apic: avoid passing CPUState from CPU code
[qemu.git] / target-i386 / kvm.c
blob545323961aaee42ec322241b98075024c9aad4dd
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"
25 #include "gdbstub.h"
26 #include "host-utils.h"
27 #include "hw/pc.h"
28 #include "ioport.h"
30 #ifdef CONFIG_KVM_PARA
31 #include <linux/kvm_para.h>
32 #endif
34 //#define DEBUG_KVM
36 #ifdef DEBUG_KVM
37 #define DPRINTF(fmt, ...) \
38 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
39 #else
40 #define DPRINTF(fmt, ...) \
41 do { } while (0)
42 #endif
44 #define MSR_KVM_WALL_CLOCK 0x11
45 #define MSR_KVM_SYSTEM_TIME 0x12
47 #ifdef KVM_CAP_EXT_CPUID
49 static struct kvm_cpuid2 *try_get_cpuid(KVMState *s, int max)
51 struct kvm_cpuid2 *cpuid;
52 int r, size;
54 size = sizeof(*cpuid) + max * sizeof(*cpuid->entries);
55 cpuid = (struct kvm_cpuid2 *)qemu_mallocz(size);
56 cpuid->nent = max;
57 r = kvm_ioctl(s, KVM_GET_SUPPORTED_CPUID, cpuid);
58 if (r == 0 && cpuid->nent >= max) {
59 r = -E2BIG;
61 if (r < 0) {
62 if (r == -E2BIG) {
63 qemu_free(cpuid);
64 return NULL;
65 } else {
66 fprintf(stderr, "KVM_GET_SUPPORTED_CPUID failed: %s\n",
67 strerror(-r));
68 exit(1);
71 return cpuid;
74 uint32_t kvm_arch_get_supported_cpuid(CPUState *env, uint32_t function, int reg)
76 struct kvm_cpuid2 *cpuid;
77 int i, max;
78 uint32_t ret = 0;
79 uint32_t cpuid_1_edx;
81 if (!kvm_check_extension(env->kvm_state, KVM_CAP_EXT_CPUID)) {
82 return -1U;
85 max = 1;
86 while ((cpuid = try_get_cpuid(env->kvm_state, max)) == NULL) {
87 max *= 2;
90 for (i = 0; i < cpuid->nent; ++i) {
91 if (cpuid->entries[i].function == function) {
92 switch (reg) {
93 case R_EAX:
94 ret = cpuid->entries[i].eax;
95 break;
96 case R_EBX:
97 ret = cpuid->entries[i].ebx;
98 break;
99 case R_ECX:
100 ret = cpuid->entries[i].ecx;
101 break;
102 case R_EDX:
103 ret = cpuid->entries[i].edx;
104 switch (function) {
105 case 1:
106 /* KVM before 2.6.30 misreports the following features */
107 ret |= CPUID_MTRR | CPUID_PAT | CPUID_MCE | CPUID_MCA;
108 break;
109 case 0x80000001:
110 /* On Intel, kvm returns cpuid according to the Intel spec,
111 * so add missing bits according to the AMD spec:
113 cpuid_1_edx = kvm_arch_get_supported_cpuid(env, 1, R_EDX);
114 ret |= cpuid_1_edx & 0x183f7ff;
115 break;
117 break;
122 qemu_free(cpuid);
124 return ret;
127 #else
129 uint32_t kvm_arch_get_supported_cpuid(CPUState *env, uint32_t function, int reg)
131 return -1U;
134 #endif
136 #ifdef CONFIG_KVM_PARA
137 struct kvm_para_features {
138 int cap;
139 int feature;
140 } para_features[] = {
141 #ifdef KVM_CAP_CLOCKSOURCE
142 { KVM_CAP_CLOCKSOURCE, KVM_FEATURE_CLOCKSOURCE },
143 #endif
144 #ifdef KVM_CAP_NOP_IO_DELAY
145 { KVM_CAP_NOP_IO_DELAY, KVM_FEATURE_NOP_IO_DELAY },
146 #endif
147 #ifdef KVM_CAP_PV_MMU
148 { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
149 #endif
150 { -1, -1 }
153 static int get_para_features(CPUState *env)
155 int i, features = 0;
157 for (i = 0; i < ARRAY_SIZE(para_features) - 1; i++) {
158 if (kvm_check_extension(env->kvm_state, para_features[i].cap))
159 features |= (1 << para_features[i].feature);
162 return features;
164 #endif
166 int kvm_arch_init_vcpu(CPUState *env)
168 struct {
169 struct kvm_cpuid2 cpuid;
170 struct kvm_cpuid_entry2 entries[100];
171 } __attribute__((packed)) cpuid_data;
172 uint32_t limit, i, j, cpuid_i;
173 uint32_t unused;
174 struct kvm_cpuid_entry2 *c;
175 #ifdef KVM_CPUID_SIGNATURE
176 uint32_t signature[3];
177 #endif
179 env->mp_state = KVM_MP_STATE_RUNNABLE;
181 env->cpuid_features &= kvm_arch_get_supported_cpuid(env, 1, R_EDX);
183 i = env->cpuid_ext_features & CPUID_EXT_HYPERVISOR;
184 env->cpuid_ext_features &= kvm_arch_get_supported_cpuid(env, 1, R_ECX);
185 env->cpuid_ext_features |= i;
187 env->cpuid_ext2_features &= kvm_arch_get_supported_cpuid(env, 0x80000001,
188 R_EDX);
189 env->cpuid_ext3_features &= kvm_arch_get_supported_cpuid(env, 0x80000001,
190 R_ECX);
192 cpuid_i = 0;
194 #ifdef CONFIG_KVM_PARA
195 /* Paravirtualization CPUIDs */
196 memcpy(signature, "KVMKVMKVM\0\0\0", 12);
197 c = &cpuid_data.entries[cpuid_i++];
198 memset(c, 0, sizeof(*c));
199 c->function = KVM_CPUID_SIGNATURE;
200 c->eax = 0;
201 c->ebx = signature[0];
202 c->ecx = signature[1];
