target-alpha: Fix load-locked/store-conditional.

Use an exception plus start_exclusive to implement the compare-and-swap.
This follows the example set by the MIPS and PPC ports.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>

5 files changed, 193 insertions, 79 deletions

diff --git a/linux-user/main.c b/linux-user/main.c
index 5680d8e0cd..18b52c00a4 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -2349,6 +2349,51 @@ void cpu_loop(CPUM68KState *env)
 #endif /* TARGET_M68K */
 
 #ifdef TARGET_ALPHA
+static void do_store_exclusive(CPUAlphaState *env, int reg, int quad)
+{
+    target_ulong addr, val, tmp;
+    target_siginfo_t info;
+    int ret = 0;
+
+    addr = env->lock_addr;
+    tmp = env->lock_st_addr;
+    env->lock_addr = -1;
+    env->lock_st_addr = 0;
+
+    start_exclusive();
+    mmap_lock();
+
+    if (addr == tmp) {
+        if (quad ? get_user_s64(val, addr) : get_user_s32(val, addr)) {
+            goto do_sigsegv;
+        }
+
+        if (val == env->lock_value) {
+            tmp = env->ir[reg];
+            if (quad ? put_user_u64(tmp, addr) : put_user_u32(tmp, addr)) {
+                goto do_sigsegv;
+            }
+            ret = 1;
+        }
+    }
+    env->ir[reg] = ret;
+    env->pc += 4;
+
+    mmap_unlock();
+    end_exclusive();
+    return;
+
+ do_sigsegv:
+    mmap_unlock();
+    end_exclusive();
+
+    info.si_signo = TARGET_SIGSEGV;
+    info.si_errno = 0;
+    info.si_code = TARGET_SEGV_MAPERR;
+    info._sifields._sigfault._addr = addr;
+    queue_signal(env, TARGET_SIGSEGV, &info);
+}
+
 void cpu_loop (CPUState *env)
 {
     int trapnr;
@@ -2373,6 +2418,7 @@ void cpu_loop (CPUState *env)
             exit(1);
             break;
         case EXCP_ARITH:
+            env->lock_addr = -1;
             info.si_signo = TARGET_SIGFPE;
             info.si_errno = 0;
             info.si_code = TARGET_FPE_FLTINV;
@@ -2384,6 +2430,7 @@ void cpu_loop (CPUState *env)
             exit(1);
             break;
         case EXCP_DFAULT:
+            env->lock_addr = -1;
             info.si_signo = TARGET_SIGSEGV;
             info.si_errno = 0;
             info.si_code = 0;  /* ??? SEGV_MAPERR vs SEGV_ACCERR.  */
@@ -2407,6 +2454,7 @@ void cpu_loop (CPUState *env)
             exit(1);
             break;
         case EXCP_UNALIGN:
+            env->lock_addr = -1;
             info.si_signo = TARGET_SIGBUS;
             info.si_errno = 0;
             info.si_code = TARGET_BUS_ADRALN;
@@ -2415,6 +2463,7 @@ void cpu_loop (CPUState *env)
             break;
         case EXCP_OPCDEC:
         do_sigill:
+            env->lock_addr = -1;
             info.si_signo = TARGET_SIGILL;
             info.si_errno = 0;
             info.si_code = TARGET_ILL_ILLOPC;
@@ -2425,6 +2474,7 @@ void cpu_loop (CPUState *env)
             /* No-op.  Linux simply re-enables the FPU.  */
             break;
         case EXCP_CALL_PAL ... (EXCP_CALL_PALP - 1):
+            env->lock_addr = -1;
             switch ((trapnr >> 6) | 0x80) {
             case 0x80:
                 /* BPT */
@@ -2514,11 +2564,16 @@ void cpu_loop (CPUState *env)
         case EXCP_DEBUG:
             info.si_signo = gdb_handlesig (env, TARGET_SIGTRAP);
             if (info.si_signo) {
+                env->lock_addr = -1;
                 info.si_errno = 0;
                 info.si_code = TARGET_TRAP_BRKPT;
                 queue_signal(env, info.si_signo, &info);
             }
             break;
+        case EXCP_STL_C:
+        case EXCP_STQ_C:
+            do_store_exclusive(env, env->error_code, trapnr - EXCP_STL_C);
+            break;
         default:
             printf ("Unhandled trap: 0x%x\n", trapnr);
             cpu_dump_state(env, stderr, fprintf, 0);
diff --git a/target-alpha/cpu.h b/target-alpha/cpu.h
index c3979300f1..817504b02b 100644
--- a/target-alpha/cpu.h
+++ b/target-alpha/cpu.h
@@ -355,11 +355,13 @@ struct CPUAlphaState {
     uint64_t ir[31];
     float64 fir[31];
     uint64_t pc;
-    uint64_t lock;
     uint32_t pcc[2];
     uint64_t ipr[IPR_LAST];
     uint64_t ps;
     uint64_t unique;
+    uint64_t lock_addr;
+    uint64_t lock_st_addr;
+    uint64_t lock_value;
     float_status fp_status;
     /* The following fields make up the FPCR, but in FP_STATUS format.  */
     uint8_t fpcr_exc_status;
@@ -440,6 +442,8 @@ enum {
     /* Pseudo exception for console */
     EXCP_CONSOLE_DISPATCH = 0x4001,
     EXCP_CONSOLE_FIXUP    = 0x4002,
+    EXCP_STL_C            = 0x4003,
+    EXCP_STQ_C            = 0x4004,
 };
 
