[Qemu-devel] [4431] Remove someexplicit alignment checks ( initial patch by Fabrice Bellard)

[Blue Swirl]

Revision: 4431
          http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=4431
Author:   blueswir1
Date:     2008-05-11 19:24:10 +0000 (Sun, 11 May 2008)

Log Message:
-----------
Remove someexplicit alignment checks (initial patch by Fabrice Bellard)

Modified Paths:
--------------
    trunk/target-sparc/op_helper.c
    trunk/target-sparc/translate.c

Modified: trunk/target-sparc/op_helper.c
===================================================================
--- trunk/target-sparc/op_helper.c      2008-05-11 15:07:19 UTC (rev 4430)
+++ trunk/target-sparc/op_helper.c      2008-05-11 19:24:10 UTC (rev 4431)
@@ -33,6 +33,12 @@
 #define DPRINTF_ASI(fmt, args...) do {} while (0)
 #endif
 
+#ifdef TARGET_ABI32
+#define ABI32_MASK(addr) do { (addr) &= 0xffffffffULL; } while (0)
+#else
+#define ABI32_MASK(addr) do {} while (0)
+#endif
+
 void raise_exception(int tt)
 {
     env->exception_index = tt;
@@ -55,8 +61,13 @@
 
 void helper_check_align(target_ulong addr, uint32_t align)
 {
-    if (addr & align)
+    if (addr & align) {
+#ifdef DEBUG_UNALIGNED
+    printf("Unaligned access to 0x" TARGET_FMT_lx " from 0x" TARGET_FMT_lx
+           "\n", addr, env->pc);
+#endif
         raise_exception(TT_UNALIGNED);
+    }
 }
 
 #define F_HELPER(name, p) void helper_f##name##p(void)
@@ -832,6 +843,7 @@
     uint32_t last_addr = addr;
 #endif
 
+    helper_check_align(addr, size - 1);
     switch (asi) {
     case 2: /* SuperSparc MXCC registers */
         switch (addr) {
@@ -1030,6 +1042,7 @@
 
 void helper_st_asi(target_ulong addr, uint64_t val, int asi, int size)
 {
+    helper_check_align(addr, size - 1);
     switch(asi) {
     case 2: /* SuperSparc MXCC registers */
         switch (addr) {
@@ -1335,6 +1348,9 @@
     if (asi < 0x80)
         raise_exception(TT_PRIV_ACT);
 
+    helper_check_align(addr, size - 1);
+    ABI32_MASK(addr);
+
     switch (asi) {
     case 0x80: // Primary
     case 0x82: // Primary no-fault
@@ -1421,6 +1437,9 @@
     if (asi < 0x80)
         raise_exception(TT_PRIV_ACT);
 
+    helper_check_align(addr, size - 1);
+    ABI32_MASK(addr);
+
     /* Convert to little endian */
     switch (asi) {
     case 0x88: // Primary LE
@@ -1491,6 +1510,7 @@
         || (asi >= 0x30 && asi < 0x80 && !(env->hpstate & HS_PRIV)))
         raise_exception(TT_PRIV_ACT);
 
+    helper_check_align(addr, size - 1);
     switch (asi) {
     case 0x10: // As if user primary
     case 0x18: // As if user primary LE
@@ -1714,6 +1734,7 @@
         || (asi >= 0x30 && asi < 0x80 && !(env->hpstate & HS_PRIV)))
         raise_exception(TT_PRIV_ACT);
 
+    helper_check_align(addr, size - 1);
     /* Convert to little endian */
     switch (asi) {
     case 0x0c: // Nucleus Little Endian (LE)
@@ -2009,6 +2030,7 @@
     unsigned int i;
     target_ulong val;
 
+    helper_check_align(addr, 3);
     switch (asi) {
     case 0xf0: // Block load primary
     case 0xf1: // Block load secondary
@@ -2018,10 +2040,7 @@
             raise_exception(TT_ILL_INSN);
             return;
         }
-        if (addr & 0x3f) {
-            raise_exception(TT_UNALIGNED);
-            return;
-        }
+        helper_check_align(addr, 0x3f);
         for (i = 0; i < 16; i++) {
             *(uint32_t *)&env->fpr[rd++] = helper_ld_asi(addr, asi & 0x8f, 4, 
0);
             addr += 4;
@@ -2052,6 +2071,7 @@
     unsigned int i;
     target_ulong val = 0;
 