203 c->edx = signature[2];
205 c = &cpuid_data.entries[cpuid_i++];
206 memset(c, 0, sizeof(*c));
207 c->function = KVM_CPUID_FEATURES;
208 c->eax = env->cpuid_kvm_features & get_para_features(env);
209 #endif
211 cpu_x86_cpuid(env, 0, 0, &limit, &unused, &unused, &unused);
213 for (i = 0; i <= limit; i++) {
214 c = &cpuid_data.entries[cpuid_i++];
216 switch (i) {
217 case 2: {
218 /* Keep reading function 2 till all the input is received */
219 int times;
221 c->function = i;
222 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC |
223 KVM_CPUID_FLAG_STATE_READ_NEXT;
224 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
225 times = c->eax & 0xff;
227 for (j = 1; j < times; ++j) {
228 c = &cpuid_data.entries[cpuid_i++];
229 c->function = i;
230 c->flags = KVM_CPUID_FLAG_STATEFUL_FUNC;
231 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
233 break;
235 case 4:
236 case 0xb:
237 case 0xd:
238 for (j = 0; ; j++) {
239 c->function = i;
240 c->flags = KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
241 c->index = j;
242 cpu_x86_cpuid(env, i, j, &c->eax, &c->ebx, &c->ecx, &c->edx);
244 if (i == 4 && c->eax == 0)
245 break;
246 if (i == 0xb && !(c->ecx & 0xff00))
247 break;
248 if (i == 0xd && c->eax == 0)
249 break;
251 c = &cpuid_data.entries[cpuid_i++];
253 break;
254 default:
255 c->function = i;
256 c->flags = 0;
257 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
258 break;
261 cpu_x86_cpuid(env, 0x80000000, 0, &limit, &unused, &unused, &unused);
263 for (i = 0x80000000; i <= limit; i++) {
264 c = &cpuid_data.entries[cpuid_i++];
266 c->function = i;
267 c->flags = 0;
268 cpu_x86_cpuid(env, i, 0, &c->eax, &c->ebx, &c->ecx, &c->edx);
271 cpuid_data.cpuid.nent = cpuid_i;
273 return kvm_vcpu_ioctl(env, KVM_SET_CPUID2, &cpuid_data);
276 void kvm_arch_reset_vcpu(CPUState *env)
278 env->exception_injected = -1;
279 env->interrupt_injected = -1;
280 env->nmi_injected = 0;
281 env->nmi_pending = 0;
284 static int kvm_has_msr_star(CPUState *env)
286 static int has_msr_star;
287 int ret;
289 /* first time */
290 if (has_msr_star == 0) {
291 struct kvm_msr_list msr_list, *kvm_msr_list;
293 has_msr_star = -1;
295 /* Obtain MSR list from KVM. These are the MSRs that we must
296 * save/restore */
297 msr_list.nmsrs = 0;
298 ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, &msr_list);
299 if (ret < 0 && ret != -E2BIG) {
300 return 0;
302 /* Old kernel modules had a bug and could write beyond the provided
303 memory. Allocate at least a safe amount of 1K. */
304 kvm_msr_list = qemu_mallocz(MAX(1024, sizeof(msr_list) +
305 msr_list.nmsrs *
306 sizeof(msr_list.indices[0])));
308 kvm_msr_list->nmsrs = msr_list.nmsrs;
309 ret = kvm_ioctl(env->kvm_state, KVM_GET_MSR_INDEX_LIST, kvm_msr_list);
310 if (ret >= 0) {
311 int i;
313 for (i = 0; i < kvm_msr_list->nmsrs; i++) {
314 if (kvm_msr_list->indices[i] == MSR_STAR) {
315 has_msr_star = 1;
316 break;
321 free(kvm_msr_list);
324 if (has_msr_star == 1)
325 return 1;
326 return 0;
329 static int kvm_init_identity_map_page(KVMState *s)
331 #ifdef KVM_CAP_SET_IDENTITY_MAP_ADDR
332 int ret;
333 uint64_t addr = 0xfffbc000;
335 if (!kvm_check_extension(s, KVM_CAP_SET_IDENTITY_MAP_ADDR)) {
336 return 0;
339 ret = kvm_vm_ioctl(s, KVM_SET_IDENTITY_MAP_ADDR, &addr);
340 if (ret < 0) {
341 fprintf(stderr, "kvm_set_identity_map_addr: %s\n", strerror(ret));
342 return ret;
344 #endif
345 return 0;
348 int kvm_arch_init(KVMState *s, int smp_cpus)
350 int ret;
352 /* create vm86 tss. KVM uses vm86 mode to emulate 16-bit code
353 * directly. In order to use vm86 mode, a TSS is needed. Since this
354 * must be part of guest physical memory, we need to allocate it. Older
355 * versions of KVM just assumed that it would be at the end of physical
356 * memory but that doesn't work with more than 4GB of memory. We simply
357 * refuse to work with those older versions of KVM. */
358 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, KVM_CAP_SET_TSS_ADDR);
359 if (ret <= 0) {
360 fprintf(stderr, "kvm does not support KVM_CAP_SET_TSS_ADDR\n");
361 return ret;
364 /* this address is 3 pages before the bios, and the bios should present
365 * as unavaible memory. FIXME, need to ensure the e820 map deals with
366 * this?
