projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: MMU: Remove unused prev_shadow_ent variable from fetch()
[linux-2.6-block.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
f77bc6a4
ZX		 1004static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1005 struct kvm_interrupt *irq)
1006{
1007 if (irq->irq < 0 || irq->irq >= 256)
1008 return -EINVAL;
1009 if (irqchip_in_kernel(vcpu->kvm))
1010 return -ENXIO;
1011 vcpu_load(vcpu);
1012
1013 set_bit(irq->irq, vcpu->irq_pending);
1014 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1015
1016 vcpu_put(vcpu);
1017
1018 return 0;
1019}
1020
313a3dc7
CO		 1021long kvm_arch_vcpu_ioctl(struct file *filp,
1022 unsigned int ioctl, unsigned long arg)
1023{
1024 struct kvm_vcpu *vcpu = filp->private_data;
1025 void __user *argp = (void __user *)arg;
1026 int r;
1027
1028 switch (ioctl) {
1029 case KVM_GET_LAPIC: {
1030 struct kvm_lapic_state lapic;
1031
1032 memset(&lapic, 0, sizeof lapic);
1033 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1034 if (r)
1035 goto out;
1036 r = -EFAULT;
1037 if (copy_to_user(argp, &lapic, sizeof lapic))
1038 goto out;
1039 r = 0;
1040 break;
1041 }
1042 case KVM_SET_LAPIC: {
1043 struct kvm_lapic_state lapic;
1044
1045 r = -EFAULT;
1046 if (copy_from_user(&lapic, argp, sizeof lapic))
1047 goto out;
1048 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1049 if (r)
1050 goto out;
1051 r = 0;
1052 break;
1053 }
f77bc6a4
ZX		 1054 case KVM_INTERRUPT: {
1055 struct kvm_interrupt irq;
1056
1057 r = -EFAULT;
1058 if (copy_from_user(&irq, argp, sizeof irq))
1059 goto out;
1060 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1061 if (r)
1062 goto out;
1063 r = 0;
1064 break;
1065 }
313a3dc7
CO		 1066 case KVM_SET_CPUID: {
1067 struct kvm_cpuid __user *cpuid_arg = argp;
1068 struct kvm_cpuid cpuid;
1069
1070 r = -EFAULT;
1071 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1072 goto out;
1073 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1074 if (r)
1075 goto out;
1076 break;
1077 }
07716717
DK		 1078 case KVM_SET_CPUID2: {
1079 struct kvm_cpuid2 __user *cpuid_arg = argp;
1080 struct kvm_cpuid2 cpuid;
1081
1082 r = -EFAULT;
1083 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1084 goto out;
1085 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1086 cpuid_arg->entries);
1087 if (r)
1088 goto out;
1089 break;
1090 }
1091 case KVM_GET_CPUID2: {
1092 struct kvm_cpuid2 __user *cpuid_arg = argp;
1093 struct kvm_cpuid2 cpuid;
1094
1095 r = -EFAULT;
1096 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1097 goto out;
1098 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1099 cpuid_arg->entries);
1100 if (r)
1101 goto out;
1102 r = -EFAULT;
1103 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1104 goto out;
1105 r = 0;
1106 break;
1107 }
313a3dc7
CO		 1108 case KVM_GET_MSRS:
1109 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1110 break;
1111 case KVM_SET_MSRS:
1112 r = msr_io(vcpu, argp, do_set_msr, 0);
1113 break;
1114 default:
1115 r = -EINVAL;
1116 }
1117out:
1118 return r;
1119}
1120
1fe779f8
CO		 1121static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1122{
1123 int ret;
1124
1125 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1126 return -1;
1127 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1128 return ret;
1129}
1130
1131static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1132 u32 kvm_nr_mmu_pages)
1133{
1134 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1135 return -EINVAL;
1136
1137 mutex_lock(&kvm->lock);
1138
1139 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1140 kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
1141
1142 mutex_unlock(&kvm->lock);
1143 return 0;
1144}
1145
1146static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1147{
1148 return kvm->n_alloc_mmu_pages;
1149}
1150
e9f85cde
ZX		 1151gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1152{
1153 int i;
1154 struct kvm_mem_alias *alias;
1155
1156 for (i = 0; i < kvm->naliases; ++i) {
1157 alias = &kvm->aliases[i];
1158 if (gfn >= alias->base_gfn
1159 && gfn < alias->base_gfn + alias->npages)
1160 return alias->target_gfn + gfn - alias->base_gfn;
1161 }
1162 return gfn;
1163}
1164
1fe779f8
CO		 1165/*
1166 * Set a new alias region. Aliases map a portion of physical memory into
1167 * another portion. This is useful for memory windows, for example the PC
1168 * VGA region.
1169 */
1170static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1171 struct kvm_memory_alias *alias)
1172{
1173 int r, n;
1174 struct kvm_mem_alias *p;
1175
1176 r = -EINVAL;
1177 /* General sanity checks */
1178 if (alias->memory_size & (PAGE_SIZE - 1))
1179 goto out;
1180 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1181 goto out;
1182 if (alias->slot >= KVM_ALIAS_SLOTS)
1183 goto out;
1184 if (alias->guest_phys_addr + alias->memory_size
1185 < alias->guest_phys_addr)
1186 goto out;
1187 if (alias->target_phys_addr + alias->memory_size
1188 < alias->target_phys_addr)
1189 goto out;
1190
1191 mutex_lock(&kvm->lock);
1192
1193 p = &kvm->aliases[alias->slot];
1194 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1195 p->npages = alias->memory_size >> PAGE_SHIFT;
1196 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1197
1198 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1199 if (kvm->aliases[n - 1].npages)
1200 break;
1201 kvm->naliases = n;
1202
1203 kvm_mmu_zap_all(kvm);
1204
1205 mutex_unlock(&kvm->lock);
1206
1207 return 0;
1208
1209out:
1210 return r;
1211}
1212
1213static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1214{
1215 int r;
1216
1217 r = 0;
1218 switch (chip->chip_id) {
1219 case KVM_IRQCHIP_PIC_MASTER:
1220 memcpy(&chip->chip.pic,
1221 &pic_irqchip(kvm)->pics[0],
1222 sizeof(struct kvm_pic_state));
1223 break;
1224 case KVM_IRQCHIP_PIC_SLAVE:
1225 memcpy(&chip->chip.pic,
1226 &pic_irqchip(kvm)->pics[1],
1227 sizeof(struct kvm_pic_state));
1228 break;
1229 case KVM_IRQCHIP_IOAPIC:
1230 memcpy(&chip->chip.ioapic,
1231 ioapic_irqchip(kvm),
1232 sizeof(struct kvm_ioapic_state));
1233 break;
1234 default:
1235 r = -EINVAL;
1236 break;
1237 }
1238 return r;
1239}
1240
1241static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1242{
1243 int r;
1244
1245 r = 0;
1246 switch (chip->chip_id) {
1247 case KVM_IRQCHIP_PIC_MASTER:
1248 memcpy(&pic_irqchip(kvm)->pics[0],
1249 &chip->chip.pic,
1250 sizeof(struct kvm_pic_state));
1251 break;
1252 case KVM_IRQCHIP_PIC_SLAVE:
1253 memcpy(&pic_irqchip(kvm)->pics[1],
1254 &chip->chip.pic,
1255 sizeof(struct kvm_pic_state));
1256 break;
1257 case KVM_IRQCHIP_IOAPIC:
1258 memcpy(ioapic_irqchip(kvm),
1259 &chip->chip.ioapic,
1260 sizeof(struct kvm_ioapic_state));
1261 break;
1262 default:
1263 r = -EINVAL;
1264 break;
1265 }
1266 kvm_pic_update_irq(pic_irqchip(kvm));
1267 return r;
1268}
1269
5bb064dc
ZX		 1270/*
1271 * Get (and clear) the dirty memory log for a memory slot.
