target-sparc: Implement ldqf and stqf inline

At the same time, fix a problem with stqf_asi, when
a write might access two pages.

Tested-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Signed-off-by: Richard Henderson <rth@twiddle.net>

3 files changed, 67 insertions, 111 deletions

diff --git a/target-sparc/helper.h b/target-sparc/helper.h
index b7bff84efa..0cf1bfb73a 100644
--- a/target-sparc/helper.h
+++ b/target-sparc/helper.h
@@ -38,8 +38,6 @@ DEF_HELPER_3(tsubcctv, tl, env, tl, tl)
 DEF_HELPER_FLAGS_3(sdivx, TCG_CALL_NO_WG, s64, env, s64, s64)
 DEF_HELPER_FLAGS_3(udivx, TCG_CALL_NO_WG, i64, env, i64, i64)
 #endif
-DEF_HELPER_FLAGS_3(ldqf, TCG_CALL_NO_WG, void, env, tl, int)
-DEF_HELPER_FLAGS_3(stqf, TCG_CALL_NO_WG, void, env, tl, int)
 #if !defined(CONFIG_USER_ONLY) || defined(TARGET_SPARC64)
 DEF_HELPER_FLAGS_4(ld_asi, TCG_CALL_NO_WG, i64, env, tl, int, i32)
 DEF_HELPER_FLAGS_5(st_asi, TCG_CALL_NO_WG, void, env, tl, i64, int, i32)
diff --git a/target-sparc/ldst_helper.c b/target-sparc/ldst_helper.c
index 725bb49e9d..de7d53ae20 100644
--- a/target-sparc/ldst_helper.c
+++ b/target-sparc/ldst_helper.c
@@ -254,18 +254,6 @@ static void replace_tlb_1bit_lru(SparcTLBEntry *tlb,
 
 #endif
 
-#if defined(TARGET_SPARC64) || defined(CONFIG_USER_ONLY)
-static inline target_ulong address_mask(CPUSPARCState *env1, target_ulong addr)
-{
-#ifdef TARGET_SPARC64
-    if (AM_CHECK(env1)) {
-        addr &= 0xffffffffULL;
-    }
-#endif
-    return addr;
-}
-#endif
-
 #ifdef TARGET_SPARC64
 /* returns true if access using this ASI is to have address translated by MMU
    otherwise access is to raw physical address */
@@ -290,14 +278,21 @@ static inline int is_translating_asi(int asi)
     }
 }
 
+static inline target_ulong address_mask(CPUSPARCState *env1, target_ulong addr)
+{
+    if (AM_CHECK(env1)) {
+        addr &= 0xffffffffULL;
+    }
+    return addr;
+}
+
 static inline target_ulong asi_address_mask(CPUSPARCState *env,
                                             int asi, target_ulong addr)
 {
     if (is_translating_asi(asi)) {
-        return address_mask(env, addr);
-    } else {
-        return addr;
+        addr = address_mask(env, addr);
     }
+    return addr;
 }
 #endif
 
@@ -1601,78 +1596,6 @@ void helper_st_asi(CPUSPARCState *env, target_ulong addr, target_ulong val,
 #endif /* CONFIG_USER_ONLY */
 #endif /* TARGET_SPARC64 */
 
