On 2023/03/10 6:54, Niklas Cassel wrote:
> From: Damien Le Moal <damien.lemoal@xxxxxxxxxxxxxxxxxx>
>
> IO priorities currently only use 6-bits of the 16-bits ioprio value: the
> 3-upper bits are used to define up to 8 priority classes (4 of which are
> valid) and the 3 lower bits of the value are used to define a priority
> level for the real-time and best-effort class.
>
> The remaining 10-bits between the IO priority class and level are
> unused, and in fact, cannot be used by the user as doing so would
> either result in the value being completely ignored, or in an error
> returned by ioprio_check_cap().
>
> Use these 10-bits of an ioprio value to allow a user to specify IO
> hints. An IO hint is defined as a 10-bits value, allowing up to 1023
> different hints to be specified, with the value 0 being reserved as the
> "no hint" case. An IO hint can apply to any IO that specifies a valid
> priority class other than NONE, regardless of the IO priority level
> specified.
>
> To do so, the macros IOPRIO_PRIO_HINT() and IOPRIO_PRIO_VALUE_HINT() are
> introduced in include/uapi/linux/ioprio.h to respectively allow a user
> to get and set a hint in an ioprio value.
>
> To support the ATA and SCSI command duration limits feature, 7 hints
> are defined: IOPRIO_HINT_DEV_DURATION_LIMIT_1 to
> IOPRIO_HINT_DEV_DURATION_LIMIT_7, allowing a user to specify which
> command duration limit descriptor should be applied to the commands
> serving an IO. Specifying these hints has for now no effect whatsoever
> if the target block devices do not support the command duration limits
> feature. However, in the future, block IO schedulers can be modified to
> optimize IO issuing order based on these hints, even for devices that
> do not support the command duration limits feature.
>
> Given that the 7 duration limits hints defined have no effect on any
> block layer component, the actual definition of the duration limits
> implied by these hints remains at the device level.
>
> Signed-off-by: Damien Le Moal <damien.lemoal@xxxxxxxxxxxxxxxxxx>
> Signed-off-by: Niklas Cassel <niklas.cassel@xxxxxxx>
Jens, Martin,
Any thoughts on this new interface ?
Bart,
I would like to hear your opinion as well.
> ---
> include/uapi/linux/ioprio.h | 49 +++++++++++++++++++++++++++++++++++++
> 1 file changed, 49 insertions(+)
>
> diff --git a/include/uapi/linux/ioprio.h b/include/uapi/linux/ioprio.h
> index 4444b4e4fdad..607c7617b9d2 100644
> --- a/include/uapi/linux/ioprio.h
> +++ b/include/uapi/linux/ioprio.h
> @@ -58,4 +58,53 @@ enum {
> #define IOPRIO_NORM 4
> #define IOPRIO_BE_NORM IOPRIO_NORM
>
> +/*
> + * The 10-bits between the priority class and the priority level are used to
> + * optionally define IO hints for any combination of IO priority class and
> + * level. Depending on the kernel configuration, IO scheduler being used and
> + * the target IO device being used, hints can influence how IOs are processed
> + * without affecting the IO scheduling ordering defined by the IO priority
> + * class and level.
> + */
> +#define IOPRIO_HINT_SHIFT IOPRIO_LEVEL_NR_BITS
> +#define IOPRIO_HINT_NR_BITS 10
> +#define IOPRIO_NR_HINTS (1 << IOPRIO_HINT_NR_BITS)
> +#define IOPRIO_HINT_MASK (IOPRIO_NR_HINTS - 1)
> +#define IOPRIO_PRIO_HINT(ioprio) \
> + (((ioprio) >> IOPRIO_HINT_SHIFT) & IOPRIO_HINT_MASK)
> +
> +/*
> + * Alternate macro for IOPRIO_PRIO_VALUE() to define an IO priority with
> + * a class, level and hint.
> + */
> +#define IOPRIO_PRIO_VALUE_HINT(class, level, hint) \
> + ((((class) & IOPRIO_CLASS_MASK) << IOPRIO_CLASS_SHIFT) | \
> + (((hint) & IOPRIO_HINT_MASK) << IOPRIO_HINT_SHIFT) | \
> + ((level) & IOPRIO_LEVEL_MASK))
> +
> +/*
> + * IO hints.
> + */
> +enum {
> + /* No hint */
> + IOPRIO_HINT_NONE = 0,
> +
> + /*
> + * Device command duration limits: indicate to the device a desired
> + * duration limit for the commands that will be used to process an IO.
> + * These will currently only be effective for SCSI and ATA devices that
> + * support the command duration limits feature. If this feature is
> + * enabled, then the commands issued to the device to process an IO with
> + * one of these hints set will have the duration limit index (dld field)
> + * set to the value of the hint.
> + */
> + IOPRIO_HINT_DEV_DURATION_LIMIT_1 = 1,
> + IOPRIO_HINT_DEV_DURATION_LIMIT_2 = 2,
> + IOPRIO_HINT_DEV_DURATION_LIMIT_3 = 3,
> + IOPRIO_HINT_DEV_DURATION_LIMIT_4 = 4,
> + IOPRIO_HINT_DEV_DURATION_LIMIT_5 = 5,
> + IOPRIO_HINT_DEV_DURATION_LIMIT_6 = 6,
> + IOPRIO_HINT_DEV_DURATION_LIMIT_7 = 7,
> +};
> +
> #endif /* _UAPI_LINUX_IOPRIO_H */
--
Damien Le Moal
Western Digital Research
