projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
KVM: VMX: Handle #SS faults from real mode
[deliverable/linux.git] / drivers / kvm / vmx.c
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18 #include "kvm.h"
19 #include "vmx.h"
20 #include <linux/module.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/highmem.h>
24 #include <linux/profile.h>
25 #include <linux/sched.h>
26 #include <asm/io.h>
27 #include <asm/desc.h>
28
29 #include "segment_descriptor.h"
30
31 MODULE_AUTHOR("Qumranet");
32 MODULE_LICENSE("GPL");
33
34 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
35 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
36
37 static struct page *vmx_io_bitmap_a;
38 static struct page *vmx_io_bitmap_b;
39
40 #ifdef CONFIG_X86_64
41 #define HOST_IS_64 1
42 #else
43 #define HOST_IS_64 0
44 #endif
45
46 static struct vmcs_descriptor {
47 int size;
48 int order;
49 u32 revision_id;
50 } vmcs_descriptor;
51
52 #define VMX_SEGMENT_FIELD(seg) \
53 [VCPU_SREG_##seg] = { \
54 .selector = GUEST_##seg##_SELECTOR, \
55 .base = GUEST_##seg##_BASE, \
56 .limit = GUEST_##seg##_LIMIT, \
57 .ar_bytes = GUEST_##seg##_AR_BYTES, \
58 }
59
60 static struct kvm_vmx_segment_field {
61 unsigned selector;
62 unsigned base;
63 unsigned limit;
64 unsigned ar_bytes;
65 } kvm_vmx_segment_fields[] = {
66 VMX_SEGMENT_FIELD(CS),
67 VMX_SEGMENT_FIELD(DS),
68 VMX_SEGMENT_FIELD(ES),
69 VMX_SEGMENT_FIELD(FS),
70 VMX_SEGMENT_FIELD(GS),
71 VMX_SEGMENT_FIELD(SS),
72 VMX_SEGMENT_FIELD(TR),
73 VMX_SEGMENT_FIELD(LDTR),
74 };
75
76 /*
77 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
78 * away by decrementing the array size.
79 */
80 static const u32 vmx_msr_index[] = {
81 #ifdef CONFIG_X86_64
82 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
83 #endif
84 MSR_EFER, MSR_K6_STAR,
85 };
86 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
87
88 #ifdef CONFIG_X86_64
89 static unsigned msr_offset_kernel_gs_base;
90 #define NR_64BIT_MSRS 4
91 /*
92 * avoid save/load MSR_SYSCALL_MASK and MSR_LSTAR by std vt
93 * mechanism (cpu bug AA24)
94 */
95 #define NR_BAD_MSRS 2
96 #else
97 #define NR_64BIT_MSRS 0
98 #define NR_BAD_MSRS 0
99 #endif
100
101 static inline int is_page_fault(u32 intr_info)
102 {
103 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
104 INTR_INFO_VALID_MASK)) ==
105 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
106 }
107
108 static inline int is_no_device(u32 intr_info)
109 {
110 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
111 INTR_INFO_VALID_MASK)) ==
112 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
113 }
114
115 static inline int is_external_interrupt(u32 intr_info)
116 {
117 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
118 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
119 }
120
121 static struct vmx_msr_entry *find_msr_entry(struct kvm_vcpu *vcpu, u32 msr)
122 {
123 int i;
124
125 for (i = 0; i < vcpu->nmsrs; ++i)
126 if (vcpu->guest_msrs[i].index == msr)
127 return &vcpu->guest_msrs[i];
128 return NULL;
129 }
130
131 static void vmcs_clear(struct vmcs *vmcs)
132 {
133 u64 phys_addr = __pa(vmcs);
134 u8 error;
135
136 asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
137 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
138 : "cc", "memory");
139 if (error)
140 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
141 vmcs, phys_addr);
142 }
143
144 static void __vcpu_clear(void *arg)
145 {
146 struct kvm_vcpu *vcpu = arg;
147 int cpu = raw_smp_processor_id();
148
149 if (vcpu->cpu == cpu)
150 vmcs_clear(vcpu->vmcs);
151 if (per_cpu(current_vmcs, cpu) == vcpu->vmcs)
152 per_cpu(current_vmcs, cpu) = NULL;
153 }
154
155 static void vcpu_clear(struct kvm_vcpu *vcpu)
156 {
157 if (vcpu->cpu != raw_smp_processor_id() && vcpu->cpu != -1)
158 smp_call_function_single(vcpu->cpu, __vcpu_clear, vcpu, 0, 1);
159 else
160 __vcpu_clear(vcpu);
161 vcpu->launched = 0;
162 }
163
164 static unsigned long vmcs_readl(unsigned long field)
165 {
166 unsigned long value;
167
168 asm volatile (ASM_VMX_VMREAD_RDX_RAX
169 : "=a"(value) : "d"(field) : "cc");
170 return value;
171 }
172
173 static u16 vmcs_read16(unsigned long field)
174 {
175 return vmcs_readl(field);
176 }
177
178 static u32 vmcs_read32(unsigned long field)
179 {
180 return vmcs_readl(field);
181 }
182
183 static u64 vmcs_read64(unsigned long field)
184 {
185 #ifdef CONFIG_X86_64
186 return vmcs_readl(field);
187 #else
188 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
189 #endif
190 }
191
192 static noinline void vmwrite_error(unsigned long field, unsigned long value)
193 {
194 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
195 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
196 dump_stack();
197 }
198
199 static void vmcs_writel(unsigned long field, unsigned long value)
200 {
201 u8 error;
202
203 asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
204 : "=q"(error) : "a"(value), "d"(field) : "cc" );
205 if (unlikely(error))
206 vmwrite_error(field, value);
207 }
208
209 static void vmcs_write16(unsigned long field, u16 value)
210 {
211 vmcs_writel(field, value);
212 }
213
214 static void vmcs_write32(unsigned long field, u32 value)
215 {
216 vmcs_writel(field, value);
217 }
218
219 static void vmcs_write64(unsigned long field, u64 value)
220 {
221 #ifdef CONFIG_X86_64
222 vmcs_writel(field, value);
223 #else
224 vmcs_writel(field, value);
225 asm volatile ("");
226 vmcs_writel(field+1, value >> 32);
227 #endif
228 }
229
230 static void vmcs_clear_bits(unsigned long field, u32 mask)
231 {
232 vmcs_writel(field, vmcs_readl(field) & ~mask);
233 }
234
235 static void vmcs_set_bits(unsigned long field, u32 mask)
236 {
237 vmcs_writel(field, vmcs_readl(field) | mask);
238 }
239
240 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
241 {
242 u32 eb;
243
244 eb = 1u << PF_VECTOR;
245 if (!vcpu->fpu_active)
246 eb |= 1u << NM_VECTOR;
247 if (vcpu->guest_debug.enabled)
248 eb |= 1u << 1;
249 if (vcpu->rmode.active)
250 eb = ~0;
251 vmcs_write32(EXCEPTION_BITMAP, eb);
252 }
253
254 static void reload_tss(void)
255 {
256 #ifndef CONFIG_X86_64
257
258 /*
259 * VT restores TR but not its size. Useless.
260 */
261 struct descriptor_table gdt;
262 struct segment_descriptor *descs;
263
264 get_gdt(&gdt);
265 descs = (void *)gdt.base;
266 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
267 load_TR_desc();
268 #endif
269 }
270
271 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
272 {
273 struct vmx_host_state *hs = &vcpu->vmx_host_state;
274
275 if (hs->loaded)
276 return;
277
278 hs->loaded = 1;
279 /*
280 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
281 * allow segment selectors with cpl > 0 or ti == 1.
282 */
283 hs->ldt_sel = read_ldt();
284 hs->fs_gs_ldt_reload_needed = hs->ldt_sel;
285 hs->fs_sel = read_fs();
286 if (!(hs->fs_sel & 7))
287 vmcs_write16(HOST_FS_SELECTOR, hs->fs_sel);
288 else {
289 vmcs_write16(HOST_FS_SELECTOR, 0);
290 hs->fs_gs_ldt_reload_needed = 1;
291 }
292 hs->gs_sel = read_gs();
293 if (!(hs->gs_sel & 7))
294 vmcs_write16(HOST_GS_SELECTOR, hs->gs_sel);
295 else {
296 vmcs_write16(HOST_GS_SELECTOR, 0);
297 hs->fs_gs_ldt_reload_needed = 1;
298 }
299
300 #ifdef CONFIG_X86_64
301 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
302 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
303 #else
304 vmcs_writel(HOST_FS_BASE, segment_base(hs->fs_sel));
305 vmcs_writel(HOST_GS_BASE, segment_base(hs->gs_sel));
306 #endif
307
308 #ifdef CONFIG_X86_64
309 if (is_long_mode(vcpu)) {
310 save_msrs(vcpu->host_msrs + msr_offset_kernel_gs_base, 1);
311 load_msrs(vcpu->guest_msrs, NR_BAD_MSRS);
312 }
313 #endif
314 }
315
316 static void vmx_load_host_state(struct kvm_vcpu *vcpu)
317 {
318 struct vmx_host_state *hs = &vcpu->vmx_host_state;
319
320 if (!hs->loaded)
321 return;
322
323 hs->loaded = 0;
324 if (hs->fs_gs_ldt_reload_needed) {
325 load_ldt(hs->ldt_sel);
326 load_fs(hs->fs_sel);
327 /*
328 * If we have to reload gs, we must take care to
329 * preserve our gs base.
