Re: [bug] ata subsystem related crash with latest -git

[Jens Axboe <jens.axboe@oracle.com>]

On Thu, Oct 18 2007, Jeff Garzik wrote:
> Ingo Molnar wrote:
>> * Jens Axboe <jens.axboe@oracle.com> wrote:
>>> --- a/drivers/scsi/scsi_lib.c
>>> +++ b/drivers/scsi/scsi_lib.c
>>> @@ -39,7 +39,7 @@
>>>   * (unless chaining is used). Should ideally fit inside a single page, 
>>> to
>>>   * avoid a higher order allocation.
>>>   */
>>> -#define SCSI_MAX_SG_SEGMENTS	128
>>> +#define SCSI_MAX_SG_SEGMENTS	129
>> this one finally made the trick and it's booting fine now, without any 
>> crashes!
>
> Alas, this didn't help me here.  I did manage to capture the error messages 
> this time, and in the two machines I'm actively testing on, sata_mv is the 
> driver that's dying in both cases.  Machine A additionally has sata_nv, 
> which is working.  Machine B additionally has ata_piix, which is working.  
> So in both cases, its sata_mv that is throwing SError complaints:

Theory - ata_sg_is_last() isn't returning true for the last entry. Can
you double check that it correcly marks the last entry in mv_fill_sg()?
Alternatively, just try this patch.


diff --git a/drivers/ata/sata_mv.c b/drivers/ata/sata_mv.c
index 4df8311..b858183 100644
--- a/drivers/ata/sata_mv.c
+++ b/drivers/ata/sata_mv.c
@@ -1139,6 +1139,7 @@ static void mv_fill_sg(struct ata_queued_cmd *qc)
 	struct mv_port_priv *pp = qc->ap->private_data;
 	struct scatterlist *sg;
 	struct mv_sg *mv_sg;
+	int end_marked = 0;
 
 	mv_sg = pp->sg_tbl;
 	ata_for_each_sg(sg, qc) {
@@ -1159,13 +1160,15 @@ static void mv_fill_sg(struct ata_queued_cmd *qc)
 			sg_len -= len;
 			addr += len;
 
-			if (!sg_len && ata_sg_is_last(sg, qc))
+			if (!sg_len && ata_sg_is_last(sg, qc)) {
 				mv_sg->flags_size |= cpu_to_le32(EPRD_FLAG_END_OF_TBL);
+				end_marked++;
+			}
 
 			mv_sg++;
 		}
-
 	}
+	BUG_ON(end_marked != 1);
 }
 
 static inline void mv_crqb_pack_cmd(__le16 *cmdw, u8 data, u8 addr, unsigned last)

-- 
Jens Axboe