Major overhaul of the aic7xxx driver:

        - Report valid residual byte counts.  We actually pause the sequencer
	  when the residual is non-zero.  I thought about using DMA to do this,
	  bus sequencer program space is tight.

	- Fix embarassing off by one error in the computation of a 2's
	  compliment variable.  This was most likely the cause of the
	  many problems reported with the tagged queuing code.

	- Handle "MAX_SYNC" as a special case (ie we are the ones starting
	  the sync negotiation sequence).  This was done so that the target
	  scratch area can be initialed to 0 offset (asyncronous transfers)
	  safely.  The initialization to 0 (was 15) is necessary since in
	  some cases a Wide negotiation could run into problems if SCSIRATE
	  was set wrong and we went into data(in/out).

	- Trim the DMA routines a little by using some procedures.  Net
	  effect is more functionality with 3 less instructions after this
	  update.

	- Toggle the WIDEODD bit of the DFCNTRL whenever this is not the
	  last SG block.  It has no effect in the 8bit bus configuration,
	  but in the Wide configuration ensures that the overlap byte is
	  held in the SCSI block if the transfer is odd so it will end
	  up in the next SG (the correct behavior).

4 files changed, 126 insertions, 90 deletions

diff --git a/sys/dev/aic7xxx/aic7xxx.seq b/sys/dev/aic7xxx/aic7xxx.seq
index d328693..42d0d82 100644
--- a/sys/dev/aic7xxx/aic7xxx.seq
+++ b/sys/dev/aic7xxx/aic7xxx.seq
@@ -22,7 +22,7 @@
 # optimizations provided by Justin T. Gibbs (gibbs@FreeBSD.org)
 #
 
-VERSION AIC7XXX_SEQ_VER "$Id: aic7xxx.seq,v 1.9 1995/03/07 09:00:44 gibbs Exp $"
+VERSION AIC7XXX_SEQ_VER "$Id: aic7xxx.seq,v 1.10 1995/03/17 23:54:17 gibbs Exp $"
 
 SCBMASK		= 0x1f
 
@@ -56,6 +56,9 @@ SINDIR		= 0x6c
 DINDIR		= 0x6d
 FUNCTION1	= 0x6e
 HADDR		= 0x88
+HADDR+1		= 0x89
+HADDR+2		= 0x8a
+HADDR+3		= 0x8b
 HCNT		= 0x8c
 HCNT+0		= 0x8c
 HCNT+1		= 0x8d
@@ -113,8 +116,12 @@ SCBARRAY+23	= 0xb7
 SCBARRAY+24	= 0xb8
 SCBARRAY+25	= 0xb9
 SCBARRAY+26	= 0xba
+SCBARRAY+27	= 0xbb
+SCBARRAY+28	= 0xbc
+SCBARRAY+29	= 0xbd
 
 BAD_PHASE	= 0x01				# unknown scsi bus phase
+CMDCMPLT	= 0x02
 SEND_REJECT	= 0x11				# sending a message reject
 NO_IDENT	= 0x21				# no IDENTIFY after reconnect
 NO_MATCH	= 0x31				# no cmd match for reconnect
@@ -122,6 +129,7 @@ MSG_SDTR	= 0x41				# SDTR message recieved
 MSG_WDTR	= 0x51				# WDTR message recieved
 MSG_REJECT	= 0x61				# Reject message recieved
 BAD_STATUS	= 0x71				# Bad status from target
+RESIDUAL	= 0x81				# Residual byte count != 0
 
 #  The host adapter card (at least the BIOS) uses 20-2f for SCSI
 #  device information, 32-33 and 5a-5f as well. As it turns out, the
@@ -164,7 +172,7 @@ MSG_START+2	= 0x37
 MSG_START+3	= 0x38
 MSG_START+4	= 0x39
 MSG_START+5	= 0x3a
--MSG_START+0	= 0xcb				# 2's complement of MSG_START+0
+-MSG_START+0	= 0xcc				# 2's complement of MSG_START+0
 
