?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.31.6
[linux/fpc-iii.git] / arch / x86 / kvm / x86.c
blob26e454ccd0a842c6c65fa8dd64eeb2ba85134d7f
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 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
15 *
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
18 *
19 */
20
21 #include <linux/kvm_host.h>
22 #include "irq.h"
23 #include "mmu.h"
24 #include "i8254.h"
25 #include "tss.h"
26 #include "kvm_cache_regs.h"
27 #include "x86.h"
28
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
32 #include <linux/fs.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
40
41 #include <asm/uaccess.h>
42 #include <asm/msr.h>
43 #include <asm/desc.h>
44 #include <asm/mtrr.h>
45
46 #define MAX_IO_MSRS 256
47 #define CR0_RESERVED_BITS \
48 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
49 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
50 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
51 #define CR4_RESERVED_BITS \
52 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
53 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
54 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
55 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
56
57 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
58 /* EFER defaults:
59 * - enable syscall per default because its emulated by KVM
60 * - enable LME and LMA per default on 64 bit KVM
61 */
62 #ifdef CONFIG_X86_64
63 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
64 #else
65 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
66 #endif
67
68 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
69 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
70
71 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
72 struct kvm_cpuid_entry2 __user *entries);
73 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
74 u32 function, u32 index);
75
76 struct kvm_x86_ops *kvm_x86_ops;
77 EXPORT_SYMBOL_GPL(kvm_x86_ops);
78
79 struct kvm_stats_debugfs_item debugfs_entries[] = {
80 { "pf_fixed", VCPU_STAT(pf_fixed) },
81 { "pf_guest", VCPU_STAT(pf_guest) },
82 { "tlb_flush", VCPU_STAT(tlb_flush) },
83 { "invlpg", VCPU_STAT(invlpg) },
84 { "exits", VCPU_STAT(exits) },
85 { "io_exits", VCPU_STAT(io_exits) },
86 { "mmio_exits", VCPU_STAT(mmio_exits) },
87 { "signal_exits", VCPU_STAT(signal_exits) },
88 { "irq_window", VCPU_STAT(irq_window_exits) },
89 { "nmi_window", VCPU_STAT(nmi_window_exits) },
90 { "halt_exits", VCPU_STAT(halt_exits) },
91 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
92 { "hypercalls", VCPU_STAT(hypercalls) },
93 { "request_irq", VCPU_STAT(request_irq_exits) },
94 { "irq_exits", VCPU_STAT(irq_exits) },
95 { "host_state_reload", VCPU_STAT(host_state_reload) },
96 { "efer_reload", VCPU_STAT(efer_reload) },
97 { "fpu_reload", VCPU_STAT(fpu_reload) },
98 { "insn_emulation", VCPU_STAT(insn_emulation) },
99 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
100 { "irq_injections", VCPU_STAT(irq_injections) },
101 { "nmi_injections", VCPU_STAT(nmi_injections) },
102 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
103 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
104 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
105 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
106 { "mmu_flooded", VM_STAT(mmu_flooded) },
107 { "mmu_recycled", VM_STAT(mmu_recycled) },
108 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
109 { "mmu_unsync", VM_STAT(mmu_unsync) },
110 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
111 { "largepages", VM_STAT(lpages) },
112 { NULL }
113 };
114
115 unsigned long segment_base(u16 selector)
116 {
117 struct descriptor_table gdt;
118 struct desc_struct *d;
119 unsigned long table_base;
120 unsigned long v;
121
122 if (selector == 0)
123 return 0;
124
125 asm("sgdt %0" : "=m"(gdt));
126 table_base = gdt.base;
127
128 if (selector & 4) { /* from ldt */
129 u16 ldt_selector;
130
131 asm("sldt %0" : "=g"(ldt_selector));
132 table_base = segment_base(ldt_selector);
133 }
134 d = (struct desc_struct *)(table_base + (selector & ~7));
135 v = d->base0 | ((unsigned long)d->base1 << 16) |
136 ((unsigned long)d->base2 << 24);
137 #ifdef CONFIG_X86_64
138 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
139 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
140 #endif
141 return v;
142 }
143 EXPORT_SYMBOL_GPL(segment_base);
144
145 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
146 {
147 if (irqchip_in_kernel(vcpu->kvm))
148 return vcpu->arch.apic_base;
149 else
150 return vcpu->arch.apic_base;
151 }
152 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
153
154 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
155 {
156 /* TODO: reserve bits check */
157 if (irqchip_in_kernel(vcpu->kvm))
158 kvm_lapic_set_base(vcpu, data);
159 else
160 vcpu->arch.apic_base = data;
161 }
162 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
163
164 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
165 {
166 WARN_ON(vcpu->arch.exception.pending);
167 vcpu->arch.exception.pending = true;
168 vcpu->arch.exception.has_error_code = false;
169 vcpu->arch.exception.nr = nr;
170 }
171 EXPORT_SYMBOL_GPL(kvm_queue_exception);
172
173 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
174 u32 error_code)
175 {
176 ++vcpu->stat.pf_guest;
177
178 if (vcpu->arch.exception.pending) {
179 if (vcpu->arch.exception.nr == PF_VECTOR) {
180 printk(KERN_DEBUG "kvm: inject_page_fault:"
181 " double fault 0x%lx\n", addr);
182 vcpu->arch.exception.nr = DF_VECTOR;
183 vcpu->arch.exception.error_code = 0;
184 } else if (vcpu->arch.exception.nr == DF_VECTOR) {
185 /* triple fault -> shutdown */
186 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
187 }
188 return;
189 }
190 vcpu->arch.cr2 = addr;
191 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
192 }
193
194 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
195 {
196 vcpu->arch.nmi_pending = 1;
197 }
198 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
199
200 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
201 {
202 WARN_ON(vcpu->arch.exception.pending);
203 vcpu->arch.exception.pending = true;
204 vcpu->arch.exception.has_error_code = true;
205 vcpu->arch.exception.nr = nr;
206 vcpu->arch.exception.error_code = error_code;
207 }
208 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
209
210 static void __queue_exception(struct kvm_vcpu *vcpu)
211 {
212 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
213 vcpu->arch.exception.has_error_code,
214 vcpu->arch.exception.error_code);
215 }
216
217 /*
218 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
219 * a #GP and return false.
220 */
221 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
222 {
223 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
224 return true;
225 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
226 return false;
227 }
228 EXPORT_SYMBOL_GPL(kvm_require_cpl);
229
230 /*
231 * Load the pae pdptrs. Return true is they are all valid.
232 */
233 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
234 {
235 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
236 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
237 int i;
238 int ret;
239 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
240
241 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
242 offset * sizeof(u64), sizeof(pdpte));
243 if (ret < 0) {
244 ret = 0;
245 goto out;
246 }
247 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
248 if (is_present_pte(pdpte[i]) &&
249 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
250 ret = 0;
251 goto out;
252 }
253 }
254 ret = 1;
255
256 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
257 out:
258
259 return ret;
260 }
261 EXPORT_SYMBOL_GPL(load_pdptrs);
262
263 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
264 {
265 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
266 bool changed = true;
267 int r;
268
269 if (is_long_mode(vcpu) || !is_pae(vcpu))
270 return false;
271
272 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
273 if (r < 0)
274 goto out;
275 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
276 out:
277
278 return changed;
279 }
280
281 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
282 {
283 if (cr0 & CR0_RESERVED_BITS) {
284 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
285 cr0, vcpu->arch.cr0);
286 kvm_inject_gp(vcpu, 0);
287 return;
288 }
289
290 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
291 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
292 kvm_inject_gp(vcpu, 0);
293 return;
294 }
295
296 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
297 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
298 "and a clear PE flag\n");
299 kvm_inject_gp(vcpu, 0);
300 return;
301 }
302
303 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
304 #ifdef CONFIG_X86_64
305 if ((vcpu->arch.shadow_efer & EFER_LME)) {
306 int cs_db, cs_l;
307
308 if (!is_pae(vcpu)) {
309 printk(KERN_DEBUG "set_cr0: #GP, start paging "
310 "in long mode while PAE is disabled\n");
311 kvm_inject_gp(vcpu, 0);
312 return;
313 }
314 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
315 if (cs_l) {
316 printk(KERN_DEBUG "set_cr0: #GP, start paging "
317 "in long mode while CS.L == 1\n");
318 kvm_inject_gp(vcpu, 0);
319 return;
320
321 }
322 } else
323 #endif
324 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
325 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
326 "reserved bits\n");
327 kvm_inject_gp(vcpu, 0);
328 return;
329 }
330
331 }
332
333 kvm_x86_ops->set_cr0(vcpu, cr0);
334 vcpu->arch.cr0 = cr0;
335
336 kvm_mmu_reset_context(vcpu);
337 return;
338 }
339 EXPORT_SYMBOL_GPL(kvm_set_cr0);
340
341 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
342 {
343 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
344 KVMTRACE_1D(LMSW, vcpu,
345 (u32)((vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f)),
346 handler);
347 }
348 EXPORT_SYMBOL_GPL(kvm_lmsw);
349
350 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
351 {
352 unsigned long old_cr4 = vcpu->arch.cr4;
353 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
354
355 if (cr4 & CR4_RESERVED_BITS) {
356 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
357 kvm_inject_gp(vcpu, 0);
358 return;
359 }
360
361 if (is_long_mode(vcpu)) {
362 if (!(cr4 & X86_CR4_PAE)) {
363 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
364 "in long mode\n");
365 kvm_inject_gp(vcpu, 0);
366 return;
367 }
368 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
369 && ((cr4 ^ old_cr4) & pdptr_bits)
370 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
371 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
372 kvm_inject_gp(vcpu, 0);
373 return;
374 }
375
376 if (cr4 & X86_CR4_VMXE) {
377 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
378 kvm_inject_gp(vcpu, 0);
379 return;
380 }
381 kvm_x86_ops->set_cr4(vcpu, cr4);
382 vcpu->arch.cr4 = cr4;
383 vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
384 kvm_mmu_reset_context(vcpu);
385 }
386 EXPORT_SYMBOL_GPL(kvm_set_cr4);
387
388 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
389 {
390 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
391 kvm_mmu_sync_roots(vcpu);
392 kvm_mmu_flush_tlb(vcpu);
393 return;
394 }
395
396 if (is_long_mode(vcpu)) {
397 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
398 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
399 kvm_inject_gp(vcpu, 0);
400 return;
401 }
402 } else {
403 if (is_pae(vcpu)) {
404 if (cr3 & CR3_PAE_RESERVED_BITS) {
405 printk(KERN_DEBUG
406 "set_cr3: #GP, reserved bits\n");
407 kvm_inject_gp(vcpu, 0);
408 return;
409 }
410 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
411 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
412 "reserved bits\n");
413 kvm_inject_gp(vcpu, 0);
414 return;
415 }
416 }
417 /*
418 * We don't check reserved bits in nonpae mode, because
419 * this isn't enforced, and VMware depends on this.
420 */
421 }
422
423 /*
424 * Does the new cr3 value map to physical memory? (Note, we
425 * catch an invalid cr3 even in real-mode, because it would
426 * cause trouble later on when we turn on paging anyway.)
427 *
428 * A real CPU would silently accept an invalid cr3 and would
429 * attempt to use it - with largely undefined (and often hard
430 * to debug) behavior on the guest side.
431 */
432 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
433 kvm_inject_gp(vcpu, 0);
434 else {
435 vcpu->arch.cr3 = cr3;
436 vcpu->arch.mmu.new_cr3(vcpu);
437 }
438 }
439 EXPORT_SYMBOL_GPL(kvm_set_cr3);
440
441 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
442 {
443 if (cr8 & CR8_RESERVED_BITS) {
444 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
445 kvm_inject_gp(vcpu, 0);
446 return;
447 }
448 if (irqchip_in_kernel(vcpu->kvm))
449 kvm_lapic_set_tpr(vcpu, cr8);
450 else
451 vcpu->arch.cr8 = cr8;
452 }
453 EXPORT_SYMBOL_GPL(kvm_set_cr8);
454
455 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
456 {
457 if (irqchip_in_kernel(vcpu->kvm))
458 return kvm_lapic_get_cr8(vcpu);
459 else
460 return vcpu->arch.cr8;
461 }
462 EXPORT_SYMBOL_GPL(kvm_get_cr8);
463
464 static inline u32 bit(int bitno)
465 {
466 return 1 << (bitno & 31);
467 }
468
469 /*
470 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
471 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
472 *
473 * This list is modified at module load time to reflect the
474 * capabilities of the host cpu.
475 */
476 static u32 msrs_to_save[] = {
477 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
478 MSR_K6_STAR,
479 #ifdef CONFIG_X86_64
480 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
481 #endif
482 MSR_IA32_TIME_STAMP_COUNTER, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
483 MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
484 };
485
486 static unsigned num_msrs_to_save;
487
488 static u32 emulated_msrs[] = {
489 MSR_IA32_MISC_ENABLE,
490 };
491
492 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
493 {
494 if (efer & efer_reserved_bits) {
495 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
496 efer);
497 kvm_inject_gp(vcpu, 0);
498 return;
499 }
500
501 if (is_paging(vcpu)
502 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
503 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
504 kvm_inject_gp(vcpu, 0);
505 return;
506 }
507
508 if (efer & EFER_FFXSR) {
509 struct kvm_cpuid_entry2 *feat;
510
511 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
512 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
513 printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
514 kvm_inject_gp(vcpu, 0);
515 return;
516 }
517 }
518
519 if (efer & EFER_SVME) {
520 struct kvm_cpuid_entry2 *feat;
521
522 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
523 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
524 printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
525 kvm_inject_gp(vcpu, 0);
526 return;
527 }
528 }
529
530 kvm_x86_ops->set_efer(vcpu, efer);
531
532 efer &= ~EFER_LMA;
533 efer |= vcpu->arch.shadow_efer & EFER_LMA;
534
535 vcpu->arch.shadow_efer = efer;
536
537 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
538 kvm_mmu_reset_context(vcpu);
539 }
540
541 void kvm_enable_efer_bits(u64 mask)
542 {
543 efer_reserved_bits &= ~mask;
544 }
545 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
546
547
548 /*
549 * Writes msr value into into the appropriate "register".
550 * Returns 0 on success, non-0 otherwise.
551 * Assumes vcpu_load() was already called.
552 */
553 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
554 {
555 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
556 }
557
558 /*
559 * Adapt set_msr() to msr_io()'s calling convention
560 */
561 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
562 {
563 return kvm_set_msr(vcpu, index, *data);
564 }
565
566 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
567 {
568 static int version;
569 struct pvclock_wall_clock wc;
570 struct timespec now, sys, boot;
571
572 if (!wall_clock)
573 return;
574
575 version++;
576
577 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
578
579 /*
580 * The guest calculates current wall clock time by adding
581 * system time (updated by kvm_write_guest_time below) to the
582 * wall clock specified here. guest system time equals host
583 * system time for us, thus we must fill in host boot time here.