1272 */
1273int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1274 struct kvm_dirty_log *log)
1275{
1276 int r;
1277 int n;
1278 struct kvm_memory_slot *memslot;
1279 int is_dirty = 0;
1280
1281 mutex_lock(&kvm->lock);
1282
1283 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1284 if (r)
1285 goto out;
1286
1287 /* If nothing is dirty, don't bother messing with page tables. */
1288 if (is_dirty) {
1289 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1290 kvm_flush_remote_tlbs(kvm);
1291 memslot = &kvm->memslots[log->slot];
1292 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1293 memset(memslot->dirty_bitmap, 0, n);
1294 }
1295 r = 0;
1296out:
1297 mutex_unlock(&kvm->lock);
1298 return r;
1299}
1300
1fe779f8
CO		 1301long kvm_arch_vm_ioctl(struct file *filp,
1302 unsigned int ioctl, unsigned long arg)
1303{
1304 struct kvm *kvm = filp->private_data;
1305 void __user *argp = (void __user *)arg;
1306 int r = -EINVAL;
1307
1308 switch (ioctl) {
1309 case KVM_SET_TSS_ADDR:
1310 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1311 if (r < 0)
1312 goto out;
1313 break;
1314 case KVM_SET_MEMORY_REGION: {
1315 struct kvm_memory_region kvm_mem;
1316 struct kvm_userspace_memory_region kvm_userspace_mem;
1317
1318 r = -EFAULT;
1319 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1320 goto out;
1321 kvm_userspace_mem.slot = kvm_mem.slot;
1322 kvm_userspace_mem.flags = kvm_mem.flags;
1323 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1324 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1325 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1326 if (r)
1327 goto out;
1328 break;
1329 }
1330 case KVM_SET_NR_MMU_PAGES:
1331 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1332 if (r)
1333 goto out;
1334 break;
1335 case KVM_GET_NR_MMU_PAGES:
1336 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1337 break;
1338 case KVM_SET_MEMORY_ALIAS: {
1339 struct kvm_memory_alias alias;
1340
1341 r = -EFAULT;
1342 if (copy_from_user(&alias, argp, sizeof alias))
1343 goto out;
1344 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1345 if (r)
1346 goto out;
1347 break;
1348 }
1349 case KVM_CREATE_IRQCHIP:
1350 r = -ENOMEM;
1351 kvm->vpic = kvm_create_pic(kvm);
1352 if (kvm->vpic) {
1353 r = kvm_ioapic_init(kvm);
1354 if (r) {
1355 kfree(kvm->vpic);
1356 kvm->vpic = NULL;
1357 goto out;
1358 }
1359 } else
1360 goto out;
1361 break;
1362 case KVM_IRQ_LINE: {
1363 struct kvm_irq_level irq_event;
1364
1365 r = -EFAULT;
1366 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1367 goto out;
1368 if (irqchip_in_kernel(kvm)) {
1369 mutex_lock(&kvm->lock);
1370 if (irq_event.irq < 16)
1371 kvm_pic_set_irq(pic_irqchip(kvm),
1372 irq_event.irq,
1373 irq_event.level);
1374 kvm_ioapic_set_irq(kvm->vioapic,
1375 irq_event.irq,
1376 irq_event.level);
1377 mutex_unlock(&kvm->lock);
1378 r = 0;
1379 }
1380 break;
1381 }
1382 case KVM_GET_IRQCHIP: {
1383 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1384 struct kvm_irqchip chip;
1385
1386 r = -EFAULT;
1387 if (copy_from_user(&chip, argp, sizeof chip))
1388 goto out;
1389 r = -ENXIO;
1390 if (!irqchip_in_kernel(kvm))
1391 goto out;
1392 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1393 if (r)
1394 goto out;
1395 r = -EFAULT;
1396 if (copy_to_user(argp, &chip, sizeof chip))
1397 goto out;
1398 r = 0;
1399 break;
1400 }
1401 case KVM_SET_IRQCHIP: {
1402 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1403 struct kvm_irqchip chip;
1404
1405 r = -EFAULT;
1406 if (copy_from_user(&chip, argp, sizeof chip))
1407 goto out;
1408 r = -ENXIO;
1409 if (!irqchip_in_kernel(kvm))
1410 goto out;
1411 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1412 if (r)
1413 goto out;
1414 r = 0;
1415 break;
1416 }
07716717
DK		 1417 case KVM_GET_SUPPORTED_CPUID: {
1418 struct kvm_cpuid2 __user *cpuid_arg = argp;
1419 struct kvm_cpuid2 cpuid;
1420
1421 r = -EFAULT;
1422 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1423 goto out;
1424 r = kvm_vm_ioctl_get_supported_cpuid(kvm, &cpuid,
1425 cpuid_arg->entries);
1426 if (r)
1427 goto out;
1428
1429 r = -EFAULT;
1430 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1431 goto out;
1432 r = 0;
1433 break;
1434 }
1fe779f8
CO		 1435 default:
1436 ;
1437 }
1438out:
1439 return r;
1440}
1441
a16b043c 1442static void kvm_init_msr_list(void)
043405e1
CO		 1443{
1444 u32 dummy[2];
1445 unsigned i, j;
1446
1447 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1448 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1449 continue;
1450 if (j < i)
1451 msrs_to_save[j] = msrs_to_save[i];
1452 j++;
1453 }
1454 num_msrs_to_save = j;
1455}
1456
bbd9b64e
CO		 1457/*
1458 * Only apic need an MMIO device hook, so shortcut now..
1459 */
1460static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1461 gpa_t addr)
1462{
1463 struct kvm_io_device *dev;
1464
1465 if (vcpu->apic) {
1466 dev = &vcpu->apic->dev;
1467 if (dev->in_range(dev, addr))
1468 return dev;
1469 }
1470 return NULL;
1471}
1472
1473
1474static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1475 gpa_t addr)
1476{
1477 struct kvm_io_device *dev;
1478
1479 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1480 if (dev == NULL)
1481 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1482 return dev;
1483}
1484
1485int emulator_read_std(unsigned long addr,
1486 void *val,
1487 unsigned int bytes,
1488 struct kvm_vcpu *vcpu)
1489{
1490 void *data = val;
1491
1492 while (bytes) {
1493 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1494 unsigned offset = addr & (PAGE_SIZE-1);
1495 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1496 int ret;
1497
1498 if (gpa == UNMAPPED_GVA)
1499 return X86EMUL_PROPAGATE_FAULT;
1500 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1501 if (ret < 0)
1502 return X86EMUL_UNHANDLEABLE;
1503
1504 bytes -= tocopy;
1505 data += tocopy;
1506 addr += tocopy;
1507 }
1508
1509 return X86EMUL_CONTINUE;
1510}
1511EXPORT_SYMBOL_GPL(emulator_read_std);
1512
bbd9b64e
CO		 1513static int emulator_read_emulated(unsigned long addr,
1514 void *val,
1515 unsigned int bytes,
1516 struct kvm_vcpu *vcpu)
1517{
1518 struct kvm_io_device *mmio_dev;
1519 gpa_t gpa;
1520
1521 if (vcpu->mmio_read_completed) {
1522 memcpy(val, vcpu->mmio_data, bytes);
1523 vcpu->mmio_read_completed = 0;
1524 return X86EMUL_CONTINUE;
1525 }
1526
1527 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1528
1529 /* For APIC access vmexit */
1530 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1531 goto mmio;
1532
1533 if (emulator_read_std(addr, val, bytes, vcpu)
1534 == X86EMUL_CONTINUE)
1535 return X86EMUL_CONTINUE;
1536 if (gpa == UNMAPPED_GVA)
1537 return X86EMUL_PROPAGATE_FAULT;
1538
1539mmio:
1540 /*
1541 * Is this MMIO handled locally?