 ARG_1		= 0x4a				# sdtr conversion args & return
 BUS_16_BIT	= 0x01
@@ -189,6 +197,7 @@ SCBCOUNT	= 0x52				# the actual number of SCBs
 FLAGS		= 0x53				# Device configuration flags
 TWIN_BUS	= 0x01
 WIDE_BUS	= 0x02
+MAX_SYNC	= 0x08
 SENSE		= 0x10
 ACTIVE_MSG	= 0x20
 IDENTIFY_SEEN	= 0x40
@@ -236,34 +245,16 @@ test	SCBARRAY+0,NEEDDMA	jz test_busy
 #  Wait for DMA from host memory to data FIFO to complete, then disable
 #  DMA and wait for it to acknowledge that it's off.
 #
-scb_load1:
-	test	DFSTATUS,0x8	jz scb_load1	# HDONE
-
-	clr	DFCNTRL				# disable DMA
-scb_load2:
-	test	DFCNTRL,0x8	jnz scb_load2	# HDMAENACK
+	call	dma_finish
 
 # Copy the SCB from the FIFO to  the SCBARRAY
 
-	mov	SCBARRAY+0, DFDAT
-	mov	SCBARRAY+1, DFDAT
-	mov	SCBARRAY+2, DFDAT
-	mov	SCBARRAY+3, DFDAT
-	mov	SCBARRAY+4, DFDAT
-	mov	SCBARRAY+5, DFDAT
-	mov	SCBARRAY+6, DFDAT
-	mov	SCBARRAY+7, DFDAT
-	mov	SCBARRAY+8, DFDAT
-	mov	SCBARRAY+9, DFDAT
-	mov	SCBARRAY+10, DFDAT
-	mov	SCBARRAY+11, DFDAT
-	mov	SCBARRAY+12, DFDAT
-	mov	SCBARRAY+13, DFDAT
-	mov	SCBARRAY+14, DFDAT
-	mov	SCBARRAY+15, DFDAT
-	mov	SCBARRAY+16, DFDAT
-	mov	SCBARRAY+17, DFDAT
-	mov	SCBARRAY+18, DFDAT
+	mvi	DINDEX, SCBARRAY+0
+	call	bcopy_3_dfdat
+	call	bcopy_4_dfdat
+	call	bcopy_4_dfdat
+	call	bcopy_4_dfdat
+	call	bcopy_4_dfdat
    
 # See if there is not already an active SCB for this target.  This code
 # locks out on a per target basis instead of target/lun.  Although this
@@ -338,7 +329,6 @@ mk_tag:
 	mov	DINDIR,SCBPTR
 
 	add	MSG_LEN,-MSG_START+0,DINDEX	# update message length
-	jmp	!message			# Can't do DTR when taged
 
 mk_tag_done:
 
@@ -441,9 +431,17 @@ p_dataout:
 	mvi	DINDEX,STCNT
 	mvi	SCBARRAY+23	call bcopy_3
 
+# If we are the last SG block, don't set wideodd.
+	test    SCBARRAY+18,0xff jnz p_dataout_wideodd
 	mvi	0x3d		call dma	# SCSIEN|SDMAEN|HDMAEN|
 						#   DIRECTION|FIFORESET
+	jmp	p_dataout_rest
+
+p_dataout_wideodd:
+	mvi	0xbd		call dma	# WIDEODD|SCSIEN|SDMAEN|HDMAEN|
+						#   DIRECTION|FIFORESET
 
+p_dataout_rest:
 #  After a DMA finishes, save the final transfer pointer and count
 #  back into the SCB, in case a device disconnects in the middle of
 #  a transfer.  Use SHADDR and STCNT instead of HADDR and HCNT, since
@@ -475,8 +473,15 @@ p_datain:
 	mvi	DINDEX,STCNT
 	mvi	SCBARRAY+23	call bcopy_3
 
+# If we are the last SG block, don't set wideodd.
+	test	SCBARRAY+18,0xff jnz p_datain_wideodd
 	mvi	0x39		call dma	# SCSIEN|SDMAEN|HDMAEN|
 						#   !DIRECTION|FIFORESET
+	jmp	p_datain_rest
+p_datain_wideodd:
+	mvi	0xb9		call dma	# WIDEODD|SCSIEN|SDMAEN|HDMAEN|
+						#   !DIRECTION|FIFORESET
+p_datain_rest:
 	mvi	DINDEX,SCBARRAY+23
 	mvi	STCNT		call bcopy_3
 