+    helper_check_align(addr, 3);
     switch (asi) {
     case 0xf0: // Block store primary
     case 0xf1: // Block store secondary
@@ -2061,10 +2081,7 @@
             raise_exception(TT_ILL_INSN);
             return;
         }
-        if (addr & 0x3f) {
-            raise_exception(TT_UNALIGNED);
-            return;
-        }
+        helper_check_align(addr, 0x3f);
         for (i = 0; i < 16; i++) {
             val = *(uint32_t *)&env->fpr[rd++];
             helper_st_asi(addr, val, asi & 0x8f, 4);
@@ -2183,55 +2200,53 @@
     return ((uint64_t)high << 32) | (uint64_t)(low & 0xffffffff);
 }
 
-#ifdef TARGET_ABI32
-#define ADDR(x) ((x) & 0xffffffff)
-#else
-#define ADDR(x) (x)
-#endif
-
 void helper_stdf(target_ulong addr, int mem_idx)
 {
+    helper_check_align(addr, 7);
 #if !defined(CONFIG_USER_ONLY)
     switch (mem_idx) {
     case 0:
-        stfq_user(ADDR(addr), DT0);
+        stfq_user(addr, DT0);
         break;
     case 1:
-        stfq_kernel(ADDR(addr), DT0);
+        stfq_kernel(addr, DT0);
         break;
 #ifdef TARGET_SPARC64
     case 2:
-        stfq_hypv(ADDR(addr), DT0);
+        stfq_hypv(addr, DT0);
         break;
 #endif
     default:
         break;
     }
 #else
-    stfq_raw(ADDR(addr), DT0);
+    ABI32_MASK(addr);
+    stfq_raw(addr, DT0);
 #endif
 }
 
 void helper_lddf(target_ulong addr, int mem_idx)
 {
+    helper_check_align(addr, 7);
 #if !defined(CONFIG_USER_ONLY)
     switch (mem_idx) {
     case 0:
-        DT0 = ldfq_user(ADDR(addr));
+        DT0 = ldfq_user(addr);
         break;
     case 1:
-        DT0 = ldfq_kernel(ADDR(addr));
+        DT0 = ldfq_kernel(addr);
         break;
 #ifdef TARGET_SPARC64
     case 2:
-        DT0 = ldfq_hypv(ADDR(addr));
+        DT0 = ldfq_hypv(addr);
         break;
 #endif
     default:
         break;
     }
 #else
-    DT0 = ldfq_raw(ADDR(addr));
+    ABI32_MASK(addr);
+    DT0 = ldfq_raw(addr);
 #endif
 }
 
@@ -2240,22 +2255,23 @@
     // XXX add 128 bit load
     CPU_QuadU u;
 
+    helper_check_align(addr, 7);
 #if !defined(CONFIG_USER_ONLY)
     switch (mem_idx) {
     case 0:
-        u.ll.upper = ldq_user(ADDR(addr));
-        u.ll.lower = ldq_user(ADDR(addr + 8));
+        u.ll.upper = ldq_user(addr);
+        u.ll.lower = ldq_user(addr + 8);
         QT0 = u.q;
         break;
     case 1:
-        u.ll.upper = ldq_kernel(ADDR(addr));
-        u.ll.lower = ldq_kernel(ADDR(addr + 8));
+        u.ll.upper = ldq_kernel(addr);
+        u.ll.lower = ldq_kernel(addr + 8);
         QT0 = u.q;
         break;
 #ifdef TARGET_SPARC64
     case 2:
-        u.ll.upper = ldq_hypv(ADDR(addr));
-        u.ll.lower = ldq_hypv(ADDR(addr + 8));
+        u.ll.upper = ldq_hypv(addr);
+        u.ll.lower = ldq_hypv(addr + 8);
         QT0 = u.q;
         break;
 #endif
@@ -2263,8 +2279,9 @@
         break;
     }
 #else
-    u.ll.upper = ldq_raw(ADDR(addr));
-    u.ll.lower = ldq_raw(ADDR(addr + 8));
+    ABI32_MASK(addr);
+    u.ll.upper = ldq_raw(addr);
+    u.ll.lower = ldq_raw((addr + 8) & 0xffffffffULL);
     QT0 = u.q;
 #endif
 }
@@ -2274,23 +2291,24 @@
     // XXX add 128 bit store
     CPU_QuadU u;
 
