PERFORCE change 110005 for review

[Matt Jacob]

http://perforce.freebsd.org/chv.cgi?CH=110005

Change 110005 by mjacob at newisp on 2006/11/15 05:36:29

	Ignore zero-valued FCP RSP INFO codes. FCP-2 claims
	that this is 'Task Management Complete'. This is
	nonsensical when it is getting returned from a
	normal I/O (non-task management) command. I would
	guess that targets just set RSP LEN VALID (0x100)
	in the status word, set the accepted length field
	for response info (4 or 8) and call it a day.
	
	Non-zero values, if we ever see them, *do* indicate
	a problem with the transfer that has just occurred,
	so in that case it is indeed correct to set a generic
	I/O error for the transfer and let the upper layers
	decide whether to retry.

Affected files ...

.. //depot/projects/newisp/dev/isp/isp.c#35 edit

Differences ...

==== //depot/projects/newisp/dev/isp/isp.c#35 (text+ko) ====

@@ -4861,7 +4861,8 @@
 		switch (etype) {
 		case RQSTYPE_RESPONSE:
 			XS_SET_STATE_STAT(isp, xs, sp);
-			if (resp && rlen >= 4) {
+			if (resp && rlen >= 4 &&
+			    resp[FCP_RSPNS_CODE_OFFSET] != 0) {
 				isp_prt(isp, ISP_LOGWARN,
 				    "%d.%d FCP RESPONSE: 0x%x",
 				    XS_TGT(xs), XS_LUN(xs),