UNMAPPED_GVA) {
1549 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1550 return X86EMUL_PROPAGATE_FAULT;
1551 }
1552
1553 /* For APIC access vmexit */
1554 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1555 goto mmio;
1556
1557 if (emulator_write_phys(vcpu, gpa, val, bytes))
1558 return X86EMUL_CONTINUE;
1559
1560mmio:
1561 /*
1562 * Is this MMIO handled locally?
1563 */
1564 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1565 if (mmio_dev) {
1566 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1567 return X86EMUL_CONTINUE;
1568 }
1569
1570 vcpu->mmio_needed = 1;
1571 vcpu->mmio_phys_addr = gpa;
1572 vcpu->mmio_size = bytes;
1573 vcpu->mmio_is_write = 1;
1574 memcpy(vcpu->mmio_data, val, bytes);
1575
1576 return X86EMUL_CONTINUE;
1577}
1578
1579int emulator_write_emulated(unsigned long addr,
1580 const void *val,
1581 unsigned int bytes,
1582 struct kvm_vcpu *vcpu)
1583{
1584 /* Crossing a page boundary? */
1585 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1586 int rc, now;
1587
1588 now = -addr & ~PAGE_MASK;
1589 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1590 if (rc != X86EMUL_CONTINUE)
1591 return rc;
1592 addr += now;
1593 val += now;
1594 bytes -= now;
1595 }
1596 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1597}
1598EXPORT_SYMBOL_GPL(emulator_write_emulated);
1599
1600static int emulator_cmpxchg_emulated(unsigned long addr,
1601 const void *old,
1602 const void *new,
1603 unsigned int bytes,
1604 struct kvm_vcpu *vcpu)
1605{
1606 static int reported;
1607
1608 if (!reported) {
1609 reported = 1;
1610 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1611 }
1612 return emulator_write_emulated(addr, new, bytes, vcpu);
1613}
1614
1615static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1616{
1617 return kvm_x86_ops->get_segment_base(vcpu, seg);
1618}
1619
1620int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1621{
1622 return X86EMUL_CONTINUE;
1623}
1624
1625int emulate_clts(struct kvm_vcpu *vcpu)
1626{
1627 kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1628 return X86EMUL_CONTINUE;
1629}
1630
1631int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1632{
1633 struct kvm_vcpu *vcpu = ctxt->vcpu;
1634
1635 switch (dr) {
1636 case 0 ... 3:
1637 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1638 return X86EMUL_CONTINUE;
1639 default:
1640 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1641 return X86EMUL_UNHANDLEABLE;
1642 }
1643}
1644
1645int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1646{
1647 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1648 int exception;
1649
1650 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1651 if (exception) {
1652 /* FIXME: better handling */
1653 return X86EMUL_UNHANDLEABLE;
1654 }
1655 return X86EMUL_CONTINUE;
1656}
1657
1658void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1659{
1660 static int reported;
1661 u8 opcodes[4];
1662 unsigned long rip = vcpu->rip;
1663 unsigned long rip_linear;
1664
1665 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1666
1667 if (reported)
1668 return;
1669
1670 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1671
1672 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1673 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1674 reported = 1;
1675}
1676EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1677
1678struct x86_emulate_ops emulate_ops = {
1679 .read_std = emulator_read_std,
bbd9b64e
CO		 1680 .read_emulated = emulator_read_emulated,
1681 .write_emulated = emulator_write_emulated,
1682 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1683};
1684
1685int emulate_instruction(struct kvm_vcpu *vcpu,
1686 struct kvm_run *run,
1687 unsigned long cr2,
1688 u16 error_code,
1689 int no_decode)
1690{
1691 int r;
1692
1693 vcpu->mmio_fault_cr2 = cr2;
1694 kvm_x86_ops->cache_regs(vcpu);
1695
1696 vcpu->mmio_is_write = 0;
1697 vcpu->pio.string = 0;
1698
1699 if (!no_decode) {
1700 int cs_db, cs_l;
1701 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1702
1703 vcpu->emulate_ctxt.vcpu = vcpu;
1704 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
bbd9b64e
CO		 1705 vcpu->emulate_ctxt.mode =
1706 (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1707 ? X86EMUL_MODE_REAL : cs_l
1708 ? X86EMUL_MODE_PROT64 : cs_db
1709 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1710
1711 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1712 vcpu->emulate_ctxt.cs_base = 0;
1713 vcpu->emulate_ctxt.ds_base = 0;
1714 vcpu->emulate_ctxt.es_base = 0;
1715 vcpu->emulate_ctxt.ss_base = 0;
1716 } else {
1717 vcpu->emulate_ctxt.cs_base =
1718 get_segment_base(vcpu, VCPU_SREG_CS);
1719 vcpu->emulate_ctxt.ds_base =
1720 get_segment_base(vcpu, VCPU_SREG_DS);
1721 vcpu->emulate_ctxt.es_base =
1722 get_segment_base(vcpu, VCPU_SREG_ES);
1723 vcpu->emulate_ctxt.ss_base =
1724 get_segment_base(vcpu, VCPU_SREG_SS);
1725 }
1726
1727 vcpu->emulate_ctxt.gs_base =
1728 get_segment_base(vcpu, VCPU_SREG_GS);
1729 vcpu->emulate_ctxt.fs_base =
1730 get_segment_base(vcpu, VCPU_SREG_FS);
1731
1732 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
f2b5756b 1733 ++vcpu->stat.insn_emulation;
