Merge branch 'qemu-cvs'
[qemu-kvm/fedora.git] / target-i386 / kvm.c
blobac215eba2e6c7ba133e5d8617111e6c6d657cfe0
1 /*
2 * QEMU KVM support
4 * Copyright (C) 2006-2008 Qumranet Technologies
5 * Copyright IBM, Corp. 2008
7 * Authors:
8 * Anthony Liguori <aliguori@us.ibm.com>
10 * This work is licensed under the terms of the GNU GPL, version 2 or later.
11 * See the COPYING file in the top-level directory.
15 #include <sys/types.h>
16 #include <sys/ioctl.h>
17 #include <sys/mman.h>
19 #include <linux/kvm.h>
21 #include "qemu-common.h"
22 #include "sysemu.h"
23 #include "kvm.h"
24 #include "cpu.h"
26 //#define DEBUG_KVM
28 #ifdef DEBUG_KVM
29 #define dprintf(fmt, ...) \
30 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
31 #else
32 #define dprintf(fmt, ...) \
33 do { } while (0)
34 #endif
36 int kvm_arch_init_vcpu(CPUState *env)
38 struct {
39 struct kvm_cpuid cpuid;
40 struct kvm_cpuid_entry entries[100];
41 } __attribute__((packed)) cpuid_data;
42 uint32_t limit, i, cpuid_i;
43 uint32_t eax, ebx, ecx, edx;
45 cpuid_i = 0;
47 cpu_x86_cpuid(env, 0, &eax, &ebx, &ecx, &edx);
48 limit = eax;
50 for (i = 0; i <= limit; i++) {
51 struct kvm_cpuid_entry *c = &cpuid_data.entries[cpuid_i++];
53 cpu_x86_cpuid(env, i, &eax, &ebx, &ecx, &edx);
54 c->function = i;
55 c->eax = eax;
56 c->ebx = ebx;
57 c->ecx = ecx;
58 c->edx = edx;
61 cpu_x86_cpuid(env, 0x80000000, &eax, &ebx, &ecx, &edx);
62 limit = eax;
64 for (i = 0x80000000; i <= limit; i++) {
65 struct kvm_cpuid_entry *c = &cpuid_data.entries[cpuid_i++];
67 cpu_x86_cpuid(env, i, &eax, &ebx, &ecx, &edx);
68 c->function = i;
69 c->eax = eax;
70 c->ebx = ebx;
71 c->ecx = ecx;
72 c->edx = edx;
75 cpuid_data.cpuid.nent = cpuid_i;
77 return kvm_vcpu_ioctl(env, KVM_SET_CPUID, &cpuid_data);
80 static int kvm_has_msr_star(CPUState *env)
82 static int has_msr_star;
83 int ret;
85 /* first time */
86 if (has_msr_star == 0) {
87 struct kvm_msr_list msr_list, *kvm_msr_list;
89 has_msr_star = -1;
91 /* Obtain MSR list from KVM. These are the MSRs that we must
92 * save/restore */
93 msr_list.nmsrs = 0;
94 ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, &msr_list);
95 if (ret < 0)
96 return 0;
98 kvm_msr_list = qemu_mallocz(sizeof(msr_list) +
99 msr_list.nmsrs * sizeof(msr_list.indices[0]));
100 if (kvm_msr_list == NULL)
101 return 0;
103 kvm_msr_list->nmsrs = msr_list.nmsrs;
104 ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
105 if (ret >= 0) {
106 int i;
108 for (i = 0; i < kvm_msr_list->nmsrs; i++) {
109 if (kvm_msr_list->indices[i] == MSR_STAR) {
110 has_msr_star = 1;
111 break;
116 free(kvm_msr_list);
119 if (has_msr_star == 1)
120 return 1;
121 return 0;
124 int kvm_arch_init(KVMState *s, int smp_cpus)
126 int ret;
128 /* create vm86 tss. KVM uses vm86 mode to emulate 16-bit code
129 * directly. In order to use vm86 mode, a TSS is needed. Since this
130 * must be part of guest physical memory, we need to allocate it. Older
131 * versions of KVM just assumed that it would be at the end of physical
132 * memory but that doesn't work with more than 4GB of memory. We simply
133 * refuse to work with those older versions of KVM. */
134 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_SET_TSS_ADDR);
135 if (ret <= 0) {
136 fprintf(stderr, "kvm does not support KVM_CAP_SET_TSS_ADDR\n");
137 return ret;
140 /* this address is 3 pages before the bios, and the bios should present
141 * as unavaible memory. FIXME, need to ensure the e820 map deals with
142 * this?