-void helper_ldqf(CPUSPARCState *env, target_ulong addr, int mem_idx)
-{
-    /* XXX add 128 bit load */
-    CPU_QuadU u;
-
-    do_check_align(env, addr, 7, GETPC());
-#if !defined(CONFIG_USER_ONLY)
-    switch (mem_idx) {
-    case MMU_USER_IDX:
-        u.ll.upper = cpu_ldq_user(env, addr);
-        u.ll.lower = cpu_ldq_user(env, addr + 8);
-        QT0 = u.q;
-        break;
-    case MMU_KERNEL_IDX:
-        u.ll.upper = cpu_ldq_kernel(env, addr);
-        u.ll.lower = cpu_ldq_kernel(env, addr + 8);
-        QT0 = u.q;
-        break;
-#ifdef TARGET_SPARC64
-    case MMU_HYPV_IDX:
-        u.ll.upper = cpu_ldq_hypv(env, addr);
-        u.ll.lower = cpu_ldq_hypv(env, addr + 8);
-        QT0 = u.q;
-        break;
-#endif
-    default:
-        DPRINTF_MMU("helper_ldqf: need to check MMU idx %d\n", mem_idx);
-        break;
-    }
-#else
-    u.ll.upper = cpu_ldq_data(env, address_mask(env, addr));
-    u.ll.lower = cpu_ldq_data(env, address_mask(env, addr + 8));
-    QT0 = u.q;
-#endif
-}
-
-void helper_stqf(CPUSPARCState *env, target_ulong addr, int mem_idx)
-{
-    /* XXX add 128 bit store */
-    CPU_QuadU u;
-
-    do_check_align(env, addr, 7, GETPC());
-#if !defined(CONFIG_USER_ONLY)
-    switch (mem_idx) {
-    case MMU_USER_IDX:
-        u.q = QT0;
-        cpu_stq_user(env, addr, u.ll.upper);
-        cpu_stq_user(env, addr + 8, u.ll.lower);
-        break;
-    case MMU_KERNEL_IDX:
-        u.q = QT0;
-        cpu_stq_kernel(env, addr, u.ll.upper);
-        cpu_stq_kernel(env, addr + 8, u.ll.lower);
-        break;
-#ifdef TARGET_SPARC64
-    case MMU_HYPV_IDX:
-        u.q = QT0;
-        cpu_stq_hypv(env, addr, u.ll.upper);
-        cpu_stq_hypv(env, addr + 8, u.ll.lower);
-        break;
-#endif
-    default:
-        DPRINTF_MMU("helper_stqf: need to check MMU idx %d\n", mem_idx);
-        break;
-    }
-#else
-    u.q = QT0;
-    cpu_stq_data(env, address_mask(env, addr), u.ll.upper);
-    cpu_stq_data(env, address_mask(env, addr + 8), u.ll.lower);
-#endif
-}
-
 #if !defined(CONFIG_USER_ONLY)
 #ifndef TARGET_SPARC64
 void sparc_cpu_unassigned_access(CPUState *cs, hwaddr addr,
diff --git a/target-sparc/translate.c b/target-sparc/translate.c
index ecdc7e1d8a..a869d7d6dc 100644
--- a/target-sparc/translate.c
+++ b/target-sparc/translate.c
@@ -242,7 +242,29 @@ static void gen_op_store_QT0_fpr(unsigned int dst)
                    offsetof(CPU_QuadU, ll.lower));
 }
 