1542 */
1543 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1544 if (mmio_dev) {
1545 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1546 return X86EMUL_CONTINUE;
1547 }
1548
1549 vcpu->mmio_needed = 1;
1550 vcpu->mmio_phys_addr = gpa;
1551 vcpu->mmio_size = bytes;
1552 vcpu->mmio_is_write = 0;
1553
1554 return X86EMUL_UNHANDLEABLE;
1555}
1556
1557static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1558 const void *val, int bytes)
1559{
1560 int ret;
1561
1562 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1563 if (ret < 0)
1564 return 0;
1565 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1566 return 1;
1567}
1568
1569static int emulator_write_emulated_onepage(unsigned long addr,
1570 const void *val,
1571 unsigned int bytes,
1572 struct kvm_vcpu *vcpu)
1573{
1574 struct kvm_io_device *mmio_dev;
1575 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1576
1577 if (gpa == UNMAPPED_GVA) {
1578 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1579 return X86EMUL_PROPAGATE_FAULT;
1580 }
1581
1582 /* For APIC access vmexit */
1583 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1584 goto mmio;
1585
1586 if (emulator_write_phys(vcpu, gpa, val, bytes))
1587 return X86EMUL_CONTINUE;
1588
1589mmio:
1590 /*
1591 * Is this MMIO handled locally?
1592 */
1593 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1594 if (mmio_dev) {
1595 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1596 return X86EMUL_CONTINUE;
1597 }
1598
1599 vcpu->mmio_needed = 1;
1600 vcpu->mmio_phys_addr = gpa;
1601 vcpu->mmio_size = bytes;
1602 vcpu->mmio_is_write = 1;
1603 memcpy(vcpu->mmio_data, val, bytes);
1604
1605 return X86EMUL_CONTINUE;
1606}
1607
1608int emulator_write_emulated(unsigned long addr,
1609 const void *val,
1610 unsigned int bytes,
1611 struct kvm_vcpu *vcpu)
1612{
1613 /* Crossing a page boundary? */
1614 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1615 int rc, now;
1616
1617 now = -addr & ~PAGE_MASK;
1618 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1619 if (rc != X86EMUL_CONTINUE)
1620 return rc;
1621 addr += now;
1622 val += now;
1623 bytes -= now;
1624 }
1625 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1626}
1627EXPORT_SYMBOL_GPL(emulator_write_emulated);
1628
1629static int emulator_cmpxchg_emulated(unsigned long addr,
1630 const void *old,
1631 const void *new,
1632 unsigned int bytes,
1633 struct kvm_vcpu *vcpu)
1634{
1635 static int reported;
1636
1637 if (!reported) {
1638 reported = 1;
1639 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1640 }
1641 return emulator_write_emulated(addr, new, bytes, vcpu);
1642}
1643
1644static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1645{
1646 return kvm_x86_ops->get_segment_base(vcpu, seg);
1647}
1648
1649int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1650{
1651 return X86EMUL_CONTINUE;
1652}
1653
1654int emulate_clts(struct kvm_vcpu *vcpu)
1655{
1656 kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1657 return X86EMUL_CONTINUE;
1658}
1659
1660int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1661{
1662 struct kvm_vcpu *vcpu = ctxt->vcpu;
1663
1664 switch (dr) {
1665 case 0 ... 3:
1666 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1667 return X86EMUL_CONTINUE;
1668 default:
1669 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1670 return X86EMUL_UNHANDLEABLE;
1671 }
1672}
1673
1674int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1675{
1676 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1677 int exception;
1678
1679 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1680 if (exception) {
1681 /* FIXME: better handling */
1682 return X86EMUL_UNHANDLEABLE;
1683 }
1684 return X86EMUL_CONTINUE;
1685}
1686
1687void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1688{
1689 static int reported;
1690 u8 opcodes[4];
1691 unsigned long rip = vcpu->rip;
1692 unsigned long rip_linear;
1693
1694 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1695
1696 if (reported)
1697 return;
1698
1699 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1700
1701 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1702 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1703 reported = 1;
1704}
1705EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1706
1707struct x86_emulate_ops emulate_ops = {
1708 .read_std = emulator_read_std,
bbd9b64e
CO		 1709 .read_emulated = emulator_read_emulated,
1710 .write_emulated = emulator_write_emulated,
1711 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1712};
1713
1714int emulate_instruction(struct kvm_vcpu *vcpu,
1715 struct kvm_run *run,
1716 unsigned long cr2,
1717 u16 error_code,
1718 int no_decode)
1719{
1720 int r;
1721
1722 vcpu->mmio_fault_cr2 = cr2;
1723 kvm_x86_ops->cache_regs(vcpu);
1724
1725 vcpu->mmio_is_write = 0;
1726 vcpu->pio.string = 0;
1727
1728 if (!no_decode) {
1729 int cs_db, cs_l;
1730 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1731
1732 vcpu->emulate_ctxt.vcpu = vcpu;
1733 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
bbd9b64e
CO		 1734 vcpu->emulate_ctxt.mode =
1735 (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1736 ? X86EMUL_MODE_REAL : cs_l
1737 ? X86EMUL_MODE_PROT64 : cs_db
1738 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1739
1740 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1741 vcpu->emulate_ctxt.cs_base = 0;
1742 vcpu->emulate_ctxt.ds_base = 0;
1743 vcpu->emulate_ctxt.es_base = 0;
1744 vcpu->emulate_ctxt.ss_base = 0;
1745 } else {
1746 vcpu->emulate_ctxt.cs_base =
1747 get_segment_base(vcpu, VCPU_SREG_CS);
1748 vcpu->emulate_ctxt.ds_base =
1749 get_segment_base(vcpu, VCPU_SREG_DS);
1750 vcpu->emulate_ctxt.es_base =
1751 get_segment_base(vcpu, VCPU_SREG_ES);
1752 vcpu->emulate_ctxt.ss_base =
1753 get_segment_base(vcpu, VCPU_SREG_SS);
1754 }
1755
1756 vcpu->emulate_ctxt.gs_base =
1757 get_segment_base(vcpu, VCPU_SREG_GS);
1758 vcpu->emulate_ctxt.fs_base =
1759 get_segment_base(vcpu, VCPU_SREG_FS);
1760
1761 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
f2b5756b 1762 ++vcpu->stat.insn_emulation;
bbd9b64e 1763 if (r) {
f2b5756b 1764 ++vcpu->stat.insn_emulation_fail;
bbd9b64e
CO		 1765 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1766 return EMULATE_DONE;
1767 return EMULATE_FAIL;
1768 }
1769 }
1770