 /* Arithmetic exception */
diff --git a/target-alpha/helper.c b/target-alpha/helper.c
index 9a039cb96b..b6d21601d5 100644
--- a/target-alpha/helper.c
+++ b/target-alpha/helper.c
@@ -557,12 +557,15 @@ void cpu_dump_state (CPUState *env, FILE *f,
         if ((i % 3) == 2)
             cpu_fprintf(f, "\n");
     }
-    cpu_fprintf(f, "\n");
+
+    cpu_fprintf(f, "lock_a   " TARGET_FMT_lx " lock_v   " TARGET_FMT_lx "\n",
+                env->lock_addr, env->lock_value);
+
     for (i = 0; i < 31; i++) {
         cpu_fprintf(f, "FIR%02d    " TARGET_FMT_lx " ", i,
                     *((uint64_t *)(&env->fir[i])));
         if ((i % 3) == 2)
             cpu_fprintf(f, "\n");
     }
-    cpu_fprintf(f, "\nlock     " TARGET_FMT_lx "\n", env->lock);
+    cpu_fprintf(f, "\n");
 }
diff --git a/target-alpha/op_helper.c b/target-alpha/op_helper.c
index a20913033d..bfc095cc8a 100644
--- a/target-alpha/op_helper.c
+++ b/target-alpha/op_helper.c
@@ -1159,6 +1159,7 @@ void helper_hw_rei (void)
     env->pc = env->ipr[IPR_EXC_ADDR] & ~3;
     env->ipr[IPR_EXC_ADDR] = env->ipr[IPR_EXC_ADDR] & 1;
     env->intr_flag = 0;
+    env->lock_addr = -1;
     /* XXX: re-enable interrupts and memory mapping */
 }
 
@@ -1167,6 +1168,7 @@ void helper_hw_ret (uint64_t a)
     env->pc = a & ~3;
     env->ipr[IPR_EXC_ADDR] = a & 1;
     env->intr_flag = 0;
+    env->lock_addr = -1;
     /* XXX: re-enable interrupts and memory mapping */
 }
 
diff --git a/target-alpha/translate.c b/target-alpha/translate.c
index 924fc706ae..3a1c625f73 100644
--- a/target-alpha/translate.c
+++ b/target-alpha/translate.c
@@ -86,7 +86,9 @@ static TCGv_ptr cpu_env;
 static TCGv cpu_ir[31];
 static TCGv cpu_fir[31];
 static TCGv cpu_pc;
-static TCGv cpu_lock;
+static TCGv cpu_lock_addr;
+static TCGv cpu_lock_st_addr;
+static TCGv cpu_lock_value;
 #ifdef CONFIG_USER_ONLY
 static TCGv cpu_uniq;
 #endif
@@ -123,8 +125,15 @@ static void alpha_translate_init(void)
     cpu_pc = tcg_global_mem_new_i64(TCG_AREG0,
                                     offsetof(CPUState, pc), "pc");
 