584 */
585 now = current_kernel_time();
586 ktime_get_ts(&sys);
587 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
588
589 wc.sec = boot.tv_sec;
590 wc.nsec = boot.tv_nsec;
591 wc.version = version;
592
593 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
594
595 version++;
596 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
597 }
598
599 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
600 {
601 uint32_t quotient, remainder;
602
603 /* Don't try to replace with do_div(), this one calculates
604 * "(dividend << 32) / divisor" */
605 __asm__ ( "divl %4"
606 : "=a" (quotient), "=d" (remainder)
607 : "0" (0), "1" (dividend), "r" (divisor) );
608 return quotient;
609 }
610
611 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
612 {
613 uint64_t nsecs = 1000000000LL;
614 int32_t shift = 0;
615 uint64_t tps64;
616 uint32_t tps32;
617
618 tps64 = tsc_khz * 1000LL;
619 while (tps64 > nsecs*2) {
620 tps64 >>= 1;
621 shift--;
622 }
623
624 tps32 = (uint32_t)tps64;
625 while (tps32 <= (uint32_t)nsecs) {
626 tps32 <<= 1;
627 shift++;
628 }
629
630 hv_clock->tsc_shift = shift;
631 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
632
633 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
634 __func__, tsc_khz, hv_clock->tsc_shift,
635 hv_clock->tsc_to_system_mul);
636 }
637
638 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
639
640 static void kvm_write_guest_time(struct kvm_vcpu *v)
641 {
642 struct timespec ts;
643 unsigned long flags;
644 struct kvm_vcpu_arch *vcpu = &v->arch;
645 void *shared_kaddr;
646 unsigned long this_tsc_khz;
647
648 if ((!vcpu->time_page))
649 return;
650
651 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
652 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
653 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
654 vcpu->hv_clock_tsc_khz = this_tsc_khz;
655 }
656 put_cpu_var(cpu_tsc_khz);
657
658 /* Keep irq disabled to prevent changes to the clock */
659 local_irq_save(flags);
660 kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
661 &vcpu->hv_clock.tsc_timestamp);
662 ktime_get_ts(&ts);
663 local_irq_restore(flags);
664
665 /* With all the info we got, fill in the values */
666
667 vcpu->hv_clock.system_time = ts.tv_nsec +
668 (NSEC_PER_SEC * (u64)ts.tv_sec);
669 /*
670 * The interface expects us to write an even number signaling that the
671 * update is finished. Since the guest won't see the intermediate
672 * state, we just increase by 2 at the end.
673 */
674 vcpu->hv_clock.version += 2;
675
676 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
677
678 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
679 sizeof(vcpu->hv_clock));
680
681 kunmap_atomic(shared_kaddr, KM_USER0);
682
683 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
684 }
685
686 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
687 {
688 struct kvm_vcpu_arch *vcpu = &v->arch;
689
690 if (!vcpu->time_page)
691 return 0;
692 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
693 return 1;
694 }
695
696 static bool msr_mtrr_valid(unsigned msr)
697 {
698 switch (msr) {
699 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
700 case MSR_MTRRfix64K_00000:
701 case MSR_MTRRfix16K_80000:
702 case MSR_MTRRfix16K_A0000:
703 case MSR_MTRRfix4K_C0000:
704 case MSR_MTRRfix4K_C8000:
705 case MSR_MTRRfix4K_D0000:
706 case MSR_MTRRfix4K_D8000:
707 case MSR_MTRRfix4K_E0000:
708 case MSR_MTRRfix4K_E8000:
709 case MSR_MTRRfix4K_F0000:
710 case MSR_MTRRfix4K_F8000:
711 case MSR_MTRRdefType:
712 case MSR_IA32_CR_PAT:
713 return true;
714 case 0x2f8:
715 return true;
716 }
717 return false;
718 }
719
720 static bool valid_pat_type(unsigned t)
721 {
722 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
723 }
724
725 static bool valid_mtrr_type(unsigned t)
726 {
727 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
728 }
729
730 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
731 {
732 int i;
733
734 if (!msr_mtrr_valid(msr))
735 return false;
736
737 if (msr == MSR_IA32_CR_PAT) {
738 for (i = 0; i < 8; i++)
739 if (!valid_pat_type((data >> (i * 8)) & 0xff))
740 return false;
741 return true;
742 } else if (msr == MSR_MTRRdefType) {
743 if (data & ~0xcff)
744 return false;
745 return valid_mtrr_type(data & 0xff);
746 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
747 for (i = 0; i < 8 ; i++)
748 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
749 return false;
750 return true;
751 }
752
753 /* variable MTRRs */
754 return valid_mtrr_type(data & 0xff);
755 }
756
757 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
758 {
759 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
760
761 if (!mtrr_valid(vcpu, msr, data))
762 return 1;
763
764 if (msr == MSR_MTRRdefType) {
765 vcpu->arch.mtrr_state.def_type = data;
766 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
767 } else if (msr == MSR_MTRRfix64K_00000)
768 p[0] = data;
769 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
770 p[1 + msr - MSR_MTRRfix16K_80000] = data;
771 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
772 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
773 else if (msr == MSR_IA32_CR_PAT)
774 vcpu->arch.pat = data;
775 else { /* Variable MTRRs */
776 int idx, is_mtrr_mask;
777 u64 *pt;
778
779 idx = (msr - 0x200) / 2;
780 is_mtrr_mask = msr - 0x200 - 2 * idx;
781 if (!is_mtrr_mask)
782 pt =
783 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
784 else
785 pt =
786 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
787 *pt = data;
788 }
789
790 kvm_mmu_reset_context(vcpu);
791 return 0;
792 }
793
794 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
795 {
796 switch (msr) {
797 case MSR_EFER:
798 set_efer(vcpu, data);
799 break;
800 case MSR_IA32_MC0_STATUS:
801 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
802 __func__, data);
803 break;
804 case MSR_IA32_MCG_STATUS:
805 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
806 __func__, data);
807 break;
808 case MSR_IA32_MCG_CTL:
809 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_CTL 0x%llx, nop\n",
810 __func__, data);
811 break;
812 case MSR_IA32_DEBUGCTLMSR:
813 if (!data) {
814 /* We support the non-activated case already */
815 break;
816 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
817 /* Values other than LBR and BTF are vendor-specific,
818 thus reserved and should throw a #GP */
819 return 1;
820 }
821 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
822 __func__, data);
823 break;
824 case MSR_IA32_UCODE_REV:
825 case MSR_IA32_UCODE_WRITE:
826 case MSR_VM_HSAVE_PA:
827 break;
828 case 0x200 ... 0x2ff:
829 return set_msr_mtrr(vcpu, msr, data);
830 case MSR_IA32_APICBASE:
831 kvm_set_apic_base(vcpu, data);
832 break;
833 case MSR_IA32_MISC_ENABLE:
834 vcpu->arch.ia32_misc_enable_msr = data;
835 break;
836 case MSR_KVM_WALL_CLOCK:
837 vcpu->kvm->arch.wall_clock = data;
838 kvm_write_wall_clock(vcpu->kvm, data);
839 break;
840 case MSR_KVM_SYSTEM_TIME: {
841 if (vcpu->arch.time_page) {
842 kvm_release_page_dirty(vcpu->arch.time_page);
843 vcpu->arch.time_page = NULL;
844 }
845
846 vcpu->arch.time = data;
847
848 /* we verify if the enable bit is set... */
849 if (!(data & 1))
850 break;
851
852 /* ...but clean it before doing the actual write */
853 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
854
855 vcpu->arch.time_page =
856 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
857
858 if (is_error_page(vcpu->arch.time_page)) {
859 kvm_release_page_clean(vcpu->arch.time_page);
860 vcpu->arch.time_page = NULL;
861 }
862
863 kvm_request_guest_time_update(vcpu);
864 break;
865 }
866 default:
867 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
868 return 1;
869 }
870 return 0;
871 }
872 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
873
874
875 /*
876 * Reads an msr value (of 'msr_index') into 'pdata'.
877 * Returns 0 on success, non-0 otherwise.
878 * Assumes vcpu_load() was already called.
879 */
880 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
881 {
882 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
883 }
884
885 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
886 {
887 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
888
889 if (!msr_mtrr_valid(msr))
890 return 1;
891
892 if (msr == MSR_MTRRdefType)
893 *pdata = vcpu->arch.mtrr_state.def_type +
894 (vcpu->arch.mtrr_state.enabled << 10);
895 else if (msr == MSR_MTRRfix64K_00000)
896 *pdata = p[0];
897 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
898 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
899 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
900 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
901 else if (msr == MSR_IA32_CR_PAT)
902 *pdata = vcpu->arch.pat;
903 else { /* Variable MTRRs */
904 int idx, is_mtrr_mask;
905 u64 *pt;
906
907 idx = (msr - 0x200) / 2;
908 is_mtrr_mask = msr - 0x200 - 2 * idx;
909 if (!is_mtrr_mask)
910 pt =
911 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
912 else
913 pt =
914 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
915 *pdata = *pt;
916 }
917
918 return 0;
919 }
920
921 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
922 {
923 u64 data;
924
925 switch (msr) {
926 case 0xc0010010: /* SYSCFG */
927 case 0xc0010015: /* HWCR */
928 case MSR_IA32_PLATFORM_ID:
929 case MSR_IA32_P5_MC_ADDR:
930 case MSR_IA32_P5_MC_TYPE:
931 case MSR_IA32_MC0_CTL:
932 case MSR_IA32_MCG_STATUS:
933 case MSR_IA32_MCG_CAP:
934 case MSR_IA32_MCG_CTL:
935 case MSR_IA32_MC0_MISC:
936 case MSR_IA32_MC0_MISC+4:
937 case MSR_IA32_MC0_MISC+8:
938 case MSR_IA32_MC0_MISC+12:
939 case MSR_IA32_MC0_MISC+16:
940 case MSR_IA32_MC0_MISC+20:
941 case MSR_IA32_UCODE_REV:
942 case MSR_IA32_EBL_CR_POWERON:
943 case MSR_IA32_DEBUGCTLMSR:
944 case MSR_IA32_LASTBRANCHFROMIP:
945 case MSR_IA32_LASTBRANCHTOIP:
946 case MSR_IA32_LASTINTFROMIP:
947 case MSR_IA32_LASTINTTOIP:
948 case MSR_VM_HSAVE_PA:
949 case MSR_P6_EVNTSEL0:
950 case MSR_P6_EVNTSEL1:
951 case MSR_K7_EVNTSEL0:
952 case MSR_K8_INT_PENDING_MSG:
953 data = 0;
954 break;
955 case MSR_MTRRcap:
956 data = 0x500 | KVM_NR_VAR_MTRR;
957 break;
958 case 0x200 ... 0x2ff:
959 return get_msr_mtrr(vcpu, msr, pdata);
960 case 0xcd: /* fsb frequency */
961 data = 3;
962 break;
963 case MSR_IA32_APICBASE:
964 data = kvm_get_apic_base(vcpu);
965 break;
966 case MSR_IA32_MISC_ENABLE:
967 data = vcpu->arch.ia32_misc_enable_msr;
968 break;
969 case MSR_IA32_PERF_STATUS:
970 /* TSC increment by tick */
971 data = 1000ULL;
972 /* CPU multiplier */
973 data |= (((uint64_t)4ULL) << 40);
974 break;
975 case MSR_EFER:
976 data = vcpu->arch.shadow_efer;
977 break;
978 case MSR_KVM_WALL_CLOCK:
979 data = vcpu->kvm->arch.wall_clock;
980 break;
981 case MSR_KVM_SYSTEM_TIME:
982 data = vcpu->arch.time;
983 break;
984 default:
985 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
986 return 1;
987 }
988 *pdata = data;
989 return 0;
990 }
991 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
992
993 /*
994 * Read or write a bunch of msrs. All parameters are kernel addresses.
995 *
996 * @return number of msrs set successfully.
997 */
998 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
999 struct kvm_msr_entry *entries,
1000 int (*do_msr)(struct kvm_vcpu *vcpu,
1001 unsigned index, u64 *data))
1002 {
1003 int i;
1004
1005 vcpu_load(vcpu);
1006
1007 down_read(&vcpu->kvm->slots_lock);
1008 for (i = 0; i < msrs->nmsrs; ++i)
1009 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1010 break;
1011 up_read(&vcpu->kvm->slots_lock);
1012
1013 vcpu_put(vcpu);
1014
1015 return i;
1016 }
1017
1018 /*
1019 * Read or write a bunch of msrs. Parameters are user addresses.
1020 *
1021 * @return number of msrs set successfully.
