This patch adds little-endian mode support to PPC emulation.

This is needed by OS/2 and Windows NT and some programs like VirtualPC.
This patch has been tested using OS/2 bootloader (thanks to Tero
Kaarlela).
(Jocelyn Mayer)


git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@1379 c046a42c-6fe2-441c-8c8c-71466251a162

3 files changed, 327 insertions, 26 deletions

diff --git a/target-ppc/op_helper_mem.h b/target-ppc/op_helper_mem.h
index 85ac91163c..fa7f076762 100644
--- a/target-ppc/op_helper_mem.h
+++ b/target-ppc/op_helper_mem.h
@@ -40,4 +40,53 @@ void glue(do_stsw, MEMSUFFIX) (int src)
     }
 }
 
+void glue(do_lsw_le, MEMSUFFIX) (int dst)
+{
+    uint32_t tmp;
+    int sh;
+
+    if (loglevel > 0) {
+        fprintf(logfile, "%s: addr=0x%08x count=%d reg=%d\n",
+                __func__, T0, T1, dst);
+    }
+    for (; T1 > 3; T1 -= 4, T0 += 4) {
+        tmp = glue(ldl, MEMSUFFIX)(T0);
+        ugpr(dst++) = ((tmp & 0xFF000000) >> 24) | ((tmp & 0x00FF0000) >> 8) |
+            ((tmp & 0x0000FF00) << 8) | ((tmp & 0x000000FF) << 24);
+        if (dst == 32)
+            dst = 0;
+    }
+    if (T1 > 0) {
+        tmp = 0;
+        for (sh = 0; T1 > 0; T1--, T0++, sh += 8) {
+            tmp |= glue(ldub, MEMSUFFIX)(T0) << sh;
+        }
+        ugpr(dst) = tmp;
+    }
+}
+
+void glue(do_stsw_le, MEMSUFFIX) (int src)
+{
+    uint32_t tmp;
+    int sh;
+
+    if (loglevel > 0) {
+        fprintf(logfile, "%s: addr=0x%08x count=%d reg=%d\n",
+                __func__, T0, T1, src);
+    }
+    for (; T1 > 3; T1 -= 4, T0 += 4) {
+        tmp = ((ugpr(src++) & 0xFF000000) >> 24);
+        tmp |= ((ugpr(src++) & 0x00FF0000) >> 8);
+        tmp |= ((ugpr(src++) & 0x0000FF00) << 8);
+        tmp |= ((ugpr(src++) & 0x000000FF) << 24);
+        glue(stl, MEMSUFFIX)(T0, tmp);
+        if (src == 32)
+            src = 0;
+    }
+    if (T1 > 0) {
+        for (sh = 0; T1 > 0; T1--, T0++, sh += 8)
+            glue(stb, MEMSUFFIX)(T0, (ugpr(src) >> sh) & 0xFF);
+    }
+}
+
 #undef MEMSUFFIX
diff --git a/target-ppc/op_mem.h b/target-ppc/op_mem.h
index f0f0cd1b34..9b3f721e5f 100644
--- a/target-ppc/op_mem.h
+++ b/target-ppc/op_mem.h
@@ -8,6 +8,12 @@ static inline uint16_t glue(ld16r, MEMSUFFIX) (target_ulong EA)
     return ((tmp & 0xFF00) >> 8) | ((tmp & 0x00FF) << 8);
 }
 
+static inline int32_t glue(ld16rs, MEMSUFFIX) (target_ulong EA)
+{
+    int16_t tmp = glue(lduw, MEMSUFFIX)(EA);
+    return ((tmp & 0xFF00) >> 8) | ((tmp & 0x00FF) << 8);
+}
+
 static inline uint32_t glue(ld32r, MEMSUFFIX) (target_ulong EA)
 {
     uint32_t tmp = glue(ldl, MEMSUFFIX)(EA);
@@ -48,17 +54,29 @@ PPC_LD_OP(ha, ldsw);
 PPC_LD_OP(hz, lduw);
 PPC_LD_OP(wz, ldl);
 