1771 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1772
1773 if (vcpu->pio.string)
1774 return EMULATE_DO_MMIO;
1775
1776 if ((r || vcpu->mmio_is_write) && run) {
1777 run->exit_reason = KVM_EXIT_MMIO;
1778 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1779 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1780 run->mmio.len = vcpu->mmio_size;
1781 run->mmio.is_write = vcpu->mmio_is_write;
1782 }
1783
1784 if (r) {
1785 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1786 return EMULATE_DONE;
1787 if (!vcpu->mmio_needed) {
1788 kvm_report_emulation_failure(vcpu, "mmio");
1789 return EMULATE_FAIL;
1790 }
1791 return EMULATE_DO_MMIO;
1792 }
1793
1794 kvm_x86_ops->decache_regs(vcpu);
1795 kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1796
1797 if (vcpu->mmio_is_write) {
1798 vcpu->mmio_needed = 0;
1799 return EMULATE_DO_MMIO;
1800 }
1801
1802 return EMULATE_DONE;
1803}
1804EXPORT_SYMBOL_GPL(emulate_instruction);
1805
de7d789a
CO		 1806static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1807{
1808 int i;
1809
1810 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1811 if (vcpu->pio.guest_pages[i]) {
b4231d61 1812 kvm_release_page_dirty(vcpu->pio.guest_pages[i]);
de7d789a
CO		 1813 vcpu->pio.guest_pages[i] = NULL;
1814 }
1815}
1816
1817static int pio_copy_data(struct kvm_vcpu *vcpu)
1818{
1819 void *p = vcpu->pio_data;
1820 void *q;
1821 unsigned bytes;
1822 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1823
1824 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1825 PAGE_KERNEL);
1826 if (!q) {
1827 free_pio_guest_pages(vcpu);
1828 return -ENOMEM;
1829 }
1830 q += vcpu->pio.guest_page_offset;
1831 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1832 if (vcpu->pio.in)
1833 memcpy(q, p, bytes);
1834 else
1835 memcpy(p, q, bytes);
1836 q -= vcpu->pio.guest_page_offset;
1837 vunmap(q);
1838 free_pio_guest_pages(vcpu);
1839 return 0;
1840}
1841
1842int complete_pio(struct kvm_vcpu *vcpu)
1843{
1844 struct kvm_pio_request *io = &vcpu->pio;
1845 long delta;
1846 int r;
1847
1848 kvm_x86_ops->cache_regs(vcpu);
1849
1850 if (!io->string) {
1851 if (io->in)
1852 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1853 io->size);
1854 } else {
1855 if (io->in) {
1856 r = pio_copy_data(vcpu);
1857 if (r) {
1858 kvm_x86_ops->cache_regs(vcpu);
1859 return r;
1860 }
1861 }
1862
1863 delta = 1;
1864 if (io->rep) {
1865 delta *= io->cur_count;
1866 /*
1867 * The size of the register should really depend on
1868 * current address size.
1869 */
1870 vcpu->regs[VCPU_REGS_RCX] -= delta;
1871 }
1872 if (io->down)
1873 delta = -delta;
1874 delta *= io->size;
1875 if (io->in)
1876 vcpu->regs[VCPU_REGS_RDI] += delta;
1877 else
1878 vcpu->regs[VCPU_REGS_RSI] += delta;
1879 }
1880
1881 kvm_x86_ops->decache_regs(vcpu);
1882
1883 io->count -= io->cur_count;
1884 io->cur_count = 0;
1885
1886 return 0;
1887}
1888
1889static void kernel_pio(struct kvm_io_device *pio_dev,
1890 struct kvm_vcpu *vcpu,
1891 void *pd)
1892{
1893 /* TODO: String I/O for in kernel device */
1894
1895 mutex_lock(&vcpu->kvm->lock);
1896 if (vcpu->pio.in)
1897 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1898 vcpu->pio.size,
1899 pd);
1900 else
1901 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1902 vcpu->pio.size,
1903 pd);
1904 mutex_unlock(&vcpu->kvm->lock);
1905}
1906
1907static void pio_string_write(struct kvm_io_device *pio_dev,
1908 struct kvm_vcpu *vcpu)
1909{
1910 struct kvm_pio_request *io = &vcpu->pio;
1911 void *pd = vcpu->pio_data;
1912 int i;
1913
1914 mutex_lock(&vcpu->kvm->lock);
1915 for (i = 0; i < io->cur_count; i++) {
1916 kvm_iodevice_write(pio_dev, io->port,
1917 io->size,
1918 pd);
1919 pd += io->size;
1920 }
1921 mutex_unlock(&vcpu->kvm->lock);
1922}
1923
1924static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1925 gpa_t addr)
1926{
1927 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1928}
1929
1930int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1931 int size, unsigned port)
1932{
1933 struct kvm_io_device *pio_dev;
1934
1935 vcpu->run->exit_reason = KVM_EXIT_IO;
1936 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1937 vcpu->run->io.size = vcpu->pio.size = size;
1938 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1939 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1940 vcpu->run->io.port = vcpu->pio.port = port;
1941 vcpu->pio.in = in;
1942 vcpu->pio.string = 0;
1943 vcpu->pio.down = 0;
1944 vcpu->pio.guest_page_offset = 0;
1945 vcpu->pio.rep = 0;
1946
1947 kvm_x86_ops->cache_regs(vcpu);
1948 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1949 kvm_x86_ops->decache_regs(vcpu);
1950
1951 kvm_x86_ops->skip_emulated_instruction(vcpu);
1952
1953 pio_dev = vcpu_find_pio_dev(vcpu, port);
1954 if (pio_dev) {
1955 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1956 complete_pio(vcpu);
1957 return 1;
1958 }
1959 return 0;
1960}
1961EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1962
1963int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1964 int size, unsigned long count, int down,
1965 gva_t address, int rep, unsigned port)
1966{
1967 unsigned now, in_page;
1968 int i, ret = 0;
1969 int nr_pages = 1;
1970 struct page *page;
1971 struct kvm_io_device *pio_dev;
1972
1973 vcpu->run->exit_reason = KVM_EXIT_IO;
1974 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1975 vcpu->run->io.size = vcpu->pio.size = size;
1976 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1977 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1978 vcpu->run->io.port = vcpu->pio.port = port;
1979 vcpu->pio.in = in;
1980 vcpu->pio.string = 1;
1981 vcpu->pio.down = down;
1982 vcpu->pio.guest_page_offset = offset_in_page(address);
1983 vcpu->pio.rep = rep;
1984
1985 if (!count) {
1986 kvm_x86_ops->skip_emulated_instruction(vcpu);
1987 return 1;
1988 }
1989
1990 if (!down)
1991 in_page = PAGE_SIZE - offset_in_page(address);
1992 else
1993 in_page = offset_in_page(address) + size;
1994 now = min(count, (unsigned long)in_page / size);
1995 if (!now) {
1996 /*
1997 * String I/O straddles page boundary. Pin two guest pages
1998 * so that we satisfy atomicity constraints. Do just one
1999 * transaction to avoid complexity.
2000 */
2001 nr_pages = 2;
2002 now = 1;
2003 }
2004 if (down) {
2005 /*
2006 * String I/O in reverse. Yuck. Kill the guest, fix later.
