[qemu.git] / kqemu.c

/*
 *  KQEMU support
 * 
 *  Copyright (c) 2005 Fabrice Bellard
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */
#include "config.h"
#ifdef _WIN32
#include <windows.h>
#include <winioctl.h>
#else
#include <sys/types.h>
#include <sys/mman.h>
#include <sys/ioctl.h>
#endif
#include <stdlib.h>
#include <stdio.h>
#include <stdarg.h>
#include <string.h>
#include <errno.h>
#include <unistd.h>
#include <inttypes.h>

#include "cpu.h"
#include "exec-all.h"

#ifdef USE_KQEMU

#define DEBUG
//#define PROFILE

#include <unistd.h>
#include <fcntl.h>
#include "kqemu/kqemu.h"

/* compatibility stuff */
#ifndef KQEMU_RET_SYSCALL
#define KQEMU_RET_SYSCALL   0x0300 /* syscall insn */
#endif
#ifndef KQEMU_MAX_RAM_PAGES_TO_UPDATE
#define KQEMU_MAX_RAM_PAGES_TO_UPDATE 512
#define KQEMU_RAM_PAGES_UPDATE_ALL (KQEMU_MAX_RAM_PAGES_TO_UPDATE + 1)
#endif

#ifdef _WIN32
#define KQEMU_DEVICE "\\\\.\\kqemu"
#else
#define KQEMU_DEVICE "/dev/kqemu"
#endif

#ifdef _WIN32
#define KQEMU_INVALID_FD INVALID_HANDLE_VALUE
HANDLE kqemu_fd = KQEMU_INVALID_FD;
#define kqemu_closefd(x) CloseHandle(x)
#else
#define KQEMU_INVALID_FD -1
int kqemu_fd = KQEMU_INVALID_FD;
#define kqemu_closefd(x) close(x)
#endif

int kqemu_allowed = 1;
unsigned long *pages_to_flush;
unsigned int nb_pages_to_flush;
unsigned long *ram_pages_to_update;
unsigned int nb_ram_pages_to_update;
extern uint32_t **l1_phys_map;

#define cpuid(index, eax, ebx, ecx, edx) \
  asm volatile ("cpuid" \
                : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) \
                : "0" (index))

#ifdef __x86_64__
static int is_cpuid_supported(void)
{
    return 1;
}
#else
static int is_cpuid_supported(void)
{
    int v0, v1;
    asm volatile ("pushf\n"
                  "popl %0\n"
                  "movl %0, %1\n"
                  "xorl $0x00200000, %0\n"
                  "pushl %0\n"
                  "popf\n"
                  "pushf\n"
                  "popl %0\n"
                  : "=a" (v0), "=d" (v1)
                  :
                  : "cc");
    return (v0 != v1);
}
#endif

static void kqemu_update_cpuid(CPUState *env)
{
    int critical_features_mask, features;
    uint32_t eax, ebx, ecx, edx;

    /* the following features are kept identical on the host and
       target cpus because they are important for user code. Strictly
       speaking, only SSE really matters because the OS must support
       it if the user code uses it. */
    critical_features_mask = 
        CPUID_CMOV | CPUID_CX8 | 
        CPUID_FXSR | CPUID_MMX | CPUID_SSE | 
        CPUID_SSE2 | CPUID_SEP;
    if (!is_cpuid_supported()) {
        features = 0;
    } else {
        cpuid(1, eax, ebx, ecx, edx);
        features = edx;
    }
#ifdef __x86_64__
    /* NOTE: on x86_64 CPUs, SYSENTER is not supported in
       compatibility mode, so in order to have the best performances
       it is better not to use it */
    features &= ~CPUID_SEP;
#endif
    env->cpuid_features = (env->cpuid_features & ~critical_features_mask) |
        (features & critical_features_mask);
    /* XXX: we could update more of the target CPUID state so that the
       non accelerated code sees exactly the same CPU features as the
       accelerated code */
}

int kqemu_init(CPUState *env)
{
    struct kqemu_init init;
    int ret, version;
#ifdef _WIN32
    DWORD temp;
#endif

    if (!kqemu_allowed)
        return -1;

#ifdef _WIN32
    kqemu_fd = CreateFile(KQEMU_DEVICE, GENERIC_WRITE | GENERIC_READ,
                          FILE_SHARE_READ | FILE_SHARE_WRITE,
                          NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL,
                          NULL);
#else
    kqemu_fd = open(KQEMU_DEVICE, O_RDWR);
#endif
    if (kqemu_fd == KQEMU_INVALID_FD) {
        fprintf(stderr, "Could not open '%s' - QEMU acceleration layer not activated\n", KQEMU_DEVICE);
        return -1;
    }
    version = 0;
#ifdef _WIN32
    DeviceIoControl(kqemu_fd, KQEMU_GET_VERSION, NULL, 0,
                    &version, sizeof(version), &temp, NULL);
#else
    ioctl(kqemu_fd, KQEMU_GET_VERSION, &version);
#endif
    if (version != KQEMU_VERSION) {
        fprintf(stderr, "Version mismatch between kqemu module and qemu (%08x %08x) - disabling kqemu use\n",
                version, KQEMU_VERSION);
        goto fail;
    }

    pages_to_flush = qemu_vmalloc(KQEMU_MAX_PAGES_TO_FLUSH * 
                                  sizeof(unsigned long));
    if (!pages_to_flush)
        goto fail;

    ram_pages_to_update = qemu_vmalloc(KQEMU_MAX_RAM_PAGES_TO_UPDATE * 
                                       sizeof(unsigned long));
    if (!ram_pages_to_update)
        goto fail;

