RISC-V: Adding XTheadSync ISA extension

This patch adds support for the XTheadSync ISA extension.
The patch uses the T-Head specific decoder and translation.

The implementation introduces a helper to execute synchronization tasks:
helper_tlb_flush_all() performs a synchronized TLB flush on all CPUs.

Co-developed-by: LIU Zhiwei <zhiwei_liu@linux.alibaba.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Christoph Müllner <christoph.muellner@vrull.eu>
Message-Id: <20230131202013.2541053-3-christoph.muellner@vrull.eu>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>

target/riscv/cpu.c

@@ -110,6 +110,7 @@ static const struct isa_ext_data isa_edata_arr[] = {
     ISA_EXT_DATA_ENTRY(svnapot, true, PRIV_VERSION_1_12_0, ext_svnapot),
     ISA_EXT_DATA_ENTRY(svpbmt, true, PRIV_VERSION_1_12_0, ext_svpbmt),
     ISA_EXT_DATA_ENTRY(xtheadcmo, true, PRIV_VERSION_1_11_0, ext_xtheadcmo),
+    ISA_EXT_DATA_ENTRY(xtheadsync, true, PRIV_VERSION_1_11_0, ext_xtheadsync),
     ISA_EXT_DATA_ENTRY(xventanacondops, true, PRIV_VERSION_1_12_0, ext_XVentanaCondOps),
 };
 
@@ -1090,6 +1091,7 @@ static Property riscv_cpu_extensions[] = {
 
     /* Vendor-specific custom extensions */
     DEFINE_PROP_BOOL("xtheadcmo", RISCVCPU, cfg.ext_xtheadcmo, false),
+    DEFINE_PROP_BOOL("xtheadsync", RISCVCPU, cfg.ext_xtheadsync, false),
     DEFINE_PROP_BOOL("xventanacondops", RISCVCPU, cfg.ext_XVentanaCondOps, false),
 
     /* These are experimental so mark with 'x-' */

target/riscv/cpu.h

@@ -474,6 +474,7 @@ struct RISCVCPUConfig {
 
     /* Vendor-specific custom extensions */
     bool ext_xtheadcmo;
+    bool ext_xtheadsync;
     bool ext_XVentanaCondOps;
 
     uint8_t pmu_num;

target/riscv/helper.h

@@ -109,6 +109,7 @@ DEF_HELPER_1(sret, tl, env)
 DEF_HELPER_1(mret, tl, env)
 DEF_HELPER_1(wfi, void, env)
 DEF_HELPER_1(tlb_flush, void, env)
+DEF_HELPER_1(tlb_flush_all, void, env)
 /* Native Debug */
 DEF_HELPER_1(itrigger_match, void, env)
 #endif

target/riscv/insn_trans/trans_xthead.c.inc

@@ -22,6 +22,12 @@
     }                                            \
 } while (0)
 
+#define REQUIRE_XTHEADSYNC(ctx) do {             \
+    if (!ctx->cfg_ptr->ext_xtheadsync) {         \
+        return false;                            \
+    }                                            \
+} while (0)
+
 /* XTheadCmo */
 