2007 */
2008 pr_unimpl(vcpu, "guest string pio down\n");
2009 inject_gp(vcpu);
2010 return 1;
2011 }
2012 vcpu->run->io.count = now;
2013 vcpu->pio.cur_count = now;
2014
2015 if (vcpu->pio.cur_count == vcpu->pio.count)
2016 kvm_x86_ops->skip_emulated_instruction(vcpu);
2017
2018 for (i = 0; i < nr_pages; ++i) {
2019 mutex_lock(&vcpu->kvm->lock);
2020 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2021 vcpu->pio.guest_pages[i] = page;
2022 mutex_unlock(&vcpu->kvm->lock);
2023 if (!page) {
2024 inject_gp(vcpu);
2025 free_pio_guest_pages(vcpu);
2026 return 1;
2027 }
2028 }
2029
2030 pio_dev = vcpu_find_pio_dev(vcpu, port);
2031 if (!vcpu->pio.in) {
2032 /* string PIO write */
2033 ret = pio_copy_data(vcpu);
2034 if (ret >= 0 && pio_dev) {
2035 pio_string_write(pio_dev, vcpu);
2036 complete_pio(vcpu);
2037 if (vcpu->pio.count == 0)
2038 ret = 1;
2039 }
2040 } else if (pio_dev)
2041 pr_unimpl(vcpu, "no string pio read support yet, "
2042 "port %x size %d count %ld\n",
2043 port, size, count);
2044
2045 return ret;
2046}
2047EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2048
f8c16bba 2049int kvm_arch_init(void *opaque)
043405e1 2050{
56c6d28a 2051 int r;
f8c16bba
ZX		 2052 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2053
56c6d28a
ZX		 2054 r = kvm_mmu_module_init();
2055 if (r)
2056 goto out_fail;
2057
043405e1 2058 kvm_init_msr_list();
f8c16bba
ZX		 2059
2060 if (kvm_x86_ops) {
2061 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 2062 r = -EEXIST;
2063 goto out;
f8c16bba
ZX		 2064 }
2065
2066 if (!ops->cpu_has_kvm_support()) {
2067 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 2068 r = -EOPNOTSUPP;
2069 goto out;
f8c16bba
ZX		 2070 }
2071 if (ops->disabled_by_bios()) {
2072 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 2073 r = -EOPNOTSUPP;
2074 goto out;
f8c16bba
ZX		 2075 }
2076
2077 kvm_x86_ops = ops;
56c6d28a 2078 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
f8c16bba 2079 return 0;
56c6d28a
ZX		 2080
2081out:
2082 kvm_mmu_module_exit();
2083out_fail:
2084 return r;
043405e1 2085}
8776e519 2086
f8c16bba
ZX		 2087void kvm_arch_exit(void)
2088{
2089 kvm_x86_ops = NULL;
56c6d28a
ZX		 2090 kvm_mmu_module_exit();
2091}
f8c16bba 2092
8776e519
HB		 2093int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2094{
2095 ++vcpu->stat.halt_exits;
2096 if (irqchip_in_kernel(vcpu->kvm)) {
2097 vcpu->mp_state = VCPU_MP_STATE_HALTED;
2098 kvm_vcpu_block(vcpu);
2099 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
2100 return -EINTR;
2101 return 1;
2102 } else {
2103 vcpu->run->exit_reason = KVM_EXIT_HLT;
2104 return 0;
2105 }
2106}
2107EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2108
2109int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2110{
2111 unsigned long nr, a0, a1, a2, a3, ret;
2112
2113 kvm_x86_ops->cache_regs(vcpu);
2114
2115 nr = vcpu->regs[VCPU_REGS_RAX];
2116 a0 = vcpu->regs[VCPU_REGS_RBX];
2117 a1 = vcpu->regs[VCPU_REGS_RCX];
2118 a2 = vcpu->regs[VCPU_REGS_RDX];
2119 a3 = vcpu->regs[VCPU_REGS_RSI];
2120
2121 if (!is_long_mode(vcpu)) {
2122 nr &= 0xFFFFFFFF;
2123 a0 &= 0xFFFFFFFF;
2124 a1 &= 0xFFFFFFFF;
2125 a2 &= 0xFFFFFFFF;
2126 a3 &= 0xFFFFFFFF;
2127 }
2128
2129 switch (nr) {
2130 default:
2131 ret = -KVM_ENOSYS;
2132 break;
2133 }
2134 vcpu->regs[VCPU_REGS_RAX] = ret;
2135 kvm_x86_ops->decache_regs(vcpu);
2136 return 0;
2137}
2138EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2139
2140int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2141{
2142 char instruction[3];
2143 int ret = 0;
2144
2145 mutex_lock(&vcpu->kvm->lock);
2146
2147 /*
2148 * Blow out the MMU to ensure that no other VCPU has an active mapping
2149 * to ensure that the updated hypercall appears atomically across all
2150 * VCPUs.
2151 */
2152 kvm_mmu_zap_all(vcpu->kvm);
2153
2154 kvm_x86_ops->cache_regs(vcpu);
2155 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2156 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
2157 != X86EMUL_CONTINUE)
2158 ret = -EFAULT;
2159
2160 mutex_unlock(&vcpu->kvm->lock);
2161
2162 return ret;
2163}
2164
2165static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2166{
2167 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2168}
2169
2170void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2171{
2172 struct descriptor_table dt = { limit, base };
2173
2174 kvm_x86_ops->set_gdt(vcpu, &dt);
2175}
2176
2177void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2178{
2179 struct descriptor_table dt = { limit, base };
2180
2181 kvm_x86_ops->set_idt(vcpu, &dt);
2182}
2183
2184void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2185 unsigned long *rflags)
2186{
2187 lmsw(vcpu, msw);
2188 *rflags = kvm_x86_ops->get_rflags(vcpu);
2189}
2190
2191unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2192{
2193 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2194 switch (cr) {
2195 case 0:
2196 return vcpu->cr0;
2197 case 2:
2198 return vcpu->cr2;
2199 case 3:
2200 return vcpu->cr3;
2201 case 4:
2202 return vcpu->cr4;
2203 default:
2204 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2205 return 0;
2206 }
2207}
2208
2209void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2210 unsigned long *rflags)
2211{
2212 switch (cr) {
2213 case 0:
2214 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
2215 *rflags = kvm_x86_ops->get_rflags(vcpu);
2216 break;
2217 case 2:
2218 vcpu->cr2 = val;
2219 break;
2220 case 3:
2221 set_cr3(vcpu, val);
2222 break;
2223 case 4:
2224 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
2225 break;
2226 default:
2227 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2228 }
2229}
2230
07716717
DK		 2231static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2232{
2233 struct kvm_cpuid_entry2 *e = &vcpu->cpuid_entries[i];
2234 int j, nent = vcpu->cpuid_nent;
2235
2236 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2237 /* when no next entry is found, the current entry[i] is reselected */
2238 for (j = i + 1; j == i; j = (j + 1) % nent) {
2239 struct kvm_cpuid_entry2 *ej = &vcpu->cpuid_entries[j];
2240 if (ej->function == e->function) {
2241 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2242 return j;
2243 }
2244 }
2245 return 0; /* silence gcc, even though control never reaches here */
2246}
2247
2248/* find an entry with matching function, matching index (if needed), and that
2249 * should be read next (if it's stateful) */
2250static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2251 u32 function, u32 index)
2252{
2253 if (e->function != function)
2254 return 0;
2255 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2256 return 0;
2257 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2258 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2259 return 0;
2260 return 1;
2261}
2262
8776e519
HB		 2263void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2264{
2265 int i;
07716717
DK		 2266 u32 function, index;
2267 struct kvm_cpuid_entry2 *e, *best;
8776e519
HB		 2268
2269 kvm_x86_ops->cache_regs(vcpu);
2270 function = vcpu->regs[VCPU_REGS_RAX];
07716717 2271 index = vcpu->regs[VCPU_REGS_RCX];
8776e519
HB		 2272 vcpu->regs[VCPU_REGS_RAX] = 0;
2273 vcpu->regs[VCPU_REGS_RBX] = 0;
2274 vcpu->regs[VCPU_REGS_RCX] = 0;
2275 vcpu->regs[VCPU_REGS_RDX] = 0;
2276 best = NULL;
2277 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2278 e = &vcpu->cpuid_entries[i];
07716717
DK		 2279 if (is_matching_cpuid_entry(e, function, index)) {
2280 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2281 move_to_next_stateful_cpuid_entry(vcpu, i);
8776e519
HB		 2282 best = e;
2283 break;
2284 }
2285 /*
2286 * Both basic or both extended?