+PPC_LD_OP(ha_le, ld16rs);
+PPC_LD_OP(hz_le, ld16r);
+PPC_LD_OP(wz_le, ld32r);
+
 /***                              Integer store                            ***/
 PPC_ST_OP(b, stb);
 PPC_ST_OP(h, stw);
 PPC_ST_OP(w, stl);
 
+PPC_ST_OP(h_le, st16r);
+PPC_ST_OP(w_le, st32r);
+
 /***                Integer load and store with byte reverse               ***/
 PPC_LD_OP(hbr, ld16r);
 PPC_LD_OP(wbr, ld32r);
 PPC_ST_OP(hbr, st16r);
 PPC_ST_OP(wbr, st32r);
 
+PPC_LD_OP(hbr_le, lduw);
+PPC_LD_OP(wbr_le, ldl);
+PPC_ST_OP(hbr_le, stw);
+PPC_ST_OP(wbr_le, stl);
+
 /***                    Integer load and store multiple                    ***/
 PPC_OP(glue(lmw, MEMSUFFIX))
 {
@@ -80,6 +98,26 @@ PPC_OP(glue(stmw, MEMSUFFIX))
     RETURN();
 }
 
+PPC_OP(glue(lmw_le, MEMSUFFIX))
+{
+    int dst = PARAM(1);
+
+    for (; dst < 32; dst++, T0 += 4) {
+        ugpr(dst) = glue(ld32r, MEMSUFFIX)(T0);
+    }
+    RETURN();
+}
+
+PPC_OP(glue(stmw_le, MEMSUFFIX))
+{
+    int src = PARAM(1);
+
+    for (; src < 32; src++, T0 += 4) {
+        glue(st32r, MEMSUFFIX)(T0, ugpr(src));
+    }
+    RETURN();
+}
+
 /***                    Integer load and store strings                     ***/
 PPC_OP(glue(lswi, MEMSUFFIX))
 {
@@ -87,6 +125,13 @@ PPC_OP(glue(lswi, MEMSUFFIX))
     RETURN();
 }
 
+void glue(do_lsw_le, MEMSUFFIX) (int dst);
+PPC_OP(glue(lswi_le, MEMSUFFIX))
+{
+    glue(do_lsw_le, MEMSUFFIX)(PARAM(1));
+    RETURN();
+}
+
 /* PPC32 specification says we must generate an exception if
  * rA is in the range of registers to be loaded.
  * In an other hand, IBM says this is valid, but rA won't be loaded.
@@ -105,12 +150,32 @@ PPC_OP(glue(lswx, MEMSUFFIX))
     RETURN();
 }
 
+PPC_OP(glue(lswx_le, MEMSUFFIX))
+{
+    if (T1 > 0) {
+        if ((PARAM(1) < PARAM(2) && (PARAM(1) + T1) > PARAM(2)) ||
+            (PARAM(1) < PARAM(3) && (PARAM(1) + T1) > PARAM(3))) {
+            do_raise_exception_err(EXCP_PROGRAM, EXCP_INVAL | EXCP_INVAL_LSWX);
+        } else {
+            glue(do_lsw_le, MEMSUFFIX)(PARAM(1));
+        }
+    }
+    RETURN();
+}
+
 PPC_OP(glue(stsw, MEMSUFFIX))
 {
     glue(do_stsw, MEMSUFFIX)(PARAM(1));
     RETURN();
 }
 
+void glue(do_stsw_le, MEMSUFFIX) (int src);
+PPC_OP(glue(stsw_le, MEMSUFFIX))
+{
+    glue(do_stsw_le, MEMSUFFIX)(PARAM(1));
+    RETURN();
+}
+
 /***                         Floating-point store                          ***/
 #define PPC_STF_OP(name, op)                                                  \
 PPC_OP(glue(glue(st, name), MEMSUFFIX))                                       \
@@ -122,6 +187,43 @@ PPC_OP(glue(glue(st, name), MEMSUFFIX))                                       \
 PPC_STF_OP(fd, stfq);
 PPC_STF_OP(fs, stfl);
 