144 return kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, 0xfffbd000);
147 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
149 lhs->selector = rhs->selector;
150 lhs->base = rhs->base;
151 lhs->limit = rhs->limit;
152 lhs->type = 3;
153 lhs->present = 1;
154 lhs->dpl = 3;
155 lhs->db = 0;
156 lhs->s = 1;
157 lhs->l = 0;
158 lhs->g = 0;
159 lhs->avl = 0;
160 lhs->unusable = 0;
163 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
165 unsigned flags = rhs->flags;
166 lhs->selector = rhs->selector;
167 lhs->base = rhs->base;
168 lhs->limit = rhs->limit;
169 lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
170 lhs->present = (flags & DESC_P_MASK) != 0;
171 lhs->dpl = rhs->selector & 3;
172 lhs->db = (flags >> DESC_B_SHIFT) & 1;
173 lhs->s = (flags & DESC_S_MASK) != 0;
174 lhs->l = (flags >> DESC_L_SHIFT) & 1;
175 lhs->g = (flags & DESC_G_MASK) != 0;
176 lhs->avl = (flags & DESC_AVL_MASK) != 0;
177 lhs->unusable = 0;
180 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
182 lhs->selector = rhs->selector;
183 lhs->base = rhs->base;
184 lhs->limit = rhs->limit;
185 lhs->flags =
186 (rhs->type << DESC_TYPE_SHIFT)
187 | (rhs->present * DESC_P_MASK)
188 | (rhs->dpl << DESC_DPL_SHIFT)
189 | (rhs->db << DESC_B_SHIFT)
190 | (rhs->s * DESC_S_MASK)
191 | (rhs->l << DESC_L_SHIFT)
192 | (rhs->g * DESC_G_MASK)
193 | (rhs->avl * DESC_AVL_MASK);
196 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
198 if (set)
199 *kvm_reg = *qemu_reg;
200 else
201 *qemu_reg = *kvm_reg;
204 static int kvm_getput_regs(CPUState *env, int set)
206 struct kvm_regs regs;
207 int ret = 0;
209 if (!set) {
210 ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
211 if (ret < 0)
212 return ret;
215 kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
216 kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
217 kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
218 kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
219 kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
220 kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
221 kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
222 kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
223 #ifdef TARGET_X86_64
224 kvm_getput_reg(&regs.r8, &env->regs[8], set);
225 kvm_getput_reg(&regs.r9, &env->regs[9], set);
226 kvm_getput_reg(&regs.r10, &env->regs[10], set);
227 kvm_getput_reg(&regs.r11, &env->regs[11], set);
228 kvm_getput_reg(&regs.r12, &env->regs[12], set);
229 kvm_getput_reg(&regs.r13, &env->regs[13], set);
230 kvm_getput_reg(&regs.r14, &env->regs[14], set);
231 kvm_getput_reg(&regs.r15, &env->regs[15], set);
232 #endif
234 kvm_getput_reg(&regs.rflags, &env->eflags, set);
235 kvm_getput_reg(&regs.rip, &env->eip, set);
237 if (set)
238 ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, &regs);
240 return ret;
243 static int kvm_put_fpu(CPUState *env)
245 struct kvm_fpu fpu;
246 int i;
248 memset(&fpu, 0, sizeof fpu);
249 fpu.fsw = env->fpus & ~(7 << 11);
250 fpu.fsw |= (env->fpstt & 7) << 11;
251 fpu.fcw = env->fpuc;
252 for (i = 0; i < 8; ++i)
253 fpu.ftwx |= (!env->fptags[i]) << i;
254 memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
255 memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
