qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Qemu-devel] [PATCH 03/11] target/sh4: Handle user-space atomics


From: Richard Henderson
Subject: [Qemu-devel] [PATCH 03/11] target/sh4: Handle user-space atomics
Date: Wed, 5 Jul 2017 14:23:53 -1000

For uniprocessors, SH4 uses optimistic restartable atomic sequences.
Upon an interrupt, a real kernel would simply notice magic values in
the registers and reset the PC to the start of the sequence.

For QEMU, we cannot do this in quite the same way.  Instead, we notice
the normal start of such a sequence (mov #-x,r15), and start a new TB
that can be executed under cpu_exec_step_atomic.

Reported-by: Bruno Haible  <address@hidden>
LP: https://bugs.launchpad.net/bugs/1701971
Signed-off-by: Richard Henderson <address@hidden>
---
 target/sh4/cpu.h       |  21 ++++++--
 target/sh4/helper.h    |   1 +
 target/sh4/op_helper.c |   6 +++
 target/sh4/translate.c | 137 +++++++++++++++++++++++++++++++++++++++++++------
 4 files changed, 147 insertions(+), 18 deletions(-)

diff --git a/target/sh4/cpu.h b/target/sh4/cpu.h
index b15116e..0a08b12 100644
--- a/target/sh4/cpu.h
+++ b/target/sh4/cpu.h
@@ -96,6 +96,12 @@
 #define DELAY_SLOT_CONDITIONAL (1 << 1)
 #define DELAY_SLOT_RTE         (1 << 2)
 
+#define TB_FLAG_PENDING_MOVCA  (1 << 3)
+
+#define GUSA_SHIFT             4
+#define GUSA_EXCLUSIVE         (1 << 12)
+#define GUSA_MASK              ((0xff << GUSA_SHIFT) | GUSA_EXCLUSIVE)
+
 typedef struct tlb_t {
     uint32_t vpn;              /* virtual page number */
     uint32_t ppn;              /* physical page number */
@@ -367,7 +373,11 @@ static inline int cpu_ptel_pr (uint32_t ptel)
 #define PTEA_TC        (1 << 3)
 #define cpu_ptea_tc(ptea) (((ptea) & PTEA_TC) >> 3)
 
-#define TB_FLAG_PENDING_MOVCA  (1 << 4)
+#ifdef CONFIG_USER_ONLY
+#define TB_FLAG_ENVFLAGS_MASK  (DELAY_SLOT_MASK | GUSA_MASK)
+#else
+#define TB_FLAG_ENVFLAGS_MASK  DELAY_SLOT_MASK
+#endif
 
 static inline target_ulong cpu_read_sr(CPUSH4State *env)
 {
@@ -388,12 +398,17 @@ static inline void cpu_get_tb_cpu_state(CPUSH4State *env, 
target_ulong *pc,
                                         target_ulong *cs_base, uint32_t *flags)
 {
     *pc = env->pc;
+#ifdef CONFIG_USER_ONLY
+    /* For a gUSA region, notice the end of the region.  */
+    *cs_base = env->flags & GUSA_MASK ? env->gregs[0] : 0;
+#else
     *cs_base = 0;
-    *flags = (env->flags & DELAY_SLOT_MASK)                    /* Bits  0- 2 */
+#endif
+    *flags = env->flags /* TB_FLAG_ENVFLAGS_MASK: bits 0-2, 4-12 */
             | (env->fpscr & (FPSCR_FR | FPSCR_SZ | FPSCR_PR))  /* Bits 19-21 */
             | (env->sr & ((1u << SR_MD) | (1u << SR_RB)))      /* Bits 29-30 */
             | (env->sr & (1u << SR_FD))                        /* Bit 15 */
-            | (env->movcal_backup ? TB_FLAG_PENDING_MOVCA : 0); /* Bit 4 */
+            | (env->movcal_backup ? TB_FLAG_PENDING_MOVCA : 0); /* Bit 3 */
 }
 