369 * Tell fw_cfg to notify the BIOS to reserve the range.
371 if (e820_add_entry(0xfffbc000, 0x4000, E820_RESERVED) < 0) {
372 perror("e820_add_entry() table is full");
373 exit(1);
375 ret = kvm_vm_ioctl(s, KVM_SET_TSS_ADDR, 0xfffbd000);
376 if (ret < 0) {
377 return ret;
380 return kvm_init_identity_map_page(s);
383 static void set_v8086_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
385 lhs->selector = rhs->selector;
386 lhs->base = rhs->base;
387 lhs->limit = rhs->limit;
388 lhs->type = 3;
389 lhs->present = 1;
390 lhs->dpl = 3;
391 lhs->db = 0;
392 lhs->s = 1;
393 lhs->l = 0;
394 lhs->g = 0;
395 lhs->avl = 0;
396 lhs->unusable = 0;
399 static void set_seg(struct kvm_segment *lhs, const SegmentCache *rhs)
401 unsigned flags = rhs->flags;
402 lhs->selector = rhs->selector;
403 lhs->base = rhs->base;
404 lhs->limit = rhs->limit;
405 lhs->type = (flags >> DESC_TYPE_SHIFT) & 15;
406 lhs->present = (flags & DESC_P_MASK) != 0;
407 lhs->dpl = rhs->selector & 3;
408 lhs->db = (flags >> DESC_B_SHIFT) & 1;
409 lhs->s = (flags & DESC_S_MASK) != 0;
410 lhs->l = (flags >> DESC_L_SHIFT) & 1;
411 lhs->g = (flags & DESC_G_MASK) != 0;
412 lhs->avl = (flags & DESC_AVL_MASK) != 0;
413 lhs->unusable = 0;
416 static void get_seg(SegmentCache *lhs, const struct kvm_segment *rhs)
418 lhs->selector = rhs->selector;
419 lhs->base = rhs->base;
420 lhs->limit = rhs->limit;
421 lhs->flags =
422 (rhs->type << DESC_TYPE_SHIFT)
423 | (rhs->present * DESC_P_MASK)
424 | (rhs->dpl << DESC_DPL_SHIFT)
425 | (rhs->db << DESC_B_SHIFT)
426 | (rhs->s * DESC_S_MASK)
427 | (rhs->l << DESC_L_SHIFT)
428 | (rhs->g * DESC_G_MASK)
429 | (rhs->avl * DESC_AVL_MASK);
432 static void kvm_getput_reg(__u64 *kvm_reg, target_ulong *qemu_reg, int set)
434 if (set)
435 *kvm_reg = *qemu_reg;
436 else
437 *qemu_reg = *kvm_reg;
440 static int kvm_getput_regs(CPUState *env, int set)
442 struct kvm_regs regs;
443 int ret = 0;
445 if (!set) {
446 ret = kvm_vcpu_ioctl(env, KVM_GET_REGS, &regs);
447 if (ret < 0)
448 return ret;
451 kvm_getput_reg(&regs.rax, &env->regs[R_EAX], set);
452 kvm_getput_reg(&regs.rbx, &env->regs[R_EBX], set);
453 kvm_getput_reg(&regs.rcx, &env->regs[R_ECX], set);
454 kvm_getput_reg(&regs.rdx, &env->regs[R_EDX], set);
455 kvm_getput_reg(&regs.rsi, &env->regs[R_ESI], set);
456 kvm_getput_reg(&regs.rdi, &env->regs[R_EDI], set);
457 kvm_getput_reg(&regs.rsp, &env->regs[R_ESP], set);
458 kvm_getput_reg(&regs.rbp, &env->regs[R_EBP], set);
459 #ifdef TARGET_X86_64
460 kvm_getput_reg(&regs.r8, &env->regs[8], set);
461 kvm_getput_reg(&regs.r9, &env->regs[9], set);
462 kvm_getput_reg(&regs.r10, &env->regs[10], set);
463 kvm_getput_reg(&regs.r11, &env->regs[11], set);
464 kvm_getput_reg(&regs.r12, &env->regs[12], set);
465 kvm_getput_reg(&regs.r13, &env->regs[13], set);
466 kvm_getput_reg(&regs.r14, &env->regs[14], set);
467 kvm_getput_reg(&regs.r15, &env->regs[15], set);
468 #endif
470 kvm_getput_reg(&regs.rflags, &env->eflags, set);
471 kvm_getput_reg(&regs.rip, &env->eip, set);
473 if (set)
474 ret = kvm_vcpu_ioctl(env, KVM_SET_REGS, &regs);
476 return ret;
479 static int kvm_put_fpu(CPUState *env)
481 struct kvm_fpu fpu;
482 int i;
484 memset(&fpu, 0, sizeof fpu);
485 fpu.fsw = env->fpus & ~(7 << 11);
486 fpu.fsw |= (env->fpstt & 7) << 11;
487 fpu.fcw = env->fpuc;
488 for (i = 0; i < 8; ++i)
489 fpu.ftwx |= (!env->fptags[i]) << i;
490 memcpy(fpu.fpr, env->fpregs, sizeof env->fpregs);
491 memcpy(fpu.xmm, env->xmm_regs, sizeof env->xmm_regs);
492 fpu.mxcsr = env->mxcsr;
494 return kvm_vcpu_ioctl(env, KVM_SET_FPU, &fpu);
497 static int kvm_put_sregs(CPUState *env)
499 struct kvm_sregs sregs;
