Re: [Bug 15185] New: Sending a 48bit ATA-Command with "CheckCondition" through SG_IO does not return correct 48bit sense descriptor

[Douglas Gilbert <dgilbert@xxxxxxxxxxxx>]

bugzilla-daemon@xxxxxxxxxxxxxxxxxxx wrote:
> http://bugzilla.kernel.org/show_bug.cgi?id=15185
>
>            Summary: Sending a 48bit ATA-Command with "CheckCondition"
>                     through SG_IO does not return correct 48bit sense
>                     descriptor
>            Product: SCSI Drivers
>            Version: 2.5
>     Kernel Version: 2.6.31
>           Platform: All
>         OS/Version: Linux
>               Tree: Mainline
>             Status: NEW
>           Severity: normal
>           Priority: P1
>          Component: Other
>         AssignedTo: scsi_drivers-other@xxxxxxxxxxxxxxxxxxxx
>         ReportedBy: stefan.huebner@xxxxxxxxxxxxxxxxxx
>         Regression: No
>
>
> Example: sending a correct "READ_NATIVE_MAX_ADDRESS_EXT" to /dev/sdd (opened
> O_RDWR | O_NONBLOCK) via ATA_PASSTHROUGH_16 (SCSI-Command 0x85) with the EXTEND
> and the CHECK_CONDITION bits set to 1 yields sense data in descriptor-format. 
> Unfortunately, the descriptor does not have EXTEND set, and by that only
> returns 24 Bits of LBA.
>
> This obviously is a bug, as the SAT-2 Draft says: "If the sense data is for an
> ATA PASS-THROUGH (16) command with the EXTEND bit set to one, then the SATL
> shall return the 48-bit extended status and shall set the EXTEND bit to one."
>
> Contents of important data-structures for SG_IO:
> sg_io_hdr.cmdlen = 16
> *sg_io_hdr.cmdp = {0x85 0x07 0x20 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00
> 0x00 0x00 0x27 0x00}
>
> sense-data after command:
> 0x72 0x00 0x00 0x00 0x00 0x00 0x00 0x0e 0x09 0x0c 0x00 0x00 0x00 0x00 0x00 0xaf
> 0x00 0x6d 0x00 0x70 0x00 0x50
> meaning: descriptor-sense, no error
> descriptor:
>  code=0x09 -> ATA-Return descriptor
>  length=0x0c
>  EXTEND=0
>  Error = 0x00
>  SectorCount = 0x00
>  LBA_Low = 0xaf
>  LBA_Mid = 0x6d
>  LBA_High= 0x70
>  Device  = 0
>  Status  = DeviceReady | DeferredWriteError
>
> The drive used should be reporting a native max lba of 0x74706daf (1.02TB), so
> the expected sense data should look like:
> 0x72 0x00 0x00 0x00 0x00 0x00 0x00 0x0e 0x09 0x0c 0x01 0x00 0x00 0x00 0x6d 0xaf
> 0x74 0x70 0x00 0x00 0x00 0x50

This bug does not occur in lk 2.6.30 but does in lk
2.6.32 . There was a pretty large rework of libata
in that period and there is obvious bug in
drivers/ata/libata-scsi.c that causes this.

The attached patch fixes this problem in lk 2.6.32 in
my test.

Signed-off-by: Douglas Gilbert <dgilbert@xxxxxxxxxxxx>




--- linux/drivers/ata/libata-scsi.c	2009-12-03 11:11:13.000000000 -0500
+++ linux/drivers/ata/libata-scsi.c2632dpg1	2010-02-01 10:55:36.000000000 -0500
@@ -2825,7 +2825,7 @@
 	 * write indication (used for PIO/DMA setup), result TF is
 	 * copied back and we don't whine too much about its failure.
 	 */
-	tf->flags = ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
+	tf->flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
 	if (scmd->sc_data_direction == DMA_TO_DEVICE)
 		tf->flags |= ATA_TFLAG_WRITE;