Re: [PATCH 1/4] os-linux: add initial support for IO priority hints

[Niklas Cassel <Niklas.Cassel@wdc.com>]

On Thu, Jul 06, 2023 at 07:12:59AM +0900, Damien Le Moal wrote:
> From: Damien Le Moal <damien.lemoal@opensource.wdc.com>
> 
> Add initial support for Linux to allow specifying a hint for any
> priority value. With this change, a priority value becomes the
> combination of a priority class, a priority level and a hint.
> The generic os.h ioprio manipulation macros, as well as the
> os-dragonfly.h ioprio manipulation macros  are modified to ignore this
> hint.
> 
> Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
> ---
>  backend.c         |  5 +++--
>  engines/cmdprio.c |  4 ++--
>  options.c         |  2 +-
>  os/os-dragonfly.h |  4 ++--
>  os/os-linux.h     | 17 +++++++++++++----
>  os/os.h           |  4 ++--
>  6 files changed, 23 insertions(+), 13 deletions(-)
> 
> diff --git a/backend.c b/backend.c
> index d67a4a07..6641441c 100644
> --- a/backend.c
> +++ b/backend.c
> @@ -1791,12 +1791,13 @@ static void *thread_main(void *data)
>  	/* ioprio_set() has to be done before td_io_init() */
>  	if (fio_option_is_set(o, ioprio) ||
>  	    fio_option_is_set(o, ioprio_class)) {
> -		ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
> +		ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class,
> +				 o->ioprio, 0);
>  		if (ret == -1) {
>  			td_verror(td, errno, "ioprio_set");
>  			goto err;
>  		}
> -		td->ioprio = ioprio_value(o->ioprio_class, o->ioprio);
> +		td->ioprio = ioprio_value(o->ioprio_class, o->ioprio, 0);
>  		td->ts.ioprio = td->ioprio;
>  	}
>  
> diff --git a/engines/cmdprio.c b/engines/cmdprio.c
> index 979a81b6..e6ff1fc2 100644
> --- a/engines/cmdprio.c
> +++ b/engines/cmdprio.c
> @@ -342,7 +342,7 @@ static int fio_cmdprio_gen_perc(struct thread_data *td, struct cmdprio *cmdprio)
>  		prio = &cmdprio->perc_entry[ddir];
>  		prio->perc = options->percentage[ddir];
>  		prio->prio = ioprio_value(options->class[ddir],
> -					  options->level[ddir]);
> +					  options->level[ddir], 0);
>  		assign_clat_prio_index(prio, &values[ddir]);
>  
>  		ret = init_ts_clat_prio(ts, ddir, &values[ddir]);
> @@ -400,7 +400,7 @@ static int fio_cmdprio_parse_and_gen_bssplit(struct thread_data *td,
>  			goto err;
>  
>  		implicit_cmdprio = ioprio_value(options->class[ddir],
> -						options->level[ddir]);
> +						options->level[ddir], 0);
>  
>  		ret = fio_cmdprio_generate_bsprio_desc(&cmdprio->bsprio_desc[ddir],
>  						       &parse_res[ddir],
> diff --git a/options.c b/options.c
> index a7c4ef6e..c5343d4b 100644
> --- a/options.c
> +++ b/options.c
> @@ -344,7 +344,7 @@ static int parse_cmdprio_bssplit_entry(struct thread_options *o,
>  	case 4: /* bs/perc/class/level case */
>  		class = min(class, (unsigned int) IOPRIO_MAX_PRIO_CLASS);
>  		level = min(level, (unsigned int) IOPRIO_MAX_PRIO);
> -		entry->prio = ioprio_value(class, level);
> +		entry->prio = ioprio_value(class, level, 0);
>  		break;
>  	default:
>  		log_err("fio: invalid cmdprio_bssplit format\n");