256 fpu.mxcsr = env->mxcsr;
258 return kvm_vcpu_ioctl(env, KVM_SET_FPU, &fpu);
261 static int kvm_put_sregs(CPUState *env)
263 struct kvm_sregs sregs;
265 memcpy(sregs.interrupt_bitmap,
266 env->interrupt_bitmap,
267 sizeof(sregs.interrupt_bitmap));
269 if ((env->eflags & VM_MASK)) {
270 set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
271 set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
272 set_v8086_seg(&sregs.es, &env->segs[R_ES]);
273 set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
274 set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
275 set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
276 } else {
277 set_seg(&sregs.cs, &env->segs[R_CS]);
278 set_seg(&sregs.ds, &env->segs[R_DS]);
279 set_seg(&sregs.es, &env->segs[R_ES]);
280 set_seg(&sregs.fs, &env->segs[R_FS]);
281 set_seg(&sregs.gs, &env->segs[R_GS]);
282 set_seg(&sregs.ss, &env->segs[R_SS]);
284 if (env->cr[0] & CR0_PE_MASK) {
285 /* force ss cpl to cs cpl */
286 sregs.ss.selector = (sregs.ss.selector & ~3) |
287 (sregs.cs.selector & 3);
288 sregs.ss.dpl = sregs.ss.selector & 3;
292 set_seg(&sregs.tr, &env->tr);
293 set_seg(&sregs.ldt, &env->ldt);
295 sregs.idt.limit = env->idt.limit;
296 sregs.idt.base = env->idt.base;
297 sregs.gdt.limit = env->gdt.limit;
298 sregs.gdt.base = env->gdt.base;
300 sregs.cr0 = env->cr[0];
301 sregs.cr2 = env->cr[2];
302 sregs.cr3 = env->cr[3];
303 sregs.cr4 = env->cr[4];
305 sregs.cr8 = cpu_get_apic_tpr(env);
306 sregs.apic_base = cpu_get_apic_base(env);
308 sregs.efer = env->efer;
310 return kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs);
313 static void kvm_msr_entry_set(struct kvm_msr_entry *entry,
314 uint32_t index, uint64_t value)
316 entry->index = index;
317 entry->data = value;
320 static int kvm_put_msrs(CPUState *env)
322 struct {
323 struct kvm_msrs info;
324 struct kvm_msr_entry entries[100];
325 } msr_data;
326 struct kvm_msr_entry *msrs = msr_data.entries;
327 int n = 0;
329 kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_CS, env->sysenter_cs);
330 kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
331 kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
332 if (kvm_has_msr_star(env))
333 kvm_msr_entry_set(&msrs[n++], MSR_STAR, env->star);
334 kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSC, env->tsc);
335 kvm_msr_entry_set(&msrs[n++], MSR_VM_HSAVE_PA, env->vm_hsave);
336 #ifdef TARGET_X86_64
337 /* FIXME if lm capable */
338 kvm_msr_entry_set(&msrs[n++], MSR_CSTAR, env->cstar);
339 kvm_msr_entry_set(&msrs[n++], MSR_KERNELGSBASE, env->kernelgsbase);
340 kvm_msr_entry_set(&msrs[n++], MSR_FMASK, env->fmask);
341 kvm_msr_entry_set(&msrs[n++], MSR_LSTAR, env->lstar);
342 #endif
343 msr_data.info.nmsrs = n;
345 return kvm_vcpu_ioctl(env, KVM_SET_MSRS, &msr_data);
350 static int kvm_get_fpu(CPUState *env)
352 struct kvm_fpu fpu;
353 int i, ret;
355 ret = kvm_vcpu_ioctl(env, KVM_GET_FPU, &fpu);
356 if (ret < 0)
357 return ret;
359 env->fpstt = (fpu.fsw >> 11) & 7;
360 env->fpus = fpu.fsw;
361 env->fpuc = fpu.fcw;
362 for (i = 0; i < 8; ++i)
363 env->fptags[i] = !((fpu.ftwx >> i) & 1);
364 memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
365 memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