1022 */
1023 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1024 int (*do_msr)(struct kvm_vcpu *vcpu,
1025 unsigned index, u64 *data),
1026 int writeback)
1027 {
1028 struct kvm_msrs msrs;
1029 struct kvm_msr_entry *entries;
1030 int r, n;
1031 unsigned size;
1032
1033 r = -EFAULT;
1034 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1035 goto out;
1036
1037 r = -E2BIG;
1038 if (msrs.nmsrs >= MAX_IO_MSRS)
1039 goto out;
1040
1041 r = -ENOMEM;
1042 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1043 entries = vmalloc(size);
1044 if (!entries)
1045 goto out;
1046
1047 r = -EFAULT;
1048 if (copy_from_user(entries, user_msrs->entries, size))
1049 goto out_free;
1050
1051 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1052 if (r < 0)
1053 goto out_free;
1054
1055 r = -EFAULT;
1056 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1057 goto out_free;
1058
1059 r = n;
1060
1061 out_free:
1062 vfree(entries);
1063 out:
1064 return r;
1065 }
1066
1067 int kvm_dev_ioctl_check_extension(long ext)
1068 {
1069 int r;
1070
1071 switch (ext) {
1072 case KVM_CAP_IRQCHIP:
1073 case KVM_CAP_HLT:
1074 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1075 case KVM_CAP_SET_TSS_ADDR:
1076 case KVM_CAP_EXT_CPUID:
1077 case KVM_CAP_CLOCKSOURCE:
1078 case KVM_CAP_PIT:
1079 case KVM_CAP_NOP_IO_DELAY:
1080 case KVM_CAP_MP_STATE:
1081 case KVM_CAP_SYNC_MMU:
1082 case KVM_CAP_REINJECT_CONTROL:
1083 case KVM_CAP_IRQ_INJECT_STATUS:
1084 case KVM_CAP_ASSIGN_DEV_IRQ:
1085 r = 1;
1086 break;
1087 case KVM_CAP_COALESCED_MMIO:
1088 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1089 break;
1090 case KVM_CAP_VAPIC:
1091 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1092 break;
1093 case KVM_CAP_NR_VCPUS:
1094 r = KVM_MAX_VCPUS;
1095 break;
1096 case KVM_CAP_NR_MEMSLOTS:
1097 r = KVM_MEMORY_SLOTS;
1098 break;
1099 case KVM_CAP_PV_MMU:
1100 r = !tdp_enabled;
1101 break;
1102 case KVM_CAP_IOMMU:
1103 r = iommu_found();
1104 break;
1105 default:
1106 r = 0;
1107 break;
1108 }
1109 return r;
1110
1111 }
1112
1113 long kvm_arch_dev_ioctl(struct file *filp,
1114 unsigned int ioctl, unsigned long arg)
1115 {
1116 void __user *argp = (void __user *)arg;
1117 long r;
1118
1119 switch (ioctl) {
1120 case KVM_GET_MSR_INDEX_LIST: {
1121 struct kvm_msr_list __user *user_msr_list = argp;
1122 struct kvm_msr_list msr_list;
1123 unsigned n;
1124
1125 r = -EFAULT;
1126 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1127 goto out;
1128 n = msr_list.nmsrs;
1129 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1130 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1131 goto out;
1132 r = -E2BIG;
1133 if (n < msr_list.nmsrs)
1134 goto out;
1135 r = -EFAULT;
1136 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1137 num_msrs_to_save * sizeof(u32)))
1138 goto out;
1139 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1140 &emulated_msrs,
1141 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1142 goto out;
1143 r = 0;
1144 break;
1145 }
1146 case KVM_GET_SUPPORTED_CPUID: {
1147 struct kvm_cpuid2 __user *cpuid_arg = argp;
1148 struct kvm_cpuid2 cpuid;
1149
1150 r = -EFAULT;
1151 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1152 goto out;
1153 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1154 cpuid_arg->entries);
1155 if (r)
1156 goto out;
1157
1158 r = -EFAULT;
1159 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1160 goto out;
1161 r = 0;
1162 break;
1163 }
1164 default:
1165 r = -EINVAL;
1166 }
1167 out:
1168 return r;
1169 }
1170
1171 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1172 {
1173 kvm_x86_ops->vcpu_load(vcpu, cpu);
1174 kvm_request_guest_time_update(vcpu);
1175 }
1176
1177 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1178 {
1179 kvm_x86_ops->vcpu_put(vcpu);
1180 kvm_put_guest_fpu(vcpu);
1181 }
1182
1183 static int is_efer_nx(void)
1184 {
1185 unsigned long long efer = 0;
1186
1187 rdmsrl_safe(MSR_EFER, &efer);
1188 return efer & EFER_NX;
1189 }
1190
1191 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1192 {
1193 int i;
1194 struct kvm_cpuid_entry2 *e, *entry;
1195
1196 entry = NULL;
1197 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1198 e = &vcpu->arch.cpuid_entries[i];
1199 if (e->function == 0x80000001) {
1200 entry = e;
1201 break;
1202 }
1203 }
1204 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1205 entry->edx &= ~(1 << 20);
1206 printk(KERN_INFO "kvm: guest NX capability removed\n");
1207 }
1208 }
1209
1210 /* when an old userspace process fills a new kernel module */
1211 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1212 struct kvm_cpuid *cpuid,
1213 struct kvm_cpuid_entry __user *entries)
1214 {
1215 int r, i;
1216 struct kvm_cpuid_entry *cpuid_entries;
1217
1218 r = -E2BIG;
1219 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1220 goto out;
1221 r = -ENOMEM;
1222 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1223 if (!cpuid_entries)
1224 goto out;
1225 r = -EFAULT;
1226 if (copy_from_user(cpuid_entries, entries,
1227 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1228 goto out_free;
1229 for (i = 0; i < cpuid->nent; i++) {
1230 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1231 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1232 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1233 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1234 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1235 vcpu->arch.cpuid_entries[i].index = 0;
1236 vcpu->arch.cpuid_entries[i].flags = 0;
1237 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1238 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1239 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1240 }
1241 vcpu->arch.cpuid_nent = cpuid->nent;
1242 cpuid_fix_nx_cap(vcpu);
1243 r = 0;
1244
1245 out_free:
1246 vfree(cpuid_entries);
1247 out:
1248 return r;
1249 }
1250
1251 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1252 struct kvm_cpuid2 *cpuid,
1253 struct kvm_cpuid_entry2 __user *entries)
1254 {
1255 int r;
1256
1257 r = -E2BIG;
1258 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1259 goto out;
1260 r = -EFAULT;
1261 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1262 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1263 goto out;
1264 vcpu->arch.cpuid_nent = cpuid->nent;
1265 return 0;
1266
1267 out:
1268 return r;
1269 }
1270
1271 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1272 struct kvm_cpuid2 *cpuid,
1273 struct kvm_cpuid_entry2 __user *entries)
1274 {
1275 int r;
1276
1277 r = -E2BIG;
1278 if (cpuid->nent < vcpu->arch.cpuid_nent)
1279 goto out;
1280 r = -EFAULT;
1281 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1282 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1283 goto out;
1284 return 0;
1285
1286 out:
1287 cpuid->nent = vcpu->arch.cpuid_nent;
1288 return r;
1289 }
1290
1291 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1292 u32 index)
1293 {
1294 entry->function = function;
1295 entry->index = index;
1296 cpuid_count(entry->function, entry->index,
1297 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1298 entry->flags = 0;
1299 }
1300
1301 #define F(x) bit(X86_FEATURE_##x)
1302
1303 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1304 u32 index, int *nent, int maxnent)
1305 {
1306 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1307 #ifdef CONFIG_X86_64
1308 unsigned f_lm = F(LM);
1309 #else
1310 unsigned f_lm = 0;
1311 #endif
1312
1313 /* cpuid 1.edx */
1314 const u32 kvm_supported_word0_x86_features =
1315 F(FPU) | F(VME) | F(DE) | F(PSE) |
1316 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1317 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1318 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1319 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1320 0 /* Reserved, DS, ACPI */ | F(MMX) |
1321 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1322 0 /* HTT, TM, Reserved, PBE */;
1323 /* cpuid 0x80000001.edx */
1324 const u32 kvm_supported_word1_x86_features =
1325 F(FPU) | F(VME) | F(DE) | F(PSE) |
1326 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1327 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1328 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1329 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1330 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1331 F(FXSR) | F(FXSR_OPT) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
1332 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1333 /* cpuid 1.ecx */
1334 const u32 kvm_supported_word4_x86_features =
1335 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1336 0 /* DS-CPL, VMX, SMX, EST */ |
1337 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1338 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1339 0 /* Reserved, DCA */ | F(XMM4_1) |
1340 F(XMM4_2) | 0 /* x2APIC */ | F(MOVBE) | F(POPCNT) |
1341 0 /* Reserved, XSAVE, OSXSAVE */;
1342 /* cpuid 0x80000001.ecx */
1343 const u32 kvm_supported_word6_x86_features =
1344 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1345 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1346 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1347 0 /* SKINIT */ | 0 /* WDT */;
1348
1349 /* all calls to cpuid_count() should be made on the same cpu */
1350 get_cpu();
1351 do_cpuid_1_ent(entry, function, index);
1352 ++*nent;
1353
1354 switch (function) {
1355 case 0:
1356 entry->eax = min(entry->eax, (u32)0xb);
1357 break;
1358 case 1:
1359 entry->edx &= kvm_supported_word0_x86_features;
1360 entry->ecx &= kvm_supported_word4_x86_features;
1361 break;
1362 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1363 * may return different values. This forces us to get_cpu() before
1364 * issuing the first command, and also to emulate this annoying behavior
1365 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1366 case 2: {
1367 int t, times = entry->eax & 0xff;
1368
1369 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1370 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1371 for (t = 1; t < times && *nent < maxnent; ++t) {
1372 do_cpuid_1_ent(&entry[t], function, 0);
1373 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1374 ++*nent;
1375 }
1376 break;
1377 }
1378 /* function 4 and 0xb have additional index. */
1379 case 4: {
1380 int i, cache_type;
1381
1382 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1383 /* read more entries until cache_type is zero */
1384 for (i = 1; *nent < maxnent; ++i) {
1385 cache_type = entry[i - 1].eax & 0x1f;
1386 if (!cache_type)
1387 break;
1388 do_cpuid_1_ent(&entry[i], function, i);
1389 entry[i].flags |=
1390 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1391 ++*nent;
1392 }
1393 break;
1394 }
1395 case 0xb: {
1396 int i, level_type;
1397
1398 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1399 /* read more entries until level_type is zero */
1400 for (i = 1; *nent < maxnent; ++i) {
1401 level_type = entry[i - 1].ecx & 0xff00;
1402 if (!level_type)
1403 break;
1404 do_cpuid_1_ent(&entry[i], function, i);
1405 entry[i].flags |=
1406 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1407 ++*nent;
1408 }
1409 break;
1410 }
1411 case 0x80000000:
1412 entry->eax = min(entry->eax, 0x8000001a);
1413 break;
1414 case 0x80000001:
1415 entry->edx &= kvm_supported_word1_x86_features;
1416 entry->ecx &= kvm_supported_word6_x86_features;
1417 break;
1418 }
1419 put_cpu();
1420 }
1421
1422 #undef F
1423
1424 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1425 struct kvm_cpuid_entry2 __user *entries)
1426 {
1427 struct kvm_cpuid_entry2 *cpuid_entries;
1428 int limit, nent = 0, r = -E2BIG;
1429 u32 func;
1430
1431 if (cpuid->nent < 1)
1432 goto out;
1433 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1434 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1435 r = -ENOMEM;
1436 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1437 if (!cpuid_entries)
1438 goto out;
1439
1440 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1441 limit = cpuid_entries[0].eax;
1442 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1443 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1444 &nent, cpuid->nent);
1445 r = -E2BIG;
1446 if (nent >= cpuid->nent)
1447 goto out_free;
1448
1449 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1450 limit = cpuid_entries[nent - 1].eax;
1451 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1452 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1453 &nent, cpuid->nent);
1454 r = -E2BIG;
1455 if (nent >= cpuid->nent)
1456 goto out_free;
1457
1458 r = -EFAULT;
1459 if (copy_to_user(entries, cpuid_entries,
1460 nent * sizeof(struct kvm_cpuid_entry2)))
1461 goto out_free;
1462 cpuid->nent = nent;
1463 r = 0;
1464
1465 out_free:
1466 vfree(cpuid_entries);
1467 out:
1468 return r;
1469 }
1470
1471 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1472 struct kvm_lapic_state *s)
1473 {
1474 vcpu_load(vcpu);
1475 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1476 vcpu_put(vcpu);
1477
1478 return 0;
1479 }
1480
1481 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1482 struct kvm_lapic_state *s)
1483 {
1484 vcpu_load(vcpu);
1485 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1486 kvm_apic_post_state_restore(vcpu);
1487 vcpu_put(vcpu);
1488
1489 return 0;
1490 }
1491
1492 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1493 struct kvm_interrupt *irq)
1494 {
1495 if (irq->irq < 0 || irq->irq >= 256)
1496 return -EINVAL;
1497 if (irqchip_in_kernel(vcpu->kvm))
1498 return -ENXIO;
1499 vcpu_load(vcpu);
1500
1501 kvm_queue_interrupt(vcpu, irq->irq, false);
1502
1503 vcpu_put(vcpu);
1504
1505 return 0;
1506 }
1507
1508 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1509 {
1510 vcpu_load(vcpu);
1511 kvm_inject_nmi(vcpu);
1512 vcpu_put(vcpu);
1513
1514 return 0;
1515 }
1516
1517 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1518 struct kvm_tpr_access_ctl *tac)
1519 {
1520 if (tac->flags)
1521 return -EINVAL;
1522 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1523 return 0;
1524 }
1525
1526 long kvm_arch_vcpu_ioctl(struct file *filp,
1527 unsigned int ioctl, unsigned long arg)
1528 {
1529 struct kvm_vcpu *vcpu = filp->private_data;
1530 void __user *argp = (void __user *)arg;
1531 int r;
1532 struct kvm_lapic_state *lapic = NULL;
1533
1534 switch (ioctl) {
1535 case KVM_GET_LAPIC: {
1536 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1537
1538 r = -ENOMEM;
1539 if (!lapic)
1540 goto out;
1541 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1542 if (r)
1543 goto out;
1544 r = -EFAULT;
1545 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1546 goto out;
1547 r = 0;
1548 break;
1549 }
1550 case KVM_SET_LAPIC: {
1551 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1552 r = -ENOMEM;
1553 if (!lapic)
1554 goto out;
1555 r = -EFAULT;
1556 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1557 goto out;
1558 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1559 if (r)
1560 goto out;
1561 r = 0;
1562 break;
1563 }
1564 case KVM_INTERRUPT: {
1565 struct kvm_interrupt irq;
1566
1567 r = -EFAULT;
1568 if (copy_from_user(&irq, argp, sizeof irq))
1569 goto out;
1570 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1571 if (r)
1572 goto out;
1573 r = 0;
1574 break;
1575 }
1576 case KVM_NMI: {
1577 r = kvm_vcpu_ioctl_nmi(vcpu);
1578 if (r)
1579 goto out;
1580 r = 0;
1581 break;
1582 }
1583 case KVM_SET_CPUID: {
1584 struct kvm_cpuid __user *cpuid_arg = argp;
1585 struct kvm_cpuid cpuid;
1586
1587 r = -EFAULT;
1588 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1589 goto out;
1590 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1591 if (r)
1592 goto out;
1593 break;
1594 }
1595 case KVM_SET_CPUID2: {
1596 struct kvm_cpuid2 __user *cpuid_arg = argp;
1597 struct kvm_cpuid2 cpuid;
1598
1599 r = -EFAULT;
1600 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1601 goto out;
1602 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1603 cpuid_arg->entries);
1604 if (r)
1605 goto out;
1606 break;
1607 }
1608 case KVM_GET_CPUID2: {
1609 struct kvm_cpuid2 __user *cpuid_arg = argp;
1610 struct kvm_cpuid2 cpuid;
1611
1612 r = -EFAULT;
1613 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1614 goto out;
1615 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1616 cpuid_arg->entries);
1617 if (r)
1618 goto out;
1619 r = -EFAULT;
1620 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1621 goto out;
1622 r = 0;
1623 break;
1624 }
1625 case KVM_GET_MSRS:
1626 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1627 break;
1628 case KVM_SET_MSRS:
1629 r = msr_io(vcpu, argp, do_set_msr, 0);
1630 break;
1631 case KVM_TPR_ACCESS_REPORTING: {
1632 struct kvm_tpr_access_ctl tac;
1633
1634 r = -EFAULT;
1635 if (copy_from_user(&tac, argp, sizeof tac))
1636 goto out;
1637 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1638 if (r)
1639 goto out;
1640 r = -EFAULT;
1641 if (copy_to_user(argp, &tac, sizeof tac))
1642 goto out;
1643 r = 0;
1644 break;
1645 };
1646 case KVM_SET_VAPIC_ADDR: {
1647 struct kvm_vapic_addr va;
1648
1649 r = -EINVAL;
1650 if (!irqchip_in_kernel(vcpu->kvm))
1651 goto out;
1652 r = -EFAULT;
1653 if (copy_from_user(&va, argp, sizeof va))
1654 goto out;
1655 r = 0;
1656 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1657 break;
1658 }
1659 default:
1660 r = -EINVAL;
1661 }
1662 out:
1663 kfree(lapic);
1664 return r;
1665 }
1666
1667 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1668 {
1669 int ret;
1670
1671 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1672 return -1;
1673 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1674 return ret;
1675 }
1676
1677 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1678 u32 kvm_nr_mmu_pages)
1679 {
1680 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1681 return -EINVAL;
1682
1683 down_write(&kvm->slots_lock);
1684 spin_lock(&kvm->mmu_lock);
1685
1686 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1687 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1688
1689 spin_unlock(&kvm->mmu_lock);
1690 up_write(&kvm->slots_lock);
1691 return 0;
1692 }
1693
1694 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1695 {
1696 return kvm->arch.n_alloc_mmu_pages;
1697 }
1698
1699 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1700 {
1701 int i;
1702 struct kvm_mem_alias *alias;
1703
1704 for (i = 0; i < kvm->arch.naliases; ++i) {
1705 alias = &kvm->arch.aliases[i];
1706 if (gfn >= alias->base_gfn
1707 && gfn < alias->base_gfn + alias->npages)
1708 return alias->target_gfn + gfn - alias->base_gfn;
1709 }
1710 return gfn;
1711 }
1712
1713 /*
1714 * Set a new alias region. Aliases map a portion of physical memory into
1715 * another portion. This is useful for memory windows, for example the PC
1716 * VGA region.
