Hi James -
On Tue, Sep 20, 2005 at 10:11:10PM -0500, James Bottomley wrote:
> Sorry, I didn't really like either patch.
OK :-/
> The problem, essentially is this rather evil return refcounted sdev
> under certain circumstances. What I didn't like about your patch was
> that you actually extended it to be under more circumstances.
I don't follow the "certain circumstances", but OK.
> The correct fix, I think, is to make scsi_report_lun_scan() actually
> work on the target device; this will also set us up nicely for WLUN
> scans if they prove necessary. This being well into -rc for linux, I'll
Yes, I agree. I was forward porting the patch from some time ago, and
didn't think the starget was far enough along.
> resist the strong urge to take the hatchet to the sdev returning code
> and stomach the duplication of scsi_level and resist the urge to
> investigate dumping the rescan parameter ... they can all be fixed
> properly after 2.6.14.
Yes for scsi_level.
With your patch, rescan could go away, but I do not like the lookup for
each LU when we first scan a target (I think O(n^2)? i.e. for many LUs,
1000 + 999 + ... + 1 and except for LUN 0, none will match). Long term we
could avoid this, maybe still special case LUN 0 in scsi_report_lun_scan();
we could also auto-set the rescan setting (and not externalize it) when
there no LU's on a target (or host) at the start
> How does the attached look (and more importantly, does it work)?
I tried it on top of scsi-rc-fixes, it oopsed (non-scsi) on ppc/JS20, so I
patched on a 2.6.14-rc2 git tree from sept 21, that boots up OK on IDE.
But the module refcount stays non-zero for scsi-debug, probably
scsi_device_put/get. I am still debugging ...
Did your try it with the scsi_debug patch? And multiple insmod/rmmod of
it?
This chunk (and "insmod scsi_debug.ko max_luns=5" or such):
--- scsi-misc-2.6/drivers/scsi/scsi_debug.c 2005-07-26 11:08:51.000000000 -0700
+++ lun0-replun-scsi-misc-2.6/drivers/scsi/scsi_debug.c 2005-07-27 18:24:05.000000000 -0700
@@ -619,7 +619,10 @@ static int resp_inquiry(struct scsi_cmnd
alloc_len = (cmd[3] << 8) + cmd[4];
memset(arr, 0, SDEBUG_MAX_INQ_ARR_SZ);
- pq_pdt = (scsi_debug_ptype & 0x1f);
+ if (devip->lun == 0)
+ pq_pdt = 0x7f;
+ else
+ pq_pdt = (scsi_debug_ptype & 0x1f);
arr[0] = pq_pdt;
if (0x2 & cmd[1]) { /* CMDDT bit set */
mk_sense_buffer(devip, ILLEGAL_REQUEST, INVALID_FIELD_IN_CDB,
-- Patrick Mansfield
-
: 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