    init.ram_base = phys_ram_base;
    init.ram_size = phys_ram_size;
    init.ram_dirty = phys_ram_dirty;
    init.phys_to_ram_map = l1_phys_map;
    init.pages_to_flush = pages_to_flush;
#if KQEMU_VERSION >= 0x010200
    init.ram_pages_to_update = ram_pages_to_update;
#endif
#ifdef _WIN32
    ret = DeviceIoControl(kqemu_fd, KQEMU_INIT, &init, sizeof(init),
                          NULL, 0, &temp, NULL) == TRUE ? 0 : -1;
#else
    ret = ioctl(kqemu_fd, KQEMU_INIT, &init);
#endif
    if (ret < 0) {
        fprintf(stderr, "Error %d while initializing QEMU acceleration layer - disabling it for now\n", ret);
    fail:
        kqemu_closefd(kqemu_fd);
        kqemu_fd = KQEMU_INVALID_FD;
        return -1;
    }
    kqemu_update_cpuid(env);
    env->kqemu_enabled = 1;
    nb_pages_to_flush = 0;
    nb_ram_pages_to_update = 0;
    return 0;
}

void kqemu_flush_page(CPUState *env, target_ulong addr)
{
#ifdef DEBUG
    if (loglevel & CPU_LOG_INT) {
        fprintf(logfile, "kqemu_flush_page: addr=" TARGET_FMT_lx "\n", addr);
    }
#endif
    if (nb_pages_to_flush >= KQEMU_MAX_PAGES_TO_FLUSH)
        nb_pages_to_flush = KQEMU_FLUSH_ALL;
    else
        pages_to_flush[nb_pages_to_flush++] = addr;
}

void kqemu_flush(CPUState *env, int global)
{
#ifdef DEBUG
    if (loglevel & CPU_LOG_INT) {
        fprintf(logfile, "kqemu_flush:\n");
    }
#endif
    nb_pages_to_flush = KQEMU_FLUSH_ALL;
}

void kqemu_set_notdirty(CPUState *env, ram_addr_t ram_addr)
{
#ifdef DEBUG
    if (loglevel & CPU_LOG_INT) {
        fprintf(logfile, "kqemu_set_notdirty: addr=%08lx\n", ram_addr);
    }
#endif
    /* we only track transitions to dirty state */
    if (phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS] != 0xff)
        return;
    if (nb_ram_pages_to_update >= KQEMU_MAX_RAM_PAGES_TO_UPDATE)
        nb_ram_pages_to_update = KQEMU_RAM_PAGES_UPDATE_ALL;
    else
        ram_pages_to_update[nb_ram_pages_to_update++] = ram_addr;
}

struct fpstate {
    uint16_t fpuc;
    uint16_t dummy1;
    uint16_t fpus;
    uint16_t dummy2;
    uint16_t fptag;
    uint16_t dummy3;

    uint32_t fpip;
    uint32_t fpcs;
    uint32_t fpoo;
    uint32_t fpos;
    uint8_t fpregs1[8 * 10];
};

struct fpxstate {
    uint16_t fpuc;
    uint16_t fpus;
    uint16_t fptag;
    uint16_t fop;
    uint32_t fpuip;
    uint16_t cs_sel;
    uint16_t dummy0;
    uint32_t fpudp;
    uint16_t ds_sel;
    uint16_t dummy1;
    uint32_t mxcsr;
    uint32_t mxcsr_mask;
    uint8_t fpregs1[8 * 16];
    uint8_t xmm_regs[16 * 16];
    uint8_t dummy2[96];
};

static struct fpxstate fpx1 __attribute__((aligned(16)));

static void restore_native_fp_frstor(CPUState *env)
{
    int fptag, i, j;
    struct fpstate fp1, *fp = &fp1;
    
    fp->fpuc = env->fpuc;
    fp->fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
    fptag = 0;
    for (i=7; i>=0; i--) {
	fptag <<= 2;
	if (env->fptags[i]) {
            fptag |= 3;
        } else {
            /* the FPU automatically computes it */
        }
    }
    fp->fptag = fptag;
    j = env->fpstt;
    for(i = 0;i < 8; i++) {
        memcpy(&fp->fpregs1[i * 10], &env->fpregs[j].d, 10);
        j = (j + 1) & 7;
    }
    asm volatile ("frstor %0" : "=m" (*fp));
}
 
static void save_native_fp_fsave(CPUState *env)
{
    int fptag, i, j;
    uint16_t fpuc;
    struct fpstate fp1, *fp = &fp1;

    asm volatile ("fsave %0" : : "m" (*fp));
    env->fpuc = fp->fpuc;
    env->fpstt = (fp->fpus >> 11) & 7;
    env->fpus = fp->fpus & ~0x3800;
    fptag = fp->fptag;
    for(i = 0;i < 8; i++) {
        env->fptags[i] = ((fptag & 3) == 3);
        fptag >>= 2;
    }
    j = env->fpstt;
    for(i = 0;i < 8; i++) {
        memcpy(&env->fpregs[j].d, &fp->fpregs1[i * 10], 10);
        j = (j + 1) & 7;
    }
    /* we must restore the default rounding state */
    fpuc = 0x037f | (env->fpuc & (3 << 10));
    asm volatile("fldcw %0" : : "m" (fpuc));
}

static void restore_native_fp_fxrstor(CPUState *env)
{
    struct fpxstate *fp = &fpx1;
    int i, j, fptag;

    fp->fpuc = env->fpuc;
    fp->fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
    fptag = 0;
    for(i = 0; i < 8; i++)
        fptag |= (env->fptags[i] << i);
    fp->fptag = fptag ^ 0xff;

    j = env->fpstt;
    for(i = 0;i < 8; i++) {
        memcpy(&fp->fpregs1[i * 16], &env->fpregs[j].d, 10);
        j = (j + 1) & 7;
    }
    if (env->cpuid_features & CPUID_SSE) {
        fp->mxcsr = env->mxcsr;
        /* XXX: check if DAZ is not available */
        fp->mxcsr_mask = 0xffff;
        memcpy(fp->xmm_regs, env->xmm_regs, CPU_NB_REGS * 16);
    }
    asm volatile ("fxrstor %0" : "=m" (*fp));
}

