target-tilegx: Handle atomic instructions

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <rth@twiddle.net>

linux-user/main.c

@@ -3436,6 +3436,148 @@ static void gen_sigill_reg(CPUTLGState *env)
     queue_signal(env, info.si_signo, &info);
 }
 
+static void set_regval(CPUTLGState *env, uint8_t reg, uint64_t val)
+{
+    if (unlikely(reg >= TILEGX_R_COUNT)) {
+        switch (reg) {
+        case TILEGX_R_SN:
+        case TILEGX_R_ZERO:
+            return;
+        case TILEGX_R_IDN0:
+        case TILEGX_R_IDN1:
+        case TILEGX_R_UDN0:
+        case TILEGX_R_UDN1:
+        case TILEGX_R_UDN2:
+        case TILEGX_R_UDN3:
+            gen_sigill_reg(env);
+            return;
+        default:
+            g_assert_not_reached();
+        }
+    }
+    env->regs[reg] = val;
+}
+
+/*
+ * Compare the 8-byte contents of the CmpValue SPR with the 8-byte value in
+ * memory at the address held in the first source register. If the values are
+ * not equal, then no memory operation is performed. If the values are equal,
+ * the 8-byte quantity from the second source register is written into memory
+ * at the address held in the first source register. In either case, the result
+ * of the instruction is the value read from memory. The compare and write to
+ * memory are atomic and thus can be used for synchronization purposes. This
+ * instruction only operates for addresses aligned to a 8-byte boundary.
+ * Unaligned memory access causes an Unaligned Data Reference interrupt.
+ *
+ * Functional Description (64-bit)
+ *       uint64_t memVal = memoryReadDoubleWord (rf[SrcA]);
+ *       rf[Dest] = memVal;
+ *       if (memVal == SPR[CmpValueSPR])
+ *           memoryWriteDoubleWord (rf[SrcA], rf[SrcB]);
+ *
+ * Functional Description (32-bit)
+ *       uint64_t memVal = signExtend32 (memoryReadWord (rf[SrcA]));
+ *       rf[Dest] = memVal;
+ *       if (memVal == signExtend32 (SPR[CmpValueSPR]))
+ *           memoryWriteWord (rf[SrcA], rf[SrcB]);
+ *
+ *
+ * This function also processes exch and exch4 which need not process SPR.
+ */
+static void do_exch(CPUTLGState *env, bool quad, bool cmp)
+{
+    target_ulong addr;
+    target_long val, sprval;
+
+    start_exclusive();
+
+    addr = env->atomic_srca;
+    if (quad ? get_user_s64(val, addr) : get_user_s32(val, addr)) {
+        goto sigsegv_maperr;
+    }
+
+    if (cmp) {
+        if (quad) {
+            sprval = env->spregs[TILEGX_SPR_CMPEXCH];
+        } else {
+            sprval = sextract64(env->spregs[TILEGX_SPR_CMPEXCH], 0, 32);
+        }
+    }
+
+    if (!cmp || val == sprval) {
+        target_long valb = env->atomic_srcb;
+        if (quad ? put_user_u64(valb, addr) : put_user_u32(valb, addr)) {
+            goto sigsegv_maperr;
+        }
+    }
+
+    set_regval(env, env->atomic_dstr, val);