@@ -611,21 +616,25 @@ p_mesgin:
 #  This allows the driver to interpret errors only when they occur
 #  instead of always uploading the scb.  If the status is SCSI_CHECK,
 #  the driver will download a new scb requesting sense, to replace 
-#  the old one and the sequencer code will imediately jump to start 
-#  working on it.  If the kernel driver does not wish to request sense, 
-#  the sequencer program counter is incremented by 1, preventing another run 
-#  on the current SCB and the command is allowed to complete.  We don't 
+#  the old one and set the SENSE sequencer flag.  If the sense flag is
+#  set, the sequencer imediately jumps to start working on the sense
+#  command.  If the kernel driver does not wish to request sense, it need
+#  do nothing, and the command is allowed to complete.  We don't 
 #  bother to post to the QOUTFIFO in the error case since it would require 
 #  extra work in the kernel driver to ensure that the entry was removed 
 #  before the command complete code tried processing it.
 
+	test	SCBARRAY+15,0xff	jnz resid
+	test	SCBARRAY+16,0xff	jnz resid
+	test	SCBARRAY+17,0xff	jnz resid
+
+check_status:
 	test	SCBARRAY+14,0xff	jz status_ok	# 0 Status?
 	mvi	INTSTAT,BAD_STATUS			# let driver know
 	test	FLAGS,SENSE	jz status_ok
 	jmp	p_mesgin_done
 
 status_ok:
-
 #  First, mark this target as free.
 	test	SCBARRAY+0,0x20	jnz complete		# Tagged command
 	and	FUNCTION1,0x70,SCBARRAY+1
@@ -639,9 +648,20 @@ clear_a:
 
 complete:
 	mov	QOUTFIFO,SCBPTR
-	mvi	INTSTAT,0x02			# CMDCMPLT
+	mvi	INTSTAT,CMDCMPLT
 	jmp	p_mesgin_done
 
+# If we have a residual count, interrupt and tell the host.  Other
+# alternatives are to pause the sequencer on all command completes (yuck),
+# dma the resid directly to the host (slick, but a ton of instructions), or
+# have the sequencer pause itself when it encounters a non-zero resid 
+# (unecessary pause just to flag the command -- yuck, but takes few instructions
+# and since it shouldn't happen that offten is good enough for our purposes).  
+
+resid:
+	mvi	INTSTAT,RESIDUAL
+	jmp	check_status
+
 #  Is it an extended message?  We only support the synchronous and wide data
 #  transfer request messages, which will probably be in response to
 #  WDTR or SDTR message outs from us.  If it's not SDTR or WDTR, reject it -
@@ -836,6 +856,16 @@ bcopy_4:
 	mov	DINDIR,SINDIR
 	mov	DINDIR,SINDIR	ret
 	
+bcopy_3_dfdat:
+	mov	DINDIR,DFDAT
+	mov	DINDIR,DFDAT
+	mov	DINDIR,DFDAT	ret
+
+bcopy_4_dfdat:
+	mov	DINDIR,DFDAT
+	mov	DINDIR,DFDAT
+	mov	DINDIR,DFDAT
+	mov	DINDIR,DFDAT	ret
 
 #  Locking the driver out, build a one-byte message passed in SINDEX
 #  if there is no active message already.  SINDEX is returned intact.
@@ -966,6 +996,14 @@ dma6:
 
 	ret
 
+dma_finish:
+	test	DFSTATUS,0x8	jz dma_finish	# HDONE
+
+	clr	DFCNTRL				# disable DMA
+dma_finish2:
+	test	DFCNTRL,0x8	jnz dma_finish2	# HDMAENACK
+	ret
+
 #  Common SCSI initialization for selection and reselection.  Expects
 #  the target SCSI ID to be in the upper four bits of SINDEX, and A's
 #  contents are stomped on return.
@@ -1003,13 +1041,10 @@ initialize:
 #  message.
 #
 assert:
-	test	FLAGS,RESELECTED	jz assert1	# reselected?
-	test	FLAGS,IDENTIFY_SEEN	jnz assert1	# seen IDENTIFY?
-
-	mvi	INTSTAT,NO_IDENT		# no - cause a kernel panic
+	test	FLAGS,RESELECTED	jz return	# reselected?
+	test	FLAGS,IDENTIFY_SEEN	jnz return	# seen IDENTIFY?
 
