target/hppa: Implement STDBY

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

4 files changed, 213 insertions, 5 deletions

diff --git a/target/hppa/helper.h b/target/hppa/helper.h
index 647f043c85..9920d38ded 100644
--- a/target/hppa/helper.h
+++ b/target/hppa/helper.h
@@ -16,6 +16,11 @@ DEF_HELPER_FLAGS_3(stby_b_parallel, TCG_CALL_NO_WG, void, env, tl, tr)
 DEF_HELPER_FLAGS_3(stby_e, TCG_CALL_NO_WG, void, env, tl, tr)
 DEF_HELPER_FLAGS_3(stby_e_parallel, TCG_CALL_NO_WG, void, env, tl, tr)
 
+DEF_HELPER_FLAGS_3(stdby_b, TCG_CALL_NO_WG, void, env, tl, tr)
+DEF_HELPER_FLAGS_3(stdby_b_parallel, TCG_CALL_NO_WG, void, env, tl, tr)
+DEF_HELPER_FLAGS_3(stdby_e, TCG_CALL_NO_WG, void, env, tl, tr)
+DEF_HELPER_FLAGS_3(stdby_e_parallel, TCG_CALL_NO_WG, void, env, tl, tr)
+
 DEF_HELPER_FLAGS_1(ldc_check, TCG_CALL_NO_RWG, void, tl)
 
 DEF_HELPER_FLAGS_4(probe, TCG_CALL_NO_WG, tr, env, tl, i32, i32)
diff --git a/target/hppa/insns.decode b/target/hppa/insns.decode
index ba7731b517..9d8c6a1a16 100644
--- a/target/hppa/insns.decode
+++ b/target/hppa/insns.decode
@@ -228,6 +228,7 @@ lda             000011 ..... ..... .. . 0 -- 0100      ......   @ldstx size=3
 sta             000011 ..... ..... .. . 1 -- 1110      ......   @stim5 size=2
 sta             000011 ..... ..... .. . 1 -- 1111      ......   @stim5 size=3
 stby            000011 b:5 r:5 sp:2 a:1 1 -- 1100 m:1   .....   disp=%im5_0
+stdby           000011 b:5 r:5 sp:2 a:1 1 -- 1101 m:1   .....   disp=%im5_0
 
 @fldstwx        ...... b:5 x:5   sp:2 scale:1 ....... m:1 ..... \
                 &ldst t=%rt64 disp=0 size=2
diff --git a/target/hppa/op_helper.c b/target/hppa/op_helper.c
index 837e2b3117..b5b45f5120 100644
--- a/target/hppa/op_helper.c
+++ b/target/hppa/op_helper.c
@@ -56,11 +56,11 @@ void HELPER(tcond)(CPUHPPAState *env, target_ureg cond)
     }
 }
 
-static void atomic_store_3(CPUHPPAState *env, target_ulong addr,
-                           uint32_t val, uintptr_t ra)
+static void atomic_store_mask32(CPUHPPAState *env, target_ulong addr,
+                                uint32_t val, uint32_t mask, uintptr_t ra)
 {
     int mmu_idx = cpu_mmu_index(env, 0);
-    uint32_t old, new, cmp, mask, *haddr;
+    uint32_t old, new, cmp, *haddr;
     void *vaddr;
 
     vaddr = probe_access(env, addr, 3, MMU_DATA_STORE, mmu_idx, ra);
@@ -81,6 +81,35 @@ static void atomic_store_3(CPUHPPAState *env, target_ulong addr,
     }
 }
 
+static void atomic_store_mask64(CPUHPPAState *env, target_ulong addr,
+                                uint64_t val, uint64_t mask,
+                                int size, uintptr_t ra)
+{
+#ifdef CONFIG_ATOMIC64
+    int mmu_idx = cpu_mmu_index(env, 0);
+    uint64_t old, new, cmp, *haddr;
+    void *vaddr;
+
+    vaddr = probe_access(env, addr, size, MMU_DATA_STORE, mmu_idx, ra);
+    if (vaddr == NULL) {
+        cpu_loop_exit_atomic(env_cpu(env), ra);
+    }
+    haddr = (uint64_t *)((uintptr_t)vaddr & -8);
+
+    old = *haddr;
+    while (1) {
+        new = be32_to_cpu((cpu_to_be32(old) & ~mask) | (val & mask));
+        cmp = qatomic_cmpxchg__nocheck(haddr, old, new);
+        if (cmp == old) {
+            return;
+        }
+        old = cmp;
+    }
+#else
+    cpu_loop_exit_atomic(env_cpu(env), ra);
+#endif
+}
+
 static void do_stby_b(CPUHPPAState *env, target_ulong addr, target_ureg val,
                       bool parallel, uintptr_t ra)
 {
@@ -94,7 +123,7 @@ static void do_stby_b(CPUHPPAState *env, target_ulong addr, target_ureg val,
     case 1:
         /* The 3 byte store must appear atomic.  */
         if (parallel) {
-            atomic_store_3(env, addr, val, ra);
+            atomic_store_mask32(env, addr, val, 0x00ffffffu, ra);
         } else {
             cpu_stb_data_ra(env, addr, val >> 16, ra);
             cpu_stw_data_ra(env, addr + 1, val, ra);
@@ -106,6 +135,62 @@ static void do_stby_b(CPUHPPAState *env, target_ulong addr, target_ureg val,
     }
 }
 
