kvm: bios: resolve memory device roll over reporting issues with >32G guests
[qemu-kvm/fedora.git] / target-i386 / kvm.c
blobf22f30a1375013a83d7d28016da172332c877741
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 int 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 ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, &msr_list);
94 if (ret < 0)
95 return 0;
97 msr_list.nmsrs = 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 ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
104 if (ret >= 0) {
105 int i;
107 for (i = 0; i < kvm_msr_list->nmsrs; i++) {
108 if (kvm_msr_list->indices[i] == MSR_STAR) {
109 has_msr_star = 1;
110 break;
115 free(kvm_msr_list);
118 if (has_msr_star == 1)
119 return 1;
120 return 0;
123 int kvm_arch_init(KVMState *s, int smp_cpus)
125 int ret;
127 /* create vm86 tss. KVM uses vm86 mode to emulate 16-bit code
128 * directly. In order to use vm86 mode, a TSS is needed. Since this
129 * must be part of guest physical memory, we need to allocate it. Older
130 * versions of KVM just assumed that it would be at the end of physical
131 * memory but that doesn't work with more than 4GB of memory. We simply
132 * refuse to work with those older versions of KVM. */
133 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, (void *)KVM_CAP_SET_TSS_ADDR);
134 if (ret <= 0) {
135 fprintf(stderr, "kvm does not support KVM_CAP_SET_TSS_ADDR\n");
136 return ret;
139 /* this address is 3 pages before the bios, and the bios should present
140 * as unavaible memory. FIXME, need to ensure the e820 map deals with
141 * this?
143 return kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, (void *)0xfffbd000);
146 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
148 lhs->selector = rhs->selector;
149 lhs->base = rhs->base;
150 lhs->limit = rhs->limit;
151 lhs->type = 3;
152 lhs->present = 1;
153 lhs->dpl = 3;
154 lhs->db = 0;
155 lhs->s = 1;
156 lhs->l = 0;
157 lhs->g = 0;
158 lhs->avl = 0;
159 lhs->unusable = 0;
162 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
164 unsigned flags = rhs->flags;
165 lhs->selector = rhs->selector;
166 lhs->base = rhs->base;
167 lhs->limit = rhs->limit;
168 lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
169 lhs->present = (flags & DESC_P_MASK) != 0;
170 lhs->dpl = rhs->selector & 3;
171 lhs->db = (flags >> DESC_B_SHIFT) & 1;
172 lhs->s = (flags & DESC_S_MASK) != 0;
173 lhs->l = (flags >> DESC_L_SHIFT) & 1;
174 lhs->g = (flags & DESC_G_MASK) != 0;
175 lhs->avl = (flags & DESC_AVL_MASK) != 0;
176 lhs->unusable = 0;
179 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
181 lhs->selector = rhs->selector;
182 lhs->base = rhs->base;
183 lhs->limit = rhs->limit;
184 lhs->flags =
185 (rhs->type << DESC_TYPE_SHIFT)
186 | (rhs->present * DESC_P_MASK)
187 | (rhs->dpl << DESC_DPL_SHIFT)
188 | (rhs->db << DESC_B_SHIFT)
189 | (rhs->s * DESC_S_MASK)
190 | (rhs->l << DESC_L_SHIFT)
191 | (rhs->g * DESC_G_MASK)
192 | (rhs->avl * DESC_AVL_MASK);
195 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
197 if (set)
198 *kvm_reg = *qemu_reg;
199 else
200 *qemu_reg = *kvm_reg;
203 static int kvm_getput_regs(CPUState *env, int set)
205 struct kvm_regs regs;
206 int ret = 0;
208 if (!set) {
209 ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
210 if (ret < 0)
211 return ret;
214 kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
215 kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
216 kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
217 kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
218 kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
219 kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