> diff --git a/os/os-dragonfly.h b/os/os-dragonfly.h
> index bde39101..4ce72539 100644
> --- a/os/os-dragonfly.h
> +++ b/os/os-dragonfly.h
> @@ -171,8 +171,8 @@ static inline int fio_getaffinity(int pid, os_cpu_mask_t *mask)
>   * ioprio_set() with 4 arguments, so define fio's ioprio_set() as a macro.
>   * Note that there is no idea of class within ioprio_set(2) unlike Linux.
>   */
> -#define ioprio_value(ioprio_class, ioprio)	(ioprio)
> -#define ioprio_set(which, who, ioprio_class, ioprio)	\
> +#define ioprio_value(ioprio_class, ioprio, ioprio_hint)	(ioprio)
> +#define ioprio_set(which, who, ioprio_class, ioprio, ioprio_hint)	\
>  	ioprio_set(which, who, ioprio)
>  
>  #define ioprio(ioprio)		(ioprio)
> diff --git a/os/os-linux.h b/os/os-linux.h
> index 2f9f7e79..6f241d09 100644
> --- a/os/os-linux.h
> +++ b/os/os-linux.h
> @@ -125,13 +125,19 @@ enum {
>  #define IOPRIO_BITS		16
>  #define IOPRIO_CLASS_SHIFT	13
>  
> +#define IOPRIO_HINT_BITS	10
> +#define IOPRIO_HINT_SHIFT	3
> +
>  #define IOPRIO_MIN_PRIO		0	/* highest priority */
>  #define IOPRIO_MAX_PRIO		7	/* lowest priority */
>  
>  #define IOPRIO_MIN_PRIO_CLASS	0
>  #define IOPRIO_MAX_PRIO_CLASS	3
>  
> -static inline int ioprio_value(int ioprio_class, int ioprio)
> +#define IOPRIO_MIN_PRIO_HINT	0
> +#define IOPRIO_MAX_PRIO_HINT	((1 << IOPRIO_HINT_BITS) - 1)
> +
> +static inline int ioprio_value(int ioprio_class, int ioprio, int ioprio_hint)
>  {
>  	/*
>  	 * If no class is set, assume BE
> @@ -139,7 +145,9 @@ static inline int ioprio_value(int ioprio_class, int ioprio)
>          if (!ioprio_class)
>                  ioprio_class = IOPRIO_CLASS_BE;
>  
> -	return (ioprio_class << IOPRIO_CLASS_SHIFT) | ioprio;
> +	return (ioprio_class << IOPRIO_CLASS_SHIFT) |
> +		(ioprio_hint << IOPRIO_HINT_SHIFT) |
> +		ioprio;
>  }
>  
>  static inline bool ioprio_value_is_class_rt(unsigned int priority)
> @@ -147,10 +155,11 @@ static inline bool ioprio_value_is_class_rt(unsigned int priority)
>  	return (priority >> IOPRIO_CLASS_SHIFT) == IOPRIO_CLASS_RT;
>  }
>  
> -static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
> +static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio,
> +			     int ioprio_hint)
>  {
>  	return syscall(__NR_ioprio_set, which, who,
> -		       ioprio_value(ioprio_class, ioprio));
> +		       ioprio_value(ioprio_class, ioprio, ioprio_hint));
>  }
>  
>  #define ioprio_class(ioprio)	((ioprio) >> IOPRIO_CLASS_SHIFT)
> diff --git a/os/os.h b/os/os.h
> index 036fc233..2217d5f8 100644
> --- a/os/os.h
> +++ b/os/os.h
> @@ -122,9 +122,9 @@ extern int fio_cpus_split(os_cpu_mask_t *mask, unsigned int cpu);
>  #define IOPRIO_MAX_PRIO_CLASS		0
>  #endif
>  #ifndef FIO_HAVE_IOPRIO
> -#define ioprio_value(prioclass, prio)	(0)
> +#define ioprio_value(prioclass, prio, priohint)	(0)
>  #define ioprio(ioprio)			0
> -#define ioprio_set(which, who, prioclass, prio)	(0)
> +#define ioprio_set(which, who, prioclass, prio, priohint) (0)
>  #define IOPRIO_MIN_PRIO			0
>  #define IOPRIO_MAX_PRIO			0
>  #endif
> -- 
> 2.41.0
> 

Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>