+    helper_check_align(addr, 7);
 #if !defined(CONFIG_USER_ONLY)
     switch (mem_idx) {
     case 0:
         u.q = QT0;
-        stq_user(ADDR(addr), u.ll.upper);
-        stq_user(ADDR(addr + 8), u.ll.lower);
+        stq_user(addr, u.ll.upper);
+        stq_user(addr + 8, u.ll.lower);
         break;
     case 1:
         u.q = QT0;
-        stq_kernel(ADDR(addr), u.ll.upper);
-        stq_kernel(ADDR(addr + 8), u.ll.lower);
+        stq_kernel(addr, u.ll.upper);
+        stq_kernel(addr + 8, u.ll.lower);
         break;
 #ifdef TARGET_SPARC64
     case 2:
         u.q = QT0;
-        stq_hypv(ADDR(addr), u.ll.upper);
-        stq_hypv(ADDR(addr + 8), u.ll.lower);
+        stq_hypv(addr, u.ll.upper);
+        stq_hypv(addr + 8, u.ll.lower);
         break;
 #endif
     default:
@@ -2298,13 +2316,12 @@
     }
 #else
     u.q = QT0;
-    stq_raw(ADDR(addr), u.ll.upper);
-    stq_raw(ADDR(addr + 8), u.ll.lower);
+    ABI32_MASK(addr);
+    stq_raw(addr, u.ll.upper);
+    stq_raw((addr + 8) & 0xffffffffULL, u.ll.lower);
 #endif
 }
 
-#undef ADDR
-
 void helper_ldfsr(void)
 {
     int rnd_mode;
@@ -2833,12 +2850,32 @@
 #define SHIFT 3
 #include "softmmu_template.h"
 
+/* XXX: make it generic ? */
+static void cpu_restore_state2(void *retaddr)
+{
+    TranslationBlock *tb;
+    unsigned long pc;
+
+    if (retaddr) {
+        /* now we have a real cpu fault */
+        pc = (unsigned long)retaddr;
+        tb = tb_find_pc(pc);
+        if (tb) {
+            /* the PC is inside the translated code. It means that we have
+               a virtual CPU fault */
+            cpu_restore_state(tb, env, pc, (void *)(long)env->cond);
+        }
+    }
+}
+
 static void do_unaligned_access(target_ulong addr, int is_write, int is_user,
                                 void *retaddr)
 {
 #ifdef DEBUG_UNALIGNED
-    printf("Unaligned access to 0x%x from 0x%x\n", addr, env->pc);
+    printf("Unaligned access to 0x" TARGET_FMT_lx " from 0x" TARGET_FMT_lx
+           "\n", addr, env->pc);
 #endif
+    cpu_restore_state2(retaddr);
     raise_exception(TT_UNALIGNED);
 }
 
@@ -2848,9 +2885,7 @@
 /* XXX: fix it to restore all registers */
 void tlb_fill(target_ulong addr, int is_write, int mmu_idx, void *retaddr)
 {
-    TranslationBlock *tb;
     int ret;
-    unsigned long pc;
     CPUState *saved_env;
 
     /* XXX: hack to restore env in all cases, even if not called from
@@ -2860,16 +2895,7 @@
 
     ret = cpu_sparc_handle_mmu_fault(env, addr, is_write, mmu_idx, 1);
     if (ret) {
-        if (retaddr) {
-            /* now we have a real cpu fault */
-            pc = (unsigned long)retaddr;
-            tb = tb_find_pc(pc);
-            if (tb) {
-                /* the PC is inside the translated code. It means that we have
-                   a virtual CPU fault */
-                cpu_restore_state(tb, env, pc, (void *)env->cond);
-            }
-        }
+        cpu_restore_state2(retaddr);
         cpu_loop_exit();
     }
     env = saved_env;

Modified: trunk/target-sparc/translate.c
===================================================================
--- trunk/target-sparc/translate.c      2008-05-11 15:07:19 UTC (rev 4430)
+++ trunk/target-sparc/translate.c      2008-05-11 19:24:10 UTC (rev 4431)
@@ -3945,7 +3945,6 @@
         {
             unsigned int xop = GET_FIELD(insn, 7, 12);
 