220 kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
221 kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
222 #ifdef TARGET_X86_64
223 kvm_getput_reg(&regs.r8, &env->regs[8], set);
224 kvm_getput_reg(&regs.r9, &env->regs[9], set);
225 kvm_getput_reg(&regs.r10, &env->regs[10], set);
226 kvm_getput_reg(&regs.r11, &env->regs[11], set);
227 kvm_getput_reg(&regs.r12, &env->regs[12], set);
228 kvm_getput_reg(&regs.r13, &env->regs[13], set);
229 kvm_getput_reg(&regs.r14, &env->regs[14], set);
230 kvm_getput_reg(&regs.r15, &env->regs[15], set);
231 #endif
233 kvm_getput_reg(&regs.rflags, &env->eflags, set);
234 kvm_getput_reg(&regs.rip, &env->eip, set);
236 if (set)
237 ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, &regs);
239 return ret;
242 static int kvm_put_fpu(CPUState *env)
244 struct kvm_fpu fpu;
245 int i;
247 memset(&fpu, 0, sizeof fpu);
248 fpu.fsw = env->fpus & ~(7 << 11);
249 fpu.fsw |= (env->fpstt & 7) << 11;
250 fpu.fcw = env->fpuc;
251 for (i = 0; i < 8; ++i)
252 fpu.ftwx |= (!env->fptags[i]) << i;
253 memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
254 memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
255 fpu.mxcsr = env->mxcsr;
257 return kvm_vcpu_ioctl(env, KVM_SET_FPU, &fpu);
260 static int kvm_put_sregs(CPUState *env)
262 struct kvm_sregs sregs;
264 memcpy(sregs.interrupt_bitmap,
265 env->interrupt_bitmap,
266 sizeof(sregs.interrupt_bitmap));
268 if ((env->eflags & VM_MASK)) {
269 set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
270 set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
271 set_v8086_seg(&sregs.es, &env->segs[R_ES]);
272 set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
273 set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
274 set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
275 } else {
276 set_seg(&sregs.cs, &env->segs[R_CS]);
277 set_seg(&sregs.ds, &env->segs[R_DS]);
278 set_seg(&sregs.es, &env->segs[R_ES]);
279 set_seg(&sregs.fs, &env->segs[R_FS]);
280 set_seg(&sregs.gs, &env->segs[R_GS]);
281 set_seg(&sregs.ss, &env->segs[R_SS]);
283 if (env->cr[0] & CR0_PE_MASK) {
284 /* force ss cpl to cs cpl */
285 sregs.ss.selector = (sregs.ss.selector & ~3) |
286 (sregs.cs.selector & 3);
287 sregs.ss.dpl = sregs.ss.selector & 3;
291 set_seg(&sregs.tr, &env->tr);
292 set_seg(&sregs.ldt, &env->ldt);
294 sregs.idt.limit = env->idt.limit;
295 sregs.idt.base = env->idt.base;
296 sregs.gdt.limit = env->gdt.limit;
297 sregs.gdt.base = env->gdt.base;
299 sregs.cr0 = env->cr[0];
300 sregs.cr2 = env->cr[2];
301 sregs.cr3 = env->cr[3];
302 sregs.cr4 = env->cr[4];
304 sregs.cr8 = cpu_get_apic_tpr(env);
305 sregs.apic_base = cpu_get_apic_base(env);
307 sregs.efer = env->efer;
309 return kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs);
312 static void kvm_msr_entry_set(struct kvm_msr_entry *entry,
313 uint32_t index, uint64_t value)
315 entry->index = index;
316 entry->data = value;
319 static int kvm_put_msrs(CPUState *env)
321 struct {
322 struct kvm_msrs info;
323 struct kvm_msr_entry entries[100];
324 } msr_data;
325 struct kvm_msr_entry *msrs = msr_data.entries;
326 int n = 0;
328 kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_CS, env->sysenter_cs);
329 kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
330 kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
331 if (kvm_has_msr_star(env))
332 kvm_msr_entry_set(&msrs[n++], MSR_STAR, env->star);
333 kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSC, env->tsc);
334 #ifdef TARGET_X86_64
335 /* FIXME if lm capable */
336 kvm_msr_entry_set(&msrs[n++], MSR_CSTAR, env->cstar);