 #endif /* SH4_CPU_H */
diff --git a/target/sh4/helper.h b/target/sh4/helper.h
index dce859c..efbb560 100644
--- a/target/sh4/helper.h
+++ b/target/sh4/helper.h
@@ -6,6 +6,7 @@ DEF_HELPER_1(raise_slot_fpu_disable, noreturn, env)
 DEF_HELPER_1(debug, noreturn, env)
 DEF_HELPER_1(sleep, noreturn, env)
 DEF_HELPER_2(trapa, noreturn, env, i32)
+DEF_HELPER_1(exclusive, noreturn, env)
 
 DEF_HELPER_3(movcal, void, env, i32, i32)
 DEF_HELPER_1(discard_movcal_backup, void, env)
diff --git a/target/sh4/op_helper.c b/target/sh4/op_helper.c
index 528a40a..3139ad2 100644
--- a/target/sh4/op_helper.c
+++ b/target/sh4/op_helper.c
@@ -115,6 +115,12 @@ void helper_trapa(CPUSH4State *env, uint32_t tra)
     raise_exception(env, 0x160, 0);
 }
 
+void helper_exclusive(CPUSH4State *env)
+{
+    /* We do not want cpu_restore_state to run.  */
+    cpu_loop_exit_atomic(ENV_GET_CPU(env), 0);
+}
+
 void helper_movcal(CPUSH4State *env, uint32_t address, uint32_t value)
 {
     if (cpu_sh4_is_cached (env, address))
diff --git a/target/sh4/translate.c b/target/sh4/translate.c
index e1661e9..02c6efc 100644
--- a/target/sh4/translate.c
+++ b/target/sh4/translate.c
@@ -225,7 +225,7 @@ static inline void gen_save_cpu_state(DisasContext *ctx, 
bool save_pc)
     if (ctx->delayed_pc != (uint32_t) -1) {
         tcg_gen_movi_i32(cpu_delayed_pc, ctx->delayed_pc);
     }
-    if ((ctx->tbflags & DELAY_SLOT_MASK) != ctx->envflags) {
+    if ((ctx->tbflags & TB_FLAG_ENVFLAGS_MASK) != ctx->envflags) {
         tcg_gen_movi_i32(cpu_flags, ctx->envflags);
     }
 }
@@ -239,7 +239,7 @@ static inline bool use_goto_tb(DisasContext *ctx, 
target_ulong dest)
 #ifndef CONFIG_USER_ONLY
     return (ctx->tb->pc & TARGET_PAGE_MASK) == (dest & TARGET_PAGE_MASK);
 #else
-    return true;
+    return (ctx->tbflags & GUSA_EXCLUSIVE) == 0;
 #endif
 }
 
@@ -260,16 +260,17 @@ static void gen_goto_tb(DisasContext *ctx, int n, 
target_ulong dest)
 
 static void gen_jump(DisasContext * ctx)
 {
-    if (ctx->delayed_pc == (uint32_t) - 1) {
-       /* Target is not statically known, it comes necessarily from a
-          delayed jump as immediate jump are conditinal jumps */
-       tcg_gen_mov_i32(cpu_pc, cpu_delayed_pc);
+    if (ctx->delayed_pc == -1) {
+        /* Target is not statically known, it comes necessarily from a
+           delayed jump as immediate jump are conditinal jumps */
+        tcg_gen_mov_i32(cpu_pc, cpu_delayed_pc);
         tcg_gen_discard_i32(cpu_delayed_pc);
-       if (ctx->singlestep_enabled)
+        if (ctx->singlestep_enabled) {
             gen_helper_debug(cpu_env);
-       tcg_gen_exit_tb(0);
+        }
+        tcg_gen_exit_tb(0);
     } else {
-       gen_goto_tb(ctx, 0, ctx->delayed_pc);
+        gen_goto_tb(ctx, 0, ctx->delayed_pc);
     }
 }
 
