Merge remote-tracking branch 'remotes/rth/tags/pull-axp-20211013' into staging

Cleanup alpha memory ops prior to prctl PR_SET_UNALIGN

# gpg: Signature made Wed 13 Oct 2021 10:34:10 AM PDT
# gpg:                using RSA key 7A481E78868B4DB6A85A05C064DF38E8AF7E215F
# gpg:                issuer "richard.henderson@linaro.org"
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>" [ultimate]

* remotes/rth/tags/pull-axp-20211013:
  target/alpha: Reorg integer memory operations
  target/alpha: Reorg fp memory operations

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

1 files changed, 90 insertions, 83 deletions

diff --git a/target/alpha/translate.c b/target/alpha/translate.c
index b034206688..0eee3a1bcc 100644
--- a/target/alpha/translate.c
+++ b/target/alpha/translate.c
@@ -267,51 +267,51 @@ static inline DisasJumpType gen_invalid(DisasContext *ctx)
     return gen_excp(ctx, EXCP_OPCDEC, 0);
 }
 
-static inline void gen_qemu_ldf(TCGv t0, TCGv t1, int flags)
+static void gen_ldf(DisasContext *ctx, TCGv dest, TCGv addr)
 {
     TCGv_i32 tmp32 = tcg_temp_new_i32();
-    tcg_gen_qemu_ld_i32(tmp32, t1, flags, MO_LEUL);
-    gen_helper_memory_to_f(t0, tmp32);
+    tcg_gen_qemu_ld_i32(tmp32, addr, ctx->mem_idx, MO_LEUL);
+    gen_helper_memory_to_f(dest, tmp32);
     tcg_temp_free_i32(tmp32);
 }
 
-static inline void gen_qemu_ldg(TCGv t0, TCGv t1, int flags)
+static void gen_ldg(DisasContext *ctx, TCGv dest, TCGv addr)
 {
     TCGv tmp = tcg_temp_new();
-    tcg_gen_qemu_ld_i64(tmp, t1, flags, MO_LEQ);
-    gen_helper_memory_to_g(t0, tmp);
+    tcg_gen_qemu_ld_i64(tmp, addr, ctx->mem_idx, MO_LEQ);
+    gen_helper_memory_to_g(dest, tmp);
     tcg_temp_free(tmp);
 }
 
-static inline void gen_qemu_lds(TCGv t0, TCGv t1, int flags)
+static void gen_lds(DisasContext *ctx, TCGv dest, TCGv addr)
 {
     TCGv_i32 tmp32 = tcg_temp_new_i32();
-    tcg_gen_qemu_ld_i32(tmp32, t1, flags, MO_LEUL);
-    gen_helper_memory_to_s(t0, tmp32);
+    tcg_gen_qemu_ld_i32(tmp32, addr, ctx->mem_idx, MO_LEUL);
+    gen_helper_memory_to_s(dest, tmp32);
     tcg_temp_free_i32(tmp32);
 }
 
-static inline void gen_qemu_ldl_l(TCGv t0, TCGv t1, int flags)
+static void gen_ldt(DisasContext *ctx, TCGv dest, TCGv addr)
 {
-    tcg_gen_qemu_ld_i64(t0, t1, flags, MO_LESL);
-    tcg_gen_mov_i64(cpu_lock_addr, t1);
-    tcg_gen_mov_i64(cpu_lock_value, t0);
+    tcg_gen_qemu_ld_i64(dest, addr, ctx->mem_idx, MO_LEQ);
 }
 
-static inline void gen_qemu_ldq_l(TCGv t0, TCGv t1, int flags)
+static void gen_load_fp(DisasContext *ctx, int ra, int rb, int32_t disp16,
+                        void (*func)(DisasContext *, TCGv, TCGv))
 {
-    tcg_gen_qemu_ld_i64(t0, t1, flags, MO_LEQ);
-    tcg_gen_mov_i64(cpu_lock_addr, t1);
-    tcg_gen_mov_i64(cpu_lock_value, t0);
+    /* Loads to $f31 are prefetches, which we can treat as nops. */
+    if (likely(ra != 31)) {
+        TCGv addr = tcg_temp_new();
+        tcg_gen_addi_i64(addr, load_gpr(ctx, rb), disp16);
+        func(ctx, cpu_fir[ra], addr);
+        tcg_temp_free(addr);
+    }
 }
 
