Re: [PATCH v3 01/20] accel/tcg: restrict assert on icount_enabled to qemu-system

[Richard Henderson]

On 8/7/24 10:06, Deepak Gupta wrote:
> commit 16ad9788 [1] restricted icount to qemu-system only. Although
> assert in `cpu_loop_exec_tb` is on `icount_enabled()` which is 0 when
> its qemu-user and debug build starts asserting.
> Move assert for qemu-system.
>
> [1] - https://lists.gnu.org/archive/html/qemu-riscv/2024-01/msg00608.html
>
> Signed-off-by: Deepak Gupta <debug@rivosinc.com>
> ---
>   accel/tcg/cpu-exec.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/accel/tcg/cpu-exec.c b/accel/tcg/cpu-exec.c
> index 245fd6327d..8cc2a6104f 100644
> --- a/accel/tcg/cpu-exec.c
> +++ b/accel/tcg/cpu-exec.c
> @@ -927,9 +927,9 @@ static inline void cpu_loop_exec_tb(CPUState *cpu, 
> TranslationBlock *tb,
>           return;
>       }
>   
> +#ifndef CONFIG_USER_ONLY
>       /* Instruction counter expired.  */
>       assert(icount_enabled());
> -#ifndef CONFIG_USER_ONLY
>       /* Ensure global icount has gone forward */
>       icount_update(cpu);
>       /* Refill decrementer and continue execution.  */

No, this is a real bug.

Just above we handled

  (1) exit for tcg (non-)chaining (!= TB_EXIT_REQUESTED),
  (2) exit for exception/interrupt (cpu_loop_exit_requested).

The only thing that is left is exit for icount expired.
And for that we *must* have icount enabled.

How did you encounter this?


r~