]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/kvm/x86.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: Remove ptr comparisons to 0
[net-next-2.6.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>
043405e1
CO		 27
28#include <asm/uaccess.h>
d825ed0a 29#include <asm/msr.h>
043405e1 30
313a3dc7 31#define MAX_IO_MSRS 256
a03490ed
CO		 32#define CR0_RESERVED_BITS \
33 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
34 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
35 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
36#define CR4_RESERVED_BITS \
37 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
38 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
39 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
40 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
41
42#define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
15c4a640 43#define EFER_RESERVED_BITS 0xfffffffffffff2fe
313a3dc7 44
417bc304
HB		 45#define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
46
97896d04
ZX		 47struct kvm_x86_ops *kvm_x86_ops;
48
417bc304
HB		 49struct kvm_stats_debugfs_item debugfs_entries[] = {
50 { "pf_fixed", STAT_OFFSET(pf_fixed) },
51 { "pf_guest", STAT_OFFSET(pf_guest) },
52 { "tlb_flush", STAT_OFFSET(tlb_flush) },
53 { "invlpg", STAT_OFFSET(invlpg) },
54 { "exits", STAT_OFFSET(exits) },
55 { "io_exits", STAT_OFFSET(io_exits) },
56 { "mmio_exits", STAT_OFFSET(mmio_exits) },
57 { "signal_exits", STAT_OFFSET(signal_exits) },
58 { "irq_window", STAT_OFFSET(irq_window_exits) },
59 { "halt_exits", STAT_OFFSET(halt_exits) },
60 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
61 { "request_irq", STAT_OFFSET(request_irq_exits) },
62 { "irq_exits", STAT_OFFSET(irq_exits) },
63 { "light_exits", STAT_OFFSET(light_exits) },
64 { "efer_reload", STAT_OFFSET(efer_reload) },
65 { NULL }
66};
67
68
5fb76f9b
CO		 69unsigned long segment_base(u16 selector)
70{
71 struct descriptor_table gdt;
72 struct segment_descriptor *d;
73 unsigned long table_base;
74 unsigned long v;
75
76 if (selector == 0)
77 return 0;
78
79 asm("sgdt %0" : "=m"(gdt));
80 table_base = gdt.base;
81
82 if (selector & 4) { /* from ldt */
83 u16 ldt_selector;
84
85 asm("sldt %0" : "=g"(ldt_selector));
86 table_base = segment_base(ldt_selector);
87 }
88 d = (struct segment_descriptor *)(table_base + (selector & ~7));
89 v = d->base_low | ((unsigned long)d->base_mid << 16) |
90 ((unsigned long)d->base_high << 24);
91#ifdef CONFIG_X86_64
92 if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
93 v |= ((unsigned long) \
94 ((struct segment_descriptor_64 *)d)->base_higher) << 32;
95#endif
96 return v;
97}
98EXPORT_SYMBOL_GPL(segment_base);
99
6866b83e
CO		 100u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
101{
102 if (irqchip_in_kernel(vcpu->kvm))
103 return vcpu->apic_base;
104 else
105 return vcpu->apic_base;
106}
107EXPORT_SYMBOL_GPL(kvm_get_apic_base);
108
109void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
110{
111 /* TODO: reserve bits check */
112 if (irqchip_in_kernel(vcpu->kvm))
113 kvm_lapic_set_base(vcpu, data);
114 else
115 vcpu->apic_base = data;
116}
117EXPORT_SYMBOL_GPL(kvm_set_apic_base);
118
a03490ed
CO		 119static void inject_gp(struct kvm_vcpu *vcpu)
120{
121 kvm_x86_ops->inject_gp(vcpu, 0);
122}
123
124/*
125 * Load the pae pdptrs. Return true is they are all valid.
126 */
127int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
128{
129 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
130 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
131 int i;
132 int ret;
133 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
134
135 mutex_lock(&vcpu->kvm->lock);
136 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
137 offset * sizeof(u64), sizeof(pdpte));
138 if (ret < 0) {
139 ret = 0;
140 goto out;
141 }
142 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
143 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
144 ret = 0;
145 goto out;
146 }
147 }
148 ret = 1;
149
150 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
151out:
152 mutex_unlock(&vcpu->kvm->lock);
153
154 return ret;
155}
156
157void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
158{
159 if (cr0 & CR0_RESERVED_BITS) {
160 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
161 cr0, vcpu->cr0);
162 inject_gp(vcpu);
163 return;
164 }
165
166 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
167 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
168 inject_gp(vcpu);
169 return;
170 }
171
172 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
173 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
174 "and a clear PE flag\n");
175 inject_gp(vcpu);
176 return;
177 }
178
179 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
180#ifdef CONFIG_X86_64
181 if ((vcpu->shadow_efer & EFER_LME)) {
182 int cs_db, cs_l;
183
184 if (!is_pae(vcpu)) {
185 printk(KERN_DEBUG "set_cr0: #GP, start paging "
186 "in long mode while PAE is disabled\n");
187 inject_gp(vcpu);
188 return;
189 }
190 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
191 if (cs_l) {
192 printk(KERN_DEBUG "set_cr0: #GP, start paging "
193 "in long mode while CS.L == 1\n");
194 inject_gp(vcpu);
195 return;
196
197 }
198 } else
199#endif
200 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
201 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
202 "reserved bits\n");
203 inject_gp(vcpu);
204 return;
205 }
206
207 }
208
209 kvm_x86_ops->set_cr0(vcpu, cr0);
210 vcpu->cr0 = cr0;
211
212 mutex_lock(&vcpu->kvm->lock);
213 kvm_mmu_reset_context(vcpu);
214 mutex_unlock(&vcpu->kvm->lock);
215 return;
216}
217EXPORT_SYMBOL_GPL(set_cr0);
218
219void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
220{
221 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
222}
223EXPORT_SYMBOL_GPL(lmsw);
224
225void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
226{
227 if (cr4 & CR4_RESERVED_BITS) {
228 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
229 inject_gp(vcpu);
230 return;
231 }
232
233 if (is_long_mode(vcpu)) {
234 if (!(cr4 & X86_CR4_PAE)) {
235 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
236 "in long mode\n");
237 inject_gp(vcpu);
238 return;
239 }
240 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
241 && !load_pdptrs(vcpu, vcpu->cr3)) {
242 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
243 inject_gp(vcpu);
244 return;
245 }
246
247 if (cr4 & X86_CR4_VMXE) {
248 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
249 inject_gp(vcpu);
250 return;
251 }
252 kvm_x86_ops->set_cr4(vcpu, cr4);
253 vcpu->cr4 = cr4;
254 mutex_lock(&vcpu->kvm->lock);
255 kvm_mmu_reset_context(vcpu);
256 mutex_unlock(&vcpu->kvm->lock);
257}
258EXPORT_SYMBOL_GPL(set_cr4);
259
260void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
261{
262 if (is_long_mode(vcpu)) {
263 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
264 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
265 inject_gp(vcpu);
266 return;
267 }
268 } else {
269 if (is_pae(vcpu)) {
270 if (cr3 & CR3_PAE_RESERVED_BITS) {
271 printk(KERN_DEBUG
272 "set_cr3: #GP, reserved bits\n");
273 inject_gp(vcpu);
274 return;
275 }
276 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
277 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
278 "reserved bits\n");
279 inject_gp(vcpu);
280 return;
281 }
282 }
283 /*
284 * We don't check reserved bits in nonpae mode, because
285 * this isn't enforced, and VMware depends on this.
286 */
287 }
288
289 mutex_lock(&vcpu->kvm->lock);
290 /*
291 * Does the new cr3 value map to physical memory? (Note, we
292 * catch an invalid cr3 even in real-mode, because it would
293 * cause trouble later on when we turn on paging anyway.)
294 *
295 * A real CPU would silently accept an invalid cr3 and would
296 * attempt to use it - with largely undefined (and often hard
297 * to debug) behavior on the guest side.
298 */
299 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
300 inject_gp(vcpu);
301 else {
302 vcpu->cr3 = cr3;
303 vcpu->mmu.new_cr3(vcpu);
304 }
305 mutex_unlock(&vcpu->kvm->lock);
306}
307EXPORT_SYMBOL_GPL(set_cr3);
308
309void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
310{
311 if (cr8 & CR8_RESERVED_BITS) {
312 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
313 inject_gp(vcpu);
314 return;
315 }
316 if (irqchip_in_kernel(vcpu->kvm))
317 kvm_lapic_set_tpr(vcpu, cr8);
318 else
319 vcpu->cr8 = cr8;
320}
321EXPORT_SYMBOL_GPL(set_cr8);
322
323unsigned long get_cr8(struct kvm_vcpu *vcpu)
324{
325 if (irqchip_in_kernel(vcpu->kvm))
326 return kvm_lapic_get_cr8(vcpu);
327 else
328 return vcpu->cr8;
329}
330EXPORT_SYMBOL_GPL(get_cr8);
331
043405e1
CO		 332/*
333 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
334 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
335 *
336 * This list is modified at module load time to reflect the
337 * capabilities of the host cpu.