337 kvm_msr_entry_set(&msrs[n++], MSR_KERNELGSBASE, env->kernelgsbase);
338 kvm_msr_entry_set(&msrs[n++], MSR_FMASK, env->fmask);
339 kvm_msr_entry_set(&msrs[n++], MSR_LSTAR, env->lstar);
340 #endif
341 msr_data.info.nmsrs = n;
343 return kvm_vcpu_ioctl(env, KVM_SET_MSRS, &msr_data);
348 static int kvm_get_fpu(CPUState *env)
350 struct kvm_fpu fpu;
351 int i, ret;
353 ret = kvm_vcpu_ioctl(env, KVM_GET_FPU, &fpu);
354 if (ret < 0)
355 return ret;
357 env->fpstt = (fpu.fsw >> 11) & 7;
358 env->fpus = fpu.fsw;
359 env->fpuc = fpu.fcw;
360 for (i = 0; i < 8; ++i)
361 env->fptags[i] = !((fpu.ftwx >> i) & 1);
362 memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
363 memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
364 env->mxcsr = fpu.mxcsr;
366 return 0;
369 static int kvm_get_sregs(CPUState *env)
371 struct kvm_sregs sregs;
372 uint32_t hflags;
373 int ret;
375 ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
376 if (ret < 0)
377 return ret;
379 memcpy(env->interrupt_bitmap,
380 sregs.interrupt_bitmap,
381 sizeof(sregs.interrupt_bitmap));
383 get_seg(&env->segs[R_CS], &sregs.cs);
384 get_seg(&env->segs[R_DS], &sregs.ds);
385 get_seg(&env->segs[R_ES], &sregs.es);
386 get_seg(&env->segs[R_FS], &sregs.fs);
387 get_seg(&env->segs[R_GS], &sregs.gs);
388 get_seg(&env->segs[R_SS], &sregs.ss);
390 get_seg(&env->tr, &sregs.tr);
391 get_seg(&env->ldt, &sregs.ldt);
393 env->idt.limit = sregs.idt.limit;
394 env->idt.base = sregs.idt.base;
395 env->gdt.limit = sregs.gdt.limit;
396 env->gdt.base = sregs.gdt.base;
398 env->cr[0] = sregs.cr0;
399 env->cr[2] = sregs.cr2;
400 env->cr[3] = sregs.cr3;
401 env->cr[4] = sregs.cr4;
403 cpu_set_apic_base(env, sregs.apic_base);
405 env->efer = sregs.efer;
406 //cpu_set_apic_tpr(env, sregs.cr8);
408 #define HFLAG_COPY_MASK ~( \
409 HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
410 HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
411 HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
412 HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
416 hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
417 hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
418 hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
419 (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
420 hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
421 hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
422 (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
424 if (env->efer & MSR_EFER_LMA) {
425 hflags |= HF_LMA_MASK;
428 if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
429 hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
430 } else {
431 hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
432 (DESC_B_SHIFT - HF_CS32_SHIFT);
433 hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
434 (DESC_B_SHIFT - HF_SS32_SHIFT);
435 if (!(env->cr[0] & CR0_PE_MASK) ||
436 (env->eflags & VM_MASK) ||
437 !(hflags & HF_CS32_MASK)) {
438 hflags |= HF_ADDSEG_MASK;
439 } else {
440 hflags |= ((env->segs[R_DS].base |
441 env->segs[R_ES].base |
442 env->segs[R_SS].base) != 0) <<
443 HF_ADDSEG_SHIFT;
446 env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
447 env->cc_src = env->eflags & (CC_O | CC_S | CC_Z | CC_A | CC_P | CC_C);
448 env->df = 1 - (2 * ((env->eflags >> 10) & 1));
449 env->cc_op = CC_OP_EFLAGS;
450 env->eflags &= ~(DF_MASK | CC_O | CC_S | CC_Z | CC_A | CC_P | CC_C);
452 return 0;
455 static int kvm_get_msrs(CPUState *env)
457 struct {
458 struct kvm_msrs info;
459 struct kvm_msr_entry entries[100];
460 } msr_data;
461 struct kvm_msr_entry *msrs = msr_data.entries;
462 int ret, i, n;
464 n = 0;
465 msrs[n++].index = MSR_IA32_SYSENTER_CS;