1717 */
1718 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1719 struct kvm_memory_alias *alias)
1720 {
1721 int r, n;
1722 struct kvm_mem_alias *p;
1723
1724 r = -EINVAL;
1725 /* General sanity checks */
1726 if (alias->memory_size & (PAGE_SIZE - 1))
1727 goto out;
1728 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1729 goto out;
1730 if (alias->slot >= KVM_ALIAS_SLOTS)
1731 goto out;
1732 if (alias->guest_phys_addr + alias->memory_size
1733 < alias->guest_phys_addr)
1734 goto out;
1735 if (alias->target_phys_addr + alias->memory_size
1736 < alias->target_phys_addr)
1737 goto out;
1738
1739 down_write(&kvm->slots_lock);
1740 spin_lock(&kvm->mmu_lock);
1741
1742 p = &kvm->arch.aliases[alias->slot];
1743 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1744 p->npages = alias->memory_size >> PAGE_SHIFT;
1745 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1746
1747 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1748 if (kvm->arch.aliases[n - 1].npages)
1749 break;
1750 kvm->arch.naliases = n;
1751
1752 spin_unlock(&kvm->mmu_lock);
1753 kvm_mmu_zap_all(kvm);
1754
1755 up_write(&kvm->slots_lock);
1756
1757 return 0;
1758
1759 out:
1760 return r;
1761 }
1762
1763 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1764 {
1765 int r;
1766
1767 r = 0;
1768 switch (chip->chip_id) {
1769 case KVM_IRQCHIP_PIC_MASTER:
1770 memcpy(&chip->chip.pic,
1771 &pic_irqchip(kvm)->pics[0],
1772 sizeof(struct kvm_pic_state));
1773 break;
1774 case KVM_IRQCHIP_PIC_SLAVE:
1775 memcpy(&chip->chip.pic,
1776 &pic_irqchip(kvm)->pics[1],
1777 sizeof(struct kvm_pic_state));
1778 break;
1779 case KVM_IRQCHIP_IOAPIC:
1780 memcpy(&chip->chip.ioapic,
1781 ioapic_irqchip(kvm),
1782 sizeof(struct kvm_ioapic_state));
1783 break;
1784 default:
1785 r = -EINVAL;
1786 break;
1787 }
1788 return r;
1789 }
1790
1791 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1792 {
1793 int r;
1794
1795 r = 0;
1796 switch (chip->chip_id) {
1797 case KVM_IRQCHIP_PIC_MASTER:
1798 memcpy(&pic_irqchip(kvm)->pics[0],
1799 &chip->chip.pic,
1800 sizeof(struct kvm_pic_state));
1801 break;
1802 case KVM_IRQCHIP_PIC_SLAVE:
1803 memcpy(&pic_irqchip(kvm)->pics[1],
1804 &chip->chip.pic,
1805 sizeof(struct kvm_pic_state));
1806 break;
1807 case KVM_IRQCHIP_IOAPIC:
1808 memcpy(ioapic_irqchip(kvm),
1809 &chip->chip.ioapic,
1810 sizeof(struct kvm_ioapic_state));
1811 break;
1812 default:
1813 r = -EINVAL;
1814 break;
1815 }
1816 kvm_pic_update_irq(pic_irqchip(kvm));
1817 return r;
1818 }
1819
1820 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1821 {
1822 int r = 0;
1823
1824 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
1825 return r;
1826 }
1827
1828 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
1829 {
1830 int r = 0;
1831
1832 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
1833 kvm_pit_load_count(kvm, 0, ps->channels[0].count);
1834 return r;
1835 }
1836
1837 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
1838 struct kvm_reinject_control *control)
1839 {
1840 if (!kvm->arch.vpit)
1841 return -ENXIO;
1842 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
1843 return 0;
1844 }
1845
1846 /*
1847 * Get (and clear) the dirty memory log for a memory slot.
1848 */
1849 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1850 struct kvm_dirty_log *log)
1851 {
1852 int r;
1853 int n;
1854 struct kvm_memory_slot *memslot;
1855 int is_dirty = 0;
1856
1857 down_write(&kvm->slots_lock);
1858
1859 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1860 if (r)
1861 goto out;
1862
1863 /* If nothing is dirty, don't bother messing with page tables. */
1864 if (is_dirty) {
1865 spin_lock(&kvm->mmu_lock);
1866 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1867 spin_unlock(&kvm->mmu_lock);
1868 kvm_flush_remote_tlbs(kvm);
1869 memslot = &kvm->memslots[log->slot];
1870 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1871 memset(memslot->dirty_bitmap, 0, n);
1872 }
1873 r = 0;
1874 out:
1875 up_write(&kvm->slots_lock);
1876 return r;
1877 }
1878
1879 long kvm_arch_vm_ioctl(struct file *filp,
1880 unsigned int ioctl, unsigned long arg)
1881 {
1882 struct kvm *kvm = filp->private_data;
1883 void __user *argp = (void __user *)arg;
1884 int r = -EINVAL;
1885 /*
1886 * This union makes it completely explicit to gcc-3.x
1887 * that these two variables' stack usage should be
1888 * combined, not added together.
1889 */
1890 union {
1891 struct kvm_pit_state ps;
1892 struct kvm_memory_alias alias;
1893 } u;
1894
1895 switch (ioctl) {
1896 case KVM_SET_TSS_ADDR:
1897 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1898 if (r < 0)
1899 goto out;
1900 break;
1901 case KVM_SET_MEMORY_REGION: {
1902 struct kvm_memory_region kvm_mem;
1903 struct kvm_userspace_memory_region kvm_userspace_mem;
1904
1905 r = -EFAULT;
1906 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1907 goto out;
1908 kvm_userspace_mem.slot = kvm_mem.slot;
1909 kvm_userspace_mem.flags = kvm_mem.flags;
1910 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1911 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1912 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1913 if (r)
1914 goto out;
1915 break;
1916 }
1917 case KVM_SET_NR_MMU_PAGES:
1918 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1919 if (r)
1920 goto out;
1921 break;
1922 case KVM_GET_NR_MMU_PAGES:
1923 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1924 break;
1925 case KVM_SET_MEMORY_ALIAS:
1926 r = -EFAULT;
1927 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1928 goto out;
1929 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1930 if (r)
1931 goto out;
1932 break;
1933 case KVM_CREATE_IRQCHIP:
1934 r = -ENOMEM;
1935 kvm->arch.vpic = kvm_create_pic(kvm);
1936 if (kvm->arch.vpic) {
1937 r = kvm_ioapic_init(kvm);
1938 if (r) {
1939 kfree(kvm->arch.vpic);
1940 kvm->arch.vpic = NULL;
1941 goto out;
1942 }
1943 } else
1944 goto out;
1945 r = kvm_setup_default_irq_routing(kvm);
1946 if (r) {
1947 kfree(kvm->arch.vpic);
1948 kfree(kvm->arch.vioapic);
1949 goto out;
1950 }
1951 break;
1952 case KVM_CREATE_PIT:
1953 mutex_lock(&kvm->lock);
1954 r = -EEXIST;
1955 if (kvm->arch.vpit)
1956 goto create_pit_unlock;
1957 r = -ENOMEM;
1958 kvm->arch.vpit = kvm_create_pit(kvm);
1959 if (kvm->arch.vpit)
1960 r = 0;
1961 create_pit_unlock:
1962 mutex_unlock(&kvm->lock);
1963 break;
1964 case KVM_IRQ_LINE_STATUS:
1965 case KVM_IRQ_LINE: {
1966 struct kvm_irq_level irq_event;
1967
1968 r = -EFAULT;
1969 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1970 goto out;
1971 if (irqchip_in_kernel(kvm)) {
1972 __s32 status;
1973 mutex_lock(&kvm->lock);
1974 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
1975 irq_event.irq, irq_event.level);
1976 mutex_unlock(&kvm->lock);
1977 if (ioctl == KVM_IRQ_LINE_STATUS) {
1978 irq_event.status = status;
1979 if (copy_to_user(argp, &irq_event,
1980 sizeof irq_event))
1981 goto out;
1982 }
1983 r = 0;
1984 }
1985 break;
1986 }
1987 case KVM_GET_IRQCHIP: {
1988 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1989 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1990
1991 r = -ENOMEM;
1992 if (!chip)
1993 goto out;
1994 r = -EFAULT;
1995 if (copy_from_user(chip, argp, sizeof *chip))
1996 goto get_irqchip_out;
1997 r = -ENXIO;
1998 if (!irqchip_in_kernel(kvm))
1999 goto get_irqchip_out;
2000 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2001 if (r)
2002 goto get_irqchip_out;
2003 r = -EFAULT;
2004 if (copy_to_user(argp, chip, sizeof *chip))
2005 goto get_irqchip_out;
2006 r = 0;
2007 get_irqchip_out:
2008 kfree(chip);
2009 if (r)
2010 goto out;
2011 break;
2012 }
2013 case KVM_SET_IRQCHIP: {
2014 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2015 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2016
2017 r = -ENOMEM;
2018 if (!chip)
2019 goto out;
2020 r = -EFAULT;
2021 if (copy_from_user(chip, argp, sizeof *chip))
2022 goto set_irqchip_out;
2023 r = -ENXIO;
2024 if (!irqchip_in_kernel(kvm))
2025 goto set_irqchip_out;
2026 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2027 if (r)
2028 goto set_irqchip_out;
2029 r = 0;
2030 set_irqchip_out:
2031 kfree(chip);
2032 if (r)
2033 goto out;
2034 break;
2035 }
2036 case KVM_GET_PIT: {
2037 r = -EFAULT;
2038 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2039 goto out;
2040 r = -ENXIO;
2041 if (!kvm->arch.vpit)
2042 goto out;
2043 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2044 if (r)
2045 goto out;
2046 r = -EFAULT;
2047 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2048 goto out;
2049 r = 0;
2050 break;
2051 }
2052 case KVM_SET_PIT: {
2053 r = -EFAULT;
2054 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2055 goto out;
2056 r = -ENXIO;
2057 if (!kvm->arch.vpit)
2058 goto out;
2059 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2060 if (r)
2061 goto out;
2062 r = 0;
2063 break;
2064 }
2065 case KVM_REINJECT_CONTROL: {
2066 struct kvm_reinject_control control;
2067 r = -EFAULT;
2068 if (copy_from_user(&control, argp, sizeof(control)))
2069 goto out;
2070 r = kvm_vm_ioctl_reinject(kvm, &control);
2071 if (r)
2072 goto out;
2073 r = 0;
2074 break;
2075 }
2076 default:
2077 ;
2078 }
2079 out:
2080 return r;
2081 }
2082
2083 static void kvm_init_msr_list(void)
2084 {
2085 u32 dummy[2];
2086 unsigned i, j;
2087
2088 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2089 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2090 continue;
2091 if (j < i)
2092 msrs_to_save[j] = msrs_to_save[i];
2093 j++;
2094 }
2095 num_msrs_to_save = j;
2096 }
2097
2098 /*
2099 * Only apic need an MMIO device hook, so shortcut now..
2100 */
2101 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2102 gpa_t addr, int len,
2103 int is_write)
2104 {
2105 struct kvm_io_device *dev;
2106
2107 if (vcpu->arch.apic) {
2108 dev = &vcpu->arch.apic->dev;
2109 if (dev->in_range(dev, addr, len, is_write))
2110 return dev;
2111 }
2112 return NULL;
2113 }
2114
2115
2116 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2117 gpa_t addr, int len,
2118 int is_write)
2119 {
2120 struct kvm_io_device *dev;
2121
2122 dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2123 if (dev == NULL)
2124 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2125 is_write);
2126 return dev;
2127 }
2128
2129 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2130 struct kvm_vcpu *vcpu)
2131 {
2132 void *data = val;
2133 int r = X86EMUL_CONTINUE;
2134
2135 while (bytes) {
2136 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2137 unsigned offset = addr & (PAGE_SIZE-1);
2138 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2139 int ret;
2140
2141 if (gpa == UNMAPPED_GVA) {
2142 r = X86EMUL_PROPAGATE_FAULT;
2143 goto out;
2144 }
2145 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2146 if (ret < 0) {
2147 r = X86EMUL_UNHANDLEABLE;
2148 goto out;
2149 }
2150
2151 bytes -= toread;
2152 data += toread;
2153 addr += toread;
2154 }
2155 out:
2156 return r;
2157 }
2158
2159 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2160 struct kvm_vcpu *vcpu)
2161 {
2162 void *data = val;
2163 int r = X86EMUL_CONTINUE;
2164
2165 while (bytes) {
2166 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2167 unsigned offset = addr & (PAGE_SIZE-1);
2168 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2169 int ret;
2170
2171 if (gpa == UNMAPPED_GVA) {
2172 r = X86EMUL_PROPAGATE_FAULT;
2173 goto out;
2174 }
2175 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2176 if (ret < 0) {
2177 r = X86EMUL_UNHANDLEABLE;
2178 goto out;
2179 }
2180
2181 bytes -= towrite;
2182 data += towrite;
2183 addr += towrite;
2184 }
2185 out:
2186 return r;
2187 }
2188
2189
2190 static int emulator_read_emulated(unsigned long addr,
2191 void *val,
2192 unsigned int bytes,
2193 struct kvm_vcpu *vcpu)
2194 {
2195 struct kvm_io_device *mmio_dev;
2196 gpa_t gpa;
2197
2198 if (vcpu->mmio_read_completed) {
2199 memcpy(val, vcpu->mmio_data, bytes);
2200 vcpu->mmio_read_completed = 0;
2201 return X86EMUL_CONTINUE;
2202 }
2203
2204 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2205
2206 /* For APIC access vmexit */
2207 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2208 goto mmio;
2209
2210 if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2211 == X86EMUL_CONTINUE)
2212 return X86EMUL_CONTINUE;
2213 if (gpa == UNMAPPED_GVA)
2214 return X86EMUL_PROPAGATE_FAULT;
2215
2216 mmio:
2217 /*
2218 * Is this MMIO handled locally?
2219 */
2220 mutex_lock(&vcpu->kvm->lock);
2221 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2222 if (mmio_dev) {
2223 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2224 mutex_unlock(&vcpu->kvm->lock);
2225 return X86EMUL_CONTINUE;
2226 }
2227 mutex_unlock(&vcpu->kvm->lock);
2228
2229 vcpu->mmio_needed = 1;
2230 vcpu->mmio_phys_addr = gpa;
2231 vcpu->mmio_size = bytes;
2232 vcpu->mmio_is_write = 0;
2233
2234 return X86EMUL_UNHANDLEABLE;
2235 }
2236
2237 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2238 const void *val, int bytes)
2239 {
2240 int ret;
2241
2242 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2243 if (ret < 0)
2244 return 0;
2245 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2246 return 1;
2247 }
2248
2249 static int emulator_write_emulated_onepage(unsigned long addr,
2250 const void *val,
2251 unsigned int bytes,
2252 struct kvm_vcpu *vcpu)
2253 {
2254 struct kvm_io_device *mmio_dev;
2255 gpa_t gpa;
2256
2257 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2258
2259 if (gpa == UNMAPPED_GVA) {
2260 kvm_inject_page_fault(vcpu, addr, 2);
2261 return X86EMUL_PROPAGATE_FAULT;
2262 }
2263
2264 /* For APIC access vmexit */
2265 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2266 goto mmio;
2267
2268 if (emulator_write_phys(vcpu, gpa, val, bytes))
2269 return X86EMUL_CONTINUE;
2270
2271 mmio:
2272 /*
2273 * Is this MMIO handled locally?