static void save_native_fp_fxsave(CPUState *env)
{
    struct fpxstate *fp = &fpx1;
    int fptag, i, j;
    uint16_t fpuc;

    asm volatile ("fxsave %0" : : "m" (*fp));
    env->fpuc = fp->fpuc;
    env->fpstt = (fp->fpus >> 11) & 7;
    env->fpus = fp->fpus & ~0x3800;
    fptag = fp->fptag ^ 0xff;
    for(i = 0;i < 8; i++) {
        env->fptags[i] = (fptag >> i) & 1;
    }
    j = env->fpstt;
    for(i = 0;i < 8; i++) {
        memcpy(&env->fpregs[j].d, &fp->fpregs1[i * 16], 10);
        j = (j + 1) & 7;
    }
    if (env->cpuid_features & CPUID_SSE) {
        env->mxcsr = fp->mxcsr;
        memcpy(env->xmm_regs, fp->xmm_regs, CPU_NB_REGS * 16);
    }

    /* 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 PROFILE

#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;

PCRecord *pc_rec_hash[PC_REC_HASH_SIZE];
int nb_pc_records;

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++;
}

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;
}

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);
}
#else
void kqemu_record_dump(void)
{
}
#endif

int kqemu_cpu_exec(CPUState *env)
{
    struct kqemu_cpu_state kcpu_state, *kenv = &kcpu_state;
    int ret;
#ifdef _WIN32
    DWORD temp;
#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 (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];
    kenv->cpl = 3;
    kenv->nb_pages_to_flush = nb_pages_to_flush;
    nb_pages_to_flush = 0;
#if KQEMU_VERSION >= 0x010200
    kenv->user_only = 1;
    kenv->nb_ram_pages_to_update = nb_ram_pages_to_update;
#endif
    nb_ram_pages_to_update = 0;
    
    if (!(kenv->cr0 & CR0_TS_MASK)) {
        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 (!(kenv->cr0 & CR0_TS_MASK)) {
        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));
#if 0
    /* no need to restore that */
    memcpy(env->ldt, kenv->ldt, sizeof(env->ldt));
    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->cr[0] = kenv->cr0;
    env->cr[3] = kenv->cr3;
    env->cr[4] = kenv->cr4;
    env->a20_mask = kenv->a20_mask;
#endif
    env->cr[2] = kenv->cr2;
    env->dr[6] = kenv->dr6;