+static inline void glue(stfqr, MEMSUFFIX) (target_ulong EA, double d)
+{
+    union {
+        double d;
+        uint64_t u;
+    } u;
+
+    u.d = d;
+    u.u = ((u.u & 0xFF00000000000000ULL) >> 56) |
+        ((u.u & 0x00FF000000000000ULL) >> 40) |
+        ((u.u & 0x0000FF0000000000ULL) >> 24) |
+        ((u.u & 0x000000FF00000000ULL) >> 8) |
+        ((u.u & 0x00000000FF000000ULL) << 8) |
+        ((u.u & 0x0000000000FF0000ULL) << 24) |
+        ((u.u & 0x000000000000FF00ULL) << 40) |
+        ((u.u & 0x00000000000000FFULL) << 56);
+    glue(stfq, MEMSUFFIX)(EA, u.d);
+}
+
+static inline void glue(stflr, MEMSUFFIX) (target_ulong EA, float f)
+{
+    union {
+        float f;
+        uint32_t u;
+    } u;
+
+    u.f = f;
+    u.u = ((u.u & 0xFF000000UL) >> 24) |
+        ((u.u & 0x00FF0000ULL) >> 8) |
+        ((u.u & 0x0000FF00UL) << 8) |
+        ((u.u & 0x000000FFULL) << 24);
+    glue(stfl, MEMSUFFIX)(EA, u.f);
+}
+
+PPC_STF_OP(fd_le, stfqr);
+PPC_STF_OP(fs_le, stflr);
+
 /***                         Floating-point load                           ***/
 #define PPC_LDF_OP(name, op)                                                  \
 PPC_OP(glue(glue(l, name), MEMSUFFIX))                                        \
@@ -133,6 +235,45 @@ PPC_OP(glue(glue(l, name), MEMSUFFIX))                                        \
 PPC_LDF_OP(fd, ldfq);
 PPC_LDF_OP(fs, ldfl);
 
+static inline double glue(ldfqr, MEMSUFFIX) (target_ulong EA)
+{
+    union {
+        double d;
+        uint64_t u;
+    } u;
+
+    u.d = glue(ldfq, MEMSUFFIX)(EA);
+    u.u = ((u.u & 0xFF00000000000000ULL) >> 56) |
+        ((u.u & 0x00FF000000000000ULL) >> 40) |
+        ((u.u & 0x0000FF0000000000ULL) >> 24) |
+        ((u.u & 0x000000FF00000000ULL) >> 8) |
+        ((u.u & 0x00000000FF000000ULL) << 8) |
+        ((u.u & 0x0000000000FF0000ULL) << 24) |
+        ((u.u & 0x000000000000FF00ULL) << 40) |
+        ((u.u & 0x00000000000000FFULL) << 56);
+
+    return u.d;
+}
+
+static inline float glue(ldflr, MEMSUFFIX) (target_ulong EA)
+{
+    union {
+        float f;
+        uint32_t u;
+    } u;
+
+    u.f = glue(ldfl, MEMSUFFIX)(EA);
+    u.u = ((u.u & 0xFF000000UL) >> 24) |
+        ((u.u & 0x00FF0000ULL) >> 8) |
+        ((u.u & 0x0000FF00UL) << 8) |
+        ((u.u & 0x000000FFULL) << 24);
+
+    return u.f;
+}
+
+PPC_LDF_OP(fd_le, ldfqr);
+PPC_LDF_OP(fs_le, ldflr);
+
 /* Load and set reservation */
 PPC_OP(glue(lwarx, MEMSUFFIX))
 {
@@ -145,6 +286,17 @@ PPC_OP(glue(lwarx, MEMSUFFIX))
     RETURN();
 }
 
+PPC_OP(glue(lwarx_le, MEMSUFFIX))
+{
+    if (T0 & 0x03) {
+        do_raise_exception(EXCP_ALIGN);
+    } else {
+       T1 = glue(ld32r, MEMSUFFIX)(T0);
+       regs->reserve = T0;
+    }
+    RETURN();
+}
+
 /* Store with reservation */
 PPC_OP(glue(stwcx, MEMSUFFIX))
 {
@@ -162,6 +314,22 @@ PPC_OP(glue(stwcx, MEMSUFFIX))
     RETURN();
 }
 