2274 */
2275 mutex_lock(&vcpu->kvm->lock);
2276 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2277 if (mmio_dev) {
2278 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2279 mutex_unlock(&vcpu->kvm->lock);
2280 return X86EMUL_CONTINUE;
2281 }
2282 mutex_unlock(&vcpu->kvm->lock);
2283
2284 vcpu->mmio_needed = 1;
2285 vcpu->mmio_phys_addr = gpa;
2286 vcpu->mmio_size = bytes;
2287 vcpu->mmio_is_write = 1;
2288 memcpy(vcpu->mmio_data, val, bytes);
2289
2290 return X86EMUL_CONTINUE;
2291 }
2292
2293 int emulator_write_emulated(unsigned long addr,
2294 const void *val,
2295 unsigned int bytes,
2296 struct kvm_vcpu *vcpu)
2297 {
2298 /* Crossing a page boundary? */
2299 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2300 int rc, now;
2301
2302 now = -addr & ~PAGE_MASK;
2303 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2304 if (rc != X86EMUL_CONTINUE)
2305 return rc;
2306 addr += now;
2307 val += now;
2308 bytes -= now;
2309 }
2310 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2311 }
2312 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2313
2314 static int emulator_cmpxchg_emulated(unsigned long addr,
2315 const void *old,
2316 const void *new,
2317 unsigned int bytes,
2318 struct kvm_vcpu *vcpu)
2319 {
2320 static int reported;
2321
2322 if (!reported) {
2323 reported = 1;
2324 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2325 }
2326 #ifndef CONFIG_X86_64
2327 /* guests cmpxchg8b have to be emulated atomically */
2328 if (bytes == 8) {
2329 gpa_t gpa;
2330 struct page *page;
2331 char *kaddr;
2332 u64 val;
2333
2334 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2335
2336 if (gpa == UNMAPPED_GVA ||
2337 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2338 goto emul_write;
2339
2340 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2341 goto emul_write;
2342
2343 val = *(u64 *)new;
2344
2345 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2346
2347 kaddr = kmap_atomic(page, KM_USER0);
2348 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2349 kunmap_atomic(kaddr, KM_USER0);
2350 kvm_release_page_dirty(page);
2351 }
2352 emul_write:
2353 #endif
2354
2355 return emulator_write_emulated(addr, new, bytes, vcpu);
2356 }
2357
2358 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2359 {
2360 return kvm_x86_ops->get_segment_base(vcpu, seg);
2361 }
2362
2363 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2364 {
2365 kvm_mmu_invlpg(vcpu, address);
2366 return X86EMUL_CONTINUE;
2367 }
2368
2369 int emulate_clts(struct kvm_vcpu *vcpu)
2370 {
2371 KVMTRACE_0D(CLTS, vcpu, handler);
2372 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2373 return X86EMUL_CONTINUE;
2374 }
2375
2376 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2377 {
2378 struct kvm_vcpu *vcpu = ctxt->vcpu;
2379
2380 switch (dr) {
2381 case 0 ... 3:
2382 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2383 return X86EMUL_CONTINUE;
2384 default:
2385 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2386 return X86EMUL_UNHANDLEABLE;
2387 }
2388 }
2389
2390 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2391 {
2392 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2393 int exception;
2394
2395 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2396 if (exception) {
2397 /* FIXME: better handling */
2398 return X86EMUL_UNHANDLEABLE;
2399 }
2400 return X86EMUL_CONTINUE;
2401 }
2402
2403 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2404 {
2405 u8 opcodes[4];
2406 unsigned long rip = kvm_rip_read(vcpu);
2407 unsigned long rip_linear;
2408
2409 if (!printk_ratelimit())
2410 return;
2411
2412 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2413
2414 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2415
2416 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2417 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2418 }
2419 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2420
2421 static struct x86_emulate_ops emulate_ops = {
2422 .read_std = kvm_read_guest_virt,
2423 .read_emulated = emulator_read_emulated,
2424 .write_emulated = emulator_write_emulated,
2425 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2426 };
2427
2428 static void cache_all_regs(struct kvm_vcpu *vcpu)
2429 {
2430 kvm_register_read(vcpu, VCPU_REGS_RAX);
2431 kvm_register_read(vcpu, VCPU_REGS_RSP);
2432 kvm_register_read(vcpu, VCPU_REGS_RIP);
2433 vcpu->arch.regs_dirty = ~0;
2434 }
2435
2436 int emulate_instruction(struct kvm_vcpu *vcpu,
2437 struct kvm_run *run,
2438 unsigned long cr2,
2439 u16 error_code,
2440 int emulation_type)
2441 {
2442 int r, shadow_mask;
2443 struct decode_cache *c;
2444
2445 kvm_clear_exception_queue(vcpu);
2446 vcpu->arch.mmio_fault_cr2 = cr2;
2447 /*
2448 * TODO: fix x86_emulate.c to use guest_read/write_register
2449 * instead of direct ->regs accesses, can save hundred cycles
2450 * on Intel for instructions that don't read/change RSP, for
2451 * for example.
2452 */
2453 cache_all_regs(vcpu);
2454
2455 vcpu->mmio_is_write = 0;
2456 vcpu->arch.pio.string = 0;
2457
2458 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2459 int cs_db, cs_l;
2460 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2461
2462 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2463 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2464 vcpu->arch.emulate_ctxt.mode =
2465 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2466 ? X86EMUL_MODE_REAL : cs_l
2467 ? X86EMUL_MODE_PROT64 : cs_db
2468 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2469
2470 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2471
2472 /* Reject the instructions other than VMCALL/VMMCALL when
2473 * try to emulate invalid opcode */
2474 c = &vcpu->arch.emulate_ctxt.decode;
2475 if ((emulation_type & EMULTYPE_TRAP_UD) &&
2476 (!(c->twobyte && c->b == 0x01 &&
2477 (c->modrm_reg == 0 || c->modrm_reg == 3) &&
2478 c->modrm_mod == 3 && c->modrm_rm == 1)))
2479 return EMULATE_FAIL;
2480
2481 ++vcpu->stat.insn_emulation;
2482 if (r) {
2483 ++vcpu->stat.insn_emulation_fail;
2484 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2485 return EMULATE_DONE;
2486 return EMULATE_FAIL;
2487 }
2488 }
2489
2490 if (emulation_type & EMULTYPE_SKIP) {
2491 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2492 return EMULATE_DONE;
2493 }
2494
2495 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2496 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2497
2498 if (r == 0)
2499 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2500
2501 if (vcpu->arch.pio.string)
2502 return EMULATE_DO_MMIO;
2503
2504 if ((r || vcpu->mmio_is_write) && run) {
2505 run->exit_reason = KVM_EXIT_MMIO;
2506 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2507 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2508 run->mmio.len = vcpu->mmio_size;
2509 run->mmio.is_write = vcpu->mmio_is_write;
2510 }
2511
2512 if (r) {
2513 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2514 return EMULATE_DONE;
2515 if (!vcpu->mmio_needed) {
2516 kvm_report_emulation_failure(vcpu, "mmio");
2517 return EMULATE_FAIL;
2518 }
2519 return EMULATE_DO_MMIO;
2520 }
2521
2522 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2523
2524 if (vcpu->mmio_is_write) {
2525 vcpu->mmio_needed = 0;
2526 return EMULATE_DO_MMIO;
2527 }
2528
2529 return EMULATE_DONE;
2530 }
2531 EXPORT_SYMBOL_GPL(emulate_instruction);
2532
2533 static int pio_copy_data(struct kvm_vcpu *vcpu)
2534 {
2535 void *p = vcpu->arch.pio_data;
2536 gva_t q = vcpu->arch.pio.guest_gva;
2537 unsigned bytes;
2538 int ret;
2539
2540 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2541 if (vcpu->arch.pio.in)
2542 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2543 else
2544 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2545 return ret;
2546 }
2547
2548 int complete_pio(struct kvm_vcpu *vcpu)
2549 {
2550 struct kvm_pio_request *io = &vcpu->arch.pio;
2551 long delta;
2552 int r;
2553 unsigned long val;
2554
2555 if (!io->string) {
2556 if (io->in) {
2557 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2558 memcpy(&val, vcpu->arch.pio_data, io->size);
2559 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2560 }
2561 } else {
2562 if (io->in) {
2563 r = pio_copy_data(vcpu);
2564 if (r)
2565 return r;
2566 }
2567
2568 delta = 1;
2569 if (io->rep) {
2570 delta *= io->cur_count;
2571 /*
2572 * The size of the register should really depend on
2573 * current address size.
2574 */
2575 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2576 val -= delta;
2577 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2578 }
2579 if (io->down)
2580 delta = -delta;
2581 delta *= io->size;
2582 if (io->in) {
2583 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2584 val += delta;
2585 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2586 } else {
2587 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2588 val += delta;
2589 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2590 }
2591 }
2592
2593 io->count -= io->cur_count;
2594 io->cur_count = 0;
2595
2596 return 0;
2597 }
2598
2599 static void kernel_pio(struct kvm_io_device *pio_dev,
2600 struct kvm_vcpu *vcpu,
2601 void *pd)
2602 {
2603 /* TODO: String I/O for in kernel device */
2604
2605 mutex_lock(&vcpu->kvm->lock);
2606 if (vcpu->arch.pio.in)
2607 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2608 vcpu->arch.pio.size,
2609 pd);
2610 else
2611 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2612 vcpu->arch.pio.size,
2613 pd);
2614 mutex_unlock(&vcpu->kvm->lock);
2615 }
2616
2617 static void pio_string_write(struct kvm_io_device *pio_dev,
2618 struct kvm_vcpu *vcpu)
2619 {
2620 struct kvm_pio_request *io = &vcpu->arch.pio;
2621 void *pd = vcpu->arch.pio_data;
2622 int i;
2623
2624 mutex_lock(&vcpu->kvm->lock);
2625 for (i = 0; i < io->cur_count; i++) {
2626 kvm_iodevice_write(pio_dev, io->port,
2627 io->size,
2628 pd);
2629 pd += io->size;
2630 }
2631 mutex_unlock(&vcpu->kvm->lock);
2632 }
2633
2634 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2635 gpa_t addr, int len,
2636 int is_write)
2637 {
2638 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2639 }
2640
2641 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2642 int size, unsigned port)
2643 {
2644 struct kvm_io_device *pio_dev;
2645 unsigned long val;
2646
2647 vcpu->run->exit_reason = KVM_EXIT_IO;
2648 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2649 vcpu->run->io.size = vcpu->arch.pio.size = size;
2650 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2651 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2652 vcpu->run->io.port = vcpu->arch.pio.port = port;
2653 vcpu->arch.pio.in = in;
2654 vcpu->arch.pio.string = 0;
2655 vcpu->arch.pio.down = 0;
2656 vcpu->arch.pio.rep = 0;
2657
2658 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2659 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2660 handler);
2661 else
2662 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2663 handler);
2664
2665 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2666 memcpy(vcpu->arch.pio_data, &val, 4);
2667
2668 pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2669 if (pio_dev) {
2670 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2671 complete_pio(vcpu);
2672 return 1;
2673 }
2674 return 0;
2675 }
2676 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2677
2678 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2679 int size, unsigned long count, int down,
2680 gva_t address, int rep, unsigned port)
2681 {
2682 unsigned now, in_page;
2683 int ret = 0;
2684 struct kvm_io_device *pio_dev;
2685
2686 vcpu->run->exit_reason = KVM_EXIT_IO;
2687 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2688 vcpu->run->io.size = vcpu->arch.pio.size = size;
2689 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2690 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2691 vcpu->run->io.port = vcpu->arch.pio.port = port;
2692 vcpu->arch.pio.in = in;
2693 vcpu->arch.pio.string = 1;
2694 vcpu->arch.pio.down = down;
2695 vcpu->arch.pio.rep = rep;
2696
2697 if (vcpu->run->io.direction == KVM_EXIT_IO_IN)
2698 KVMTRACE_2D(IO_READ, vcpu, vcpu->run->io.port, (u32)size,
2699 handler);
2700 else
2701 KVMTRACE_2D(IO_WRITE, vcpu, vcpu->run->io.port, (u32)size,
2702 handler);
2703
2704 if (!count) {
2705 kvm_x86_ops->skip_emulated_instruction(vcpu);
2706 return 1;
2707 }
2708
2709 if (!down)
2710 in_page = PAGE_SIZE - offset_in_page(address);
2711 else
2712 in_page = offset_in_page(address) + size;
2713 now = min(count, (unsigned long)in_page / size);
2714 if (!now)
2715 now = 1;
2716 if (down) {
2717 /*
2718 * String I/O in reverse. Yuck. Kill the guest, fix later.
2719 */
2720 pr_unimpl(vcpu, "guest string pio down\n");
2721 kvm_inject_gp(vcpu, 0);
2722 return 1;
2723 }
2724 vcpu->run->io.count = now;
2725 vcpu->arch.pio.cur_count = now;
2726
2727 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2728 kvm_x86_ops->skip_emulated_instruction(vcpu);
2729
2730 vcpu->arch.pio.guest_gva = address;
2731
2732 pio_dev = vcpu_find_pio_dev(vcpu, port,
2733 vcpu->arch.pio.cur_count,
2734 !vcpu->arch.pio.in);
2735 if (!vcpu->arch.pio.in) {
2736 /* string PIO write */
2737 ret = pio_copy_data(vcpu);
2738 if (ret == X86EMUL_PROPAGATE_FAULT) {
2739 kvm_inject_gp(vcpu, 0);
2740 return 1;
2741 }
2742 if (ret == 0 && pio_dev) {
2743 pio_string_write(pio_dev, vcpu);
2744 complete_pio(vcpu);
2745 if (vcpu->arch.pio.count == 0)
2746 ret = 1;
2747 }
2748 } else if (pio_dev)
2749 pr_unimpl(vcpu, "no string pio read support yet, "
2750 "port %x size %d count %ld\n",
2751 port, size, count);
2752
2753 return ret;
2754 }
2755 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2756
2757 static void bounce_off(void *info)
2758 {
2759 /* nothing */
2760 }
2761
2762 static unsigned int ref_freq;
2763 static unsigned long tsc_khz_ref;
2764
2765 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2766 void *data)
2767 {
2768 struct cpufreq_freqs *freq = data;
2769 struct kvm *kvm;
2770 struct kvm_vcpu *vcpu;
2771 int i, send_ipi = 0;
2772
2773 if (!ref_freq)
2774 ref_freq = freq->old;
2775
2776 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2777 return 0;
2778 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
2779 return 0;
2780 per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
2781
2782 spin_lock(&kvm_lock);
2783 list_for_each_entry(kvm, &vm_list, vm_list) {
2784 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2785 vcpu = kvm->vcpus[i];
2786 if (!vcpu)
2787 continue;
2788 if (vcpu->cpu != freq->cpu)
2789 continue;
2790 if (!kvm_request_guest_time_update(vcpu))
2791 continue;
2792 if (vcpu->cpu != smp_processor_id())
2793 send_ipi++;
2794 }
2795 }
2796 spin_unlock(&kvm_lock);
2797
2798 if (freq->old < freq->new && send_ipi) {
2799 /*
2800 * We upscale the frequency. Must make the guest
2801 * doesn't see old kvmclock values while running with
2802 * the new frequency, otherwise we risk the guest sees
2803 * time go backwards.
2804 *
2805 * In case we update the frequency for another cpu
2806 * (which might be in guest context) send an interrupt
2807 * to kick the cpu out of guest context. Next time
2808 * guest context is entered kvmclock will be updated,
2809 * so the guest will not see stale values.
2810 */
2811 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
2812 }
2813 return 0;
2814 }
2815
2816 static struct notifier_block kvmclock_cpufreq_notifier_block = {
2817 .notifier_call = kvmclock_cpufreq_notifier
2818 };
2819
2820 int kvm_arch_init(void *opaque)
2821 {
2822 int r, cpu;
2823 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2824
2825 if (kvm_x86_ops) {
2826 printk(KERN_ERR "kvm: already loaded the other module\n");
2827 r = -EEXIST;
2828 goto out;
2829 }
2830
2831 if (!ops->cpu_has_kvm_support()) {
2832 printk(KERN_ERR "kvm: no hardware support\n");
2833 r = -EOPNOTSUPP;
2834 goto out;
2835 }
2836 if (ops->disabled_by_bios()) {
2837 printk(KERN_ERR "kvm: disabled by bios\n");
2838 r = -EOPNOTSUPP;
2839 goto out;
2840 }
2841
2842 r = kvm_mmu_module_init();
2843 if (r)
2844 goto out;
2845
2846 kvm_init_msr_list();
2847
2848 kvm_x86_ops = ops;
2849 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2850 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
2851 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
2852 PT_DIRTY_MASK, PT64_NX_MASK, 0);
2853
2854 for_each_possible_cpu(cpu)
2855 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
2856 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
2857 tsc_khz_ref = tsc_khz;
2858 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
2859 CPUFREQ_TRANSITION_NOTIFIER);
2860 }
2861
2862 return 0;
2863
2864 out:
2865 return r;
2866 }
2867
2868 void kvm_arch_exit(void)
2869 {
2870 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
2871 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
2872 CPUFREQ_TRANSITION_NOTIFIER);
2873 kvm_x86_ops = NULL;
2874 kvm_mmu_module_exit();
2875 }
2876
2877 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2878 {
2879 ++vcpu->stat.halt_exits;
2880 KVMTRACE_0D(HLT, vcpu, handler);
2881 if (irqchip_in_kernel(vcpu->kvm)) {
2882 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
2883 return 1;
2884 } else {
2885 vcpu->run->exit_reason = KVM_EXIT_HLT;
2886 return 0;
2887 }
2888 }
2889 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2890
2891 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
2892 unsigned long a1)
2893 {
2894 if (is_long_mode(vcpu))
2895 return a0;
2896 else
2897 return a0 | ((gpa_t)a1 << 32);
2898 }
2899
2900 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2901 {
2902 unsigned long nr, a0, a1, a2, a3, ret;
2903 int r = 1;
2904
2905 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
2906 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
2907 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
2908 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
2909 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
2910
2911 KVMTRACE_1D(VMMCALL, vcpu, (u32)nr, handler);
2912
2913 if (!is_long_mode(vcpu)) {
2914 nr &= 0xFFFFFFFF;
2915 a0 &= 0xFFFFFFFF;
2916 a1 &= 0xFFFFFFFF;
2917 a2 &= 0xFFFFFFFF;
2918 a3 &= 0xFFFFFFFF;
2919 }
2920
2921 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
2922 ret = -KVM_EPERM;
2923 goto out;
2924 }
2925
2926 switch (nr) {
2927 case KVM_HC_VAPIC_POLL_IRQ:
2928 ret = 0;
2929 break;
2930 case KVM_HC_MMU_OP:
2931 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
2932 break;
2933 default:
2934 ret = -KVM_ENOSYS;
2935 break;
2936 }
2937 out:
2938 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
2939 ++vcpu->stat.hypercalls;
2940 return r;
2941 }
2942 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2943
2944 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2945 {
2946 char instruction[3];
2947 int ret = 0;
2948 unsigned long rip = kvm_rip_read(vcpu);
2949
2950
2951 /*
2952 * Blow out the MMU to ensure that no other VCPU has an active mapping
2953 * to ensure that the updated hypercall appears atomically across all
2954 * VCPUs.