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

#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 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 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 DEBUG
        if (loglevel & CPU_LOG_INT) {
            cpu_dump_state(env, logfile, fprintf, 0);
        }
#endif
        return 0;
    } else if (ret == KQEMU_RET_SOFTMMU) { 
#ifdef PROFILE
        kqemu_record_pc(env->eip + env->segs[R_CS].base);
#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
Commit	Line	Data
9df217a3 FB	1	/*
	2	* KQEMU support
	3	*
	4	* Copyright (c) 2005 Fabrice Bellard
	5	*
	6	* This library is free software; you can redistribute it and/or
	7	* modify it under the terms of the GNU Lesser General Public
	8	* License as published by the Free Software Foundation; either
	9	* version 2 of the License, or (at your option) any later version.
	10	*
	11	* This library is distributed in the hope that it will be useful,
	12	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	14	* Lesser General Public License for more details.
	15	*
	16	* You should have received a copy of the GNU Lesser General Public
	17	* License along with this library; if not, write to the Free Software
	18	* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
	19	*/
	20	#include "config.h"
	21	#ifdef _WIN32
	22	#include <windows.h>
6e4255f6	23	#include <winioctl.h>
9df217a3 FB	24	#else
	25	#include <sys/types.h>
	26	#include <sys/mman.h>
6e4255f6	27	#include <sys/ioctl.h>
9df217a3 FB	28	#endif
	29	#include <stdlib.h>
	30	#include <stdio.h>
	31	#include <stdarg.h>
	32	#include <string.h>
	33	#include <errno.h>
	34	#include <unistd.h>
	35	#include <inttypes.h>
	36
	37	#include "cpu.h"
	38	#include "exec-all.h"
	39
	40	#ifdef USE_KQEMU
	41
	42	#define DEBUG
aa062973	43	//#define PROFILE
9df217a3 FB	44
	45	#include <unistd.h>
	46	#include <fcntl.h>
9df217a3 FB	47	#include "kqemu/kqemu.h"
9df217a3 FB	48
c28e951f FB	49	/* compatibility stuff */
	50	#ifndef KQEMU_RET_SYSCALL
	51	#define KQEMU_RET_SYSCALL 0x0300 /* syscall insn */
	52	#endif
aa062973 FB	53	#ifndef KQEMU_MAX_RAM_PAGES_TO_UPDATE
	54	#define KQEMU_MAX_RAM_PAGES_TO_UPDATE 512
	55	#define KQEMU_RAM_PAGES_UPDATE_ALL (KQEMU_MAX_RAM_PAGES_TO_UPDATE + 1)
	56	#endif
c28e951f	57
6e4255f6 FB	58	#ifdef _WIN32
	59	#define KQEMU_DEVICE "\\\\.\\kqemu"
	60	#else
9df217a3	61	#define KQEMU_DEVICE "/dev/kqemu"
6e4255f6 FB	62	#endif
	63
	64	#ifdef _WIN32
	65	#define KQEMU_INVALID_FD INVALID_HANDLE_VALUE
	66	HANDLE kqemu_fd = KQEMU_INVALID_FD;
	67	#define kqemu_closefd(x) CloseHandle(x)
	68	#else
	69	#define KQEMU_INVALID_FD -1
	70	int kqemu_fd = KQEMU_INVALID_FD;
	71	#define kqemu_closefd(x) close(x)
	72	#endif
9df217a3 FB	73
9df217a3 FB	74	int kqemu_allowed = 1;
9df217a3 FB	75	unsigned long *pages_to_flush;
9df217a3 FB	76	unsigned int nb_pages_to_flush;
aa062973 FB	77	unsigned long *ram_pages_to_update;
aa062973 FB	78	unsigned int nb_ram_pages_to_update;
9df217a3 FB	79	extern uint32_t **l1_phys_map;
	80
	81	#define cpuid(index, eax, ebx, ecx, edx) \
	82	asm volatile ("cpuid" \
	83	: "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) \
	84	: "0" (index))
	85
c28e951f FB	86	#ifdef __x86_64__
	87	static int is_cpuid_supported(void)
	88	{
	89	return 1;
	90	}
	91	#else
9df217a3 FB	92	static int is_cpuid_supported(void)
	93	{
	94	int v0, v1;
	95	asm volatile ("pushf\n"
	96	"popl %0\n"
	97	"movl %0, %1\n"
	98	"xorl $0x00200000, %0\n"
	99	"pushl %0\n"
	100	"popf\n"
	101	"pushf\n"
	102	"popl %0\n"
	103	: "=a" (v0), "=d" (v1)
	104	:
	105	: "cc");
	106	return (v0 != v1);
	107	}
c28e951f	108	#endif
9df217a3 FB	109
	110	static void kqemu_update_cpuid(CPUState *env)
	111	{
	112	int critical_features_mask, features;
	113	uint32_t eax, ebx, ecx, edx;
	114
	115	/* the following features are kept identical on the host and
	116	target cpus because they are important for user code. Strictly
	117	speaking, only SSE really matters because the OS must support
	118	it if the user code uses it. */
	119	critical_features_mask =
	120	CPUID_CMOV \| CPUID_CX8 \|
	121	CPUID_FXSR \| CPUID_MMX \| CPUID_SSE \|
ca0d1734	122	CPUID_SSE2 \| CPUID_SEP;
9df217a3 FB	123	if (!is_cpuid_supported()) {
	124	features = 0;
	125	} else {
	126	cpuid(1, eax, ebx, ecx, edx);
	127	features = edx;
	128	}
ca0d1734 FB	129	#ifdef __x86_64__
	130	/* NOTE: on x86_64 CPUs, SYSENTER is not supported in
	131	compatibility mode, so in order to have the best performances
	132	it is better not to use it */