-static inline void gen_load_mem(DisasContext *ctx,
-                                void (*tcg_gen_qemu_load)(TCGv t0, TCGv t1,
-                                                          int flags),
-                                int ra, int rb, int32_t disp16, bool fp,
-                                bool clear)
+static void gen_load_int(DisasContext *ctx, int ra, int rb, int32_t disp16,
+                         MemOp op, bool clear, bool locked)
 {
-    TCGv tmp, addr, va;
+    TCGv addr, dest;
 
     /* LDQ_U with ra $31 is UNOP.  Other various loads are forms of
        prefetches, which we can treat as nops.  No worries about
@@ -320,72 +320,75 @@ static inline void gen_load_mem(DisasContext *ctx,
         return;
     }
 
-    tmp = tcg_temp_new();
-    addr = load_gpr(ctx, rb);
-
-    if (disp16) {
-        tcg_gen_addi_i64(tmp, addr, disp16);
-        addr = tmp;
-    }
+    addr = tcg_temp_new();
+    tcg_gen_addi_i64(addr, load_gpr(ctx, rb), disp16);
     if (clear) {
-        tcg_gen_andi_i64(tmp, addr, ~0x7);
-        addr = tmp;
+        tcg_gen_andi_i64(addr, addr, ~0x7);
     }
 
-    va = (fp ? cpu_fir[ra] : ctx->ir[ra]);
-    tcg_gen_qemu_load(va, addr, ctx->mem_idx);
+    dest = ctx->ir[ra];
+    tcg_gen_qemu_ld_i64(dest, addr, ctx->mem_idx, op);
 
-    tcg_temp_free(tmp);
+    if (locked) {
+        tcg_gen_mov_i64(cpu_lock_addr, addr);
+        tcg_gen_mov_i64(cpu_lock_value, dest);
+    }
+    tcg_temp_free(addr);
 }
 
-static inline void gen_qemu_stf(TCGv t0, TCGv t1, int flags)
+static void gen_stf(DisasContext *ctx, TCGv src, TCGv addr)
 {
     TCGv_i32 tmp32 = tcg_temp_new_i32();
-    gen_helper_f_to_memory(tmp32, t0);
-    tcg_gen_qemu_st_i32(tmp32, t1, flags, MO_LEUL);
+    gen_helper_f_to_memory(tmp32, addr);
+    tcg_gen_qemu_st_i32(tmp32, addr, ctx->mem_idx, MO_LEUL);
     tcg_temp_free_i32(tmp32);
 }
 
-static inline void gen_qemu_stg(TCGv t0, TCGv t1, int flags)
+static void gen_stg(DisasContext *ctx, TCGv src, TCGv addr)
 {
     TCGv tmp = tcg_temp_new();
-    gen_helper_g_to_memory(tmp, t0);
-    tcg_gen_qemu_st_i64(tmp, t1, flags, MO_LEQ);
+    gen_helper_g_to_memory(tmp, src);
+    tcg_gen_qemu_st_i64(tmp, addr, ctx->mem_idx, MO_LEQ);
     tcg_temp_free(tmp);
 }
 
-static inline void gen_qemu_sts(TCGv t0, TCGv t1, int flags)
+static void gen_sts(DisasContext *ctx, TCGv src, TCGv addr)
 {
     TCGv_i32 tmp32 = tcg_temp_new_i32();
-    gen_helper_s_to_memory(tmp32, t0);
-    tcg_gen_qemu_st_i32(tmp32, t1, flags, MO_LEUL);
+    gen_helper_s_to_memory(tmp32, src);
+    tcg_gen_qemu_st_i32(tmp32, addr, ctx->mem_idx, MO_LEUL);
     tcg_temp_free_i32(tmp32);
 }
 
-static inline void gen_store_mem(DisasContext *ctx,
-                                 void (*tcg_gen_qemu_store)(TCGv t0, TCGv t1,
-                                                            int flags),
-                                 int ra, int rb, int32_t disp16, bool fp,
-                                 bool clear)
+static void gen_stt(DisasContext *ctx, TCGv src, TCGv addr)
 {
-    TCGv tmp, addr, va;
+    tcg_gen_qemu_st_i64(src, addr, ctx->mem_idx, MO_LEQ);
+}
 