2955 */
2956 kvm_mmu_zap_all(vcpu->kvm);
2957
2958 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2959 if (emulator_write_emulated(rip, instruction, 3, vcpu)
2960 != X86EMUL_CONTINUE)
2961 ret = -EFAULT;
2962
2963 return ret;
2964 }
2965
2966 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2967 {
2968 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2969 }
2970
2971 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2972 {
2973 struct descriptor_table dt = { limit, base };
2974
2975 kvm_x86_ops->set_gdt(vcpu, &dt);
2976 }
2977
2978 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2979 {
2980 struct descriptor_table dt = { limit, base };
2981
2982 kvm_x86_ops->set_idt(vcpu, &dt);
2983 }
2984
2985 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2986 unsigned long *rflags)
2987 {
2988 kvm_lmsw(vcpu, msw);
2989 *rflags = kvm_x86_ops->get_rflags(vcpu);
2990 }
2991
2992 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2993 {
2994 unsigned long value;
2995
2996 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2997 switch (cr) {
2998 case 0:
2999 value = vcpu->arch.cr0;
3000 break;
3001 case 2:
3002 value = vcpu->arch.cr2;
3003 break;
3004 case 3:
3005 value = vcpu->arch.cr3;
3006 break;
3007 case 4:
3008 value = vcpu->arch.cr4;
3009 break;
3010 case 8:
3011 value = kvm_get_cr8(vcpu);
3012 break;
3013 default:
3014 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3015 return 0;
3016 }
3017 KVMTRACE_3D(CR_READ, vcpu, (u32)cr, (u32)value,
3018 (u32)((u64)value >> 32), handler);
3019
3020 return value;
3021 }
3022
3023 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3024 unsigned long *rflags)
3025 {
3026 KVMTRACE_3D(CR_WRITE, vcpu, (u32)cr, (u32)val,
3027 (u32)((u64)val >> 32), handler);
3028
3029 switch (cr) {
3030 case 0:
3031 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3032 *rflags = kvm_x86_ops->get_rflags(vcpu);
3033 break;
3034 case 2:
3035 vcpu->arch.cr2 = val;
3036 break;
3037 case 3:
3038 kvm_set_cr3(vcpu, val);
3039 break;
3040 case 4:
3041 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3042 break;
3043 case 8:
3044 kvm_set_cr8(vcpu, val & 0xfUL);
3045 break;
3046 default:
3047 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3048 }
3049 }
3050
3051 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3052 {
3053 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3054 int j, nent = vcpu->arch.cpuid_nent;
3055
3056 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3057 /* when no next entry is found, the current entry[i] is reselected */
3058 for (j = i + 1; ; j = (j + 1) % nent) {
3059 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3060 if (ej->function == e->function) {
3061 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3062 return j;
3063 }
3064 }
3065 return 0; /* silence gcc, even though control never reaches here */
3066 }
3067
3068 /* find an entry with matching function, matching index (if needed), and that
3069 * should be read next (if it's stateful) */
3070 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3071 u32 function, u32 index)
3072 {
3073 if (e->function != function)
3074 return 0;
3075 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3076 return 0;
3077 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3078 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3079 return 0;
3080 return 1;
3081 }
3082
3083 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3084 u32 function, u32 index)
3085 {
3086 int i;
3087 struct kvm_cpuid_entry2 *best = NULL;
3088
3089 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3090 struct kvm_cpuid_entry2 *e;
3091
3092 e = &vcpu->arch.cpuid_entries[i];
3093 if (is_matching_cpuid_entry(e, function, index)) {
3094 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3095 move_to_next_stateful_cpuid_entry(vcpu, i);
3096 best = e;
3097 break;
3098 }
3099 /*
3100 * Both basic or both extended?
3101 */
3102 if (((e->function ^ function) & 0x80000000) == 0)
3103 if (!best || e->function > best->function)
3104 best = e;
3105 }
3106 return best;
3107 }
3108
3109 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3110 {
3111 struct kvm_cpuid_entry2 *best;
3112
3113 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3114 if (best)
3115 return best->eax & 0xff;
3116 return 36;
3117 }
3118
3119 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3120 {
3121 u32 function, index;
3122 struct kvm_cpuid_entry2 *best;
3123
3124 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3125 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3126 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3127 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3128 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3129 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3130 best = kvm_find_cpuid_entry(vcpu, function, index);
3131 if (best) {
3132 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3133 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3134 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3135 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3136 }
3137 kvm_x86_ops->skip_emulated_instruction(vcpu);
3138 KVMTRACE_5D(CPUID, vcpu, function,
3139 (u32)kvm_register_read(vcpu, VCPU_REGS_RAX),
3140 (u32)kvm_register_read(vcpu, VCPU_REGS_RBX),
3141 (u32)kvm_register_read(vcpu, VCPU_REGS_RCX),
3142 (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler);
3143 }
3144 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3145
3146 /*
3147 * Check if userspace requested an interrupt window, and that the
3148 * interrupt window is open.
3149 *
3150 * No need to exit to userspace if we already have an interrupt queued.
3151 */
3152 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3153 struct kvm_run *kvm_run)
3154 {
3155 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3156 kvm_run->request_interrupt_window &&
3157 kvm_arch_interrupt_allowed(vcpu));
3158 }
3159
3160 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3161 struct kvm_run *kvm_run)
3162 {
3163 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3164 kvm_run->cr8 = kvm_get_cr8(vcpu);
3165 kvm_run->apic_base = kvm_get_apic_base(vcpu);
3166 if (irqchip_in_kernel(vcpu->kvm))
3167 kvm_run->ready_for_interrupt_injection = 1;
3168 else
3169 kvm_run->ready_for_interrupt_injection =
3170 kvm_arch_interrupt_allowed(vcpu) &&
3171 !kvm_cpu_has_interrupt(vcpu) &&
3172 !kvm_event_needs_reinjection(vcpu);
3173 }
3174
3175 static void vapic_enter(struct kvm_vcpu *vcpu)
3176 {
3177 struct kvm_lapic *apic = vcpu->arch.apic;
3178 struct page *page;
3179
3180 if (!apic || !apic->vapic_addr)
3181 return;
3182
3183 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3184
3185 vcpu->arch.apic->vapic_page = page;
3186 }
3187
3188 static void vapic_exit(struct kvm_vcpu *vcpu)
3189 {
3190 struct kvm_lapic *apic = vcpu->arch.apic;
3191
3192 if (!apic || !apic->vapic_addr)
3193 return;
3194
3195 down_read(&vcpu->kvm->slots_lock);
3196 kvm_release_page_dirty(apic->vapic_page);
3197 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3198 up_read(&vcpu->kvm->slots_lock);
3199 }
3200
3201 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3202 {
3203 int max_irr, tpr;
3204
3205 if (!kvm_x86_ops->update_cr8_intercept)
3206 return;
3207
3208 if (!vcpu->arch.apic)
3209 return;
3210
3211 if (!vcpu->arch.apic->vapic_addr)
3212 max_irr = kvm_lapic_find_highest_irr(vcpu);
3213 else
3214 max_irr = -1;
3215
3216 if (max_irr != -1)
3217 max_irr >>= 4;
3218
3219 tpr = kvm_lapic_get_cr8(vcpu);
3220
3221 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3222 }
3223
3224 static void inject_pending_irq(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3225 {
3226 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3227 kvm_x86_ops->set_interrupt_shadow(vcpu, 0);
3228
3229 /* try to reinject previous events if any */
3230 if (vcpu->arch.nmi_injected) {
3231 kvm_x86_ops->set_nmi(vcpu);
3232 return;
3233 }
3234
3235 if (vcpu->arch.interrupt.pending) {
3236 kvm_x86_ops->set_irq(vcpu);
3237 return;
3238 }
3239
3240 /* try to inject new event if pending */
3241 if (vcpu->arch.nmi_pending) {
3242 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3243 vcpu->arch.nmi_pending = false;
3244 vcpu->arch.nmi_injected = true;
3245 kvm_x86_ops->set_nmi(vcpu);
3246 }
3247 } else if (kvm_cpu_has_interrupt(vcpu)) {
3248 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3249 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3250 false);
3251 kvm_x86_ops->set_irq(vcpu);
3252 }
3253 }
3254 }
3255
3256 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3257 {
3258 int r;
3259 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3260 kvm_run->request_interrupt_window;
3261
3262 if (vcpu->requests)
3263 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3264 kvm_mmu_unload(vcpu);
3265
3266 r = kvm_mmu_reload(vcpu);
3267 if (unlikely(r))
3268 goto out;
3269
3270 if (vcpu->requests) {
3271 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3272 __kvm_migrate_timers(vcpu);
3273 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3274 kvm_write_guest_time(vcpu);
3275 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3276 kvm_mmu_sync_roots(vcpu);
3277 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3278 kvm_x86_ops->tlb_flush(vcpu);
3279 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3280 &vcpu->requests)) {
3281 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3282 r = 0;
3283 goto out;
3284 }
3285 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3286 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3287 r = 0;
3288 goto out;
3289 }
3290 }
3291
3292 preempt_disable();
3293
3294 kvm_x86_ops->prepare_guest_switch(vcpu);
3295 kvm_load_guest_fpu(vcpu);
3296
3297 local_irq_disable();
3298
3299 clear_bit(KVM_REQ_KICK, &vcpu->requests);
3300 smp_mb__after_clear_bit();
3301
3302 if (vcpu->requests || need_resched() || signal_pending(current)) {
3303 local_irq_enable();
3304 preempt_enable();
3305 r = 1;
3306 goto out;
3307 }
3308
3309 if (vcpu->arch.exception.pending)
3310 __queue_exception(vcpu);
3311 else
3312 inject_pending_irq(vcpu, kvm_run);
3313
3314 /* enable NMI/IRQ window open exits if needed */
3315 if (vcpu->arch.nmi_pending)
3316 kvm_x86_ops->enable_nmi_window(vcpu);
3317 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3318 kvm_x86_ops->enable_irq_window(vcpu);
3319
3320 if (kvm_lapic_enabled(vcpu)) {
3321 update_cr8_intercept(vcpu);
3322 kvm_lapic_sync_to_vapic(vcpu);
3323 }
3324
3325 up_read(&vcpu->kvm->slots_lock);
3326
3327 kvm_guest_enter();
3328
3329 get_debugreg(vcpu->arch.host_dr6, 6);
3330 get_debugreg(vcpu->arch.host_dr7, 7);
3331 if (unlikely(vcpu->arch.switch_db_regs)) {
3332 get_debugreg(vcpu->arch.host_db[0], 0);
3333 get_debugreg(vcpu->arch.host_db[1], 1);
3334 get_debugreg(vcpu->arch.host_db[2], 2);
3335 get_debugreg(vcpu->arch.host_db[3], 3);
3336
3337 set_debugreg(0, 7);
3338 set_debugreg(vcpu->arch.eff_db[0], 0);
3339 set_debugreg(vcpu->arch.eff_db[1], 1);
3340 set_debugreg(vcpu->arch.eff_db[2], 2);
3341 set_debugreg(vcpu->arch.eff_db[3], 3);
3342 }
3343
3344 KVMTRACE_0D(VMENTRY, vcpu, entryexit);
3345 kvm_x86_ops->run(vcpu, kvm_run);
3346
3347 if (unlikely(vcpu->arch.switch_db_regs)) {
3348 set_debugreg(0, 7);
3349 set_debugreg(vcpu->arch.host_db[0], 0);
3350 set_debugreg(vcpu->arch.host_db[1], 1);
3351 set_debugreg(vcpu->arch.host_db[2], 2);
3352 set_debugreg(vcpu->arch.host_db[3], 3);
3353 }
3354 set_debugreg(vcpu->arch.host_dr6, 6);
3355 set_debugreg(vcpu->arch.host_dr7, 7);
3356
3357 set_bit(KVM_REQ_KICK, &vcpu->requests);
3358 local_irq_enable();
3359
3360 ++vcpu->stat.exits;
3361
3362 /*
3363 * We must have an instruction between local_irq_enable() and
3364 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3365 * the interrupt shadow. The stat.exits increment will do nicely.
3366 * But we need to prevent reordering, hence this barrier():
3367 */
3368 barrier();
3369
3370 kvm_guest_exit();
3371
3372 preempt_enable();
3373
3374 down_read(&vcpu->kvm->slots_lock);
3375
3376 /*
3377 * Profile KVM exit RIPs:
3378 */
3379 if (unlikely(prof_on == KVM_PROFILING)) {
3380 unsigned long rip = kvm_rip_read(vcpu);
3381 profile_hit(KVM_PROFILING, (void *)rip);
3382 }
3383
3384
3385 kvm_lapic_sync_from_vapic(vcpu);
3386
3387 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3388 out:
3389 return r;
3390 }
3391
3392
3393 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3394 {
3395 int r;
3396
3397 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3398 pr_debug("vcpu %d received sipi with vector # %x\n",
3399 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3400 kvm_lapic_reset(vcpu);
3401 r = kvm_arch_vcpu_reset(vcpu);
3402 if (r)
3403 return r;
3404 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3405 }
3406
3407 down_read(&vcpu->kvm->slots_lock);
3408 vapic_enter(vcpu);
3409
3410 r = 1;
3411 while (r > 0) {
3412 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3413 r = vcpu_enter_guest(vcpu, kvm_run);
3414 else {
3415 up_read(&vcpu->kvm->slots_lock);
3416 kvm_vcpu_block(vcpu);
3417 down_read(&vcpu->kvm->slots_lock);
3418 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3419 {
3420 switch(vcpu->arch.mp_state) {
3421 case KVM_MP_STATE_HALTED:
3422 vcpu->arch.mp_state =
3423 KVM_MP_STATE_RUNNABLE;
3424 case KVM_MP_STATE_RUNNABLE:
3425 break;
3426 case KVM_MP_STATE_SIPI_RECEIVED:
3427 default:
3428 r = -EINTR;
3429 break;
3430 }
3431 }
3432 }
3433
3434 if (r <= 0)
3435 break;
3436
3437 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3438 if (kvm_cpu_has_pending_timer(vcpu))
3439 kvm_inject_pending_timer_irqs(vcpu);
3440
3441 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3442 r = -EINTR;
3443 kvm_run->exit_reason = KVM_EXIT_INTR;
3444 ++vcpu->stat.request_irq_exits;
3445 }
3446 if (signal_pending(current)) {
3447 r = -EINTR;
3448 kvm_run->exit_reason = KVM_EXIT_INTR;
3449 ++vcpu->stat.signal_exits;
3450 }
3451 if (need_resched()) {
3452 up_read(&vcpu->kvm->slots_lock);
3453 kvm_resched(vcpu);
3454 down_read(&vcpu->kvm->slots_lock);
3455 }
3456 }
3457
3458 up_read(&vcpu->kvm->slots_lock);
3459 post_kvm_run_save(vcpu, kvm_run);
3460
3461 vapic_exit(vcpu);
3462
3463 return r;
3464 }
3465
3466 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3467 {
3468 int r;
3469 sigset_t sigsaved;
3470
3471 vcpu_load(vcpu);
3472
3473 if (vcpu->sigset_active)
3474 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3475
3476 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3477 kvm_vcpu_block(vcpu);
3478 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3479 r = -EAGAIN;
3480 goto out;
3481 }
3482
3483 /* re-sync apic's tpr */
3484 if (!irqchip_in_kernel(vcpu->kvm))
3485 kvm_set_cr8(vcpu, kvm_run->cr8);
3486
3487 if (vcpu->arch.pio.cur_count) {
3488 r = complete_pio(vcpu);
3489 if (r)
3490 goto out;
3491 }
3492 #if CONFIG_HAS_IOMEM
3493 if (vcpu->mmio_needed) {
3494 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3495 vcpu->mmio_read_completed = 1;
3496 vcpu->mmio_needed = 0;
3497
3498 down_read(&vcpu->kvm->slots_lock);
3499 r = emulate_instruction(vcpu, kvm_run,
3500 vcpu->arch.mmio_fault_cr2, 0,
3501 EMULTYPE_NO_DECODE);
3502 up_read(&vcpu->kvm->slots_lock);
3503 if (r == EMULATE_DO_MMIO) {
3504 /*
3505 * Read-modify-write. Back to userspace.