	133	features &= ~CPUID_SEP;
	134	#endif
9df217a3 FB	135	env->cpuid_features = (env->cpuid_features & ~critical_features_mask) \|
	136	(features & critical_features_mask);
	137	/* XXX: we could update more of the target CPUID state so that the
	138	non accelerated code sees exactly the same CPU features as the
	139	accelerated code */
	140	}
	141
	142	int kqemu_init(CPUState *env)
	143	{
	144	struct kqemu_init init;
	145	int ret, version;
6e4255f6 FB	146	#ifdef _WIN32
	147	DWORD temp;
	148	#endif
9df217a3 FB	149
	150	if (!kqemu_allowed)
	151	return -1;
	152
6e4255f6 FB	153	#ifdef _WIN32
	154	kqemu_fd = CreateFile(KQEMU_DEVICE, GENERIC_WRITE \| GENERIC_READ,
	155	FILE_SHARE_READ \| FILE_SHARE_WRITE,
	156	NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL,
	157	NULL);
	158	#else
9df217a3	159	kqemu_fd = open(KQEMU_DEVICE, O_RDWR);
6e4255f6 FB	160	#endif
6e4255f6 FB	161	if (kqemu_fd == KQEMU_INVALID_FD) {
9df217a3 FB	162	fprintf(stderr, "Could not open '%s' - QEMU acceleration layer not activated\n", KQEMU_DEVICE);
	163	return -1;
	164	}
	165	version = 0;
6e4255f6 FB	166	#ifdef _WIN32
	167	DeviceIoControl(kqemu_fd, KQEMU_GET_VERSION, NULL, 0,
	168	&version, sizeof(version), &temp, NULL);
	169	#else
9df217a3	170	ioctl(kqemu_fd, KQEMU_GET_VERSION, &version);
6e4255f6	171	#endif
9df217a3 FB	172	if (version != KQEMU_VERSION) {
	173	fprintf(stderr, "Version mismatch between kqemu module and qemu (%08x %08x) - disabling kqemu use\n",
	174	version, KQEMU_VERSION);
	175	goto fail;
	176	}
	177
	178	pages_to_flush = qemu_vmalloc(KQEMU_MAX_PAGES_TO_FLUSH *
	179	sizeof(unsigned long));
	180	if (!pages_to_flush)
	181	goto fail;
	182
aa062973 FB	183	ram_pages_to_update = qemu_vmalloc(KQEMU_MAX_RAM_PAGES_TO_UPDATE *
	184	sizeof(unsigned long));
	185	if (!ram_pages_to_update)
	186	goto fail;
	187
9df217a3 FB	188	init.ram_base = phys_ram_base;
	189	init.ram_size = phys_ram_size;
	190	init.ram_dirty = phys_ram_dirty;
	191	init.phys_to_ram_map = l1_phys_map;
	192	init.pages_to_flush = pages_to_flush;
aa062973 FB	193	#if KQEMU_VERSION >= 0x010200
	194	init.ram_pages_to_update = ram_pages_to_update;
	195	#endif
6e4255f6 FB	196	#ifdef _WIN32
	197	ret = DeviceIoControl(kqemu_fd, KQEMU_INIT, &init, sizeof(init),
	198	NULL, 0, &temp, NULL) == TRUE ? 0 : -1;
	199	#else
9df217a3	200	ret = ioctl(kqemu_fd, KQEMU_INIT, &init);
6e4255f6	201	#endif
9df217a3 FB	202	if (ret < 0) {
	203	fprintf(stderr, "Error %d while initializing QEMU acceleration layer - disabling it for now\n", ret);
	204	fail:
6e4255f6 FB	205	kqemu_closefd(kqemu_fd);
6e4255f6 FB	206	kqemu_fd = KQEMU_INVALID_FD;
9df217a3 FB	207	return -1;
	208	}
	209	kqemu_update_cpuid(env);
	210	env->kqemu_enabled = 1;
	211	nb_pages_to_flush = 0;
aa062973	212	nb_ram_pages_to_update = 0;
9df217a3 FB	213	return 0;
	214	}
	215
	216	void kqemu_flush_page(CPUState *env, target_ulong addr)
	217	{
	218	#ifdef DEBUG
	219	if (loglevel & CPU_LOG_INT) {
	220	fprintf(logfile, "kqemu_flush_page: addr=" TARGET_FMT_lx "\n", addr);
	221	}
	222	#endif
	223	if (nb_pages_to_flush >= KQEMU_MAX_PAGES_TO_FLUSH)
	224	nb_pages_to_flush = KQEMU_FLUSH_ALL;
	225	else
	226	pages_to_flush[nb_pages_to_flush++] = addr;
	227	}
	228
	229	void kqemu_flush(CPUState *env, int global)
	230	{
	231	#ifdef DEBUG
	232	if (loglevel & CPU_LOG_INT) {
	233	fprintf(logfile, "kqemu_flush:\n");
	234	}
	235	#endif
	236	nb_pages_to_flush = KQEMU_FLUSH_ALL;
	237	}
	238
aa062973 FB	239	void kqemu_set_notdirty(CPUState *env, ram_addr_t ram_addr)
	240	{
	241	#ifdef DEBUG
	242	if (loglevel & CPU_LOG_INT) {
	243	fprintf(logfile, "kqemu_set_notdirty: addr=%08lx\n", ram_addr);
	244	}
	245	#endif
fc8dc060 FB	246	/* we only track transitions to dirty state */
	247	if (phys_ram_dirty[ram_addr >> TARGET_PAGE_BITS] != 0xff)
	248	return;
aa062973 FB	249	if (nb_ram_pages_to_update >= KQEMU_MAX_RAM_PAGES_TO_UPDATE)
	250	nb_ram_pages_to_update = KQEMU_RAM_PAGES_UPDATE_ALL;
	251	else
	252	ram_pages_to_update[nb_ram_pages_to_update++] = ram_addr;
	253	}
	254
9df217a3 FB	255	struct fpstate {
	256	uint16_t fpuc;
	257	uint16_t dummy1;
	258	uint16_t fpus;
	259	uint16_t dummy2;
	260	uint16_t fptag;
	261	uint16_t dummy3;
	262
	263	uint32_t fpip;
	264	uint32_t fpcs;
	265	uint32_t fpoo;
	266	uint32_t fpos;
	267	uint8_t fpregs1[8 * 10];
	268	};
	269
	270	struct fpxstate {
	271	uint16_t fpuc;
	272	uint16_t fpus;
	273	uint16_t fptag;
	274	uint16_t fop;
	275	uint32_t fpuip;
	276	uint16_t cs_sel;
	277	uint16_t dummy0;