@@ -278,6 +279,30 @@ static void gen_conditional_jump(DisasContext * ctx,
                                 target_ulong ift, target_ulong ifnott)
 {
     TCGLabel *l1 = gen_new_label();
+
+#ifdef CONFIG_USER_ONLY
+    if (ctx->tbflags & GUSA_EXCLUSIVE) {
+        /* When in an exclusive region, we must continue to the end.
+           Therefore, exit the region on a taken branch, but otherwise
+           fall through to the next instruction.  */
+        uint32_t taken;
+        TCGCond cond;
+
+        if (ift == ctx->pc + 2) {
+            taken = ifnott;
+            cond = TCG_COND_NE;
+        } else {
+            taken = ift;
+            cond = TCG_COND_EQ;
+        }
+        tcg_gen_brcondi_i32(cond, cpu_sr_t, 0, l1);
+        tcg_gen_movi_i32(cpu_flags, ctx->envflags & ~GUSA_MASK);
+        gen_goto_tb(ctx, 0, taken);
+        gen_set_label(l1);
+        return;
+    }
+#endif
+
     gen_save_cpu_state(ctx, false);
     tcg_gen_brcondi_i32(TCG_COND_NE, cpu_sr_t, 0, l1);
     gen_goto_tb(ctx, 0, ifnott);
@@ -289,13 +314,28 @@ static void gen_conditional_jump(DisasContext * ctx,
 /* Delayed conditional jump (bt or bf) */
 static void gen_delayed_conditional_jump(DisasContext * ctx)
 {
-    TCGLabel *l1;
-    TCGv ds;
+    TCGLabel *l1 = gen_new_label();
+    TCGv ds = tcg_temp_new();
 
-    l1 = gen_new_label();
-    ds = tcg_temp_new();
     tcg_gen_mov_i32(ds, cpu_delayed_cond);
     tcg_gen_discard_i32(cpu_delayed_cond);
+
+#ifdef CONFIG_USER_ONLY
+    if (ctx->tbflags & GUSA_EXCLUSIVE) {
+        /* When in an exclusive region, we must continue to the end.
+           Therefore, exit the region on a taken branch, but otherwise
+           fall through to the next instruction.  */
+        tcg_gen_brcondi_i32(TCG_COND_EQ, ds, 0, l1);
+
+        /* Leave the gUSA region.  */
+        tcg_gen_movi_i32(cpu_flags, ctx->envflags & ~GUSA_MASK);
+        gen_jump(ctx);
+
+        gen_set_label(l1);
+        return;
+    }
+#endif
+
     tcg_gen_brcondi_i32(TCG_COND_NE, ds, 0, l1);
     gen_goto_tb(ctx, 1, ctx->pc + 2);
     gen_set_label(l1);
@@ -480,6 +520,15 @@ static void _decode_opc(DisasContext * ctx)
        }
        return;
     case 0xe000:               /* mov #imm,Rn */
+#ifdef CONFIG_USER_ONLY
+        /* Detect the start of a gUSA region.  If so, update envflags
+           and end the TB.  This will allow us to see the end of the
+           region (stored in R0) in the next TB.  */
+        if (B11_8 == 15 && B7_0s < 0) {
+            ctx->envflags = deposit32(ctx->envflags, GUSA_SHIFT, 8, B7_0s);
+            ctx->bstate = BS_STOP;
+        }
+#endif
        tcg_gen_movi_i32(REG(B11_8), B7_0s);
        return;
     case 0x9000:               /* mov.w @(disp,PC),Rn */
@@ -1822,6 +1871,20 @@ static void decode_opc(DisasContext * ctx)
     if (old_flags & DELAY_SLOT_MASK) {
         /* go out of the delay slot */
         ctx->envflags &= ~DELAY_SLOT_MASK;
+
+#ifdef CONFIG_USER_ONLY
+        /* When in an exclusive region, we must continue to the end
+           for conditional branches.  */
+        if (ctx->tbflags & GUSA_EXCLUSIVE
+            && old_flags & DELAY_SLOT_CONDITIONAL) {
+            gen_delayed_conditional_jump(ctx);
+            return;
+        }
+        /* Otherwise this is probably an invalid gUSA region.
+           Drop the GUSA bits so the next TB doesn't see them.  */
+        ctx->envflags &= ~GUSA_MASK;
+#endif
+
         tcg_gen_movi_i32(cpu_flags, ctx->envflags);
         ctx->bstate = BS_BRANCH;
         if (old_flags & DELAY_SLOT_CONDITIONAL) {
@@ -1829,10 +1892,35 @@ static void decode_opc(DisasContext * ctx)
         } else {
             gen_jump(ctx);
        }
-
     }
 }
 