501 memset(sregs.interrupt_bitmap, 0, sizeof(sregs.interrupt_bitmap));
502 if (env->interrupt_injected >= 0) {
503 sregs.interrupt_bitmap[env->interrupt_injected / 64] |=
504 (uint64_t)1 << (env->interrupt_injected % 64);
507 if ((env->eflags & VM_MASK)) {
508 set_v8086_seg(&sregs.cs, &env->segs[R_CS]);
509 set_v8086_seg(&sregs.ds, &env->segs[R_DS]);
510 set_v8086_seg(&sregs.es, &env->segs[R_ES]);
511 set_v8086_seg(&sregs.fs, &env->segs[R_FS]);
512 set_v8086_seg(&sregs.gs, &env->segs[R_GS]);
513 set_v8086_seg(&sregs.ss, &env->segs[R_SS]);
514 } else {
515 set_seg(&sregs.cs, &env->segs[R_CS]);
516 set_seg(&sregs.ds, &env->segs[R_DS]);
517 set_seg(&sregs.es, &env->segs[R_ES]);
518 set_seg(&sregs.fs, &env->segs[R_FS]);
519 set_seg(&sregs.gs, &env->segs[R_GS]);
520 set_seg(&sregs.ss, &env->segs[R_SS]);
522 if (env->cr[0] & CR0_PE_MASK) {
523 /* force ss cpl to cs cpl */
524 sregs.ss.selector = (sregs.ss.selector & ~3) |
525 (sregs.cs.selector & 3);
526 sregs.ss.dpl = sregs.ss.selector & 3;
530 set_seg(&sregs.tr, &env->tr);
531 set_seg(&sregs.ldt, &env->ldt);
533 sregs.idt.limit = env->idt.limit;
534 sregs.idt.base = env->idt.base;
535 sregs.gdt.limit = env->gdt.limit;
536 sregs.gdt.base = env->gdt.base;
538 sregs.cr0 = env->cr[0];
539 sregs.cr2 = env->cr[2];
540 sregs.cr3 = env->cr[3];
541 sregs.cr4 = env->cr[4];
543 sregs.cr8 = cpu_get_apic_tpr(env->apic_state);
544 sregs.apic_base = cpu_get_apic_base(env->apic_state);
546 sregs.efer = env->efer;
548 return kvm_vcpu_ioctl(env, KVM_SET_SREGS, &sregs);
551 static void kvm_msr_entry_set(struct kvm_msr_entry *entry,
552 uint32_t index, uint64_t value)
554 entry->index = index;
555 entry->data = value;
558 static int kvm_put_msrs(CPUState *env, int level)
560 struct {
561 struct kvm_msrs info;
562 struct kvm_msr_entry entries[100];
563 } msr_data;
564 struct kvm_msr_entry *msrs = msr_data.entries;
565 int n = 0;
567 kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_CS, env->sysenter_cs);
568 kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_ESP, env->sysenter_esp);
569 kvm_msr_entry_set(&msrs[n++], MSR_IA32_SYSENTER_EIP, env->sysenter_eip);
570 if (kvm_has_msr_star(env))
571 kvm_msr_entry_set(&msrs[n++], MSR_STAR, env->star);
572 #ifdef TARGET_X86_64
573 /* FIXME if lm capable */
574 kvm_msr_entry_set(&msrs[n++], MSR_CSTAR, env->cstar);
575 kvm_msr_entry_set(&msrs[n++], MSR_KERNELGSBASE, env->kernelgsbase);
576 kvm_msr_entry_set(&msrs[n++], MSR_FMASK, env->fmask);
577 kvm_msr_entry_set(&msrs[n++], MSR_LSTAR, env->lstar);
578 #endif
579 if (level == KVM_PUT_FULL_STATE) {
580 kvm_msr_entry_set(&msrs[n++], MSR_IA32_TSC, env->tsc);
581 kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME,
582 env->system_time_msr);
583 kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
586 msr_data.info.nmsrs = n;
588 return kvm_vcpu_ioctl(env, KVM_SET_MSRS, &msr_data);
593 static int kvm_get_fpu(CPUState *env)
595 struct kvm_fpu fpu;
596 int i, ret;
598 ret = kvm_vcpu_ioctl(env, KVM_GET_FPU, &fpu);
599 if (ret < 0)
600 return ret;
602 env->fpstt = (fpu.fsw >> 11) & 7;
603 env->fpus = fpu.fsw;
604 env->fpuc = fpu.fcw;
605 for (i = 0; i < 8; ++i)
606 env->fptags[i] = !((fpu.ftwx >> i) & 1);
607 memcpy(env->fpregs, fpu.fpr, sizeof env->fpregs);
608 memcpy(env->xmm_regs, fpu.xmm, sizeof env->xmm_regs);
609 env->mxcsr = fpu.mxcsr;
611 return 0;
614 static int kvm_get_sregs(CPUState *env)
616 struct kvm_sregs sregs;
617 uint32_t hflags;
618 int bit, i, ret;
620 ret = kvm_vcpu_ioctl(env, KVM_GET_SREGS, &sregs);
621 if (ret < 0)
622 return ret;
624 /* There can only be one pending IRQ set in the bitmap at a time, so try
625 to find it and save its number instead (-1 for none). */