330 */
331 local_irq_disable();
332 load_gs(hs->gs_sel);
333 #ifdef CONFIG_X86_64
334 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
335 #endif
336 local_irq_enable();
337
338 reload_tss();
339 }
340 #ifdef CONFIG_X86_64
341 if (is_long_mode(vcpu)) {
342 save_msrs(vcpu->guest_msrs, NR_BAD_MSRS);
343 load_msrs(vcpu->host_msrs, NR_BAD_MSRS);
344 }
345 #endif
346 }
347
348 /*
349 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
350 * vcpu mutex is already taken.
351 */
352 static void vmx_vcpu_load(struct kvm_vcpu *vcpu)
353 {
354 u64 phys_addr = __pa(vcpu->vmcs);
355 int cpu;
356
357 cpu = get_cpu();
358
359 if (vcpu->cpu != cpu)
360 vcpu_clear(vcpu);
361
362 if (per_cpu(current_vmcs, cpu) != vcpu->vmcs) {
363 u8 error;
364
365 per_cpu(current_vmcs, cpu) = vcpu->vmcs;
366 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
367 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
368 : "cc");
369 if (error)
370 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
371 vcpu->vmcs, phys_addr);
372 }
373
374 if (vcpu->cpu != cpu) {
375 struct descriptor_table dt;
376 unsigned long sysenter_esp;
377
378 vcpu->cpu = cpu;
379 /*
380 * Linux uses per-cpu TSS and GDT, so set these when switching
381 * processors.
382 */
383 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
384 get_gdt(&dt);
385 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
386
387 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
388 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
389 }
390 }
391
392 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
393 {
394 vmx_load_host_state(vcpu);
395 kvm_put_guest_fpu(vcpu);
396 put_cpu();
397 }
398
399 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
400 {
401 if (vcpu->fpu_active)
402 return;
403 vcpu->fpu_active = 1;
404 vmcs_clear_bits(GUEST_CR0, CR0_TS_MASK);
405 if (vcpu->cr0 & CR0_TS_MASK)
406 vmcs_set_bits(GUEST_CR0, CR0_TS_MASK);
407 update_exception_bitmap(vcpu);
408 }
409
410 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
411 {
412 if (!vcpu->fpu_active)
413 return;
414 vcpu->fpu_active = 0;
415 vmcs_set_bits(GUEST_CR0, CR0_TS_MASK);
416 update_exception_bitmap(vcpu);
417 }
418
419 static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
420 {
421 vcpu_clear(vcpu);
422 }
423
424 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
425 {
426 return vmcs_readl(GUEST_RFLAGS);
427 }
428
429 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
430 {
431 vmcs_writel(GUEST_RFLAGS, rflags);
432 }
433
434 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
435 {
436 unsigned long rip;
437 u32 interruptibility;
438
439 rip = vmcs_readl(GUEST_RIP);
440 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
441 vmcs_writel(GUEST_RIP, rip);
442
443 /*
444 * We emulated an instruction, so temporary interrupt blocking
445 * should be removed, if set.
446 */
447 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
448 if (interruptibility & 3)
449 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
450 interruptibility & ~3);
451 vcpu->interrupt_window_open = 1;
452 }
453
454 static void vmx_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
455 {
456 printk(KERN_DEBUG "inject_general_protection: rip 0x%lx\n",
457 vmcs_readl(GUEST_RIP));
458 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
459 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
460 GP_VECTOR |
461 INTR_TYPE_EXCEPTION |
462 INTR_INFO_DELIEVER_CODE_MASK |
463 INTR_INFO_VALID_MASK);
464 }
465
466 /*
467 * Set up the vmcs to automatically save and restore system
468 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
469 * mode, as fiddling with msrs is very expensive.
470 */
471 static void setup_msrs(struct kvm_vcpu *vcpu)
472 {
473 int nr_skip, nr_good_msrs;
474
475 if (is_long_mode(vcpu))
476 nr_skip = NR_BAD_MSRS;
477 else
478 nr_skip = NR_64BIT_MSRS;
479 nr_good_msrs = vcpu->nmsrs - nr_skip;
480
481 /*
482 * MSR_K6_STAR is only needed on long mode guests, and only
483 * if efer.sce is enabled.
484 */
485 if (find_msr_entry(vcpu, MSR_K6_STAR)) {
486 --nr_good_msrs;
487 #ifdef CONFIG_X86_64
488 if (is_long_mode(vcpu) && (vcpu->shadow_efer & EFER_SCE))
489 ++nr_good_msrs;
490 #endif
491 }
492
493 vmcs_writel(VM_ENTRY_MSR_LOAD_ADDR,
494 virt_to_phys(vcpu->guest_msrs + nr_skip));
495 vmcs_writel(VM_EXIT_MSR_STORE_ADDR,
496 virt_to_phys(vcpu->guest_msrs + nr_skip));
497 vmcs_writel(VM_EXIT_MSR_LOAD_ADDR,
498 virt_to_phys(vcpu->host_msrs + nr_skip));
499 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, nr_good_msrs); /* 22.2.2 */
500 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, nr_good_msrs); /* 22.2.2 */
501 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, nr_good_msrs); /* 22.2.2 */
502 }
503
504 /*
505 * reads and returns guest's timestamp counter "register"
506 * guest_tsc = host_tsc + tsc_offset -- 21.3
507 */
508 static u64 guest_read_tsc(void)
509 {
510 u64 host_tsc, tsc_offset;
511
512 rdtscll(host_tsc);
513 tsc_offset = vmcs_read64(TSC_OFFSET);
514 return host_tsc + tsc_offset;
515 }
516
517 /*
518 * writes 'guest_tsc' into guest's timestamp counter "register"
519 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
520 */
521 static void guest_write_tsc(u64 guest_tsc)
522 {
523 u64 host_tsc;
524
525 rdtscll(host_tsc);
526 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
527 }
528
529 /*
530 * Reads an msr value (of 'msr_index') into 'pdata'.
531 * Returns 0 on success, non-0 otherwise.
532 * Assumes vcpu_load() was already called.
533 */
534 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
535 {
536 u64 data;
537 struct vmx_msr_entry *msr;
538
539 if (!pdata) {
540 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
541 return -EINVAL;
542 }
543
544 switch (msr_index) {
545 #ifdef CONFIG_X86_64
546 case MSR_FS_BASE:
547 data = vmcs_readl(GUEST_FS_BASE);
548 break;
549 case MSR_GS_BASE:
550 data = vmcs_readl(GUEST_GS_BASE);
551 break;
552 case MSR_EFER:
553 return kvm_get_msr_common(vcpu, msr_index, pdata);
554 #endif
555 case MSR_IA32_TIME_STAMP_COUNTER:
556 data = guest_read_tsc();
557 break;
558 case MSR_IA32_SYSENTER_CS:
559 data = vmcs_read32(GUEST_SYSENTER_CS);
560 break;
561 case MSR_IA32_SYSENTER_EIP:
562 data = vmcs_readl(GUEST_SYSENTER_EIP);
563 break;
564 case MSR_IA32_SYSENTER_ESP:
565 data = vmcs_readl(GUEST_SYSENTER_ESP);
566 break;
567 default:
568 msr = find_msr_entry(vcpu, msr_index);
569 if (msr) {
570 data = msr->data;
571 break;
572 }
573 return kvm_get_msr_common(vcpu, msr_index, pdata);
574 }
575
576 *pdata = data;
577 return 0;
578 }
579
580 /*
581 * Writes msr value into into the appropriate "register".
582 * Returns 0 on success, non-0 otherwise.
583 * Assumes vcpu_load() was already called.
584 */
585 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
586 {
587 struct vmx_msr_entry *msr;
588 switch (msr_index) {
589 #ifdef CONFIG_X86_64
590 case MSR_EFER:
591 return kvm_set_msr_common(vcpu, msr_index, data);
592 case MSR_FS_BASE:
593 vmcs_writel(GUEST_FS_BASE, data);
594 break;
595 case MSR_GS_BASE:
596 vmcs_writel(GUEST_GS_BASE, data);
597 break;
598 case MSR_LSTAR:
599 case MSR_SYSCALL_MASK:
600 msr = find_msr_entry(vcpu, msr_index);
601 if (msr)
602 msr->data = data;
603 if (vcpu->vmx_host_state.loaded)
604 load_msrs(vcpu->guest_msrs, NR_BAD_MSRS);
605 break;
606 #endif
607 case MSR_IA32_SYSENTER_CS:
608 vmcs_write32(GUEST_SYSENTER_CS, data);
609 break;
610 case MSR_IA32_SYSENTER_EIP:
611 vmcs_writel(GUEST_SYSENTER_EIP, data);
612 break;
613 case MSR_IA32_SYSENTER_ESP:
614 vmcs_writel(GUEST_SYSENTER_ESP, data);
615 break;
616 case MSR_IA32_TIME_STAMP_COUNTER:
617 guest_write_tsc(data);
618 break;
619 default:
620 msr = find_msr_entry(vcpu, msr_index);
621 if (msr) {
622 msr->data = data;
623 break;
624 }
625 return kvm_set_msr_common(vcpu, msr_index, data);
626 msr->data = data;
627 break;
628 }
629
630 return 0;
631 }
632
633 /*
634 * Sync the rsp and rip registers into the vcpu structure. This allows
635 * registers to be accessed by indexing vcpu->regs.
