freebsd-skq/sys/dev/aic7xxx/aic7xxx.seq

1293 lines
39 KiB
Plaintext
Raw Normal View History

##+M#########################################################################
# Adaptec 274x/284x/294x device driver for Linux and FreeBSD.
#
# Copyright (c) 1994 John Aycock
# The University of Calgary Department of Computer Science.
# All rights reserved.
#
# Modifications/enhancements:
# Copyright (c) 1994, 1995 Justin Gibbs. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# 1. Redistributions of source code must retain the above copyright
# notice, this list of conditions, and the following disclaimer.
# 2. Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# 3. All advertising materials mentioning features or use of this software
# must display the following acknowledgement:
# This product includes software developed by the University of Calgary
# Department of Computer Science and its contributors.
# 4. Neither the name of the University nor the names of its contributors
# may be used to endorse or promote products derived from this software
# without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
# ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
# ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
# OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
# HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
# LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
# OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
# SUCH DAMAGE.
#
# FreeBSD, Twin, Wide, 2 command per target support, tagged queuing and other
# optimizations provided by Justin T. Gibbs (gibbs@FreeBSD.org)
#
##-M#########################################################################
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
VERSION AIC7XXX_SEQ_VER "$Id: aic7xxx.seq,v 1.15 1995/04/27 17:44:27 gibbs Exp $"
