The temp that gets assigned to clean_addr has been allocated with
new_tmp_a64, which means that it will be freed at the end of the
instruction.  Freeing it earlier leads to assertion failure.

The loop creates a complication, in which we allocate a new local
temp, which does need freeing, and the final code path is shared
between the loop and non-loop.

Fix this complication by adding new_tmp_a64_local so that the new
local temp is freed at the end, and can be treated exactly like
the non-loop path.

Fixes: bba87d0a0f4
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/translate-a64.h | 1 +
 target/arm/translate-a64.c | 6 ++++++
 target/arm/translate-sve.c | 8 ++------
 3 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/target/arm/translate-a64.h b/target/arm/translate-a64.h
index 49e4865918..647f0c74f6 100644
--- a/target/arm/translate-a64.h
+++ b/target/arm/translate-a64.h
@@ -30,6 +30,7 @@ void unallocated_encoding(DisasContext *s);
     } while (0)
 
 TCGv_i64 new_tmp_a64(DisasContext *s);
+TCGv_i64 new_tmp_a64_local(DisasContext *s);
 TCGv_i64 new_tmp_a64_zero(DisasContext *s);
 TCGv_i64 cpu_reg(DisasContext *s, int reg);
 TCGv_i64 cpu_reg_sp(DisasContext *s, int reg);
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index 73d753f11f..8c0764957c 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -461,6 +461,12 @@ TCGv_i64 new_tmp_a64(DisasContext *s)
     return s->tmp_a64[s->tmp_a64_count++] = tcg_temp_new_i64();
 }
 
+TCGv_i64 new_tmp_a64_local(DisasContext *s)
+{
+    assert(s->tmp_a64_count < TMP_A64_MAX);
+    return s->tmp_a64[s->tmp_a64_count++] = tcg_temp_local_new_i64();
+}
+
 TCGv_i64 new_tmp_a64_zero(DisasContext *s)
 {
     TCGv_i64 t = new_tmp_a64(s);
diff --git a/target/arm/translate-sve.c b/target/arm/translate-sve.c
index f318ca265f..08f0fd15b2 100644
--- a/target/arm/translate-sve.c
+++ b/target/arm/translate-sve.c
@@ -4372,9 +4372,8 @@ static void do_ldr(DisasContext *s, uint32_t vofs, int len, int rn, int imm)
 
         /* Copy the clean address into a local temp, live across the loop. */
         t0 = clean_addr;
-        clean_addr = tcg_temp_local_new_i64();
+        clean_addr = new_tmp_a64_local(s);
         tcg_gen_mov_i64(clean_addr, t0);
-        tcg_temp_free_i64(t0);
 
         gen_set_label(loop);
 
@@ -4422,7 +4421,6 @@ static void do_ldr(DisasContext *s, uint32_t vofs, int len, int rn, int imm)
         tcg_gen_st_i64(t0, cpu_env, vofs + len_align);
         tcg_temp_free_i64(t0);
     }
-    tcg_temp_free_i64(clean_addr);
 }
 
 /* Similarly for stores.  */
@@ -4463,9 +4461,8 @@ static void do_str(DisasContext *s, uint32_t vofs, int len, int rn, int imm)
 
         /* Copy the clean address into a local temp, live across the loop. */
         t0 = clean_addr;
-        clean_addr = tcg_temp_local_new_i64();
+        clean_addr = new_tmp_a64_local(s);
         tcg_gen_mov_i64(clean_addr, t0);
-        tcg_temp_free_i64(t0);
 
         gen_set_label(loop);
 
@@ -4509,7 +4506,6 @@ static void do_str(DisasContext *s, uint32_t vofs, int len, int rn, int imm)
         }
         tcg_temp_free_i64(t0);
     }
-    tcg_temp_free_i64(clean_addr);
 }
 
 static bool trans_LDR_zri(DisasContext *s, arg_rri *a)
-- 
2.25.1

On 7/2/20 7:56 PM, Richard Henderson wrote:
> The temp that gets assigned to clean_addr has been allocated with
> new_tmp_a64, which means that it will be freed at the end of the
> instruction.  Freeing it earlier leads to assertion failure.
> 
> The loop creates a complication, in which we allocate a new local
> temp, which does need freeing, and the final code path is shared
> between the loop and non-loop.
> 
> Fix this complication by adding new_tmp_a64_local so that the new
> local temp is freed at the end, and can be treated exactly like
> the non-loop path.
> 
> Fixes: bba87d0a0f4
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  target/arm/translate-a64.h | 1 +
>  target/arm/translate-a64.c | 6 ++++++
>  target/arm/translate-sve.c | 8 ++------
>  3 files changed, 9 insertions(+), 6 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

On Thu, 2 Jul 2020 at 18:56, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> The temp that gets assigned to clean_addr has been allocated with
> new_tmp_a64, which means that it will be freed at the end of the
> instruction.  Freeing it earlier leads to assertion failure.
>
> The loop creates a complication, in which we allocate a new local
> temp, which does need freeing, and the final code path is shared
> between the loop and non-loop.
>
> Fix this complication by adding new_tmp_a64_local so that the new
> local temp is freed at the end, and can be treated exactly like
> the non-loop path.
>
> Fixes: bba87d0a0f4
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>


Applied to target-arm.next, thanks.

-- PMM