	278	uint32_t fpudp;
	279	uint16_t ds_sel;
	280	uint16_t dummy1;
	281	uint32_t mxcsr;
	282	uint32_t mxcsr_mask;
	283	uint8_t fpregs1[8 * 16];
c28e951f FB	284	uint8_t xmm_regs[16 * 16];
c28e951f FB	285	uint8_t dummy2[96];
9df217a3 FB	286	};
	287
	288	static struct fpxstate fpx1 __attribute__((aligned(16)));
	289
	290	static void restore_native_fp_frstor(CPUState *env)
	291	{
	292	int fptag, i, j;
	293	struct fpstate fp1, *fp = &fp1;
	294
	295	fp->fpuc = env->fpuc;
	296	fp->fpus = (env->fpus & ~0x3800) \| (env->fpstt & 0x7) << 11;
	297	fptag = 0;
	298	for (i=7; i>=0; i--) {
	299	fptag <<= 2;
	300	if (env->fptags[i]) {
	301	fptag \|= 3;
	302	} else {
	303	/* the FPU automatically computes it */
	304	}
	305	}
	306	fp->fptag = fptag;
	307	j = env->fpstt;
	308	for(i = 0;i < 8; i++) {
	309	memcpy(&fp->fpregs1[i * 10], &env->fpregs[j].d, 10);
	310	j = (j + 1) & 7;
	311	}
	312	asm volatile ("frstor %0" : "=m" (*fp));
	313	}
	314
	315	static void save_native_fp_fsave(CPUState *env)
	316	{
	317	int fptag, i, j;
	318	uint16_t fpuc;
	319	struct fpstate fp1, *fp = &fp1;
	320
	321	asm volatile ("fsave %0" : : "m" (*fp));
	322	env->fpuc = fp->fpuc;
	323	env->fpstt = (fp->fpus >> 11) & 7;
	324	env->fpus = fp->fpus & ~0x3800;
	325	fptag = fp->fptag;
	326	for(i = 0;i < 8; i++) {
	327	env->fptags[i] = ((fptag & 3) == 3);
	328	fptag >>= 2;
	329	}
	330	j = env->fpstt;
	331	for(i = 0;i < 8; i++) {
	332	memcpy(&env->fpregs[j].d, &fp->fpregs1[i * 10], 10);
	333	j = (j + 1) & 7;
	334	}
	335	/* we must restore the default rounding state */
	336	fpuc = 0x037f \| (env->fpuc & (3 << 10));
	337	asm volatile("fldcw %0" : : "m" (fpuc));
	338	}
	339
	340	static void restore_native_fp_fxrstor(CPUState *env)
	341	{
	342	struct fpxstate *fp = &fpx1;
	343	int i, j, fptag;
	344
	345	fp->fpuc = env->fpuc;
	346	fp->fpus = (env->fpus & ~0x3800) \| (env->fpstt & 0x7) << 11;
	347	fptag = 0;
	348	for(i = 0; i < 8; i++)
	349	fptag \|= (env->fptags[i] << i);
350	fp->fptag = fptag ^ 0xff;
351
352	j = env->fpstt;
353	for(i = 0;i < 8; i++) {
354	memcpy(&fp->fpregs1[i * 16], &env->fpregs[j].d, 10);
355	j = (j + 1) & 7;
356	}
357	if (env->cpuid_features & CPUID_SSE) {
358	fp->mxcsr = env->mxcsr;
359	/* XXX: check if DAZ is not available */
360	fp->mxcsr_mask = 0xffff;
c28e951f	361	memcpy(fp->xmm_regs, env->xmm_regs, CPU_NB_REGS * 16);
9df217a3 FB	362	}
	363	asm volatile ("fxrstor %0" : "=m" (*fp));
	364	}
	365
	366	static void save_native_fp_fxsave(CPUState *env)
	367	{
	368	struct fpxstate *fp = &fpx1;
	369	int fptag, i, j;
	370	uint16_t fpuc;
	371
	372	asm volatile ("fxsave %0" : : "m" (*fp));
	373	env->fpuc = fp->fpuc;
	374	env->fpstt = (fp->fpus >> 11) & 7;
	375	env->fpus = fp->fpus & ~0x3800;
	376	fptag = fp->fptag ^ 0xff;
	377	for(i = 0;i < 8; i++) {
	378	env->fptags[i] = (fptag >> i) & 1;
	379	}
	380	j = env->fpstt;
	381	for(i = 0;i < 8; i++) {
	382	memcpy(&env->fpregs[j].d, &fp->fpregs1[i * 16], 10);
	383	j = (j + 1) & 7;
	384	}
	385	if (env->cpuid_features & CPUID_SSE) {
	386	env->mxcsr = fp->mxcsr;
c28e951f	387	memcpy(env->xmm_regs, fp->xmm_regs, CPU_NB_REGS * 16);
9df217a3 FB	388	}
	389
	390	/* we must restore the default rounding state */
	391	asm volatile ("fninit");
	392	fpuc = 0x037f \| (env->fpuc & (3 << 10));
	393	asm volatile("fldcw %0" : : "m" (fpuc));
	394	}
	395
c28e951f FB	396	static int do_syscall(CPUState *env,
	397	struct kqemu_cpu_state *kenv)
	398	{
	399	int selector;
	400
	401	selector = (env->star >> 32) & 0xffff;
	402	#ifdef __x86_64__
	403	if (env->hflags & HF_LMA_MASK) {
	404	env->regs[R_ECX] = kenv->next_eip;
	405	env->regs[11] = env->eflags;
	406
	407	cpu_x86_set_cpl(env, 0);
	408	cpu_x86_load_seg_cache(env, R_CS, selector & 0xfffc,
	409	0, 0xffffffff,
	410	DESC_G_MASK \| DESC_B_MASK \| DESC_P_MASK \|
	411	DESC_S_MASK \|
	412	DESC_CS_MASK \| DESC_R_MASK \| DESC_A_MASK \| DESC_L_MASK);
	413	cpu_x86_load_seg_cache(env, R_SS, (selector + 8) & 0xfffc,
	414	0, 0xffffffff,
	415	DESC_G_MASK \| DESC_B_MASK \| DESC_P_MASK \|
	416	DESC_S_MASK \|
	417	DESC_W_MASK \| DESC_A_MASK);
	418	env->eflags &= ~env->fmask;
	419	if (env->hflags & HF_CS64_MASK)
	420	env->eip = env->lstar;
	421	else
	422	env->eip = env->cstar;
	423	} else
	424	#endif
	425	{
	426	env->regs[R_ECX] = (uint32_t)kenv->next_eip;
	427
	428	cpu_x86_set_cpl(env, 0);
	429	cpu_x86_load_seg_cache(env, R_CS, selector & 0xfffc,
	430	0, 0xffffffff,
	431	DESC_G_MASK \| DESC_B_MASK \| DESC_P_MASK \|
	432	DESC_S_MASK \|
	433	DESC_CS_MASK \| DESC_R_MASK \| DESC_A_MASK);