3506 */
3507 r = 0;
3508 goto out;
3509 }
3510 }
3511 #endif
3512 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3513 kvm_register_write(vcpu, VCPU_REGS_RAX,
3514 kvm_run->hypercall.ret);
3515
3516 r = __vcpu_run(vcpu, kvm_run);
3517
3518 out:
3519 if (vcpu->sigset_active)
3520 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3521
3522 vcpu_put(vcpu);
3523 return r;
3524 }
3525
3526 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3527 {
3528 vcpu_load(vcpu);
3529
3530 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3531 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3532 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3533 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3534 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3535 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3536 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3537 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3538 #ifdef CONFIG_X86_64
3539 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3540 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3541 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3542 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3543 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3544 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3545 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3546 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3547 #endif
3548
3549 regs->rip = kvm_rip_read(vcpu);
3550 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3551
3552 /*
3553 * Don't leak debug flags in case they were set for guest debugging
3554 */
3555 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3556 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3557
3558 vcpu_put(vcpu);
3559
3560 return 0;
3561 }
3562
3563 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3564 {
3565 vcpu_load(vcpu);
3566
3567 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3568 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3569 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3570 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3571 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3572 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3573 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3574 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3575 #ifdef CONFIG_X86_64
3576 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3577 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3578 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3579 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3580 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3581 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3582 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3583 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3584
3585 #endif
3586
3587 kvm_rip_write(vcpu, regs->rip);
3588 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3589
3590
3591 vcpu->arch.exception.pending = false;
3592
3593 vcpu_put(vcpu);
3594
3595 return 0;
3596 }
3597
3598 void kvm_get_segment(struct kvm_vcpu *vcpu,
3599 struct kvm_segment *var, int seg)
3600 {
3601 kvm_x86_ops->get_segment(vcpu, var, seg);
3602 }
3603
3604 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3605 {
3606 struct kvm_segment cs;
3607
3608 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3609 *db = cs.db;
3610 *l = cs.l;
3611 }
3612 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3613
3614 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3615 struct kvm_sregs *sregs)
3616 {
3617 struct descriptor_table dt;
3618
3619 vcpu_load(vcpu);
3620
3621 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3622 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3623 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3624 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3625 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3626 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3627
3628 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3629 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3630
3631 kvm_x86_ops->get_idt(vcpu, &dt);
3632 sregs->idt.limit = dt.limit;
3633 sregs->idt.base = dt.base;
3634 kvm_x86_ops->get_gdt(vcpu, &dt);
3635 sregs->gdt.limit = dt.limit;
3636 sregs->gdt.base = dt.base;
3637
3638 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3639 sregs->cr0 = vcpu->arch.cr0;
3640 sregs->cr2 = vcpu->arch.cr2;
3641 sregs->cr3 = vcpu->arch.cr3;
3642 sregs->cr4 = vcpu->arch.cr4;
3643 sregs->cr8 = kvm_get_cr8(vcpu);
3644 sregs->efer = vcpu->arch.shadow_efer;
3645 sregs->apic_base = kvm_get_apic_base(vcpu);
3646
3647 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3648
3649 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3650 set_bit(vcpu->arch.interrupt.nr,
3651 (unsigned long *)sregs->interrupt_bitmap);
3652
3653 vcpu_put(vcpu);
3654
3655 return 0;
3656 }
3657
3658 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3659 struct kvm_mp_state *mp_state)
3660 {
3661 vcpu_load(vcpu);
3662 mp_state->mp_state = vcpu->arch.mp_state;
3663 vcpu_put(vcpu);
3664 return 0;
3665 }
3666
3667 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3668 struct kvm_mp_state *mp_state)
3669 {
3670 vcpu_load(vcpu);
3671 vcpu->arch.mp_state = mp_state->mp_state;
3672 vcpu_put(vcpu);
3673 return 0;
3674 }
3675
3676 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3677 struct kvm_segment *var, int seg)
3678 {
3679 kvm_x86_ops->set_segment(vcpu, var, seg);
3680 }
3681
3682 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3683 struct kvm_segment *kvm_desct)
3684 {
3685 kvm_desct->base = seg_desc->base0;
3686 kvm_desct->base |= seg_desc->base1 << 16;
3687 kvm_desct->base |= seg_desc->base2 << 24;
3688 kvm_desct->limit = seg_desc->limit0;
3689 kvm_desct->limit |= seg_desc->limit << 16;
3690 if (seg_desc->g) {
3691 kvm_desct->limit <<= 12;
3692 kvm_desct->limit |= 0xfff;
3693 }
3694 kvm_desct->selector = selector;
3695 kvm_desct->type = seg_desc->type;
3696 kvm_desct->present = seg_desc->p;
3697 kvm_desct->dpl = seg_desc->dpl;
3698 kvm_desct->db = seg_desc->d;
3699 kvm_desct->s = seg_desc->s;
3700 kvm_desct->l = seg_desc->l;
3701 kvm_desct->g = seg_desc->g;
3702 kvm_desct->avl = seg_desc->avl;
3703 if (!selector)
3704 kvm_desct->unusable = 1;
3705 else
3706 kvm_desct->unusable = 0;
3707 kvm_desct->padding = 0;
3708 }
3709
3710 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3711 u16 selector,
3712 struct descriptor_table *dtable)
3713 {
3714 if (selector & 1 << 2) {
3715 struct kvm_segment kvm_seg;
3716
3717 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3718
3719 if (kvm_seg.unusable)
3720 dtable->limit = 0;
3721 else
3722 dtable->limit = kvm_seg.limit;
3723 dtable->base = kvm_seg.base;
3724 }
3725 else
3726 kvm_x86_ops->get_gdt(vcpu, dtable);
3727 }
3728
3729 /* allowed just for 8 bytes segments */
3730 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3731 struct desc_struct *seg_desc)
3732 {
3733 gpa_t gpa;
3734 struct descriptor_table dtable;
3735 u16 index = selector >> 3;
3736
3737 get_segment_descriptor_dtable(vcpu, selector, &dtable);
3738
3739 if (dtable.limit < index * 8 + 7) {
3740 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3741 return 1;
3742 }
3743 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3744 gpa += index * 8;
3745 return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3746 }
3747
3748 /* allowed just for 8 bytes segments */
3749 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3750 struct desc_struct *seg_desc)
3751 {
3752 gpa_t gpa;
3753 struct descriptor_table dtable;
3754 u16 index = selector >> 3;
3755
3756 get_segment_descriptor_dtable(vcpu, selector, &dtable);
3757
3758 if (dtable.limit < index * 8 + 7)
3759 return 1;
3760 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3761 gpa += index * 8;
3762 return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3763 }
3764
3765 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
3766 struct desc_struct *seg_desc)
3767 {
3768 u32 base_addr;
3769
3770 base_addr = seg_desc->base0;
3771 base_addr |= (seg_desc->base1 << 16);
3772 base_addr |= (seg_desc->base2 << 24);
3773
3774 return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3775 }
3776
3777 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3778 {
3779 struct kvm_segment kvm_seg;
3780
3781 kvm_get_segment(vcpu, &kvm_seg, seg);
3782 return kvm_seg.selector;
3783 }
3784
3785 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3786 u16 selector,
3787 struct kvm_segment *kvm_seg)
3788 {
3789 struct desc_struct seg_desc;
3790
3791 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3792 return 1;
3793 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3794 return 0;
3795 }
3796
3797 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3798 {
3799 struct kvm_segment segvar = {
3800 .base = selector << 4,
3801 .limit = 0xffff,
3802 .selector = selector,
3803 .type = 3,
3804 .present = 1,
3805 .dpl = 3,
3806 .db = 0,
3807 .s = 1,
3808 .l = 0,
3809 .g = 0,
3810 .avl = 0,
3811 .unusable = 0,
3812 };
3813 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
3814 return 0;
3815 }
3816
3817 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3818 int type_bits, int seg)
3819 {
3820 struct kvm_segment kvm_seg;
3821
3822 if (!(vcpu->arch.cr0 & X86_CR0_PE))
3823 return kvm_load_realmode_segment(vcpu, selector, seg);
3824 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
3825 return 1;
3826 kvm_seg.type |= type_bits;
3827
3828 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
3829 seg != VCPU_SREG_LDTR)
3830 if (!kvm_seg.s)
3831 kvm_seg.unusable = 1;
3832
3833 kvm_set_segment(vcpu, &kvm_seg, seg);
3834 return 0;
3835 }
3836
3837 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
3838 struct tss_segment_32 *tss)
3839 {
3840 tss->cr3 = vcpu->arch.cr3;
3841 tss->eip = kvm_rip_read(vcpu);
3842 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
3843 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3844 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3845 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3846 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3847 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3848 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3849 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3850 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3851 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3852 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3853 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3854 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3855 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
3856 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
3857 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3858 }
3859
3860 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
3861 struct tss_segment_32 *tss)
3862 {
3863 kvm_set_cr3(vcpu, tss->cr3);
3864
3865 kvm_rip_write(vcpu, tss->eip);
3866 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
3867
3868 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
3869 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
3870 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
3871 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
3872 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
3873 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
3874 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
3875 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
3876
3877 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
3878 return 1;
3879
3880 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3881 return 1;
3882
3883 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3884 return 1;
3885
3886 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3887 return 1;
3888
3889 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3890 return 1;
3891
3892 if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
3893 return 1;
3894
3895 if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
3896 return 1;
3897 return 0;
3898 }
3899
3900 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
3901 struct tss_segment_16 *tss)
3902 {
3903 tss->ip = kvm_rip_read(vcpu);
3904 tss->flag = kvm_x86_ops->get_rflags(vcpu);
3905 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3906 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3907 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3908 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3909 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3910 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3911 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
3912 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
3913
3914 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
3915 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
3916 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
3917 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
3918 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
3919 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
3920 }
3921
3922 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
3923 struct tss_segment_16 *tss)
3924 {
3925 kvm_rip_write(vcpu, tss->ip);
3926 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
3927 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
3928 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
3929 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
3930 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
3931 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
3932 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
3933 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
3934 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
3935
3936 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
3937 return 1;
3938
3939 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
3940 return 1;
3941
3942 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
3943 return 1;
3944
3945 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
3946 return 1;
3947
3948 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
3949 return 1;
3950 return 0;
3951 }
3952
3953 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
3954 u16 old_tss_sel, u32 old_tss_base,
3955 struct desc_struct *nseg_desc)
3956 {
3957 struct tss_segment_16 tss_segment_16;
3958 int ret = 0;
3959
3960 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3961 sizeof tss_segment_16))
3962 goto out;
3963
3964 save_state_to_tss16(vcpu, &tss_segment_16);
3965
3966 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
3967 sizeof tss_segment_16))
3968 goto out;
3969
3970 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
3971 &tss_segment_16, sizeof tss_segment_16))
3972 goto out;
3973
3974 if (old_tss_sel != 0xffff) {
3975 tss_segment_16.prev_task_link = old_tss_sel;
3976
3977 if (kvm_write_guest(vcpu->kvm,
3978 get_tss_base_addr(vcpu, nseg_desc),
3979 &tss_segment_16.prev_task_link,
3980 sizeof tss_segment_16.prev_task_link))
3981 goto out;
3982 }
3983
3984 if (load_state_from_tss16(vcpu, &tss_segment_16))
3985 goto out;
3986
3987 ret = 1;
3988 out:
3989 return ret;
3990 }
3991
3992 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
3993 u16 old_tss_sel, u32 old_tss_base,
3994 struct desc_struct *nseg_desc)
3995 {
3996 struct tss_segment_32 tss_segment_32;
3997 int ret = 0;
3998
3999 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4000 sizeof tss_segment_32))
4001 goto out;
4002
4003 save_state_to_tss32(vcpu, &tss_segment_32);
4004
4005 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4006 sizeof tss_segment_32))
4007 goto out;
4008
4009 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4010 &tss_segment_32, sizeof tss_segment_32))
4011 goto out;
4012
4013 if (old_tss_sel != 0xffff) {
4014 tss_segment_32.prev_task_link = old_tss_sel;
4015
4016 if (kvm_write_guest(vcpu->kvm,
4017 get_tss_base_addr(vcpu, nseg_desc),
4018 &tss_segment_32.prev_task_link,
4019 sizeof tss_segment_32.prev_task_link))
4020 goto out;
4021 }
4022
4023 if (load_state_from_tss32(vcpu, &tss_segment_32))
4024 goto out;
4025
4026 ret = 1;
4027 out:
4028 return ret;
4029 }
4030
4031 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4032 {
4033 struct kvm_segment tr_seg;
4034 struct desc_struct cseg_desc;
4035 struct desc_struct nseg_desc;
4036 int ret = 0;
4037 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4038 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4039
4040 old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4041
4042 /* FIXME: Handle errors. Failure to read either TSS or their
4043 * descriptors should generate a pagefault.