366 env->mxcsr = fpu.mxcsr;
368 return 0;
371 static int kvm_get_sregs(CPUState *env)
373 struct kvm_sregs sregs;
374 uint32_t hflags;
375 int ret;
377 ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
378 if (ret < 0)
379 return ret;
381 memcpy(env->interrupt_bitmap,
382 sregs.interrupt_bitmap,
383 sizeof(sregs.interrupt_bitmap));
385 get_seg(&env->segs[R_CS], &sregs.cs);
386 get_seg(&env->segs[R_DS], &sregs.ds);
387 get_seg(&env->segs[R_ES], &sregs.es);
388 get_seg(&env->segs[R_FS], &sregs.fs);
389 get_seg(&env->segs[R_GS], &sregs.gs);
390 get_seg(&env->segs[R_SS], &sregs.ss);
392 get_seg(&env->tr, &sregs.tr);
393 get_seg(&env->ldt, &sregs.ldt);
395 env->idt.limit = sregs.idt.limit;
396 env->idt.base = sregs.idt.base;
397 env->gdt.limit = sregs.gdt.limit;
398 env->gdt.base = sregs.gdt.base;
400 env->cr[0] = sregs.cr0;
401 env->cr[2] = sregs.cr2;
402 env->cr[3] = sregs.cr3;
403 env->cr[4] = sregs.cr4;
405 cpu_set_apic_base(env, sregs.apic_base);
407 env->efer = sregs.efer;
408 //cpu_set_apic_tpr(env, sregs.cr8);
410 #define HFLAG_COPY_MASK ~( \
411 HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
412 HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
413 HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
414 HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
418 hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
419 hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
420 hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
421 (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
422 hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
423 hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
424 (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
426 if (env->efer & MSR_EFER_LMA) {
427 hflags |= HF_LMA_MASK;
430 if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
431 hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
432 } else {
433 hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
434 (DESC_B_SHIFT - HF_CS32_SHIFT);
435 hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
436 (DESC_B_SHIFT - HF_SS32_SHIFT);
437 if (!(env->cr[0] & CR0_PE_MASK) ||
438 (env->eflags & VM_MASK) ||
439 !(hflags & HF_CS32_MASK)) {
440 hflags |= HF_ADDSEG_MASK;
441 } else {
442 hflags |= ((env->segs[R_DS].base |
443 env->segs[R_ES].base |
444 env->segs[R_SS].base) != 0) <<
445 HF_ADDSEG_SHIFT;
448 env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
450 return 0;
453 static int kvm_get_msrs(CPUState *env)
455 struct {
456 struct kvm_msrs info;
457 struct kvm_msr_entry entries[100];
458 } msr_data;
459 struct kvm_msr_entry *msrs = msr_data.entries;
460 int ret, i, n;
462 n = 0;
463 msrs[n++].index = MSR_IA32_SYSENTER_CS;
464 msrs[n++].index = MSR_IA32_SYSENTER_ESP;
465 msrs[n++].index = MSR_IA32_SYSENTER_EIP;
466 if (kvm_has_msr_star(env))
467 msrs[n++].index = MSR_STAR;
468 msrs[n++].index = MSR_IA32_TSC;
469 msrs[n++].index = MSR_VM_HSAVE_PA;
470 #ifdef TARGET_X86_64
471 /* FIXME lm_capable_kernel */
472 msrs[n++].index = MSR_CSTAR;
473 msrs[n++].index = MSR_KERNELGSBASE;
474 msrs[n++].index = MSR_FMASK;
475 msrs[n++].index = MSR_LSTAR;
476 #endif
477 msr_data.info.nmsrs = n;
478 ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);