2287 */
2288 if (((e->function ^ function) & 0x80000000) == 0)
2289 if (!best || e->function > best->function)
2290 best = e;
2291 }
2292 if (best) {
2293 vcpu->regs[VCPU_REGS_RAX] = best->eax;
2294 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
2295 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
2296 vcpu->regs[VCPU_REGS_RDX] = best->edx;
2297 }
2298 kvm_x86_ops->decache_regs(vcpu);
2299 kvm_x86_ops->skip_emulated_instruction(vcpu);
2300}
2301EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
d0752060 2302
b6c7a5dc
HB		 2303/*
2304 * Check if userspace requested an interrupt window, and that the
2305 * interrupt window is open.
2306 *
2307 * No need to exit to userspace if we already have an interrupt queued.
2308 */
2309static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2310 struct kvm_run *kvm_run)
2311{
2312 return (!vcpu->irq_summary &&
2313 kvm_run->request_interrupt_window &&
2314 vcpu->interrupt_window_open &&
2315 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2316}
2317
2318static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2319 struct kvm_run *kvm_run)
2320{
2321 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2322 kvm_run->cr8 = get_cr8(vcpu);
2323 kvm_run->apic_base = kvm_get_apic_base(vcpu);
2324 if (irqchip_in_kernel(vcpu->kvm))
2325 kvm_run->ready_for_interrupt_injection = 1;
2326 else
2327 kvm_run->ready_for_interrupt_injection =
2328 (vcpu->interrupt_window_open &&
2329 vcpu->irq_summary == 0);
2330}
2331
2332static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2333{
2334 int r;
2335
2336 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2337 pr_debug("vcpu %d received sipi with vector # %x\n",
2338 vcpu->vcpu_id, vcpu->sipi_vector);
2339 kvm_lapic_reset(vcpu);
2340 r = kvm_x86_ops->vcpu_reset(vcpu);
2341 if (r)
2342 return r;
2343 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2344 }
2345
2346preempted:
2347 if (vcpu->guest_debug.enabled)
2348 kvm_x86_ops->guest_debug_pre(vcpu);
2349
2350again:
2351 r = kvm_mmu_reload(vcpu);
2352 if (unlikely(r))
2353 goto out;
2354
2355 kvm_inject_pending_timer_irqs(vcpu);
2356
2357 preempt_disable();
2358
2359 kvm_x86_ops->prepare_guest_switch(vcpu);
2360 kvm_load_guest_fpu(vcpu);
2361
2362 local_irq_disable();
2363
2364 if (signal_pending(current)) {
2365 local_irq_enable();
2366 preempt_enable();
2367 r = -EINTR;
2368 kvm_run->exit_reason = KVM_EXIT_INTR;
2369 ++vcpu->stat.signal_exits;
2370 goto out;
2371 }
2372
2373 if (irqchip_in_kernel(vcpu->kvm))
2374 kvm_x86_ops->inject_pending_irq(vcpu);
eb9774f0 2375 else
b6c7a5dc
HB		 2376 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2377
2378 vcpu->guest_mode = 1;
2379 kvm_guest_enter();
2380
2381 if (vcpu->requests)
2382 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2383 kvm_x86_ops->tlb_flush(vcpu);
2384
2385 kvm_x86_ops->run(vcpu, kvm_run);
2386
2387 vcpu->guest_mode = 0;
2388 local_irq_enable();
2389
2390 ++vcpu->stat.exits;
2391
2392 /*
2393 * We must have an instruction between local_irq_enable() and
2394 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2395 * the interrupt shadow. The stat.exits increment will do nicely.
2396 * But we need to prevent reordering, hence this barrier():
2397 */
2398 barrier();
2399
2400 kvm_guest_exit();
2401
2402 preempt_enable();
2403
2404 /*
2405 * Profile KVM exit RIPs:
2406 */
2407 if (unlikely(prof_on == KVM_PROFILING)) {
2408 kvm_x86_ops->cache_regs(vcpu);
2409 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2410 }
2411
2412 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2413
2414 if (r > 0) {
2415 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2416 r = -EINTR;
2417 kvm_run->exit_reason = KVM_EXIT_INTR;
2418 ++vcpu->stat.request_irq_exits;
2419 goto out;
2420 }
e1beb1d3 2421 if (!need_resched())
b6c7a5dc 2422 goto again;
b6c7a5dc
HB		 2423 }
2424
2425out:
2426 if (r > 0) {
2427 kvm_resched(vcpu);
2428 goto preempted;
2429 }
2430
2431 post_kvm_run_save(vcpu, kvm_run);
2432
2433 return r;
2434}
2435
2436int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2437{
2438 int r;
2439 sigset_t sigsaved;
2440
2441 vcpu_load(vcpu);
2442
2443 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2444 kvm_vcpu_block(vcpu);
2445 vcpu_put(vcpu);
2446 return -EAGAIN;
2447 }
2448
2449 if (vcpu->sigset_active)
2450 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2451
2452 /* re-sync apic's tpr */
2453 if (!irqchip_in_kernel(vcpu->kvm))
2454 set_cr8(vcpu, kvm_run->cr8);
2455
2456 if (vcpu->pio.cur_count) {
2457 r = complete_pio(vcpu);
2458 if (r)
2459 goto out;
2460 }
2461#if CONFIG_HAS_IOMEM
2462 if (vcpu->mmio_needed) {
2463 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2464 vcpu->mmio_read_completed = 1;
2465 vcpu->mmio_needed = 0;
2466 r = emulate_instruction(vcpu, kvm_run,
2467 vcpu->mmio_fault_cr2, 0, 1);
2468 if (r == EMULATE_DO_MMIO) {
2469 /*
2470 * Read-modify-write. Back to userspace.