bbd9b64e 1734 if (r) {
f2b5756b 1735 ++vcpu->stat.insn_emulation_fail;
bbd9b64e
CO		 1736 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1737 return EMULATE_DONE;
1738 return EMULATE_FAIL;
1739 }
1740 }
1741
1742 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1743
1744 if (vcpu->pio.string)
1745 return EMULATE_DO_MMIO;
1746
1747 if ((r || vcpu->mmio_is_write) && run) {
1748 run->exit_reason = KVM_EXIT_MMIO;
1749 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1750 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1751 run->mmio.len = vcpu->mmio_size;
1752 run->mmio.is_write = vcpu->mmio_is_write;
1753 }
1754
1755 if (r) {
1756 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1757 return EMULATE_DONE;
1758 if (!vcpu->mmio_needed) {
1759 kvm_report_emulation_failure(vcpu, "mmio");
1760 return EMULATE_FAIL;
1761 }
1762 return EMULATE_DO_MMIO;
1763 }
1764
1765 kvm_x86_ops->decache_regs(vcpu);
1766 kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1767
1768 if (vcpu->mmio_is_write) {
1769 vcpu->mmio_needed = 0;
1770 return EMULATE_DO_MMIO;
1771 }
1772
1773 return EMULATE_DONE;
1774}
1775EXPORT_SYMBOL_GPL(emulate_instruction);
1776
de7d789a
CO		 1777static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1778{
1779 int i;
1780
1781 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1782 if (vcpu->pio.guest_pages[i]) {
b4231d61 1783 kvm_release_page_dirty(vcpu->pio.guest_pages[i]);
de7d789a
CO		 1784 vcpu->pio.guest_pages[i] = NULL;
1785 }
1786}
1787
1788static int pio_copy_data(struct kvm_vcpu *vcpu)
1789{
1790 void *p = vcpu->pio_data;
1791 void *q;
1792 unsigned bytes;
1793 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1794
1795 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1796 PAGE_KERNEL);
1797 if (!q) {
1798 free_pio_guest_pages(vcpu);
1799 return -ENOMEM;
1800 }
1801 q += vcpu->pio.guest_page_offset;
1802 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1803 if (vcpu->pio.in)
1804 memcpy(q, p, bytes);
1805 else
1806 memcpy(p, q, bytes);
1807 q -= vcpu->pio.guest_page_offset;
1808 vunmap(q);
1809 free_pio_guest_pages(vcpu);
1810 return 0;
1811}
1812
1813int complete_pio(struct kvm_vcpu *vcpu)
1814{
1815 struct kvm_pio_request *io = &vcpu->pio;
1816 long delta;
1817 int r;
1818
1819 kvm_x86_ops->cache_regs(vcpu);
1820
1821 if (!io->string) {
1822 if (io->in)
1823 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1824 io->size);
1825 } else {
1826 if (io->in) {
1827 r = pio_copy_data(vcpu);
1828 if (r) {
1829 kvm_x86_ops->cache_regs(vcpu);
1830 return r;
1831 }
1832 }
1833
1834 delta = 1;
1835 if (io->rep) {
1836 delta *= io->cur_count;
1837 /*
1838 * The size of the register should really depend on
1839 * current address size.
1840 */
1841 vcpu->regs[VCPU_REGS_RCX] -= delta;
1842 }
1843 if (io->down)
1844 delta = -delta;
1845 delta *= io->size;
1846 if (io->in)
1847 vcpu->regs[VCPU_REGS_RDI] += delta;
1848 else
1849 vcpu->regs[VCPU_REGS_RSI] += delta;
1850 }
1851
1852 kvm_x86_ops->decache_regs(vcpu);
1853
1854 io->count -= io->cur_count;
1855 io->cur_count = 0;
1856
1857 return 0;
1858}
1859
1860static void kernel_pio(struct kvm_io_device *pio_dev,
1861 struct kvm_vcpu *vcpu,
1862 void *pd)
1863{
1864 /* TODO: String I/O for in kernel device */
1865
1866 mutex_lock(&vcpu->kvm->lock);
1867 if (vcpu->pio.in)
1868 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1869 vcpu->pio.size,
1870 pd);
1871 else
1872 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1873 vcpu->pio.size,
1874 pd);
1875 mutex_unlock(&vcpu->kvm->lock);
1876}
1877
1878static void pio_string_write(struct kvm_io_device *pio_dev,
1879 struct kvm_vcpu *vcpu)
1880{
1881 struct kvm_pio_request *io = &vcpu->pio;
1882 void *pd = vcpu->pio_data;
1883 int i;
1884
1885 mutex_lock(&vcpu->kvm->lock);
1886 for (i = 0; i < io->cur_count; i++) {
1887 kvm_iodevice_write(pio_dev, io->port,
1888 io->size,
1889 pd);
1890 pd += io->size;
1891 }
1892 mutex_unlock(&vcpu->kvm->lock);
1893}
1894
1895static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1896 gpa_t addr)
1897{
1898 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1899}
1900
1901int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1902 int size, unsigned port)
1903{
1904 struct kvm_io_device *pio_dev;
1905
1906 vcpu->run->exit_reason = KVM_EXIT_IO;
1907 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1908 vcpu->run->io.size = vcpu->pio.size = size;
1909 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1910 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1911 vcpu->run->io.port = vcpu->pio.port = port;
1912 vcpu->pio.in = in;
1913 vcpu->pio.string = 0;
1914 vcpu->pio.down = 0;
1915 vcpu->pio.guest_page_offset = 0;
1916 vcpu->pio.rep = 0;
1917
1918 kvm_x86_ops->cache_regs(vcpu);
1919 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1920 kvm_x86_ops->decache_regs(vcpu);
1921
1922 kvm_x86_ops->skip_emulated_instruction(vcpu);
1923
1924 pio_dev = vcpu_find_pio_dev(vcpu, port);
1925 if (pio_dev) {
1926 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1927 complete_pio(vcpu);
1928 return 1;
1929 }
1930 return 0;
1931}
1932EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1933