479 if (ret < 0)
480 return ret;
482 for (i = 0; i < ret; i++) {
483 switch (msrs[i].index) {
484 case MSR_IA32_SYSENTER_CS:
485 env->sysenter_cs = msrs[i].data;
486 break;
487 case MSR_IA32_SYSENTER_ESP:
488 env->sysenter_esp = msrs[i].data;
489 break;
490 case MSR_IA32_SYSENTER_EIP:
491 env->sysenter_eip = msrs[i].data;
492 break;
493 case MSR_STAR:
494 env->star = msrs[i].data;
495 break;
496 #ifdef TARGET_X86_64
497 case MSR_CSTAR:
498 env->cstar = msrs[i].data;
499 break;
500 case MSR_KERNELGSBASE:
501 env->kernelgsbase = msrs[i].data;
502 break;
503 case MSR_FMASK:
504 env->fmask = msrs[i].data;
505 break;
506 case MSR_LSTAR:
507 env->lstar = msrs[i].data;
508 break;
509 #endif
510 case MSR_IA32_TSC:
511 env->tsc = msrs[i].data;
512 break;
513 case MSR_VM_HSAVE_PA:
514 env->vm_hsave = msrs[i].data;
515 break;
519 return 0;
522 int kvm_arch_put_registers(CPUState *env)
524 int ret;
526 ret = kvm_getput_regs(env, 1);
527 if (ret < 0)
528 return ret;
530 ret = kvm_put_fpu(env);
531 if (ret < 0)
532 return ret;
534 ret = kvm_put_sregs(env);
535 if (ret < 0)
536 return ret;
538 ret = kvm_put_msrs(env);
539 if (ret < 0)
540 return ret;
542 return 0;
545 int kvm_arch_get_registers(CPUState *env)
547 int ret;
549 ret = kvm_getput_regs(env, 0);
550 if (ret < 0)
551 return ret;
553 ret = kvm_get_fpu(env);
554 if (ret < 0)
555 return ret;
557 ret = kvm_get_sregs(env);
558 if (ret < 0)
559 return ret;
561 ret = kvm_get_msrs(env);
562 if (ret < 0)
563 return ret;
565 return 0;
568 int kvm_arch_pre_run(CPUState *env, struct kvm_run *run)
570 /* Try to inject an interrupt if the guest can accept it */
571 if (run->ready_for_interrupt_injection &&
572 (env->interrupt_request & CPU_INTERRUPT_HARD) &&
573 (env->eflags & IF_MASK)) {
574 int irq;
576 env->interrupt_request &= ~CPU_INTERRUPT_HARD;
577 irq = cpu_get_pic_interrupt(env);
578 if (irq >= 0) {
579 struct kvm_interrupt intr;
580 intr.irq = irq;
581 /* FIXME: errors */
582 dprintf("injected interrupt %d\n", irq);
583 kvm_vcpu_ioctl(env, KVM_INTERRUPT, &intr);
587 /* If we have an interrupt but the guest is not ready to receive an
588 * interrupt, request an interrupt window exit. This will
589 * cause a return to userspace as soon as the guest is ready to
590 * receive interrupts. */
591 if ((env->interrupt_request & CPU_INTERRUPT_HARD))
592 run->request_interrupt_window = 1;
593 else
594 run->request_interrupt_window = 0;
596 dprintf("setting tpr\n");
597 run->cr8 = cpu_get_apic_tpr(env);
599 return 0;
602 int kvm_arch_post_run(CPUState *env, struct kvm_run *run)
604 if (run->if_flag)
605 env->eflags |= IF_MASK;
606 else
607 env->eflags &= ~IF_MASK;
609 cpu_set_apic_tpr(env, run->cr8);
610 cpu_set_apic_base(env, run->apic_base);
612 return 0;
615 static int kvm_handle_halt(CPUState *env)
617 if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
618 (env->eflags & IF_MASK)) &&
619 !(env->interrupt_request & CPU_INTERRUPT_NMI)) {
620 env->halted = 1;
621 env->exception_index = EXCP_HLT;
622 return 0;
625 return 1;
628 int kvm_arch_handle_exit(CPUState *env, struct kvm_run *run)
630 int ret = 0;
632 switch (run->exit_reason) {
633 case KVM_EXIT_HLT:
634 dprintf("handle_hlt\n");
635 ret = kvm_handle_halt(env);
636 break;
639 return ret;