-assert1:
-	ret
+	mvi	INTSTAT,NO_IDENT 	ret	# no - cause a kernel panic
 
 #  Find out if disconnection is ok from the information the BIOS has left
 #  us.  The tcl from SCBARRAY+1 should be in SINDEX; A will
@@ -1067,11 +1102,8 @@ sg_scb2ram:
 	mvi	SCBARRAY+3	call bcopy_4
 
 	mvi	SG_NOLOAD,0x80
-	test	SCBARRAY+0,0x10	jnz sg_scb2ram1	# don't reload s/g?
-	clr	SG_NOLOAD
-
-sg_scb2ram1:
-	ret
+	test	SCBARRAY+0,0x10	jnz return	# don't reload s/g?
+	clr	SG_NOLOAD	 ret
 
 #  Copying RAM values back to SCB, for Save Data Pointers message.
 #
@@ -1082,11 +1114,8 @@ sg_ram2scb:
 	mvi	SG_NEXT		call bcopy_4
 
 	and	SCBARRAY+0,0xef,SCBARRAY+0
-	test	SG_NOLOAD,0x80	jz sg_ram2scb1	# reload s/g?
-	or	SCBARRAY+0,SG_LOAD
-
-sg_ram2scb1:
-	ret
+	test	SG_NOLOAD,0x80	jz return	# reload s/g?
+	or	SCBARRAY+0,SG_LOAD	 ret
 
 #  Load a struct scatter if needed and set up the data address and
 #  length.  If the working value of the SG count is nonzero, then
@@ -1095,8 +1124,8 @@ sg_ram2scb1:
 #  This, like the above DMA, assumes a little-endian host data storage.
 #
 sg_load:
-	test	SG_COUNT,0xff	jz sg_load3	# SG being used?
-	test	SG_NOLOAD,0x80	jnz sg_load3	# don't reload s/g?
+	test	SG_COUNT,0xff	jz return	# SG being used?
+	test	SG_NOLOAD,0x80	jnz return	# don't reload s/g?
 
 	clr	HCNT+2
 	clr	HCNT+1
@@ -1110,12 +1139,8 @@ sg_load:
 #  Wait for DMA from host memory to data FIFO to complete, then disable
 #  DMA and wait for it to acknowledge that it's off.
 #
-sg_load1:
-	test	DFSTATUS,0x8	jz sg_load1	# HDONE
 
-	clr	DFCNTRL				# disable DMA
-sg_load2:
-	test	DFCNTRL,0x8	jnz sg_load2	# HDMAENACK
+	call	dma_finish
 
 #  Copy data from FIFO into SCB data pointer and data count.  This assumes
 #  that the struct scatterlist has this structure (this and sizeof(struct
@@ -1128,7 +1153,7 @@ sg_load2:
 #	}
 #
 
-# Not in FreeBSD.  the scatter list is only 8 bytes.
+# Not in FreeBSD.  the scatter list entry is only 8 bytes.
 # 
 # struct ahc_dma_seg {
 #       physaddr addr;                  /* four bytes, little-endian order */
@@ -1136,10 +1161,8 @@ sg_load2:
 # };
 #
 
-	mov	SCBARRAY+19,DFDAT		# new data address
-	mov	SCBARRAY+20,DFDAT
-	mov	SCBARRAY+21,DFDAT
-	mov	SCBARRAY+22,DFDAT
+	mvi	DINDEX, SCBARRAY+19
+	call	bcopy_4_dfdat
 
 # For Linux, we must throw away four bytes since there is a 32bit gap
 # in the middle of a struct scatterlist
@@ -1148,11 +1171,7 @@ sg_load2:
 #	mov	NONE,DFDAT
 #	mov	NONE,DFDAT
 