1934int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1935 int size, unsigned long count, int down,
1936 gva_t address, int rep, unsigned port)
1937{
1938 unsigned now, in_page;
1939 int i, ret = 0;
1940 int nr_pages = 1;
1941 struct page *page;
1942 struct kvm_io_device *pio_dev;
1943
1944 vcpu->run->exit_reason = KVM_EXIT_IO;
1945 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1946 vcpu->run->io.size = vcpu->pio.size = size;
1947 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1948 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1949 vcpu->run->io.port = vcpu->pio.port = port;
1950 vcpu->pio.in = in;
1951 vcpu->pio.string = 1;
1952 vcpu->pio.down = down;
1953 vcpu->pio.guest_page_offset = offset_in_page(address);
1954 vcpu->pio.rep = rep;
1955
1956 if (!count) {
1957 kvm_x86_ops->skip_emulated_instruction(vcpu);
1958 return 1;
1959 }
1960
1961 if (!down)
1962 in_page = PAGE_SIZE - offset_in_page(address);
1963 else
1964 in_page = offset_in_page(address) + size;
1965 now = min(count, (unsigned long)in_page / size);
1966 if (!now) {
1967 /*
1968 * String I/O straddles page boundary. Pin two guest pages
1969 * so that we satisfy atomicity constraints. Do just one
1970 * transaction to avoid complexity.
1971 */
1972 nr_pages = 2;
1973 now = 1;
1974 }
1975 if (down) {
1976 /*
1977 * String I/O in reverse. Yuck. Kill the guest, fix later.
1978 */
1979 pr_unimpl(vcpu, "guest string pio down\n");
1980 inject_gp(vcpu);
1981 return 1;
1982 }
1983 vcpu->run->io.count = now;
1984 vcpu->pio.cur_count = now;
1985
1986 if (vcpu->pio.cur_count == vcpu->pio.count)
1987 kvm_x86_ops->skip_emulated_instruction(vcpu);
1988
1989 for (i = 0; i < nr_pages; ++i) {
1990 mutex_lock(&vcpu->kvm->lock);
1991 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1992 vcpu->pio.guest_pages[i] = page;
1993 mutex_unlock(&vcpu->kvm->lock);
1994 if (!page) {
1995 inject_gp(vcpu);
1996 free_pio_guest_pages(vcpu);
1997 return 1;
1998 }
1999 }
2000
2001 pio_dev = vcpu_find_pio_dev(vcpu, port);
2002 if (!vcpu->pio.in) {
2003 /* string PIO write */
2004 ret = pio_copy_data(vcpu);
2005 if (ret >= 0 && pio_dev) {
2006 pio_string_write(pio_dev, vcpu);
2007 complete_pio(vcpu);
2008 if (vcpu->pio.count == 0)
2009 ret = 1;
2010 }
2011 } else if (pio_dev)
2012 pr_unimpl(vcpu, "no string pio read support yet, "
2013 "port %x size %d count %ld\n",
2014 port, size, count);
2015
2016 return ret;
2017}
2018EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2019
f8c16bba 2020int kvm_arch_init(void *opaque)
043405e1 2021{
56c6d28a 2022 int r;
f8c16bba
ZX		 2023 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2024
56c6d28a
ZX		 2025 r = kvm_mmu_module_init();
2026 if (r)
2027 goto out_fail;
2028
043405e1 2029 kvm_init_msr_list();
f8c16bba
ZX		 2030
2031 if (kvm_x86_ops) {
2032 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 2033 r = -EEXIST;
2034 goto out;
f8c16bba
ZX		 2035 }
2036
2037 if (!ops->cpu_has_kvm_support()) {
2038 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 2039 r = -EOPNOTSUPP;
2040 goto out;
f8c16bba
ZX		 2041 }
2042 if (ops->disabled_by_bios()) {
2043 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 2044 r = -EOPNOTSUPP;
2045 goto out;
f8c16bba
ZX		 2046 }
2047
2048 kvm_x86_ops = ops;
56c6d28a 2049 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
f8c16bba 2050 return 0;
56c6d28a
ZX		 2051
2052out:
2053 kvm_mmu_module_exit();
2054out_fail:
2055 return r;
043405e1 2056}
8776e519 2057
f8c16bba
ZX		 2058void kvm_arch_exit(void)
2059{
2060 kvm_x86_ops = NULL;
56c6d28a
ZX		 2061 kvm_mmu_module_exit();
2062}
f8c16bba 2063
8776e519
HB		 2064int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2065{
2066 ++vcpu->stat.halt_exits;
2067 if (irqchip_in_kernel(vcpu->kvm)) {
2068 vcpu->mp_state = VCPU_MP_STATE_HALTED;
2069 kvm_vcpu_block(vcpu);
2070 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
2071 return -EINTR;
2072 return 1;
2073 } else {
2074 vcpu->run->exit_reason = KVM_EXIT_HLT;
2075 return 0;
2076 }
2077}
2078EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2079
2080int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2081{
2082 unsigned long nr, a0, a1, a2, a3, ret;
2083
2084 kvm_x86_ops->cache_regs(vcpu);
2085
2086 nr = vcpu->regs[VCPU_REGS_RAX];
2087 a0 = vcpu->regs[VCPU_REGS_RBX];
2088 a1 = vcpu->regs[VCPU_REGS_RCX];
2089 a2 = vcpu->regs[VCPU_REGS_RDX];
2090 a3 = vcpu->regs[VCPU_REGS_RSI];
2091
2092 if (!is_long_mode(vcpu)) {
2093 nr &= 0xFFFFFFFF;
2094 a0 &= 0xFFFFFFFF;
2095 a1 &= 0xFFFFFFFF;
2096 a2 &= 0xFFFFFFFF;
2097 a3 &= 0xFFFFFFFF;
2098 }
2099
2100 switch (nr) {
2101 default:
2102 ret = -KVM_ENOSYS;
2103 break;
2104 }
2105 vcpu->regs[VCPU_REGS_RAX] = ret;
2106 kvm_x86_ops->decache_regs(vcpu);
2107 return 0;
2108}
2109EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2110
2111int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2112{
2113 char instruction[3];
2114 int ret = 0;
2115
2116 mutex_lock(&vcpu->kvm->lock);
2117
2118 /*
2119 * Blow out the MMU to ensure that no other VCPU has an active mapping
2120 * to ensure that the updated hypercall appears atomically across all
2121 * VCPUs.
