Re: [PATCH v9 06/13] debug: add -d tb_stats to control TBStatistics collection:

[Richard Henderson]

On 10/7/19 11:28 AM, Alex Bennée wrote:
> From: "Vanderson M. do Rosario" <address@hidden>
> 
>  -d tb_stats[[,level=(+all+jit+exec+time)][,dump_limit=<number>]]
> 
> "dump_limit" is used to limit the number of dumped TBStats in
> linux-user mode.
> 
> [all+jit+exec+time] control the profilling level used
> by the TBStats. Can be used as follow:
> 
> -d tb_stats
> -d tb_stats,level=jit+time
> -d tb_stats,dump_limit=15
> ...
> 
> Signed-off-by: Vanderson M. do Rosario <address@hidden>
> Message-Id: <address@hidden>
> [AJB: fix authorship, reword title]
> Signed-off-by: Alex Bennée <address@hidden>
> 
> ---
> AJB:
>   - reword title
>   - add stubs for enabling
>   - move things across to tb-stats-flags.h
> ---
>  accel/tcg/tb-stats.c          |  5 +++++
>  include/exec/gen-icount.h     |  1 +
>  include/exec/tb-stats-flags.h | 29 +++++++++++++++++++++++++++++
>  include/exec/tb-stats.h       | 16 +++-------------
>  include/qemu/log.h            |  1 +
>  stubs/Makefile.objs           |  1 +
>  stubs/tb-stats.c              | 27 +++++++++++++++++++++++++++
>  util/log.c                    | 35 +++++++++++++++++++++++++++++++++++
>  8 files changed, 102 insertions(+), 13 deletions(-)
>  create mode 100644 include/exec/tb-stats-flags.h
>  create mode 100644 stubs/tb-stats.c
> 
> diff --git a/accel/tcg/tb-stats.c b/accel/tcg/tb-stats.c
> index f431159fd2..1c66e03979 100644
> --- a/accel/tcg/tb-stats.c
> +++ b/accel/tcg/tb-stats.c
> @@ -193,3 +193,8 @@ uint32_t get_default_tbstats_flag(void)
>  {
>      return default_tbstats_flag;
>  }
> +
> +void set_default_tbstats_flag(uint32_t flags)
> +{
> +    default_tbstats_flag = flags;
> +}
> diff --git a/include/exec/gen-icount.h b/include/exec/gen-icount.h
> index be006383b9..3987adfb0e 100644
> --- a/include/exec/gen-icount.h
> +++ b/include/exec/gen-icount.h
> @@ -2,6 +2,7 @@
>  #define GEN_ICOUNT_H
>  
>  #include "qemu/timer.h"
> +#include "tb-stats-flags.h"
>  
>  /* Helpers for instruction counting code generation.  */
>  
> diff --git a/include/exec/tb-stats-flags.h b/include/exec/tb-stats-flags.h
> new file mode 100644
> index 0000000000..8455073048
> --- /dev/null
> +++ b/include/exec/tb-stats-flags.h
> @@ -0,0 +1,29 @@
> +/*
> + * QEMU System Emulator, Code Quality Monitor System
> + *
> + * We define the flags and control bits here to avoid complications of
> + * including TCG/CPU information in common code.
> + *
> + * Copyright (c) 2019 Vanderson M. do Rosario <address@hidden>
> + *
> + * SPDX-License-Identifier: GPL-2.0-or-later
> + */
> +#ifndef TB_STATS_FLAGS
> +#define TB_STATS_FLAGS
> +
> +#define TB_NOTHING    (1 << 0)

Repeating my question about TB_NOTHING -- what is it?

> +#define TB_EXEC_STATS (1 << 1)
> +#define TB_JIT_STATS  (1 << 2)
> +#define TB_JIT_TIME   (1 << 3)
> +
> +/* TBStatistic collection controls */
> +void enable_collect_tb_stats(void);
> +void disable_collect_tb_stats(void);
> +void pause_collect_tb_stats(void);
> +bool tb_stats_collection_enabled(void);
> +bool tb_stats_collection_paused(void);
> +
> +uint32_t get_default_tbstats_flag(void);
> +void set_default_tbstats_flag(uint32_t);

Is a get/set really better than an exported variable?

Should we have created this header in the first place,
rather than moving stuff here in patch 6?


> +        } else if (g_str_has_prefix(*tmp, "tb_stats")) {
> +            mask |= CPU_LOG_TB_STATS;
> +            set_default_tbstats_flag(TB_JIT_STATS | TB_EXEC_STATS | 
> TB_JIT_TIME);

Surely TB_ALL_STATS?

> +                } else if (g_str_equal(*level_tmp, "all")) {
> +                    flags |= TB_JIT_STATS | TB_EXEC_STATS | TB_JIT_TIME;

Likewise.


r~