338 */
339static u32 msrs_to_save[] = {
340 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
341 MSR_K6_STAR,
342#ifdef CONFIG_X86_64
343 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
344#endif
345 MSR_IA32_TIME_STAMP_COUNTER,
346};
347
348static unsigned num_msrs_to_save;
349
350static u32 emulated_msrs[] = {
351 MSR_IA32_MISC_ENABLE,
352};
353
15c4a640
CO		 354#ifdef CONFIG_X86_64
355
356static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
357{
358 if (efer & EFER_RESERVED_BITS) {
359 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
360 efer);
361 inject_gp(vcpu);
362 return;
363 }
364
365 if (is_paging(vcpu)
366 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
367 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
368 inject_gp(vcpu);
369 return;
370 }
371
372 kvm_x86_ops->set_efer(vcpu, efer);
373
374 efer &= ~EFER_LMA;
375 efer |= vcpu->shadow_efer & EFER_LMA;
376
377 vcpu->shadow_efer = efer;
378}
379
380#endif
381
382/*
383 * Writes msr value into into the appropriate "register".
384 * Returns 0 on success, non-0 otherwise.
385 * Assumes vcpu_load() was already called.
386 */
387int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
388{
389 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
390}
391
313a3dc7
CO		 392/*
393 * Adapt set_msr() to msr_io()'s calling convention
394 */
395static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
396{
397 return kvm_set_msr(vcpu, index, *data);
398}
399
15c4a640
CO		 400
401int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
402{
403 switch (msr) {
404#ifdef CONFIG_X86_64
405 case MSR_EFER:
406 set_efer(vcpu, data);
407 break;
408#endif
409 case MSR_IA32_MC0_STATUS:
410 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
411 __FUNCTION__, data);
412 break;
413 case MSR_IA32_MCG_STATUS:
414 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
415 __FUNCTION__, data);
416 break;
417 case MSR_IA32_UCODE_REV:
418 case MSR_IA32_UCODE_WRITE:
419 case 0x200 ... 0x2ff: /* MTRRs */
420 break;
421 case MSR_IA32_APICBASE:
422 kvm_set_apic_base(vcpu, data);
423 break;
424 case MSR_IA32_MISC_ENABLE:
425 vcpu->ia32_misc_enable_msr = data;
426 break;
427 default:
428 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
429 return 1;
430 }
431 return 0;
432}
433EXPORT_SYMBOL_GPL(kvm_set_msr_common);
434
435
436/*
437 * Reads an msr value (of 'msr_index') into 'pdata'.
438 * Returns 0 on success, non-0 otherwise.
439 * Assumes vcpu_load() was already called.
440 */
441int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
442{
443 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
444}
445
446int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
447{
448 u64 data;
449
450 switch (msr) {
451 case 0xc0010010: /* SYSCFG */
452 case 0xc0010015: /* HWCR */
453 case MSR_IA32_PLATFORM_ID:
454 case MSR_IA32_P5_MC_ADDR:
455 case MSR_IA32_P5_MC_TYPE:
456 case MSR_IA32_MC0_CTL:
457 case MSR_IA32_MCG_STATUS:
458 case MSR_IA32_MCG_CAP:
459 case MSR_IA32_MC0_MISC:
460 case MSR_IA32_MC0_MISC+4:
461 case MSR_IA32_MC0_MISC+8:
462 case MSR_IA32_MC0_MISC+12:
463 case MSR_IA32_MC0_MISC+16:
464 case MSR_IA32_UCODE_REV:
465 case MSR_IA32_PERF_STATUS:
466 case MSR_IA32_EBL_CR_POWERON:
467 /* MTRR registers */
468 case 0xfe:
469 case 0x200 ... 0x2ff:
470 data = 0;
471 break;
472 case 0xcd: /* fsb frequency */
473 data = 3;
474 break;
475 case MSR_IA32_APICBASE:
476 data = kvm_get_apic_base(vcpu);
477 break;
478 case MSR_IA32_MISC_ENABLE:
479 data = vcpu->ia32_misc_enable_msr;
480 break;
481#ifdef CONFIG_X86_64
482 case MSR_EFER:
483 data = vcpu->shadow_efer;
484 break;
485#endif
486 default:
487 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
488 return 1;
489 }
490 *pdata = data;
491 return 0;
492}
493EXPORT_SYMBOL_GPL(kvm_get_msr_common);
494
313a3dc7
CO		 495/*
496 * Read or write a bunch of msrs. All parameters are kernel addresses.
497 *
498 * @return number of msrs set successfully.
499 */
500static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
501 struct kvm_msr_entry *entries,
502 int (*do_msr)(struct kvm_vcpu *vcpu,
503 unsigned index, u64 *data))
504{
505 int i;
506
507 vcpu_load(vcpu);
508
509 for (i = 0; i < msrs->nmsrs; ++i)
510 if (do_msr(vcpu, entries[i].index, &entries[i].data))
511 break;
512
513 vcpu_put(vcpu);
514
515 return i;
516}
517
518/*
519 * Read or write a bunch of msrs. Parameters are user addresses.
520 *
521 * @return number of msrs set successfully.
522 */
523static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
524 int (*do_msr)(struct kvm_vcpu *vcpu,
525 unsigned index, u64 *data),
526 int writeback)
527{
528 struct kvm_msrs msrs;
529 struct kvm_msr_entry *entries;
530 int r, n;
531 unsigned size;
532
533 r = -EFAULT;
534 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
535 goto out;
536
537 r = -E2BIG;
538 if (msrs.nmsrs >= MAX_IO_MSRS)
539 goto out;
540
541 r = -ENOMEM;
542 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
543 entries = vmalloc(size);
544 if (!entries)
545 goto out;
546
547 r = -EFAULT;
548 if (copy_from_user(entries, user_msrs->entries, size))
549 goto out_free;
550
551 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
552 if (r < 0)
553 goto out_free;
554
555 r = -EFAULT;
556 if (writeback && copy_to_user(user_msrs->entries, entries, size))
557 goto out_free;
558
559 r = n;
560
561out_free:
562 vfree(entries);
563out:
564 return r;
565}
566
e9b11c17
ZX		 567/*
568 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
569 * cached on it.
570 */
571void decache_vcpus_on_cpu(int cpu)
572{
573 struct kvm *vm;
574 struct kvm_vcpu *vcpu;
575 int i;
576
577 spin_lock(&kvm_lock);
578 list_for_each_entry(vm, &vm_list, vm_list)
579 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
580 vcpu = vm->vcpus[i];
581 if (!vcpu)
582 continue;
583 /*
584 * If the vcpu is locked, then it is running on some
585 * other cpu and therefore it is not cached on the
586 * cpu in question.
587 *
588 * If it's not locked, check the last cpu it executed
589 * on.