626 env->interrupt_injected = -1;
627 for (i = 0; i < ARRAY_SIZE(sregs.interrupt_bitmap); i++) {
628 if (sregs.interrupt_bitmap[i]) {
629 bit = ctz64(sregs.interrupt_bitmap[i]);
630 env->interrupt_injected = i * 64 + bit;
631 break;
635 get_seg(&env->segs[R_CS], &sregs.cs);
636 get_seg(&env->segs[R_DS], &sregs.ds);
637 get_seg(&env->segs[R_ES], &sregs.es);
638 get_seg(&env->segs[R_FS], &sregs.fs);
639 get_seg(&env->segs[R_GS], &sregs.gs);
640 get_seg(&env->segs[R_SS], &sregs.ss);
642 get_seg(&env->tr, &sregs.tr);
643 get_seg(&env->ldt, &sregs.ldt);
645 env->idt.limit = sregs.idt.limit;
646 env->idt.base = sregs.idt.base;
647 env->gdt.limit = sregs.gdt.limit;
648 env->gdt.base = sregs.gdt.base;
650 env->cr[0] = sregs.cr0;
651 env->cr[2] = sregs.cr2;
652 env->cr[3] = sregs.cr3;
653 env->cr[4] = sregs.cr4;
655 cpu_set_apic_base(env->apic_state, sregs.apic_base);
657 env->efer = sregs.efer;
658 //cpu_set_apic_tpr(env->apic_state, sregs.cr8);
660 #define HFLAG_COPY_MASK ~( \
661 HF_CPL_MASK | HF_PE_MASK | HF_MP_MASK | HF_EM_MASK | \
662 HF_TS_MASK | HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK | \
663 HF_OSFXSR_MASK | HF_LMA_MASK | HF_CS32_MASK | \
664 HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)
668 hflags = (env->segs[R_CS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
669 hflags |= (env->cr[0] & CR0_PE_MASK) << (HF_PE_SHIFT - CR0_PE_SHIFT);
670 hflags |= (env->cr[0] << (HF_MP_SHIFT - CR0_MP_SHIFT)) &
671 (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK);
672 hflags |= (env->eflags & (HF_TF_MASK | HF_VM_MASK | HF_IOPL_MASK));
673 hflags |= (env->cr[4] & CR4_OSFXSR_MASK) <<
674 (HF_OSFXSR_SHIFT - CR4_OSFXSR_SHIFT);
676 if (env->efer & MSR_EFER_LMA) {
677 hflags |= HF_LMA_MASK;
680 if ((hflags & HF_LMA_MASK) && (env->segs[R_CS].flags & DESC_L_MASK)) {
681 hflags |= HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
682 } else {
683 hflags |= (env->segs[R_CS].flags & DESC_B_MASK) >>
684 (DESC_B_SHIFT - HF_CS32_SHIFT);
685 hflags |= (env->segs[R_SS].flags & DESC_B_MASK) >>
686 (DESC_B_SHIFT - HF_SS32_SHIFT);
687 if (!(env->cr[0] & CR0_PE_MASK) ||
688 (env->eflags & VM_MASK) ||
689 !(hflags & HF_CS32_MASK)) {
690 hflags |= HF_ADDSEG_MASK;
691 } else {
692 hflags |= ((env->segs[R_DS].base |
693 env->segs[R_ES].base |
694 env->segs[R_SS].base) != 0) <<
695 HF_ADDSEG_SHIFT;
698 env->hflags = (env->hflags & HFLAG_COPY_MASK) | hflags;
700 return 0;
703 static int kvm_get_msrs(CPUState *env)
705 struct {
706 struct kvm_msrs info;
707 struct kvm_msr_entry entries[100];
708 } msr_data;
709 struct kvm_msr_entry *msrs = msr_data.entries;
710 int ret, i, n;
712 n = 0;
713 msrs[n++].index = MSR_IA32_SYSENTER_CS;
714 msrs[n++].index = MSR_IA32_SYSENTER_ESP;
715 msrs[n++].index = MSR_IA32_SYSENTER_EIP;
716 if (kvm_has_msr_star(env))
717 msrs[n++].index = MSR_STAR;
718 msrs[n++].index = MSR_IA32_TSC;
719 #ifdef TARGET_X86_64
720 /* FIXME lm_capable_kernel */
721 msrs[n++].index = MSR_CSTAR;
722 msrs[n++].index = MSR_KERNELGSBASE;
723 msrs[n++].index = MSR_FMASK;
724 msrs[n++].index = MSR_LSTAR;
725 #endif
726 msrs[n++].index = MSR_KVM_SYSTEM_TIME;
727 msrs[n++].index = MSR_KVM_WALL_CLOCK;
729 msr_data.info.nmsrs = n;
730 ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);
731 if (ret < 0)
732 return ret;
734 for (i = 0; i < ret; i++) {
735 switch (msrs[i].index) {
736 case MSR_IA32_SYSENTER_CS:
737 env->sysenter_cs = msrs[i].data;
738 break;
739 case MSR_IA32_SYSENTER_ESP:
740 env->sysenter_esp = msrs[i].data;
741 break;
742 case MSR_IA32_SYSENTER_EIP:
743 env->sysenter_eip = msrs[i].data;
744 break;
745 case MSR_STAR:
746 env->star = msrs[i].data;