-            save_state(dc, cpu_cond);
             cpu_src1 = get_src1(insn, cpu_src1);
             if (xop == 0x3c || xop == 0x3e)
             {
@@ -3968,7 +3967,6 @@
                 (xop > 0x2c && xop <= 0x33) || xop == 0x1f || xop == 0x3d) {
                 switch (xop) {
                 case 0x0:       /* load unsigned word */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
                     ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_ld32u(cpu_val, cpu_addr, dc->mem_idx);
                     break;
@@ -3977,7 +3975,6 @@
                     tcg_gen_qemu_ld8u(cpu_val, cpu_addr, dc->mem_idx);
                     break;
                 case 0x2:       /* load unsigned halfword */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(1));
                     ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_ld16u(cpu_val, cpu_addr, dc->mem_idx);
                     break;
@@ -3985,7 +3982,9 @@
                     if (rd & 1)
                         goto illegal_insn;
                     else {
-                        tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(7));
+                        save_state(dc, cpu_cond);
+                        tcg_gen_helper_0_2(helper_check_align, cpu_addr,
+                                           tcg_const_i32(7)); // XXX remove
                         ABI32_MASK(cpu_addr);
                         tcg_gen_qemu_ld64(cpu_tmp64, cpu_addr, dc->mem_idx);
                         tcg_gen_trunc_i64_tl(cpu_tmp0, cpu_tmp64);
@@ -4001,7 +4000,6 @@
                     tcg_gen_qemu_ld8s(cpu_val, cpu_addr, dc->mem_idx);
                     break;
                 case 0xa:       /* load signed halfword */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(1));
                     ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_ld16s(cpu_val, cpu_addr, dc->mem_idx);
                     break;
@@ -4012,7 +4010,6 @@
                     break;
                 case 0x0f:      /* swap register with memory. Also atomically 
*/
                     CHECK_IU_FEATURE(dc, SWAP);
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
                     gen_movl_reg_TN(rd, cpu_val);
                     ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_ld32u(cpu_tmp32, cpu_addr, dc->mem_idx);
@@ -4027,7 +4024,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    save_state(dc, cpu_cond);
                     gen_ld_asi(cpu_val, cpu_addr, insn, 4, 0);
                     break;
                 case 0x11:      /* load unsigned byte alternate */
@@ -4037,6 +4034,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
+                    save_state(dc, cpu_cond);
                     gen_ld_asi(cpu_val, cpu_addr, insn, 1, 0);
                     break;
                 case 0x12:      /* load unsigned halfword alternate */
@@ -4046,7 +4044,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(1));
+                    save_state(dc, cpu_cond);
                     gen_ld_asi(cpu_val, cpu_addr, insn, 2, 0);
                     break;
                 case 0x13:      /* load double word alternate */
@@ -4058,7 +4056,7 @@
 #endif
                     if (rd & 1)
                         goto illegal_insn;
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(7));
+                    save_state(dc, cpu_cond);
                     gen_ldda_asi(cpu_tmp0, cpu_val, cpu_addr, insn);
                     gen_movl_TN_reg(rd + 1, cpu_tmp0);
                     break;
@@ -4069,6 +4067,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
+                    save_state(dc, cpu_cond);
                     gen_ld_asi(cpu_val, cpu_addr, insn, 1, 1);
                     break;
                 case 0x1a:      /* load signed halfword alternate */
@@ -4078,7 +4077,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(1));
+                    save_state(dc, cpu_cond);
                     gen_ld_asi(cpu_val, cpu_addr, insn, 2, 1);
                     break;
                 case 0x1d:      /* ldstuba -- XXX: should be atomically */
@@ -4088,6 +4087,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
+                    save_state(dc, cpu_cond);
                     gen_ldstub_asi(cpu_val, cpu_addr, insn);
                     break;
                 case 0x1f:      /* swap reg with alt. memory. Also atomically 
*/
@@ -4098,7 +4098,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    save_state(dc, cpu_cond);