590 */
591 if (mutex_trylock(&vcpu->mutex)) {
592 if (vcpu->cpu == cpu) {
593 kvm_x86_ops->vcpu_decache(vcpu);
594 vcpu->cpu = -1;
595 }
596 mutex_unlock(&vcpu->mutex);
597 }
598 }
599 spin_unlock(&kvm_lock);
600}
601
018d00d2
ZX		 602int kvm_dev_ioctl_check_extension(long ext)
603{
604 int r;
605
606 switch (ext) {
607 case KVM_CAP_IRQCHIP:
608 case KVM_CAP_HLT:
609 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
610 case KVM_CAP_USER_MEMORY:
611 case KVM_CAP_SET_TSS_ADDR:
612 r = 1;
613 break;
614 default:
615 r = 0;
616 break;
617 }
618 return r;
619
620}
621
043405e1
CO		 622long kvm_arch_dev_ioctl(struct file *filp,
623 unsigned int ioctl, unsigned long arg)
624{
625 void __user *argp = (void __user *)arg;
626 long r;
627
628 switch (ioctl) {
629 case KVM_GET_MSR_INDEX_LIST: {
630 struct kvm_msr_list __user *user_msr_list = argp;
631 struct kvm_msr_list msr_list;
632 unsigned n;
633
634 r = -EFAULT;
635 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
636 goto out;
637 n = msr_list.nmsrs;
638 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
639 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
640 goto out;
641 r = -E2BIG;
642 if (n < num_msrs_to_save)
643 goto out;
644 r = -EFAULT;
645 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
646 num_msrs_to_save * sizeof(u32)))
647 goto out;
648 if (copy_to_user(user_msr_list->indices
649 + num_msrs_to_save * sizeof(u32),
650 &emulated_msrs,
651 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
652 goto out;
653 r = 0;
654 break;
655 }
656 default:
657 r = -EINVAL;
658 }
659out:
660 return r;
661}
662
313a3dc7
CO		 663void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
664{
665 kvm_x86_ops->vcpu_load(vcpu, cpu);
666}
667
668void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
669{
670 kvm_x86_ops->vcpu_put(vcpu);
671}
672
673static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
674{
675 u64 efer;
676 int i;
677 struct kvm_cpuid_entry *e, *entry;
678
679 rdmsrl(MSR_EFER, efer);
680 entry = NULL;
681 for (i = 0; i < vcpu->cpuid_nent; ++i) {
682 e = &vcpu->cpuid_entries[i];
683 if (e->function == 0x80000001) {
684 entry = e;
685 break;
686 }
687 }
688 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
689 entry->edx &= ~(1 << 20);
690 printk(KERN_INFO "kvm: guest NX capability removed\n");
691 }
692}
693
694static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
695 struct kvm_cpuid *cpuid,
696 struct kvm_cpuid_entry __user *entries)
697{
698 int r;
699
700 r = -E2BIG;
701 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
702 goto out;
703 r = -EFAULT;
704 if (copy_from_user(&vcpu->cpuid_entries, entries,
705 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
706 goto out;
707 vcpu->cpuid_nent = cpuid->nent;
708 cpuid_fix_nx_cap(vcpu);
709 return 0;
710
711out:
712 return r;
713}
714
715static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
716 struct kvm_lapic_state *s)
717{
718 vcpu_load(vcpu);
719 memcpy(s->regs, vcpu->apic->regs, sizeof *s);
720 vcpu_put(vcpu);
721
722 return 0;
723}
724
725static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
726 struct kvm_lapic_state *s)
727{
728 vcpu_load(vcpu);
729 memcpy(vcpu->apic->regs, s->regs, sizeof *s);
730 kvm_apic_post_state_restore(vcpu);
731 vcpu_put(vcpu);
732
733 return 0;
734}
735
736long kvm_arch_vcpu_ioctl(struct file *filp,
737 unsigned int ioctl, unsigned long arg)
738{
739 struct kvm_vcpu *vcpu = filp->private_data;
740 void __user *argp = (void __user *)arg;
741 int r;
742
743 switch (ioctl) {
744 case KVM_GET_LAPIC: {
745 struct kvm_lapic_state lapic;
746
747 memset(&lapic, 0, sizeof lapic);
748 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
749 if (r)
750 goto out;
751 r = -EFAULT;
752 if (copy_to_user(argp, &lapic, sizeof lapic))
753 goto out;
754 r = 0;
755 break;
756 }
757 case KVM_SET_LAPIC: {
758 struct kvm_lapic_state lapic;
759
760 r = -EFAULT;
761 if (copy_from_user(&lapic, argp, sizeof lapic))
762 goto out;
763 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
764 if (r)
765 goto out;
766 r = 0;
767 break;
768 }
769 case KVM_SET_CPUID: {
770 struct kvm_cpuid __user *cpuid_arg = argp;
771 struct kvm_cpuid cpuid;
772
773 r = -EFAULT;
774 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
775 goto out;
776 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
777 if (r)
778 goto out;
779 break;
780 }
781 case KVM_GET_MSRS:
782 r = msr_io(vcpu, argp, kvm_get_msr, 1);
783 break;
784 case KVM_SET_MSRS:
785 r = msr_io(vcpu, argp, do_set_msr, 0);
786 break;
787 default:
788 r = -EINVAL;
789 }
790out:
791 return r;
792}
793
1fe779f8
CO		 794static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
795{
796 int ret;
797
798 if (addr > (unsigned int)(-3 * PAGE_SIZE))
799 return -1;
800 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
801 return ret;
802}
803
804static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
805 u32 kvm_nr_mmu_pages)
806{
807 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
808 return -EINVAL;
809
810 mutex_lock(&kvm->lock);
811
812 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
813 kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
814
815 mutex_unlock(&kvm->lock);
816 return 0;
817}
818
819static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
820{
821 return kvm->n_alloc_mmu_pages;
822}
823
824/*
825 * Set a new alias region. Aliases map a portion of physical memory into
826 * another portion. This is useful for memory windows, for example the PC
827 * VGA region.
828 */
829static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
830 struct kvm_memory_alias *alias)
831{
832 int r, n;
833 struct kvm_mem_alias *p;
834
835 r = -EINVAL;
836 /* General sanity checks */
837 if (alias->memory_size & (PAGE_SIZE - 1))
838 goto out;
839 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
840 goto out;
841 if (alias->slot >= KVM_ALIAS_SLOTS)
842 goto out;
843 if (alias->guest_phys_addr + alias->memory_size
844 < alias->guest_phys_addr)
845 goto out;
846 if (alias->target_phys_addr + alias->memory_size
847 < alias->target_phys_addr)
848 goto out;
849
850 mutex_lock(&kvm->lock);
851
852 p = &kvm->aliases[alias->slot];
853 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
854 p->npages = alias->memory_size >> PAGE_SHIFT;
855 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
856
857 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
858 if (kvm->aliases[n - 1].npages)
859 break;
860 kvm->naliases = n;
861
862 kvm_mmu_zap_all(kvm);
863
864 mutex_unlock(&kvm->lock);
865
866 return 0;
867
868out:
869 return r;
870}
871
872static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
873{
874 int r;
875
876 r = 0;
877 switch (chip->chip_id) {
878 case KVM_IRQCHIP_PIC_MASTER:
879 memcpy(&chip->chip.pic,
880 &pic_irqchip(kvm)->pics[0],
881 sizeof(struct kvm_pic_state));
882 break;
883 case KVM_IRQCHIP_PIC_SLAVE:
884 memcpy(&chip->chip.pic,
885 &pic_irqchip(kvm)->pics[1],
886 sizeof(struct kvm_pic_state));
887 break;
888 case KVM_IRQCHIP_IOAPIC:
889 memcpy(&chip->chip.ioapic,
890 ioapic_irqchip(kvm),
891 sizeof(struct kvm_ioapic_state));
892 break;
893 default:
894 r = -EINVAL;
895 break;
896 }
897 return r;
898}
899
900static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
901{
902 int r;
903
904 r = 0;
905 switch (chip->chip_id) {
906 case KVM_IRQCHIP_PIC_MASTER:
907 memcpy(&pic_irqchip(kvm)->pics[0],
908 &chip->chip.pic,
909 sizeof(struct kvm_pic_state));
910 break;
911 case KVM_IRQCHIP_PIC_SLAVE:
912 memcpy(&pic_irqchip(kvm)->pics[1],
913 &chip->chip.pic,
914 sizeof(struct kvm_pic_state));
915 break;
916 case KVM_IRQCHIP_IOAPIC:
917 memcpy(ioapic_irqchip(kvm),
918 &chip->chip.ioapic,
919 sizeof(struct kvm_ioapic_state));
920 break;
921 default:
922 r = -EINVAL;
923 break;
924 }
925 kvm_pic_update_irq(pic_irqchip(kvm));
926 return r;
927}
928
929long kvm_arch_vm_ioctl(struct file *filp,
930 unsigned int ioctl, unsigned long arg)
931{
932 struct kvm *kvm = filp->private_data;
933 void __user *argp = (void __user *)arg;
934 int r = -EINVAL;
935
936 switch (ioctl) {
937 case KVM_SET_TSS_ADDR:
938 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
939 if (r < 0)
940 goto out;
941 break;
942 case KVM_SET_MEMORY_REGION: {
943 struct kvm_memory_region kvm_mem;
944 struct kvm_userspace_memory_region kvm_userspace_mem;
945
946 r = -EFAULT;
947 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
948 goto out;
949 kvm_userspace_mem.slot = kvm_mem.slot;
950 kvm_userspace_mem.flags = kvm_mem.flags;
951 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
952 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
953 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
954 if (r)
955 goto out;
956 break;
957 }
958 case KVM_SET_NR_MMU_PAGES:
959 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
960 if (r)
961 goto out;
962 break;
963 case KVM_GET_NR_MMU_PAGES:
964 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
965 break;
966 case KVM_SET_MEMORY_ALIAS: {
967 struct kvm_memory_alias alias;
968
969 r = -EFAULT;
970 if (copy_from_user(&alias, argp, sizeof alias))
971 goto out;
972 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
973 if (r)
974 goto out;
975 break;
976 }
977 case KVM_CREATE_IRQCHIP:
978 r = -ENOMEM;
979 kvm->vpic = kvm_create_pic(kvm);
980 if (kvm->vpic) {
981 r = kvm_ioapic_init(kvm);
982 if (r) {
983 kfree(kvm->vpic);
984 kvm->vpic = NULL;
985 goto out;
986 }
987 } else
988 goto out;
989 break;
990 case KVM_IRQ_LINE: {
991 struct kvm_irq_level irq_event;
992
993 r = -EFAULT;
994 if (copy_from_user(&irq_event, argp, sizeof irq_event))
995 goto out;
996 if (irqchip_in_kernel(kvm)) {
997 mutex_lock(&kvm->lock);
998 if (irq_event.irq < 16)
999 kvm_pic_set_irq(pic_irqchip(kvm),
1000 irq_event.irq,
1001 irq_event.level);
1002 kvm_ioapic_set_irq(kvm->vioapic,
1003 irq_event.irq,
1004 irq_event.level);
1005 mutex_unlock(&kvm->lock);
1006 r = 0;
1007 }
1008 break;
1009 }
1010 case KVM_GET_IRQCHIP: {
1011 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1012 struct kvm_irqchip chip;
1013
1014 r = -EFAULT;
1015 if (copy_from_user(&chip, argp, sizeof chip))
1016 goto out;
1017 r = -ENXIO;
1018 if (!irqchip_in_kernel(kvm))
1019 goto out;
1020 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1021 if (r)
1022 goto out;
1023 r = -EFAULT;
1024 if (copy_to_user(argp, &chip, sizeof chip))
1025 goto out;
1026 r = 0;
1027 break;
1028 }
1029 case KVM_SET_IRQCHIP: {
1030 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1031 struct kvm_irqchip chip;
1032
1033 r = -EFAULT;
1034 if (copy_from_user(&chip, argp, sizeof chip))
1035 goto out;
1036 r = -ENXIO;
1037 if (!irqchip_in_kernel(kvm))
1038 goto out;
1039 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1040 if (r)
1041 goto out;
1042 r = 0;
1043 break;
1044 }
1045 default:
1046 ;
1047 }
1048out:
1049 return r;
1050}
1051
043405e1
CO		 1052static __init void kvm_init_msr_list(void)
1053{
1054 u32 dummy[2];
1055 unsigned i, j;
1056
1057 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1058 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1059 continue;
1060 if (j < i)
1061 msrs_to_save[j] = msrs_to_save[i];
1062 j++;
1063 }
1064 num_msrs_to_save = j;
1065}
1066
bbd9b64e
CO		 1067/*
1068 * Only apic need an MMIO device hook, so shortcut now..
