libata: scsi error handling

Jeff,
Here is a follow on patch based on lk 2.6.13-rc7 plus the start_stop_unit patch and the rw_zero_length patch.

It adds more general error processing, typically for problems (or an early finish) detected while a SCSI command is being processed prior to an ATA command being executed.

Changelog:
- add extern ata_scsi_set_sense() to build SCSI sense data and corresponding status
- this allows removal of extern ata_scsi_badcmd() and static inline ata_bad_scsiop() and ata_bad_cdb()
- change "xlat" functions in libata-scsi so they are responsible for SCSI status and sense data when they return 1. This allows GOOD status or a specialized error to be set.
- set DRIVER_SENSE when SAM_STAT_CHECK_CONDITION is flagged in scsi_cmnd::result
- yield an error for mode sense requests for saved

    values [sat-r05]

- change recent rw_zero_length patch to do nothing (yield GOOD status) when transfer_length==0 for 10 and 16 byte READ and WRITE commands (SBC-2).

Doug Gilbert

@@ -69,21 +69,9 @@
 			        unsigned int buflen);
 extern unsigned int ata_scsiop_report_luns(struct ata_scsi_args *args, u8 *rbuf,
 				   unsigned int buflen);
-extern void ata_scsi_badcmd(struct scsi_cmnd *cmd,
-			    void (*done)(struct scsi_cmnd *),
-			    u8 asc, u8 ascq);
+extern void ata_scsi_set_sense(struct scsi_cmnd *cmd, u8 sk, u8 asc, u8 ascq);
 extern void ata_scsi_rbuf_fill(struct ata_scsi_args *args, 
                         unsigned int (*actor) (struct ata_scsi_args *args,
                                            u8 *rbuf, unsigned int buflen));
 
-static inline void ata_bad_scsiop(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *))
-{
-	ata_scsi_badcmd(cmd, done, 0x20, 0x00);
-}
-
-static inline void ata_bad_cdb(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *))
-{
-	ata_scsi_badcmd(cmd, done, 0x24, 0x00);
-}
-
 #endif /* __LIBATA_H__ */

All this is OK, though in the future, please split changes such as the creation and explicit use of ata_scsi_set_sense() to a separate patch. We need to separate logical changes. Patches get applied more quickly when they are split up, too.

However, I really wanted to go ahead and merge the C/H/S support patch, which updates the R/W xlat function, causing conflicts with your patch in several places. (as a side effect, C/H/S updates clean up the R/W xlat function rather nicely)

If you would be kind enough to rediff against the 'upstream' branch of rsync://rsync.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev.git

I would really appreciate it. I'll merge it as soon as I receive the rediff.

Jeff