On Tue, 2010-05-04 at 16:30 -0400, James Bottomley wrote: > On Tue, 2010-05-04 at 15:27 -0500, Mike Christie wrote: > > On 05/04/2010 03:15 PM, Mike Christie wrote: > > > On 05/04/2010 11:26 AM, James Bottomley wrote: > > >> The other patch is fine, but I don't think this is necessary. The > > >> reason is that even returning SUCCESS here, we go straight into > > >> scsi_finish_command() (which passes it up to the driver handler) and > > >> then scsi_io_completion(). There's a catch for UNIT_ATTENTION in > > >> scsi_io_completion > > > > > > The request is sent as a REQ_TYPE_BLOCK_PC (this flag is set for the > > > request in sd_prepare_flush), and scsi_io_completion's blk_pc_request > > > check() that returns the request upwards is before the UNIT_ATTENTION > > > > I was looking at the wrong source. > > > > scsi_finish_command checks for REQ_TYPE_BLOCK_PC and sets good_bytes to > > scsi_bufflen, so when scsi_io_completion calls scsi_end_request, it > > fails the request before we can get to the UNIT_ATTENTION. > > Ah, yes, missed that. > > However there are other problems then. We can't just eat all unit > attentions on the BLOCK_PC path because some of the user programs want > to see them (CD burners for one). I know the patch allows some to > proceed upwards, but it's risky making all except device not started do > a retry. > > I'm a bit stumped on this one ... the intention of BLOCK_PC is that the > caller simply retries if they get a unit attention (which is what all > SCSI code does). The block code doesn't want to look at the sense data > but at the error return. I suppose we could make UNIT_ATTENTION > translate to -EAGAIN and have block do the right thing? I tried this, it gets very messy in block, so it looks to be a hack too far. What about this then ... it only retries UAs for barrier commands (thus preserving standard semantics for SG_IO requests)? James --- diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c index d45c69c..7ad53fa 100644 --- a/drivers/scsi/scsi_error.c +++ b/drivers/scsi/scsi_error.c @@ -302,7 +302,20 @@ static int scsi_check_sense(struct scsi_cmnd *scmd) if (scmd->device->allow_restart && (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) return FAILED; - return SUCCESS; + + if (blk_barrier_rq(scmd->request)) + /* + * barrier requests should always retry on UA + * otherwise block will get a spurious error + */ + return NEEDS_RETRY; + else + /* + * for normal (non barrier) commands, pass the + * UA upwards for a determination in the + * completion functions + */ + return SUCCESS; /* these three are not supported */ case COPY_ABORTED: -- 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
