This patch series effects the retry behavior of IO when blk_noretry_request is used.
The patches effect the retries in two areas. The first is that it allows
the scsi_dh check_sense routines to have more control of the retry of a
IO. Currently if the scsi_dh check_sense functions return NEEDS_RETRY
this will not lead to a retry in the DM-MP configurations where they are
most likely to be used because the blk_noretry_request will be true for
these IOs which will prevent the retry. The second change in behavior of
the HARDWARE_ERROR sense key is to avoid path failures of all DM-MP paths
for the IBM device(s) returning this sense key. The utilization of
ADD_TO_MLQUEUE for the HARDWARE_ERROR sense key could lead to longer retry
times for the other user of BLIST_RETRY_HWERROR.
I have tried to summarize the scsi_decide_disposition current retry behavior
below.
1.) Based on host_byte
The host bytes listed below will return ADD_TO_MLQUEUE or NEEDS_RETRY
which is always retried up to the max scsi mid-layer in-flight time
(unsigned long wait_for = (cmd->allowed + 1) * cmd->timeout_per_command;).
DID_REQUEUE:
(returns ADD_TO_MLQUEUE)
DID_IMM_RETRY:
(returns NEEDS_RETRY)
The host bytes listed below will use the maybe_retry goto which will
not be retried if blk_noretry_request is true.
DID_ERROR:
DID_SOFT_ERROR:
DID_BUS_BUSY:
DID_PARITY:
2.) Based on status_byte
The status bytes listed below will return ADD_TO_MLQUEUE which is
always retried up to the max scsi mid-layer in-flight time (unsigned
long wait_for = (cmd->allowed + 1) * cmd->timeout_per_command;).
BUSY:
QUEUE_FULL:
3.) Based on sense_key
The sense keys and sense checks listed below will return NEEDS_RETRY
which will be not be retried if blk_noretry_request is true.
Note1: There is some if checks removed to simplify the list. Please see
code for more detailed behavior.
Note2: All scsi_dh sense_check functions will be effected by the
blk_noretry_request check.
scsi_sense_is_deferred
ABORTED_COMMAND:
NOT_READY:
UNIT_ATTENTION:
MEDIUM_ERROR:
HARDWARE_ERROR:
I have summarized the change in retry behavior below.
1.) Based on sense_key
The current scsi_dh sense_check functions will return
ADD_TO_MLQUEUE.
HARDWARE_ERROR will return ADD_TO_MLQUEUE if retry_hwerror is set.
-andmike
--
Michael Anderson
andmike@xxxxxxxxxxxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