-	mov	SCBARRAY+23,DFDAT
-	mov	SCBARRAY+24,DFDAT
-	mov	SCBARRAY+25,DFDAT		#Only support 24 bit length.
-
-sg_load3:
+	call	bcopy_3_dfdat		#Only support 24 bit length.
 	ret
 
 #  Advance the scatter-gather pointers only IF NEEDED.  If SG is enabled,
@@ -1163,7 +1182,7 @@ sg_load3:
 #  next time.
 #
 sg_advance:
-	test	SG_COUNT,0xff	jz sg_advance2	# s/g enabled?
+	test	SG_COUNT,0xff	jz return	# s/g enabled?
 
 	test	STCNT+0,0xff	jnz sg_advance1	# SCSI transfer count nonzero?
 	test	STCNT+1,0xff	jnz sg_advance1
@@ -1179,9 +1198,7 @@ sg_advance:
 	adc	SG_NEXT+3,A,SG_NEXT+3	ret
 
 sg_advance1:
-	mvi	SG_NOLOAD,0x80			# don't reload s/g next time
-sg_advance2:
-	ret
+	mvi	SG_NOLOAD,0x80	ret		# don't reload s/g next time
 
 #  Add the array base SYNCNEG to the target offset (the target address
 #  is in SCSIID), and return the result in SINDEX.  The accumulator
@@ -1204,31 +1221,38 @@ ndx_dtr_2:
 #  reject, you wouldn't be able to tell which message was the culpret.
 #
 mk_dtr:
-	mov	DINDEX,SINDEX			# save SINDEX
-
+	test	SCBARRAY+0,0xc0 jz return	# NEEDWDTR|NEEDSDTR
 	test	SCBARRAY+0,NEEDWDTR jnz  mk_wdtr_16bit
-	jmp	mk_sdtr
-
-mk_wdtr_16bit:
-	mvi	ARG_1,BUS_16_BIT
-mk_wdtr:
-	mvi	DINDIR,1			# extended message
-	mvi	DINDIR,2			# extended message length = 2
-	mvi	DINDIR,3			# WDTR code
-	mov	DINDIR,ARG_1			# bus width
+	or	FLAGS, MAX_SYNC		 # Force an offset of 15
 
-	add	MSG_LEN,-MSG_START+0,DINDEX ret	# update message length
-	
 mk_sdtr:
 	mvi	DINDIR,1			# extended message
 	mvi	DINDIR,3			# extended message length = 3
 	mvi	DINDIR,1			# SDTR code
 	call	sdtr_to_rate
 	mov	DINDIR,RETURN_1			# REQ/ACK transfer period
+	test	FLAGS, MAX_SYNC	jnz mk_sdtr_max_sync
 	and	DINDIR,0xf,SINDIR		# Sync Offset
 
+mk_sdtr_done:
 	add	MSG_LEN,-MSG_START+0,DINDEX ret	# update message length
 
+mk_sdtr_max_sync:
+# We're initiating sync negotiation, so request the max offset we can (15)
+	mvi	DINDIR, 0x0f
+	xor	FLAGS, MAX_SYNC
+	jmp	mk_sdtr_done
+
+mk_wdtr_16bit:
+	mvi	ARG_1,BUS_16_BIT
+mk_wdtr:
+	mvi	DINDIR,1			# extended message
+	mvi	DINDIR,2			# extended message length = 2
+	mvi	DINDIR,3			# WDTR code
+	mov	DINDIR,ARG_1			# bus width
+
+	add	MSG_LEN,-MSG_START+0,DINDEX ret	# update message length
+	
 #  Set SCSI bus control signal state.  This also saves the last-written
 #  value into a location where the higher-level driver can read it - if
 #  it has to send an ABORT or RESET message, then it needs to know this
@@ -1254,3 +1278,6 @@ sdtr_to_rate_loop:
 sdtr_to_rate_done:
 	shr	RETURN_1,0x2
 	add	RETURN_1,0x18	ret
+
+return:
+	ret
diff --git a/sys/dev/aic7xxx/aic7xxx_asm.c b/sys/dev/aic7xxx/aic7xxx_asm.c
index 9eb9630..48f4f67 100644
--- a/sys/dev/aic7xxx/aic7xxx_asm.c
+++ b/sys/dev/aic7xxx/aic7xxx_asm.c
@@ -26,7 +26,7 @@
  *  A <label> is an <undef-sym> ending in a colon.  Spaces, tabs, and commas
  *  are token separators.
  *	
- *	$Id: aic7xxx.c,v 1.5 1995/01/22 00:46:52 gibbs Exp $
+ *	$Id: aic7xxx.c,v 1.6 1995/03/17 23:54:16 gibbs Exp $
  */
 