+    end_exclusive();
+    return;
+
+ sigsegv_maperr:
+    end_exclusive();
+    gen_sigsegv_maperr(env, addr);
+}
+
+static void do_fetch(CPUTLGState *env, int trapnr, bool quad)
+{
+    int8_t write = 1;
+    target_ulong addr;
+    target_long val, valb;
+
+    start_exclusive();
+
+    addr = env->atomic_srca;
+    valb = env->atomic_srcb;
+    if (quad ? get_user_s64(val, addr) : get_user_s32(val, addr)) {
+        goto sigsegv_maperr;
+    }
+
+    switch (trapnr) {
+    case TILEGX_EXCP_OPCODE_FETCHADD:
+    case TILEGX_EXCP_OPCODE_FETCHADD4:
+        valb += val;
+        break;
+    case TILEGX_EXCP_OPCODE_FETCHADDGEZ:
+        valb += val;
+        if (valb < 0) {
+            write = 0;
+        }
+        break;
+    case TILEGX_EXCP_OPCODE_FETCHADDGEZ4:
+        valb += val;
+        if ((int32_t)valb < 0) {
+            write = 0;
+        }
+        break;
+    case TILEGX_EXCP_OPCODE_FETCHAND:
+    case TILEGX_EXCP_OPCODE_FETCHAND4:
+        valb &= val;
+        break;
+    case TILEGX_EXCP_OPCODE_FETCHOR:
+    case TILEGX_EXCP_OPCODE_FETCHOR4:
+        valb |= val;
+        break;
+    default:
+        g_assert_not_reached();
+    }
+
+    if (write) {
+        if (quad ? put_user_u64(valb, addr) : put_user_u32(valb, addr)) {
+            goto sigsegv_maperr;
+        }
+    }
+
+    set_regval(env, env->atomic_dstr, val);
+    end_exclusive();
+    return;
+
+ sigsegv_maperr:
+    end_exclusive();
+    gen_sigsegv_maperr(env, addr);
+}
+
 void cpu_loop(CPUTLGState *env)
 {
     CPUState *cs = CPU(tilegx_env_get_cpu(env));
@@ -3456,6 +3598,30 @@ void cpu_loop(CPUTLGState *env)
                                                       ? - env->regs[TILEGX_R_RE]
                                                       : 0;
             break;
+        case TILEGX_EXCP_OPCODE_EXCH:
+            do_exch(env, true, false);
+            break;
+        case TILEGX_EXCP_OPCODE_EXCH4:
+            do_exch(env, false, false);
+            break;
+        case TILEGX_EXCP_OPCODE_CMPEXCH:
+            do_exch(env, true, true);
+            break;
+        case TILEGX_EXCP_OPCODE_CMPEXCH4:
+            do_exch(env, false, true);
+            break;
+        case TILEGX_EXCP_OPCODE_FETCHADD:
+        case TILEGX_EXCP_OPCODE_FETCHADDGEZ:
+        case TILEGX_EXCP_OPCODE_FETCHAND:
+        case TILEGX_EXCP_OPCODE_FETCHOR:
+            do_fetch(env, trapnr, true);
+            break;
+        case TILEGX_EXCP_OPCODE_FETCHADD4:
+        case TILEGX_EXCP_OPCODE_FETCHADDGEZ4:
+        case TILEGX_EXCP_OPCODE_FETCHAND4:
+        case TILEGX_EXCP_OPCODE_FETCHOR4:
+            do_fetch(env, trapnr, false);
+            break;
         case TILEGX_EXCP_REG_IDN_ACCESS:
         case TILEGX_EXCP_REG_UDN_ACCESS:
             gen_sigill_reg(env);