747 break;
748 #ifdef TARGET_X86_64
749 case MSR_CSTAR:
750 env->cstar = msrs[i].data;
751 break;
752 case MSR_KERNELGSBASE:
753 env->kernelgsbase = msrs[i].data;
754 break;
755 case MSR_FMASK:
756 env->fmask = msrs[i].data;
757 break;
758 case MSR_LSTAR:
759 env->lstar = msrs[i].data;
760 break;
761 #endif
762 case MSR_IA32_TSC:
763 env->tsc = msrs[i].data;
764 break;
765 case MSR_KVM_SYSTEM_TIME:
766 env->system_time_msr = msrs[i].data;
767 break;
768 case MSR_KVM_WALL_CLOCK:
769 env->wall_clock_msr = msrs[i].data;
770 break;
774 return 0;
777 static int kvm_put_mp_state(CPUState *env)
779 struct kvm_mp_state mp_state = { .mp_state = env->mp_state };
781 return kvm_vcpu_ioctl(env, KVM_SET_MP_STATE, &mp_state);
784 static int kvm_get_mp_state(CPUState *env)
786 struct kvm_mp_state mp_state;
787 int ret;
789 ret = kvm_vcpu_ioctl(env, KVM_GET_MP_STATE, &mp_state);
790 if (ret < 0) {
791 return ret;
793 env->mp_state = mp_state.mp_state;
794 return 0;
797 static int kvm_put_vcpu_events(CPUState *env, int level)
799 #ifdef KVM_CAP_VCPU_EVENTS
800 struct kvm_vcpu_events events;
802 if (!kvm_has_vcpu_events()) {
803 return 0;
806 events.exception.injected = (env->exception_injected >= 0);
807 events.exception.nr = env->exception_injected;
808 events.exception.has_error_code = env->has_error_code;
809 events.exception.error_code = env->error_code;
811 events.interrupt.injected = (env->interrupt_injected >= 0);
812 events.interrupt.nr = env->interrupt_injected;
813 events.interrupt.soft = env->soft_interrupt;
815 events.nmi.injected = env->nmi_injected;
816 events.nmi.pending = env->nmi_pending;
817 events.nmi.masked = !!(env->hflags2 & HF2_NMI_MASK);
819 events.sipi_vector = env->sipi_vector;
821 events.flags = 0;
822 if (level >= KVM_PUT_RESET_STATE) {
823 events.flags |=
824 KVM_VCPUEVENT_VALID_NMI_PENDING | KVM_VCPUEVENT_VALID_SIPI_VECTOR;
827 return kvm_vcpu_ioctl(env, KVM_SET_VCPU_EVENTS, &events);
828 #else
829 return 0;
830 #endif
833 static int kvm_get_vcpu_events(CPUState *env)
835 #ifdef KVM_CAP_VCPU_EVENTS
836 struct kvm_vcpu_events events;
837 int ret;
839 if (!kvm_has_vcpu_events()) {
840 return 0;
843 ret = kvm_vcpu_ioctl(env, KVM_GET_VCPU_EVENTS, &events);
844 if (ret < 0) {
845 return ret;
847 env->exception_injected =
848 events.exception.injected ? events.exception.nr : -1;
849 env->has_error_code = events.exception.has_error_code;
850 env->error_code = events.exception.error_code;
852 env->interrupt_injected =
853 events.interrupt.injected ? events.interrupt.nr : -1;
854 env->soft_interrupt = events.interrupt.soft;
856 env->nmi_injected = events.nmi.injected;
857 env->nmi_pending = events.nmi.pending;
858 if (events.nmi.masked) {
859 env->hflags2 |= HF2_NMI_MASK;
860 } else {
861 env->hflags2 &= ~HF2_NMI_MASK;
864 env->sipi_vector = events.sipi_vector;
865 #endif
867 return 0;
870 static int kvm_guest_debug_workarounds(CPUState *env)
872 int ret = 0;
873 #ifdef KVM_CAP_SET_GUEST_DEBUG
874 unsigned long reinject_trap = 0;
876 if (!kvm_has_vcpu_events()) {
877 if (env->exception_injected == 1) {
878 reinject_trap = KVM_GUESTDBG_INJECT_DB;
879 } else if (env->exception_injected == 3) {
880 reinject_trap = KVM_GUESTDBG_INJECT_BP;
882 env->exception_injected = -1;
886 * Kernels before KVM_CAP_X86_ROBUST_SINGLESTEP overwrote flags.TF
887 * injected via SET_GUEST_DEBUG while updating GP regs. Work around this
888 * by updating the debug state once again if single-stepping is on.
889 * Another reason to call kvm_update_guest_debug here is a pending debug
890 * trap raise by the guest. On kernels without SET_VCPU_EVENTS we have to
891 * reinject them via SET_GUEST_DEBUG.