-    cpu_lock = tcg_global_mem_new_i64(TCG_AREG0,
-                                      offsetof(CPUState, lock), "lock");
+    cpu_lock_addr = tcg_global_mem_new_i64(TCG_AREG0,
+					   offsetof(CPUState, lock_addr),
+					   "lock_addr");
+    cpu_lock_st_addr = tcg_global_mem_new_i64(TCG_AREG0,
+					      offsetof(CPUState, lock_st_addr),
+					      "lock_st_addr");
+    cpu_lock_value = tcg_global_mem_new_i64(TCG_AREG0,
+					    offsetof(CPUState, lock_value),
+					    "lock_value");
 
 #ifdef CONFIG_USER_ONLY
     cpu_uniq = tcg_global_mem_new_i64(TCG_AREG0,
@@ -189,14 +198,16 @@ static inline void gen_qemu_lds(TCGv t0, TCGv t1, int flags)
 
 static inline void gen_qemu_ldl_l(TCGv t0, TCGv t1, int flags)
 {
-    tcg_gen_mov_i64(cpu_lock, t1);
     tcg_gen_qemu_ld32s(t0, t1, flags);
+    tcg_gen_mov_i64(cpu_lock_addr, t1);
+    tcg_gen_mov_i64(cpu_lock_value, t0);
 }
 
 static inline void gen_qemu_ldq_l(TCGv t0, TCGv t1, int flags)
 {
-    tcg_gen_mov_i64(cpu_lock, t1);
     tcg_gen_qemu_ld64(t0, t1, flags);
+    tcg_gen_mov_i64(cpu_lock_addr, t1);
+    tcg_gen_mov_i64(cpu_lock_value, t0);
 }
 
 static inline void gen_load_mem(DisasContext *ctx,
@@ -205,25 +216,31 @@ static inline void gen_load_mem(DisasContext *ctx,
                                 int ra, int rb, int32_t disp16, int fp,
                                 int clear)
 {
-    TCGv addr;
+    TCGv addr, va;
 
-    if (unlikely(ra == 31))
+    /* LDQ_U with ra $31 is UNOP.  Other various loads are forms of
+       prefetches, which we can treat as nops.  No worries about
+       missed exceptions here.  */
+    if (unlikely(ra == 31)) {
         return;
+    }
 
     addr = tcg_temp_new();
     if (rb != 31) {
         tcg_gen_addi_i64(addr, cpu_ir[rb], disp16);
-        if (clear)
+        if (clear) {
             tcg_gen_andi_i64(addr, addr, ~0x7);
+        }
     } else {
-        if (clear)
+        if (clear) {
             disp16 &= ~0x7;
+        }
         tcg_gen_movi_i64(addr, disp16);
     }
-    if (fp)
-        tcg_gen_qemu_load(cpu_fir[ra], addr, ctx->mem_idx);
-    else
-        tcg_gen_qemu_load(cpu_ir[ra], addr, ctx->mem_idx);
+
+    va = (fp ? cpu_fir[ra] : cpu_ir[ra]);
+    tcg_gen_qemu_load(va, addr, ctx->mem_idx);
+
     tcg_temp_free(addr);
 }
 
@@ -257,73 +274,105 @@ static inline void gen_qemu_sts(TCGv t0, TCGv t1, int flags)
     tcg_temp_free_i32(tmp32);
 }
 