-    tmp = tcg_temp_new();
-    addr = load_gpr(ctx, rb);
+static void gen_store_fp(DisasContext *ctx, int ra, int rb, int32_t disp16,
+                         void (*func)(DisasContext *, TCGv, TCGv))
+{
+    TCGv addr = tcg_temp_new();
+    tcg_gen_addi_i64(addr, load_gpr(ctx, rb), disp16);
+    func(ctx, load_fpr(ctx, ra), addr);
+    tcg_temp_free(addr);
+}
 
-    if (disp16) {
-        tcg_gen_addi_i64(tmp, addr, disp16);
-        addr = tmp;
-    }
+static void gen_store_int(DisasContext *ctx, int ra, int rb, int32_t disp16,
+                          MemOp op, bool clear)
+{
+    TCGv addr, src;
+
+    addr = tcg_temp_new();
+    tcg_gen_addi_i64(addr, load_gpr(ctx, rb), disp16);
     if (clear) {
-        tcg_gen_andi_i64(tmp, addr, ~0x7);
-        addr = tmp;
+        tcg_gen_andi_i64(addr, addr, ~0x7);
     }
 
-    va = (fp ? load_fpr(ctx, ra) : load_gpr(ctx, ra));
-    tcg_gen_qemu_store(va, addr, ctx->mem_idx);
+    src = load_gpr(ctx, ra);
+    tcg_gen_qemu_st_i64(src, addr, ctx->mem_idx, op);
 
-    tcg_temp_free(tmp);
+    tcg_temp_free(addr);
 }
 
 static DisasJumpType gen_store_conditional(DisasContext *ctx, int ra, int rb,
@@ -1480,30 +1483,30 @@ static DisasJumpType translate_one(DisasContext *ctx, uint32_t insn)
     case 0x0A:
         /* LDBU */
         REQUIRE_AMASK(BWX);
-        gen_load_mem(ctx, &tcg_gen_qemu_ld8u, ra, rb, disp16, 0, 0);
+        gen_load_int(ctx, ra, rb, disp16, MO_UB, 0, 0);
         break;
     case 0x0B:
         /* LDQ_U */
-        gen_load_mem(ctx, &tcg_gen_qemu_ld64, ra, rb, disp16, 0, 1);
+        gen_load_int(ctx, ra, rb, disp16, MO_LEQ, 1, 0);
         break;
     case 0x0C:
         /* LDWU */
         REQUIRE_AMASK(BWX);
-        gen_load_mem(ctx, &tcg_gen_qemu_ld16u, ra, rb, disp16, 0, 0);
+        gen_load_int(ctx, ra, rb, disp16, MO_LEUW, 0, 0);
         break;
     case 0x0D:
         /* STW */
         REQUIRE_AMASK(BWX);
-        gen_store_mem(ctx, &tcg_gen_qemu_st16, ra, rb, disp16, 0, 0);
+        gen_store_int(ctx, ra, rb, disp16, MO_LEUW, 0);
         break;
     case 0x0E:
         /* STB */
         REQUIRE_AMASK(BWX);
-        gen_store_mem(ctx, &tcg_gen_qemu_st8, ra, rb, disp16, 0, 0);
+        gen_store_int(ctx, ra, rb, disp16, MO_UB, 0);
         break;
     case 0x0F:
         /* STQ_U */
-        gen_store_mem(ctx, &tcg_gen_qemu_st64, ra, rb, disp16, 0, 1);
+        gen_store_int(ctx, ra, rb, disp16, MO_LEQ, 1);
         break;
 