636 */
637 static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
638 {
639 vcpu->regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
640 vcpu->rip = vmcs_readl(GUEST_RIP);
641 }
642
643 /*
644 * Syncs rsp and rip back into the vmcs. Should be called after possible
645 * modification.
646 */
647 static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
648 {
649 vmcs_writel(GUEST_RSP, vcpu->regs[VCPU_REGS_RSP]);
650 vmcs_writel(GUEST_RIP, vcpu->rip);
651 }
652
653 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
654 {
655 unsigned long dr7 = 0x400;
656 int old_singlestep;
657
658 old_singlestep = vcpu->guest_debug.singlestep;
659
660 vcpu->guest_debug.enabled = dbg->enabled;
661 if (vcpu->guest_debug.enabled) {
662 int i;
663
664 dr7 |= 0x200; /* exact */
665 for (i = 0; i < 4; ++i) {
666 if (!dbg->breakpoints[i].enabled)
667 continue;
668 vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
669 dr7 |= 2 << (i*2); /* global enable */
670 dr7 |= 0 << (i*4+16); /* execution breakpoint */
671 }
672
673 vcpu->guest_debug.singlestep = dbg->singlestep;
674 } else
675 vcpu->guest_debug.singlestep = 0;
676
677 if (old_singlestep && !vcpu->guest_debug.singlestep) {
678 unsigned long flags;
679
680 flags = vmcs_readl(GUEST_RFLAGS);
681 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
682 vmcs_writel(GUEST_RFLAGS, flags);
683 }
684
685 update_exception_bitmap(vcpu);
686 vmcs_writel(GUEST_DR7, dr7);
687
688 return 0;
689 }
690
691 static __init int cpu_has_kvm_support(void)
692 {
693 unsigned long ecx = cpuid_ecx(1);
694 return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
695 }
696
697 static __init int vmx_disabled_by_bios(void)
698 {
699 u64 msr;
700
701 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
702 return (msr & 5) == 1; /* locked but not enabled */
703 }
704
705 static void hardware_enable(void *garbage)
706 {
707 int cpu = raw_smp_processor_id();
708 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
709 u64 old;
710
711 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
712 if ((old & 5) != 5)
713 /* enable and lock */
714 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | 5);
715 write_cr4(read_cr4() | CR4_VMXE); /* FIXME: not cpu hotplug safe */
716 asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
717 : "memory", "cc");
718 }
719
720 static void hardware_disable(void *garbage)
721 {
722 asm volatile (ASM_VMX_VMXOFF : : : "cc");
723 }
724
725 static __init void setup_vmcs_descriptor(void)
726 {
727 u32 vmx_msr_low, vmx_msr_high;
728
729 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
730 vmcs_descriptor.size = vmx_msr_high & 0x1fff;
731 vmcs_descriptor.order = get_order(vmcs_descriptor.size);
732 vmcs_descriptor.revision_id = vmx_msr_low;
733 }
734
735 static struct vmcs *alloc_vmcs_cpu(int cpu)
736 {
737 int node = cpu_to_node(cpu);
738 struct page *pages;
739 struct vmcs *vmcs;
740
741 pages = alloc_pages_node(node, GFP_KERNEL, vmcs_descriptor.order);
742 if (!pages)
743 return NULL;
744 vmcs = page_address(pages);
745 memset(vmcs, 0, vmcs_descriptor.size);
746 vmcs->revision_id = vmcs_descriptor.revision_id; /* vmcs revision id */
747 return vmcs;
748 }
749
750 static struct vmcs *alloc_vmcs(void)
751 {
752 return alloc_vmcs_cpu(raw_smp_processor_id());
753 }
754
755 static void free_vmcs(struct vmcs *vmcs)
756 {
757 free_pages((unsigned long)vmcs, vmcs_descriptor.order);
758 }
759
760 static void free_kvm_area(void)
761 {
762 int cpu;
763
764 for_each_online_cpu(cpu)
765 free_vmcs(per_cpu(vmxarea, cpu));
766 }
767
768 extern struct vmcs *alloc_vmcs_cpu(int cpu);
769
770 static __init int alloc_kvm_area(void)
771 {
772 int cpu;
773
774 for_each_online_cpu(cpu) {
775 struct vmcs *vmcs;
776
777 vmcs = alloc_vmcs_cpu(cpu);
778 if (!vmcs) {
779 free_kvm_area();
780 return -ENOMEM;
781 }
782
783 per_cpu(vmxarea, cpu) = vmcs;
784 }
785 return 0;
786 }
787
788 static __init int hardware_setup(void)
789 {
790 setup_vmcs_descriptor();
791 return alloc_kvm_area();
792 }
793
794 static __exit void hardware_unsetup(void)
795 {
796 free_kvm_area();
797 }
798
799 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
800 {
801 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
802
803 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
804 vmcs_write16(sf->selector, save->selector);
805 vmcs_writel(sf->base, save->base);
806 vmcs_write32(sf->limit, save->limit);
807 vmcs_write32(sf->ar_bytes, save->ar);
808 } else {
809 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
810 << AR_DPL_SHIFT;
811 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
812 }
813 }
814
815 static void enter_pmode(struct kvm_vcpu *vcpu)
816 {
817 unsigned long flags;
818
819 vcpu->rmode.active = 0;
820
821 vmcs_writel(GUEST_TR_BASE, vcpu->rmode.tr.base);
822 vmcs_write32(GUEST_TR_LIMIT, vcpu->rmode.tr.limit);
823 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->rmode.tr.ar);
824
825 flags = vmcs_readl(GUEST_RFLAGS);
826 flags &= ~(IOPL_MASK | X86_EFLAGS_VM);
827 flags |= (vcpu->rmode.save_iopl << IOPL_SHIFT);
828 vmcs_writel(GUEST_RFLAGS, flags);
829
830 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~CR4_VME_MASK) |
831 (vmcs_readl(CR4_READ_SHADOW) & CR4_VME_MASK));
832
833 update_exception_bitmap(vcpu);
834
835 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->rmode.es);
836 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->rmode.ds);
837 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->rmode.gs);
838 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->rmode.fs);
839
840 vmcs_write16(GUEST_SS_SELECTOR, 0);
841 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
842
843 vmcs_write16(GUEST_CS_SELECTOR,
844 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
845 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
846 }
847
848 static int rmode_tss_base(struct kvm* kvm)
849 {
850 gfn_t base_gfn = kvm->memslots[0].base_gfn + kvm->memslots[0].npages - 3;
851 return base_gfn << PAGE_SHIFT;
852 }
853
854 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
855 {
856 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
857
858 save->selector = vmcs_read16(sf->selector);
859 save->base = vmcs_readl(sf->base);
860 save->limit = vmcs_read32(sf->limit);
861 save->ar = vmcs_read32(sf->ar_bytes);
862 vmcs_write16(sf->selector, vmcs_readl(sf->base) >> 4);
863 vmcs_write32(sf->limit, 0xffff);
864 vmcs_write32(sf->ar_bytes, 0xf3);
865 }
866
867 static void enter_rmode(struct kvm_vcpu *vcpu)
868 {
869 unsigned long flags;
870
871 vcpu->rmode.active = 1;
872
873 vcpu->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
874 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
875
876 vcpu->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
877 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
878
879 vcpu->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
880 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
881
882 flags = vmcs_readl(GUEST_RFLAGS);
883 vcpu->rmode.save_iopl = (flags & IOPL_MASK) >> IOPL_SHIFT;
884
885 flags |= IOPL_MASK | X86_EFLAGS_VM;
886
887 vmcs_writel(GUEST_RFLAGS, flags);
888 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | CR4_VME_MASK);
889 update_exception_bitmap(vcpu);
890
891 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
892 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
893 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
894
895 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
896 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
897 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
898 vmcs_writel(GUEST_CS_BASE, 0xf0000);
899 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
900
901 fix_rmode_seg(VCPU_SREG_ES, &vcpu->rmode.es);
902 fix_rmode_seg(VCPU_SREG_DS, &vcpu->rmode.ds);
903 fix_rmode_seg(VCPU_SREG_GS, &vcpu->rmode.gs);
904 fix_rmode_seg(VCPU_SREG_FS, &vcpu->rmode.fs);
905 }
906
907 #ifdef CONFIG_X86_64
908
909 static void enter_lmode(struct kvm_vcpu *vcpu)
910 {
911 u32 guest_tr_ar;
912
913 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
914 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
915 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
916 __FUNCTION__);
917 vmcs_write32(GUEST_TR_AR_BYTES,
918 (guest_tr_ar & ~AR_TYPE_MASK)
919 | AR_TYPE_BUSY_64_TSS);
920 }
921
922 vcpu->shadow_efer |= EFER_LMA;
923
924 find_msr_entry(vcpu, MSR_EFER)->data |= EFER_LMA | EFER_LME;
925 vmcs_write32(VM_ENTRY_CONTROLS,
926 vmcs_read32(VM_ENTRY_CONTROLS)
927 | VM_ENTRY_CONTROLS_IA32E_MASK);
928 }
929
930 static void exit_lmode(struct kvm_vcpu *vcpu)
931 {
932 vcpu->shadow_efer &= ~EFER_LMA;
933
934 vmcs_write32(VM_ENTRY_CONTROLS,
935 vmcs_read32(VM_ENTRY_CONTROLS)
936 & ~VM_ENTRY_CONTROLS_IA32E_MASK);
937 }
938
939 #endif
940
941 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
942 {
943 vcpu->cr4 &= KVM_GUEST_CR4_MASK;
944 vcpu->cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
945 }
946
947 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
948 {
949 vmx_fpu_deactivate(vcpu);
950
951 if (vcpu->rmode.active && (cr0 & CR0_PE_MASK))
952 enter_pmode(vcpu);
953
954 if (!vcpu->rmode.active && !(cr0 & CR0_PE_MASK))
955 enter_rmode(vcpu);
956
957 #ifdef CONFIG_X86_64
958 if (vcpu->shadow_efer & EFER_LME) {
959 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK))
960 enter_lmode(vcpu);
961 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK))
962 exit_lmode(vcpu);
963 }
964 #endif
965
966 vmcs_writel(CR0_READ_SHADOW, cr0);
967 vmcs_writel(GUEST_CR0,
968 (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
969 vcpu->cr0 = cr0;
970
971 if (!(cr0 & CR0_TS_MASK) || !(cr0 & CR0_PE_MASK))
972 vmx_fpu_activate(vcpu);
973 }
974
975 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
976 {
977 vmcs_writel(GUEST_CR3, cr3);
978 if (vcpu->cr0 & CR0_PE_MASK)
979 vmx_fpu_deactivate(vcpu);
980 }
981
982 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
983 {
984 vmcs_writel(CR4_READ_SHADOW, cr4);
985 vmcs_writel(GUEST_CR4, cr4 | (vcpu->rmode.active ?