2471 */
2472 r = 0;
2473 goto out;
2474 }
2475 }
2476#endif
2477 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2478 kvm_x86_ops->cache_regs(vcpu);
2479 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2480 kvm_x86_ops->decache_regs(vcpu);
2481 }
2482
2483 r = __vcpu_run(vcpu, kvm_run);
2484
2485out:
2486 if (vcpu->sigset_active)
2487 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2488
2489 vcpu_put(vcpu);
2490 return r;
2491}
2492
2493int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2494{
2495 vcpu_load(vcpu);
2496
2497 kvm_x86_ops->cache_regs(vcpu);
2498
2499 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2500 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2501 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2502 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2503 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2504 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2505 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2506 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2507#ifdef CONFIG_X86_64
2508 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2509 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2510 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2511 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2512 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2513 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2514 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2515 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2516#endif
2517
2518 regs->rip = vcpu->rip;
2519 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2520
2521 /*
2522 * Don't leak debug flags in case they were set for guest debugging
2523 */
2524 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2525 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2526
2527 vcpu_put(vcpu);
2528
2529 return 0;
2530}
2531
2532int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2533{
2534 vcpu_load(vcpu);
2535
2536 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2537 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2538 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2539 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2540 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2541 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2542 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2543 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2544#ifdef CONFIG_X86_64
2545 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2546 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2547 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2548 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2549 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2550 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2551 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2552 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2553#endif
2554
2555 vcpu->rip = regs->rip;
2556 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2557
2558 kvm_x86_ops->decache_regs(vcpu);
2559
2560 vcpu_put(vcpu);
2561
2562 return 0;
2563}
2564
2565static void get_segment(struct kvm_vcpu *vcpu,
2566 struct kvm_segment *var, int seg)
2567{
2568 return kvm_x86_ops->get_segment(vcpu, var, seg);
2569}
2570
2571void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2572{
2573 struct kvm_segment cs;
2574
2575 get_segment(vcpu, &cs, VCPU_SREG_CS);
2576 *db = cs.db;
2577 *l = cs.l;
2578}
2579EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2580
2581int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2582 struct kvm_sregs *sregs)
2583{
2584 struct descriptor_table dt;
2585 int pending_vec;
2586
2587 vcpu_load(vcpu);
2588
2589 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2590 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2591 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2592 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2593 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2594 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2595
2596 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2597 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2598
2599 kvm_x86_ops->get_idt(vcpu, &dt);
2600 sregs->idt.limit = dt.limit;
2601 sregs->idt.base = dt.base;
2602 kvm_x86_ops->get_gdt(vcpu, &dt);
2603 sregs->gdt.limit = dt.limit;
2604 sregs->gdt.base = dt.base;
2605
2606 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2607 sregs->cr0 = vcpu->cr0;
2608 sregs->cr2 = vcpu->cr2;
2609 sregs->cr3 = vcpu->cr3;
2610 sregs->cr4 = vcpu->cr4;
2611 sregs->cr8 = get_cr8(vcpu);
2612 sregs->efer = vcpu->shadow_efer;
2613 sregs->apic_base = kvm_get_apic_base(vcpu);
2614
2615 if (irqchip_in_kernel(vcpu->kvm)) {
2616 memset(sregs->interrupt_bitmap, 0,
2617 sizeof sregs->interrupt_bitmap);
2618 pending_vec = kvm_x86_ops->get_irq(vcpu);
2619 if (pending_vec >= 0)
2620 set_bit(pending_vec,
2621 (unsigned long *)sregs->interrupt_bitmap);
2622 } else
2623 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2624 sizeof sregs->interrupt_bitmap);
2625
2626 vcpu_put(vcpu);
2627
2628 return 0;
2629}
2630
2631static void set_segment(struct kvm_vcpu *vcpu,
2632 struct kvm_segment *var, int seg)
2633{
2634 return kvm_x86_ops->set_segment(vcpu, var, seg);
2635}
2636
2637int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2638 struct kvm_sregs *sregs)
2639{
2640 int mmu_reset_needed = 0;
2641 int i, pending_vec, max_bits;
2642 struct descriptor_table dt;
2643
2644 vcpu_load(vcpu);
2645
2646 dt.limit = sregs->idt.limit;
2647 dt.base = sregs->idt.base;
2648 kvm_x86_ops->set_idt(vcpu, &dt);
2649 dt.limit = sregs->gdt.limit;
2650 dt.base = sregs->gdt.base;
2651 kvm_x86_ops->set_gdt(vcpu, &dt);
2652
2653 vcpu->cr2 = sregs->cr2;
2654 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2655 vcpu->cr3 = sregs->cr3;
2656
2657 set_cr8(vcpu, sregs->cr8);
2658
2659 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2660#ifdef CONFIG_X86_64
2661 kvm_x86_ops->set_efer(vcpu, sregs->efer);
2662#endif
2663 kvm_set_apic_base(vcpu, sregs->apic_base);
2664
2665 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2666
2667 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2668 vcpu->cr0 = sregs->cr0;
2669 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2670
2671 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2672 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2673 if (!is_long_mode(vcpu) && is_pae(vcpu))
2674 load_pdptrs(vcpu, vcpu->cr3);
2675
2676 if (mmu_reset_needed)
2677 kvm_mmu_reset_context(vcpu);
2678
2679 if (!irqchip_in_kernel(vcpu->kvm)) {
2680 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2681 sizeof vcpu->irq_pending);
2682 vcpu->irq_summary = 0;
2683 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2684 if (vcpu->irq_pending[i])
2685 __set_bit(i, &vcpu->irq_summary);
2686 } else {
2687 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2688 pending_vec = find_first_bit(
2689 (const unsigned long *)sregs->interrupt_bitmap,
2690 max_bits);
2691 /* Only pending external irq is handled here */
2692 if (pending_vec < max_bits) {
2693 kvm_x86_ops->set_irq(vcpu, pending_vec);
2694 pr_debug("Set back pending irq %d\n",
2695 pending_vec);
2696 }
2697 }
2698
2699 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2700 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2701 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2702 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2703 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2704 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2705
2706 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2707 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2708
2709 vcpu_put(vcpu);
2710
2711 return 0;
2712}
2713
2714int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2715 struct kvm_debug_guest *dbg)
2716{
2717 int r;
2718
2719 vcpu_load(vcpu);
2720
2721 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2722
2723 vcpu_put(vcpu);
2724
2725 return r;
2726}
2727
d0752060
HB		 2728/*
2729 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2730 * we have asm/x86/processor.h
2731 */
2732struct fxsave {
2733 u16 cwd;
2734 u16 swd;
2735 u16 twd;
2736 u16 fop;
2737 u64 rip;
2738 u64 rdp;
2739 u32 mxcsr;
2740 u32 mxcsr_mask;
2741 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2742#ifdef CONFIG_X86_64
2743 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2744#else
2745 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2746#endif
2747};
2748
8b006791
ZX		 2749/*
2750 * Translate a guest virtual address to a guest physical address.