2122 */
2123 kvm_mmu_zap_all(vcpu->kvm);
2124
2125 kvm_x86_ops->cache_regs(vcpu);
2126 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2127 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
2128 != X86EMUL_CONTINUE)
2129 ret = -EFAULT;
2130
2131 mutex_unlock(&vcpu->kvm->lock);
2132
2133 return ret;
2134}
2135
2136static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2137{
2138 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2139}
2140
2141void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2142{
2143 struct descriptor_table dt = { limit, base };
2144
2145 kvm_x86_ops->set_gdt(vcpu, &dt);
2146}
2147
2148void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2149{
2150 struct descriptor_table dt = { limit, base };
2151
2152 kvm_x86_ops->set_idt(vcpu, &dt);
2153}
2154
2155void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2156 unsigned long *rflags)
2157{
2158 lmsw(vcpu, msw);
2159 *rflags = kvm_x86_ops->get_rflags(vcpu);
2160}
2161
2162unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2163{
2164 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2165 switch (cr) {
2166 case 0:
2167 return vcpu->cr0;
2168 case 2:
2169 return vcpu->cr2;
2170 case 3:
2171 return vcpu->cr3;
2172 case 4:
2173 return vcpu->cr4;
2174 default:
2175 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2176 return 0;
2177 }
2178}
2179
2180void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2181 unsigned long *rflags)
2182{
2183 switch (cr) {
2184 case 0:
2185 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
2186 *rflags = kvm_x86_ops->get_rflags(vcpu);
2187 break;
2188 case 2:
2189 vcpu->cr2 = val;
2190 break;
2191 case 3:
2192 set_cr3(vcpu, val);
2193 break;
2194 case 4:
2195 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
2196 break;
2197 default:
2198 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2199 }
2200}
2201
07716717
DK		 2202static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2203{
2204 struct kvm_cpuid_entry2 *e = &vcpu->cpuid_entries[i];
2205 int j, nent = vcpu->cpuid_nent;
2206
2207 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2208 /* when no next entry is found, the current entry[i] is reselected */
2209 for (j = i + 1; j == i; j = (j + 1) % nent) {
2210 struct kvm_cpuid_entry2 *ej = &vcpu->cpuid_entries[j];
2211 if (ej->function == e->function) {
2212 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2213 return j;
2214 }
2215 }
2216 return 0; /* silence gcc, even though control never reaches here */
2217}
2218
2219/* find an entry with matching function, matching index (if needed), and that
2220 * should be read next (if it's stateful) */
2221static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2222 u32 function, u32 index)
2223{
2224 if (e->function != function)
2225 return 0;
2226 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2227 return 0;
2228 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2229 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2230 return 0;
2231 return 1;
2232}
2233
8776e519
HB		 2234void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2235{
2236 int i;
07716717
DK		 2237 u32 function, index;
2238 struct kvm_cpuid_entry2 *e, *best;
8776e519
HB		 2239
2240 kvm_x86_ops->cache_regs(vcpu);
2241 function = vcpu->regs[VCPU_REGS_RAX];
07716717 2242 index = vcpu->regs[VCPU_REGS_RCX];
8776e519
HB		 2243 vcpu->regs[VCPU_REGS_RAX] = 0;
2244 vcpu->regs[VCPU_REGS_RBX] = 0;
2245 vcpu->regs[VCPU_REGS_RCX] = 0;
2246 vcpu->regs[VCPU_REGS_RDX] = 0;
2247 best = NULL;
2248 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2249 e = &vcpu->cpuid_entries[i];
07716717
DK		 2250 if (is_matching_cpuid_entry(e, function, index)) {
2251 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2252 move_to_next_stateful_cpuid_entry(vcpu, i);
8776e519
HB		 2253 best = e;
2254 break;
2255 }
2256 /*
2257 * Both basic or both extended?
2258 */
2259 if (((e->function ^ function) & 0x80000000) == 0)
2260 if (!best || e->function > best->function)
2261 best = e;
2262 }
2263 if (best) {
2264 vcpu->regs[VCPU_REGS_RAX] = best->eax;
2265 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
2266 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
2267 vcpu->regs[VCPU_REGS_RDX] = best->edx;
2268 }
2269 kvm_x86_ops->decache_regs(vcpu);
2270 kvm_x86_ops->skip_emulated_instruction(vcpu);
2271}
2272EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
d0752060 2273
b6c7a5dc
HB		 2274/*
2275 * Check if userspace requested an interrupt window, and that the
2276 * interrupt window is open.