1069 */
1070static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1071 gpa_t addr)
1072{
1073 struct kvm_io_device *dev;
1074
1075 if (vcpu->apic) {
1076 dev = &vcpu->apic->dev;
1077 if (dev->in_range(dev, addr))
1078 return dev;
1079 }
1080 return NULL;
1081}
1082
1083
1084static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1085 gpa_t addr)
1086{
1087 struct kvm_io_device *dev;
1088
1089 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1090 if (dev == NULL)
1091 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1092 return dev;
1093}
1094
1095int emulator_read_std(unsigned long addr,
1096 void *val,
1097 unsigned int bytes,
1098 struct kvm_vcpu *vcpu)
1099{
1100 void *data = val;
1101
1102 while (bytes) {
1103 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1104 unsigned offset = addr & (PAGE_SIZE-1);
1105 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1106 int ret;
1107
1108 if (gpa == UNMAPPED_GVA)
1109 return X86EMUL_PROPAGATE_FAULT;
1110 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1111 if (ret < 0)
1112 return X86EMUL_UNHANDLEABLE;
1113
1114 bytes -= tocopy;
1115 data += tocopy;
1116 addr += tocopy;
1117 }
1118
1119 return X86EMUL_CONTINUE;
1120}
1121EXPORT_SYMBOL_GPL(emulator_read_std);
1122
1123static int emulator_write_std(unsigned long addr,
1124 const void *val,
1125 unsigned int bytes,
1126 struct kvm_vcpu *vcpu)
1127{
1128 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1129 return X86EMUL_UNHANDLEABLE;
1130}
1131
1132static int emulator_read_emulated(unsigned long addr,
1133 void *val,
1134 unsigned int bytes,
1135 struct kvm_vcpu *vcpu)
1136{
1137 struct kvm_io_device *mmio_dev;
1138 gpa_t gpa;
1139
1140 if (vcpu->mmio_read_completed) {
1141 memcpy(val, vcpu->mmio_data, bytes);
1142 vcpu->mmio_read_completed = 0;
1143 return X86EMUL_CONTINUE;
1144 }
1145
1146 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1147
1148 /* For APIC access vmexit */
1149 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1150 goto mmio;
1151
1152 if (emulator_read_std(addr, val, bytes, vcpu)
1153 == X86EMUL_CONTINUE)
1154 return X86EMUL_CONTINUE;
1155 if (gpa == UNMAPPED_GVA)
1156 return X86EMUL_PROPAGATE_FAULT;
1157
1158mmio:
1159 /*
1160 * Is this MMIO handled locally?
1161 */
1162 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1163 if (mmio_dev) {
1164 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1165 return X86EMUL_CONTINUE;
1166 }
1167
1168 vcpu->mmio_needed = 1;
1169 vcpu->mmio_phys_addr = gpa;
1170 vcpu->mmio_size = bytes;
1171 vcpu->mmio_is_write = 0;
1172
1173 return X86EMUL_UNHANDLEABLE;
1174}
1175
1176static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1177 const void *val, int bytes)
1178{
1179 int ret;
1180
1181 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1182 if (ret < 0)
1183 return 0;
1184 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1185 return 1;
1186}
1187
1188static int emulator_write_emulated_onepage(unsigned long addr,
1189 const void *val,
1190 unsigned int bytes,
1191 struct kvm_vcpu *vcpu)
1192{
1193 struct kvm_io_device *mmio_dev;
1194 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1195
1196 if (gpa == UNMAPPED_GVA) {
1197 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1198 return X86EMUL_PROPAGATE_FAULT;
1199 }
1200
1201 /* For APIC access vmexit */
1202 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1203 goto mmio;
1204
1205 if (emulator_write_phys(vcpu, gpa, val, bytes))
1206 return X86EMUL_CONTINUE;
1207
1208mmio:
1209 /*
1210 * Is this MMIO handled locally?
1211 */
1212 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1213 if (mmio_dev) {
1214 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1215 return X86EMUL_CONTINUE;
1216 }
1217
1218 vcpu->mmio_needed = 1;
1219 vcpu->mmio_phys_addr = gpa;
1220 vcpu->mmio_size = bytes;
1221 vcpu->mmio_is_write = 1;
1222 memcpy(vcpu->mmio_data, val, bytes);
1223
1224 return X86EMUL_CONTINUE;
1225}
1226
1227int emulator_write_emulated(unsigned long addr,
1228 const void *val,
1229 unsigned int bytes,
1230 struct kvm_vcpu *vcpu)
1231{
1232 /* Crossing a page boundary? */
1233 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1234 int rc, now;
1235
1236 now = -addr & ~PAGE_MASK;
1237 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1238 if (rc != X86EMUL_CONTINUE)
1239 return rc;
1240 addr += now;
1241 val += now;
1242 bytes -= now;
1243 }
1244 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1245}
1246EXPORT_SYMBOL_GPL(emulator_write_emulated);
1247
1248static int emulator_cmpxchg_emulated(unsigned long addr,
1249 const void *old,
1250 const void *new,
1251 unsigned int bytes,
1252 struct kvm_vcpu *vcpu)
1253{
1254 static int reported;
1255
1256 if (!reported) {
1257 reported = 1;
1258 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1259 }
1260 return emulator_write_emulated(addr, new, bytes, vcpu);
1261}
1262
1263static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1264{
1265 return kvm_x86_ops->get_segment_base(vcpu, seg);
1266}
1267
1268int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1269{
1270 return X86EMUL_CONTINUE;
1271}
1272
1273int emulate_clts(struct kvm_vcpu *vcpu)
1274{
1275 kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1276 return X86EMUL_CONTINUE;
1277}
1278
1279int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1280{
1281 struct kvm_vcpu *vcpu = ctxt->vcpu;
1282
1283 switch (dr) {
1284 case 0 ... 3:
1285 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1286 return X86EMUL_CONTINUE;
1287 default:
1288 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1289 return X86EMUL_UNHANDLEABLE;
1290 }
1291}
1292
1293int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1294{
1295 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1296 int exception;
1297
1298 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1299 if (exception) {
1300 /* FIXME: better handling */
1301 return X86EMUL_UNHANDLEABLE;
1302 }
1303 return X86EMUL_CONTINUE;
1304}
1305
1306void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1307{
1308 static int reported;
1309 u8 opcodes[4];
1310 unsigned long rip = vcpu->rip;
1311 unsigned long rip_linear;
1312
1313 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1314
1315 if (reported)
1316 return;
1317
1318 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1319
1320 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1321 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1322 reported = 1;
1323}
1324EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1325
1326struct x86_emulate_ops emulate_ops = {
1327 .read_std = emulator_read_std,
1328 .write_std = emulator_write_std,
1329 .read_emulated = emulator_read_emulated,
1330 .write_emulated = emulator_write_emulated,
1331 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1332};
1333
1334int emulate_instruction(struct kvm_vcpu *vcpu,
1335 struct kvm_run *run,
1336 unsigned long cr2,
1337 u16 error_code,
1338 int no_decode)
1339{
1340 int r;
1341
1342 vcpu->mmio_fault_cr2 = cr2;
1343 kvm_x86_ops->cache_regs(vcpu);
1344
1345 vcpu->mmio_is_write = 0;
1346 vcpu->pio.string = 0;
1347
1348 if (!no_decode) {
1349 int cs_db, cs_l;
1350 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1351
1352 vcpu->emulate_ctxt.vcpu = vcpu;
1353 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1354 vcpu->emulate_ctxt.cr2 = cr2;