+PPC_OP(glue(stwcx_le, MEMSUFFIX))
+{
+    if (T0 & 0x03) {
+        do_raise_exception(EXCP_ALIGN);
+    } else {
+        if (regs->reserve != T0) {
+            env->crf[0] = xer_ov;
+        } else {
+            glue(st32r, MEMSUFFIX)(T0, T1);
+            env->crf[0] = xer_ov | 0x02;
+        }
+    }
+    regs->reserve = 0;
+    RETURN();
+}
+
 PPC_OP(glue(dcbz, MEMSUFFIX))
 {
     glue(stl, MEMSUFFIX)(T0 + 0x00, 0);
@@ -188,4 +356,16 @@ PPC_OP(glue(ecowx, MEMSUFFIX))
     RETURN();
 }
 
+PPC_OP(glue(eciwx_le, MEMSUFFIX))
+{
+    T1 = glue(ld32r, MEMSUFFIX)(T0);
+    RETURN();
+}
+
+PPC_OP(glue(ecowx_le, MEMSUFFIX))
+{
+    glue(st32r, MEMSUFFIX)(T0, T1);
+    RETURN();
+}
+
 #undef MEMSUFFIX
diff --git a/target-ppc/translate.c b/target-ppc/translate.c
index 6cfbcc2b2d..7d2e62b958 100644
--- a/target-ppc/translate.c
+++ b/target-ppc/translate.c
@@ -1046,22 +1046,41 @@ GEN_HANDLER(mtfsfi, 0x3F, 0x06, 0x04, 0x006f0800, PPC_FLOAT)
 }
 
 /***                             Integer load                              ***/
+#define op_ldst(name)        (*gen_op_##name[ctx->mem_idx])()
 #if defined(CONFIG_USER_ONLY)
-#define op_ldst(name)        gen_op_##name##_raw()
-#define OP_LD_TABLE(width)
-#define OP_ST_TABLE(width)
+#define OP_LD_TABLE(width)                                                    \
+static GenOpFunc *gen_op_l##width[] = {                                       \
+    &gen_op_l##width##_raw,                                                   \
+    &gen_op_l##width##_le_raw,                                                \
+};
+#define OP_ST_TABLE(width)                                                    \
+static GenOpFunc *gen_op_st##width[] = {                                      \
+    &gen_op_st##width##_raw,                                                  \
+    &gen_op_st##width##_le_raw,                                               \
+};
+/* Byte access routine are endian safe */
+#define gen_op_stb_le_raw gen_op_stb_raw
+#define gen_op_lbz_le_raw gen_op_lbz_raw
 #else
-#define op_ldst(name)        (*gen_op_##name[ctx->mem_idx])()
 #define OP_LD_TABLE(width)                                                    \
 static GenOpFunc *gen_op_l##width[] = {                                       \
     &gen_op_l##width##_user,                                                  \
+    &gen_op_l##width##_le_user,                                               \
     &gen_op_l##width##_kernel,                                                \
-}
+    &gen_op_l##width##_le_kernel,                                             \
+};
 #define OP_ST_TABLE(width)                                                    \
 static GenOpFunc *gen_op_st##width[] = {                                      \
     &gen_op_st##width##_user,                                                 \
+    &gen_op_st##width##_le_user,                                              \
     &gen_op_st##width##_kernel,                                               \
-}
+    &gen_op_st##width##_le_kernel,                                            \
+};
+/* Byte access routine are endian safe */
+#define gen_op_stb_le_user gen_op_stb_user
+#define gen_op_lbz_le_user gen_op_lbz_user
+#define gen_op_stb_le_kernel gen_op_stb_kernel
+#define gen_op_lbz_le_kernel gen_op_lbz_kernel
 #endif
 
 #define GEN_LD(width, opc)                                                    \
@@ -1232,17 +1251,28 @@ OP_ST_TABLE(wbr);
 GEN_STX(wbr, 0x16, 0x14);
 