4044 */
4045 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4046 goto out;
4047
4048 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4049 goto out;
4050
4051 if (reason != TASK_SWITCH_IRET) {
4052 int cpl;
4053
4054 cpl = kvm_x86_ops->get_cpl(vcpu);
4055 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4056 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4057 return 1;
4058 }
4059 }
4060
4061 if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
4062 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4063 return 1;
4064 }
4065
4066 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4067 cseg_desc.type &= ~(1 << 1); //clear the B flag
4068 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4069 }
4070
4071 if (reason == TASK_SWITCH_IRET) {
4072 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4073 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4074 }
4075
4076 /* set back link to prev task only if NT bit is set in eflags
4077 note that old_tss_sel is not used afetr this point */
4078 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4079 old_tss_sel = 0xffff;
4080
4081 /* set back link to prev task only if NT bit is set in eflags
4082 note that old_tss_sel is not used afetr this point */
4083 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4084 old_tss_sel = 0xffff;
4085
4086 if (nseg_desc.type & 8)
4087 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4088 old_tss_base, &nseg_desc);
4089 else
4090 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4091 old_tss_base, &nseg_desc);
4092
4093 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4094 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4095 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4096 }
4097
4098 if (reason != TASK_SWITCH_IRET) {
4099 nseg_desc.type |= (1 << 1);
4100 save_guest_segment_descriptor(vcpu, tss_selector,
4101 &nseg_desc);
4102 }
4103
4104 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4105 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4106 tr_seg.type = 11;
4107 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4108 out:
4109 return ret;
4110 }
4111 EXPORT_SYMBOL_GPL(kvm_task_switch);
4112
4113 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4114 struct kvm_sregs *sregs)
4115 {
4116 int mmu_reset_needed = 0;
4117 int pending_vec, max_bits;
4118 struct descriptor_table dt;
4119
4120 vcpu_load(vcpu);
4121
4122 dt.limit = sregs->idt.limit;
4123 dt.base = sregs->idt.base;
4124 kvm_x86_ops->set_idt(vcpu, &dt);
4125 dt.limit = sregs->gdt.limit;
4126 dt.base = sregs->gdt.base;
4127 kvm_x86_ops->set_gdt(vcpu, &dt);
4128
4129 vcpu->arch.cr2 = sregs->cr2;
4130 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4131 vcpu->arch.cr3 = sregs->cr3;
4132
4133 kvm_set_cr8(vcpu, sregs->cr8);
4134
4135 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4136 kvm_x86_ops->set_efer(vcpu, sregs->efer);
4137 kvm_set_apic_base(vcpu, sregs->apic_base);
4138
4139 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4140
4141 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4142 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4143 vcpu->arch.cr0 = sregs->cr0;
4144
4145 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4146 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4147 if (!is_long_mode(vcpu) && is_pae(vcpu))
4148 load_pdptrs(vcpu, vcpu->arch.cr3);
4149
4150 if (mmu_reset_needed)
4151 kvm_mmu_reset_context(vcpu);
4152
4153 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4154 pending_vec = find_first_bit(
4155 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4156 if (pending_vec < max_bits) {
4157 kvm_queue_interrupt(vcpu, pending_vec, false);
4158 pr_debug("Set back pending irq %d\n", pending_vec);
4159 if (irqchip_in_kernel(vcpu->kvm))
4160 kvm_pic_clear_isr_ack(vcpu->kvm);
4161 }
4162
4163 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4164 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4165 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4166 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4167 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4168 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4169
4170 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4171 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4172
4173 /* Older userspace won't unhalt the vcpu on reset. */
4174 if (vcpu->vcpu_id == 0 && kvm_rip_read(vcpu) == 0xfff0 &&
4175 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4176 !(vcpu->arch.cr0 & X86_CR0_PE))
4177 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4178
4179 vcpu_put(vcpu);
4180
4181 return 0;
4182 }
4183
4184 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4185 struct kvm_guest_debug *dbg)
4186 {
4187 int i, r;
4188
4189 vcpu_load(vcpu);
4190
4191 if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4192 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4193 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4194 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4195 vcpu->arch.switch_db_regs =
4196 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4197 } else {
4198 for (i = 0; i < KVM_NR_DB_REGS; i++)
4199 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4200 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4201 }
4202
4203 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4204
4205 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4206 kvm_queue_exception(vcpu, DB_VECTOR);
4207 else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4208 kvm_queue_exception(vcpu, BP_VECTOR);
4209
4210 vcpu_put(vcpu);
4211
4212 return r;
4213 }
4214
4215 /*
4216 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4217 * we have asm/x86/processor.h
4218 */
4219 struct fxsave {
4220 u16 cwd;
4221 u16 swd;
4222 u16 twd;
4223 u16 fop;
4224 u64 rip;
4225 u64 rdp;
4226 u32 mxcsr;
4227 u32 mxcsr_mask;
4228 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4229 #ifdef CONFIG_X86_64
4230 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4231 #else
4232 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4233 #endif
4234 };
4235
4236 /*
4237 * Translate a guest virtual address to a guest physical address.
4238 */
4239 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4240 struct kvm_translation *tr)
4241 {
4242 unsigned long vaddr = tr->linear_address;
4243 gpa_t gpa;
4244
4245 vcpu_load(vcpu);
4246 down_read(&vcpu->kvm->slots_lock);
4247 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4248 up_read(&vcpu->kvm->slots_lock);
4249 tr->physical_address = gpa;
4250 tr->valid = gpa != UNMAPPED_GVA;
4251 tr->writeable = 1;
4252 tr->usermode = 0;
4253 vcpu_put(vcpu);
4254
4255 return 0;
4256 }
4257
4258 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4259 {
4260 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4261
4262 vcpu_load(vcpu);
4263
4264 memcpy(fpu->fpr, fxsave->st_space, 128);
4265 fpu->fcw = fxsave->cwd;
4266 fpu->fsw = fxsave->swd;
4267 fpu->ftwx = fxsave->twd;
4268 fpu->last_opcode = fxsave->fop;
4269 fpu->last_ip = fxsave->rip;
4270 fpu->last_dp = fxsave->rdp;
4271 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4272
4273 vcpu_put(vcpu);
4274
4275 return 0;
4276 }
4277
4278 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4279 {
4280 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4281
4282 vcpu_load(vcpu);
4283
4284 memcpy(fxsave->st_space, fpu->fpr, 128);
4285 fxsave->cwd = fpu->fcw;
4286 fxsave->swd = fpu->fsw;
4287 fxsave->twd = fpu->ftwx;
4288 fxsave->fop = fpu->last_opcode;
4289 fxsave->rip = fpu->last_ip;
4290 fxsave->rdp = fpu->last_dp;
4291 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4292
4293 vcpu_put(vcpu);
4294
4295 return 0;
4296 }
4297
4298 void fx_init(struct kvm_vcpu *vcpu)
4299 {
4300 unsigned after_mxcsr_mask;
4301
4302 /*
4303 * Touch the fpu the first time in non atomic context as if
4304 * this is the first fpu instruction the exception handler
4305 * will fire before the instruction returns and it'll have to
4306 * allocate ram with GFP_KERNEL.
4307 */
4308 if (!used_math())
4309 kvm_fx_save(&vcpu->arch.host_fx_image);
4310
4311 /* Initialize guest FPU by resetting ours and saving into guest's */
4312 preempt_disable();
4313 kvm_fx_save(&vcpu->arch.host_fx_image);
4314 kvm_fx_finit();
4315 kvm_fx_save(&vcpu->arch.guest_fx_image);
4316 kvm_fx_restore(&vcpu->arch.host_fx_image);
4317 preempt_enable();
4318
4319 vcpu->arch.cr0 |= X86_CR0_ET;
4320 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4321 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4322 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4323 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4324 }
4325 EXPORT_SYMBOL_GPL(fx_init);
4326
4327 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4328 {
4329 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4330 return;
4331
4332 vcpu->guest_fpu_loaded = 1;
4333 kvm_fx_save(&vcpu->arch.host_fx_image);
4334 kvm_fx_restore(&vcpu->arch.guest_fx_image);
4335 }
4336 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4337
4338 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4339 {
4340 if (!vcpu->guest_fpu_loaded)
4341 return;
4342
4343 vcpu->guest_fpu_loaded = 0;
4344 kvm_fx_save(&vcpu->arch.guest_fx_image);
4345 kvm_fx_restore(&vcpu->arch.host_fx_image);
4346 ++vcpu->stat.fpu_reload;
4347 }
4348 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4349
4350 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4351 {
4352 if (vcpu->arch.time_page) {
4353 kvm_release_page_dirty(vcpu->arch.time_page);
4354 vcpu->arch.time_page = NULL;
4355 }
4356
4357 kvm_x86_ops->vcpu_free(vcpu);
4358 }
4359
4360 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4361 unsigned int id)
4362 {
4363 return kvm_x86_ops->vcpu_create(kvm, id);
4364 }
4365
4366 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4367 {
4368 int r;
4369
4370 /* We do fxsave: this must be aligned. */
4371 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4372
4373 vcpu->arch.mtrr_state.have_fixed = 1;
4374 vcpu_load(vcpu);
4375 r = kvm_arch_vcpu_reset(vcpu);
4376 if (r == 0)
4377 r = kvm_mmu_setup(vcpu);
4378 vcpu_put(vcpu);
4379 if (r < 0)
4380 goto free_vcpu;
4381
4382 return 0;
4383 free_vcpu:
4384 kvm_x86_ops->vcpu_free(vcpu);
4385 return r;
4386 }
4387
4388 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4389 {
4390 vcpu_load(vcpu);
4391 kvm_mmu_unload(vcpu);
4392 vcpu_put(vcpu);
4393
4394 kvm_x86_ops->vcpu_free(vcpu);
4395 }
4396
4397 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4398 {
4399 vcpu->arch.nmi_pending = false;
4400 vcpu->arch.nmi_injected = false;
4401
4402 vcpu->arch.switch_db_regs = 0;
4403 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4404 vcpu->arch.dr6 = DR6_FIXED_1;
4405 vcpu->arch.dr7 = DR7_FIXED_1;
4406
4407 return kvm_x86_ops->vcpu_reset(vcpu);
4408 }
4409
4410 void kvm_arch_hardware_enable(void *garbage)
4411 {
4412 kvm_x86_ops->hardware_enable(garbage);
4413 }
4414
4415 void kvm_arch_hardware_disable(void *garbage)
4416 {
4417 kvm_x86_ops->hardware_disable(garbage);
4418 }
4419
4420 int kvm_arch_hardware_setup(void)
4421 {
4422 return kvm_x86_ops->hardware_setup();
4423 }
4424
4425 void kvm_arch_hardware_unsetup(void)
4426 {
4427 kvm_x86_ops->hardware_unsetup();
4428 }
4429
4430 void kvm_arch_check_processor_compat(void *rtn)
4431 {
4432 kvm_x86_ops->check_processor_compatibility(rtn);
4433 }
4434
4435 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4436 {
4437 struct page *page;
4438 struct kvm *kvm;
4439 int r;
4440
4441 BUG_ON(vcpu->kvm == NULL);
4442 kvm = vcpu->kvm;
4443
4444 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4445 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
4446 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4447 else
4448 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4449
4450 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4451 if (!page) {
4452 r = -ENOMEM;
4453 goto fail;
4454 }
4455 vcpu->arch.pio_data = page_address(page);
4456
4457 r = kvm_mmu_create(vcpu);
4458 if (r < 0)
4459 goto fail_free_pio_data;
4460
4461 if (irqchip_in_kernel(kvm)) {
4462 r = kvm_create_lapic(vcpu);
4463 if (r < 0)
4464 goto fail_mmu_destroy;
4465 }
4466
4467 return 0;
4468
4469 fail_mmu_destroy:
4470 kvm_mmu_destroy(vcpu);
4471 fail_free_pio_data:
4472 free_page((unsigned long)vcpu->arch.pio_data);
4473 fail:
4474 return r;
4475 }
4476
4477 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4478 {
4479 kvm_free_lapic(vcpu);
4480 down_read(&vcpu->kvm->slots_lock);
4481 kvm_mmu_destroy(vcpu);
4482 up_read(&vcpu->kvm->slots_lock);
4483 free_page((unsigned long)vcpu->arch.pio_data);
4484 }
4485
4486 struct kvm *kvm_arch_create_vm(void)
4487 {
4488 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4489
4490 if (!kvm)
4491 return ERR_PTR(-ENOMEM);
4492
4493 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4494 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4495
4496 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4497 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4498
4499 rdtscll(kvm->arch.vm_init_tsc);
4500
4501 return kvm;
4502 }
4503
4504 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4505 {
4506 vcpu_load(vcpu);
4507 kvm_mmu_unload(vcpu);
4508 vcpu_put(vcpu);
4509 }
4510
4511 static void kvm_free_vcpus(struct kvm *kvm)
4512 {
4513 unsigned int i;
4514
4515 /*
4516 * Unpin any mmu pages first.
4517 */
4518 for (i = 0; i < KVM_MAX_VCPUS; ++i)
4519 if (kvm->vcpus[i])
4520 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
4521 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
4522 if (kvm->vcpus[i]) {
4523 kvm_arch_vcpu_free(kvm->vcpus[i]);
4524 kvm->vcpus[i] = NULL;
4525 }
4526 }
4527
4528 }
4529
4530 void kvm_arch_sync_events(struct kvm *kvm)
4531 {
4532 kvm_free_all_assigned_devices(kvm);
4533 }
4534
4535 void kvm_arch_destroy_vm(struct kvm *kvm)
4536 {
4537 kvm_iommu_unmap_guest(kvm);
4538 kvm_free_pit(kvm);
4539 kfree(kvm->arch.vpic);
4540 kfree(kvm->arch.vioapic);
4541 kvm_free_vcpus(kvm);
4542 kvm_free_physmem(kvm);
4543 if (kvm->arch.apic_access_page)
4544 put_page(kvm->arch.apic_access_page);
4545 if (kvm->arch.ept_identity_pagetable)
4546 put_page(kvm->arch.ept_identity_pagetable);
4547 kfree(kvm);
4548 }
4549
4550 int kvm_arch_set_memory_region(struct kvm *kvm,
4551 struct kvm_userspace_memory_region *mem,
4552 struct kvm_memory_slot old,
4553 int user_alloc)
4554 {
4555 int npages = mem->memory_size >> PAGE_SHIFT;
4556 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4557
4558 /*To keep backward compatibility with older userspace,
4559 *x86 needs to hanlde !user_alloc case.
4560 */
4561 if (!user_alloc) {
4562 if (npages && !old.rmap) {
4563 unsigned long userspace_addr;
4564
4565 down_write(&current->mm->mmap_sem);
4566 userspace_addr = do_mmap(NULL, 0,
4567 npages * PAGE_SIZE,
4568 PROT_READ | PROT_WRITE,
4569 MAP_PRIVATE | MAP_ANONYMOUS,
4570 0);
4571 up_write(&current->mm->mmap_sem);
4572
4573 if (IS_ERR((void *)userspace_addr))
4574 return PTR_ERR((void *)userspace_addr);
4575
4576 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4577 spin_lock(&kvm->mmu_lock);
4578 memslot->userspace_addr = userspace_addr;
4579 spin_unlock(&kvm->mmu_lock);
4580 } else {
4581 if (!old.user_alloc && old.rmap) {
4582 int ret;
4583
4584 down_write(&current->mm->mmap_sem);
4585 ret = do_munmap(current->mm, old.userspace_addr,
4586 old.npages * PAGE_SIZE);
4587 up_write(&current->mm->mmap_sem);
4588 if (ret < 0)
4589 printk(KERN_WARNING
4590 "kvm_vm_ioctl_set_memory_region: "
4591 "failed to munmap memory\n");
4592 }
4593 }
4594 }
4595
4596 spin_lock(&kvm->mmu_lock);
4597 if (!kvm->arch.n_requested_mmu_pages) {
4598 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4599 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4600 }
4601
4602 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4603 spin_unlock(&kvm->mmu_lock);
4604 kvm_flush_remote_tlbs(kvm);
4605
4606 return 0;
4607 }
4608
4609 void kvm_arch_flush_shadow(struct kvm *kvm)
4610 {
4611 kvm_mmu_zap_all(kvm);
4612 kvm_reload_remote_mmus(kvm);
4613 }
4614
4615 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4616 {
4617 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4618 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4619 || vcpu->arch.nmi_pending;
4620 }
4621
4622 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4623 {
4624 int me;
4625 int cpu = vcpu->cpu;
4626
4627 if (waitqueue_active(&vcpu->wq)) {
4628 wake_up_interruptible(&vcpu->wq);
4629 ++vcpu->stat.halt_wakeup;
4630 }
4631
4632 me = get_cpu();
4633 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4634 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4635 smp_send_reschedule(cpu);
4636 put_cpu();
4637 }
4638
4639 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4640 {
4641 return kvm_x86_ops->interrupt_allowed(vcpu);
4642 }
Linux with added FPC-III support