1355 vcpu->emulate_ctxt.mode =
1356 (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1357 ? X86EMUL_MODE_REAL : cs_l
1358 ? X86EMUL_MODE_PROT64 : cs_db
1359 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1360
1361 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1362 vcpu->emulate_ctxt.cs_base = 0;
1363 vcpu->emulate_ctxt.ds_base = 0;
1364 vcpu->emulate_ctxt.es_base = 0;
1365 vcpu->emulate_ctxt.ss_base = 0;
1366 } else {
1367 vcpu->emulate_ctxt.cs_base =
1368 get_segment_base(vcpu, VCPU_SREG_CS);
1369 vcpu->emulate_ctxt.ds_base =
1370 get_segment_base(vcpu, VCPU_SREG_DS);
1371 vcpu->emulate_ctxt.es_base =
1372 get_segment_base(vcpu, VCPU_SREG_ES);
1373 vcpu->emulate_ctxt.ss_base =
1374 get_segment_base(vcpu, VCPU_SREG_SS);
1375 }
1376
1377 vcpu->emulate_ctxt.gs_base =
1378 get_segment_base(vcpu, VCPU_SREG_GS);
1379 vcpu->emulate_ctxt.fs_base =
1380 get_segment_base(vcpu, VCPU_SREG_FS);
1381
1382 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1383 if (r) {
1384 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1385 return EMULATE_DONE;
1386 return EMULATE_FAIL;
1387 }
1388 }
1389
1390 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1391
1392 if (vcpu->pio.string)
1393 return EMULATE_DO_MMIO;
1394
1395 if ((r || vcpu->mmio_is_write) && run) {
1396 run->exit_reason = KVM_EXIT_MMIO;
1397 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1398 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1399 run->mmio.len = vcpu->mmio_size;
1400 run->mmio.is_write = vcpu->mmio_is_write;
1401 }
1402
1403 if (r) {
1404 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1405 return EMULATE_DONE;
1406 if (!vcpu->mmio_needed) {
1407 kvm_report_emulation_failure(vcpu, "mmio");
1408 return EMULATE_FAIL;
1409 }
1410 return EMULATE_DO_MMIO;
1411 }
1412
1413 kvm_x86_ops->decache_regs(vcpu);
1414 kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1415
1416 if (vcpu->mmio_is_write) {
1417 vcpu->mmio_needed = 0;
1418 return EMULATE_DO_MMIO;
1419 }
1420
1421 return EMULATE_DONE;
1422}
1423EXPORT_SYMBOL_GPL(emulate_instruction);
1424
de7d789a
CO		 1425static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1426{
1427 int i;
1428
1429 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1430 if (vcpu->pio.guest_pages[i]) {
1431 kvm_release_page(vcpu->pio.guest_pages[i]);
1432 vcpu->pio.guest_pages[i] = NULL;
1433 }
1434}
1435
1436static int pio_copy_data(struct kvm_vcpu *vcpu)
1437{
1438 void *p = vcpu->pio_data;
1439 void *q;
1440 unsigned bytes;
1441 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1442
1443 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1444 PAGE_KERNEL);
1445 if (!q) {
1446 free_pio_guest_pages(vcpu);
1447 return -ENOMEM;
1448 }
1449 q += vcpu->pio.guest_page_offset;
1450 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1451 if (vcpu->pio.in)
1452 memcpy(q, p, bytes);
1453 else
1454 memcpy(p, q, bytes);
1455 q -= vcpu->pio.guest_page_offset;
1456 vunmap(q);
1457 free_pio_guest_pages(vcpu);
1458 return 0;
1459}
1460
1461int complete_pio(struct kvm_vcpu *vcpu)
1462{
1463 struct kvm_pio_request *io = &vcpu->pio;
1464 long delta;
1465 int r;
1466
1467 kvm_x86_ops->cache_regs(vcpu);
1468
1469 if (!io->string) {
1470 if (io->in)
1471 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1472 io->size);
1473 } else {
1474 if (io->in) {
1475 r = pio_copy_data(vcpu);
1476 if (r) {
1477 kvm_x86_ops->cache_regs(vcpu);
1478 return r;
1479 }
1480 }
1481
1482 delta = 1;
1483 if (io->rep) {
1484 delta *= io->cur_count;
1485 /*
1486 * The size of the register should really depend on
1487 * current address size.
1488 */
1489 vcpu->regs[VCPU_REGS_RCX] -= delta;
1490 }
1491 if (io->down)
1492 delta = -delta;
1493 delta *= io->size;
1494 if (io->in)
1495 vcpu->regs[VCPU_REGS_RDI] += delta;
1496 else
1497 vcpu->regs[VCPU_REGS_RSI] += delta;
1498 }
1499
1500 kvm_x86_ops->decache_regs(vcpu);
1501
1502 io->count -= io->cur_count;
1503 io->cur_count = 0;
1504
1505 return 0;
1506}
1507
1508static void kernel_pio(struct kvm_io_device *pio_dev,
1509 struct kvm_vcpu *vcpu,
1510 void *pd)
1511{
1512 /* TODO: String I/O for in kernel device */
1513
1514 mutex_lock(&vcpu->kvm->lock);
1515 if (vcpu->pio.in)
1516 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1517 vcpu->pio.size,
1518 pd);
1519 else
1520 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1521 vcpu->pio.size,
1522 pd);
1523 mutex_unlock(&vcpu->kvm->lock);
1524}
1525
1526static void pio_string_write(struct kvm_io_device *pio_dev,
1527 struct kvm_vcpu *vcpu)
1528{
1529 struct kvm_pio_request *io = &vcpu->pio;
1530 void *pd = vcpu->pio_data;
1531 int i;
1532
1533 mutex_lock(&vcpu->kvm->lock);
1534 for (i = 0; i < io->cur_count; i++) {
1535 kvm_iodevice_write(pio_dev, io->port,
1536 io->size,
1537 pd);
1538 pd += io->size;
1539 }
1540 mutex_unlock(&vcpu->kvm->lock);
1541}
1542
1543static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1544 gpa_t addr)
1545{
1546 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1547}
1548
1549int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1550 int size, unsigned port)
1551{
1552 struct kvm_io_device *pio_dev;
1553
1554 vcpu->run->exit_reason = KVM_EXIT_IO;
1555 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1556 vcpu->run->io.size = vcpu->pio.size = size;
1557 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1558 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1559 vcpu->run->io.port = vcpu->pio.port = port;
1560 vcpu->pio.in = in;
1561 vcpu->pio.string = 0;
1562 vcpu->pio.down = 0;
1563 vcpu->pio.guest_page_offset = 0;
1564 vcpu->pio.rep = 0;
1565
1566 kvm_x86_ops->cache_regs(vcpu);
1567 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1568 kvm_x86_ops->decache_regs(vcpu);
1569
1570 kvm_x86_ops->skip_emulated_instruction(vcpu);
1571
1572 pio_dev = vcpu_find_pio_dev(vcpu, port);
1573 if (pio_dev) {
1574 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1575 complete_pio(vcpu);
1576 return 1;
1577 }
1578 return 0;
1579}
1580EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1581
1582int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1583 int size, unsigned long count, int down,
1584 gva_t address, int rep, unsigned port)
1585{
1586 unsigned now, in_page;
1587 int i, ret = 0;
1588 int nr_pages = 1;
1589 struct page *page;
1590 struct kvm_io_device *pio_dev;
1591
1592 vcpu->run->exit_reason = KVM_EXIT_IO;
1593 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1594 vcpu->run->io.size = vcpu->pio.size = size;
1595 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1596 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1597 vcpu->run->io.port = vcpu->pio.port = port;
1598 vcpu->pio.in = in;
1599 vcpu->pio.string = 1;
1600 vcpu->pio.down = down;
1601 vcpu->pio.guest_page_offset = offset_in_page(address);
1602 vcpu->pio.rep = rep;
1603
1604 if (!count) {
1605 kvm_x86_ops->skip_emulated_instruction(vcpu);
1606 return 1;
1607 }
1608
1609 if (!down)
1610 in_page = PAGE_SIZE - offset_in_page(address);
1611 else
1612 in_page = offset_in_page(address) + size;
1613 now = min(count, (unsigned long)in_page / size);
1614 if (!now) {
1615 /*
1616 * String I/O straddles page boundary. Pin two guest pages
1617 * so that we satisfy atomicity constraints. Do just one
1618 * transaction to avoid complexity.