 /***                    Integer load and store multiple                    ***/
+#define op_ldstm(name, reg) (*gen_op_##name[ctx->mem_idx])(reg)
 #if defined(CONFIG_USER_ONLY)
-#define op_ldstm(name, reg) gen_op_##name##_raw(reg)
+static GenOpFunc1 *gen_op_lmw[] = {
+    &gen_op_lmw_raw,
+    &gen_op_lmw_le_raw,
+};
+static GenOpFunc1 *gen_op_stmw[] = {
+    &gen_op_stmw_raw,
+    &gen_op_stmw_le_raw,
+};
 #else
-#define op_ldstm(name, reg) (*gen_op_##name[ctx->mem_idx])(reg)
 static GenOpFunc1 *gen_op_lmw[] = {
     &gen_op_lmw_user,
+    &gen_op_lmw_le_user,
     &gen_op_lmw_kernel,
+    &gen_op_lmw_le_kernel,
 };
 static GenOpFunc1 *gen_op_stmw[] = {
     &gen_op_stmw_user,
+    &gen_op_stmw_le_user,
     &gen_op_stmw_kernel,
+    &gen_op_stmw_le_kernel,
 };
 #endif
 
@@ -1277,23 +1307,39 @@ GEN_HANDLER(stmw, 0x2F, 0xFF, 0xFF, 0x00000000, PPC_INTEGER)
 }
 
 /***                    Integer load and store strings                     ***/
-#if defined(CONFIG_USER_ONLY)
-#define op_ldsts(name, start) gen_op_##name##_raw(start)
-#define op_ldstsx(name, rd, ra, rb) gen_op_##name##_raw(rd, ra, rb)
-#else
 #define op_ldsts(name, start) (*gen_op_##name[ctx->mem_idx])(start)
 #define op_ldstsx(name, rd, ra, rb) (*gen_op_##name[ctx->mem_idx])(rd, ra, rb)
+#if defined(CONFIG_USER_ONLY)
+static GenOpFunc1 *gen_op_lswi[] = {
+    &gen_op_lswi_raw,
+    &gen_op_lswi_le_raw,
+};
+static GenOpFunc3 *gen_op_lswx[] = {
+    &gen_op_lswx_raw,
+    &gen_op_lswx_le_raw,
+};
+static GenOpFunc1 *gen_op_stsw[] = {
+    &gen_op_stsw_raw,
+    &gen_op_stsw_le_raw,
+};
+#else
 static GenOpFunc1 *gen_op_lswi[] = {
     &gen_op_lswi_user,
+    &gen_op_lswi_le_user,
     &gen_op_lswi_kernel,
+    &gen_op_lswi_le_kernel,
 };
 static GenOpFunc3 *gen_op_lswx[] = {
     &gen_op_lswx_user,
+    &gen_op_lswx_le_user,
     &gen_op_lswx_kernel,
+    &gen_op_lswx_le_kernel,
 };
 static GenOpFunc1 *gen_op_stsw[] = {
     &gen_op_stsw_user,
+    &gen_op_stsw_le_user,
     &gen_op_stsw_kernel,
+    &gen_op_stsw_le_kernel,
 };
 #endif
 
@@ -1389,23 +1435,33 @@ GEN_HANDLER(isync, 0x13, 0x16, 0xFF, 0x03FF0801, PPC_MEM)
 {
 }
 
-/* lwarx */
+#define op_lwarx() (*gen_op_lwarx[ctx->mem_idx])()
+#define op_stwcx() (*gen_op_stwcx[ctx->mem_idx])()
 #if defined(CONFIG_USER_ONLY)
-#define op_lwarx() gen_op_lwarx_raw()
-#define op_stwcx() gen_op_stwcx_raw()
+static GenOpFunc *gen_op_lwarx[] = {
+    &gen_op_lwarx_raw,
+    &gen_op_lwarx_le_raw,
+};
+static GenOpFunc *gen_op_stwcx[] = {
+    &gen_op_stwcx_raw,
+    &gen_op_stwcx_le_raw,
+};
 #else
-#define op_lwarx() (*gen_op_lwarx[ctx->mem_idx])()
 static GenOpFunc *gen_op_lwarx[] = {
     &gen_op_lwarx_user,
+    &gen_op_lwarx_le_user,
     &gen_op_lwarx_kernel,
+    &gen_op_lwarx_le_kernel,
 };
-#define op_stwcx() (*gen_op_stwcx[ctx->mem_idx])()
 static GenOpFunc *gen_op_stwcx[] = {
     &gen_op_stwcx_user,
+    &gen_op_stwcx_le_user,
     &gen_op_stwcx_kernel,
+    &gen_op_stwcx_le_kernel,
 };
 #endif
 