SCBMASK = 0x1f
SCSISEQ = 0x00
1995-04-27 17:44:27 +00:00
ENRSELI = 0x10
SXFRCTL0 = 0x01
SXFRCTL1 = 0x02
SCSISIGI = 0x03
SCSISIGO = 0x03
SCSIRATE = 0x04
SCSIID = 0x05
SCSIDATL = 0x06
STCNT = 0x08
STCNT+0 = 0x08
STCNT+1 = 0x09
STCNT+2 = 0x0a
1995-04-27 17:44:27 +00:00
CLRSINT0 = 0x0b
SSTAT0 = 0x0b
1995-04-27 17:44:27 +00:00
SELDO = 0x40
SELDI = 0x20
CLRSINT1 = 0x0c
SSTAT1 = 0x0c
SIMODE1 = 0x11
SCSIBUSL = 0x12
SHADDR = 0x14
SELID = 0x19
SBLKCTL = 0x1f
SEQCTL = 0x60
A = 0x64 # == ACCUM
SINDEX = 0x65
DINDEX = 0x66
ALLZEROS = 0x6a
NONE = 0x6a
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
HCNT+2 = 0x8e
SCBPTR = 0x90
INTSTAT = 0x91
DFCNTRL = 0x93
DFSTATUS = 0x94
DFDAT = 0x99
QINFIFO = 0x9b
QINCNT = 0x9c
QOUTFIFO = 0x9d
SCSICONF_A = 0x5a
SCSICONF_B = 0x5b
# The two reserved bytes at SCBARRAY+1[23] are expected to be set to
# zero, and the reserved bit in SCBARRAY+0 is used as an internal flag
# to indicate whether or not to reload scatter-gather parameters after
# a disconnect. We also use bits 6 & 7 to indicate whether or not to
# initiate SDTR or WDTR repectively when starting this command.
#
SCBARRAY+0 = 0xa0
DISCONNECTED = 0x04
NEEDDMA = 0x08
SG_LOAD = 0x10
TAG_ENB = 0x20
NEEDSDTR = 0x40
NEEDWDTR = 0x80
SCBARRAY+1 = 0xa1
SCBARRAY+2 = 0xa2
SCBARRAY+3 = 0xa3
SCBARRAY+4 = 0xa4
SCBARRAY+5 = 0xa5
SCBARRAY+6 = 0xa6
SCBARRAY+7 = 0xa7
SCBARRAY+8 = 0xa8
SCBARRAY+9 = 0xa9
SCBARRAY+10 = 0xaa
SCBARRAY+11 = 0xab
SCBARRAY+12 = 0xac
SCBARRAY+13 = 0xad
SCBARRAY+14 = 0xae
SCBARRAY+15 = 0xaf
SCBARRAY+16 = 0xb0
SCBARRAY+17 = 0xb1
SCBARRAY+18 = 0xb2
SCBARRAY+19 = 0xb3
SCBARRAY+20 = 0xb4
SCBARRAY+21 = 0xb5
SCBARRAY+22 = 0xb6
SCBARRAY+23 = 0xb7
SCBARRAY+24 = 0xb8
SCBARRAY+25 = 0xb9
SCBARRAY+26 = 0xba
SCBARRAY+27 = 0xbb
SCBARRAY+28 = 0xbc
SCBARRAY+29 = 0xbd
1995-04-27 17:44:27 +00:00
SCBARRAY+30 = 0xbe
BAD_PHASE = 0x01 # unknown scsi bus phase
1995-04-27 17:44:27 +00:00
CMDCMPLT = 0x02 # Command Complete
SEND_REJECT = 0x11 # sending a message reject
NO_IDENT = 0x21 # no IDENTIFY after reconnect
NO_MATCH = 0x31 # no cmd match for reconnect
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
ABORT_TAG = 0x91 # Sent an ABORT_TAG message
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
AWAITING_MSG = 0xa1 # Kernel requested to specify
# a message to this target
# (command was null), so tell
# it that it can fill the
# message buffer.
# 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
# BIOS trashes 20-2f, writing the synchronous negotiation results
# on top of the BIOS values, so we re-use those for our per-target
# scratchspace (actually a value that can be copied directly into
# SCSIRATE). The kernel driver will enable synchronous negotiation
# for all targets that have a value other than 0 in the lower four
# bits of the target scratch space. This should work irregardless of
# whether the bios has been installed. NEEDWDTR and NEEDSDTR are the top
# two bits of the SCB control byte. The kernel driver will set these
# when a WDTR or SDTR message should be sent to the target the SCB's
# command references.
#
1995-04-27 17:44:27 +00:00
# REJBYTE contains the first byte of a MESSAGE IN message, so the driver
# can report an intelligible error if a message is rejected.
#
# FLAGS's high bit is true if we are currently handling a reselect;
# its next-highest bit is true ONLY IF we've seen an IDENTIFY message
# from the reselecting target. If we haven't had IDENTIFY, then we have
# no idea what the lun is, and we can't select the right SCB register
# bank, so force a kernel panic if the target attempts a data in/out or
1995-04-27 17:44:27 +00:00
# command phase instead of corrupting something. FLAGS also contains
# configuration bits so that we can optimize for TWIN and WIDE controllers
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
# as well as the MAX_OFFSET bit which we set when we want to negotiate for
# maximum sync offset irregardless of what the per target scratch space says.
#
# Note that SG_NEXT occupies four bytes.
#
SYNCNEG = 0x20
REJBYTE = 0x31
DISC_DSB_A = 0x32
DISC_DSB_B = 0x33
MSG_LEN = 0x34
MSG_START+0 = 0x35
MSG_START+1 = 0x36
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
ARG_1 = 0x4a # sdtr conversion args & return
BUS_16_BIT = 0x01
RETURN_1 = 0x4a
SIGSTATE = 0x4b # value written to SCSISIGO
# Linux users should use 0xc (12) for SG_SIZEOF
SG_SIZEOF = 0x8 # sizeof(struct ahc_dma)
#SG_SIZEOF = 0xc # sizeof(struct scatterlist)
SCB_SIZEOF = 0x13 # sizeof SCB to DMA (19 bytes)
SG_NOLOAD = 0x4c # load SG pointer/length?
SG_COUNT = 0x4d # working value of SG count
SG_NEXT = 0x4e # working value of SG pointer
SG_NEXT+0 = 0x4e
SG_NEXT+1 = 0x4f
SG_NEXT+2 = 0x50
SG_NEXT+3 = 0x51
SCBCOUNT = 0x52 # the actual number of SCBs
FLAGS = 0x53 # Device configuration flags
TWIN_BUS = 0x01
WIDE_BUS = 0x02
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
MAX_OFFSET = 0x08
ACTIVE_MSG = 0x20
IDENTIFY_SEEN = 0x40
RESELECTED = 0x80
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
MAX_OFFSET_8BIT = 0x0f
MAX_OFFSET_WIDE = 0x08
ACTIVE_A = 0x54
ACTIVE_B = 0x55
1995-04-27 17:44:27 +00:00
SAVED_TCL = 0x56 # Temporary storage for the
# target/channel/lun of a
# reconnecting target
# After starting the selection hardware, we return to the "poll_for_work"
# loop so that we can check for reconnecting targets as well as for our
# selection to complete just in case the reselection wins bus arbitration.
# The problem with this is that we must keep track of the SCB that we've
# already pulled from the QINFIFO and started the selection on just in case
# the reselection wins so that we can retry the selection at a later time.
# This problem cannot be resolved by holding a single entry in scratch
# ram since a reconnecting target can request sense and this will create
# yet another SCB waiting for selection. The solution used here is to
# use byte 31 of the SCB as a psuedo-next pointer and to thread a list
# of SCBs that are awaiting selection. Since 0 is a valid SCB offset,
# SCB_LIST_NULL is 0x10 which is out of range. The kernel driver must
# add an entry to this list everytime a request sense occurs. The sequencer
# will automatically consume the entries.
WAITING_SCBH = 0x57 # head of list of SCBs awaiting
# selection
WAITING_SCBT = 0x58 # tail of list of SCBs awaiting
# selection
SCB_LIST_NULL = 0x10
# Poll QINCNT for work - the lower bits contain
# the number of entries in the Queue In FIFO.
#
start:
1995-04-27 17:44:27 +00:00
test WAITING_SCBH,SCB_LIST_NULL jz start_waiting
poll_for_work:
test FLAGS,TWIN_BUS jz start2 # Are we a twin channel device?
# For fairness, we check the other bus first, since we just finished a
# transaction on the current channel.
xor SBLKCTL,0x08 # Toggle to the other bus
1995-04-27 17:44:27 +00:00
test SSTAT0,SELDI jnz reselect
test SSTAT0,SELDO jnz select
xor SBLKCTL,0x08 # Toggle to the original bus
start2:
1995-04-27 17:44:27 +00:00
test SSTAT0,SELDI jnz reselect
test SSTAT0,SELDO jnz select
test WAITING_SCBH,SCB_LIST_NULL jz start_waiting
test QINCNT,SCBMASK jz poll_for_work
1995-04-27 17:44:27 +00:00
# We have at least one queued SCB now and we don't have any
# SCBs in the list of SCBs awaiting selection. Set the SCB
# pointer from the FIFO so we see the right bank of SCB
# registers, then set SCSI options and set the initiator and
# target SCSI IDs.
#
mov SCBPTR,QINFIFO
# If the control byte of this SCB has the NEEDDMA flag set, we have
# yet to DMA it from host memory
1995-04-27 17:44:27 +00:00
test SCBARRAY+0,NEEDDMA jz test_busy
clr HCNT+2
clr HCNT+1
mvi HCNT+0,SCB_SIZEOF
1995-04-27 17:44:27 +00:00
mvi DINDEX,HADDR
mvi SCBARRAY+26 call bcopy_4
mvi DFCNTRL,0xd # HDMAEN|DIRECTION|FIFORESET
# Wait for DMA from host memory to data FIFO to complete, then disable
# DMA and wait for it to acknowledge that it's off.
#
call dma_finish
# Copy the SCB from the FIFO to the SCBARRAY
mvi DINDEX, SCBARRAY+0
1995-04-27 17:44:27 +00:00
call bcopy_3_dfdat
call bcopy_4_dfdat
call bcopy_4_dfdat
1995-04-27 17:44:27 +00:00
call bcopy_4_dfdat
call bcopy_4_dfdat
1995-04-27 17:44:27 +00:00
# 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
# is not ideal for devices that have multiple luns active at the same
# time, it is faster than looping through all SCB's looking for active
# commands. It may be benificial to make findscb a more general procedure
# to see if the added cost of the search is negligible. This code also
# assumes that the kernel driver will clear the active flags on board
# initialization, board reset, and a target's SELTO.
test_busy:
test SCBARRAY+0,0x20 jnz start_scb
and FUNCTION1,0x70,SCBARRAY+1
mov A,FUNCTION1
test SCBARRAY+1,0x88 jz test_a # Id < 8 && A channel
test ACTIVE_B,A jnz requeue
or ACTIVE_B,A # Mark the current target as busy
jmp start_scb
# Place the currently active back on the queue for later processing
requeue:
mov QINFIFO, SCBPTR
1995-04-27 17:44:27 +00:00
jmp poll_for_work
# Pull the first entry off of the waiting for selection list
start_waiting:
mov SCBPTR,WAITING_SCBH
jmp start_scb
test_a:
test ACTIVE_A,A jnz requeue
or ACTIVE_A,A # Mark the current target as busy
start_scb:
and SINDEX,0xf7,SBLKCTL #Clear the channel select bit
and A,0x08,SCBARRAY+1 #Get new channel bit
or SINDEX,A
mov SBLKCTL,SINDEX # select channel
1995-04-27 17:44:27 +00:00
mov SCBARRAY+1 call initialize_scsiid
# Enable selection phase as an initiator, and do automatic ATN
# after the selection. We do this now so that we can overlap the
# rest of our work to set up this target with the arbitration and
# selection bus phases.
#
start_selection:
or SCSISEQ,0x48 # ENSELO|ENAUTOATNO
mov WAITING_SCBH, SCBPTR
clr SG_NOLOAD
and FLAGS,0x3f # !RESELECTING
# As soon as we get a successful selection, the target should go
# into the message out phase since we have ATN asserted. Prepare
# the message to send, locking out the device driver. If the device
# driver hasn't beaten us with an ABORT or RESET message, then tack
# on an SDTR negotiation if required.
#
# Messages are stored in scratch RAM starting with a flag byte (high bit
# set means active message), one length byte, and then the message itself.
#
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
test SCBARRAY+11,0xff jnz identify # 0 Length Command?
# The kernel has sent us an SCB with no command attached. This implies
# that the kernel wants to send a message of some sort to this target,
# so we interrupt the driver, allow it to fill the message buffer, and
# then go back into the arbitration loop
mvi INTSTAT,AWAITING_MSG
jmp poll_for_work
identify:
mov SCBARRAY+1 call disconnect # disconnect ok?
and SINDEX,0x7,SCBARRAY+1 # lun
or SINDEX,A # return value from disconnect
or SINDEX,0x80 call mk_mesg # IDENTIFY message
mov A,SINDEX
test SCBARRAY+0,0xe0 jz !message # WDTR, SDTR or TAG??
cmp MSG_START+0,A jne !message # did driver beat us?
# Tag Message if Tag enabled in SCB control block. Use SCBPTR as the tag
# value
mk_tag:
mvi DINDEX, MSG_START+1
test SCBARRAY+0,TAG_ENB jz mk_tag_done
and A,0x23,SCBARRAY+0
mov DINDIR,A
mov DINDIR,SCBPTR
add MSG_LEN,-MSG_START+0,DINDEX # update message length
mk_tag_done:
mov DINDEX call mk_dtr # build DTR message if needed
!message:
1995-04-27 17:44:27 +00:00
jmp poll_for_work
1995-04-27 17:44:27 +00:00
# Reselection has been initiated by a target. Make a note that we've been
# reselected, but haven't seen an IDENTIFY message from the target
# yet.
#
reselect:
1995-04-27 17:44:27 +00:00
mov SELID call initialize_scsiid
and FLAGS,0x3f # reselected, no IDENTIFY
1995-04-27 17:44:27 +00:00
or FLAGS,RESELECTED jmp select2
1995-04-27 17:44:27 +00:00
# After the selection, remove this SCB from the "waiting for selection"
# list. This is achieved by simply moving our "next" pointer into
# WAITING_SCBH and setting our next pointer to null so that the next
# time this SCB is used, we don't get confused.
#
1995-04-27 17:44:27 +00:00
select:
or SCBARRAY+0,NEEDDMA
mov WAITING_SCBH,SCBARRAY+30
mvi SCBARRAY+30,SCB_LIST_NULL
select2:
call initialize_for_target
mvi SCSISEQ,ENRSELI
mvi CLRSINT0,0x60 # CLRSELDI|CLRSELDO
mvi CLRSINT1,0x8 # CLRBUSFREE
# Main loop for information transfer phases. If BSY is false, then
# we have a bus free condition, expected or not. Otherwise, wait
# for the target to assert REQ before checking MSG, C/D and I/O
# for the bus phase.
#
# We can't simply look at the values of SCSISIGI here (if we want
# to do synchronous data transfer), because the target won't assert
# REQ if it's already sent us some data that we haven't acknowledged
# yet.
#
ITloop:
test SSTAT1,0x8 jnz p_busfree # BUSFREE
test SSTAT1,0x1 jz ITloop # REQINIT
and A,0xe0,SCSISIGI # CDI|IOI|MSGI
cmp ALLZEROS,A je p_dataout
cmp A,0x40 je p_datain
cmp A,0x80 je p_command
cmp A,0xc0 je p_status
cmp A,0xa0 je p_mesgout
cmp A,0xe0 je p_mesgin
mvi INTSTAT,BAD_PHASE # unknown - signal driver
p_dataout:
mvi 0 call scsisig # !CDO|!IOO|!MSGO
call assert
call sg_load
mvi DINDEX,HADDR
mvi SCBARRAY+19 call bcopy_4
# mvi DINDEX,HCNT # implicit since HCNT is next to HADDR
mvi SCBARRAY+23 call bcopy_3
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
# it's a reflection of how many bytes were transferred on the SCSI
# (as opposed to the host) bus.
#
mvi DINDEX,SCBARRAY+23
mvi STCNT call bcopy_3
mvi DINDEX,SCBARRAY+19
mvi SHADDR call bcopy_4
call sg_advance
mov SCBARRAY+18,SG_COUNT # residual S/G count
jmp ITloop
p_datain:
mvi 0x40 call scsisig # !CDO|IOO|!MSGO
call assert
call sg_load
mvi DINDEX,HADDR
mvi SCBARRAY+19 call bcopy_4
# mvi DINDEX,HCNT # implicit since HCNT is next to HADDR
mvi SCBARRAY+23 call bcopy_3
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
mvi DINDEX,SCBARRAY+19
mvi SHADDR call bcopy_4
call sg_advance
mov SCBARRAY+18,SG_COUNT # residual S/G count
jmp ITloop
# Command phase. Set up the DMA registers and let 'er rip - the
# two bytes after the SCB SCSI_cmd_length are zeroed by the driver,
# so we can copy those three bytes directly into HCNT.
#
p_command:
mvi 0x80 call scsisig # CDO|!IOO|!MSGO
call assert
mvi DINDEX,HADDR
mvi SCBARRAY+7 call bcopy_4
# mvi DINDEX,HCNT # implicit since HCNT is next to HADDR
mvi SCBARRAY+11 call bcopy_3
mvi DINDEX,STCNT
mvi SCBARRAY+11 call bcopy_3
mvi 0x3d call dma # SCSIEN|SDMAEN|HDMAEN|
# DIRECTION|FIFORESET
jmp ITloop
# Status phase. Wait for the data byte to appear, then read it
# and store it into the SCB.
#
p_status:
mvi 0xc0 call scsisig # CDO|IOO|!MSGO
mvi SCBARRAY+14 call inb_first
jmp p_mesgin_done
# Message out phase. If there is no active message, but the target
# took us into this phase anyway, build a no-op message and send it.
#
p_mesgout:
mvi 0xa0 call scsisig # CDO|!IOO|MSGO
mvi 0x8 call mk_mesg # build NOP message
clr STCNT+2
clr STCNT+1
# Set up automatic PIO transfer from MSG_START. Bit 3 in
# SXFRCTL0 (SPIOEN) is already on.
#
mvi SINDEX,MSG_START+0
mov DINDEX,MSG_LEN
# When target asks for a byte, drop ATN if it's the last one in
# the message. Otherwise, keep going until the message is exhausted.
# (We can't use outb for this since it wants the input in SINDEX.)
#
# Keep an eye out for a phase change, in case the target issues
# a MESSAGE REJECT.
#
p_mesgout2:
test SSTAT0,0x2 jz p_mesgout2 # SPIORDY
test SSTAT1,0x10 jnz p_mesgout6 # PHASEMIS
cmp DINDEX,1 jne p_mesgout3 # last byte?
mvi CLRSINT1,0x40 # CLRATNO - drop ATN
# Write a byte to the SCSI bus. The AIC-7770 refuses to automatically
# send ACKs in automatic PIO or DMA mode unless you make sure that the
# "expected" bus phase in SCSISIGO matches the actual bus phase. This
# behaviour is completely undocumented and caused me several days of
# grief.
#
# After plugging in different drives to test with and using a longer
# SCSI cable, I found that I/O in Automatic PIO mode ceased to function,
# especially when transferring >1 byte. It seems to be much more stable
# if STCNT is set to one before the transfer, and SDONE (in SSTAT0) is
# polled for transfer completion - for both output _and_ input. The
# only theory I have is that SPIORDY doesn't drop right away when SCSIDATL
# is accessed (like the documentation says it does), and that on a longer
# cable run, the sequencer code was fast enough to loop back and see
# an SPIORDY that hadn't dropped yet.
#
p_mesgout3:
mvi STCNT+0, 0x01
mov SCSIDATL,SINDIR
p_mesgout4:
test SSTAT0,0x4 jz p_mesgout4 # SDONE
dec DINDEX
test DINDEX,0xff jnz p_mesgout2
# If the next bus phase after ATN drops is a message out, it means
# that the target is requesting that the last message(s) be resent.
#
p_mesgout5:
test SSTAT1,0x8 jnz p_mesgout6 # BUSFREE
test SSTAT1,0x1 jz p_mesgout5 # REQINIT
and A,0xe0,SCSISIGI # CDI|IOI|MSGI
cmp A,0xa0 jne p_mesgout6
mvi 0x10 call scsisig # ATNO - re-assert ATN
jmp ITloop
p_mesgout6:
mvi CLRSINT1,0x40 # CLRATNO - in case of PHASEMIS
and FLAGS,0xdf # no active msg
jmp ITloop
# Message in phase. Bytes are read using Automatic PIO mode, but not
# using inb. This alleviates a race condition, namely that if ATN had
# to be asserted under Automatic PIO mode, it had to beat the SCSI
# circuitry sending an ACK to the target. This showed up under heavy
# loads and really confused things, since ABORT commands wouldn't be
# seen by the drive after an IDENTIFY message in until it had changed
# to a data I/O phase.
#
p_mesgin:
mvi 0xe0 call scsisig # CDO|IOO|MSGO
mvi A call inb_first # read the 1st message byte
mvi REJBYTE,A # save it for the driver
cmp ALLZEROS,A jne p_mesgin1
# We got a "command complete" message, so put the SCB pointer
# into the Queue Out, and trigger a completion interrupt.
# Check status for non zero return and interrupt driver if needed
# This allows the driver to interpret errors only when they occur
# instead of always uploading the scb. If the status is SCSI_CHECK,
1995-04-27 17:44:27 +00:00
# the driver will download a new scb requesting sense to replace
# the old one, modify the "waiting for selection" SCB list and set
# RETURN_1 to 0x80. If RETURN_1 is set to 0x80 the sequencer imediately
# jumps to main loop where it will run down the waiting SCB list.
# If the kernel driver does not wish to request sense, it need
# only clear RETURN_1, 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.
# First check for residuals
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
1995-04-27 17:44:27 +00:00
test RETURN_1, 0x80 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
mov A,FUNCTION1
test SCBARRAY+1,0x88 jz clear_a
xor ACTIVE_B,A
jmp complete
clear_a:
xor ACTIVE_A,A
complete:
mov QOUTFIFO,SCBPTR
mvi INTSTAT,CMDCMPLT
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
test SCBARRAY+11,0xff jz start # Immediate message complete
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 often 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 -
# apparently this can be done after any message in byte, according
# to the SCSI-2 spec.
#
p_mesgin1:
cmp A,1 jne p_mesgin2 # extended message code?
mvi ARG_1 call inb_next # extended message length
mvi A call inb_next # extended message code
cmp A,1 je p_mesginSDTR # Syncronous negotiation message
cmp A,3 je p_mesginWDTR # Wide negotiation message
jmp p_mesginN
p_mesginWDTR:
cmp ARG_1,2 jne p_mesginN # extended mesg length = 2
mvi A call inb_next # Width of bus
mvi INTSTAT,MSG_WDTR # let driver know
test RETURN_1,0x80 jz p_mesgin_done# Do we need to send WDTR?
# We didn't initiate the wide negotiation, so we must respond to the request
and RETURN_1,0x7f # Clear the SEND_WDTR Flag
or FLAGS,ACTIVE_MSG
mvi DINDEX,MSG_START+0
mvi MSG_START+0 call mk_wdtr # build WDTR message
or SINDEX,0x10,SIGSTATE # turn on ATNO
call scsisig
jmp p_mesgin_done
p_mesginSDTR:
cmp ARG_1,3 jne p_mesginN # extended mesg length = 3
mvi ARG_1 call inb_next # xfer period
mvi A call inb_next # REQ/ACK offset
mvi INTSTAT,MSG_SDTR # call driver to convert
test RETURN_1,0xc0 jz p_mesgin_done# Do we need to mk_sdtr or rej?
test RETURN_1,0x40 jnz p_mesginN # Requested SDTR too small - rej
or FLAGS,ACTIVE_MSG
mvi DINDEX, MSG_START+0
mvi MSG_START+0 call mk_sdtr
or SINDEX,0x10,SIGSTATE # turn on ATNO
call scsisig
jmp p_mesgin_done
# Is it a disconnect message? Set a flag in the SCB to remind us
# and await the bus going free.
#
p_mesgin2:
cmp A,4 jne p_mesgin3 # disconnect code?
or SCBARRAY+0,0x4 # set "disconnected" bit
jmp p_mesgin_done
# Save data pointers message? Copy working values into the SCB,
# usually in preparation for a disconnect.
#
p_mesgin3:
cmp A,2 jne p_mesgin4 # save data pointers code?
call sg_ram2scb
jmp p_mesgin_done
# Restore pointers message? Data pointers are recopied from the
# SCB anyway at the start of any DMA operation, so the only thing
# to copy is the scatter-gather values.
#
p_mesgin4:
cmp A,3 jne p_mesgin5 # restore pointers code?
call sg_scb2ram
jmp p_mesgin_done
# Identify message? For a reconnecting target, this tells us the lun
# that the reconnection is for - find the correct SCB and switch to it,
# clearing the "disconnected" bit so we don't "find" it by accident later.
#
p_mesgin5:
test A,0x80 jz p_mesgin6 # identify message?
test A,0x78 jnz p_mesginN # !DiscPriv|!LUNTAR|!Reserved
and A,0x07 # lun in lower three bits
or SAVED_TCL,A,SELID
and SAVED_TCL,0xf7
and A,0x08,SBLKCTL # B Channel??
or SAVED_TCL,A
call inb_last # ACK
mov ALLZEROS call findSCB
setup_SCB:
and SCBARRAY+0,0xfb # clear disconnect bit in SCB
or FLAGS,IDENTIFY_SEEN # make note of IDENTIFY
call sg_scb2ram # implied restore pointers
# required on reselect
jmp ITloop
get_tag:
mvi A call inb_first
cmp A,0x20 jne return # Simple Tag message?