986 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
987 vcpu->cr4 = cr4;
988 }
989
990 #ifdef CONFIG_X86_64
991
992 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
993 {
994 struct vmx_msr_entry *msr = find_msr_entry(vcpu, MSR_EFER);
995
996 vcpu->shadow_efer = efer;
997 if (efer & EFER_LMA) {
998 vmcs_write32(VM_ENTRY_CONTROLS,
999 vmcs_read32(VM_ENTRY_CONTROLS) |
1000 VM_ENTRY_CONTROLS_IA32E_MASK);
1001 msr->data = efer;
1002
1003 } else {
1004 vmcs_write32(VM_ENTRY_CONTROLS,
1005 vmcs_read32(VM_ENTRY_CONTROLS) &
1006 ~VM_ENTRY_CONTROLS_IA32E_MASK);
1007
1008 msr->data = efer & ~EFER_LME;
1009 }
1010 setup_msrs(vcpu);
1011 }
1012
1013 #endif
1014
1015 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1016 {
1017 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1018
1019 return vmcs_readl(sf->base);
1020 }
1021
1022 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1023 struct kvm_segment *var, int seg)
1024 {
1025 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1026 u32 ar;
1027
1028 var->base = vmcs_readl(sf->base);
1029 var->limit = vmcs_read32(sf->limit);
1030 var->selector = vmcs_read16(sf->selector);
1031 ar = vmcs_read32(sf->ar_bytes);
1032 if (ar & AR_UNUSABLE_MASK)
1033 ar = 0;
1034 var->type = ar & 15;
1035 var->s = (ar >> 4) & 1;
1036 var->dpl = (ar >> 5) & 3;
1037 var->present = (ar >> 7) & 1;
1038 var->avl = (ar >> 12) & 1;
1039 var->l = (ar >> 13) & 1;
1040 var->db = (ar >> 14) & 1;
1041 var->g = (ar >> 15) & 1;
1042 var->unusable = (ar >> 16) & 1;
1043 }
1044
1045 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1046 {
1047 u32 ar;
1048
1049 if (var->unusable)
1050 ar = 1 << 16;
1051 else {
1052 ar = var->type & 15;
1053 ar |= (var->s & 1) << 4;
1054 ar |= (var->dpl & 3) << 5;
1055 ar |= (var->present & 1) << 7;
1056 ar |= (var->avl & 1) << 12;
1057 ar |= (var->l & 1) << 13;
1058 ar |= (var->db & 1) << 14;
1059 ar |= (var->g & 1) << 15;
1060 }
1061 if (ar == 0) /* a 0 value means unusable */
1062 ar = AR_UNUSABLE_MASK;
1063
1064 return ar;
1065 }
1066
1067 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1068 struct kvm_segment *var, int seg)
1069 {
1070 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1071 u32 ar;
1072
1073 if (vcpu->rmode.active && seg == VCPU_SREG_TR) {
1074 vcpu->rmode.tr.selector = var->selector;
1075 vcpu->rmode.tr.base = var->base;
1076 vcpu->rmode.tr.limit = var->limit;
1077 vcpu->rmode.tr.ar = vmx_segment_access_rights(var);
1078 return;
1079 }
1080 vmcs_writel(sf->base, var->base);
1081 vmcs_write32(sf->limit, var->limit);
1082 vmcs_write16(sf->selector, var->selector);
1083 if (vcpu->rmode.active && var->s) {
1084 /*
1085 * Hack real-mode segments into vm86 compatibility.
1086 */
1087 if (var->base == 0xffff0000 && var->selector == 0xf000)
1088 vmcs_writel(sf->base, 0xf0000);
1089 ar = 0xf3;
1090 } else
1091 ar = vmx_segment_access_rights(var);
1092 vmcs_write32(sf->ar_bytes, ar);
1093 }
1094
1095 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1096 {
1097 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1098
1099 *db = (ar >> 14) & 1;
1100 *l = (ar >> 13) & 1;
1101 }
1102
1103 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1104 {
1105 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1106 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1107 }
1108
1109 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1110 {
1111 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1112 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1113 }
1114
1115 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1116 {
1117 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1118 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1119 }
1120
1121 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1122 {
1123 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1124 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1125 }
1126
1127 static int init_rmode_tss(struct kvm* kvm)
1128 {
1129 struct page *p1, *p2, *p3;
1130 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1131 char *page;
1132
1133 p1 = gfn_to_page(kvm, fn++);
1134 p2 = gfn_to_page(kvm, fn++);
1135 p3 = gfn_to_page(kvm, fn);
1136
1137 if (!p1 || !p2 || !p3) {
1138 kvm_printf(kvm,"%s: gfn_to_page failed\n", __FUNCTION__);
1139 return 0;
1140 }
1141
1142 page = kmap_atomic(p1, KM_USER0);
1143 memset(page, 0, PAGE_SIZE);
1144 *(u16*)(page + 0x66) = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1145 kunmap_atomic(page, KM_USER0);
1146
1147 page = kmap_atomic(p2, KM_USER0);
1148 memset(page, 0, PAGE_SIZE);
1149 kunmap_atomic(page, KM_USER0);
1150
1151 page = kmap_atomic(p3, KM_USER0);
1152 memset(page, 0, PAGE_SIZE);
1153 *(page + RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1) = ~0;
1154 kunmap_atomic(page, KM_USER0);
1155
1156 return 1;
1157 }
1158
1159 static void vmcs_write32_fixedbits(u32 msr, u32 vmcs_field, u32 val)
1160 {
1161 u32 msr_high, msr_low;
1162
1163 rdmsr(msr, msr_low, msr_high);
1164
1165 val &= msr_high;
1166 val |= msr_low;
1167 vmcs_write32(vmcs_field, val);
1168 }
1169
1170 static void seg_setup(int seg)
1171 {
1172 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1173
1174 vmcs_write16(sf->selector, 0);
1175 vmcs_writel(sf->base, 0);
1176 vmcs_write32(sf->limit, 0xffff);
1177 vmcs_write32(sf->ar_bytes, 0x93);
1178 }
1179
1180 /*
1181 * Sets up the vmcs for emulated real mode.