893 if (reinject_trap ||
894 (!kvm_has_robust_singlestep() && env->singlestep_enabled)) {
895 ret = kvm_update_guest_debug(env, reinject_trap);
897 #endif /* KVM_CAP_SET_GUEST_DEBUG */
898 return ret;
901 static int kvm_put_debugregs(CPUState *env)
903 #ifdef KVM_CAP_DEBUGREGS
904 struct kvm_debugregs dbgregs;
905 int i;
907 if (!kvm_has_debugregs()) {
908 return 0;
911 for (i = 0; i < 4; i++) {
912 dbgregs.db[i] = env->dr[i];
914 dbgregs.dr6 = env->dr[6];
915 dbgregs.dr7 = env->dr[7];
916 dbgregs.flags = 0;
918 return kvm_vcpu_ioctl(env, KVM_SET_DEBUGREGS, &dbgregs);
919 #else
920 return 0;
921 #endif
924 static int kvm_get_debugregs(CPUState *env)
926 #ifdef KVM_CAP_DEBUGREGS
927 struct kvm_debugregs dbgregs;
928 int i, ret;
930 if (!kvm_has_debugregs()) {
931 return 0;
934 ret = kvm_vcpu_ioctl(env, KVM_GET_DEBUGREGS, &dbgregs);
935 if (ret < 0) {
936 return ret;
938 for (i = 0; i < 4; i++) {
939 env->dr[i] = dbgregs.db[i];
941 env->dr[4] = env->dr[6] = dbgregs.dr6;
942 env->dr[5] = env->dr[7] = dbgregs.dr7;
943 #endif
945 return 0;
948 int kvm_arch_put_registers(CPUState *env, int level)
950 int ret;
952 assert(cpu_is_stopped(env) || qemu_cpu_self(env));
954 ret = kvm_getput_regs(env, 1);
955 if (ret < 0)
956 return ret;
958 ret = kvm_put_fpu(env);
959 if (ret < 0)
960 return ret;
962 ret = kvm_put_sregs(env);
963 if (ret < 0)
964 return ret;
966 ret = kvm_put_msrs(env, level);
967 if (ret < 0)
968 return ret;
970 if (level >= KVM_PUT_RESET_STATE) {
971 ret = kvm_put_mp_state(env);
972 if (ret < 0)
973 return ret;
976 ret = kvm_put_vcpu_events(env, level);
977 if (ret < 0)
978 return ret;
980 /* must be last */
981 ret = kvm_guest_debug_workarounds(env);
982 if (ret < 0)
983 return ret;
985 ret = kvm_put_debugregs(env);
986 if (ret < 0)
987 return ret;
989 return 0;
992 int kvm_arch_get_registers(CPUState *env)
994 int ret;
996 assert(cpu_is_stopped(env) || qemu_cpu_self(env));
998 ret = kvm_getput_regs(env, 0);
999 if (ret < 0)
1000 return ret;
1002 ret = kvm_get_fpu(env);
1003 if (ret < 0)
1004 return ret;
1006 ret = kvm_get_sregs(env);
1007 if (ret < 0)
1008 return ret;
1010 ret = kvm_get_msrs(env);
1011 if (ret < 0)
1012 return ret;
1014 ret = kvm_get_mp_state(env);
1015 if (ret < 0)
1016 return ret;
1018 ret = kvm_get_vcpu_events(env);
1019 if (ret < 0)
1020 return ret;
1022 ret = kvm_get_debugregs(env);
1023 if (ret < 0)
1024 return ret;
1026 return 0;
1029 int kvm_arch_pre_run(CPUState *env, struct kvm_run *run)
1031 /* Try to inject an interrupt if the guest can accept it */
1032 if (run->ready_for_interrupt_injection &&
1033 (env->interrupt_request & CPU_INTERRUPT_HARD) &&
1034 (env->eflags & IF_MASK)) {
1035 int irq;
1037 env->interrupt_request &= ~CPU_INTERRUPT_HARD;
1038 irq = cpu_get_pic_interrupt(env);
1039 if (irq >= 0) {
1040 struct kvm_interrupt intr;
1041 intr.irq = irq;
1042 /* FIXME: errors */
1043 DPRINTF("injected interrupt %d\n", irq);
1044 kvm_vcpu_ioctl(env, KVM_INTERRUPT, &intr);
1048 /* If we have an interrupt but the guest is not ready to receive an
1049 * interrupt, request an interrupt window exit. This will
1050 * cause a return to userspace as soon as the guest is ready to
1051 * receive interrupts. */
1052 if ((env->interrupt_request & CPU_INTERRUPT_HARD))
1053 run->request_interrupt_window = 1;
1054 else
1055 run->request_interrupt_window = 0;
1057 DPRINTF("setting tpr\n");
1058 run->cr8 = cpu_get_apic_tpr(env->apic_state);
1060 return 0;
1063 int kvm_arch_post_run(CPUState *env, struct kvm_run *run)
1065 if (run->if_flag)
1066 env->eflags |= IF_MASK;
1067 else
1068 env->eflags &= ~IF_MASK;
1070 cpu_set_apic_tpr(env->apic_state, run->cr8);
1071 cpu_set_apic_base(env->apic_state, run->apic_base);
1073 return 0;
1076 int kvm_arch_process_irqchip_events(CPUState *env)
1078 if (env->interrupt_request & CPU_INTERRUPT_INIT) {
1079 kvm_cpu_synchronize_state(env);
1080 do_cpu_init(env);
1081 env->exception_index = EXCP_HALTED;
1084 if (env->interrupt_request & CPU_INTERRUPT_SIPI) {
1085 kvm_cpu_synchronize_state(env);
1086 do_cpu_sipi(env);
1089 return env->halted;
1092 static int kvm_handle_halt(CPUState *env)
1094 if (!((env->interrupt_request & CPU_INTERRUPT_HARD) &&
1095 (env->eflags & IF_MASK)) &&
1096 !(env->interrupt_request & CPU_INTERRUPT_NMI)) {
1097 env->halted = 1;
1098 env->exception_index = EXCP_HLT;
1099 return 0;
1102 return 1;
1105 int kvm_arch_handle_exit(CPUState *env, struct kvm_run *run)
1107 int ret = 0;
1109 switch (run->exit_reason) {
