From mboxrd@z Thu Jan  1 00:00:00 1970
From: tih@Hamartun.Priv.NO (Tom Ivar Helbekkmo)
Date: 06 Nov 2001 20:55:27 +0100
Subject: [pups] Trouble with 2.11BSD and CQD-220
In-Reply-To: <200111050415.fA54Fxd02290@moe.2bsd.com> ("Steven M. Schultz"'s message of "Sun, 4 Nov 2001 20:15:59 -0800 (PST)")
References: <200111050415.fA54Fxd02290@moe.2bsd.com>
Message-ID: <m24ro73c9c.fsf@barsoom.Hamartun.Priv.NO>

"Steven M. Schultz" <sms at 2BSD.COM> writes:

>> When booting from the SCSI drive, the boot sector is loaded into memory
>> and then relocated. It hangs waiting for the MSCP controller to respond.
>> I have not diagnosed it to the command that hangs.
> 
> 	Bug in the CMD controller but CMD isn't unique - others have had
> 	the problem as well.   The bug is that the controller insists on
> 	an interrupt vector being presented during the 3 or 4 step 
> 	initialization protocol.

Not exactly.  You and I actually worked this out together back in
1995, when I was figuring out how to get Reno to boot my VAX using a
CQD-220.  :-)  You don't have to present any interrupt vector, but you
shouldn't lie to the controller about this.  Instead of setting the
"please interrupt" bit in the MSCP datagram, and waiting for the
controller to set the "I've interrupted now" bit, you should set the
"controller owns this datagram now" bit, and wait for it to clear it.
This make much better sense, anyway, while running in polled mode.

I don't have any of my 2.11 systems running right now (but with winter
coming up, I shortly will, and will then have quite some work to do
patching them up to the current level), but the change I had to make
to the Reno uda.c illustrates it well:

*** uda.c.ORIG	Tue Nov 25 20:10:01 1997
--- uda.c	Tue Nov 25 20:11:50 1997
***************
*** 140,163 ****
  	u->uda1_cmd.mscp_opcode = op;
  	u->uda1_cmd.mscp_msglen = MSCP_MSGLEN;
  	u->uda1_rsp.mscp_msglen = MSCP_MSGLEN;
! 	u->uda1_ca.ca_rspdsc |= MSCP_OWN|MSCP_INT;
! 	u->uda1_ca.ca_cmddsc |= MSCP_OWN|MSCP_INT;
  	i = udaddr[io->i_adapt][io->i_ctlr]->udaip;	/* start uda polling */
  #ifdef lint
  	i = i;
  #endif
  	mp = &u->uda1_rsp;
  	for (;;) {
  		if (u->uda1_ca.ca_cmdint)
  			u->uda1_ca.ca_cmdint = 0;
! 		if (u->uda1_ca.ca_rspint == 0)
  			continue;
  		u->uda1_ca.ca_rspint = 0;
  		if (mp->mscp_opcode == (op | M_OP_END))
  			break;
  		printf("unexpected rsp type %x op %x ignored\n",
  			MSCP_MSGTYPE(mp->mscp_msgtc), mp->mscp_opcode);
! 		u->uda1_ca.ca_rspdsc |= MSCP_OWN | MSCP_INT;
  	}
  	if ((mp->mscp_status&M_ST_MASK) != M_ST_SUCCESS)
  		return (-1);
--- 140,165 ----
  	u->uda1_cmd.mscp_opcode = op;
  	u->uda1_cmd.mscp_msglen = MSCP_MSGLEN;
  	u->uda1_rsp.mscp_msglen = MSCP_MSGLEN;
! 	u->uda1_ca.ca_rspdsc |= MSCP_OWN;
! 	u->uda1_ca.ca_cmddsc |= MSCP_OWN;
! 
  	i = udaddr[io->i_adapt][io->i_ctlr]->udaip;	/* start uda polling */
  #ifdef lint
  	i = i;
  #endif
  	mp = &u->uda1_rsp;
+ 
  	for (;;) {
  		if (u->uda1_ca.ca_cmdint)
  			u->uda1_ca.ca_cmdint = 0;
! 		if (u->uda1_ca.ca_rspdsc & MSCP_OWN)
  			continue;
  		u->uda1_ca.ca_rspint = 0;
  		if (mp->mscp_opcode == (op | M_OP_END))
  			break;
  		printf("unexpected rsp type %x op %x ignored\n",
  			MSCP_MSGTYPE(mp->mscp_msgtc), mp->mscp_opcode);
! 		u->uda1_ca.ca_rspdsc |= MSCP_OWN;
  	}
  	if ((mp->mscp_status&M_ST_MASK) != M_ST_SUCCESS)
  		return (-1);

Oh, and that very VAX is still running Reno, happily booting off the
CQD-220 whenever I ask it to.  :-)

-tih
-- 
Popularity is the hallmark of mediocrity.  --Niles Crane, "Frasier"