Re: [PATCH v2 22/35] scsi: Have scsi-ml retry read_capacity_16 errors

[Martin Wilck <mwilck@suse.com>]

On Wed, 2022-09-28 at 21:53 -0500, Mike Christie wrote:
> This has read_capacity_16 have scsi-ml retry errors instead of
> driving
> them itself.
> 
> Signed-off-by: Mike Christie <michael.christie@oracle.com>
> ---
>  drivers/scsi/sd.c | 82 +++++++++++++++++++++++++--------------------
> --
>  1 file changed, 43 insertions(+), 39 deletions(-)
> 
> diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
> index 37eafa968116..a35c089c3097 100644
> --- a/drivers/scsi/sd.c
> +++ b/drivers/scsi/sd.c
> @@ -2283,55 +2283,59 @@ static int read_capacity_16(struct scsi_disk
> *sdkp, struct scsi_device *sdp,
>         struct scsi_sense_hdr sshdr;
>         int sense_valid = 0;
>         int the_result;
> -       int retries = 3, reset_retries =
> READ_CAPACITY_RETRIES_ON_RESET;
>         unsigned int alignment;
>         unsigned long long lba;
>         unsigned sector_size;
> +       struct scsi_failure failures[] = {
> +               {
> +                       .sense = UNIT_ATTENTION,
> +                       .asc = 0x29,
> +                       .ascq = 0,
> +                       /* Device reset might occur several times */
> +                       .allowed = READ_CAPACITY_RETRIES_ON_RESET,
> +                       .result = SAM_STAT_CHECK_CONDITION,
> +               },
> +               {
> +                       .result = SCMD_FAILURE_ANY,
> +                       .allowed = 3,
> +               },
> +               {},
> +       };

I first wondered whether this was correct, until I realized that
the logic in patch 02/35 actually treats the counts for different
failures independently, so that the maximum overall retry count is
the sum of the individual retry counts.

I wonder if we should give callers the chance to set a limit for the
overall retry count in addition to the retry counts for individual 
failures.

Martin