 static inline int priv_level(DisasContext *ctx)
@@ -79,3 +85,82 @@ NOP_PRIVCHECK(th_icache_iva, REQUIRE_XTHEADCMO, REQUIRE_PRIV_MSU)
 NOP_PRIVCHECK(th_l2cache_call, REQUIRE_XTHEADCMO, REQUIRE_PRIV_MS)
 NOP_PRIVCHECK(th_l2cache_ciall, REQUIRE_XTHEADCMO, REQUIRE_PRIV_MS)
 NOP_PRIVCHECK(th_l2cache_iall, REQUIRE_XTHEADCMO, REQUIRE_PRIV_MS)
+
+/* XTheadSync */
+
+static bool trans_th_sfence_vmas(DisasContext *ctx, arg_th_sfence_vmas *a)
+{
+    (void) a;
+    REQUIRE_XTHEADSYNC(ctx);
+
+#ifndef CONFIG_USER_ONLY
+    REQUIRE_PRIV_MS(ctx);
+    gen_helper_tlb_flush_all(cpu_env);
+    return true;
+#else
+    return false;
+#endif
+}
+
+#ifndef CONFIG_USER_ONLY
+static void gen_th_sync_local(DisasContext *ctx)
+{
+    /*
+     * Emulate out-of-order barriers with pipeline flush
+     * by exiting the translation block.
+     */
+    gen_set_pc_imm(ctx, ctx->pc_succ_insn);
+    tcg_gen_exit_tb(NULL, 0);
+    ctx->base.is_jmp = DISAS_NORETURN;
+}
+#endif
+
+static bool trans_th_sync(DisasContext *ctx, arg_th_sync *a)
+{
+    (void) a;
+    REQUIRE_XTHEADSYNC(ctx);
+
+#ifndef CONFIG_USER_ONLY
+    REQUIRE_PRIV_MSU(ctx);
+
+    /*
+     * th.sync is an out-of-order barrier.
+     */
+    gen_th_sync_local(ctx);
+
+    return true;
+#else
+    return false;
+#endif
+}
+
+static bool trans_th_sync_i(DisasContext *ctx, arg_th_sync_i *a)
+{
+    (void) a;
+    REQUIRE_XTHEADSYNC(ctx);
+
+#ifndef CONFIG_USER_ONLY
+    REQUIRE_PRIV_MSU(ctx);
+
+    /*
+     * th.sync.i is th.sync plus pipeline flush.
+     */
+    gen_th_sync_local(ctx);
+
+    return true;
+#else
+    return false;
+#endif
+}
+
+static bool trans_th_sync_is(DisasContext *ctx, arg_th_sync_is *a)
+{
+    /* This instruction has the same behaviour like th.sync.i. */
+    return trans_th_sync_i(ctx, a);
+}
+
+static bool trans_th_sync_s(DisasContext *ctx, arg_th_sync_s *a)
+{
+    /* This instruction has the same behaviour like th.sync. */
+    return trans_th_sync(ctx, a);
+}

target/riscv/op_helper.c

@@ -258,6 +258,12 @@ void helper_tlb_flush(CPURISCVState *env)
     }
 }
 
+void helper_tlb_flush_all(CPURISCVState *env)
+{
+    CPUState *cs = env_cpu(env);
+    tlb_flush_all_cpus_synced(cs);
+}
+
 void helper_hyp_tlb_flush(CPURISCVState *env)
 {
     CPUState *cs = env_cpu(env);

target/riscv/translate.c

@@ -132,7 +132,7 @@ static bool always_true_p(DisasContext *ctx  __attribute__((__unused__)))
 
 static bool has_xthead_p(DisasContext *ctx  __attribute__((__unused__)))
 {
-    return ctx->cfg_ptr->ext_xtheadcmo;
+    return ctx->cfg_ptr->ext_xtheadcmo || ctx->cfg_ptr->ext_xtheadsync;
 }
 
 #define MATERIALISE_EXT_PREDICATE(ext)  \

target/riscv/xthead.decode

@@ -10,9 +10,11 @@
 
 # Fields:
 %rs1       15:5
+%rs2       20:5
 
 # Formats
 @sfence_vm  ....... ..... .....   ... ..... ....... %rs1
+@rs2_s      ....... ..... ..... ... ..... .......   %rs2 %rs1
 
 # XTheadCmo
 th_dcache_call   0000000 00001 00000 000 00000 0001011
@@ -36,3 +38,10 @@ th_icache_iva    0000001 10000 ..... 000 00000 0001011 @sfence_vm
 th_l2cache_call  0000000 10101 00000 000 00000 0001011
 th_l2cache_ciall 0000000 10111 00000 000 00000 0001011
 th_l2cache_iall  0000000 10110 00000 000 00000 0001011
+
+# XTheadSync
+th_sfence_vmas   0000010 ..... ..... 000 00000 0001011 @rs2_s
+th_sync          0000000 11000 00000 000 00000 0001011
+th_sync_i        0000000 11010 00000 000 00000 0001011
+th_sync_is       0000000 11011 00000 000 00000 0001011
+th_sync_s        0000000 11001 00000 000 00000 0001011