-static inline void gen_qemu_stl_c(TCGv t0, TCGv t1, int flags)
-{
-    int l1, l2;
-
-    l1 = gen_new_label();
-    l2 = gen_new_label();
-    tcg_gen_brcond_i64(TCG_COND_NE, cpu_lock, t1, l1);
-    tcg_gen_qemu_st32(t0, t1, flags);
-    tcg_gen_movi_i64(t0, 1);
-    tcg_gen_br(l2);
-    gen_set_label(l1);
-    tcg_gen_movi_i64(t0, 0);
-    gen_set_label(l2);
-    tcg_gen_movi_i64(cpu_lock, -1);
-}
-
-static inline void gen_qemu_stq_c(TCGv t0, TCGv t1, int flags)
-{
-    int l1, l2;
-
-    l1 = gen_new_label();
-    l2 = gen_new_label();
-    tcg_gen_brcond_i64(TCG_COND_NE, cpu_lock, t1, l1);
-    tcg_gen_qemu_st64(t0, t1, flags);
-    tcg_gen_movi_i64(t0, 1);
-    tcg_gen_br(l2);
-    gen_set_label(l1);
-    tcg_gen_movi_i64(t0, 0);
-    gen_set_label(l2);
-    tcg_gen_movi_i64(cpu_lock, -1);
-}
-
 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, int fp,
-                                 int clear, int local)
+                                 int clear)
 {
-    TCGv addr;
-    if (local)
-        addr = tcg_temp_local_new();
-    else
-        addr = tcg_temp_new();
+    TCGv addr, va;
+
+    addr = tcg_temp_new();
     if (rb != 31) {
         tcg_gen_addi_i64(addr, cpu_ir[rb], disp16);
-        if (clear)
+        if (clear) {
             tcg_gen_andi_i64(addr, addr, ~0x7);
+        }
     } else {
-        if (clear)
+        if (clear) {
             disp16 &= ~0x7;
+        }
         tcg_gen_movi_i64(addr, disp16);
     }
-    if (ra != 31) {
-        if (fp)
-            tcg_gen_qemu_store(cpu_fir[ra], addr, ctx->mem_idx);
-        else
-            tcg_gen_qemu_store(cpu_ir[ra], addr, ctx->mem_idx);
+
+    if (ra == 31) {
+        va = tcg_const_i64(0);
     } else {
-        TCGv zero;
-        if (local)
-            zero = tcg_const_local_i64(0);
-        else
-            zero = tcg_const_i64(0);
-        tcg_gen_qemu_store(zero, addr, ctx->mem_idx);
-        tcg_temp_free(zero);
+        va = (fp ? cpu_fir[ra] : cpu_ir[ra]);
     }
+    tcg_gen_qemu_store(va, addr, ctx->mem_idx);
+
     tcg_temp_free(addr);
+    if (ra == 31) {
+        tcg_temp_free(va);
+    }
+}
+
+static ExitStatus gen_store_conditional(DisasContext *ctx, int ra, int rb,
+                                        int32_t disp16, int quad)
+{
+    TCGv addr;
+
+    if (ra == 31) {
+        /* ??? Don't bother storing anything.  The user can't tell
+           the difference, since the zero register always reads zero.  */
+        return NO_EXIT;
+    }
+
+#if defined(CONFIG_USER_ONLY)
+    addr = cpu_lock_st_addr;
+#else
+    addr = tcg_local_new();
+#endif
+
+    if (rb != 31) {
+        tcg_gen_addi_i64(addr, cpu_ir[rb], disp16);
+    } else {
+        tcg_gen_movi_i64(addr, disp16);
+    }
+
+#if defined(CONFIG_USER_ONLY)
+    /* ??? This is handled via a complicated version of compare-and-swap
+       in the cpu_loop.  Hopefully one day we'll have a real CAS opcode
+       in TCG so that this isn't necessary.  */
+    return gen_excp(ctx, quad ? EXCP_STQ_C : EXCP_STL_C, ra);
+#else
+    /* ??? In system mode we are never multi-threaded, so CAS can be
+       implemented via a non-atomic load-compare-store sequence.  */
+    {
+        int lab_fail, lab_done;
+        TCGv val;
+
+        lab_fail = gen_new_label();
+        lab_done = gen_new_label();
+        tcg_gen_brcond(TCG_COND_NE, addr, cpu_lock_addr, lab_fail);
+
+        val = tcg_temp_new();
+        if (quad) {
+            tcg_gen_qemu_ld64(val, addr, ctx->mem_idx);
+        } else {
+            tcg_gen_qemu_ld32s(val, addr, ctx->mem_idx);
+        }
+        tcg_gen_brcond(TCG_COND_NE, val, cpu_lock_value, lab_fail);
+
+        if (quad) {
+            tcg_gen_qemu_st64(cpu_ir[ra], addr, ctx->mem_idx);
+        } else {
+            tcg_gen_qemu_st32(cpu_ir[ra], addr, ctx->mem_idx);
+        }
+        tcg_gen_movi_i64(cpu_ir[ra], 1);
+        tcg_gen_br(lab_done);
+
+        gen_set_label(lab_fail);
+        tcg_gen_movi_i64(cpu_ir[ra], 0);
+
+        gen_set_label(lab_done);
+        tcg_gen_movi_i64(cpu_lock_addr, -1);
+
+        tcg_temp_free(addr);
+        return NO_EXIT;
+    }
+#endif
 }
 