1182 */
1183 static int vmx_vcpu_setup(struct kvm_vcpu *vcpu)
1184 {
1185 u32 host_sysenter_cs;
1186 u32 junk;
1187 unsigned long a;
1188 struct descriptor_table dt;
1189 int i;
1190 int ret = 0;
1191 unsigned long kvm_vmx_return;
1192
1193 if (!init_rmode_tss(vcpu->kvm)) {
1194 ret = -ENOMEM;
1195 goto out;
1196 }
1197
1198 memset(vcpu->regs, 0, sizeof(vcpu->regs));
1199 vcpu->regs[VCPU_REGS_RDX] = get_rdx_init_val();
1200 vcpu->cr8 = 0;
1201 vcpu->apic_base = 0xfee00000 |
1202 /*for vcpu 0*/ MSR_IA32_APICBASE_BSP |
1203 MSR_IA32_APICBASE_ENABLE;
1204
1205 fx_init(vcpu);
1206
1207 /*
1208 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1209 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
1210 */
1211 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1212 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1213 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1214 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1215
1216 seg_setup(VCPU_SREG_DS);
1217 seg_setup(VCPU_SREG_ES);
1218 seg_setup(VCPU_SREG_FS);
1219 seg_setup(VCPU_SREG_GS);
1220 seg_setup(VCPU_SREG_SS);
1221
1222 vmcs_write16(GUEST_TR_SELECTOR, 0);
1223 vmcs_writel(GUEST_TR_BASE, 0);
1224 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1225 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1226
1227 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1228 vmcs_writel(GUEST_LDTR_BASE, 0);
1229 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1230 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1231
1232 vmcs_write32(GUEST_SYSENTER_CS, 0);
1233 vmcs_writel(GUEST_SYSENTER_ESP, 0);
1234 vmcs_writel(GUEST_SYSENTER_EIP, 0);
1235
1236 vmcs_writel(GUEST_RFLAGS, 0x02);
1237 vmcs_writel(GUEST_RIP, 0xfff0);
1238 vmcs_writel(GUEST_RSP, 0);
1239
1240 //todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0
1241 vmcs_writel(GUEST_DR7, 0x400);
1242
1243 vmcs_writel(GUEST_GDTR_BASE, 0);
1244 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1245
1246 vmcs_writel(GUEST_IDTR_BASE, 0);
1247 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1248
1249 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1250 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1251 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1252
1253 /* I/O */
1254 vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
1255 vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
1256
1257 guest_write_tsc(0);
1258
1259 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1260
1261 /* Special registers */
1262 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1263
1264 /* Control */
1265 vmcs_write32_fixedbits(MSR_IA32_VMX_PINBASED_CTLS,
1266 PIN_BASED_VM_EXEC_CONTROL,
1267 PIN_BASED_EXT_INTR_MASK /* 20.6.1 */
1268 | PIN_BASED_NMI_EXITING /* 20.6.1 */
1269 );
1270 vmcs_write32_fixedbits(MSR_IA32_VMX_PROCBASED_CTLS,
1271 CPU_BASED_VM_EXEC_CONTROL,
1272 CPU_BASED_HLT_EXITING /* 20.6.2 */
1273 | CPU_BASED_CR8_LOAD_EXITING /* 20.6.2 */
1274 | CPU_BASED_CR8_STORE_EXITING /* 20.6.2 */
1275 | CPU_BASED_ACTIVATE_IO_BITMAP /* 20.6.2 */
1276 | CPU_BASED_MOV_DR_EXITING
1277 | CPU_BASED_USE_TSC_OFFSETING /* 21.3 */
1278 );
1279
1280 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
1281 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
1282 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
1283
1284 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
1285 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
1286 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
1287
1288 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
1289 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1290 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1291 vmcs_write16(HOST_FS_SELECTOR, read_fs()); /* 22.2.4 */
1292 vmcs_write16(HOST_GS_SELECTOR, read_gs()); /* 22.2.4 */
1293 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1294 #ifdef CONFIG_X86_64
1295 rdmsrl(MSR_FS_BASE, a);
1296 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1297 rdmsrl(MSR_GS_BASE, a);
1298 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1299 #else
1300 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1301 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1302 #endif
1303
1304 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
1305
1306 get_idt(&dt);
1307 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
1308
1309 asm ("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
1310 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
1311
1312 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1313 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1314 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1315 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
1316 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1317 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
1318
1319 for (i = 0; i < NR_VMX_MSR; ++i) {
1320 u32 index = vmx_msr_index[i];
1321 u32 data_low, data_high;
1322 u64 data;
1323 int j = vcpu->nmsrs;
1324
1325 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1326 continue;
1327 if (wrmsr_safe(index, data_low, data_high) < 0)
1328 continue;
1329 data = data_low | ((u64)data_high << 32);
1330 vcpu->host_msrs[j].index = index;
1331 vcpu->host_msrs[j].reserved = 0;
1332 vcpu->host_msrs[j].data = data;
1333 vcpu->guest_msrs[j] = vcpu->host_msrs[j];
1334 #ifdef CONFIG_X86_64
1335 if (index == MSR_KERNEL_GS_BASE)
1336 msr_offset_kernel_gs_base = j;
1337 #endif
1338 ++vcpu->nmsrs;
1339 }
1340
1341 setup_msrs(vcpu);
1342
1343 vmcs_write32_fixedbits(MSR_IA32_VMX_EXIT_CTLS, VM_EXIT_CONTROLS,
1344 (HOST_IS_64 << 9)); /* 22.2,1, 20.7.1 */
1345
1346 /* 22.2.1, 20.8.1 */
1347 vmcs_write32_fixedbits(MSR_IA32_VMX_ENTRY_CTLS,
1348 VM_ENTRY_CONTROLS, 0);
1349 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
1350
1351 #ifdef CONFIG_X86_64
1352 vmcs_writel(VIRTUAL_APIC_PAGE_ADDR, 0);
1353 vmcs_writel(TPR_THRESHOLD, 0);
1354 #endif
1355
1356 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
1357 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1358
1359 vcpu->cr0 = 0x60000010;
1360 vmx_set_cr0(vcpu, vcpu->cr0); // enter rmode
1361 vmx_set_cr4(vcpu, 0);
1362 #ifdef CONFIG_X86_64
1363 vmx_set_efer(vcpu, 0);
1364 #endif
1365 vmx_fpu_activate(vcpu);
1366 update_exception_bitmap(vcpu);
1367
1368 return 0;
1369
1370 out:
1371 return ret;
1372 }
1373
1374 static void inject_rmode_irq(struct kvm_vcpu *vcpu, int irq)
1375 {
1376 u16 ent[2];
1377 u16 cs;
1378 u16 ip;
1379 unsigned long flags;
1380 unsigned long ss_base = vmcs_readl(GUEST_SS_BASE);
1381 u16 sp = vmcs_readl(GUEST_RSP);
1382 u32 ss_limit = vmcs_read32(GUEST_SS_LIMIT);
1383
1384 if (sp > ss_limit || sp < 6 ) {
1385 vcpu_printf(vcpu, "%s: #SS, rsp 0x%lx ss 0x%lx limit 0x%x\n",
1386 __FUNCTION__,
1387 vmcs_readl(GUEST_RSP),
1388 vmcs_readl(GUEST_SS_BASE),
1389 vmcs_read32(GUEST_SS_LIMIT));
1390 return;
1391 }
1392
1393 if (kvm_read_guest(vcpu, irq * sizeof(ent), sizeof(ent), &ent) !=
1394 sizeof(ent)) {
1395 vcpu_printf(vcpu, "%s: read guest err\n", __FUNCTION__);
1396 return;
1397 }
1398
1399 flags = vmcs_readl(GUEST_RFLAGS);
1400 cs = vmcs_readl(GUEST_CS_BASE) >> 4;
1401 ip = vmcs_readl(GUEST_RIP);
1402
1403
1404 if (kvm_write_guest(vcpu, ss_base + sp - 2, 2, &flags) != 2 ||
1405 kvm_write_guest(vcpu, ss_base + sp - 4, 2, &cs) != 2 ||
1406 kvm_write_guest(vcpu, ss_base + sp - 6, 2, &ip) != 2) {
1407 vcpu_printf(vcpu, "%s: write guest err\n", __FUNCTION__);
1408 return;
1409 }
1410
1411 vmcs_writel(GUEST_RFLAGS, flags &
1412 ~( X86_EFLAGS_IF | X86_EFLAGS_AC | X86_EFLAGS_TF));
1413 vmcs_write16(GUEST_CS_SELECTOR, ent[1]) ;
1414 vmcs_writel(GUEST_CS_BASE, ent[1] << 4);
1415 vmcs_writel(GUEST_RIP, ent[0]);
1416 vmcs_writel(GUEST_RSP, (vmcs_readl(GUEST_RSP) & ~0xffff) | (sp - 6));
1417 }
1418
1419 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1420 {
1421 int word_index = __ffs(vcpu->irq_summary);
1422 int bit_index = __ffs(vcpu->irq_pending[word_index]);
1423 int irq = word_index * BITS_PER_LONG + bit_index;
1424
1425 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1426 if (!vcpu->irq_pending[word_index])
1427 clear_bit(word_index, &vcpu->irq_summary);
1428
1429 if (vcpu->rmode.active) {
1430 inject_rmode_irq(vcpu, irq);
1431 return;
1432 }
1433 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1434 irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1435 }
1436
1437
1438 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1439 struct kvm_run *kvm_run)
1440 {
1441 u32 cpu_based_vm_exec_control;
1442
1443 vcpu->interrupt_window_open =
1444 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
1445 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1446
1447 if (vcpu->interrupt_window_open &&
1448 vcpu->irq_summary &&
1449 !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
1450 /*
1451 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1452 */
1453 kvm_do_inject_irq(vcpu);
1454
1455 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
1456 if (!vcpu->interrupt_window_open &&
1457 (vcpu->irq_summary || kvm_run->request_interrupt_window))
1458 /*
1459 * Interrupts blocked. Wait for unblock.
1460 */
1461 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
1462 else
1463 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
1464 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
1465 }
1466
1467 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1468 {
1469 struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1470
1471 set_debugreg(dbg->bp[0], 0);
1472 set_debugreg(dbg->bp[1], 1);
1473 set_debugreg(dbg->bp[2], 2);
1474 set_debugreg(dbg->bp[3], 3);
1475
1476 if (dbg->singlestep) {
1477 unsigned long flags;
1478
1479 flags = vmcs_readl(GUEST_RFLAGS);
1480 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1481 vmcs_writel(GUEST_RFLAGS, flags);
1482 }
1483 }
1484
1485 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1486 int vec, u32 err_code)
1487 {
1488 if (!vcpu->rmode.active)
1489 return 0;
1490
1491 /*
1492 * Instruction with address size override prefix opcode 0x67
1493 * Cause the #SS fault with 0 error code in VM86 mode.