                     gen_movl_reg_TN(rd, cpu_val);
                     gen_swap_asi(cpu_val, cpu_addr, insn);
                     break;
@@ -4112,38 +4112,36 @@
 #endif
 #ifdef TARGET_SPARC64
                 case 0x08: /* V9 ldsw */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
                     ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_ld32s(cpu_val, cpu_addr, dc->mem_idx);
                     break;
                 case 0x0b: /* V9 ldx */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(7));
                     ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_ld64(cpu_val, cpu_addr, dc->mem_idx);
                     break;
                 case 0x18: /* V9 ldswa */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    save_state(dc, cpu_cond);
                     gen_ld_asi(cpu_val, cpu_addr, insn, 4, 1);
                     break;
                 case 0x1b: /* V9 ldxa */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(7));
+                    save_state(dc, cpu_cond);
                     gen_ld_asi(cpu_val, cpu_addr, insn, 8, 0);
                     break;
                 case 0x2d: /* V9 prefetch, no effect */
                     goto skip_move;
                 case 0x30: /* V9 ldfa */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    save_state(dc, cpu_cond);
                     gen_ldf_asi(cpu_addr, insn, 4, rd);
                     goto skip_move;
                 case 0x33: /* V9 lddfa */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    save_state(dc, cpu_cond);
                     gen_ldf_asi(cpu_addr, insn, 8, DFPREG(rd));
                     goto skip_move;
                 case 0x3d: /* V9 prefetcha, no effect */
                     goto skip_move;
                 case 0x32: /* V9 ldqfa */
                     CHECK_FPU_FEATURE(dc, FLOAT128);
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    save_state(dc, cpu_cond);
                     gen_ldf_asi(cpu_addr, insn, 16, QFPREG(rd));
                     goto skip_move;
 #endif
@@ -4157,15 +4155,16 @@
             } else if (xop >= 0x20 && xop < 0x24) {
                 if (gen_trap_ifnofpu(dc, cpu_cond))
                     goto jmp_insn;
+                save_state(dc, cpu_cond);
                 switch (xop) {
                 case 0x20:      /* load fpreg */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_ld32u(cpu_tmp32, cpu_addr, dc->mem_idx);
                     tcg_gen_st_i32(cpu_tmp32, cpu_env,
                                    offsetof(CPUState, fpr[rd]));
                     break;
                 case 0x21:      /* load fsr */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_ld32u(cpu_tmp32, cpu_addr, dc->mem_idx);
                     tcg_gen_st_i32(cpu_tmp32, cpu_env,
                                    offsetof(CPUState, ft0));
@@ -4173,14 +4172,10 @@
                     break;
                 case 0x22:      /* load quad fpreg */
                     CHECK_FPU_FEATURE(dc, FLOAT128);
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr,
-                                       tcg_const_i32(7));
                     tcg_gen_helper_0_2(helper_ldqf, cpu_addr, dc->mem_idx);
                     gen_op_store_QT0_fpr(QFPREG(rd));
                     break;
                 case 0x23:      /* load double fpreg */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr,
-                                       tcg_const_i32(7));
                     tcg_gen_helper_0_2(helper_lddf, cpu_addr,
                                        tcg_const_i32(dc->mem_idx));
                     gen_op_store_DT0_fpr(DFPREG(rd));
@@ -4193,7 +4188,6 @@
                 gen_movl_reg_TN(rd, cpu_val);
                 switch (xop) {
                 case 0x4: /* store word */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
                     ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_st32(cpu_val, cpu_addr, dc->mem_idx);
                     break;
@@ -4202,7 +4196,6 @@
                     tcg_gen_qemu_st8(cpu_val, cpu_addr, dc->mem_idx);
                     break;
                 case 0x6: /* store halfword */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(1));
                     ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_st16(cpu_val, cpu_addr, dc->mem_idx);
                     break;
@@ -4212,7 +4205,10 @@
                     else {
                         TCGv r_low;
 