2751 */
2752int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2753 struct kvm_translation *tr)
2754{
2755 unsigned long vaddr = tr->linear_address;
2756 gpa_t gpa;
2757
2758 vcpu_load(vcpu);
2759 mutex_lock(&vcpu->kvm->lock);
2760 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2761 tr->physical_address = gpa;
2762 tr->valid = gpa != UNMAPPED_GVA;
2763 tr->writeable = 1;
2764 tr->usermode = 0;
2765 mutex_unlock(&vcpu->kvm->lock);
2766 vcpu_put(vcpu);
2767
2768 return 0;
2769}
2770
d0752060
HB		 2771int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2772{
2773 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2774
2775 vcpu_load(vcpu);
2776
2777 memcpy(fpu->fpr, fxsave->st_space, 128);
2778 fpu->fcw = fxsave->cwd;
2779 fpu->fsw = fxsave->swd;
2780 fpu->ftwx = fxsave->twd;
2781 fpu->last_opcode = fxsave->fop;
2782 fpu->last_ip = fxsave->rip;
2783 fpu->last_dp = fxsave->rdp;
2784 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2785
2786 vcpu_put(vcpu);
2787
2788 return 0;
2789}
2790
2791int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2792{
2793 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2794
2795 vcpu_load(vcpu);
2796
2797 memcpy(fxsave->st_space, fpu->fpr, 128);
2798 fxsave->cwd = fpu->fcw;
2799 fxsave->swd = fpu->fsw;
2800 fxsave->twd = fpu->ftwx;
2801 fxsave->fop = fpu->last_opcode;
2802 fxsave->rip = fpu->last_ip;
2803 fxsave->rdp = fpu->last_dp;
2804 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2805
2806 vcpu_put(vcpu);
2807
2808 return 0;
2809}
2810
2811void fx_init(struct kvm_vcpu *vcpu)
2812{
2813 unsigned after_mxcsr_mask;
2814
2815 /* Initialize guest FPU by resetting ours and saving into guest's */
2816 preempt_disable();
2817 fx_save(&vcpu->host_fx_image);
2818 fpu_init();
2819 fx_save(&vcpu->guest_fx_image);
2820 fx_restore(&vcpu->host_fx_image);
2821 preempt_enable();
2822
2823 vcpu->cr0 |= X86_CR0_ET;
2824 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2825 vcpu->guest_fx_image.mxcsr = 0x1f80;
2826 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2827 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2828}
2829EXPORT_SYMBOL_GPL(fx_init);
2830
2831void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2832{
2833 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2834 return;
2835
2836 vcpu->guest_fpu_loaded = 1;
2837 fx_save(&vcpu->host_fx_image);
2838 fx_restore(&vcpu->guest_fx_image);
2839}
2840EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2841
2842void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2843{
2844 if (!vcpu->guest_fpu_loaded)
2845 return;
2846
2847 vcpu->guest_fpu_loaded = 0;
2848 fx_save(&vcpu->guest_fx_image);
2849 fx_restore(&vcpu->host_fx_image);
f096ed85 2850 ++vcpu->stat.fpu_reload;
d0752060
HB		 2851}
2852EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
e9b11c17
ZX		 2853
2854void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
2855{
2856 kvm_x86_ops->vcpu_free(vcpu);
2857}
2858
2859struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2860 unsigned int id)
2861{
26e5215f
AK		 2862 return kvm_x86_ops->vcpu_create(kvm, id);
2863}
e9b11c17 2864
26e5215f
AK		 2865int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
2866{
2867 int r;
e9b11c17
ZX		 2868
2869 /* We do fxsave: this must be aligned. */
2870 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2871
2872 vcpu_load(vcpu);
2873 r = kvm_arch_vcpu_reset(vcpu);
2874 if (r == 0)
2875 r = kvm_mmu_setup(vcpu);
2876 vcpu_put(vcpu);
2877 if (r < 0)
2878 goto free_vcpu;
2879
26e5215f 2880 return 0;
e9b11c17
ZX		 2881free_vcpu:
2882 kvm_x86_ops->vcpu_free(vcpu);
26e5215f 2883 return r;
e9b11c17
ZX		 2884}
2885
d40ccc62 2886void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17
ZX		 2887{
2888 vcpu_load(vcpu);
2889 kvm_mmu_unload(vcpu);
2890 vcpu_put(vcpu);
2891
2892 kvm_x86_ops->vcpu_free(vcpu);
2893}
2894
2895int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
2896{
2897 return kvm_x86_ops->vcpu_reset(vcpu);
2898}
2899
2900void kvm_arch_hardware_enable(void *garbage)
2901{
2902 kvm_x86_ops->hardware_enable(garbage);
2903}
2904
2905void kvm_arch_hardware_disable(void *garbage)
2906{
2907 kvm_x86_ops->hardware_disable(garbage);
2908}
2909
2910int kvm_arch_hardware_setup(void)
2911{
2912 return kvm_x86_ops->hardware_setup();
2913}
2914
2915void kvm_arch_hardware_unsetup(void)
2916{
2917 kvm_x86_ops->hardware_unsetup();
2918}
2919
2920void kvm_arch_check_processor_compat(void *rtn)
2921{
2922 kvm_x86_ops->check_processor_compatibility(rtn);
2923}
2924
2925int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2926{
2927 struct page *page;
2928 struct kvm *kvm;
2929 int r;
2930
2931 BUG_ON(vcpu->kvm == NULL);
2932 kvm = vcpu->kvm;
2933
2934 vcpu->mmu.root_hpa = INVALID_PAGE;
2935 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
2936 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2937 else
2938 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
2939
2940 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2941 if (!page) {
2942 r = -ENOMEM;
2943 goto fail;
2944 }
2945 vcpu->pio_data = page_address(page);
2946
2947 r = kvm_mmu_create(vcpu);
2948 if (r < 0)
2949 goto fail_free_pio_data;
2950
2951 if (irqchip_in_kernel(kvm)) {
2952 r = kvm_create_lapic(vcpu);
2953 if (r < 0)
2954 goto fail_mmu_destroy;
2955 }
2956
2957 return 0;
2958
2959fail_mmu_destroy:
2960 kvm_mmu_destroy(vcpu);
2961fail_free_pio_data:
2962 free_page((unsigned long)vcpu->pio_data);
2963fail:
2964 return r;
2965}
2966
2967void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
2968{
2969 kvm_free_lapic(vcpu);
2970 kvm_mmu_destroy(vcpu);
2971 free_page((unsigned long)vcpu->pio_data);
2972}
d19a9cd2
ZX		 2973
2974struct kvm *kvm_arch_create_vm(void)
2975{
2976 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
2977
2978 if (!kvm)
2979 return ERR_PTR(-ENOMEM);
2980
2981 INIT_LIST_HEAD(&kvm->active_mmu_pages);
2982
2983 return kvm;
2984}
2985
2986static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
2987{
2988 vcpu_load(vcpu);
2989 kvm_mmu_unload(vcpu);
2990 vcpu_put(vcpu);
2991}
2992
2993static void kvm_free_vcpus(struct kvm *kvm)
2994{
2995 unsigned int i;
2996
2997 /*
2998 * Unpin any mmu pages first.
2999 */
3000 for (i = 0; i < KVM_MAX_VCPUS; ++i)
3001 if (kvm->vcpus[i])
3002 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3003 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3004 if (kvm->vcpus[i]) {
3005 kvm_arch_vcpu_free(kvm->vcpus[i]);
3006 kvm->vcpus[i] = NULL;
3007 }
3008 }
3009
3010}
3011
3012void kvm_arch_destroy_vm(struct kvm *kvm)
3013{
3014 kfree(kvm->vpic);
3015 kfree(kvm->vioapic);
3016 kvm_free_vcpus(kvm);
3017 kvm_free_physmem(kvm);
3018 kfree(kvm);
3019}
0de10343
ZX		 3020
3021int kvm_arch_set_memory_region(struct kvm *kvm,
3022 struct kvm_userspace_memory_region *mem,
3023 struct kvm_memory_slot old,
3024 int user_alloc)
3025{
3026 int npages = mem->memory_size >> PAGE_SHIFT;
3027 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3028
3029 /*To keep backward compatibility with older userspace,
3030 *x86 needs to hanlde !user_alloc case.
3031 */
3032 if (!user_alloc) {
3033 if (npages && !old.rmap) {
3034 down_write(&current->mm->mmap_sem);
3035 memslot->userspace_addr = do_mmap(NULL, 0,
3036 npages * PAGE_SIZE,
3037 PROT_READ | PROT_WRITE,
3038 MAP_SHARED | MAP_ANONYMOUS,
3039 0);
3040 up_write(&current->mm->mmap_sem);
3041
3042 if (IS_ERR((void *)memslot->userspace_addr))
3043 return PTR_ERR((void *)memslot->userspace_addr);
3044 } else {
3045 if (!old.user_alloc && old.rmap) {
3046 int ret;
3047
3048 down_write(&current->mm->mmap_sem);
3049 ret = do_munmap(current->mm, old.userspace_addr,
3050 old.npages * PAGE_SIZE);
3051 up_write(&current->mm->mmap_sem);
3052 if (ret < 0)
3053 printk(KERN_WARNING
3054 "kvm_vm_ioctl_set_memory_region: "
3055 "failed to munmap memory\n");
3056 }
3057 }
3058 }
3059
3060 if (!kvm->n_requested_mmu_pages) {
3061 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3062 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3063 }
3064
3065 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3066 kvm_flush_remote_tlbs(kvm);
3067
3068 return 0;
3069}
Linux 6.x block layer and io_uring tree(s)