1110 case KVM_EXIT_HLT:
1111 DPRINTF("handle_hlt\n");
1112 ret = kvm_handle_halt(env);
1113 break;
1116 return ret;
1119 #ifdef KVM_CAP_SET_GUEST_DEBUG
1120 int kvm_arch_insert_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
1122 static const uint8_t int3 = 0xcc;
1124 if (cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 0) ||
1125 cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&int3, 1, 1))
1126 return -EINVAL;
1127 return 0;
1130 int kvm_arch_remove_sw_breakpoint(CPUState *env, struct kvm_sw_breakpoint *bp)
1132 uint8_t int3;
1134 if (cpu_memory_rw_debug(env, bp->pc, &int3, 1, 0) || int3 != 0xcc ||
1135 cpu_memory_rw_debug(env, bp->pc, (uint8_t *)&bp->saved_insn, 1, 1))
1136 return -EINVAL;
1137 return 0;
1140 static struct {
1141 target_ulong addr;
1142 int len;
1143 int type;
1144 } hw_breakpoint[4];
1146 static int nb_hw_breakpoint;
1148 static int find_hw_breakpoint(target_ulong addr, int len, int type)
1150 int n;
1152 for (n = 0; n < nb_hw_breakpoint; n++)
1153 if (hw_breakpoint[n].addr == addr && hw_breakpoint[n].type == type &&
1154 (hw_breakpoint[n].len == len || len == -1))
1155 return n;
1156 return -1;
1159 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
1160 target_ulong len, int type)
1162 switch (type) {
1163 case GDB_BREAKPOINT_HW:
1164 len = 1;
1165 break;
1166 case GDB_WATCHPOINT_WRITE:
1167 case GDB_WATCHPOINT_ACCESS:
1168 switch (len) {
1169 case 1:
1170 break;
1171 case 2:
1172 case 4:
1173 case 8:
1174 if (addr & (len - 1))
1175 return -EINVAL;
1176 break;
1177 default:
1178 return -EINVAL;
1180 break;
1181 default:
1182 return -ENOSYS;
1185 if (nb_hw_breakpoint == 4)
1186 return -ENOBUFS;
1188 if (find_hw_breakpoint(addr, len, type) >= 0)
1189 return -EEXIST;
1191 hw_breakpoint[nb_hw_breakpoint].addr = addr;
1192 hw_breakpoint[nb_hw_breakpoint].len = len;
1193 hw_breakpoint[nb_hw_breakpoint].type = type;
1194 nb_hw_breakpoint++;
1196 return 0;
1199 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
1200 target_ulong len, int type)
1202 int n;
1204 n = find_hw_breakpoint(addr, (type == GDB_BREAKPOINT_HW) ? 1 : len, type);
1205 if (n < 0)
1206 return -ENOENT;
1208 nb_hw_breakpoint--;
1209 hw_breakpoint[n] = hw_breakpoint[nb_hw_breakpoint];
1211 return 0;
1214 void kvm_arch_remove_all_hw_breakpoints(void)
1216 nb_hw_breakpoint = 0;
1219 static CPUWatchpoint hw_watchpoint;
1221 int kvm_arch_debug(struct kvm_debug_exit_arch *arch_info)
1223 int handle = 0;
1224 int n;
1226 if (arch_info->exception == 1) {
1227 if (arch_info->dr6 & (1 << 14)) {
1228 if (cpu_single_env->singlestep_enabled)
1229 handle = 1;
1230 } else {
1231 for (n = 0; n < 4; n++)
1232 if (arch_info->dr6 & (1 << n))
1233 switch ((arch_info->dr7 >> (16 + n*4)) & 0x3) {
1234 case 0x0:
1235 handle = 1;
1236 break;
1237 case 0x1:
1238 handle = 1;
1239 cpu_single_env->watchpoint_hit = &hw_watchpoint;
1240 hw_watchpoint.vaddr = hw_breakpoint[n].addr;
1241 hw_watchpoint.flags = BP_MEM_WRITE;
1242 break;
1243 case 0x3:
1244 handle = 1;
1245 cpu_single_env->watchpoint_hit = &hw_watchpoint;
1246 hw_watchpoint.vaddr = hw_breakpoint[n].addr;
1247 hw_watchpoint.flags = BP_MEM_ACCESS;
1248 break;
1251 } else if (kvm_find_sw_breakpoint(cpu_single_env, arch_info->pc))
1252 handle = 1;
1254 if (!handle) {
1255 cpu_synchronize_state(cpu_single_env);
1256 assert(cpu_single_env->exception_injected == -1);
1258 cpu_single_env->exception_injected = arch_info->exception;
1259 cpu_single_env->has_error_code = 0;
1262 return handle;
1265 void kvm_arch_update_guest_debug(CPUState *env, struct kvm_guest_debug *dbg)
1267 const uint8_t type_code[] = {
1268 [GDB_BREAKPOINT_HW] = 0x0,
1269 [GDB_WATCHPOINT_WRITE] = 0x1,
1270 [GDB_WATCHPOINT_ACCESS] = 0x3
1272 const uint8_t len_code[] = {
1273 [1] = 0x0, [2] = 0x1, [4] = 0x3, [8] = 0x2
1275 int n;
1277 if (kvm_sw_breakpoints_active(env))
1278 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP;
1280 if (nb_hw_breakpoint > 0) {
1281 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
1282 dbg->arch.debugreg[7] = 0x0600;
1283 for (n = 0; n < nb_hw_breakpoint; n++) {
1284 dbg->arch.debugreg[n] = hw_breakpoint[n].addr;
1285 dbg->arch.debugreg[7] |= (2 << (n * 2)) |
1286 (type_code[hw_breakpoint[n].type] << (16 + n*4)) |
1287 (len_code[hw_breakpoint[n].len] << (18 + n*4));
1291 #endif /* KVM_CAP_SET_GUEST_DEBUG */
1293 bool kvm_arch_stop_on_emulation_error(CPUState *env)
1295 return !(env->cr[0] & CR0_PE_MASK) ||
1296 ((env->segs[R_CS].selector & 3) != 3);