2277 *
2278 * No need to exit to userspace if we already have an interrupt queued.
2279 */
2280static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2281 struct kvm_run *kvm_run)
2282{
2283 return (!vcpu->irq_summary &&
2284 kvm_run->request_interrupt_window &&
2285 vcpu->interrupt_window_open &&
2286 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2287}
2288
2289static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2290 struct kvm_run *kvm_run)
2291{
2292 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2293 kvm_run->cr8 = get_cr8(vcpu);
2294 kvm_run->apic_base = kvm_get_apic_base(vcpu);
2295 if (irqchip_in_kernel(vcpu->kvm))
2296 kvm_run->ready_for_interrupt_injection = 1;
2297 else
2298 kvm_run->ready_for_interrupt_injection =
2299 (vcpu->interrupt_window_open &&
2300 vcpu->irq_summary == 0);
2301}
2302
2303static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2304{
2305 int r;
2306
2307 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2308 pr_debug("vcpu %d received sipi with vector # %x\n",
2309 vcpu->vcpu_id, vcpu->sipi_vector);
2310 kvm_lapic_reset(vcpu);
2311 r = kvm_x86_ops->vcpu_reset(vcpu);
2312 if (r)
2313 return r;
2314 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2315 }
2316
2317preempted:
2318 if (vcpu->guest_debug.enabled)
2319 kvm_x86_ops->guest_debug_pre(vcpu);
2320
2321again:
2322 r = kvm_mmu_reload(vcpu);
2323 if (unlikely(r))
2324 goto out;
2325
2326 kvm_inject_pending_timer_irqs(vcpu);
2327
2328 preempt_disable();
2329
2330 kvm_x86_ops->prepare_guest_switch(vcpu);
2331 kvm_load_guest_fpu(vcpu);
2332
2333 local_irq_disable();
2334
2335 if (signal_pending(current)) {
2336 local_irq_enable();
2337 preempt_enable();
2338 r = -EINTR;
2339 kvm_run->exit_reason = KVM_EXIT_INTR;
2340 ++vcpu->stat.signal_exits;
2341 goto out;
2342 }
2343
2344 if (irqchip_in_kernel(vcpu->kvm))
2345 kvm_x86_ops->inject_pending_irq(vcpu);
eb9774f0 2346 else
b6c7a5dc
HB		 2347 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2348
2349 vcpu->guest_mode = 1;
2350 kvm_guest_enter();
2351
2352 if (vcpu->requests)
2353 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2354 kvm_x86_ops->tlb_flush(vcpu);
2355
2356 kvm_x86_ops->run(vcpu, kvm_run);
2357
2358 vcpu->guest_mode = 0;
2359 local_irq_enable();
2360
2361 ++vcpu->stat.exits;
2362
2363 /*
2364 * We must have an instruction between local_irq_enable() and
2365 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2366 * the interrupt shadow. The stat.exits increment will do nicely.
2367 * But we need to prevent reordering, hence this barrier():
2368 */
2369 barrier();
2370
2371 kvm_guest_exit();
2372
2373 preempt_enable();
2374
2375 /*
2376 * Profile KVM exit RIPs:
2377 */
2378 if (unlikely(prof_on == KVM_PROFILING)) {
2379 kvm_x86_ops->cache_regs(vcpu);
2380 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2381 }
2382
2383 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2384
2385 if (r > 0) {
2386 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2387 r = -EINTR;
2388 kvm_run->exit_reason = KVM_EXIT_INTR;
2389 ++vcpu->stat.request_irq_exits;
2390 goto out;
2391 }
e1beb1d3 2392 if (!need_resched())
b6c7a5dc 2393 goto again;
b6c7a5dc
HB		 2394 }
2395
2396out:
2397 if (r > 0) {
2398 kvm_resched(vcpu);
2399 goto preempted;
2400 }
2401
2402 post_kvm_run_save(vcpu, kvm_run);
2403
2404 return r;
2405}
2406
2407int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2408{
2409 int r;
2410 sigset_t sigsaved;
2411
2412 vcpu_load(vcpu);
2413
2414 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2415 kvm_vcpu_block(vcpu);
2416 vcpu_put(vcpu);
2417 return -EAGAIN;
2418 }
2419
2420 if (vcpu->sigset_active)
2421 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2422
2423 /* re-sync apic's tpr */
2424 if (!irqchip_in_kernel(vcpu->kvm))
2425 set_cr8(vcpu, kvm_run->cr8);
2426
2427 if (vcpu->pio.cur_count) {
2428 r = complete_pio(vcpu);
2429 if (r)
2430 goto out;
2431 }
2432#if CONFIG_HAS_IOMEM
2433 if (vcpu->mmio_needed) {
2434 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2435 vcpu->mmio_read_completed = 1;
2436 vcpu->mmio_needed = 0;
2437 r = emulate_instruction(vcpu, kvm_run,
2438 vcpu->mmio_fault_cr2, 0, 1);
2439 if (r == EMULATE_DO_MMIO) {
2440 /*
2441 * Read-modify-write. Back to userspace.