1494 */
1495 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
1496 if (emulate_instruction(vcpu, NULL, 0, 0) == EMULATE_DONE)
1497 return 1;
1498 return 0;
1499 }
1500
1501 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1502 {
1503 u32 intr_info, error_code;
1504 unsigned long cr2, rip;
1505 u32 vect_info;
1506 enum emulation_result er;
1507 int r;
1508
1509 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1510 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1511
1512 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
1513 !is_page_fault(intr_info)) {
1514 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
1515 "intr info 0x%x\n", __FUNCTION__, vect_info, intr_info);
1516 }
1517
1518 if (is_external_interrupt(vect_info)) {
1519 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
1520 set_bit(irq, vcpu->irq_pending);
1521 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
1522 }
1523
1524 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) { /* nmi */
1525 asm ("int $2");
1526 return 1;
1527 }
1528
1529 if (is_no_device(intr_info)) {
1530 vmx_fpu_activate(vcpu);
1531 return 1;
1532 }
1533
1534 error_code = 0;
1535 rip = vmcs_readl(GUEST_RIP);
1536 if (intr_info & INTR_INFO_DELIEVER_CODE_MASK)
1537 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1538 if (is_page_fault(intr_info)) {
1539 cr2 = vmcs_readl(EXIT_QUALIFICATION);
1540
1541 spin_lock(&vcpu->kvm->lock);
1542 r = kvm_mmu_page_fault(vcpu, cr2, error_code);
1543 if (r < 0) {
1544 spin_unlock(&vcpu->kvm->lock);
1545 return r;
1546 }
1547 if (!r) {
1548 spin_unlock(&vcpu->kvm->lock);
1549 return 1;
1550 }
1551
1552 er = emulate_instruction(vcpu, kvm_run, cr2, error_code);
1553 spin_unlock(&vcpu->kvm->lock);
1554
1555 switch (er) {
1556 case EMULATE_DONE:
1557 return 1;
1558 case EMULATE_DO_MMIO:
1559 ++vcpu->stat.mmio_exits;
1560 kvm_run->exit_reason = KVM_EXIT_MMIO;
1561 return 0;
1562 case EMULATE_FAIL:
1563 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
1564 break;
1565 default:
1566 BUG();
1567 }
1568 }
1569
1570 if (vcpu->rmode.active &&
1571 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
1572 error_code))
1573 return 1;
1574
1575 if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) == (INTR_TYPE_EXCEPTION | 1)) {
1576 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1577 return 0;
1578 }
1579 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1580 kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1581 kvm_run->ex.error_code = error_code;
1582 return 0;
1583 }
1584
1585 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1586 struct kvm_run *kvm_run)
1587 {
1588 ++vcpu->stat.irq_exits;
1589 return 1;
1590 }
1591
1592 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1593 {
1594 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1595 return 0;
1596 }
1597
1598 static int get_io_count(struct kvm_vcpu *vcpu, unsigned long *count)
1599 {
1600 u64 inst;
1601 gva_t rip;
1602 int countr_size;
1603 int i, n;
1604
1605 if ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_VM)) {
1606 countr_size = 2;
1607 } else {
1608 u32 cs_ar = vmcs_read32(GUEST_CS_AR_BYTES);
1609
1610 countr_size = (cs_ar & AR_L_MASK) ? 8:
1611 (cs_ar & AR_DB_MASK) ? 4: 2;
1612 }
1613
1614 rip = vmcs_readl(GUEST_RIP);
1615 if (countr_size != 8)
1616 rip += vmcs_readl(GUEST_CS_BASE);
1617
1618 n = kvm_read_guest(vcpu, rip, sizeof(inst), &inst);
1619
1620 for (i = 0; i < n; i++) {
1621 switch (((u8*)&inst)[i]) {
1622 case 0xf0:
1623 case 0xf2:
1624 case 0xf3:
1625 case 0x2e:
1626 case 0x36:
1627 case 0x3e:
1628 case 0x26:
1629 case 0x64:
1630 case 0x65:
1631 case 0x66:
1632 break;
1633 case 0x67:
1634 countr_size = (countr_size == 2) ? 4: (countr_size >> 1);
1635 default:
1636 goto done;
1637 }
1638 }
1639 return 0;
1640 done:
1641 countr_size *= 8;
1642 *count = vcpu->regs[VCPU_REGS_RCX] & (~0ULL >> (64 - countr_size));
1643 //printk("cx: %lx\n", vcpu->regs[VCPU_REGS_RCX]);
1644 return 1;
1645 }
1646
1647 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1648 {
1649 u64 exit_qualification;
1650 int size, down, in, string, rep;
1651 unsigned port;
1652 unsigned long count;
1653 gva_t address;
1654
1655 ++vcpu->stat.io_exits;
1656 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1657 in = (exit_qualification & 8) != 0;
1658 size = (exit_qualification & 7) + 1;
1659 string = (exit_qualification & 16) != 0;
1660 down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
1661 count = 1;
1662 rep = (exit_qualification & 32) != 0;
1663 port = exit_qualification >> 16;
1664 address = 0;
1665 if (string) {
1666 if (rep && !get_io_count(vcpu, &count))
1667 return 1;
1668 address = vmcs_readl(GUEST_LINEAR_ADDRESS);
1669 }
1670 return kvm_setup_pio(vcpu, kvm_run, in, size, count, string, down,
1671 address, rep, port);
1672 }
1673
1674 static void
1675 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1676 {
1677 /*
1678 * Patch in the VMCALL instruction:
1679 */
1680 hypercall[0] = 0x0f;
1681 hypercall[1] = 0x01;
1682 hypercall[2] = 0xc1;
1683 hypercall[3] = 0xc3;
1684 }
1685
1686 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1687 {
1688 u64 exit_qualification;
1689 int cr;
1690 int reg;
1691
1692 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1693 cr = exit_qualification & 15;
1694 reg = (exit_qualification >> 8) & 15;
1695 switch ((exit_qualification >> 4) & 3) {
1696 case 0: /* mov to cr */
1697 switch (cr) {
1698 case 0:
1699 vcpu_load_rsp_rip(vcpu);
1700 set_cr0(vcpu, vcpu->regs[reg]);
1701 skip_emulated_instruction(vcpu);
1702 return 1;
1703 case 3:
1704 vcpu_load_rsp_rip(vcpu);
1705 set_cr3(vcpu, vcpu->regs[reg]);
1706 skip_emulated_instruction(vcpu);
1707 return 1;
1708 case 4:
1709 vcpu_load_rsp_rip(vcpu);
1710 set_cr4(vcpu, vcpu->regs[reg]);
1711 skip_emulated_instruction(vcpu);
1712 return 1;
1713 case 8:
1714 vcpu_load_rsp_rip(vcpu);
1715 set_cr8(vcpu, vcpu->regs[reg]);
1716 skip_emulated_instruction(vcpu);
1717 return 1;
1718 };
1719 break;
1720 case 2: /* clts */
1721 vcpu_load_rsp_rip(vcpu);
1722 vmx_fpu_deactivate(vcpu);
1723 vcpu->cr0 &= ~CR0_TS_MASK;
1724 vmcs_writel(CR0_READ_SHADOW, vcpu->cr0);
1725 vmx_fpu_activate(vcpu);
1726 skip_emulated_instruction(vcpu);
1727 return 1;
1728 case 1: /*mov from cr*/
1729 switch (cr) {
1730 case 3:
1731 vcpu_load_rsp_rip(vcpu);
1732 vcpu->regs[reg] = vcpu->cr3;
1733 vcpu_put_rsp_rip(vcpu);
1734 skip_emulated_instruction(vcpu);
1735 return 1;
1736 case 8:
1737 vcpu_load_rsp_rip(vcpu);
1738 vcpu->regs[reg] = vcpu->cr8;
1739 vcpu_put_rsp_rip(vcpu);
1740 skip_emulated_instruction(vcpu);
1741 return 1;
1742 }
1743 break;
1744 case 3: /* lmsw */
1745 lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
1746
1747 skip_emulated_instruction(vcpu);
1748 return 1;
1749 default:
1750 break;
1751 }
1752 kvm_run->exit_reason = 0;
1753 printk(KERN_ERR "kvm: unhandled control register: op %d cr %d\n",
1754 (int)(exit_qualification >> 4) & 3, cr);
1755 return 0;
1756 }
1757
1758 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1759 {
1760 u64 exit_qualification;
1761 unsigned long val;
1762 int dr, reg;
1763
1764 /*
1765 * FIXME: this code assumes the host is debugging the guest.
1766 * need to deal with guest debugging itself too.
