kqemu.c

qemu虚拟机代码

    }

    /* we must restore the default rounding state */
    asm volatile ("fninit");
    fpuc = 0x037f | (env->fpuc & (3 << 10));
    asm volatile("fldcw %0" : : "m" (fpuc));
}

static int do_syscall(CPUState *env,
                      struct kqemu_cpu_state *kenv)
{
    int selector;
    
    selector = (env->star >> 32) & 0xffff;
#ifdef __x86_64__
    if (env->hflags & HF_LMA_MASK) {
        env->regs[R_ECX] = kenv->next_eip;
        env->regs[11] = env->eflags;

        cpu_x86_set_cpl(env, 0);
        cpu_x86_load_seg_cache(env, R_CS, selector & 0xfffc, 
                               0, 0xffffffff, 
                               DESC_G_MASK | DESC_B_MASK | DESC_P_MASK |
                               DESC_S_MASK |
                               DESC_CS_MASK | DESC_R_MASK | DESC_A_MASK | DESC_L_MASK);
        cpu_x86_load_seg_cache(env, R_SS, (selector + 8) & 0xfffc, 
                               0, 0xffffffff,
                               DESC_G_MASK | DESC_B_MASK | DESC_P_MASK |
                               DESC_S_MASK |
                               DESC_W_MASK | DESC_A_MASK);
        env->eflags &= ~env->fmask;
        if (env->hflags & HF_CS64_MASK)
            env->eip = env->lstar;
        else
            env->eip = env->cstar;
    } else 
#endif
    {
        env->regs[R_ECX] = (uint32_t)kenv->next_eip;
        
        cpu_x86_set_cpl(env, 0);
        cpu_x86_load_seg_cache(env, R_CS, selector & 0xfffc, 
                           0, 0xffffffff, 
                               DESC_G_MASK | DESC_B_MASK | DESC_P_MASK |
                               DESC_S_MASK |
                               DESC_CS_MASK | DESC_R_MASK | DESC_A_MASK);
        cpu_x86_load_seg_cache(env, R_SS, (selector + 8) & 0xfffc, 
                               0, 0xffffffff,
                               DESC_G_MASK | DESC_B_MASK | DESC_P_MASK |
                               DESC_S_MASK |
                               DESC_W_MASK | DESC_A_MASK);
        env->eflags &= ~(IF_MASK | RF_MASK | VM_MASK);
        env->eip = (uint32_t)env->star;
    }
    return 2;
}

#ifdef CONFIG_PROFILER

#define PC_REC_SIZE 1
#define PC_REC_HASH_BITS 16
#define PC_REC_HASH_SIZE (1 << PC_REC_HASH_BITS)

typedef struct PCRecord {
    unsigned long pc;
    int64_t count;
    struct PCRecord *next;
} PCRecord;

static PCRecord *pc_rec_hash[PC_REC_HASH_SIZE];
static int nb_pc_records;

static void kqemu_record_pc(unsigned long pc)
{
    unsigned long h;
    PCRecord **pr, *r;

    h = pc / PC_REC_SIZE;
    h = h ^ (h >> PC_REC_HASH_BITS);
    h &= (PC_REC_HASH_SIZE - 1);
    pr = &pc_rec_hash[h];
    for(;;) {
        r = *pr;
        if (r == NULL)
            break;
        if (r->pc == pc) {
            r->count++;
            return;
        }
        pr = &r->next;
    }
    r = malloc(sizeof(PCRecord));
    r->count = 1;
    r->pc = pc;
    r->next = NULL;
    *pr = r;
    nb_pc_records++;
}

static int pc_rec_cmp(const void *p1, const void *p2)
{
    PCRecord *r1 = *(PCRecord **)p1;
    PCRecord *r2 = *(PCRecord **)p2;
    if (r1->count < r2->count)
        return 1;
    else if (r1->count == r2->count)
        return 0;
    else
        return -1;
}

static void kqemu_record_flush(void)
{
    PCRecord *r, *r_next;
    int h;