2442 */
2443 r = 0;
2444 goto out;
2445 }
2446 }
2447#endif
2448 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2449 kvm_x86_ops->cache_regs(vcpu);
2450 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2451 kvm_x86_ops->decache_regs(vcpu);
2452 }
2453
2454 r = __vcpu_run(vcpu, kvm_run);
2455
2456out:
2457 if (vcpu->sigset_active)
2458 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2459
2460 vcpu_put(vcpu);
2461 return r;
2462}
2463
2464int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2465{
2466 vcpu_load(vcpu);
2467
2468 kvm_x86_ops->cache_regs(vcpu);
2469
2470 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2471 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2472 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2473 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2474 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2475 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2476 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2477 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2478#ifdef CONFIG_X86_64
2479 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2480 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2481 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2482 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2483 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2484 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2485 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2486 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2487#endif
2488
2489 regs->rip = vcpu->rip;
2490 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2491
2492 /*
2493 * Don't leak debug flags in case they were set for guest debugging
2494 */
2495 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2496 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2497
2498 vcpu_put(vcpu);
2499
2500 return 0;
2501}
2502
2503int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2504{
2505 vcpu_load(vcpu);
2506
2507 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2508 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2509 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2510 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2511 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2512 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2513 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2514 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2515#ifdef CONFIG_X86_64
2516 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2517 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2518 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2519 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2520 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2521 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2522 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2523 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2524#endif
2525
2526 vcpu->rip = regs->rip;
2527 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2528
2529 kvm_x86_ops->decache_regs(vcpu);
2530
2531 vcpu_put(vcpu);
2532
2533 return 0;
2534}
2535
2536static void get_segment(struct kvm_vcpu *vcpu,
2537 struct kvm_segment *var, int seg)
2538{
2539 return kvm_x86_ops->get_segment(vcpu, var, seg);
2540}
2541
2542void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2543{
2544 struct kvm_segment cs;
2545
2546 get_segment(vcpu, &cs, VCPU_SREG_CS);
2547 *db = cs.db;
2548 *l = cs.l;
2549}
2550EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2551
2552int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2553 struct kvm_sregs *sregs)
2554{
2555 struct descriptor_table dt;
2556 int pending_vec;
2557
2558 vcpu_load(vcpu);
2559
2560 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2561 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2562 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2563 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2564 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2565 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2566
2567 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2568 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2569
2570 kvm_x86_ops->get_idt(vcpu, &dt);
2571 sregs->idt.limit = dt.limit;
2572 sregs->idt.base = dt.base;
2573 kvm_x86_ops->get_gdt(vcpu, &dt);
2574 sregs->gdt.limit = dt.limit;
2575 sregs->gdt.base = dt.base;
2576
2577 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2578 sregs->cr0 = vcpu->cr0;
2579 sregs->cr2 = vcpu->cr2;
2580 sregs->cr3 = vcpu->cr3;
2581 sregs->cr4 = vcpu->cr4;
2582 sregs->cr8 = get_cr8(vcpu);
2583 sregs->efer = vcpu->shadow_efer;
2584 sregs->apic_base = kvm_get_apic_base(vcpu);
2585
2586 if (irqchip_in_kernel(vcpu->kvm)) {
2587 memset(sregs->interrupt_bitmap, 0,
2588 sizeof sregs->interrupt_bitmap);
2589 pending_vec = kvm_x86_ops->get_irq(vcpu);
2590 if (pending_vec >= 0)
2591 set_bit(pending_vec,
2592 (unsigned long *)sregs->interrupt_bitmap);
2593 } else
2594 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2595 sizeof sregs->interrupt_bitmap);
2596
2597 vcpu_put(vcpu);
2598
2599 return 0;
2600}
2601
2602static void set_segment(struct kvm_vcpu *vcpu,
2603 struct kvm_segment *var, int seg)
2604{
2605 return kvm_x86_ops->set_segment(vcpu, var, seg);
2606}
2607
2608int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2609 struct kvm_sregs *sregs)
2610{
2611 int mmu_reset_needed = 0;
2612 int i, pending_vec, max_bits;
2613 struct descriptor_table dt;
2614
2615 vcpu_load(vcpu);
2616
2617 dt.limit = sregs->idt.limit;
2618 dt.base = sregs->idt.base;
2619 kvm_x86_ops->set_idt(vcpu, &dt);
2620 dt.limit = sregs->gdt.limit;
2621 dt.base = sregs->gdt.base;
2622 kvm_x86_ops->set_gdt(vcpu, &dt);
2623
2624 vcpu->cr2 = sregs->cr2;
2625 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2626 vcpu->cr3 = sregs->cr3;
2627
2628 set_cr8(vcpu, sregs->cr8);
2629
2630 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2631#ifdef CONFIG_X86_64
2632 kvm_x86_ops->set_efer(vcpu, sregs->efer);
2633#endif
2634 kvm_set_apic_base(vcpu, sregs->apic_base);
2635
2636 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2637
2638 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2639 vcpu->cr0 = sregs->cr0;
2640 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2641
2642 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2643 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2644 if (!is_long_mode(vcpu) && is_pae(vcpu))
2645 load_pdptrs(vcpu, vcpu->cr3);
2646
2647 if (mmu_reset_needed)
2648 kvm_mmu_reset_context(vcpu);
2649
2650 if (!irqchip_in_kernel(vcpu->kvm)) {
2651 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2652 sizeof vcpu->irq_pending);
2653 vcpu->irq_summary = 0;
2654 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2655 if (vcpu->irq_pending[i])
2656 __set_bit(i, &vcpu->irq_summary);
2657 } else {
2658 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2659 pending_vec = find_first_bit(
2660 (const unsigned long *)sregs->interrupt_bitmap,
2661 max_bits);
2662 /* Only pending external irq is handled here */
2663 if (pending_vec < max_bits) {
2664 kvm_x86_ops->set_irq(vcpu, pending_vec);
2665 pr_debug("Set back pending irq %d\n",
2666 pending_vec);
2667 }
2668 }
2669
2670 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2671 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2672 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2673 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2674 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2675 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2676
2677 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2678 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2679
2680 vcpu_put(vcpu);
2681
2682 return 0;
2683}
2684
2685int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2686 struct kvm_debug_guest *dbg)
2687{
2688 int r;
2689
2690 vcpu_load(vcpu);
2691
2692 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2693
2694 vcpu_put(vcpu);
2695
2696 return r;
2697}
2698
d0752060
HB		 2699/*
2700 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2701 * we have asm/x86/processor.h
2702 */
2703struct fxsave {
2704 u16 cwd;
2705 u16 swd;
2706 u16 twd;
2707 u16 fop;
2708 u64 rip;
2709 u64 rdp;
2710 u32 mxcsr;
2711 u32 mxcsr_mask;
2712 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2713#ifdef CONFIG_X86_64
2714 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2715#else
2716 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2717#endif
2718};
2719
8b006791
ZX		 2720/*
2721 * Translate a guest virtual address to a guest physical address.