+static void do_stdby_b(CPUHPPAState *env, target_ulong addr, uint64_t val,
+                       bool parallel, uintptr_t ra)
+{
+    switch (addr & 7) {
+    case 7:
+        cpu_stb_data_ra(env, addr, val, ra);
+        break;
+    case 6:
+        cpu_stw_data_ra(env, addr, val, ra);
+        break;
+    case 5:
+        /* The 3 byte store must appear atomic.  */
+        if (parallel) {
+            atomic_store_mask32(env, addr, val, 0x00ffffffu, ra);
+        } else {
+            cpu_stb_data_ra(env, addr, val >> 16, ra);
+            cpu_stw_data_ra(env, addr + 1, val, ra);
+        }
+        break;
+    case 4:
+        cpu_stl_data_ra(env, addr, val, ra);
+        break;
+    case 3:
+        /* The 5 byte store must appear atomic.  */
+        if (parallel) {
+            atomic_store_mask64(env, addr, val, 0x000000ffffffffffull, 5, ra);
+        } else {
+            cpu_stb_data_ra(env, addr, val >> 32, ra);
+            cpu_stl_data_ra(env, addr + 1, val, ra);
+        }
+        break;
+    case 2:
+        /* The 6 byte store must appear atomic.  */
+        if (parallel) {
+            atomic_store_mask64(env, addr, val, 0x0000ffffffffffffull, 6, ra);
+        } else {
+            cpu_stw_data_ra(env, addr, val >> 32, ra);
+            cpu_stl_data_ra(env, addr + 2, val, ra);
+        }
+        break;
+    case 1:
+        /* The 7 byte store must appear atomic.  */
+        if (parallel) {
+            atomic_store_mask64(env, addr, val, 0x00ffffffffffffffull, 7, ra);
+        } else {
+            cpu_stb_data_ra(env, addr, val >> 48, ra);
+            cpu_stw_data_ra(env, addr + 1, val >> 32, ra);
+            cpu_stl_data_ra(env, addr + 3, val, ra);
+        }
+        break;
+    default:
+        cpu_stq_data_ra(env, addr, val, ra);
+        break;
+    }
+}
+
 void HELPER(stby_b)(CPUHPPAState *env, target_ulong addr, target_ureg val)
 {
     do_stby_b(env, addr, val, false, GETPC());
@@ -117,6 +202,17 @@ void HELPER(stby_b_parallel)(CPUHPPAState *env, target_ulong addr,
     do_stby_b(env, addr, val, true, GETPC());
 }
 