    for(h = 0; h < PC_REC_HASH_SIZE; h++) {
        for(r = pc_rec_hash[h]; r != NULL; r = r_next) {
            r_next = r->next;
            free(r);
        }
        pc_rec_hash[h] = NULL;
    }
    nb_pc_records = 0;
}

void kqemu_record_dump(void)
{
    PCRecord **pr, *r;
    int i, h;
    FILE *f;
    int64_t total, sum;

    pr = malloc(sizeof(PCRecord *) * nb_pc_records);
    i = 0;
    total = 0;
    for(h = 0; h < PC_REC_HASH_SIZE; h++) {
        for(r = pc_rec_hash[h]; r != NULL; r = r->next) {
            pr[i++] = r;
            total += r->count;
        }
    }
    qsort(pr, nb_pc_records, sizeof(PCRecord *), pc_rec_cmp);
    
    f = fopen("/tmp/kqemu.stats", "w");
    if (!f) {
        perror("/tmp/kqemu.stats");
        exit(1);
    }
    fprintf(f, "total: %lld\n", total);
    sum = 0;
    for(i = 0; i < nb_pc_records; i++) {
        r = pr[i];
        sum += r->count;
        fprintf(f, "%08lx: %lld %0.2f%% %0.2f%%\n", 
                r->pc, 
                r->count, 
                (double)r->count / (double)total * 100.0,
                (double)sum / (double)total * 100.0);
    }
    fclose(f);
    free(pr);

    kqemu_record_flush();
}
#endif

int kqemu_cpu_exec(CPUState *env)
{
    struct kqemu_cpu_state kcpu_state, *kenv = &kcpu_state;
    int ret, cpl, i;
#ifdef CONFIG_PROFILER
    int64_t ti;
#endif

#ifdef _WIN32
    DWORD temp;
#endif

#ifdef CONFIG_PROFILER
    ti = profile_getclock();
#endif
#ifdef DEBUG
    if (loglevel & CPU_LOG_INT) {
        fprintf(logfile, "kqemu: cpu_exec: enter\n");
        cpu_dump_state(env, logfile, fprintf, 0);
    }
#endif
    memcpy(kenv->regs, env->regs, sizeof(kenv->regs));
    kenv->eip = env->eip;
    kenv->eflags = env->eflags;
    memcpy(&kenv->segs, &env->segs, sizeof(env->segs));
    memcpy(&kenv->ldt, &env->ldt, sizeof(env->ldt));
    memcpy(&kenv->tr, &env->tr, sizeof(env->tr));
    memcpy(&kenv->gdt, &env->gdt, sizeof(env->gdt));
    memcpy(&kenv->idt, &env->idt, sizeof(env->idt));
    kenv->cr0 = env->cr[0];
    kenv->cr2 = env->cr[2];
    kenv->cr3 = env->cr[3];
    kenv->cr4 = env->cr[4];
    kenv->a20_mask = env->a20_mask;
#if KQEMU_VERSION >= 0x010100
    kenv->efer = env->efer;
#endif
#if KQEMU_VERSION >= 0x010300
    kenv->tsc_offset = 0;
    kenv->star = env->star;
    kenv->sysenter_cs = env->sysenter_cs;
    kenv->sysenter_esp = env->sysenter_esp;
    kenv->sysenter_eip = env->sysenter_eip;
#ifdef __x86_64__
    kenv->lstar = env->lstar;
    kenv->cstar = env->cstar;
    kenv->fmask = env->fmask;
    kenv->kernelgsbase = env->kernelgsbase;
#endif
#endif
    if (env->dr[7] & 0xff) {
        kenv->dr7 = env->dr[7];
        kenv->dr0 = env->dr[0];
        kenv->dr1 = env->dr[1];
        kenv->dr2 = env->dr[2];
        kenv->dr3 = env->dr[3];
    } else {
        kenv->dr7 = 0;
    }
    kenv->dr6 = env->dr[6];
    cpl = (env->hflags & HF_CPL_MASK);
    kenv->cpl = cpl;
    kenv->nb_pages_to_flush = nb_pages_to_flush;
#if KQEMU_VERSION >= 0x010200
    kenv->user_only = (env->kqemu_enabled == 1);
    kenv->nb_ram_pages_to_update = nb_ram_pages_to_update;
#endif
    nb_ram_pages_to_update = 0;
    