2722 */
2723int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2724 struct kvm_translation *tr)
2725{
2726 unsigned long vaddr = tr->linear_address;
2727 gpa_t gpa;
2728
2729 vcpu_load(vcpu);
2730 mutex_lock(&vcpu->kvm->lock);
2731 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2732 tr->physical_address = gpa;
2733 tr->valid = gpa != UNMAPPED_GVA;
2734 tr->writeable = 1;
2735 tr->usermode = 0;
2736 mutex_unlock(&vcpu->kvm->lock);
2737 vcpu_put(vcpu);
2738
2739 return 0;
2740}
2741
d0752060
HB		 2742int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2743{
2744 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2745
2746 vcpu_load(vcpu);
2747
2748 memcpy(fpu->fpr, fxsave->st_space, 128);
2749 fpu->fcw = fxsave->cwd;
2750 fpu->fsw = fxsave->swd;
2751 fpu->ftwx = fxsave->twd;
2752 fpu->last_opcode = fxsave->fop;
2753 fpu->last_ip = fxsave->rip;
2754 fpu->last_dp = fxsave->rdp;
2755 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2756
2757 vcpu_put(vcpu);
2758
2759 return 0;
2760}
2761
2762int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2763{
2764 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2765
2766 vcpu_load(vcpu);
2767
2768 memcpy(fxsave->st_space, fpu->fpr, 128);
2769 fxsave->cwd = fpu->fcw;
2770 fxsave->swd = fpu->fsw;
2771 fxsave->twd = fpu->ftwx;
2772 fxsave->fop = fpu->last_opcode;
2773 fxsave->rip = fpu->last_ip;
2774 fxsave->rdp = fpu->last_dp;
2775 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2776
2777 vcpu_put(vcpu);
2778
2779 return 0;
2780}
2781
2782void fx_init(struct kvm_vcpu *vcpu)
2783{
2784 unsigned after_mxcsr_mask;
2785
2786 /* Initialize guest FPU by resetting ours and saving into guest's */
2787 preempt_disable();
2788 fx_save(&vcpu->host_fx_image);
2789 fpu_init();
2790 fx_save(&vcpu->guest_fx_image);
2791 fx_restore(&vcpu->host_fx_image);
2792 preempt_enable();
2793
2794 vcpu->cr0 |= X86_CR0_ET;
2795 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2796 vcpu->guest_fx_image.mxcsr = 0x1f80;
2797 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2798 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2799}
2800EXPORT_SYMBOL_GPL(fx_init);
2801
2802void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2803{
2804 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2805 return;
2806
2807 vcpu->guest_fpu_loaded = 1;
2808 fx_save(&vcpu->host_fx_image);
2809 fx_restore(&vcpu->guest_fx_image);
2810}
2811EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2812
2813void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2814{
2815 if (!vcpu->guest_fpu_loaded)
2816 return;
2817
2818 vcpu->guest_fpu_loaded = 0;
2819 fx_save(&vcpu->guest_fx_image);
2820 fx_restore(&vcpu->host_fx_image);
f096ed85 2821 ++vcpu->stat.fpu_reload;
d0752060
HB		 2822}
2823EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
e9b11c17
ZX		 2824
2825void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
2826{
2827 kvm_x86_ops->vcpu_free(vcpu);
2828}
2829
2830struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2831 unsigned int id)
2832{
26e5215f
AK		 2833 return kvm_x86_ops->vcpu_create(kvm, id);
2834}
e9b11c17 2835
26e5215f
AK		 2836int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
2837{
2838 int r;
e9b11c17
ZX		 2839
2840 /* We do fxsave: this must be aligned. */
2841 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2842
2843 vcpu_load(vcpu);
2844 r = kvm_arch_vcpu_reset(vcpu);
2845 if (r == 0)
2846 r = kvm_mmu_setup(vcpu);
2847 vcpu_put(vcpu);
2848 if (r < 0)
2849 goto free_vcpu;
2850
26e5215f 2851 return 0;
e9b11c17
ZX		 2852free_vcpu:
2853 kvm_x86_ops->vcpu_free(vcpu);
26e5215f 2854 return r;
e9b11c17
ZX		 2855}
2856
d40ccc62 2857void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17
ZX		 2858{
2859 vcpu_load(vcpu);
2860 kvm_mmu_unload(vcpu);
2861 vcpu_put(vcpu);
2862
2863 kvm_x86_ops->vcpu_free(vcpu);
2864}
2865
2866int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
2867{
2868 return kvm_x86_ops->vcpu_reset(vcpu);
2869}
2870
2871void kvm_arch_hardware_enable(void *garbage)
2872{
2873 kvm_x86_ops->hardware_enable(garbage);
2874}