+/* lwarx */
 GEN_HANDLER(lwarx, 0x1F, 0x14, 0xFF, 0x00000001, PPC_RES)
 {
     if (rA(ctx->opcode) == 0) {
@@ -2498,23 +2554,33 @@ GEN_HANDLER(tlbsync, 0x1F, 0x16, 0x11, 0x03FFF801, PPC_MEM)
 
 /***                              External control                         ***/
 /* Optional: */
-/* eciwx */
-#if defined(CONFIG_USER_ONLY)
-#define op_eciwx() gen_op_eciwx_raw()
-#define op_ecowx() gen_op_ecowx_raw()
-#else
 #define op_eciwx() (*gen_op_eciwx[ctx->mem_idx])()
 #define op_ecowx() (*gen_op_ecowx[ctx->mem_idx])()
+#if defined(CONFIG_USER_ONLY)
+static GenOpFunc *gen_op_eciwx[] = {
+    &gen_op_eciwx_raw,
+    &gen_op_eciwx_le_raw,
+};
+static GenOpFunc *gen_op_ecowx[] = {
+    &gen_op_ecowx_raw,
+    &gen_op_ecowx_le_raw,
+};
+#else
 static GenOpFunc *gen_op_eciwx[] = {
     &gen_op_eciwx_user,
+    &gen_op_eciwx_le_user,
     &gen_op_eciwx_kernel,
+    &gen_op_eciwx_le_kernel,
 };
 static GenOpFunc *gen_op_ecowx[] = {
     &gen_op_ecowx_user,
+    &gen_op_ecowx_le_user,
     &gen_op_ecowx_kernel,
+    &gen_op_ecowx_le_kernel,
 };
 #endif
 
+/* eciwx */
 GEN_HANDLER(eciwx, 0x1F, 0x16, 0x0D, 0x00000001, PPC_EXTERN)
 {
     /* Should check EAR[E] & alignment ! */
@@ -3143,10 +3209,10 @@ int gen_intermediate_code_internal (CPUState *env, TranslationBlock *tb,
     ctx.tb = tb;
     ctx.exception = EXCP_NONE;
 #if defined(CONFIG_USER_ONLY)
-    ctx.mem_idx = 0;
+    ctx.mem_idx = msr_le;
 #else
     ctx.supervisor = 1 - msr_pr;
-    ctx.mem_idx = 1 - msr_pr;
+    ctx.mem_idx = ((1 - msr_pr) << 1) | msr_le;
 #endif
     ctx.fpu_enabled = msr_fp;
 #if defined (DO_SINGLE_STEP)
@@ -3173,11 +3239,17 @@ int gen_intermediate_code_internal (CPUState *env, TranslationBlock *tb,
         }
 #endif
         ctx.opcode = ldl_code(ctx.nip);
+        if (msr_le) {
+            ctx.opcode = ((ctx.opcode & 0xFF000000) >> 24) |
+                ((ctx.opcode & 0x00FF0000) >> 8) |
+                ((ctx.opcode & 0x0000FF00) << 8) |
+                ((ctx.opcode & 0x000000FF) << 24);
+        }
 #if defined PPC_DEBUG_DISAS
         if (loglevel & CPU_LOG_TB_IN_ASM) {
-            fprintf(logfile, "translate opcode %08x (%02x %02x %02x)\n",
+            fprintf(logfile, "translate opcode %08x (%02x %02x %02x) (%s)\n",
                     ctx.opcode, opc1(ctx.opcode), opc2(ctx.opcode),
-                    opc3(ctx.opcode));
+                    opc3(ctx.opcode), msr_le ? "little" : "big");
         }
 #endif
         ctx.nip += 4;