1767 */
1768 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1769 dr = exit_qualification & 7;
1770 reg = (exit_qualification >> 8) & 15;
1771 vcpu_load_rsp_rip(vcpu);
1772 if (exit_qualification & 16) {
1773 /* mov from dr */
1774 switch (dr) {
1775 case 6:
1776 val = 0xffff0ff0;
1777 break;
1778 case 7:
1779 val = 0x400;
1780 break;
1781 default:
1782 val = 0;
1783 }
1784 vcpu->regs[reg] = val;
1785 } else {
1786 /* mov to dr */
1787 }
1788 vcpu_put_rsp_rip(vcpu);
1789 skip_emulated_instruction(vcpu);
1790 return 1;
1791 }
1792
1793 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1794 {
1795 kvm_emulate_cpuid(vcpu);
1796 return 1;
1797 }
1798
1799 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1800 {
1801 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1802 u64 data;
1803
1804 if (vmx_get_msr(vcpu, ecx, &data)) {
1805 vmx_inject_gp(vcpu, 0);
1806 return 1;
1807 }
1808
1809 /* FIXME: handling of bits 32:63 of rax, rdx */
1810 vcpu->regs[VCPU_REGS_RAX] = data & -1u;
1811 vcpu->regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
1812 skip_emulated_instruction(vcpu);
1813 return 1;
1814 }
1815
1816 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1817 {
1818 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1819 u64 data = (vcpu->regs[VCPU_REGS_RAX] & -1u)
1820 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1821
1822 if (vmx_set_msr(vcpu, ecx, data) != 0) {
1823 vmx_inject_gp(vcpu, 0);
1824 return 1;
1825 }
1826
1827 skip_emulated_instruction(vcpu);
1828 return 1;
1829 }
1830
1831 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1832 struct kvm_run *kvm_run)
1833 {
1834 kvm_run->if_flag = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) != 0;
1835 kvm_run->cr8 = vcpu->cr8;
1836 kvm_run->apic_base = vcpu->apic_base;
1837 kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1838 vcpu->irq_summary == 0);
1839 }
1840
1841 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
1842 struct kvm_run *kvm_run)
1843 {
1844 /*
1845 * If the user space waits to inject interrupts, exit as soon as
1846 * possible
1847 */
1848 if (kvm_run->request_interrupt_window &&
1849 !vcpu->irq_summary) {
1850 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1851 ++vcpu->stat.irq_window_exits;
1852 return 0;
1853 }
1854 return 1;
1855 }
1856
1857 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1858 {
1859 skip_emulated_instruction(vcpu);
1860 if (vcpu->irq_summary)
1861 return 1;
1862
1863 kvm_run->exit_reason = KVM_EXIT_HLT;
1864 ++vcpu->stat.halt_exits;
1865 return 0;
1866 }
1867
1868 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1869 {
1870 skip_emulated_instruction(vcpu);
1871 return kvm_hypercall(vcpu, kvm_run);
1872 }
1873
1874 /*
1875 * The exit handlers return 1 if the exit was handled fully and guest execution
1876 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
1877 * to be done to userspace and return 0.
1878 */
1879 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
1880 struct kvm_run *kvm_run) = {
1881 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
1882 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
1883 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
1884 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
1885 [EXIT_REASON_CR_ACCESS] = handle_cr,
1886 [EXIT_REASON_DR_ACCESS] = handle_dr,
1887 [EXIT_REASON_CPUID] = handle_cpuid,
1888 [EXIT_REASON_MSR_READ] = handle_rdmsr,
1889 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
1890 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
1891 [EXIT_REASON_HLT] = handle_halt,
1892 [EXIT_REASON_VMCALL] = handle_vmcall,
1893 };
1894
1895 static const int kvm_vmx_max_exit_handlers =
1896 sizeof(kvm_vmx_exit_handlers) / sizeof(*kvm_vmx_exit_handlers);
1897
1898 /*
1899 * The guest has exited. See if we can fix it or if we need userspace
1900 * assistance.
1901 */
1902 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1903 {
1904 u32 vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1905 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
1906
1907 if ( (vectoring_info & VECTORING_INFO_VALID_MASK) &&
1908 exit_reason != EXIT_REASON_EXCEPTION_NMI )
1909 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
1910 "exit reason is 0x%x\n", __FUNCTION__, exit_reason);
1911 if (exit_reason < kvm_vmx_max_exit_handlers
1912 && kvm_vmx_exit_handlers[exit_reason])
1913 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
1914 else {
1915 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1916 kvm_run->hw.hardware_exit_reason = exit_reason;
1917 }
1918 return 0;
1919 }
1920
1921 /*
1922 * Check if userspace requested an interrupt window, and that the
1923 * interrupt window is open.
1924 *
1925 * No need to exit to userspace if we already have an interrupt queued.
1926 */
1927 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1928 struct kvm_run *kvm_run)
1929 {
1930 return (!vcpu->irq_summary &&
1931 kvm_run->request_interrupt_window &&
1932 vcpu->interrupt_window_open &&
1933 (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF));
1934 }
1935
1936 static int vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1937 {
1938 u8 fail;
1939 int r;
1940
1941 preempted:
1942 if (!vcpu->mmio_read_completed)
1943 do_interrupt_requests(vcpu, kvm_run);
1944
1945 if (vcpu->guest_debug.enabled)
1946 kvm_guest_debug_pre(vcpu);
1947
1948 again:
1949 vmx_save_host_state(vcpu);
1950 kvm_load_guest_fpu(vcpu);
1951
1952 /*
1953 * Loading guest fpu may have cleared host cr0.ts
1954 */
1955 vmcs_writel(HOST_CR0, read_cr0());
1956
1957 asm (
1958 /* Store host registers */
1959 "pushf \n\t"
1960 #ifdef CONFIG_X86_64
1961 "push %%rax; push %%rbx; push %%rdx;"
1962 "push %%rsi; push %%rdi; push %%rbp;"
1963 "push %%r8; push %%r9; push %%r10; push %%r11;"
1964 "push %%r12; push %%r13; push %%r14; push %%r15;"
1965 "push %%rcx \n\t"
1966 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
1967 #else
1968 "pusha; push %%ecx \n\t"
1969 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
1970 #endif
1971 /* Check if vmlaunch of vmresume is needed */
1972 "cmp $0, %1 \n\t"
1973 /* Load guest registers. Don't clobber flags. */
1974 #ifdef CONFIG_X86_64
1975 "mov %c[cr2](%3), %%rax \n\t"
1976 "mov %%rax, %%cr2 \n\t"
1977 "mov %c[rax](%3), %%rax \n\t"
1978 "mov %c[rbx](%3), %%rbx \n\t"
1979 "mov %c[rdx](%3), %%rdx \n\t"
1980 "mov %c[rsi](%3), %%rsi \n\t"
1981 "mov %c[rdi](%3), %%rdi \n\t"
1982 "mov %c[rbp](%3), %%rbp \n\t"
1983 "mov %c[r8](%3), %%r8 \n\t"
1984 "mov %c[r9](%3), %%r9 \n\t"
1985 "mov %c[r10](%3), %%r10 \n\t"
1986 "mov %c[r11](%3), %%r11 \n\t"
1987 "mov %c[r12](%3), %%r12 \n\t"
1988 "mov %c[r13](%3), %%r13 \n\t"
1989 "mov %c[r14](%3), %%r14 \n\t"
1990 "mov %c[r15](%3), %%r15 \n\t"
1991 "mov %c[rcx](%3), %%rcx \n\t" /* kills %3 (rcx) */
1992 #else
1993 "mov %c[cr2](%3), %%eax \n\t"
1994 "mov %%eax, %%cr2 \n\t"
1995 "mov %c[rax](%3), %%eax \n\t"
1996 "mov %c[rbx](%3), %%ebx \n\t"
1997 "mov %c[rdx](%3), %%edx \n\t"
1998 "mov %c[rsi](%3), %%esi \n\t"
1999 "mov %c[rdi](%3), %%edi \n\t"
2000 "mov %c[rbp](%3), %%ebp \n\t"
2001 "mov %c[rcx](%3), %%ecx \n\t" /* kills %3 (ecx) */
2002 #endif
2003 /* Enter guest mode */
2004 "jne .Llaunched \n\t"
2005 ASM_VMX_VMLAUNCH "\n\t"
2006 "jmp .Lkvm_vmx_return \n\t"
2007 ".Llaunched: " ASM_VMX_VMRESUME "\n\t"
2008 ".Lkvm_vmx_return: "
2009 /* Save guest registers, load host registers, keep flags */
2010 #ifdef CONFIG_X86_64
2011 "xchg %3, (%%rsp) \n\t"
2012 "mov %%rax, %c[rax](%3) \n\t"
2013 "mov %%rbx, %c[rbx](%3) \n\t"
2014 "pushq (%%rsp); popq %c[rcx](%3) \n\t"
2015 "mov %%rdx, %c[rdx](%3) \n\t"
2016 "mov %%rsi, %c[rsi](%3) \n\t"
2017 "mov %%rdi, %c[rdi](%3) \n\t"
2018 "mov %%rbp, %c[rbp](%3) \n\t"
2019 "mov %%r8, %c[r8](%3) \n\t"
2020 "mov %%r9, %c[r9](%3) \n\t"
2021 "mov %%r10, %c[r10](%3) \n\t"
2022 "mov %%r11, %c[r11](%3) \n\t"
2023 "mov %%r12, %c[r12](%3) \n\t"
2024 "mov %%r13, %c[r13](%3) \n\t"
2025 "mov %%r14, %c[r14](%3) \n\t"
2026 "mov %%r15, %c[r15](%3) \n\t"
2027 "mov %%cr2, %%rax \n\t"