+#ifdef CONFIG_USER_ONLY
+static int decode_gusa(DisasContext *ctx)
+{
+    uint32_t pc = ctx->pc;
+    uint32_t pc_end = ctx->tb->cs_base;
+
+    qemu_log_mask(LOG_UNIMP, "Unrecognized gUSA sequence %08x-%08x\n",
+                  pc, pc_end);
+
+    /* Restart with the EXCLUSIVE bit set, within a TB run via
+       cpu_exec_step_atomic holding the exclusive lock.  */
+    tcg_gen_insn_start(pc, ctx->envflags);
+    ctx->envflags |= GUSA_EXCLUSIVE;
+    gen_save_cpu_state(ctx, false);
+    gen_helper_exclusive(cpu_env);
+    ctx->bstate = BS_EXCP;
+
+    /* We're not executing an instruction, but we must report one for the
+       purposes of accounting within the TB.  At which point we might as
+       well report the entire region so that it's immediately available
+       in the disassembly dump.  */
+    ctx->pc = pc_end;
+    return 1;
+}
+#endif
+
 void gen_intermediate_code(CPUSH4State * env, struct TranslationBlock *tb)
 {
     SuperHCPU *cpu = sh_env_get_cpu(env);
@@ -1845,7 +1933,7 @@ void gen_intermediate_code(CPUSH4State * env, struct 
TranslationBlock *tb)
     pc_start = tb->pc;
     ctx.pc = pc_start;
     ctx.tbflags = (uint32_t)tb->flags;
-    ctx.envflags = tb->flags & DELAY_SLOT_MASK;
+    ctx.envflags = tb->flags & TB_FLAG_ENVFLAGS_MASK;
     ctx.bstate = BS_NONE;
     ctx.memidx = (ctx.tbflags & (1u << SR_MD)) == 0 ? 1 : 0;
     /* We don't know if the delayed pc came from a dynamic or static branch,
@@ -1877,6 +1965,17 @@ void gen_intermediate_code(CPUSH4State * env, struct 
TranslationBlock *tb)
     gen_tb_start(tb);
     num_insns = 0;
 
+#ifdef CONFIG_USER_ONLY
+    if (ctx.tbflags & GUSA_EXCLUSIVE) {
+        /* Regardless of single-stepping or the end of the page,
+           we must complete execution of the gUSA region while
+           holding the exclusive lock.  */
+        max_insns = (tb->cs_base - ctx.pc) / 2;
+    } else if (ctx.tbflags & GUSA_MASK) {
+        num_insns = decode_gusa(&ctx);
+    }
+#endif
+
     while (ctx.bstate == BS_NONE
            && num_insns < max_insns
            && !tcg_op_buf_full()) {
@@ -1907,6 +2006,14 @@ void gen_intermediate_code(CPUSH4State * env, struct 
TranslationBlock *tb)
     if (tb->cflags & CF_LAST_IO) {
         gen_io_end();
     }
+
+#ifdef CONFIG_USER_ONLY
+    if ((ctx.tbflags & GUSA_EXCLUSIVE) && ctx.bstate == BS_NONE) {
+        /* Ending the region of exclusivity.  Clear the bits.  */
+        ctx.envflags &= ~GUSA_MASK;
+    }
+#endif
+
     if (cs->singlestep_enabled) {
         gen_save_cpu_state(&ctx, true);
         gen_helper_debug(cpu_env);
-- 
2.9.4




reply via email to

[Prev in Thread] Current Thread [Next in Thread]