1619 */
1620 nr_pages = 2;
1621 now = 1;
1622 }
1623 if (down) {
1624 /*
1625 * String I/O in reverse. Yuck. Kill the guest, fix later.
1626 */
1627 pr_unimpl(vcpu, "guest string pio down\n");
1628 inject_gp(vcpu);
1629 return 1;
1630 }
1631 vcpu->run->io.count = now;
1632 vcpu->pio.cur_count = now;
1633
1634 if (vcpu->pio.cur_count == vcpu->pio.count)
1635 kvm_x86_ops->skip_emulated_instruction(vcpu);
1636
1637 for (i = 0; i < nr_pages; ++i) {
1638 mutex_lock(&vcpu->kvm->lock);
1639 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1640 vcpu->pio.guest_pages[i] = page;
1641 mutex_unlock(&vcpu->kvm->lock);
1642 if (!page) {
1643 inject_gp(vcpu);
1644 free_pio_guest_pages(vcpu);
1645 return 1;
1646 }
1647 }
1648
1649 pio_dev = vcpu_find_pio_dev(vcpu, port);
1650 if (!vcpu->pio.in) {
1651 /* string PIO write */
1652 ret = pio_copy_data(vcpu);
1653 if (ret >= 0 && pio_dev) {
1654 pio_string_write(pio_dev, vcpu);
1655 complete_pio(vcpu);
1656 if (vcpu->pio.count == 0)
1657 ret = 1;
1658 }
1659 } else if (pio_dev)
1660 pr_unimpl(vcpu, "no string pio read support yet, "
1661 "port %x size %d count %ld\n",
1662 port, size, count);
1663
1664 return ret;
1665}
1666EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1667
f8c16bba 1668int kvm_arch_init(void *opaque)
043405e1 1669{
f8c16bba
ZX		 1670 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
1671
043405e1 1672 kvm_init_msr_list();
f8c16bba
ZX		 1673
1674 if (kvm_x86_ops) {
1675 printk(KERN_ERR "kvm: already loaded the other module\n");
1676 return -EEXIST;
1677 }
1678
1679 if (!ops->cpu_has_kvm_support()) {
1680 printk(KERN_ERR "kvm: no hardware support\n");
1681 return -EOPNOTSUPP;
1682 }
1683 if (ops->disabled_by_bios()) {
1684 printk(KERN_ERR "kvm: disabled by bios\n");
1685 return -EOPNOTSUPP;
1686 }
1687
1688 kvm_x86_ops = ops;
1689
1690 return 0;
043405e1 1691}
8776e519 1692
f8c16bba
ZX		 1693void kvm_arch_exit(void)
1694{
1695 kvm_x86_ops = NULL;
1696 }
1697
8776e519
HB		 1698int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1699{
1700 ++vcpu->stat.halt_exits;
1701 if (irqchip_in_kernel(vcpu->kvm)) {
1702 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1703 kvm_vcpu_block(vcpu);
1704 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1705 return -EINTR;
1706 return 1;
1707 } else {
1708 vcpu->run->exit_reason = KVM_EXIT_HLT;
1709 return 0;
1710 }
1711}
1712EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1713
1714int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1715{
1716 unsigned long nr, a0, a1, a2, a3, ret;
1717
1718 kvm_x86_ops->cache_regs(vcpu);
1719
1720 nr = vcpu->regs[VCPU_REGS_RAX];
1721 a0 = vcpu->regs[VCPU_REGS_RBX];
1722 a1 = vcpu->regs[VCPU_REGS_RCX];
1723 a2 = vcpu->regs[VCPU_REGS_RDX];
1724 a3 = vcpu->regs[VCPU_REGS_RSI];
1725
1726 if (!is_long_mode(vcpu)) {
1727 nr &= 0xFFFFFFFF;
1728 a0 &= 0xFFFFFFFF;
1729 a1 &= 0xFFFFFFFF;
1730 a2 &= 0xFFFFFFFF;
1731 a3 &= 0xFFFFFFFF;
1732 }
1733
1734 switch (nr) {
1735 default:
1736 ret = -KVM_ENOSYS;
1737 break;
1738 }
1739 vcpu->regs[VCPU_REGS_RAX] = ret;
1740 kvm_x86_ops->decache_regs(vcpu);
1741 return 0;
1742}
1743EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1744
1745int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1746{
1747 char instruction[3];
1748 int ret = 0;
1749
1750 mutex_lock(&vcpu->kvm->lock);
1751
1752 /*
1753 * Blow out the MMU to ensure that no other VCPU has an active mapping
1754 * to ensure that the updated hypercall appears atomically across all
1755 * VCPUs.
1756 */
1757 kvm_mmu_zap_all(vcpu->kvm);
1758
1759 kvm_x86_ops->cache_regs(vcpu);
1760 kvm_x86_ops->patch_hypercall(vcpu, instruction);
1761 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1762 != X86EMUL_CONTINUE)
1763 ret = -EFAULT;
1764
1765 mutex_unlock(&vcpu->kvm->lock);
1766
1767 return ret;
1768}
1769
1770static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1771{
1772 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1773}
1774
1775void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1776{
1777 struct descriptor_table dt = { limit, base };
1778
1779 kvm_x86_ops->set_gdt(vcpu, &dt);
1780}
1781
1782void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1783{
1784 struct descriptor_table dt = { limit, base };
1785
1786 kvm_x86_ops->set_idt(vcpu, &dt);
1787}
1788
1789void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1790 unsigned long *rflags)
1791{
1792 lmsw(vcpu, msw);
1793 *rflags = kvm_x86_ops->get_rflags(vcpu);
1794}
1795
1796unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1797{
1798 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1799 switch (cr) {
1800 case 0:
1801 return vcpu->cr0;
1802 case 2:
1803 return vcpu->cr2;
1804 case 3:
1805 return vcpu->cr3;
1806 case 4:
1807 return vcpu->cr4;
1808 default:
1809 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1810 return 0;
1811 }
1812}
1813
1814void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1815 unsigned long *rflags)
1816{
1817 switch (cr) {
1818 case 0:
1819 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1820 *rflags = kvm_x86_ops->get_rflags(vcpu);
1821 break;
1822 case 2:
1823 vcpu->cr2 = val;
1824 break;
1825 case 3:
1826 set_cr3(vcpu, val);
1827 break;
1828 case 4:
1829 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1830 break;
1831 default:
1832 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1833 }
1834}
1835
1836void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1837{
1838 int i;
1839 u32 function;
1840 struct kvm_cpuid_entry *e, *best;
1841
1842 kvm_x86_ops->cache_regs(vcpu);
1843 function = vcpu->regs[VCPU_REGS_RAX];
1844 vcpu->regs[VCPU_REGS_RAX] = 0;
1845 vcpu->regs[VCPU_REGS_RBX] = 0;
1846 vcpu->regs[VCPU_REGS_RCX] = 0;
1847 vcpu->regs[VCPU_REGS_RDX] = 0;
1848 best = NULL;
1849 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1850 e = &vcpu->cpuid_entries[i];
1851 if (e->function == function) {
1852 best = e;
1853 break;
1854 }
1855 /*
1856 * Both basic or both extended?
1857 */
1858 if (((e->function ^ function) & 0x80000000) == 0)
1859 if (!best || e->function > best->function)
1860 best = e;
1861 }
1862 if (best) {
1863 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1864 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1865 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1866 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1867 }
1868 kvm_x86_ops->decache_regs(vcpu);
1869 kvm_x86_ops->skip_emulated_instruction(vcpu);
1870}
1871EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
d0752060 1872
b6c7a5dc
HB		 1873/*
1874 * Check if userspace requested an interrupt window, and that the
1875 * interrupt window is open.
1876 *
1877 * No need to exit to userspace if we already have an interrupt queued.