	434	cpu_x86_load_seg_cache(env, R_SS, (selector + 8) & 0xfffc,
	435	0, 0xffffffff,
	436	DESC_G_MASK \| DESC_B_MASK \| DESC_P_MASK \|
	437	DESC_S_MASK \|
	438	DESC_W_MASK \| DESC_A_MASK);
	439	env->eflags &= ~(IF_MASK \| RF_MASK \| VM_MASK);
	440	env->eip = (uint32_t)env->star;
	441	}
	442	return 2;
	443	}
	444
aa062973 FB	445	#ifdef PROFILE
	446
	447	#define PC_REC_SIZE 1
	448	#define PC_REC_HASH_BITS 16
	449	#define PC_REC_HASH_SIZE (1 << PC_REC_HASH_BITS)
	450
	451	typedef struct PCRecord {
	452	unsigned long pc;
	453	int64_t count;
	454	struct PCRecord *next;
	455	} PCRecord;
	456
	457	PCRecord *pc_rec_hash[PC_REC_HASH_SIZE];
	458	int nb_pc_records;
	459
	460	void kqemu_record_pc(unsigned long pc)
	461	{
	462	unsigned long h;
	463	PCRecord *pr, r;
	464
	465	h = pc / PC_REC_SIZE;
	466	h = h ^ (h >> PC_REC_HASH_BITS);
	467	h &= (PC_REC_HASH_SIZE - 1);
	468	pr = &pc_rec_hash[h];
	469	for(;;) {
	470	r = *pr;
	471	if (r == NULL)
	472	break;
	473	if (r->pc == pc) {
	474	r->count++;
	475	return;
	476	}
	477	pr = &r->next;
	478	}
	479	r = malloc(sizeof(PCRecord));
	480	r->count = 1;
	481	r->pc = pc;
	482	r->next = NULL;
	483	*pr = r;
	484	nb_pc_records++;
	485	}
	486
	487	int pc_rec_cmp(const void p1, const void p2)
	488	{
	489	PCRecord r1 = (PCRecord **)p1;
	490	PCRecord r2 = (PCRecord **)p2;
	491	if (r1->count < r2->count)
	492	return 1;
	493	else if (r1->count == r2->count)
	494	return 0;
	495	else
	496	return -1;
	497	}
	498
	499	void kqemu_record_dump(void)
	500	{
	501	PCRecord *pr, r;
	502	int i, h;
	503	FILE *f;
	504	int64_t total, sum;
	505
	506	pr = malloc(sizeof(PCRecord ) nb_pc_records);
	507	i = 0;
	508	total = 0;
509	for(h = 0; h < PC_REC_HASH_SIZE; h++) {
510	for(r = pc_rec_hash[h]; r != NULL; r = r->next) {
511	pr[i++] = r;
512	total += r->count;
513	}
514	}
515	qsort(pr, nb_pc_records, sizeof(PCRecord *), pc_rec_cmp);
516
517	f = fopen("/tmp/kqemu.stats", "w");
518	if (!f) {
519	perror("/tmp/kqemu.stats");
520	exit(1);
521	}
522	fprintf(f, "total: %lld\n", total);
523	sum = 0;
524	for(i = 0; i < nb_pc_records; i++) {
525	r = pr[i];
526	sum += r->count;
527	fprintf(f, "%08lx: %lld %0.2f%% %0.2f%%\n",
528	r->pc,
529	r->count,
530	(double)r->count / (double)total * 100.0,
531	(double)sum / (double)total * 100.0);
532	}
533	fclose(f);
534	free(pr);
535	}
536	#else
537	void kqemu_record_dump(void)
538	{
539	}
540	#endif
541
9df217a3 FB	542	int kqemu_cpu_exec(CPUState *env)
	543	{
	544	struct kqemu_cpu_state kcpu_state, *kenv = &kcpu_state;
	545	int ret;
6e4255f6 FB	546	#ifdef _WIN32
	547	DWORD temp;
	548	#endif
9df217a3 FB	549
	550	#ifdef DEBUG
	551	if (loglevel & CPU_LOG_INT) {
	552	fprintf(logfile, "kqemu: cpu_exec: enter\n");
	553	cpu_dump_state(env, logfile, fprintf, 0);
	554	}
	555	#endif
	556	memcpy(kenv->regs, env->regs, sizeof(kenv->regs));
	557	kenv->eip = env->eip;
	558	kenv->eflags = env->eflags;
	559	memcpy(&kenv->segs, &env->segs, sizeof(env->segs));
	560	memcpy(&kenv->ldt, &env->ldt, sizeof(env->ldt));
	561	memcpy(&kenv->tr, &env->tr, sizeof(env->tr));
	562	memcpy(&kenv->gdt, &env->gdt, sizeof(env->gdt));
	563	memcpy(&kenv->idt, &env->idt, sizeof(env->idt));
	564	kenv->cr0 = env->cr[0];
	565	kenv->cr2 = env->cr[2];
	566	kenv->cr3 = env->cr[3];
	567	kenv->cr4 = env->cr[4];
	568	kenv->a20_mask = env->a20_mask;
c45b3c0e	569	#if KQEMU_VERSION >= 0x010100
c28e951f FB	570	kenv->efer = env->efer;
c28e951f FB	571	#endif
9df217a3 FB	572	if (env->dr[7] & 0xff) {
	573	kenv->dr7 = env->dr[7];
	574	kenv->dr0 = env->dr[0];
	575	kenv->dr1 = env->dr[1];
	576	kenv->dr2 = env->dr[2];
	577	kenv->dr3 = env->dr[3];
	578	} else {
	579	kenv->dr7 = 0;
	580	}
	581	kenv->dr6 = env->dr[6];
	582	kenv->cpl = 3;
	583	kenv->nb_pages_to_flush = nb_pages_to_flush;
	584	nb_pages_to_flush = 0;
aa062973 FB	585	#if KQEMU_VERSION >= 0x010200
	586	kenv->user_only = 1;
	587	kenv->nb_ram_pages_to_update = nb_ram_pages_to_update;
	588	#endif
	589	nb_ram_pages_to_update = 0;
9df217a3 FB	590
	591	if (!(kenv->cr0 & CR0_TS_MASK)) {
	592	if (env->cpuid_features & CPUID_FXSR)
	593	restore_native_fp_fxrstor(env);
	594	else
	595	restore_native_fp_frstor(env);
	596	}
	597
6e4255f6	598	#ifdef _WIN32
a332e112 FB	599	if (DeviceIoControl(kqemu_fd, KQEMU_EXEC,
	600	kenv, sizeof(struct kqemu_cpu_state),
	601	kenv, sizeof(struct kqemu_cpu_state),
	602	&temp, NULL)) {
	603	ret = kenv->retval;
	604	} else {
	605	ret = -1;
	606	}
6e4255f6 FB	607	#else
	608	#if KQEMU_VERSION >= 0x010100
	609	ioctl(kqemu_fd, KQEMU_EXEC, kenv);