+static void gen_store_fpr_Q(DisasContext *dc, unsigned int dst,
+                            TCGv_i64 v1, TCGv_i64 v2)
+{
+    dst = QFPREG(dst);
+
+    tcg_gen_mov_i64(cpu_fpr[dst / 2], v1);
+    tcg_gen_mov_i64(cpu_fpr[dst / 2 + 1], v2);
+    gen_update_fprs_dirty(dc, dst);
+}
+
 #ifdef TARGET_SPARC64
+static TCGv_i64 gen_load_fpr_Q0(DisasContext *dc, unsigned int src)
+{
+    src = QFPREG(src);
+    return cpu_fpr[src / 2];
+}
+
+static TCGv_i64 gen_load_fpr_Q1(DisasContext *dc, unsigned int src)
+{
+    src = QFPREG(src);
+    return cpu_fpr[src / 2 + 1];
+}
+
 static void gen_move_Q(DisasContext *dc, unsigned int rd, unsigned int rs)
 {
     rd = QFPREG(rd);
@@ -2557,10 +2579,17 @@ static void gen_stf_asi(DisasContext *dc, TCGv addr,
             tcg_gen_qemu_st_i32(d32, addr, da.mem_idx, da.memop);
             break;
         case 8:
+            /* ??? Only 4-byte alignment required.  However, it is legal
+               for the cpu to signal the alignment fault, and the OS trap
+               handler is required to fix it up.  */
             tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da.mem_idx, da.memop);
             break;
         case 16:
-            tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da.mem_idx, da.memop);
+            /* Only 4-byte alignment required.  See above.  Requiring
+               16-byte alignment here avoids having to probe the second
+               page before performing the first write.  */
+            tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da.mem_idx,
+                                da.memop | MO_ALIGN_16);
             tcg_gen_addi_tl(addr, addr, 8);
             tcg_gen_qemu_st_i64(cpu_fpr[rd/2+1], addr, da.mem_idx, da.memop);
             break;
@@ -5426,17 +5455,16 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn)
                     gen_helper_ldfsr(cpu_fsr, cpu_env, cpu_fsr, cpu_dst_32);
                     break;
                 case 0x22:      /* ldqf, load quad fpreg */
-                    {
-                        TCGv_i32 r_const;
-
-                        CHECK_FPU_FEATURE(dc, FLOAT128);
-                        r_const = tcg_const_i32(dc->mem_idx);
-                        gen_address_mask(dc, cpu_addr);
-                        gen_helper_ldqf(cpu_env, cpu_addr, r_const);
-                        tcg_temp_free_i32(r_const);
-                        gen_op_store_QT0_fpr(QFPREG(rd));
-                        gen_update_fprs_dirty(dc, QFPREG(rd));
-                    }
+                    CHECK_FPU_FEATURE(dc, FLOAT128);
+                    gen_address_mask(dc, cpu_addr);
+                    cpu_src1_64 = tcg_temp_new_i64();
+                    tcg_gen_qemu_ld64(cpu_src1_64, cpu_addr, dc->mem_idx);
+                    tcg_gen_addi_tl(cpu_addr, cpu_addr, 8);
+                    cpu_src2_64 = tcg_temp_new_i64();
+                    tcg_gen_qemu_ld64(cpu_src2_64, cpu_addr, dc->mem_idx);
+                    gen_store_fpr_Q(dc, rd, cpu_src1_64, cpu_src2_64);
+                    tcg_temp_free_i64(cpu_src1_64);
+                    tcg_temp_free_i64(cpu_src2_64);
                     break;
                 case 0x23:      /* lddf, load double fpreg */
                     gen_address_mask(dc, cpu_addr);
@@ -5537,16 +5565,20 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn)
                 case 0x26:
 #ifdef TARGET_SPARC64
                     /* V9 stqf, store quad fpreg */
-                    {
-                        TCGv_i32 r_const;
-
-                        CHECK_FPU_FEATURE(dc, FLOAT128);
-                        gen_op_load_fpr_QT0(QFPREG(rd));
-                        r_const = tcg_const_i32(dc->mem_idx);
-                        gen_address_mask(dc, cpu_addr);
-                        gen_helper_stqf(cpu_env, cpu_addr, r_const);
-                        tcg_temp_free_i32(r_const);
-                    }
+                    CHECK_FPU_FEATURE(dc, FLOAT128);
+                    gen_address_mask(dc, cpu_addr);
+                    /* ??? While stqf only requires 4-byte alignment, it is
+                       legal for the cpu to signal the unaligned exception.
+                       The OS trap handler is then required to fix it up.
+                       For qemu, this avoids having to probe the second page
+                       before performing the first write.  */
+                    cpu_src1_64 = gen_load_fpr_Q0(dc, rd);
+                    tcg_gen_qemu_st_i64(cpu_src1_64, cpu_addr,
+                                        dc->mem_idx, MO_TEQ | MO_ALIGN_16);
+                    tcg_gen_addi_tl(cpu_addr, cpu_addr, 8);
+                    cpu_src2_64 = gen_load_fpr_Q1(dc, rd);
+                    tcg_gen_qemu_st_i64(cpu_src1_64, cpu_addr,
+                                        dc->mem_idx, MO_TEQ);
                     break;
 #else /* !TARGET_SPARC64 */
                     /* stdfq, store floating point queue */
@@ -5562,6 +5594,9 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn)
 #endif
 #endif
                 case 0x27: /* stdf, store double fpreg */
+                    /* ??? Only 4-byte alignment required.  However, it is
+                       legal for the cpu to signal the alignment fault, and
+                       the OS trap handler is required to fix it up.  */
                     gen_address_mask(dc, cpu_addr);
                     cpu_src1_64 = gen_load_fpr_D(dc, rd);
                     tcg_gen_qemu_st64(cpu_src1_64, cpu_addr, dc->mem_idx);