1878 */
1879static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1880 struct kvm_run *kvm_run)
1881{
1882 return (!vcpu->irq_summary &&
1883 kvm_run->request_interrupt_window &&
1884 vcpu->interrupt_window_open &&
1885 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1886}
1887
1888static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1889 struct kvm_run *kvm_run)
1890{
1891 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1892 kvm_run->cr8 = get_cr8(vcpu);
1893 kvm_run->apic_base = kvm_get_apic_base(vcpu);
1894 if (irqchip_in_kernel(vcpu->kvm))
1895 kvm_run->ready_for_interrupt_injection = 1;
1896 else
1897 kvm_run->ready_for_interrupt_injection =
1898 (vcpu->interrupt_window_open &&
1899 vcpu->irq_summary == 0);
1900}
1901
1902static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1903{
1904 int r;
1905
1906 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1907 pr_debug("vcpu %d received sipi with vector # %x\n",
1908 vcpu->vcpu_id, vcpu->sipi_vector);
1909 kvm_lapic_reset(vcpu);
1910 r = kvm_x86_ops->vcpu_reset(vcpu);
1911 if (r)
1912 return r;
1913 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1914 }
1915
1916preempted:
1917 if (vcpu->guest_debug.enabled)
1918 kvm_x86_ops->guest_debug_pre(vcpu);
1919
1920again:
1921 r = kvm_mmu_reload(vcpu);
1922 if (unlikely(r))
1923 goto out;
1924
1925 kvm_inject_pending_timer_irqs(vcpu);
1926
1927 preempt_disable();
1928
1929 kvm_x86_ops->prepare_guest_switch(vcpu);
1930 kvm_load_guest_fpu(vcpu);
1931
1932 local_irq_disable();
1933
1934 if (signal_pending(current)) {
1935 local_irq_enable();
1936 preempt_enable();
1937 r = -EINTR;
1938 kvm_run->exit_reason = KVM_EXIT_INTR;
1939 ++vcpu->stat.signal_exits;
1940 goto out;
1941 }
1942
1943 if (irqchip_in_kernel(vcpu->kvm))
1944 kvm_x86_ops->inject_pending_irq(vcpu);
1945 else if (!vcpu->mmio_read_completed)
1946 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1947
1948 vcpu->guest_mode = 1;
1949 kvm_guest_enter();
1950
1951 if (vcpu->requests)
1952 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1953 kvm_x86_ops->tlb_flush(vcpu);
1954
1955 kvm_x86_ops->run(vcpu, kvm_run);
1956
1957 vcpu->guest_mode = 0;
1958 local_irq_enable();
1959
1960 ++vcpu->stat.exits;
1961
1962 /*
1963 * We must have an instruction between local_irq_enable() and
1964 * kvm_guest_exit(), so the timer interrupt isn't delayed by
1965 * the interrupt shadow. The stat.exits increment will do nicely.
1966 * But we need to prevent reordering, hence this barrier():
1967 */
1968 barrier();
1969
1970 kvm_guest_exit();
1971
1972 preempt_enable();
1973
1974 /*
1975 * Profile KVM exit RIPs:
1976 */
1977 if (unlikely(prof_on == KVM_PROFILING)) {
1978 kvm_x86_ops->cache_regs(vcpu);
1979 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1980 }
1981
1982 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1983
1984 if (r > 0) {
1985 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1986 r = -EINTR;
1987 kvm_run->exit_reason = KVM_EXIT_INTR;
1988 ++vcpu->stat.request_irq_exits;
1989 goto out;
1990 }
1991 if (!need_resched()) {
1992 ++vcpu->stat.light_exits;
1993 goto again;
1994 }
1995 }
1996
1997out:
1998 if (r > 0) {
1999 kvm_resched(vcpu);
2000 goto preempted;
2001 }
2002
2003 post_kvm_run_save(vcpu, kvm_run);
2004
2005 return r;
2006}
2007
2008int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2009{
2010 int r;
2011 sigset_t sigsaved;
2012
2013 vcpu_load(vcpu);
2014
2015 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2016 kvm_vcpu_block(vcpu);
2017 vcpu_put(vcpu);
2018 return -EAGAIN;
2019 }
2020
2021 if (vcpu->sigset_active)
2022 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2023
2024 /* re-sync apic's tpr */
2025 if (!irqchip_in_kernel(vcpu->kvm))
2026 set_cr8(vcpu, kvm_run->cr8);
2027
2028 if (vcpu->pio.cur_count) {
2029 r = complete_pio(vcpu);
2030 if (r)
2031 goto out;
2032 }
2033#if CONFIG_HAS_IOMEM
2034 if (vcpu->mmio_needed) {
2035 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2036 vcpu->mmio_read_completed = 1;
2037 vcpu->mmio_needed = 0;
2038 r = emulate_instruction(vcpu, kvm_run,
2039 vcpu->mmio_fault_cr2, 0, 1);
2040 if (r == EMULATE_DO_MMIO) {
2041 /*
2042 * Read-modify-write. Back to userspace.
2043 */
2044 r = 0;
2045 goto out;
2046 }
2047 }
2048#endif
2049 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2050 kvm_x86_ops->cache_regs(vcpu);
2051 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2052 kvm_x86_ops->decache_regs(vcpu);
2053 }
2054
2055 r = __vcpu_run(vcpu, kvm_run);
2056
2057out:
2058 if (vcpu->sigset_active)
2059 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2060
2061 vcpu_put(vcpu);
2062 return r;
2063}
2064
2065int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2066{
2067 vcpu_load(vcpu);
2068
2069 kvm_x86_ops->cache_regs(vcpu);
2070
2071 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2072 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2073 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2074 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2075 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2076 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2077 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2078 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2079#ifdef CONFIG_X86_64
2080 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2081 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2082 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2083 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2084 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2085 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2086 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2087 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2088#endif
2089
2090 regs->rip = vcpu->rip;
2091 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2092
2093 /*
2094 * Don't leak debug flags in case they were set for guest debugging
2095 */
2096 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2097 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2098
2099 vcpu_put(vcpu);
2100
2101 return 0;
2102}
2103
2104int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2105{
2106 vcpu_load(vcpu);
2107
2108 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2109 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2110 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2111 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2112 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2113 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2114 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2115 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2116#ifdef CONFIG_X86_64
2117 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2118 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2119 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2120 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2121 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2122 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2123 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2124 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2125#endif
2126
2127 vcpu->rip = regs->rip;
2128 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2129
2130 kvm_x86_ops->decache_regs(vcpu);
2131
2132 vcpu_put(vcpu);
2133
2134 return 0;
2135}
2136
2137static void get_segment(struct kvm_vcpu *vcpu,
2138 struct kvm_segment *var, int seg)
2139{
2140 return kvm_x86_ops->get_segment(vcpu, var, seg);
2141}
2142
2143void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2144{
2145 struct kvm_segment cs;
2146
2147 get_segment(vcpu, &cs, VCPU_SREG_CS);
2148 *db = cs.db;
2149 *l = cs.l;
2150}
2151EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2152
2153int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2154 struct kvm_sregs *sregs)
2155{
2156 struct descriptor_table dt;
2157 int pending_vec;
2158
2159 vcpu_load(vcpu);
2160
2161 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2162 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2163 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2164 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2165 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2166 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2167
2168 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2169 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2170
2171 kvm_x86_ops->get_idt(vcpu, &dt);
2172 sregs->idt.limit = dt.limit;
2173 sregs->idt.base = dt.base;
2174 kvm_x86_ops->get_gdt(vcpu, &dt);
2175 sregs->gdt.limit = dt.limit;
2176 sregs->gdt.base = dt.base;
2177
2178 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2179 sregs->cr0 = vcpu->cr0;
2180 sregs->cr2 = vcpu->cr2;
2181 sregs->cr3 = vcpu->cr3;
2182 sregs->cr4 = vcpu->cr4;
2183 sregs->cr8 = get_cr8(vcpu);
2184 sregs->efer = vcpu->shadow_efer;
2185 sregs->apic_base = kvm_get_apic_base(vcpu);
2186
2187 if (irqchip_in_kernel(vcpu->kvm)) {
2188 memset(sregs->interrupt_bitmap, 0,
2189 sizeof sregs->interrupt_bitmap);
2190 pending_vec = kvm_x86_ops->get_irq(vcpu);
2191 if (pending_vec >= 0)
2192 set_bit(pending_vec,
2193 (unsigned long *)sregs->interrupt_bitmap);
2194 } else
2195 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2196 sizeof sregs->interrupt_bitmap);
2197
2198 vcpu_put(vcpu);
2199
2200 return 0;
2201}
2202
2203static void set_segment(struct kvm_vcpu *vcpu,
2204 struct kvm_segment *var, int seg)
2205{
2206 return kvm_x86_ops->set_segment(vcpu, var, seg);
2207}
2208
2209int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2210 struct kvm_sregs *sregs)
2211{
2212 int mmu_reset_needed = 0;
2213 int i, pending_vec, max_bits;
2214 struct descriptor_table dt;
2215
2216 vcpu_load(vcpu);
2217
2218 dt.limit = sregs->idt.limit;
2219 dt.base = sregs->idt.base;
2220 kvm_x86_ops->set_idt(vcpu, &dt);
2221 dt.limit = sregs->gdt.limit;
2222 dt.base = sregs->gdt.base;
2223 kvm_x86_ops->set_gdt(vcpu, &dt);
2224
2225 vcpu->cr2 = sregs->cr2;
2226 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2227 vcpu->cr3 = sregs->cr3;
2228
2229 set_cr8(vcpu, sregs->cr8);
2230
2231 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2232#ifdef CONFIG_X86_64
2233 kvm_x86_ops->set_efer(vcpu, sregs->efer);
2234#endif
2235 kvm_set_apic_base(vcpu, sregs->apic_base);
2236
2237 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2238
2239 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2240 vcpu->cr0 = sregs->cr0;
2241 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2242
2243 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2244 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2245 if (!is_long_mode(vcpu) && is_pae(vcpu))
2246 load_pdptrs(vcpu, vcpu->cr3);
2247
2248 if (mmu_reset_needed)
2249 kvm_mmu_reset_context(vcpu);
2250
2251 if (!irqchip_in_kernel(vcpu->kvm)) {
2252 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2253 sizeof vcpu->irq_pending);
2254 vcpu->irq_summary = 0;
2255 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2256 if (vcpu->irq_pending[i])
2257 __set_bit(i, &vcpu->irq_summary);
2258 } else {
2259 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2260 pending_vec = find_first_bit(
2261 (const unsigned long *)sregs->interrupt_bitmap,
2262 max_bits);
2263 /* Only pending external irq is handled here */
2264 if (pending_vec < max_bits) {
2265 kvm_x86_ops->set_irq(vcpu, pending_vec);
2266 pr_debug("Set back pending irq %d\n",
2267 pending_vec);
2268 }
2269 }
2270
2271 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2272 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2273 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2274 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2275 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2276 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2277
2278 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2279 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2280
2281 vcpu_put(vcpu);
2282
2283 return 0;
2284}
2285
2286int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2287 struct kvm_debug_guest *dbg)
2288{
2289 int r;
2290
2291 vcpu_load(vcpu);
2292
2293 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2294
2295 vcpu_put(vcpu);
2296
2297 return r;
2298}
2299
d0752060
HB		 2300/*
2301 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2302 * we have asm/x86/processor.h
2303 */
2304struct fxsave {
2305 u16 cwd;
2306 u16 swd;
2307 u16 twd;
2308 u16 fop;
2309 u64 rip;
2310 u64 rdp;
2311 u32 mxcsr;
2312 u32 mxcsr_mask;
2313 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2314#ifdef CONFIG_X86_64
2315 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2316#else
2317 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2318#endif
2319};
2320
8b006791
ZX		 2321/*
2322 * Translate a guest virtual address to a guest physical address.
