On 8/28/24 01:48, Niklas Cassel wrote:
> Every time I see libata code calling scsi_check_sense(), I get confused
> why the code path that is working fine for SCSI code, is not sufficient
> for libata code.
>
> The reason is that SCSI usually gets the sense data as part of the
> completion, and will thus automatically call scsi_check_sense(), which
> will set the SCSI ML byte (if any).
>
> However, for libata queued commands, we always need to fetch the sense
> data via SCSI EH, and thus do not get the luxury of having
> scsi_check_sense() called automatically.
>
> Add a new helper, ata_eh_decide_disposition(), that has a ata_eh_ prefix
> to more clearly highlight that this is only needed for code called by EH,
> while also having a similar name to scsi_decide_disposition(), such that
> it is easier to compare the libata code with the equivalent SCSI code.
>
> Also add a big kdoc comment explaining why this helper is called/needed in
> the first place.
>
> Signed-off-by: Niklas Cassel <cassel@xxxxxxxxxx>
> ---
> drivers/ata/libata-eh.c | 46 ++++++++++++++++++++++++++++++++++-----
> drivers/ata/libata-sata.c | 8 +++----
> drivers/ata/libata.h | 1 +
> 3 files changed, 45 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/ata/libata-eh.c b/drivers/ata/libata-eh.c
> index 214b935c2ced..173769999c13 100644
> --- a/drivers/ata/libata-eh.c
> +++ b/drivers/ata/libata-eh.c
> @@ -1401,6 +1401,43 @@ unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
> return err_mask;
> }
>
> +/**
> + * ata_eh_decide_disposition - Disposition a qc based on sense data
> + * @qc: qc to examine
> + *
> + * For a regular SCSI command, the SCSI completion callback (scsi_done())
> + * will call scsi_complete(), which will call scsi_decide_disposition(),
> + * which will call scsi_check_sense(). scsi_complete() finally calls
> + * scsi_finish_command(). This is fine for SCSI, since any eventual sense
> + * data is usually returned in the completion itself (without invoking SCSI
> + * EH). However, for a QC, we always need to fetch the sense data
> + * explicitly using SCSI EH.
> + *
> + * A command that is completed via SCSI EH will instead be completed using
> + * scsi_eh_flush_done_q(), which will call scsi_finish_command() directly
> + * (without ever calling scsi_check_sense()).
> + *
> + * For a command that went through SCSI EH, it is the responsibility of the
> + * SCSI EH strategy handler to call scsi_decide_disposition(), see e.g. how
> + * scsi_eh_get_sense() calls scsi_decide_disposition() for SCSI LLDDs that
> + * do not get the sense data as part of the completion.
> + *
> + * Thus, for QC commands that went via SCSI EH, we need to call
> + * scsi_check_sense() ourselves, similar to how scsi_eh_get_sense() calls
> + * scsi_decide_disposition(), which calls scsi_check_sense(), in order to
> + * set the correct SCSI ML byte (if any).
> + *
> + * LOCKING:
> + * EH context.
> + *
> + * RETURNS:
> + * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
> + */
> +enum scsi_disposition ata_eh_decide_disposition(struct ata_queued_cmd *qc)
> +{
> + return scsi_check_sense(qc->scsicmd);
> +}
> +
> /**
> * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
> * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
> @@ -1627,7 +1664,7 @@ static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
> }
>
> if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
> - enum scsi_disposition ret = scsi_check_sense(qc->scsicmd);
> + enum scsi_disposition ret = ata_eh_decide_disposition(qc);
While at it, please add the missing blank line after this declaration.
Other than this, looks fine to me.
> /*
> * SUCCESS here means that the sense code could be
> * evaluated and should be passed to the upper layers
> @@ -1942,11 +1979,10 @@ static int ata_eh_read_sense_success_non_ncq(struct ata_link *link)
> return -EIO;
>
> /*
> - * If we have sense data, call scsi_check_sense() in order to set the
> - * correct SCSI ML byte (if any). No point in checking the return value,
> - * since the command has already completed successfully.
> + * No point in checking the return value, since the command has already
> + * completed successfully.
> */
> - scsi_check_sense(qc->scsicmd);
> + ata_eh_decide_disposition(qc);
>
> return 0;
> }
> diff --git a/drivers/ata/libata-sata.c b/drivers/ata/libata-sata.c
> index 48660d445602..76904dce017e 100644
> --- a/drivers/ata/libata-sata.c
> +++ b/drivers/ata/libata-sata.c
> @@ -1455,12 +1455,10 @@ int ata_eh_read_sense_success_ncq_log(struct ata_link *link)
> qc->flags |= ATA_QCFLAG_SENSE_VALID;
>
> /*
> - * If we have sense data, call scsi_check_sense() in order to
> - * set the correct SCSI ML byte (if any). No point in checking
> - * the return value, since the command has already completed
> - * successfully.
> + * No point in checking the return value, since the command has
> + * already completed successfully.
> */
> - scsi_check_sense(qc->scsicmd);
> + ata_eh_decide_disposition(qc);
> }
>
> return ret;
> diff --git a/drivers/ata/libata.h b/drivers/ata/libata.h
> index 6abf265f626e..e4029626641d 100644
> --- a/drivers/ata/libata.h
> +++ b/drivers/ata/libata.h
> @@ -162,6 +162,7 @@ extern void ata_eh_finish(struct ata_port *ap);
> extern int ata_ering_map(struct ata_ering *ering,
> int (*map_fn)(struct ata_ering_entry *, void *),
> void *arg);
> +enum scsi_disposition ata_eh_decide_disposition(struct ata_queued_cmd *qc);
> extern unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key);
> extern unsigned int atapi_eh_request_sense(struct ata_device *dev,
> u8 *sense_buf, u8 dfl_sense_key);
--
Damien Le Moal
Western Digital Research