mvi A call inb_next
call inb_last
test A,0xf0 jnz abort_tag # Tag in range?
mov SCBPTR,A
mov A,SAVED_TCL
cmp SCBARRAY+1,A jne abort_tag
test SCBARRAY+0,TAG_ENB jz abort_tag
ret
abort_tag:
or SINDEX,0x10,SIGSTATE # turn on ATNO
call scsisig
mvi INTSTAT,ABORT_TAG # let driver know
mvi 0xd call mk_mesg # ABORT TAG message
ret
# Message reject? Let the kernel driver handle this. If we have an
# outstanding WDTR or SDTR negotiation, assume that it's a response from
# the target selecting 8bit or asynchronous transfer, otherwise just ignore
# it since we have no clue what it pertains to.
#
p_mesgin6:
cmp A,7 jne p_mesgin7 # message reject code?
mvi INTSTAT, MSG_REJECT
jmp p_mesgin_done
# [ ADD MORE MESSAGE HANDLING HERE ]
#
p_mesgin7:
# We have no idea what this message in is, and there's no way
# to pass it up to the kernel, so we issue a message reject and
# hope for the best. Since we're now using manual PIO mode to
# read in the message, there should no longer be a race condition
# present when we assert ATN. In any case, rejection should be a
# rare occurrence - signal the driver when it happens.
#
p_mesginN:
or SINDEX,0x10,SIGSTATE # turn on ATNO
call scsisig
mvi INTSTAT,SEND_REJECT # let driver know
mvi 0x7 call mk_mesg # MESSAGE REJECT message
p_mesgin_done:
call inb_last # ack & turn auto PIO back on
jmp ITloop
# Bus free phase. It might be useful to interrupt the device
# driver if we aren't expecting this. For now, make sure that
# ATN isn't being asserted and look for a new command.
#
p_busfree:
mvi CLRSINT1,0x40 # CLRATNO
clr SIGSTATE
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
# if this is an immediate command, perform a psuedo command complete to
# notify the driver.
test SCBARRAY+11,0xff jz status_ok
jmp start
# Instead of a generic bcopy routine that requires an argument, we unroll
# the two cases that are actually used, and call them explicitly. This
# not only reduces the overhead of doing a bcopy by 2/3rds, but ends up
# saving space in the program since you don't have to put the argument
# into the accumulator before the call. Both functions expect DINDEX to
# contain the destination address and SINDEX to contain the source
# address.
bcopy_3:
mov DINDIR,SINDIR
mov DINDIR,SINDIR
mov DINDIR,SINDIR ret
bcopy_4:
mov DINDIR,SINDIR
mov DINDIR,SINDIR
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.
#
mk_mesg:
mvi SEQCTL,0x50 # PAUSEDIS|FASTMODE
test FLAGS,ACTIVE_MSG jnz mk_mesg1 # active message?
or FLAGS,ACTIVE_MSG # if not, there is now
mvi MSG_LEN,1 # length = 1
mov MSG_START+0,SINDEX # 1-byte message
mk_mesg1:
mvi SEQCTL,0x10 ret # !PAUSEDIS|FASTMODE
# Carefully read data in Automatic PIO mode. I first tried this using
# Manual PIO mode, but it gave me continual underrun errors, probably
# indicating that I did something wrong, but I feel more secure leaving
# Automatic PIO on all the time.
#
# According to Adaptec's documentation, an ACK is not sent on input from
# the target until SCSIDATL is read from. So we wait until SCSIDATL is
# latched (the usual way), then read the data byte directly off the bus
# using SCSIBUSL. When we have pulled the ATN line, or we just want to
# acknowledge the byte, then we do a dummy read from SCISDATL. The SCSI
# spec guarantees that the target will hold the data byte on the bus until
# we send our ACK.
#
# The assumption here is that these are called in a particular sequence,
# and that REQ is already set when inb_first is called. inb_{first,next}
# use the same calling convention as inb.
#
inb_first:
clr STCNT+2
clr STCNT+1
mov DINDEX,SINDEX
mov DINDIR,SCSIBUSL ret # read byte directly from bus
inb_next:
mov DINDEX,SINDEX # save SINDEX
mvi STCNT+0,1 # xfer one byte
mov NONE,SCSIDATL # dummy read from latch to ACK
inb_next1:
test SSTAT0,0x4 jz inb_next1 # SDONE
inb_next2:
test SSTAT0,0x2 jz inb_next2 # SPIORDY - wait for next byte
mov DINDIR,SCSIBUSL ret # read byte directly from bus
inb_last:
mvi STCNT+0,1 # ACK with dummy read
mov NONE,SCSIDATL
inb_last1:
test SSTAT0,0x4 jz inb_last1 # wait for completion
ret
# DMA data transfer. HADDR and HCNT must be loaded first, and
# SINDEX should contain the value to load DFCNTRL with - 0x3d for
# host->scsi, or 0x39 for scsi->host. The SCSI channel is cleared
# during initialization.
#
dma:
mov DFCNTRL,SINDEX
dma1:
dma2:
test SSTAT0,0x1 jnz dma3 # DMADONE
test SSTAT1,0x10 jz dma1 # PHASEMIS, ie. underrun
# We will be "done" DMAing when the transfer count goes to zero, or
# the target changes the phase (in light of this, it makes sense that
# the DMA circuitry doesn't ACK when PHASEMIS is active). If we are
# doing a SCSI->Host transfer, the data FIFO should be flushed auto-
# magically on STCNT=0 or a phase change, so just wait for FIFO empty
# status.
#
dma3:
test SINDEX,0x4 jnz dma5 # DIRECTION
dma4:
test DFSTATUS,0x1 jz dma4 # !FIFOEMP
# Now shut the DMA enables off, and copy STCNT (ie. the underrun
# amount, if any) to the SCB registers; SG_COUNT will get copied to
# the SCB's residual S/G count field after sg_advance is called. Make
# sure that the DMA enables are actually off first lest we get an ILLSADDR.
#
dma5:
clr DFCNTRL # disable DMA
dma6:
test DFCNTRL,0x38 jnz dma6 # SCSIENACK|SDMAENACK|HDMAENACK
mvi DINDEX,SCBARRAY+15
mvi STCNT call bcopy_3
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.