2323 */
2324int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2325 struct kvm_translation *tr)
2326{
2327 unsigned long vaddr = tr->linear_address;
2328 gpa_t gpa;
2329
2330 vcpu_load(vcpu);
2331 mutex_lock(&vcpu->kvm->lock);
2332 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2333 tr->physical_address = gpa;
2334 tr->valid = gpa != UNMAPPED_GVA;
2335 tr->writeable = 1;
2336 tr->usermode = 0;
2337 mutex_unlock(&vcpu->kvm->lock);
2338 vcpu_put(vcpu);
2339
2340 return 0;
2341}
2342
d0752060
HB		 2343int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2344{
2345 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2346
2347 vcpu_load(vcpu);
2348
2349 memcpy(fpu->fpr, fxsave->st_space, 128);
2350 fpu->fcw = fxsave->cwd;
2351 fpu->fsw = fxsave->swd;
2352 fpu->ftwx = fxsave->twd;
2353 fpu->last_opcode = fxsave->fop;
2354 fpu->last_ip = fxsave->rip;
2355 fpu->last_dp = fxsave->rdp;
2356 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2357
2358 vcpu_put(vcpu);
2359
2360 return 0;
2361}
2362
2363int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2364{
2365 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2366
2367 vcpu_load(vcpu);
2368
2369 memcpy(fxsave->st_space, fpu->fpr, 128);
2370 fxsave->cwd = fpu->fcw;
2371 fxsave->swd = fpu->fsw;
2372 fxsave->twd = fpu->ftwx;
2373 fxsave->fop = fpu->last_opcode;
2374 fxsave->rip = fpu->last_ip;
2375 fxsave->rdp = fpu->last_dp;
2376 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2377
2378 vcpu_put(vcpu);
2379
2380 return 0;
2381}
2382
2383void fx_init(struct kvm_vcpu *vcpu)
2384{
2385 unsigned after_mxcsr_mask;
2386
2387 /* Initialize guest FPU by resetting ours and saving into guest's */
2388 preempt_disable();
2389 fx_save(&vcpu->host_fx_image);
2390 fpu_init();
2391 fx_save(&vcpu->guest_fx_image);
2392 fx_restore(&vcpu->host_fx_image);
2393 preempt_enable();
2394
2395 vcpu->cr0 |= X86_CR0_ET;
2396 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2397 vcpu->guest_fx_image.mxcsr = 0x1f80;
2398 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2399 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2400}
2401EXPORT_SYMBOL_GPL(fx_init);
2402
2403void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2404{
2405 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2406 return;
2407
2408 vcpu->guest_fpu_loaded = 1;
2409 fx_save(&vcpu->host_fx_image);
2410 fx_restore(&vcpu->guest_fx_image);
2411}
2412EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2413
2414void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2415{
2416 if (!vcpu->guest_fpu_loaded)
2417 return;
2418
2419 vcpu->guest_fpu_loaded = 0;
2420 fx_save(&vcpu->guest_fx_image);
2421 fx_restore(&vcpu->host_fx_image);
2422}
2423EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
e9b11c17
ZX		 2424
2425void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
2426{
2427 kvm_x86_ops->vcpu_free(vcpu);
2428}
2429
2430struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2431 unsigned int id)
2432{
2433 int r;
2434 struct kvm_vcpu *vcpu = kvm_x86_ops->vcpu_create(kvm, id);
2435
2436 if (IS_ERR(vcpu)) {
2437 r = -ENOMEM;
2438 goto fail;
2439 }
2440
2441 /* We do fxsave: this must be aligned. */
2442 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2443
2444 vcpu_load(vcpu);
2445 r = kvm_arch_vcpu_reset(vcpu);
2446 if (r == 0)
2447 r = kvm_mmu_setup(vcpu);
2448 vcpu_put(vcpu);
2449 if (r < 0)
2450 goto free_vcpu;
2451
2452 return vcpu;
2453free_vcpu:
2454 kvm_x86_ops->vcpu_free(vcpu);
2455fail:
2456 return ERR_PTR(r);
2457}
2458
2459void kvm_arch_vcpu_destory(struct kvm_vcpu *vcpu)
2460{
2461 vcpu_load(vcpu);
2462 kvm_mmu_unload(vcpu);
2463 vcpu_put(vcpu);
2464
2465 kvm_x86_ops->vcpu_free(vcpu);
2466}
2467
2468int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
2469{
2470 return kvm_x86_ops->vcpu_reset(vcpu);
2471}
2472
2473void kvm_arch_hardware_enable(void *garbage)
2474{
2475 kvm_x86_ops->hardware_enable(garbage);
2476}
2477
2478void kvm_arch_hardware_disable(void *garbage)
2479{
2480 kvm_x86_ops->hardware_disable(garbage);
2481}
2482
2483int kvm_arch_hardware_setup(void)
2484{
2485 return kvm_x86_ops->hardware_setup();
2486}
2487
2488void kvm_arch_hardware_unsetup(void)
2489{
2490 kvm_x86_ops->hardware_unsetup();
2491}
2492
2493void kvm_arch_check_processor_compat(void *rtn)
2494{
2495 kvm_x86_ops->check_processor_compatibility(rtn);
2496}
2497
2498int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2499{
2500 struct page *page;
2501 struct kvm *kvm;
2502 int r;
2503
2504 BUG_ON(vcpu->kvm == NULL);
2505 kvm = vcpu->kvm;
2506
2507 vcpu->mmu.root_hpa = INVALID_PAGE;
2508 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
2509 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2510 else
2511 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
2512
2513 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2514 if (!page) {
2515 r = -ENOMEM;
2516 goto fail;
2517 }
2518 vcpu->pio_data = page_address(page);
2519
2520 r = kvm_mmu_create(vcpu);
2521 if (r < 0)
2522 goto fail_free_pio_data;
2523
2524 if (irqchip_in_kernel(kvm)) {
2525 r = kvm_create_lapic(vcpu);
2526 if (r < 0)
2527 goto fail_mmu_destroy;
2528 }
2529
2530 return 0;
2531
2532fail_mmu_destroy:
2533 kvm_mmu_destroy(vcpu);
2534fail_free_pio_data:
2535 free_page((unsigned long)vcpu->pio_data);
2536fail:
2537 return r;
2538}
2539
2540void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
2541{
2542 kvm_free_lapic(vcpu);
2543 kvm_mmu_destroy(vcpu);
2544 free_page((unsigned long)vcpu->pio_data);
2545}
Clone of davem's net-next-2.6 tree