     case 0x10:
@@ -2458,11 +2461,15 @@ static DisasJumpType translate_one(DisasContext *ctx, uint32_t insn)
                 break;
             case 0x2:
                 /* Longword physical access with lock (hw_ldl_l/p) */
-                gen_qemu_ldl_l(va, addr, MMU_PHYS_IDX);
+                tcg_gen_qemu_ld_i64(va, addr, MMU_PHYS_IDX, MO_LESL);
+                tcg_gen_mov_i64(cpu_lock_addr, addr);
+                tcg_gen_mov_i64(cpu_lock_value, va);
                 break;
             case 0x3:
                 /* Quadword physical access with lock (hw_ldq_l/p) */
-                gen_qemu_ldq_l(va, addr, MMU_PHYS_IDX);
+                tcg_gen_qemu_ld_i64(va, addr, MMU_PHYS_IDX, MO_LEQ);
+                tcg_gen_mov_i64(cpu_lock_addr, addr);
+                tcg_gen_mov_i64(cpu_lock_value, va);
                 break;
             case 0x4:
                 /* Longword virtual PTE fetch (hw_ldl/v) */
@@ -2776,66 +2783,66 @@ static DisasJumpType translate_one(DisasContext *ctx, uint32_t insn)
     case 0x20:
         /* LDF */
         REQUIRE_FEN;
-        gen_load_mem(ctx, &gen_qemu_ldf, ra, rb, disp16, 1, 0);
+        gen_load_fp(ctx, ra, rb, disp16, gen_ldf);
         break;
     case 0x21:
         /* LDG */
         REQUIRE_FEN;
-        gen_load_mem(ctx, &gen_qemu_ldg, ra, rb, disp16, 1, 0);
+        gen_load_fp(ctx, ra, rb, disp16, gen_ldg);
         break;
     case 0x22:
         /* LDS */
         REQUIRE_FEN;
-        gen_load_mem(ctx, &gen_qemu_lds, ra, rb, disp16, 1, 0);
+        gen_load_fp(ctx, ra, rb, disp16, gen_lds);
         break;
     case 0x23:
         /* LDT */
         REQUIRE_FEN;
-        gen_load_mem(ctx, &tcg_gen_qemu_ld64, ra, rb, disp16, 1, 0);
+        gen_load_fp(ctx, ra, rb, disp16, gen_ldt);
         break;
     case 0x24:
         /* STF */
         REQUIRE_FEN;
-        gen_store_mem(ctx, &gen_qemu_stf, ra, rb, disp16, 1, 0);
+        gen_store_fp(ctx, ra, rb, disp16, gen_stf);
         break;
     case 0x25:
         /* STG */
         REQUIRE_FEN;
-        gen_store_mem(ctx, &gen_qemu_stg, ra, rb, disp16, 1, 0);
+        gen_store_fp(ctx, ra, rb, disp16, gen_stg);
         break;
     case 0x26:
         /* STS */
         REQUIRE_FEN;
-        gen_store_mem(ctx, &gen_qemu_sts, ra, rb, disp16, 1, 0);
+        gen_store_fp(ctx, ra, rb, disp16, gen_sts);
         break;
     case 0x27:
         /* STT */
         REQUIRE_FEN;
-        gen_store_mem(ctx, &tcg_gen_qemu_st64, ra, rb, disp16, 1, 0);
+        gen_store_fp(ctx, ra, rb, disp16, gen_stt);
         break;
     case 0x28:
         /* LDL */
-        gen_load_mem(ctx, &tcg_gen_qemu_ld32s, ra, rb, disp16, 0, 0);
+        gen_load_int(ctx, ra, rb, disp16, MO_LESL, 0, 0);
         break;
     case 0x29:
         /* LDQ */
-        gen_load_mem(ctx, &tcg_gen_qemu_ld64, ra, rb, disp16, 0, 0);
+        gen_load_int(ctx, ra, rb, disp16, MO_LEQ, 0, 0);
         break;
     case 0x2A:
         /* LDL_L */
-        gen_load_mem(ctx, &gen_qemu_ldl_l, ra, rb, disp16, 0, 0);
+        gen_load_int(ctx, ra, rb, disp16, MO_LESL, 0, 1);
         break;
     case 0x2B:
         /* LDQ_L */
-        gen_load_mem(ctx, &gen_qemu_ldq_l, ra, rb, disp16, 0, 0);
+        gen_load_int(ctx, ra, rb, disp16, MO_LEQ, 0, 1);
         break;
     case 0x2C:
         /* STL */
-        gen_store_mem(ctx, &tcg_gen_qemu_st32, ra, rb, disp16, 0, 0);
+        gen_store_int(ctx, ra, rb, disp16, MO_LEUL, 0);
         break;
     case 0x2D:
         /* STQ */
-        gen_store_mem(ctx, &tcg_gen_qemu_st64, ra, rb, disp16, 0, 0);
+        gen_store_int(ctx, ra, rb, disp16, MO_LEQ, 0);
         break;
     case 0x2E:
         /* STL_C */