target-tilegx/cpu.h

@@ -87,7 +87,9 @@ typedef struct CPUTLGState {
     uint64_t pc;                       /* Current pc */
 
 #if defined(CONFIG_USER_ONLY)
-    uint32_t excparam;                 /* exception parameter */
+    uint64_t atomic_srca;              /* Arguments to atomic "exceptions" */
+    uint64_t atomic_srcb;
+    uint32_t atomic_dstr;
     uint64_t excaddr;                  /* exception address */
 #endif
 

target-tilegx/translate.c

@@ -61,6 +61,7 @@ typedef struct {
     int num_wb;
     int mmuidx;
     bool exit_tb;
+    TileExcp atomic_excp;
 
     struct {
         TCGCond cond;    /* branch condition */
@@ -180,6 +181,32 @@ static void gen_saturate_op(TCGv tdest, TCGv tsrca, TCGv tsrcb,
     tcg_temp_free(t0);
 }
 
+static void gen_atomic_excp(DisasContext *dc, unsigned dest, TCGv tdest,
+                            TCGv tsrca, TCGv tsrcb, TileExcp excp)
+{
+#ifdef CONFIG_USER_ONLY
+    TCGv_i32 t;
+
+    tcg_gen_st_tl(tsrca, cpu_env, offsetof(CPUTLGState, atomic_srca));
+    tcg_gen_st_tl(tsrcb, cpu_env, offsetof(CPUTLGState, atomic_srcb));
+    t = tcg_const_i32(dest);
+    tcg_gen_st_i32(t, cpu_env, offsetof(CPUTLGState, atomic_dstr));
+    tcg_temp_free_i32(t);
+
+    /* We're going to write the real result in the exception.  But in
+       the meantime we've already created a writeback register, and
+       we don't want that to remain uninitialized.  */
+    tcg_gen_movi_tl(tdest, 0);
+
+    /* Note that we need to delay issuing the exception that implements
+       the atomic operation until after writing back the results of the
+       instruction occupying the X0 pipe.  */
+    dc->atomic_excp = excp;
+#else
+    gen_exception(dc, TILEGX_EXCP_OPCODE_UNIMPLEMENTED);
+#endif
+}
+
 /* Shift the 128-bit value TSRCA:TSRCD right by the number of bytes
    specified by the bottom 3 bits of TSRCB, and set TDEST to the
    low 64 bits of the resulting value.  */
@@ -591,8 +618,15 @@ static TileExcp gen_rrr_opcode(DisasContext *dc, unsigned opext,
         mnemonic = "cmpeq";
         break;
     case OE_RRR(CMPEXCH4, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_CMPEXCH4);
+        mnemonic = "cmpexch4";
+        break;
     case OE_RRR(CMPEXCH, 0, X1):
-        return TILEGX_EXCP_OPCODE_UNIMPLEMENTED;
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_CMPEXCH);
+        mnemonic = "cmpexch";
+        break;
     case OE_RRR(CMPLES, 0, X0):
     case OE_RRR(CMPLES, 0, X1):
     case OE_RRR(CMPLES, 2, Y0):
@@ -658,7 +692,15 @@ static TileExcp gen_rrr_opcode(DisasContext *dc, unsigned opext,
         mnemonic = "dblalign";
         break;
     case OE_RRR(EXCH4, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_EXCH4);
+        mnemonic = "exch4";
+        break;
     case OE_RRR(EXCH, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_EXCH);
+        mnemonic = "exch";
+        break;
     case OE_RRR(FDOUBLE_ADDSUB, 0, X0):
     case OE_RRR(FDOUBLE_ADD_FLAGS, 0, X0):
     case OE_RRR(FDOUBLE_MUL_FLAGS, 0, X0):
@@ -667,14 +709,47 @@ static TileExcp gen_rrr_opcode(DisasContext *dc, unsigned opext,
     case OE_RRR(FDOUBLE_SUB_FLAGS, 0, X0):
     case OE_RRR(FDOUBLE_UNPACK_MAX, 0, X0):
     case OE_RRR(FDOUBLE_UNPACK_MIN, 0, X0):
+        return TILEGX_EXCP_OPCODE_UNIMPLEMENTED;
     case OE_RRR(FETCHADD4, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_FETCHADD4);
+        mnemonic = "fetchadd4";
+        break;
     case OE_RRR(FETCHADDGEZ4, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_FETCHADDGEZ4);
+        mnemonic = "fetchaddgez4";
+        break;
     case OE_RRR(FETCHADDGEZ, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_FETCHADDGEZ);
+        mnemonic = "fetchaddgez";
+        break;
     case OE_RRR(FETCHADD, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_FETCHADD);
+        mnemonic = "fetchadd";
+        break;
     case OE_RRR(FETCHAND4, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_FETCHAND4);
+        mnemonic = "fetchand4";
+        break;
     case OE_RRR(FETCHAND, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_FETCHAND);
+        mnemonic = "fetchand";
+        break;
     case OE_RRR(FETCHOR4, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_FETCHOR4);
+        mnemonic = "fetchor4";
+        break;
     case OE_RRR(FETCHOR, 0, X1):
+        gen_atomic_excp(dc, dest, tdest, tsrca, tsrcb,
+                        TILEGX_EXCP_OPCODE_FETCHOR);
+        mnemonic = "fetchor";
+        break;
     case OE_RRR(FSINGLE_ADD1, 0, X0):
     case OE_RRR(FSINGLE_ADDSUB2, 0, X0):
     case OE_RRR(FSINGLE_MUL1, 0, X0):
@@ -1936,6 +2011,8 @@ static void translate_one_bundle(DisasContext *dc, uint64_t bundle)
         tcg_temp_free_i64(dc->jmp.dest);
         tcg_gen_exit_tb(0);
         dc->exit_tb = true;
+    } else if (dc->atomic_excp != TILEGX_EXCP_NONE) {
+        gen_exception(dc, dc->atomic_excp);
     }
 }
 
@@ -1956,6 +2033,7 @@ static inline void gen_intermediate_code_internal(TileGXCPU *cpu,
     dc->pc = pc_start;
     dc->mmuidx = 0;
     dc->exit_tb = false;
+    dc->atomic_excp = TILEGX_EXCP_NONE;
     dc->jmp.cond = TCG_COND_NEVER;
     TCGV_UNUSED_I64(dc->jmp.dest);
     TCGV_UNUSED_I64(dc->jmp.val1);