#
1995-04-27 17:44:27 +00:00
initialize_scsiid:
and SINDEX,0xf0 # Get target ID
and A,0x0f,SCSIID
or SINDEX,A
1995-04-27 17:44:27 +00:00
mov SCSIID,SINDEX ret
1995-04-27 17:44:27 +00:00
initialize_for_target:
# Turn on Automatic PIO mode now, before we expect to see a REQ
# from the target. It shouldn't hurt anything to leave it on. Set
# CLRCHN here before the target has entered a data transfer mode -
# with synchronous SCSI, if you do it later, you blow away some
# data in the SCSI FIFO that the target has already sent to you.
#
1995-04-27 17:44:27 +00:00
clr SIGSTATE
mvi SXFRCTL0,0x8a # DFON|SPIOEN|CLRCHN
# Initialize scatter-gather pointers by setting up the working copy
# in scratch RAM.
#
call sg_scb2ram
# Initialize SCSIRATE with the appropriate value for this target.
#
call ndx_dtr
mov SCSIRATE,SINDIR ret
# Assert that if we've been reselected, then we've seen an IDENTIFY
# message.
#
assert:
test FLAGS,RESELECTED jz return # reselected?
test FLAGS,IDENTIFY_SEEN jnz return # seen IDENTIFY?
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
# contain either 0x40 (disconnection ok) or 0x00 (disconnection not ok)
# on exit.
#
# To allow for wide or twin busses, we check the upper bit of the target ID
# and the channel ID and look at the appropriate disconnect register.
#
disconnect:
and FUNCTION1,0x70,SINDEX # strip off extra just in case
mov A,FUNCTION1
test SINDEX, 0x88 jz disconnect_a
test DISC_DSB_B,A jz disconnect1 # bit nonzero if DISabled
clr A ret
disconnect_a:
test DISC_DSB_A,A jz disconnect1 # bit nonzero if DISabled
clr A ret
disconnect1:
mvi A,0x40 ret
# Locate the SCB matching the target ID/channel/lun in SAVED_TCL and switch
# the SCB to it. Have the kernel print a warning message if it can't be
# found, and generate an ABORT message to the target. SINDEX should be
# cleared on call.
#
findSCB:
mov A,SAVED_TCL
mov SCBPTR,SINDEX # switch to new SCB
cmp SCBARRAY+1,A jne findSCB1 # target ID/channel/lun match?
test SCBARRAY+0,0x4 jz findSCB1 # should be disconnected
test SCBARRAY+0,TAG_ENB jnz get_tag
ret
findSCB1:
inc SINDEX
mov A,SCBCOUNT
cmp SINDEX,A jne findSCB
mvi INTSTAT,NO_MATCH # not found - signal kernel
mvi 0x6 call mk_mesg # ABORT message
or SINDEX,0x10,SIGSTATE # assert ATNO
call scsisig
ret
# Make a working copy of the scatter-gather parameters in the SCB.
#
sg_scb2ram:
mov SG_COUNT,SCBARRAY+2
mvi DINDEX,SG_NEXT
mvi SCBARRAY+3 call bcopy_4
mvi SG_NOLOAD,0x80
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.
#
sg_ram2scb:
mov SCBARRAY+2,SG_COUNT
mvi DINDEX,SCBARRAY+3
mvi SG_NEXT call bcopy_4
and SCBARRAY+0,0xef,SCBARRAY+0
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
# we need to load a new set of values.
#
# This, like the above DMA, assumes a little-endian host data storage.
#
sg_load:
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
mvi HCNT+0,SG_SIZEOF
mvi DINDEX,HADDR
mvi SG_NEXT call bcopy_4
mvi DFCNTRL,0xd # HDMAEN|DIRECTION|FIFORESET
# Wait for DMA from host memory to data FIFO to complete, then disable
# DMA and wait for it to acknowledge that it's off.
#
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
# scatterlist) == 12 are asserted in aic7xxx.c):
#
# struct scatterlist {
# char *address; /* four bytes, little-endian order */
# ... /* four bytes, ignored */
# unsigned short length; /* two bytes, little-endian order */
# }
#
# Not in FreeBSD. the scatter list entry is only 8 bytes.
#
# struct ahc_dma_seg {
# physaddr addr; /* four bytes, little-endian order */
# long len; /* four bytes, little endian order */
# };
#
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
# mov NONE,DFDAT
# mov NONE,DFDAT
# mov NONE,DFDAT
# mov NONE,DFDAT
call bcopy_3_dfdat #Only support 24 bit length.