 static int use_goto_tb(DisasContext *ctx, uint64_t dest)
@@ -1533,15 +1582,15 @@ static ExitStatus translate_one(DisasContext *ctx, uint32_t insn)
         break;
     case 0x0D:
         /* STW */
-        gen_store_mem(ctx, &tcg_gen_qemu_st16, ra, rb, disp16, 0, 0, 0);
+        gen_store_mem(ctx, &tcg_gen_qemu_st16, ra, rb, disp16, 0, 0);
         break;
     case 0x0E:
         /* STB */
-        gen_store_mem(ctx, &tcg_gen_qemu_st8, ra, rb, disp16, 0, 0, 0);
+        gen_store_mem(ctx, &tcg_gen_qemu_st8, ra, rb, disp16, 0, 0);
         break;
     case 0x0F:
         /* STQ_U */
-        gen_store_mem(ctx, &tcg_gen_qemu_st64, ra, rb, disp16, 0, 1, 0);
+        gen_store_mem(ctx, &tcg_gen_qemu_st64, ra, rb, disp16, 0, 1);
         break;
     case 0x10:
         switch (fn7) {
@@ -2974,19 +3023,19 @@ static ExitStatus translate_one(DisasContext *ctx, uint32_t insn)
         break;
     case 0x24:
         /* STF */
-        gen_store_mem(ctx, &gen_qemu_stf, ra, rb, disp16, 1, 0, 0);
+        gen_store_mem(ctx, &gen_qemu_stf, ra, rb, disp16, 1, 0);
         break;
     case 0x25:
         /* STG */
-        gen_store_mem(ctx, &gen_qemu_stg, ra, rb, disp16, 1, 0, 0);
+        gen_store_mem(ctx, &gen_qemu_stg, ra, rb, disp16, 1, 0);
         break;
     case 0x26:
         /* STS */
-        gen_store_mem(ctx, &gen_qemu_sts, ra, rb, disp16, 1, 0, 0);
+        gen_store_mem(ctx, &gen_qemu_sts, ra, rb, disp16, 1, 0);
         break;
     case 0x27:
         /* STT */
-        gen_store_mem(ctx, &tcg_gen_qemu_st64, ra, rb, disp16, 1, 0, 0);
+        gen_store_mem(ctx, &tcg_gen_qemu_st64, ra, rb, disp16, 1, 0);
         break;
     case 0x28:
         /* LDL */
@@ -3006,19 +3055,19 @@ static ExitStatus translate_one(DisasContext *ctx, uint32_t insn)
         break;
     case 0x2C:
         /* STL */
-        gen_store_mem(ctx, &tcg_gen_qemu_st32, ra, rb, disp16, 0, 0, 0);
+        gen_store_mem(ctx, &tcg_gen_qemu_st32, ra, rb, disp16, 0, 0);
         break;
     case 0x2D:
         /* STQ */
-        gen_store_mem(ctx, &tcg_gen_qemu_st64, ra, rb, disp16, 0, 0, 0);
+        gen_store_mem(ctx, &tcg_gen_qemu_st64, ra, rb, disp16, 0, 0);
         break;
     case 0x2E:
         /* STL_C */
-        gen_store_mem(ctx, &gen_qemu_stl_c, ra, rb, disp16, 0, 0, 1);
+        ret = gen_store_conditional(ctx, ra, rb, disp16, 0);
         break;
     case 0x2F:
         /* STQ_C */
-        gen_store_mem(ctx, &gen_qemu_stq_c, ra, rb, disp16, 0, 0, 1);
+        ret = gen_store_conditional(ctx, ra, rb, disp16, 1);
         break;
     case 0x30:
         /* BR */
@@ -3284,6 +3333,7 @@ CPUAlphaState * cpu_alpha_init (const char *cpu_model)
 #else
     pal_init(env);
 #endif
+    env->lock_addr = -1;
 
     /* Initialize IPR */
 #if defined (CONFIG_USER_ONLY)