	610	ret = kenv->retval;
	611	#else
9df217a3	612	ret = ioctl(kqemu_fd, KQEMU_EXEC, kenv);
6e4255f6 FB	613	#endif
6e4255f6 FB	614	#endif
9df217a3 FB	615	if (!(kenv->cr0 & CR0_TS_MASK)) {
	616	if (env->cpuid_features & CPUID_FXSR)
	617	save_native_fp_fxsave(env);
	618	else
	619	save_native_fp_fsave(env);
	620	}
	621
	622	memcpy(env->regs, kenv->regs, sizeof(env->regs));
	623	env->eip = kenv->eip;
	624	env->eflags = kenv->eflags;
	625	memcpy(env->segs, kenv->segs, sizeof(env->segs));
	626	#if 0
	627	/* no need to restore that */
	628	memcpy(env->ldt, kenv->ldt, sizeof(env->ldt));
	629	memcpy(env->tr, kenv->tr, sizeof(env->tr));
	630	memcpy(env->gdt, kenv->gdt, sizeof(env->gdt));
	631	memcpy(env->idt, kenv->idt, sizeof(env->idt));
	632	env->cr[0] = kenv->cr0;
	633	env->cr[3] = kenv->cr3;
	634	env->cr[4] = kenv->cr4;
	635	env->a20_mask = kenv->a20_mask;
	636	#endif
	637	env->cr[2] = kenv->cr2;
	638	env->dr[6] = kenv->dr6;
	639
aa062973 FB	640	#if KQEMU_VERSION >= 0x010200
	641	if (kenv->nb_ram_pages_to_update > 0) {
	642	cpu_tlb_update_dirty(env);
	643	}
	644	#endif
	645
	646	/* restore the hidden flags */
	647	{
	648	unsigned int new_hflags;
	649	#ifdef TARGET_X86_64
	650	if ((env->hflags & HF_LMA_MASK) &&
	651	(env->segs[R_CS].flags & DESC_L_MASK)) {
	652	/* long mode */
	653	new_hflags = HF_CS32_MASK \| HF_SS32_MASK \| HF_CS64_MASK;
	654	} else
	655	#endif
	656	{
	657	/* legacy / compatibility case */
	658	new_hflags = (env->segs[R_CS].flags & DESC_B_MASK)
	659	>> (DESC_B_SHIFT - HF_CS32_SHIFT);
	660	new_hflags \|= (env->segs[R_SS].flags & DESC_B_MASK)
	661	>> (DESC_B_SHIFT - HF_SS32_SHIFT);
	662	if (!(env->cr[0] & CR0_PE_MASK) \|\|
	663	(env->eflags & VM_MASK) \|\|
	664	!(env->hflags & HF_CS32_MASK)) {
	665	/* XXX: try to avoid this test. The problem comes from the
	666	fact that is real mode or vm86 mode we only modify the
	667	'base' and 'selector' fields of the segment cache to go
	668	faster. A solution may be to force addseg to one in
	669	translate-i386.c. */
	670	new_hflags \|= HF_ADDSEG_MASK;
	671	} else {
	672	new_hflags \|= ((env->segs[R_DS].base \|
	673	env->segs[R_ES].base \|
	674	env->segs[R_SS].base) != 0) <<
	675	HF_ADDSEG_SHIFT;
	676	}
	677	}
	678	env->hflags = (env->hflags &
	679	~(HF_CS32_MASK \| HF_SS32_MASK \| HF_CS64_MASK \| HF_ADDSEG_MASK)) \|
	680	new_hflags;
	681	}
	682
9df217a3 FB	683	#ifdef DEBUG
	684	if (loglevel & CPU_LOG_INT) {
	685	fprintf(logfile, "kqemu: kqemu_cpu_exec: ret=0x%x\n", ret);
	686	}
	687	#endif
c28e951f FB	688	if (ret == KQEMU_RET_SYSCALL) {
	689	/* syscall instruction */
	690	return do_syscall(env, kenv);
	691	} else
9df217a3 FB	692	if ((ret & 0xff00) == KQEMU_RET_INT) {
	693	env->exception_index = ret & 0xff;
	694	env->error_code = 0;
	695	env->exception_is_int = 1;
	696	env->exception_next_eip = kenv->next_eip;
	697	#ifdef DEBUG
c28e951f FB	698	if (loglevel & CPU_LOG_INT) {
	699	fprintf(logfile, "kqemu: interrupt v=%02x:\n",
	700	env->exception_index);
	701	cpu_dump_state(env, logfile, fprintf, 0);
	702	}
9df217a3 FB	703	#endif
	704	return 1;
	705	} else if ((ret & 0xff00) == KQEMU_RET_EXCEPTION) {
	706	env->exception_index = ret & 0xff;
	707	env->error_code = kenv->error_code;
	708	env->exception_is_int = 0;
	709	env->exception_next_eip = 0;
	710	#ifdef DEBUG
	711	if (loglevel & CPU_LOG_INT) {
	712	fprintf(logfile, "kqemu: exception v=%02x e=%04x:\n",
	713	env->exception_index, env->error_code);
	714	cpu_dump_state(env, logfile, fprintf, 0);
	715	}
	716	#endif
	717	return 1;
	718	} else if (ret == KQEMU_RET_INTR) {
c45b3c0e FB	719	#ifdef DEBUG
	720	if (loglevel & CPU_LOG_INT) {
	721	cpu_dump_state(env, logfile, fprintf, 0);
	722	}
	723	#endif
9df217a3 FB	724	return 0;
9df217a3 FB	725	} else if (ret == KQEMU_RET_SOFTMMU) {
aa062973 FB	726	#ifdef PROFILE
	727	kqemu_record_pc(env->eip + env->segs[R_CS].base);
	728	#endif
	729	#ifdef DEBUG
	730	if (loglevel & CPU_LOG_INT) {
	731	cpu_dump_state(env, logfile, fprintf, 0);
	732	}
	733	#endif
9df217a3 FB	734	return 2;
	735	} else {
	736	cpu_dump_state(env, stderr, fprintf, 0);
	737	fprintf(stderr, "Unsupported return value: 0x%x\n", ret);
	738	exit(1);
	739	}
	740	return 0;
	741	}
	742
a332e112 FB	743	void kqemu_cpu_interrupt(CPUState *env)
	744	{
	745	#if defined(_WIN32) && KQEMU_VERSION >= 0x010101
	746	/* cancelling the I/O request causes KQEMU to finish executing the
	747	current block and successfully returning. */
	748	CancelIo(kqemu_fd);
	749	#endif
	750	}
	751
9df217a3	752	#endif