2875
2876void kvm_arch_hardware_disable(void *garbage)
2877{
2878 kvm_x86_ops->hardware_disable(garbage);
2879}
2880
2881int kvm_arch_hardware_setup(void)
2882{
2883 return kvm_x86_ops->hardware_setup();
2884}
2885
2886void kvm_arch_hardware_unsetup(void)
2887{
2888 kvm_x86_ops->hardware_unsetup();
2889}
2890
2891void kvm_arch_check_processor_compat(void *rtn)
2892{
2893 kvm_x86_ops->check_processor_compatibility(rtn);
2894}
2895
2896int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2897{
2898 struct page *page;
2899 struct kvm *kvm;
2900 int r;
2901
2902 BUG_ON(vcpu->kvm == NULL);
2903 kvm = vcpu->kvm;
2904
2905 vcpu->mmu.root_hpa = INVALID_PAGE;
2906 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
2907 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2908 else
2909 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
2910
2911 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2912 if (!page) {
2913 r = -ENOMEM;
2914 goto fail;
2915 }
2916 vcpu->pio_data = page_address(page);
2917
2918 r = kvm_mmu_create(vcpu);
2919 if (r < 0)
2920 goto fail_free_pio_data;
2921
2922 if (irqchip_in_kernel(kvm)) {
2923 r = kvm_create_lapic(vcpu);
2924 if (r < 0)
2925 goto fail_mmu_destroy;
2926 }
2927
2928 return 0;
2929
2930fail_mmu_destroy:
2931 kvm_mmu_destroy(vcpu);
2932fail_free_pio_data:
2933 free_page((unsigned long)vcpu->pio_data);
2934fail:
2935 return r;
2936}
2937
2938void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
2939{
2940 kvm_free_lapic(vcpu);
2941 kvm_mmu_destroy(vcpu);
2942 free_page((unsigned long)vcpu->pio_data);
2943}
d19a9cd2
ZX		 2944
2945struct kvm *kvm_arch_create_vm(void)
2946{
2947 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
2948
2949 if (!kvm)
2950 return ERR_PTR(-ENOMEM);
2951
2952 INIT_LIST_HEAD(&kvm->active_mmu_pages);
2953
2954 return kvm;
2955}
2956
2957static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
2958{
2959 vcpu_load(vcpu);
2960 kvm_mmu_unload(vcpu);
2961 vcpu_put(vcpu);
2962}
2963
2964static void kvm_free_vcpus(struct kvm *kvm)
2965{
2966 unsigned int i;
2967
2968 /*
2969 * Unpin any mmu pages first.
2970 */
2971 for (i = 0; i < KVM_MAX_VCPUS; ++i)
2972 if (kvm->vcpus[i])
2973 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
2974 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2975 if (kvm->vcpus[i]) {
2976 kvm_arch_vcpu_free(kvm->vcpus[i]);
2977 kvm->vcpus[i] = NULL;
2978 }
2979 }
2980
2981}
2982
2983void kvm_arch_destroy_vm(struct kvm *kvm)
2984{
2985 kfree(kvm->vpic);
2986 kfree(kvm->vioapic);
2987 kvm_free_vcpus(kvm);
2988 kvm_free_physmem(kvm);
2989 kfree(kvm);
2990}
0de10343
ZX		 2991
2992int kvm_arch_set_memory_region(struct kvm *kvm,
2993 struct kvm_userspace_memory_region *mem,
2994 struct kvm_memory_slot old,
2995 int user_alloc)
2996{
2997 int npages = mem->memory_size >> PAGE_SHIFT;
2998 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
2999
3000 /*To keep backward compatibility with older userspace,
3001 *x86 needs to hanlde !user_alloc case.
3002 */
3003 if (!user_alloc) {
3004 if (npages && !old.rmap) {
3005 down_write(&current->mm->mmap_sem);
3006 memslot->userspace_addr = do_mmap(NULL, 0,
3007 npages * PAGE_SIZE,
3008 PROT_READ | PROT_WRITE,
3009 MAP_SHARED | MAP_ANONYMOUS,
3010 0);
3011 up_write(&current->mm->mmap_sem);
3012
3013 if (IS_ERR((void *)memslot->userspace_addr))
3014 return PTR_ERR((void *)memslot->userspace_addr);
3015 } else {
3016 if (!old.user_alloc && old.rmap) {
3017 int ret;
3018
3019 down_write(&current->mm->mmap_sem);
3020 ret = do_munmap(current->mm, old.userspace_addr,
3021 old.npages * PAGE_SIZE);
3022 up_write(&current->mm->mmap_sem);
3023 if (ret < 0)
3024 printk(KERN_WARNING
3025 "kvm_vm_ioctl_set_memory_region: "
3026 "failed to munmap memory\n");
3027 }
3028 }
3029 }
3030
3031 if (!kvm->n_requested_mmu_pages) {
3032 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3033 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3034 }
3035
3036 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3037 kvm_flush_remote_tlbs(kvm);
3038
3039 return 0;
3040}
Ubuntu Artful kernel mirror