#if KQEMU_VERSION >= 0x010300
    kenv->nb_modified_ram_pages = nb_modified_ram_pages;
#endif
    kqemu_reset_modified_ram_pages();

    if (env->cpuid_features & CPUID_FXSR)
        restore_native_fp_fxrstor(env);
    else
        restore_native_fp_frstor(env);

#ifdef _WIN32
    if (DeviceIoControl(kqemu_fd, KQEMU_EXEC,
                        kenv, sizeof(struct kqemu_cpu_state),
                        kenv, sizeof(struct kqemu_cpu_state),
                        &temp, NULL)) {
        ret = kenv->retval;
    } else {
        ret = -1;
    }
#else
#if KQEMU_VERSION >= 0x010100
    ioctl(kqemu_fd, KQEMU_EXEC, kenv);
    ret = kenv->retval;
#else
    ret = ioctl(kqemu_fd, KQEMU_EXEC, kenv);
#endif
#endif
    if (env->cpuid_features & CPUID_FXSR)
        save_native_fp_fxsave(env);
    else
        save_native_fp_fsave(env);

    memcpy(env->regs, kenv->regs, sizeof(env->regs));
    env->eip = kenv->eip;
    env->eflags = kenv->eflags;
    memcpy(env->segs, kenv->segs, sizeof(env->segs));
    cpu_x86_set_cpl(env, kenv->cpl);
    memcpy(&env->ldt, &kenv->ldt, sizeof(env->ldt));
#if 0
    /* no need to restore that */
    memcpy(env->tr, kenv->tr, sizeof(env->tr));
    memcpy(env->gdt, kenv->gdt, sizeof(env->gdt));
    memcpy(env->idt, kenv->idt, sizeof(env->idt));
    env->a20_mask = kenv->a20_mask;
#endif
    env->cr[0] = kenv->cr0;
    env->cr[4] = kenv->cr4;
    env->cr[3] = kenv->cr3;
    env->cr[2] = kenv->cr2;
    env->dr[6] = kenv->dr6;
#if KQEMU_VERSION >= 0x010300
#ifdef __x86_64__
    env->kernelgsbase = kenv->kernelgsbase;
#endif
#endif

    /* flush pages as indicated by kqemu */
    if (kenv->nb_pages_to_flush >= KQEMU_FLUSH_ALL) {
        tlb_flush(env, 1);
    } else {
        for(i = 0; i < kenv->nb_pages_to_flush; i++) {
            tlb_flush_page(env, pages_to_flush[i]);
        }
    }
    nb_pages_to_flush = 0;

#ifdef CONFIG_PROFILER
    kqemu_time += profile_getclock() - ti;
    kqemu_exec_count++;
#endif

#if KQEMU_VERSION >= 0x010200
    if (kenv->nb_ram_pages_to_update > 0) {
        cpu_tlb_update_dirty(env);
    }
#endif

#if KQEMU_VERSION >= 0x010300
    if (kenv->nb_modified_ram_pages > 0) {
        for(i = 0; i < kenv->nb_modified_ram_pages; i++) {
            unsigned long addr;
            addr = modified_ram_pages[i];
            tb_invalidate_phys_page_range(addr, addr + TARGET_PAGE_SIZE, 0);
        }
    }
#endif