ret
# Advance the scatter-gather pointers only IF NEEDED. If SG is enabled,
# and the SCSI transfer count is zero (note that this should be called
# right after a DMA finishes), then move the working copies of the SG
# pointer/length along. If the SCSI transfer count is not zero, then
# presumably the target is disconnecting - do not reload the SG values
# next time.
#
sg_advance:
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
test STCNT+2,0xff jnz sg_advance1
clr SG_NOLOAD # reload s/g next time
dec SG_COUNT # one less segment to go
clr A # add sizeof(struct scatter)
add SG_NEXT+0,SG_SIZEOF,SG_NEXT+0
adc SG_NEXT+1,A,SG_NEXT+1
adc SG_NEXT+2,A,SG_NEXT+2
adc SG_NEXT+3,A,SG_NEXT+3 ret
sg_advance1:
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
# contains the 3->8 decoding of the target ID on return.
#
ndx_dtr:
shr A,SCSIID,4
test SBLKCTL,0x08 jz ndx_dtr_2
or A,0x08 # Channel B entries add 8
ndx_dtr_2:
add SINDEX,SYNCNEG,A
and FUNCTION1,0x70,SCSIID # 3-bit target address decode
mov A,FUNCTION1 ret
# If we need to negotiate transfer parameters, build the WDTR or SDTR message
# starting at the address passed in SINDEX. DINDEX is modified on return.
# The SCSI-II spec requires that Wide negotiation occur first and you can
# only negotiat one or the other at a time otherwise in the event of a message
# reject, you wouldn't be able to tell which message was the culpret.
#
mk_dtr:
test SCBARRAY+0,0xc0 jz return # NEEDWDTR|NEEDSDTR
test SCBARRAY+0,NEEDWDTR jnz mk_wdtr_16bit
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
or FLAGS, MAX_OFFSET # Force an offset of 15 or 8 if WIDE
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
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
test FLAGS, MAX_OFFSET jnz mk_sdtr_max_offset
and DINDIR,0x0f,SINDIR # Sync Offset
mk_sdtr_done:
add MSG_LEN,-MSG_START+0,DINDEX ret # update message length
Fixes to the aic7xxx sequencer code and device driver from Justin Gibbs: 1) If a target initiated a sync negotiation with us and happened to chose a value above 15, the old code inadvertantly truncated it with an "& 0x0f". If the periferal picked something really bad like 0x32, you'd end up with an offset of 2 which would hang the drive since it didn't expect to ever get something so low. We now do a MIN(maxoffset, given_offset). 2) In the case of Wide cards, we were turning on sync transfers after a sucessfull wide negotiation. Now we leave the offset alone in the per target scratch space (which implies asyncronous transfers since we initialize it that way) until a syncronous negotation occurs. 3) We were advertizing a max offset of 15 instead of 8 for wide devices. 4) If the upper level SCSI code sent down a "SCSI_RESET", it would hang the system because we would end up sending a null command to the sequencer. Now we handle SCSI_RESET correctly by having the sequencer interrupt us when it is about to fill the message buffer so that we can fill it in ourselves. The sequencer will also "simulate" a command complete for these "message only" SCBs so that the kernel driver can finish up properly. The cdplay utility will send a "SCSI_REST" to the cdplayer if you use the reset command. 5) The code that handles SCSIINTs was broken in that if more than one type of error was true at once, we'd do outbs without the card being paused. The else clause after the busfree case was also an accident waiting to happen. I've now turned this into an if, else if, else type of thing, since in most cases when we handle one type of error, it should be okay to ignore the rest (ie if we have a SELTO, who cares if there was a parity error on the transaction?), but the section should really be rewritten after 2.0.5. This fix was the least obtrusive way to patch the problem. 6) Only tag either SDTR or WDTR negotiation on an SCB. The real problem is that I don't account for the case when an SCB that is tagged to do a particular type of negotiation completes or SELTOs (selection timeout) without the negotiation taking place, so the accounting of sdtrpending and wdtrpending gets screwed up. In the wide case, if we tag it to do both wdtr and sdtr, it only performs wdtr (since wdtr must occur first and we spread out the negotiation over two commands) so we always have sdtrpending set for that target and we never do a real SDTR. I fill properly fix the accounting after 2.0.5 goes out the door, but this works (as confirmed by Dan) on wide targets. Other stuff that is also included: 1) Don't do a bzero when recycling SCBs. The only thing that must explicitly be set to zero is the scb control byte which is done in ahc_get_scb. We also need to set the SG_list_pointer and SG_list_count to 0 for commands that do not transfer data. 2) Mask the interrupt type printout for the aic7870 case. The bit we were using to determine interrupt type is only valid for the aic7770. Submitted by: Justin Gibbs
1995-05-17 07:06:02 +00:00
mk_sdtr_max_offset:
# We're initiating sync negotiation, so request the max offset we can (15 or 8)
xor FLAGS, MAX_OFFSET
test SCSIRATE, 0x80 jnz wmax_offset # Talking to a WIDE device?
mvi DINDIR, MAX_OFFSET_8BIT
jmp mk_sdtr_done
wmax_offset:
mvi DINDIR, MAX_OFFSET_WIDE
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
# so it can assert ATN without upsetting SCSISIGO. The new value is
# expected in SINDEX. Change the actual state last to avoid contention
# from the driver.
#
scsisig:
mov SIGSTATE,SINDEX
mov SCSISIGO,SINDEX ret
sdtr_to_rate:
call ndx_dtr # index scratch space for target
shr A,SINDIR,0x4
dec SINDEX #Preserve SINDEX
and A,0x7
clr RETURN_1
sdtr_to_rate_loop:
test A,0x0f jz sdtr_to_rate_done
add RETURN_1,0x18
dec A
jmp sdtr_to_rate_loop
sdtr_to_rate_done:
shr RETURN_1,0x2
add RETURN_1,0x18 ret
return:
ret