+void HELPER(stdby_b)(CPUHPPAState *env, target_ulong addr, target_ureg val)
+{
+    do_stdby_b(env, addr, val, false, GETPC());
+}
+
+void HELPER(stdby_b_parallel)(CPUHPPAState *env, target_ulong addr,
+                              target_ureg val)
+{
+    do_stdby_b(env, addr, val, true, GETPC());
+}
+
 static void do_stby_e(CPUHPPAState *env, target_ulong addr, target_ureg val,
                       bool parallel, uintptr_t ra)
 {
@@ -124,7 +220,68 @@ static void do_stby_e(CPUHPPAState *env, target_ulong addr, target_ureg val,
     case 3:
         /* The 3 byte store must appear atomic.  */
         if (parallel) {
-            atomic_store_3(env, addr - 3, val, ra);
+            atomic_store_mask32(env, addr - 3, val, 0xffffff00u, ra);
+        } else {
+            cpu_stw_data_ra(env, addr - 3, val >> 16, ra);
+            cpu_stb_data_ra(env, addr - 1, val >> 8, ra);
+        }
+        break;
+    case 2:
+        cpu_stw_data_ra(env, addr - 2, val >> 16, ra);
+        break;
+    case 1:
+        cpu_stb_data_ra(env, addr - 1, val >> 24, ra);
+        break;
+    default:
+        /* Nothing is stored, but protection is checked and the
+           cacheline is marked dirty.  */
+        probe_write(env, addr, 0, cpu_mmu_index(env, 0), ra);
+        break;
+    }
+}
+
+static void do_stdby_e(CPUHPPAState *env, target_ulong addr, uint64_t val,
+                       bool parallel, uintptr_t ra)
+{
+    switch (addr & 7) {
+    case 7:
+        /* The 7 byte store must appear atomic.  */
+        if (parallel) {
+            atomic_store_mask64(env, addr - 7, val,
+                                0xffffffffffffff00ull, 7, ra);
+        } else {
+            cpu_stl_data_ra(env, addr - 7, val >> 32, ra);
+            cpu_stw_data_ra(env, addr - 3, val >> 16, ra);
+            cpu_stb_data_ra(env, addr - 1, val >> 8, ra);
+        }
+        break;
+    case 6:
+        /* The 6 byte store must appear atomic.  */
+        if (parallel) {
+            atomic_store_mask64(env, addr - 6, val,
+                                0xffffffffffff0000ull, 6, ra);
+        } else {
+            cpu_stl_data_ra(env, addr - 6, val >> 32, ra);
+            cpu_stw_data_ra(env, addr - 2, val >> 16, ra);
+        }
+        break;
+    case 5:
+        /* The 5 byte store must appear atomic.  */
+        if (parallel) {
+            atomic_store_mask64(env, addr - 5, val,
+                                0xffffffffff000000ull, 5, ra);
+        } else {
+            cpu_stl_data_ra(env, addr - 5, val >> 32, ra);
+            cpu_stb_data_ra(env, addr - 1, val >> 24, ra);
+        }
+        break;
+    case 4:
+        cpu_stl_data_ra(env, addr - 4, val >> 32, ra);
+        break;
+    case 3:
+        /* The 3 byte store must appear atomic.  */
+        if (parallel) {
+            atomic_store_mask32(env, addr - 3, val, 0xffffff00u, ra);
         } else {
             cpu_stw_data_ra(env, addr - 3, val >> 16, ra);
             cpu_stb_data_ra(env, addr - 1, val >> 8, ra);
@@ -155,6 +312,17 @@ void HELPER(stby_e_parallel)(CPUHPPAState *env, target_ulong addr,
     do_stby_e(env, addr, val, true, GETPC());
 }
 
+void HELPER(stdby_e)(CPUHPPAState *env, target_ulong addr, target_ureg val)
+{
+    do_stdby_e(env, addr, val, false, GETPC());
+}
+
+void HELPER(stdby_e_parallel)(CPUHPPAState *env, target_ulong addr,
+                              target_ureg val)
+{
+    do_stdby_e(env, addr, val, true, GETPC());
+}
+
 void HELPER(ldc_check)(target_ulong addr)
 {
     if (unlikely(addr & 0xf)) {
diff --git a/target/hppa/translate.c b/target/hppa/translate.c
index 91249d89ca..485251bded 100644
--- a/target/hppa/translate.c
+++ b/target/hppa/translate.c
@@ -3052,6 +3052,40 @@ static bool trans_stby(DisasContext *ctx, arg_stby *a)
     return nullify_end(ctx);
 }
 
+static bool trans_stdby(DisasContext *ctx, arg_stby *a)
+{
+    TCGv_reg ofs, val;
+    TCGv_tl addr;
+
+    if (!ctx->is_pa20) {
+        return false;
+    }
+    nullify_over(ctx);
+
+    form_gva(ctx, &addr, &ofs, a->b, 0, 0, a->disp, a->sp, a->m,
+             ctx->mmu_idx == MMU_PHYS_IDX);
+    val = load_gpr(ctx, a->r);
+    if (a->a) {
+        if (tb_cflags(ctx->base.tb) & CF_PARALLEL) {
+            gen_helper_stdby_e_parallel(tcg_env, addr, val);
+        } else {
+            gen_helper_stdby_e(tcg_env, addr, val);
+        }
+    } else {
+        if (tb_cflags(ctx->base.tb) & CF_PARALLEL) {
+            gen_helper_stdby_b_parallel(tcg_env, addr, val);
+        } else {
+            gen_helper_stdby_b(tcg_env, addr, val);
+        }
+    }
+    if (a->m) {
+        tcg_gen_andi_reg(ofs, ofs, ~7);
+        save_gpr(ctx, a->b, ofs);
+    }
+
+    return nullify_end(ctx);
+}
+
 static bool trans_lda(DisasContext *ctx, arg_ldst *a)
 {
     int hold_mmu_idx = ctx->mmu_idx;