-                        tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(7));
+                        save_state(dc, cpu_cond);
+                        ABI32_MASK(cpu_addr);
+                        tcg_gen_helper_0_2(helper_check_align, cpu_addr,
+                                           tcg_const_i32(7)); // XXX remove
                         r_low = tcg_temp_new(TCG_TYPE_I32);
                         gen_movl_reg_TN(rd + 1, r_low);
                         tcg_gen_helper_1_2(helper_pack64, cpu_tmp64, cpu_val,
@@ -4228,7 +4224,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    save_state(dc, cpu_cond);
                     gen_st_asi(cpu_val, cpu_addr, insn, 4);
                     break;
                 case 0x15: /* store byte alternate */
@@ -4238,6 +4234,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
+                    save_state(dc, cpu_cond);
                     gen_st_asi(cpu_val, cpu_addr, insn, 1);
                     break;
                 case 0x16: /* store halfword alternate */
@@ -4247,7 +4244,7 @@
                     if (!supervisor(dc))
                         goto priv_insn;
 #endif
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(1));
+                    save_state(dc, cpu_cond);
                     gen_st_asi(cpu_val, cpu_addr, insn, 2);
                     break;
                 case 0x17: /* store double word alternate */
@@ -4260,19 +4257,18 @@
                     if (rd & 1)
                         goto illegal_insn;
                     else {
-                        tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(7));
+                        save_state(dc, cpu_cond);
                         gen_stda_asi(cpu_val, cpu_addr, insn, rd);
                     }
                     break;
 #endif
 #ifdef TARGET_SPARC64
                 case 0x0e: /* V9 stx */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(7));
                     ABI32_MASK(cpu_addr);
                     tcg_gen_qemu_st64(cpu_val, cpu_addr, dc->mem_idx);
                     break;
                 case 0x1e: /* V9 stxa */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(7));
+                    save_state(dc, cpu_cond);
                     gen_st_asi(cpu_val, cpu_addr, insn, 8);
                     break;
 #endif
@@ -4282,17 +4278,16 @@
             } else if (xop > 0x23 && xop < 0x28) {
                 if (gen_trap_ifnofpu(dc, cpu_cond))
                     goto jmp_insn;
+                save_state(dc, cpu_cond);
                 switch (xop) {
                 case 0x24: /* store fpreg */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
+                    ABI32_MASK(cpu_addr);
                     tcg_gen_ld_i32(cpu_tmp32, cpu_env,
                                    offsetof(CPUState, fpr[rd]));
                     tcg_gen_qemu_st32(cpu_tmp32, cpu_addr, dc->mem_idx);
                     break;
                 case 0x25: /* stfsr, V9 stxfsr */
-#ifdef CONFIG_USER_ONLY
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
-#endif
+                    ABI32_MASK(cpu_addr);
                     tcg_gen_helper_0_0(helper_stfsr);
                     tcg_gen_ld_i32(cpu_tmp32, cpu_env,
                                    offsetof(CPUState, ft0));
@@ -4302,8 +4297,6 @@
 #ifdef TARGET_SPARC64
                     /* V9 stqf, store quad fpreg */
                     CHECK_FPU_FEATURE(dc, FLOAT128);
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr,
-                                       tcg_const_i32(7));
                     gen_op_load_fpr_QT0(QFPREG(rd));
                     tcg_gen_helper_0_2(helper_stqf, cpu_addr, dc->mem_idx);
                     break;
@@ -4320,8 +4313,6 @@
 #endif
 #endif
                 case 0x27: /* store double fpreg */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr,
-                                       tcg_const_i32(7));
                     gen_op_load_fpr_DT0(DFPREG(rd));
                     tcg_gen_helper_0_2(helper_stdf, cpu_addr,
                                        tcg_const_i32(dc->mem_idx));
@@ -4330,10 +4321,10 @@
                     goto illegal_insn;
                 }
             } else if (xop > 0x33 && xop < 0x3f) {
+                save_state(dc, cpu_cond);
                 switch (xop) {
 #ifdef TARGET_SPARC64
                 case 0x34: /* V9 stfa */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
                     gen_op_load_fpr_FT0(rd);
                     gen_stf_asi(cpu_addr, insn, 4, rd);
                     break;
@@ -4345,17 +4336,14 @@
                     gen_stf_asi(cpu_addr, insn, 16, QFPREG(rd));
                     break;
                 case 0x37: /* V9 stdfa */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
                     gen_op_load_fpr_DT0(DFPREG(rd));
                     gen_stf_asi(cpu_addr, insn, 8, DFPREG(rd));
                     break;
                 case 0x3c: /* V9 casa */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(3));
                     gen_cas_asi(cpu_val, cpu_addr, cpu_val, insn, rd);
                     gen_movl_TN_reg(rd, cpu_val);
                     break;
                 case 0x3e: /* V9 casxa */
-                    tcg_gen_helper_0_2(helper_check_align, cpu_addr, 
tcg_const_i32(7));
                     gen_casx_asi(cpu_val, cpu_addr, cpu_val, insn, rd);
                     gen_movl_TN_reg(rd, cpu_val);
                     break;