    /* restore the hidden flags */
    {
        unsigned int new_hflags;
#ifdef TARGET_X86_64
        if ((env->hflags & HF_LMA_MASK) && 
            (env->segs[R_CS].flags & DESC_L_MASK)) {
            /* long mode */
            new_hflags = HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK;
        } else
#endif
        {
            /* legacy / compatibility case */
            new_hflags = (env->segs[R_CS].flags & DESC_B_MASK)
                >> (DESC_B_SHIFT - HF_CS32_SHIFT);
            new_hflags |= (env->segs[R_SS].flags & DESC_B_MASK)
                >> (DESC_B_SHIFT - HF_SS32_SHIFT);
            if (!(env->cr[0] & CR0_PE_MASK) || 
                   (env->eflags & VM_MASK) ||
                   !(env->hflags & HF_CS32_MASK)) {
                /* XXX: try to avoid this test. The problem comes from the
                   fact that is real mode or vm86 mode we only modify the
                   'base' and 'selector' fields of the segment cache to go
                   faster. A solution may be to force addseg to one in
                   translate-i386.c. */
                new_hflags |= HF_ADDSEG_MASK;
            } else {
                new_hflags |= ((env->segs[R_DS].base | 
                                env->segs[R_ES].base |
                                env->segs[R_SS].base) != 0) << 
                    HF_ADDSEG_SHIFT;
            }
        }
        env->hflags = (env->hflags & 
           ~(HF_CS32_MASK | HF_SS32_MASK | HF_CS64_MASK | HF_ADDSEG_MASK)) |
            new_hflags;
    }
    /* update FPU flags */
    env->hflags = (env->hflags & ~(HF_MP_MASK | HF_EM_MASK | HF_TS_MASK)) |
        ((env->cr[0] << (HF_MP_SHIFT - 1)) & (HF_MP_MASK | HF_EM_MASK | HF_TS_MASK));
    if (env->cr[4] & CR4_OSFXSR_MASK)
        env->hflags |= HF_OSFXSR_MASK;
    else
        env->hflags &= ~HF_OSFXSR_MASK;
        
#ifdef DEBUG
    if (loglevel & CPU_LOG_INT) {
        fprintf(logfile, "kqemu: kqemu_cpu_exec: ret=0x%x\n", ret);
    }
#endif
    if (ret == KQEMU_RET_SYSCALL) {
        /* syscall instruction */
        return do_syscall(env, kenv);
    } else 
    if ((ret & 0xff00) == KQEMU_RET_INT) {
        env->exception_index = ret & 0xff;
        env->error_code = 0;
        env->exception_is_int = 1;
        env->exception_next_eip = kenv->next_eip;
#ifdef CONFIG_PROFILER
        kqemu_ret_int_count++;
#endif
#ifdef DEBUG
        if (loglevel & CPU_LOG_INT) {
            fprintf(logfile, "kqemu: interrupt v=%02x:\n", 
                    env->exception_index);
            cpu_dump_state(env, logfile, fprintf, 0);
        }
#endif
        return 1;
    } else if ((ret & 0xff00) == KQEMU_RET_EXCEPTION) {
        env->exception_index = ret & 0xff;
        env->error_code = kenv->error_code;
        env->exception_is_int = 0;
        env->exception_next_eip = 0;
#ifdef CONFIG_PROFILER
        kqemu_ret_excp_count++;
#endif
#ifdef DEBUG
        if (loglevel & CPU_LOG_INT) {
            fprintf(logfile, "kqemu: exception v=%02x e=%04x:\n",
                    env->exception_index, env->error_code);
            cpu_dump_state(env, logfile, fprintf, 0);
        }
#endif
        return 1;
    } else if (ret == KQEMU_RET_INTR) {
#ifdef CONFIG_PROFILER
        kqemu_ret_intr_count++;
#endif
#ifdef DEBUG
        if (loglevel & CPU_LOG_INT) {
            cpu_dump_state(env, logfile, fprintf, 0);
        }
#endif
        return 0;
    } else if (ret == KQEMU_RET_SOFTMMU) { 
#ifdef CONFIG_PROFILER
        {
            unsigned long pc = env->eip + env->segs[R_CS].base;
            kqemu_record_pc(pc);
        }
#endif
#ifdef DEBUG
        if (loglevel & CPU_LOG_INT) {
            cpu_dump_state(env, logfile, fprintf, 0);
        }
#endif
        return 2;
    } else {
        cpu_dump_state(env, stderr, fprintf, 0);
        fprintf(stderr, "Unsupported return value: 0x%x\n", ret);
        exit(1);
    }
    return 0;
}

void kqemu_cpu_interrupt(CPUState *env)
{
#if defined(_WIN32) && KQEMU_VERSION >= 0x010101
    /* cancelling the I/O request causes KQEMU to finish executing the 
       current block and successfully returning. */
    CancelIo(kqemu_fd);
#endif
}

#endif