----- Forwarded message from Steve Cameron <[email protected]> -----
Date: Fri, 24 Aug 2007 11:10:44 -0500
From: Steve Cameron <[email protected]>
To: [email protected]
Reply-To: [email protected]
Cc: [email protected]
Subject: Re: [PATCH 1/1] cciss: fix error reporting for SG_IO
This fixes a problem with the way cciss was filling out the "errors"
field of the request structure upon completion of requests.
Previously, it just put a 1 or a 0 in there and used the negation
of this as the uptodate parameter to one of the functions in the
block layer, being a block device. For the SG_IO ioctl, this was not
sufficient, and we noticed that, for example, sg_turs from sg3_utils
did not correctly detect problems due to cciss having set rq->errors
incorrectly.
Signed-off-by: Stephen M. Cameron <[email protected]>
Acked-by: Mike Miller <[email protected]>
drivers/block/cciss.c | 79 ++++++++++++++++++++++++++++++++++++++------------
1 files changed, 61 insertions(+), 18 deletions(-)
diff -puN drivers/block/cciss.c~sg_io_fix_error_reporting drivers/block/cciss.c
--- linux-2.6.23-rc2/drivers/block/cciss.c~sg_io_fix_error_reporting 2007-08-13 09:44:58.000000000 -0500
+++ linux-2.6.23-rc2-scameron/drivers/block/cciss.c 2007-08-24 10:56:56.000000000 -0500
@@ -2366,30 +2366,55 @@ static inline void resend_cciss_cmd(ctlr
start_io(h);
}
+static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
+ unsigned int msg_byte, unsigned int host_byte,
+ unsigned int driver_byte)
+{
+ /* inverse of macros in scsi.h */
+ return (scsi_status_byte & 0xff) |
+ ((msg_byte & 0xff) << 8) |
+ ((host_byte & 0xff) << 16) |
+ ((driver_byte & 0xff) << 24);
+}
+
static inline int evaluate_target_status(CommandList_struct *cmd)
{
unsigned char sense_key;
- int error_count = 1;
+ unsigned char status_byte, msg_byte, host_byte, driver_byte;
+ int error_value;
+
+ /* If we get in here, it means we got "target status", that is, scsi status */
+ status_byte = cmd->err_info->ScsiStatus;
+ driver_byte = DRIVER_OK;
+ msg_byte = cmd->err_info->CommandStatus; /* correct? seems too device specific */
+
+ if (blk_pc_request(cmd->rq))
+ host_byte = DID_PASSTHROUGH;
+ else
+ host_byte = DID_OK;
+
+ error_value = make_status_bytes(status_byte, msg_byte,
+ host_byte, driver_byte);
- if (cmd->err_info->ScsiStatus != 0x02) { /* not check condition? */
+ if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
if (!blk_pc_request(cmd->rq))
printk(KERN_WARNING "cciss: cmd %p "
"has SCSI Status 0x%x\n",
cmd, cmd->err_info->ScsiStatus);
- return error_count;
+ return error_value;
}
/* check the sense key */
sense_key = 0xf & cmd->err_info->SenseInfo[2];
/* no status or recovered error */
- if ((sense_key == 0x0) || (sense_key == 0x1))
- error_count = 0;
+ if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
+ error_value = 0;
if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
- if (error_count != 0)
+ if (error_value != 0)
printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
" sense key = 0x%x\n", cmd, sense_key);
- return error_count;
+ return error_value;
}
/* SG_IO or similar, copy sense data back */
@@ -2401,7 +2426,7 @@ static inline int evaluate_target_status
} else
cmd->rq->sense_len = 0;
- return error_count;
+ return error_value;
}
/* checks the status of the job and calls complete buffers to mark all
@@ -2417,7 +2442,7 @@ static inline void complete_command(ctlr
rq->errors = 0;
if (timeout)
- rq->errors = 1;
+ rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
if (cmd->err_info->CommandStatus == 0) /* no error has occurred */
goto after_error_processing;
@@ -2443,32 +2468,44 @@ static inline void complete_command(ctlr
case CMD_INVALID:
printk(KERN_WARNING "cciss: cmd %p is "
"reported invalid\n", cmd);
- rq->errors = 1;
+ rq->errors = make_status_bytes(SAM_STAT_GOOD,
+ cmd->err_info->CommandStatus, DRIVER_OK,
+ blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
break;
case CMD_PROTOCOL_ERR:
printk(KERN_WARNING "cciss: cmd %p has "
"protocol error \n", cmd);
- rq->errors = 1;
+ rq->errors = make_status_bytes(SAM_STAT_GOOD,
+ cmd->err_info->CommandStatus, DRIVER_OK,
+ blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
break;
case CMD_HARDWARE_ERR:
printk(KERN_WARNING "cciss: cmd %p had "
" hardware error\n", cmd);
- rq->errors = 1;
+ rq->errors = make_status_bytes(SAM_STAT_GOOD,
+ cmd->err_info->CommandStatus, DRIVER_OK,
+ blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
break;
case CMD_CONNECTION_LOST:
printk(KERN_WARNING "cciss: cmd %p had "
"connection lost\n", cmd);
- rq->errors = 1;
+ rq->errors = make_status_bytes(SAM_STAT_GOOD,
+ cmd->err_info->CommandStatus, DRIVER_OK,
+ blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
break;
case CMD_ABORTED:
printk(KERN_WARNING "cciss: cmd %p was "
"aborted\n", cmd);
- rq->errors = 1;
+ rq->errors = make_status_bytes(SAM_STAT_GOOD,
+ cmd->err_info->CommandStatus, DRIVER_OK,
+ blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
break;
case CMD_ABORT_FAILED:
printk(KERN_WARNING "cciss: cmd %p reports "
"abort failed\n", cmd);
- rq->errors = 1;
+ rq->errors = make_status_bytes(SAM_STAT_GOOD,
+ cmd->err_info->CommandStatus, DRIVER_OK,
+ blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
break;
case CMD_UNSOLICITED_ABORT:
printk(KERN_WARNING "cciss%d: unsolicited "
@@ -2482,17 +2519,23 @@ static inline void complete_command(ctlr
printk(KERN_WARNING
"cciss%d: %p retried too "
"many times\n", h->ctlr, cmd);
- rq->errors = 1;
+ rq->errors = make_status_bytes(SAM_STAT_GOOD,
+ cmd->err_info->CommandStatus, DRIVER_OK,
+ blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
break;
case CMD_TIMEOUT:
printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
- rq->errors = 1;
+ rq->errors = make_status_bytes(SAM_STAT_GOOD,
+ cmd->err_info->CommandStatus, DRIVER_OK,
+ blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
break;
default:
printk(KERN_WARNING "cciss: cmd %p returned "
"unknown status %x\n", cmd,
cmd->err_info->CommandStatus);
- rq->errors = 1;
+ rq->errors = make_status_bytes(SAM_STAT_GOOD,
+ cmd->err_info->CommandStatus, DRIVER_OK,
+ blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
}
after_error_processing:
_
----- End forwarded message -----
On Fri, 24 Aug 2007 12:53:54 -0500
"Mike Miller (OS Dev)" <[email protected]> wrote:
> This fixes a problem with the way cciss was filling out the "errors"
> field of the request structure upon completion of requests.
> Previously, it just put a 1 or a 0 in there and used the negation
> of this as the uptodate parameter to one of the functions in the
> block layer, being a block device. For the SG_IO ioctl, this was not
> sufficient, and we noticed that, for example, sg_turs from sg3_utils
> did not correctly detect problems due to cciss having set rq->errors
> incorrectly.
Do we think this problem is sufficiently serious to merit merging
this (largeish) patch into 2.6.23?
I'm thinking "no", but that might be wrong...
On Fri, Aug 24, 2007 at 05:10:47PM -0700, Andrew Morton wrote:
> On Fri, 24 Aug 2007 12:53:54 -0500
> "Mike Miller (OS Dev)" <[email protected]> wrote:
>
> > This fixes a problem with the way cciss was filling out the "errors"
> > field of the request structure upon completion of requests.
> > Previously, it just put a 1 or a 0 in there and used the negation
> > of this as the uptodate parameter to one of the functions in the
> > block layer, being a block device. For the SG_IO ioctl, this was not
> > sufficient, and we noticed that, for example, sg_turs from sg3_utils
> > did not correctly detect problems due to cciss having set rq->errors
> > incorrectly.
>
> Do we think this problem is sufficiently serious to merit merging
> this (largeish) patch into 2.6.23?
>
> I'm thinking "no", but that might be wrong...
We'd like to see it 2.6.23. It may help insulate me from lots of questions
about where it is. :) But it's entirely up to you.
mikem