2028 "mov %%rax, %c[cr2](%3) \n\t"
2029 "mov (%%rsp), %3 \n\t"
2030
2031 "pop %%rcx; pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
2032 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
2033 "pop %%rbp; pop %%rdi; pop %%rsi;"
2034 "pop %%rdx; pop %%rbx; pop %%rax \n\t"
2035 #else
2036 "xchg %3, (%%esp) \n\t"
2037 "mov %%eax, %c[rax](%3) \n\t"
2038 "mov %%ebx, %c[rbx](%3) \n\t"
2039 "pushl (%%esp); popl %c[rcx](%3) \n\t"
2040 "mov %%edx, %c[rdx](%3) \n\t"
2041 "mov %%esi, %c[rsi](%3) \n\t"
2042 "mov %%edi, %c[rdi](%3) \n\t"
2043 "mov %%ebp, %c[rbp](%3) \n\t"
2044 "mov %%cr2, %%eax \n\t"
2045 "mov %%eax, %c[cr2](%3) \n\t"
2046 "mov (%%esp), %3 \n\t"
2047
2048 "pop %%ecx; popa \n\t"
2049 #endif
2050 "setbe %0 \n\t"
2051 "popf \n\t"
2052 : "=q" (fail)
2053 : "r"(vcpu->launched), "d"((unsigned long)HOST_RSP),
2054 "c"(vcpu),
2055 [rax]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RAX])),
2056 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
2057 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
2058 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
2059 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
2060 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
2061 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP])),
2062 #ifdef CONFIG_X86_64
2063 [r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
2064 [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
2065 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
2066 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
2067 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
2068 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
2069 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
2070 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15])),
2071 #endif
2072 [cr2]"i"(offsetof(struct kvm_vcpu, cr2))
2073 : "cc", "memory" );
2074
2075 ++vcpu->stat.exits;
2076
2077 vcpu->interrupt_window_open = (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
2078
2079 asm ("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
2080
2081 if (unlikely(fail)) {
2082 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2083 kvm_run->fail_entry.hardware_entry_failure_reason
2084 = vmcs_read32(VM_INSTRUCTION_ERROR);
2085 r = 0;
2086 goto out;
2087 }
2088 /*
2089 * Profile KVM exit RIPs:
2090 */
2091 if (unlikely(prof_on == KVM_PROFILING))
2092 profile_hit(KVM_PROFILING, (void *)vmcs_readl(GUEST_RIP));
2093
2094 vcpu->launched = 1;
2095 r = kvm_handle_exit(kvm_run, vcpu);
2096 if (r > 0) {
2097 /* Give scheduler a change to reschedule. */
2098 if (signal_pending(current)) {
2099 r = -EINTR;
2100 kvm_run->exit_reason = KVM_EXIT_INTR;
2101 ++vcpu->stat.signal_exits;
2102 goto out;
2103 }
2104
2105 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2106 r = -EINTR;
2107 kvm_run->exit_reason = KVM_EXIT_INTR;
2108 ++vcpu->stat.request_irq_exits;
2109 goto out;
2110 }
2111 if (!need_resched()) {
2112 ++vcpu->stat.light_exits;
2113 goto again;
2114 }
2115 }
2116
2117 out:
2118 if (r > 0) {
2119 kvm_resched(vcpu);
2120 goto preempted;
2121 }
2122
2123 post_kvm_run_save(vcpu, kvm_run);
2124 return r;
2125 }
2126
2127 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2128 {
2129 vmcs_writel(GUEST_CR3, vmcs_readl(GUEST_CR3));
2130 }
2131
2132 static void vmx_inject_page_fault(struct kvm_vcpu *vcpu,
2133 unsigned long addr,
2134 u32 err_code)
2135 {
2136 u32 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2137
2138 ++vcpu->stat.pf_guest;
2139
2140 if (is_page_fault(vect_info)) {
2141 printk(KERN_DEBUG "inject_page_fault: "
2142 "double fault 0x%lx @ 0x%lx\n",
2143 addr, vmcs_readl(GUEST_RIP));
2144 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, 0);
2145 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2146 DF_VECTOR |
2147 INTR_TYPE_EXCEPTION |
2148 INTR_INFO_DELIEVER_CODE_MASK |
2149 INTR_INFO_VALID_MASK);
2150 return;
2151 }
2152 vcpu->cr2 = addr;
2153 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, err_code);
2154 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2155 PF_VECTOR |
2156 INTR_TYPE_EXCEPTION |
2157 INTR_INFO_DELIEVER_CODE_MASK |
2158 INTR_INFO_VALID_MASK);
2159
2160 }
2161
2162 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
2163 {
2164 if (vcpu->vmcs) {
2165 on_each_cpu(__vcpu_clear, vcpu, 0, 1);
2166 free_vmcs(vcpu->vmcs);
2167 vcpu->vmcs = NULL;
2168 }
2169 }
2170
2171 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
2172 {
2173 vmx_free_vmcs(vcpu);
2174 }
2175
2176 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
2177 {
2178 struct vmcs *vmcs;
2179
2180 vcpu->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2181 if (!vcpu->guest_msrs)
2182 return -ENOMEM;
2183
2184 vcpu->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2185 if (!vcpu->host_msrs)
2186 goto out_free_guest_msrs;
2187
2188 vmcs = alloc_vmcs();
2189 if (!vmcs)
2190 goto out_free_msrs;
2191
2192 vmcs_clear(vmcs);
2193 vcpu->vmcs = vmcs;
2194 vcpu->launched = 0;
2195
2196 return 0;
2197
2198 out_free_msrs:
2199 kfree(vcpu->host_msrs);
2200 vcpu->host_msrs = NULL;
2201
2202 out_free_guest_msrs:
2203 kfree(vcpu->guest_msrs);
2204 vcpu->guest_msrs = NULL;
2205
2206 return -ENOMEM;
2207 }
2208
2209 static struct kvm_arch_ops vmx_arch_ops = {
2210 .cpu_has_kvm_support = cpu_has_kvm_support,
2211 .disabled_by_bios = vmx_disabled_by_bios,
2212 .hardware_setup = hardware_setup,
2213 .hardware_unsetup = hardware_unsetup,
2214 .hardware_enable = hardware_enable,
2215 .hardware_disable = hardware_disable,
2216
2217 .vcpu_create = vmx_create_vcpu,
2218 .vcpu_free = vmx_free_vcpu,
2219
2220 .vcpu_load = vmx_vcpu_load,
2221 .vcpu_put = vmx_vcpu_put,
2222 .vcpu_decache = vmx_vcpu_decache,
2223
2224 .set_guest_debug = set_guest_debug,
2225 .get_msr = vmx_get_msr,
2226 .set_msr = vmx_set_msr,
2227 .get_segment_base = vmx_get_segment_base,
2228 .get_segment = vmx_get_segment,
2229 .set_segment = vmx_set_segment,
2230 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
2231 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
2232 .set_cr0 = vmx_set_cr0,
2233 .set_cr3 = vmx_set_cr3,
2234 .set_cr4 = vmx_set_cr4,
2235 #ifdef CONFIG_X86_64
2236 .set_efer = vmx_set_efer,
2237 #endif
2238 .get_idt = vmx_get_idt,
2239 .set_idt = vmx_set_idt,
2240 .get_gdt = vmx_get_gdt,
2241 .set_gdt = vmx_set_gdt,
2242 .cache_regs = vcpu_load_rsp_rip,
2243 .decache_regs = vcpu_put_rsp_rip,
2244 .get_rflags = vmx_get_rflags,
2245 .set_rflags = vmx_set_rflags,
2246
2247 .tlb_flush = vmx_flush_tlb,
2248 .inject_page_fault = vmx_inject_page_fault,
2249
2250 .inject_gp = vmx_inject_gp,
2251
2252 .run = vmx_vcpu_run,
2253 .skip_emulated_instruction = skip_emulated_instruction,
2254 .vcpu_setup = vmx_vcpu_setup,
2255 .patch_hypercall = vmx_patch_hypercall,
2256 };
2257
2258 static int __init vmx_init(void)
2259 {
2260 void *iova;
2261 int r;
2262
2263 vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2264 if (!vmx_io_bitmap_a)
2265 return -ENOMEM;
2266
2267 vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2268 if (!vmx_io_bitmap_b) {
2269 r = -ENOMEM;
2270 goto out;
2271 }
2272
2273 /*
2274 * Allow direct access to the PC debug port (it is often used for I/O
2275 * delays, but the vmexits simply slow things down).
2276 */
2277 iova = kmap(vmx_io_bitmap_a);
2278 memset(iova, 0xff, PAGE_SIZE);
2279 clear_bit(0x80, iova);
2280 kunmap(vmx_io_bitmap_a);
2281
2282 iova = kmap(vmx_io_bitmap_b);
2283 memset(iova, 0xff, PAGE_SIZE);
2284 kunmap(vmx_io_bitmap_b);
2285
2286 r = kvm_init_arch(&vmx_arch_ops, THIS_MODULE);
2287 if (r)
2288 goto out1;
2289
2290 return 0;
2291
2292 out1:
2293 __free_page(vmx_io_bitmap_b);
2294 out:
2295 __free_page(vmx_io_bitmap_a);
2296 return r;
2297 }
2298
2299 static void __exit vmx_exit(void)
2300 {
2301 __free_page(vmx_io_bitmap_b);
2302 __free_page(vmx_io_bitmap_a);
2303
2304 kvm_exit_arch();
2305 }
2306
2307 module_init(vmx_init)
2308 module_exit(vmx_exit)
Linux kernel - Deliverables
This page took 0.147969 seconds and 6 git commands to generate.