466 msrs[n++].index = MSR_IA32_SYSENTER_ESP;
467 msrs[n++].index = MSR_IA32_SYSENTER_EIP;
468 if (kvm_has_msr_star(env))
469 msrs[n++].index = MSR_STAR;
470 msrs[n++].index = MSR_IA32_TSC;
471 #ifdef TARGET_X86_64
472 /* FIXME lm_capable_kernel */
473 msrs[n++].index = MSR_CSTAR;
474 msrs[n++].index = MSR_KERNELGSBASE;
475 msrs[n++].index = MSR_FMASK;
476 msrs[n++].index = MSR_LSTAR;
477 #endif
478 msr_data.info.nmsrs = n;
479 ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);
480 if (ret < 0)
481 return ret;
483 for (i = 0; i < ret; i++) {
484 switch (msrs[i].index) {
485 case MSR_IA32_SYSENTER_CS:
486 env->sysenter_cs = msrs[i].data;
487 break;
488 case MSR_IA32_SYSENTER_ESP:
489 env->sysenter_esp = msrs[i].data;
490 break;
491 case MSR_IA32_SYSENTER_EIP:
492 env->sysenter_eip = msrs[i].data;
493 break;
494 case MSR_STAR:
495 env->star = msrs[i].data;
496 break;
497 #ifdef TARGET_X86_64
498 case MSR_CSTAR:
499 env->cstar = msrs[i].data;
500 break;
501 case MSR_KERNELGSBASE:
502 env->kernelgsbase = msrs[i].data;
503 break;
504 case MSR_FMASK:
505 env->fmask = msrs[i].data;
506 break;
507 case MSR_LSTAR:
508 env->lstar = msrs[i].data;
509 break;
510 #endif
511 case MSR_IA32_TSC:
512 env->tsc = msrs[i].data;
513 break;
517 return 0;
520 int kvm_arch_put_registers(CPUState *env)
522 int ret;
524 ret = kvm_getput_regs(env, 1);
525 if (ret < 0)
526 return ret;
528 ret = kvm_put_fpu(env);
529 if (ret < 0)
530 return ret;
532 ret = kvm_put_sregs(env);
533 if (ret < 0)
534 return ret;
536 ret = kvm_put_msrs(env);
537 if (ret < 0)
538 return ret;
540 return 0;
543 int kvm_arch_get_registers(CPUState *env)
545 int ret;
547 ret = kvm_getput_regs(env, 0);
548 if (ret < 0)
549 return ret;
551 ret = kvm_get_fpu(env);
552 if (ret < 0)
553 return ret;
555 ret = kvm_get_sregs(env);
556 if (ret < 0)
557 return ret;
559 ret = kvm_get_msrs(env);
560 if (ret < 0)
561 return ret;
563 return 0;
566 int kvm_arch_pre_run(CPUState *env, struct kvm_run *run)
568 /* Try to inject an interrupt if the guest can accept it */
569 if (run->ready_for_interrupt_injection &&
570 (env->interrupt_request & CPU_INTERRUPT_HARD) &&
571 (env->eflags & IF_MASK)) {
572 int irq;
574 env->interrupt_request &= ~CPU_INTERRUPT_HARD;
575 irq = cpu_get_pic_interrupt(env);
576 if (irq >= 0) {
577 struct kvm_interrupt intr;
578 intr.irq = irq;
579 /* FIXME: errors */
580 dprintf("injected interrupt %d\n", irq);
581 kvm_vcpu_ioctl(env, KVM_INTERRUPT, &intr);
585 /* If we have an interrupt but the guest is not ready to receive an
586 * interrupt, request an interrupt window exit. This will
587 * cause a return to userspace as soon as the guest is ready to
588 * receive interrupts. */
589 if ((env->interrupt_request & CPU_INTERRUPT_HARD))
590 run->request_interrupt_window = 1;
591 else
592 run->request_interrupt_window = 0;
594 dprintf("setting tpr\n");
595 run->cr8 = cpu_get_apic_tpr(env);
597 return 0;
600 int kvm_arch_post_run(CPUState *env, struct kvm_run *run)
602 if (run->if_flag)
603 env->eflags |= IF_MASK;
604 else
605 env->eflags &= ~IF_MASK;
607 cpu_set_apic_tpr(env, run->cr8);
608 cpu_set_apic_base(env, run->apic_base);
610 return 0;
613 static int kvm_handle_halt(CPUState *env)
615 if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
616 (env->eflags & IF_MASK)) &&
617 !(env->interrupt_request & CPU_INTERRUPT_NMI)) {
618 env->halted = 1;
619 env->exception_index = EXCP_HLT;
620 return 0;
623 return 1;
626 int kvm_arch_handle_exit(CPUState *env, struct kvm_run *run)
628 int ret = 0;
630 switch (run->exit_reason) {
631 case KVM_EXIT_HLT:
632 dprintf("handle_hlt\n");
633 ret = kvm_handle_halt(env);
634 break;
637 return ret;