 /* #define _POSIX_SOURCE	1 */
@@ -38,7 +38,7 @@
 #include <stdlib.h>
 #include <unistd.h>
 
-#define MEMORY		512		/* 2^9 29-bit words */
+#define MEMORY		448
 #define MAXLINE		1024
 #define MAXTOKEN	32
 #define ADOTOUT		"a.out"
@@ -480,6 +480,9 @@ int crack(char **a, int n)
 	I_dest	= eval_sdi(a, instr[i].dest);
 	I_addr	= eval_addr(a, instr[i].addr);
 
+	if( LC >= MEMORY )
+		error("Memory exhausted!\n");
+
 	switch (instr[i].fmt) {
 	    case 1:
 	    case 2:
diff --git a/sys/dev/aic7xxx/aicasm.c b/sys/dev/aic7xxx/aicasm.c
index 9eb9630..48f4f67 100644
--- a/sys/dev/aic7xxx/aicasm.c
+++ b/sys/dev/aic7xxx/aicasm.c
@@ -26,7 +26,7 @@
  *  A <label> is an <undef-sym> ending in a colon.  Spaces, tabs, and commas
  *  are token separators.
  *	
- *	$Id: aic7xxx.c,v 1.5 1995/01/22 00:46:52 gibbs Exp $
+ *	$Id: aic7xxx.c,v 1.6 1995/03/17 23:54:16 gibbs Exp $
  */
 
 /* #define _POSIX_SOURCE	1 */
@@ -38,7 +38,7 @@
 #include <stdlib.h>
 #include <unistd.h>
 
-#define MEMORY		512		/* 2^9 29-bit words */
+#define MEMORY		448
 #define MAXLINE		1024
 #define MAXTOKEN	32
 #define ADOTOUT		"a.out"
@@ -480,6 +480,9 @@ int crack(char **a, int n)
 	I_dest	= eval_sdi(a, instr[i].dest);
 	I_addr	= eval_addr(a, instr[i].addr);
 
+	if( LC >= MEMORY )
+		error("Memory exhausted!\n");
+
 	switch (instr[i].fmt) {
 	    case 1:
 	    case 2:
diff --git a/sys/dev/aic7xxx/aicasm/aicasm.c b/sys/dev/aic7xxx/aicasm/aicasm.c
index 9eb9630..48f4f67 100644
--- a/sys/dev/aic7xxx/aicasm/aicasm.c
+++ b/sys/dev/aic7xxx/aicasm/aicasm.c
@@ -26,7 +26,7 @@
  *  A <label> is an <undef-sym> ending in a colon.  Spaces, tabs, and commas
  *  are token separators.
  *	
- *	$Id: aic7xxx.c,v 1.5 1995/01/22 00:46:52 gibbs Exp $
+ *	$Id: aic7xxx.c,v 1.6 1995/03/17 23:54:16 gibbs Exp $
  */
 
 /* #define _POSIX_SOURCE	1 */
@@ -38,7 +38,7 @@
 #include <stdlib.h>
 #include <unistd.h>
 
-#define MEMORY		512		/* 2^9 29-bit words */
+#define MEMORY		448
 #define MAXLINE		1024
 #define MAXTOKEN	32
 #define ADOTOUT		"a.out"
@@ -480,6 +480,9 @@ int crack(char **a, int n)
 	I_dest	= eval_sdi(a, instr[i].dest);
 	I_addr	= eval_addr(a, instr[i].addr);
 
+	if( LC >= MEMORY )
+		error("Memory exhausted!\n");
+
 	switch (instr[i].fmt) {
 	    case 1:
 	    case 2: