d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
2e92a4ad78
freebsd-nq/sys/i386/scsi/aic7xxx.c

2965 lines
76 KiB
C
Raw Normal View History

My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
* Generic driver for the aic7xxx based adaptec SCSI controllers
* Product specific probe and attach routines can be found in:
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
* i386/eisa/aic7770.c 27/284X and aic7770 motherboard controllers
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* pci/aic7870.c 3940, 2940, aic7880, aic7870, aic7860,
* and aic7850 controllers
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
*
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
* Copyright (c) 1994, 1995, 1996 Justin T. Gibbs.
* All rights reserved.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
* 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 immediately at the beginning of the file, without modification,
* 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. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
* 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.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* $Id: aic7xxx.c,v 1.75.2.5 1996/10/06 22:48:12 gibbs Exp $
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
/*
* TODO:
* Implement Target Mode
*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* A few notes on features of the driver.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
*
* SCB paging takes advantage of the fact that devices stay disconnected
* from the bus a relatively long time and that while they're disconnected,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* having the SCBs for these transactions down on the host adapter is of
* little use. Instead of leaving this idle SCB down on the card we copy
* it back up into kernel memory and reuse the SCB slot on the card to
* schedule another transaction. This can be a real payoff when doing random
* I/O to tagged queueing devices since there are more transactions active at
* once for the device to sort for optimal seek reduction. The algorithm goes
* like this...
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* The sequencer maintains two lists of its hardware SCBs. The first is the
* singly linked free list which tracks all SCBs that are not currently in
* use. The second is the doubly linked disconnected list which holds the
* SCBs of transactions that are in the disconnected state sorted most
* recently disconnected first. When the kernel queues a transaction to
* the card, a hardware SCB to "house" this transaction is retrieved from
* either of these two lists. If the SCB came from the disconnected list,
* a check is made to see if any data transfer or SCB linking (more on linking
* in a bit) information has been changed since it was copied from the host
* and if so, DMAs the SCB back up before it can be used. Once a hardware
* SCB has been obtained, the SCB is DMAed from the host. Before any work
* can begin on this SCB, the sequencer must ensure that either the SCB is
* for a tagged transaction or the target is not already working on another
* non-tagged transaction. If a conflict arises in the non-tagged case, the
* sequencer finds the SCB for the active transactions and sets the SCB_LINKED
* field in that SCB to this next SCB to execute. To facilitate finding
* active non-tagged SCBs, the last four bytes of up to the first four hardware
* SCBs serve as a storage area for the currently active SCB ID for each
* target.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* When a device reconnects, a search is made of the hardware SCBs to find
* the SCB for this transaction. If the search fails, a hardware SCB is
* pulled from either the free or disconnected SCB list and the proper
* SCB is DMAed from the host. If the SCB_ABORTED control bit is set
* in the control byte of the SCB while it was disconnected, the sequencer
* will send an abort or abort tag message to the target during the
* reconnection and signal the kernel that the abort was successfull.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* When a command completes, a check for non-zero status and residuals is
* made. If either of these conditions exists, the SCB is DMAed back up to
* the host so that it can interpret this information. Additionally, in the
* case of bad status, the sequencer generates a special interrupt and pauses
* itself. This allows the host to setup a request sense command if it
* chooses for this target synchronously with the error so that sense
* information isn't lost.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
*
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
#include <sys/param.h>
#include <sys/systm.h>
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__NetBSD__)
#include <sys/device.h>
#include <machine/bus.h>
#include <machine/intr.h>
#endif /* defined(__NetBSD__) */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
#include <sys/malloc.h>
#include <sys/buf.h>
#include <sys/proc.h>
Replaced #includes of <sys/user.h> by less gross headers, usually <sys/vm.h>. Many device drivers need only the definition of vtophys() from vm. Added nearby #includes of <sys/conf.h> where appropriate.
1995-12-06 23:52:35 +00:00
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
#include <scsi/scsi_all.h>
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
#include <scsi/scsi_message.h>
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__NetBSD__)
#include <scsi/scsi_debug.h>
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
#include <scsi/scsiconf.h>
Replaced #includes of <sys/user.h> by less gross headers, usually <sys/vm.h>. Many device drivers need only the definition of vtophys() from vm. Added nearby #includes of <sys/conf.h> where appropriate.
1995-12-06 23:52:35 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
#include <machine/clock.h>
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
Replaced #includes of <sys/user.h> by less gross headers, usually <sys/vm.h>. Many device drivers need only the definition of vtophys() from vm. Added nearby #includes of <sys/conf.h> where appropriate.
1995-12-06 23:52:35 +00:00
#include <vm/vm.h>
Untangled the vm.h include file spaghetti.
1995-12-07 12:48:31 +00:00
#include <vm/vm_param.h>
#include <vm/pmap.h>
Replaced #includes of <sys/user.h> by less gross headers, usually <sys/vm.h>. Many device drivers need only the definition of vtophys() from vm. Added nearby #includes of <sys/conf.h> where appropriate.
1995-12-06 23:52:35 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Document the Adaptec driver options for tagged command queueing and SCB paging (with a warning not to use SCB paging) and create an opt_aic7xxx.h file for these options.
1996-10-01 03:01:06 +00:00
#include "opt_aic7xxx.h"
Point include to i386/scsi
1995-01-13 02:27:08 +00:00
#include <i386/scsi/aic7xxx.h>
Replaced #includes of <sys/user.h> by less gross headers, usually <sys/vm.h>. Many device drivers need only the definition of vtophys() from vm. Added nearby #includes of <sys/conf.h> where appropriate.
1995-12-06 23:52:35 +00:00
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
#include <dev/aic7xxx/aic7xxx_reg.h>
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif /* defined(__FreeBSD__) */
#if defined(__NetBSD__)
#include <dev/ic/aic7xxxreg.h>
#include <dev/ic/aic7xxxvar.h>
#define bootverbose 1
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
#if DEBUGTARGET < 0 /* Negative numbrs for disabling cause warnings */
#define DEBUGTARGET 17
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
#endif /* defined(__NetBSD__) */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
#include <sys/kernel.h>
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
#define MIN(a,b) (((a) < (b)) ? (a) : (b))
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
#define ALL_TARGETS -1
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
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
-Wall fixes.
1995-11-06 05:21:13 +00:00
u_long ahc_unit = 0;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
#ifdef AHC_DEBUG
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
static int ahc_debug = AHC_DEBUG;
#endif
#ifdef AHC_BROKEN_CACHE
int ahc_broken_cache = 1;
/*
* "wbinvd" cause writing back whole cache (both CPU internal & external)
* to memory, so that the instruction takes a lot of time.
* This makes machine slow.
*/
#define INVALIDATE_CACHE() __asm __volatile("wbinvd")
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/**** bit definitions for SCSIDEF ****/
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
#define HSCSIID 0x07 /* our SCSI ID */
#define HWSCSIID 0x0f /* our SCSI ID if Wide Bus */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
static void ahcminphys __P((struct buf *bp));
static int32_t ahc_scsi_cmd __P((struct scsi_xfer *xs));
static void ahc_run_waiting_queue __P((struct ahc_data *ahc));
static struct scb *
ahc_get_scb __P((struct ahc_data *ahc, u_int32_t flags));
static void ahc_free_scb __P((struct ahc_data *ahc, struct scb *scb));
static struct scb *
ahc_alloc_scb __P((struct ahc_data *ahc));
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
static inline void pause_sequencer __P((struct ahc_data *ahc));
static inline void unpause_sequencer __P((struct ahc_data *ahc,
int unpause_always));
static inline void restart_sequencer __P((struct ahc_data *ahc));
Completed function declarations and added prototypes. Sorted prototypes.
1995-12-15 23:49:42 +00:00
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static struct scsi_adapter ahc_switch =
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
ahc_scsi_cmd,
ahcminphys,
NULL,
NULL,
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
NULL,
"ahc",
{ 0, 0 }
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
};
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static struct scsi_device ahc_dev =
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
NULL, /* Use default error handler */
NULL, /* have a queue, served by this */
NULL, /* have no async handler */
NULL, /* Use default 'done' routine */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
"ahc",
0,
{ 0, 0 }
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
};
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
static inline void
pause_sequencer(ahc)
struct ahc_data *ahc;
{
AHC_OUTB(ahc, HCNTRL, ahc->pause);
/*
* Since the sequencer can disable pausing in a critical section, we
* must loop until it actually stops.
*/
while ((AHC_INB(ahc, HCNTRL) & PAUSE) == 0)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
static inline void
unpause_sequencer(ahc, unpause_always)
struct ahc_data *ahc;
int unpause_always;
{
if (unpause_always
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
|| (AHC_INB(ahc, INTSTAT) & (SCSIINT | SEQINT | BRKADRINT)) == 0)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
AHC_OUTB(ahc, HCNTRL, ahc->unpause);
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Restart the sequencer program from address zero
*/
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
static inline void
restart_sequencer(ahc)
struct ahc_data *ahc;
{
do {
AHC_OUTB(ahc, SEQCTL, SEQRESET|FASTMODE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} while ((AHC_INB(ahc, SEQADDR0) != 0)
|| (AHC_INB(ahc, SEQADDR1) != 0));
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
#if defined(__FreeBSD__)
#define IS_SCSIBUS_B(ahc, sc_link) \
(((u_int32_t)(sc_link)->fordriver) & SELBUSB)
#else /* NetBSD/OpenBSD */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#define IS_SCSIBUS_B(ahc, sc_link) \
((sc_link)->scsibus == (ahc)->sc_link_b.scsibus)
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
static u_int8_t ahc_abort_wscb __P((struct ahc_data *ahc, struct scb *scbp,
u_int8_t scbpos, u_int8_t prev,
struct scb *timedout_scb,
u_int32_t xs_error));
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
static void ahc_done __P((struct ahc_data *ahc, struct scb *scbp));
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
static void ahc_handle_seqint __P((struct ahc_data *ahc, u_int8_t intstat));
static void ahc_handle_scsiint __P((struct ahc_data *ahc,
u_int8_t intstat));
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
static void ahc_loadseq __P((struct ahc_data *ahc));
Completed function declarations and added prototypes. Sorted prototypes.
1995-12-15 23:49:42 +00:00
static int ahc_match_scb __P((struct scb *scb, int target, char channel));
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
static int ahc_poll __P((struct ahc_data *ahc, int wait));
Completed function declarations and added prototypes. Sorted prototypes.
1995-12-15 23:49:42 +00:00
#ifdef AHC_DEBUG
static void ahc_print_scb __P((struct scb *scb));
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
static u_int8_t find_scb __P((struct ahc_data *ahc, struct scb *scb));
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
static int ahc_reset_channel __P((struct ahc_data *ahc, char channel,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
struct scb *timedout_scb,
u_int32_t xs_error,
int initiate_reset));
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
static int ahc_reset_device __P((struct ahc_data *ahc, int target,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
char channel, struct scb *timedout_scb,
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
u_int32_t xs_error));
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
static void ahc_reset_current_bus __P((struct ahc_data *ahc));
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
static void ahc_run_done_queue __P((struct ahc_data *ahc));
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
static void ahc_scsirate __P((struct ahc_data* ahc, u_int8_t *scsirate,
u_int8_t *period, u_int8_t *offset,
char channel, int target));
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Completed function declarations and added prototypes. Sorted prototypes.
1995-12-15 23:49:42 +00:00
static timeout_t
ahc_timeout;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
static void ahc_timeout __P((void *));
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
static void ahc_unbusy_target __P((struct ahc_data *ahc,
int target, char channel));
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
static void ahc_construct_sdtr __P((struct ahc_data *ahc, int start_byte,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t period, u_int8_t offset));
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
static void ahc_construct_wdtr __P((struct ahc_data *ahc, int start_byte,
u_int8_t bus_width));
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
static void ahc_calc_residual __P((struct scb *scb));
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
char *ahc_name(ahc)
struct ahc_data *ahc;
{
static char name[10];
sprintf(name, "ahc%d", ahc->unit);
return (name);
}
#elif defined(__NetBSD__)
struct cfdriver ahc_cd = {
NULL, "ahc", DV_DULL
};
#endif
Completed function declarations and added prototypes. Sorted prototypes.
1995-12-15 23:49:42 +00:00
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
#ifdef AHC_DEBUG
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc_print_scb(scb)
struct scb *scb;
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
printf("scb:%p control:0x%x tcl:0x%x cmdlen:%d cmdpointer:0x%lx\n",
scb,
scb->control,
scb->tcl,
scb->cmdlen,
scb->cmdpointer );
printf(" datlen:%d data:0x%lx segs:0x%x segp:0x%lx\n",
scb->datalen,
scb->data,
scb->SG_segment_count,
scb->SG_list_pointer);
printf(" sg_addr:%lx sg_len:%ld\n",
scb->ahc_dma[0].addr,
scb->ahc_dma[0].len);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
static struct {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t errno;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
char *errmesg;
} hard_error[] = {
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
{ ILLHADDR, "Illegal Host Access" },
{ ILLSADDR, "Illegal Sequencer Address referrenced" },
{ ILLOPCODE, "Illegal Opcode in sequencer program" },
{ PARERR, "Sequencer Ram Parity Error" }
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
};
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Valid SCSIRATE values. (p. 3-17)
* Provides a mapping of tranfer periods in ns to the proper value to
* stick in the scsiscfr reg to use that transfer rate.
*/
static struct {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
int sxfr;
Properly deal with the Ultra series of adapters. We should now understand the new seeprom format and negotiate up to 20MHz sync if set in SCSI-Select. Reduce the complexity of the timeout code by running it at splhigh(). Fix a bug that caused rescheduled timeouts at 0 clock ticks in the future causing an infinite loop. Obtained from: Timeout bug noticed by David Greenman and wcarchive.
1995-10-26 23:57:18 +00:00
/* Rates in Ultra mode have bit 8 of sxfr set */
#define ULTRA_SXFR 0x100
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t period; /* Period to send to SCSI target */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
char *rate;
} ahc_syncrates[] = {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
{ 0x100, 12, "20.0" },
{ 0x110, 15, "16.0" },
{ 0x120, 18, "13.4" },
{ 0x000, 25, "10.0" },
{ 0x010, 31, "8.0" },
{ 0x020, 37, "6.67" },
{ 0x030, 43, "5.7" },
{ 0x040, 50, "5.0" },
{ 0x050, 56, "4.4" },
{ 0x060, 62, "4.0" },
{ 0x070, 68, "3.6" }
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
};
static int ahc_num_syncrates =
sizeof(ahc_syncrates) / sizeof(ahc_syncrates[0]);
/*
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
* Allocate a controller structures for a new device and initialize it.
* ahc_reset should be called before now since we assume that the card
* is paused.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
struct ahc_data *
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
ahc_alloc(unit, iobase, maddr, type, flags)
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
int unit;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int32_t iobase;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
void
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
ahc_construct(ahc, bc, ioh, maddr, type, flags)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
struct ahc_data *ahc;
bus_chipset_tag_t bc;
bus_io_handle_t ioh;
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
vm_offset_t maddr;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc_type type;
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-05 23:52:03 +00:00
ahc_flag flags;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/*
* find unit and check we have that many defined
*/
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
struct ahc_data *ahc;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/*
* Allocate a storage area for us
*/
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc = malloc(sizeof(struct ahc_data), M_TEMP, M_NOWAIT);
if (!ahc) {
printf("ahc%d: cannot malloc!\n", unit);
return NULL;
}
bzero(ahc, sizeof(struct ahc_data));
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
STAILQ_INIT(&ahc->free_scbs);
STAILQ_INIT(&ahc->waiting_scbs);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
ahc->unit = unit;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
#if defined(__FreeBSD__)
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
ahc->baseport = iobase;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
ahc->sc_bc = bc;
ahc->sc_ioh = ioh;
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
ahc->maddr = (volatile u_int8_t *)maddr;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc->type = type;
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-05 23:52:03 +00:00
ahc->flags = flags;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc->unpause = (AHC_INB(ahc, HCNTRL) & IRQMS) | INTEN;
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
ahc->pause = ahc->unpause | PAUSE;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
return (ahc);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
void
ahc_free(ahc)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
struct ahc_data *ahc;
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
{
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
free(ahc, M_DEVBUF);
return;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
void
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
ahc_reset(iobase)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int32_t iobase;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
ahc_reset(devname, bc, ioh)
char *devname;
bus_chipset_tag_t bc;
bus_io_handle_t ioh;
#endif
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t hcntrl;
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
int wait;
/* Retain the IRQ type accross the chip reset */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
hcntrl = (inb(HCNTRL + iobase) & IRQMS) | INTEN;
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
outb(HCNTRL + iobase, CHIPRST | PAUSE);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
hcntrl = (bus_io_read_1(bc, ioh, HCNTRL) & IRQMS) | INTEN;
bus_io_write_1(bc, ioh, HCNTRL, CHIPRST | PAUSE);
#endif
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
/*
* Ensure that the reset has finished
*/
wait = 1000;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
while (--wait && !(inb(HCNTRL + iobase) & CHIPRSTACK))
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
while (--wait && !(bus_io_read_1(bc, ioh, HCNTRL) & CHIPRSTACK))
#endif
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
DELAY(1000);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (wait == 0) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
-Wall fixes.
1995-11-06 05:21:13 +00:00
printf("ahc at 0x%lx: WARNING - Failed chip reset! "
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
"Trying to initialize anyway.\n", iobase);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
printf("%s: WARNING - Failed chip reset! "
"Trying to initialize anyway.\n", devname);
#endif
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
outb(HCNTRL + iobase, hcntrl | PAUSE);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
bus_io_write_1(bc, ioh, HCNTRL, hcntrl | PAUSE);
#endif
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Look up the valid period to SCSIRATE conversion in our table.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 18:37:25 +00:00
ahc_scsirate(ahc, scsirate, period, offset, channel, target )
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
struct ahc_data *ahc;
u_int8_t *scsirate;
u_int8_t *period;
u_int8_t *offset;
char channel;
int target;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
int i;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
u_int32_t ultra_enb_addr;
u_int8_t sxfrctl0;
u_int8_t ultra_enb;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
i = ahc_num_syncrates; /* Default to async */
if (*period >= ahc_syncrates[0].period && *offset != 0) {
for (i = 0; i < ahc_num_syncrates; i++) {
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
if (*period <= ahc_syncrates[i].period) {
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 18:37:25 +00:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
* Watch out for Ultra speeds when ultra is not
* enabled and vice-versa.
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 18:37:25 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (!(ahc->type & AHC_ULTRA)
&& (ahc_syncrates[i].sxfr & ULTRA_SXFR)) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
/*
* This should only happen if the
* drive is the first to negotiate
* and chooses a high rate. We'll
* just move down the table util
* we hit a non ultra speed.
*/
continue;
}
*scsirate = (ahc_syncrates[i].sxfr & 0xF0)
| (*offset & 0x0f);
*period = ahc_syncrates[i].period;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bootverbose) {
printf("%s: target %d synchronous at "
"%sMHz, offset = 0x%x\n",
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
ahc_name(ahc), target,
ahc_syncrates[i].rate, *offset );
}
break;
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-05 23:52:03 +00:00
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
}
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
if (i >= ahc_num_syncrates) {
/* Use asyncronous transfers. */
*scsirate = 0;
*period = 0;
*offset = 0;
Bring in bug fix from 'SCSI' branch.
1996-10-06 22:50:56 +00:00
if (bootverbose)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
printf("%s: target %d using asyncronous transfers\n",
ahc_name(ahc), target );
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-05 23:52:03 +00:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
/*
* Ensure Ultra mode is set properly for
* this target.
*/
ultra_enb_addr = ULTRA_ENB;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (channel == 'B' || target > 7)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
ultra_enb_addr++;
ultra_enb = AHC_INB(ahc, ultra_enb_addr);
sxfrctl0 = AHC_INB(ahc, SXFRCTL0);
if (*scsirate != 0 && ahc_syncrates[i].sxfr & ULTRA_SXFR) {
ultra_enb |= 0x01 << (target & 0x07);
sxfrctl0 |= ULTRAEN;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
ultra_enb &= ~(0x01 << (target & 0x07));
sxfrctl0 &= ~ULTRAEN;
}
AHC_OUTB(ahc, ultra_enb_addr, ultra_enb);
AHC_OUTB(ahc, SXFRCTL0, sxfrctl0);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__NetBSD__)
int
ahcprint(aux, name)
void *aux;
char *name;
{
if (name != NULL)
printf("%s: scsibus ", name);
return UNCONF;
}
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Attach all the sub-devices we can find
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
*/
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
int
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc_attach(ahc)
struct ahc_data *ahc;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
struct scsibus_data *scbus;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#ifdef AHC_BROKEN_CACHE
if (cpu_class == CPUCLASS_386) /* doesn't have "wbinvd" instruction */
ahc_broken_cache = 0;
#endif
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
/*
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
* fill in the prototype scsi_links.
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc->sc_link.adapter_unit = ahc->unit;
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
ahc->sc_link.adapter_targ = ahc->our_id;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc->sc_link.fordriver = 0;
#elif defined(__NetBSD__)
ahc->sc_link.adapter_target = ahc->our_id;
#endif
Use the new adapter_softc field in the scsi_link structure so that these drivers don't need to maintain an array of configured units. The bt driver still needs to because ISA interrupt handlers take a unit number. :(
1996-01-07 19:24:36 +00:00
ahc->sc_link.adapter_softc = ahc;
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
ahc->sc_link.adapter = &ahc_switch;
Increase the DELAY used in ahc_poll. This is only used during intial probe of the device and shouldn't affect normal operation. It seems that some version of the aic7870 are extreamly slow in clearing their interrupt register so the added delay is required to ensure we don't poll a stale interrupt value. Clean up debugging printfs and stuff superflous output into #ifdef AHC_DEBUG. Ensure that sdtrpending and wdtrpending are cleared in all cases after we recieve an SDTR or WDTR message. With this commit, the 294x should work or at least be much closer to working properly.
1995-03-17 23:58:27 +00:00
ahc->sc_link.opennings = 2;
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
ahc->sc_link.device = &ahc_dev;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc->sc_link.flags = DEBUGLEVEL;
Register with devconfg so we show up in lsdev.
1994-11-18 20:34:30 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type & AHC_TWIN) {
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
/* Configure the second scsi bus */
ahc->sc_link_b = ahc->sc_link;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
ahc->sc_link_b.adapter_targ = ahc->our_id_b;
ahc->sc_link_b.adapter_bus = 1;
ahc->sc_link_b.fordriver = (void *)SELBUSB;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
ahc->sc_link_b.adapter_target = ahc->our_id_b;
#endif
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
/*
* Prepare the scsibus_data area for the upperlevel
* scsi code.
*/
scbus = scsi_alloc_bus();
if(!scbus)
return 0;
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
scbus->adapter_link = (ahc->flags & AHC_CHANNEL_B_PRIMARY) ?
&ahc->sc_link_b : &ahc->sc_link;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type & AHC_WIDE)
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
scbus->maxtarg = 15;
/*
* ask the adapter what subunits are present
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bootverbose)
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
printf("ahc%d: Probing channel %c\n", ahc->unit,
(ahc->flags & AHC_CHANNEL_B_PRIMARY) ? 'B' : 'A');
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
scsi_attachdevs(scbus);
scbus = NULL; /* Upper-level SCSI code owns this now */
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type & AHC_TWIN) {
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
scbus = scsi_alloc_bus();
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (!scbus)
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
return 0;
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
scbus->adapter_link = (ahc->flags & AHC_CHANNEL_B_PRIMARY) ?
&ahc->sc_link : &ahc->sc_link_b;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type & AHC_WIDE)
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
scbus->maxtarg = 15;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bootverbose)
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
printf("ahc%d: Probing Channel %c\n", ahc->unit,
(ahc->flags & AHC_CHANNEL_B_PRIMARY) ? 'A': 'B');
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
scsi_attachdevs(scbus);
scbus = NULL; /* Upper-level SCSI code owns this now */
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
/*
* XXX - Update MI SCSI code
*
* if(ahc->type & AHC_WIDE)
* max target of both channel A and B = 15;
*/
/*
* ask the adapter what subunits are present
*/
if ((ahc->flags & AHC_CHANNEL_B_PRIMARY) == 0) {
/* make IS_SCSIBUS_B() == false, while probing channel A */
ahc->sc_link_b.scsibus = 0xff;
if (ahc->type & AHC_TWIN)
printf("%s: Probing channel A\n", ahc_name(ahc));
config_found((void *)ahc, &ahc->sc_link, ahcprint);
if (ahc->type & AHC_TWIN) {
printf("%s: Probing channel B\n", ahc_name(ahc));
config_found((void *)ahc, &ahc->sc_link_b, ahcprint);
}
} else {
/*
* if implementation of IS_SCSIBUS_B() is changed to use
* ahc->sc_link.scsibus, then "ahc->sc_link.scsibus = 0xff;"
* is needed, here.
*/
/* assert(ahc->type & AHC_TWIN); */
printf("%s: Probing channel B\n", ahc_name(ahc));
config_found((void *)ahc, &ahc->sc_link_b, ahcprint);
printf("%s: Probing channel A\n", ahc_name(ahc));
config_found((void *)ahc, &ahc->sc_link, ahcprint);
}
#endif
Remove hard coded assumption that SCSI busses have 7 targets. Change some leading spaces to tabs. This change forces the controller drivers to allocate a scsibus_data struct via a call to scsi_alloc_bus(), fill in the adapter_link field, and optionally modify any other fields of the struct. Scsi_alloc_bus() initializes all fields to the default, so the changes in most drivers are very minimal. For drivers that support Wide controllers, the maxtarg field will have to be updated to allow probing of all targets (for an example, look at the aic7xxx driver). Scsi_attachdevs() now takes a scsibus_data* as its argument instead of an sc_link*. This allows us to expand the role of the scsibus_data struct for other bus level configuration setings (max number of transactions, current transaction opennings, etc for better tagged queuing support). Reviewed by: Rodney Grimes <rgrimes>, Peter Dufault <dufault>, Julian Elischer <julian>
1995-08-23 23:03:17 +00:00
return 1;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
/*
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* Catch an interrupt from the adapter
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Make PCI interrupt handlers return void like everybody else does. Reviewed by: davidg
1996-01-23 21:48:28 +00:00
void
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined (__NetBSD__)
int
#endif
Make PCI interrupt handlers return void like everybody else does. Reviewed by: davidg
1996-01-23 21:48:28 +00:00
ahc_intr(arg)
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
void *arg;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
struct ahc_data *ahc;
u_int8_t intstat;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
ahc = (struct ahc_data *)arg;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
intstat = AHC_INB(ahc, INTSTAT);
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 13:54:41 +00:00
/*
* Is this interrupt for me? or for
* someone who is sharing my interrupt
*/
if (!(intstat & INT_PEND))
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Make PCI interrupt handlers return void like everybody else does. Reviewed by: davidg
1996-01-23 21:48:28 +00:00
return;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
return 0;
#endif
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
if (intstat & BRKADRINT) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/*
* We upset the sequencer :-(
* Lookup the error message
*/
int i, error, num_errors;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
error = AHC_INB(ahc, ERROR);
num_errors = sizeof(hard_error)/sizeof(hard_error[0]);
for (i = 0; error != 1 && i < num_errors; i++)
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
error >>= 1;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
panic("%s: brkadrint, %s at seqaddr = 0x%x\n",
ahc_name(ahc), hard_error[i].errmesg,
(AHC_INB(ahc, SEQADDR1) << 8) |
AHC_INB(ahc, SEQADDR0));
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (intstat & SEQINT)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
ahc_handle_seqint(ahc, intstat);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (intstat & SCSIINT)
ahc_handle_scsiint(ahc, intstat);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (intstat & CMDCMPLT) {
struct scb *scb;
u_int8_t scb_index;
u_int8_t qoutcnt;
int int_cleared;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
int_cleared = 0;
while (qoutcnt = (AHC_INB(ahc, QOUTCNT) & ahc->qcntmask)) {
for (; qoutcnt > 0; qoutcnt--) {
scb_index = AHC_INB(ahc, QOUTFIFO);
scb = ahc->scbarray[scb_index];
if (!scb || !(scb->flags & SCB_ACTIVE)) {
printf("%s: WARNING "
"no command for scb %d "
"(cmdcmplt)\nQOUTCNT == %d\n",
ahc_name(ahc), scb_index,
qoutcnt);
continue;
}
untimeout(ahc_timeout, (caddr_t)scb);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
/*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* Save off the residual if there is one.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (scb->hscb->residual_SG_segment_count != 0)
ahc_calc_residual(scb);
ahc_done(ahc, scb);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
}
AHC_OUTB(ahc, CLRINT, CLRCMDINT);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
int_cleared++;
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (int_cleared == 0)
AHC_OUTB(ahc, CLRINT, CLRCMDINT);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->waiting_scbs.stqh_first != NULL)
ahc_run_waiting_queue(ahc);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
#if defined(__NetBSD__)
return 1;
#endif
}
static void
ahc_handle_seqint(ahc, intstat)
struct ahc_data *ahc;
u_int8_t intstat;
{
struct scb *scb;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int16_t targ_mask;
u_int8_t target = (AHC_INB(ahc, SCSIID) >> 4) & 0x0f;
int scratch_offset = target;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
char channel = AHC_INB(ahc, SBLKCTL) & SELBUSB ? 'B': 'A';
if (channel == 'B')
scratch_offset += 8;
targ_mask = (0x01 << scratch_offset);
switch (intstat & SEQINT_MASK) {
case NO_MATCH:
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
printf("%s:%c:%d: no active SCB for reconnecting "
"target - issuing ABORT\n",
ahc_name(ahc), channel, target);
printf("SAVED_TCL == 0x%x\n",
AHC_INB(ahc, SAVED_TCL));
ahc_unbusy_target(ahc, target, channel);
AHC_OUTB(ahc, SCB_CONTROL, 0);
AHC_OUTB(ahc, CLRSINT1, CLRSELTIMEO);
AHC_OUTB(ahc, RETURN_1, 0);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
break;
case SEND_REJECT:
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t rejbyte = AHC_INB(ahc, REJBYTE);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
printf("%s:%c:%d: Warning - unknown message recieved from "
"target (0x%x). Rejecting\n",
ahc_name(ahc), channel, target, rejbyte);
break;
}
case NO_IDENT:
panic("%s:%c:%d: Target did not send an IDENTIFY message. "
"SAVED_TCL == 0x%x\n",
ahc_name(ahc), channel, target,
AHC_INB(ahc, SAVED_TCL));
break;
case BAD_PHASE:
printf("%s:%c:%d: unknown scsi bus phase. Attempting to "
"continue\n", ahc_name(ahc), channel, target);
break;
case EXTENDED_MSG:
{
u_int8_t message_length;
u_int8_t message_code;
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
message_length = AHC_INB(ahc, MSGIN_EXT_LEN);
message_code = AHC_INB(ahc, MSGIN_EXT_OPCODE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
switch (message_code) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
case MSG_EXT_SDTR:
{
u_int8_t period;
u_int8_t offset;
If, during an SDTR negotiation, the target comes back with a response that is too low for the aic7xxx chip to handle with sync transfers, negotiate async transfers.
1996-10-06 19:43:36 +00:00
u_int8_t saved_offset;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
u_int8_t targ_scratch;
u_int8_t maxoffset;
u_int8_t rate;
if (message_length != MSG_EXT_SDTR_LEN) {
AHC_OUTB(ahc, RETURN_1, SEND_REJ);
ahc->sdtrpending &= ~targ_mask;
break;
}
period = AHC_INB(ahc, MSGIN_EXT_BYTE0);
If, during an SDTR negotiation, the target comes back with a response that is too low for the aic7xxx chip to handle with sync transfers, negotiate async transfers.
1996-10-06 19:43:36 +00:00
saved_offset = AHC_INB(ahc, MSGIN_EXT_BYTE1);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
targ_scratch = AHC_INB(ahc, TARG_SCRATCH
+ scratch_offset);
if (targ_scratch & WIDEXFER)
maxoffset = MAX_OFFSET_16BIT;
else
maxoffset = MAX_OFFSET_8BIT;
If, during an SDTR negotiation, the target comes back with a response that is too low for the aic7xxx chip to handle with sync transfers, negotiate async transfers.
1996-10-06 19:43:36 +00:00
offset = MIN(saved_offset, maxoffset);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
ahc_scsirate(ahc, &rate, &period, &offset,
channel, target);
/* Preserve the WideXfer flag */
targ_scratch = rate | (targ_scratch & WIDEXFER);
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
/*
* Update both the target scratch area and the
* current SCSIRATE.
*/
AHC_OUTB(ahc, TARG_SCRATCH + scratch_offset,
targ_scratch);
AHC_OUTB(ahc, SCSIRATE, targ_scratch);
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
If, during an SDTR negotiation, the target comes back with a response that is too low for the aic7xxx chip to handle with sync transfers, negotiate async transfers.
1996-10-06 19:43:36 +00:00
/*
* See if we initiated Sync Negotiation
* and didn't have to fall down to async
* transfers.
*/
if ((ahc->sdtrpending & targ_mask) != 0
&& (saved_offset == offset)) {
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
* Don't send an SDTR back to
* the target
*/
AHC_OUTB(ahc, RETURN_1, 0);
If, during an SDTR negotiation, the target comes back with a response that is too low for the aic7xxx chip to handle with sync transfers, negotiate async transfers.
1996-10-06 19:43:36 +00:00
ahc->needsdtr &= ~targ_mask;
ahc->sdtrpending &= ~targ_mask;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
} else {
/*
* Send our own SDTR in reply
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
printf("Sending SDTR!!\n");
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
ahc_construct_sdtr(ahc, /*start_byte*/0,
period, offset);
AHC_OUTB(ahc, RETURN_1, SEND_MSG);
If, during an SDTR negotiation, the target comes back with a response that is too low for the aic7xxx chip to handle with sync transfers, negotiate async transfers.
1996-10-06 19:43:36 +00:00
/*
* If we aren't starting a re-negotiation
* because we had to go async in response
* to a "too low" response from the target
* clear the needsdtr flag for this target.
*/
if ((ahc->sdtrpending & targ_mask) == 0)
ahc->needsdtr &= ~targ_mask;
else
ahc->sdtrpending |= targ_mask;
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
break;
}
case MSG_EXT_WDTR:
{
u_int8_t scratch, bus_width;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
if (message_length != MSG_EXT_WDTR_LEN) {
AHC_OUTB(ahc, RETURN_1, SEND_REJ);
ahc->wdtrpending &= ~targ_mask;
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
}
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
bus_width = AHC_INB(ahc, MSGIN_EXT_BYTE0);
scratch = AHC_INB(ahc, TARG_SCRATCH
+ scratch_offset);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
if (ahc->wdtrpending & targ_mask) {
/*
* Don't send a WDTR back to the
* target, since we asked first.
*/
AHC_OUTB(ahc, RETURN_1, 0);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
switch (bus_width){
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
case BUS_8_BIT:
scratch &= 0x7f;
break;
case BUS_16_BIT:
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bootverbose)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
printf("%s: target %d using "
"16Bit transfers\n",
ahc_name(ahc), target);
scratch |= WIDEXFER;
break;
case BUS_32_BIT:
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
* How can we do 32bit transfers
* on a 16bit bus?
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
*/
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
AHC_OUTB(ahc, RETURN_1, SEND_REJ);
printf("%s: target %d requested 32Bit "
"transfers. Rejecting...\n",
ahc_name(ahc), target);
break;
default:
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
break;
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
} else {
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
* Send our own WDTR in reply
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
switch (bus_width) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
case BUS_8_BIT:
scratch &= 0x7f;
break;
case BUS_32_BIT:
case BUS_16_BIT:
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type & AHC_WIDE) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
/* Negotiate 16_BITS */
bus_width = BUS_16_BIT;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bootverbose)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
printf("%s: target %d "
"using 16Bit "
"transfers\n",
ahc_name(ahc),
target);
scratch |= WIDEXFER;
} else
bus_width = BUS_8_BIT;
break;
default:
break;
}
ahc_construct_wdtr(ahc, /*start_byte*/0,
bus_width);
AHC_OUTB(ahc, RETURN_1, SEND_MSG);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
ahc->needwdtr &= ~targ_mask;
ahc->wdtrpending &= ~targ_mask;
AHC_OUTB(ahc, TARG_SCRATCH + scratch_offset, scratch);
AHC_OUTB(ahc, SCSIRATE, scratch);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
}
default:
/* Unknown extended message. Reject it. */
AHC_OUTB(ahc, RETURN_1, SEND_REJ);
}
}
case REJECT_MSG:
{
/*
* What we care about here is if we had an
* outstanding SDTR or WDTR message for this
* target. If we did, this is a signal that
* the target is refusing negotiation.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t targ_scratch;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
targ_scratch = AHC_INB(ahc, TARG_SCRATCH
+ scratch_offset);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->wdtrpending & targ_mask) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
/* note 8bit xfers and clear flag */
targ_scratch &= 0x7f;
ahc->needwdtr &= ~targ_mask;
ahc->wdtrpending &= ~targ_mask;
printf("%s:%c:%d: refuses WIDE negotiation. Using "
"8bit transfers\n", ahc_name(ahc),
channel, target);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else if (ahc->sdtrpending & targ_mask) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
/* note asynch xfers and clear flag */
targ_scratch &= 0xf0;
ahc->needsdtr &= ~targ_mask;
ahc->sdtrpending &= ~targ_mask;
printf("%s:%c:%d: refuses syncronous negotiation. "
"Using asyncronous transfers\n",
ahc_name(ahc),
channel, target);
} else {
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 13:54:41 +00:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
* Otherwise, we ignore it.
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 13:54:41 +00:00
*/
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc_debug & AHC_SHOWMISC)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
printf("%s:%c:%d: Message reject -- ignored\n",
ahc_name(ahc), channel, target);
#endif
break;
}
AHC_OUTB(ahc, TARG_SCRATCH + scratch_offset, targ_scratch);
AHC_OUTB(ahc, SCSIRATE, targ_scratch);
break;
}
case BAD_STATUS:
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t scb_index;
struct scsi_xfer *xs;
struct hardware_scb *hscb;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/*
* The sequencer will notify us when a command
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
* has an error that would be of interest to
* the kernel. This allows us to leave the sequencer
* running in the common case of command completes
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* without error. The sequencer will already have
* dma'd the SCB back up to us, so we can reference
* the in kernel copy directly.
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
*/
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
scb_index = AHC_INB(ahc, SCB_TAG);
scb = ahc->scbarray[scb_index];
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
hscb = scb->hscb;
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
/*
* Set the default return value to 0 (don't
* send sense). The sense code will change
* this if needed and this reduces code
* duplication.
*/
AHC_OUTB(ahc, RETURN_1, 0);
if (!(scb && (scb->flags & SCB_ACTIVE))) {
printf("%s:%c:%d: ahc_intr - referenced scb "
"not valid during seqint 0x%x scb(%d)\n",
ahc_name(ahc),
channel, target, intstat,
scb_index);
goto clear;
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
xs = scb->xs;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
xs->status = hscb->status;
switch (hscb->status){
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
case SCSI_OK:
printf("%s: Interrupted for staus of"
" 0???\n", ahc_name(ahc));
break;
case SCSI_CHECK:
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc_debug & AHC_SHOWSENSE) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
sc_print_addr(xs->sc_link);
printf("requests Check Status\n");
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
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
#endif
if ((xs->error == XS_NOERROR)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
&& !(scb->flags & SCB_SENSE)) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
struct ahc_dma_seg *sg = scb->ahc_dma;
struct scsi_sense *sc = &(scb->sense_cmd);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/*
* Save off the residual if there is one.
*/
if (hscb->residual_SG_segment_count != 0)
ahc_calc_residual(scb);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc_debug & AHC_SHOWSENSE) {
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
sc_print_addr(xs->sc_link);
printf("Sending Sense\n");
}
#endif
#if defined(__FreeBSD__)
sc->op_code = REQUEST_SENSE;
#elif defined(__NetBSD__)
sc->opcode = REQUEST_SENSE;
#endif
sc->byte2 = xs->sc_link->lun << 5;
sc->length = sizeof(struct scsi_sense_data);
sc->control = 0;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
sg->addr = vtophys(&xs->sense);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
sg->len = sizeof(struct scsi_sense_data);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
hscb->control &= DISCENB;
hscb->status = 0;
hscb->SG_segment_count = 1;
hscb->SG_list_pointer = vtophys(sg);
hscb->data = sg->addr;
hscb->datalen &= 0xFF000000;
hscb->datalen |= sg->len;
hscb->cmdpointer = vtophys(sc);
hscb->cmdlen = sizeof(*sc);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
scb->flags |= SCB_SENSE;
AHC_OUTB(ahc, RETURN_1, SEND_SENSE);
break;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
Detect and report dataphase overruns. Put the adapter into 'Bit Bucket' mode when this occurs and allow the target to complete the transaction. Force a retry on overruns since they are usually caused by termination or cable problems.
1996-06-09 17:33:18 +00:00
/*
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
* Clear the SCB_SENSE Flag and have
* the sequencer do a normal command
* complete with either a "DRIVER_STUFFUP"
* error or whatever other error condition
* we already had.
Detect and report dataphase overruns. Put the adapter into 'Bit Bucket' mode when this occurs and allow the target to complete the transaction. Force a retry on overruns since they are usually caused by termination or cable problems.
1996-06-09 17:33:18 +00:00
*/
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
scb->flags &= ~SCB_SENSE;
if (xs->error == XS_NOERROR)
xs->error = XS_DRIVER_STUFFUP;
Detect and report dataphase overruns. Put the adapter into 'Bit Bucket' mode when this occurs and allow the target to complete the transaction. Force a retry on overruns since they are usually caused by termination or cable problems.
1996-06-09 17:33:18 +00:00
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
case SCSI_BUSY:
xs->error = XS_BUSY;
sc_print_addr(xs->sc_link);
printf("Target Busy\n");
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
case SCSI_QUEUE_FULL:
/*
* The upper level SCSI code will someday
* handle this properly.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
xs->error = XS_BUSY;
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
break;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
default:
sc_print_addr(xs->sc_link);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
printf("unexpected targ_status: %x\n", hscb->status);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
xs->error = XS_DRIVER_STUFFUP;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
break;
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
break;
}
case ABORT_TAG:
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t scb_index;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
struct scsi_xfer *xs;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
scb_index = AHC_INB(ahc, SCB_TAG);
scb = ahc->scbarray[scb_index];
xs = scb->xs;
/*
* We didn't recieve a valid tag back from
* the target on a reconnect.
*/
sc_print_addr(xs->sc_link);
printf("invalid tag recieved -- sending ABORT_TAG\n");
xs->error = XS_DRIVER_STUFFUP;
untimeout(ahc_timeout, (caddr_t)scb);
ahc_done(ahc, scb);
break;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
case AWAITING_MSG:
{
int scb_index;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
scb_index = AHC_INB(ahc, SCB_TAG);
scb = ahc->scbarray[scb_index];
/*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* This SCB had MK_MESSAGE set in its control byte,
* informing the sequencer that we wanted to send a
* special message to this target.
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
*/
if (scb->flags & SCB_DEVICE_RESET) {
AHC_OUTB(ahc, MSG0,
MSG_BUS_DEV_RESET);
AHC_OUTB(ahc, MSG_LEN, 1);
printf("Bus Device Reset Message Sent\n");
} else if (scb->flags & SCB_MSGOUT_WDTR) {
ahc_construct_wdtr(ahc, AHC_INB(ahc, MSG_LEN),
BUS_16_BIT);
} else if (scb->flags & SCB_MSGOUT_SDTR) {
u_int8_t target_scratch;
u_int8_t ultraenable;
int sxfr;
int i;
/* Pull the user defined setting */
target_scratch = AHC_INB(ahc, TARG_SCRATCH
+ scratch_offset);
sxfr = target_scratch & SXFR;
if (scratch_offset < 8)
ultraenable = AHC_INB(ahc, ULTRA_ENB);
else
ultraenable = AHC_INB(ahc, ULTRA_ENB + 1);
if (ultraenable & targ_mask)
/* Want an ultra speed in the table */
sxfr |= 0x100;
for (i = 0; i < ahc_num_syncrates; i++)
Bring in bug fix from 'SCSI' branch.
1996-10-06 22:50:56 +00:00
if (sxfr == ahc_syncrates[i].sxfr)
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
break;
ahc_construct_sdtr(ahc, AHC_INB(ahc, MSG_LEN),
ahc_syncrates[i].period,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
(target_scratch & WIDEXFER) ?
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
MAX_OFFSET_16BIT : MAX_OFFSET_8BIT);
} else
panic("ahc_intr: AWAITING_MSG for an SCB that "
"does not have a waiting message");
break;
}
case IMMEDDONE:
{
/*
* Take care of device reset messages
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t scbindex = AHC_INB(ahc, SCB_TAG);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
scb = ahc->scbarray[scbindex];
if (scb->flags & SCB_DEVICE_RESET) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t targ_scratch;
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
int found;
/*
* Go back to async/narrow transfers and
* renegotiate.
*/
ahc_unbusy_target(ahc, target, channel);
ahc->needsdtr |= ahc->needsdtr_orig & targ_mask;
ahc->needwdtr |= ahc->needwdtr_orig & targ_mask;
ahc->sdtrpending &= ~targ_mask;
ahc->wdtrpending &= ~targ_mask;
targ_scratch = AHC_INB(ahc, TARG_SCRATCH
+ scratch_offset);
targ_scratch &= SXFR;
AHC_OUTB(ahc, TARG_SCRATCH + scratch_offset,
targ_scratch);
found = ahc_reset_device(ahc, target,
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
channel, NULL,
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
XS_NOERROR);
sc_print_addr(scb->xs->sc_link);
printf("Bus Device Reset delivered. "
"%d SCBs aborted\n", found);
ahc->in_timeout = FALSE;
ahc_run_done_queue(ahc);
} else
panic("ahc_intr: Immediate complete for "
"unknown operation.");
break;
}
case DATA_OVERRUN:
{
/*
* When the sequencer detects an overrun, it
* sets STCNT to 0x00ffffff and allows the
* target to complete its transfer in
* BITBUCKET mode.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t scbindex = AHC_INB(ahc, SCB_TAG);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
u_int32_t overrun;
scb = ahc->scbarray[scbindex];
overrun = AHC_INB(ahc, STCNT0)
| (AHC_INB(ahc, STCNT1) << 8)
| (AHC_INB(ahc, STCNT2) << 16);
overrun = 0x00ffffff - overrun;
sc_print_addr(scb->xs->sc_link);
printf("data overrun of %d bytes detected."
" Forcing a retry.\n", overrun);
/*
* Set this and it will take affect when the
* target does a command complete.
*/
scb->xs->error = XS_DRIVER_STUFFUP;
break;
}
#if NOT_YET
/* XXX Fill these in later */
case MESG_BUFFER_BUSY:
break;
case MSGIN_PHASEMIS:
break;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#endif
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
default:
printf("ahc_intr: seqint, "
"intstat == 0x%x, scsisigi = 0x%x\n",
intstat, AHC_INB(ahc, SCSISIGI));
break;
}
clear:
/*
* Clear the upper byte that holds SEQINT status
* codes and clear the SEQINT bit.
*/
AHC_OUTB(ahc, CLRINT, CLRSEQINT);
/*
* The sequencer is paused immediately on
* a SEQINT, so we should restart it when
* we're done.
*/
unpause_sequencer(ahc, /*unpause_always*/TRUE);
Fixed the type of ahcintr(). The type of an ISA interrupt handler is incompatible with the type of a PCI interrupt handler. A new entry point `ahc_pci_intr()' is used for PCI. ISA and PCI interrupts are penalized equally (:-) by calling a common handler `ahc_intr()'. This should be reorganized. Some strings now name the wrong function...
1995-11-04 14:43:30 +00:00
}
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
static void
ahc_handle_scsiint(ahc, intstat)
struct ahc_data *ahc;
u_int8_t intstat;
{
u_int8_t scb_index;
u_int8_t status;
struct scb *scb;
scb_index = AHC_INB(ahc, SCB_TAG);
status = AHC_INB(ahc, SSTAT1);
scb = ahc->scbarray[scb_index];
if (status & SCSIRSTI) {
char channel;
channel = (AHC_INB(ahc, SBLKCTL) & SELBUSB) ? 'B' : 'A';
printf("%s: Someone reset channel %c\n",
ahc_name(ahc), channel);
ahc_reset_channel(ahc,
channel,
NULL,
XS_BUSY,
/* Initiate Reset */FALSE);
scb = NULL;
} else if (!(scb && (scb->flags & SCB_ACTIVE))){
printf("%s: ahc_intr - referenced scb not "
"valid during scsiint 0x%x scb(%d)\n",
ahc_name(ahc), status, scb_index);
AHC_OUTB(ahc, CLRSINT1, status);
unpause_sequencer(ahc, /*unpause_always*/TRUE);
AHC_OUTB(ahc, CLRINT, CLRSCSIINT);
scb = NULL;
} else if (status & SCSIPERR) {
/*
* Determine the bus phase and
* queue an appropriate message
*/
char *phase;
u_int8_t mesg_out = MSG_NOOP;
u_int8_t lastphase = AHC_INB(ahc, LASTPHASE);
struct scsi_xfer *xs;
xs = scb->xs;
sc_print_addr(xs->sc_link);
switch (lastphase) {
case P_DATAOUT:
phase = "Data-Out";
break;
case P_DATAIN:
phase = "Data-In";
mesg_out = MSG_INITIATOR_DET_ERR;
break;
case P_COMMAND:
phase = "Command";
break;
case P_MESGOUT:
phase = "Message-Out";
break;
case P_STATUS:
phase = "Status";
mesg_out = MSG_INITIATOR_DET_ERR;
break;
case P_MESGIN:
phase = "Message-In";
mesg_out = MSG_PARITY_ERROR;
break;
default:
phase = "unknown";
break;
}
printf("parity error during %s phase.\n", phase);
/*
* We've set the hardware to assert ATN if we
* get a parity error on "in" phases, so all we
* need to do is stuff the message buffer with
* the appropriate message. "In" phases have set
* mesg_out to something other than MSG_NOP.
*/
if (mesg_out != MSG_NOOP) {
AHC_OUTB(ahc, MSG0, mesg_out);
AHC_OUTB(ahc, MSG_LEN, 1);
} else
/*
* Should we allow the target to make
* this decision for us?
*/
xs->error = XS_DRIVER_STUFFUP;
} else if (status & SELTO) {
struct scsi_xfer *xs;
u_int8_t scbptr;
u_int8_t nextscb;
u_int8_t flags;
xs = scb->xs;
xs->error = XS_SELTIMEOUT;
/*
* Clear any pending messages for the timed out
* target, and mark the target as free
*/
flags = AHC_INB(ahc, FLAGS);
AHC_OUTB(ahc, MSG_LEN, 0);
ahc_unbusy_target(ahc, xs->sc_link->target,
IS_SCSIBUS_B(ahc, xs->sc_link) ? 'B' : 'A');
/* Stop the selection */
AHC_OUTB(ahc, SCSISEQ, 0);
AHC_OUTB(ahc, SCB_CONTROL, 0);
AHC_OUTB(ahc, CLRSINT1, CLRSELTIMEO);
AHC_OUTB(ahc, CLRINT, CLRSCSIINT);
/* Shift the waiting Q forward. */
scbptr = AHC_INB(ahc, WAITING_SCBH);
AHC_OUTB(ahc, SCBPTR, scbptr);
nextscb = AHC_INB(ahc, SCB_NEXT);
AHC_OUTB(ahc, WAITING_SCBH, nextscb);
/* Put this SCB back on the free list */
nextscb = AHC_INB(ahc, FREE_SCBH);
AHC_OUTB(ahc, SCB_NEXT, nextscb);
AHC_OUTB(ahc, FREE_SCBH, scbptr);
restart_sequencer(ahc);
} else if (!(status & BUSFREE)) {
sc_print_addr(scb->xs->sc_link);
printf("Unknown SCSIINT. Status = 0x%x\n", status);
AHC_OUTB(ahc, CLRSINT1, status);
unpause_sequencer(ahc, /*unpause_always*/TRUE);
AHC_OUTB(ahc, CLRINT, CLRSCSIINT);
scb = NULL;
}
if (scb != NULL) {
/* We want to process the command */
untimeout(ahc_timeout, (caddr_t)scb);
ahc_done(ahc, scb);
}
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* We have a scb which has been processed by the
* adaptor, now we look to see how the operation
* went.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc_done(ahc, scb)
struct ahc_data *ahc;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
struct scb *scb;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
struct scsi_xfer *xs = scb->xs;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
SC_DEBUG(xs->sc_link, SDEV_DB2, ("ahc_done\n"));
/*
* Put the results of the operation
* into the xfer and call whoever started it
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__NetBSD__)
if (xs->error != XS_NOERROR) {
/* Don't override the error value. */
} else if (scb->flags & SCB_ABORTED) {
xs->error = XS_DRIVER_STUFFUP;
} else
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (scb->flags & SCB_SENSE)
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
xs->error = XS_SENSE;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (scb->flags & SCB_SENTORDEREDTAG)
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 19:21:20 +00:00
ahc->in_timeout = FALSE;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__FreeBSD__)
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
if ((xs->flags & SCSI_ERR_OK) && !(xs->error == XS_SENSE)) {
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/* All went correctly OR errors expected */
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
xs->error = XS_NOERROR;
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#elif defined(__NetBSD__)
/*
* Since NetBSD doesn't have error ignoring operation mode
* (SCSI_ERR_OK in FreeBSD), we don't have to care this case.
*/
#endif
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
xs->flags |= ITSDONE;
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 19:21:20 +00:00
#ifdef AHC_TAGENABLE
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/*
* This functionality is provided by the generic SCSI layer
* in FreeBSD 2.2.
*/
if (xs->cmd->opcode == INQUIRY && xs->error == XS_NOERROR) {
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
struct scsi_inquiry_data *inq_data;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int16_t mask = 0x01 << (xs->sc_link->target |
(scb->hscb->tcl & 0x08));
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
/*
* Sneak a look at the results of the SCSI Inquiry
* command and see if we can do Tagged queing. This
* should really be done by the higher level drivers.
*/
inq_data = (struct scsi_inquiry_data *)xs->data;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if ((inq_data->flags & SID_CmdQue)
&& !(ahc->tagenable & mask)) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
printf("%s: target %d Tagged Queuing Device\n",
ahc_name(ahc), xs->sc_link->target);
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->tagenable |= mask;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->maxhscbs >= 16 || (ahc->flags&AHC_PAGESCBS)) {
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
/* Default to 8 tags */
The default number of tags in the SCB paging case was supposed to be 8 not the 18 I was using during some of my own testing. Ooops. For those that want to change the number for experimentation, you can set the value on line 1553 of this file.
1996-04-22 13:21:40 +00:00
xs->sc_link->opennings += 6;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else {
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
/*
* Default to 4 tags on whimpy
* cards that don't have much SCB
* space and can't page. This prevents
* a single device from hogging all
* slots. We should really have a better
* way of providing fairness.
*/
xs->sc_link->opennings += 2;
}
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
}
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
#endif /* AHC_TAGENABLE */
ahc_free_scb(ahc, scb);
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
scsi_done(xs);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
/*
* Start the board, ready for normal operation
*/
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
int
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc_init(ahc)
struct ahc_data *ahc;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
u_int8_t scsi_conf, sblkctl, i;
u_int16_t ultraenable = 0;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
int max_targ = 15;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
Move aic7xxx register definitions to sys/dev/aic7xxx/aic7xxx_reg.h. Start the revamp of the initialiation process. New routines include ahc_alloc, ahc_free, and ahc_reset. These help divide the work of staring up a board more logically between probe and attach. ahcintr now takes a (void *) and returns int. The pci code uses it directly. Until the PCI code for shared edged triggered interrupts is removed, the eisa code uses a stub (ahc_eisa_intr) that throws away the int returned by ahcintr. Use MHz instead of MB/s for printing out sync rates. Print out "aic7880" instead of "aic7870" for the new aic7880 chips.
1995-11-05 04:50:55 +00:00
* Assume we have a board at this stage and it has been reset.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
/* Handle the SCBPAGING option */
#ifndef AHC_SCBPAGING_ENABLE
ahc->flags &= ~AHC_PAGESCBS;
#endif
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
/* Determine channel configuration and who we are on the scsi bus. */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
switch ((sblkctl = AHC_INB(ahc, SBLKCTL) & 0x0a)) {
case 0:
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc->our_id = (AHC_INB(ahc, SCSICONF) & HSCSIID);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
ahc->flags &= ~AHC_CHANNEL_B_PRIMARY;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type == AHC_394)
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-05 23:52:03 +00:00
printf("Channel %c, SCSI Id=%d, ",
ahc->flags & AHC_CHNLB ? 'B' : 'A',
ahc->our_id);
else
printf("Single Channel, SCSI Id=%d, ", ahc->our_id);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, FLAGS, SINGLE_BUS | (ahc->flags & AHC_PAGESCBS));
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
break;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
case 2:
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc->our_id = (AHC_INB(ahc, SCSICONF + 1) & HWSCSIID);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
ahc->flags &= ~AHC_CHANNEL_B_PRIMARY;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type == AHC_394)
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-05 23:52:03 +00:00
printf("Wide Channel %c, SCSI Id=%d, ",
ahc->flags & AHC_CHNLB ? 'B' : 'A',
ahc->our_id);
else
printf("Wide Channel, SCSI Id=%d, ", ahc->our_id);
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->type |= AHC_WIDE;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, FLAGS, WIDE_BUS | (ahc->flags & AHC_PAGESCBS));
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
break;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
case 8:
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc->our_id = (AHC_INB(ahc, SCSICONF) & HSCSIID);
ahc->our_id_b = (AHC_INB(ahc, SCSICONF + 1) & HSCSIID);
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
printf("Twin Channel, A SCSI Id=%d, B SCSI Id=%d, ",
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
ahc->our_id, ahc->our_id_b);
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->type |= AHC_TWIN;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, FLAGS, TWIN_BUS | (ahc->flags & AHC_PAGESCBS));
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
break;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
default:
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
printf(" Unsupported adapter type. Ignoring\n");
return(-1);
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/* Determine the number of SCBs and initialize them */
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/* SCB 0 heads the free list */
AHC_OUTB(ahc, FREE_SCBH, 0);
for (i = 0; i < AHC_SCB_MAX; i++) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCBPTR, i);
AHC_OUTB(ahc, SCB_CONTROL, i);
if(AHC_INB(ahc, SCB_CONTROL) != i)
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
break;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCBPTR, 0);
if(AHC_INB(ahc, SCB_CONTROL) != 0)
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
break;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCBPTR, i);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/* Clear the control byte. */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCB_CONTROL, 0);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/* Set the next pointer */
AHC_OUTB(ahc, SCB_NEXT, i+1);
/* No Busy non-tagged targets yet */
AHC_OUTB(ahc, SCB_ACTIVE0, SCB_LIST_NULL);
AHC_OUTB(ahc, SCB_ACTIVE1, SCB_LIST_NULL);
AHC_OUTB(ahc, SCB_ACTIVE2, SCB_LIST_NULL);
AHC_OUTB(ahc, SCB_ACTIVE3, SCB_LIST_NULL);
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/* Make that the last SCB terminates the free list */
AHC_OUTB(ahc, SCBPTR, i-1);
AHC_OUTB(ahc, SCB_NEXT, SCB_LIST_NULL);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
/* Ensure we clear the 0 SCB's control byte. */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCBPTR, 0);
AHC_OUTB(ahc, SCB_CONTROL, 0);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
ahc->maxhscbs = i;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if ((ahc->maxhscbs < AHC_SCB_MAX) && (ahc->flags & AHC_PAGESCBS)) {
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
ahc->maxscbs = AHC_SCB_MAX;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
printf("%d/%d SCBs\n", ahc->maxhscbs, ahc->maxscbs);
} else {
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
ahc->maxscbs = ahc->maxhscbs;
ahc->flags &= ~AHC_PAGESCBS;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
printf("%d SCBs\n", ahc->maxhscbs);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
}
Additions to support the WAITING_SCB list that the sequencer maintains. It is the kernel driver's responsibility to do the list manipulation whenever a selection timeout or a request sense occurs. Print out the interrupt type that the device has been set to. It seems that one of the Asus motherboards botches this and David thought a diagnostic would be nice. Fix a bug in my diagnostic code that David found. Reviewed by: Wcarchive and David Greenman
1995-04-27 17:47:17 +00:00
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc_debug & AHC_SHOWMISC) {
printf("%s: hardware scb %d bytes; kernel scb %d bytes; "
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
"ahc_dma %d bytes\n",
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc_name(ahc),
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
sizeof(struct hardware_scb),
sizeof(struct scb),
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
sizeof(struct ahc_dma_seg));
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
}
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
#endif /* AHC_DEBUG */
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Properly deal with the Ultra series of adapters. We should now understand the new seeprom format and negotiate up to 20MHz sync if set in SCSI-Select. Reduce the complexity of the timeout code by running it at splhigh(). Fix a bug that caused rescheduled timeouts at 0 clock ticks in the future causing an infinite loop. Obtained from: Timeout bug noticed by David Greenman and wcarchive.
1995-10-26 23:57:18 +00:00
/* Set the SCSI Id, SXFRCTL0, SXFRCTL1, and SIMODE1, for both channels*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type & AHC_TWIN) {
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
/*
Initialize SIMODE1, SCSIID, and SXFRCTL1 in ahc_init instead of at each reconnection/selection with a target.
1995-01-22 00:48:39 +00:00
* The device is gated to channel B after a chip reset,
* so set those values first
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCSIID, ahc->our_id_b);
scsi_conf = AHC_INB(ahc, SCSICONF + 1);
AHC_OUTB(ahc, SXFRCTL1, (scsi_conf & (ENSPCHK|STIMESEL))
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
| ENSTIMER|ACTNEGEN|STPWEN);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SIMODE1, ENSELTIMO|ENSCSIRST|ENSCSIPERR);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type & AHC_ULTRA)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SXFRCTL0, DFON|SPIOEN|ULTRAEN);
Initialize SXFRCTL0 with DFON|SPIOEN|ULTRAEN.
1995-10-29 05:57:48 +00:00
else
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SXFRCTL0, DFON|SPIOEN);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (scsi_conf & RESET_SCSI) {
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
/* Reset the bus */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bootverbose)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
printf("%s: Reseting Channel B\n",
ahc_name(ahc));
AHC_OUTB(ahc, SCSISEQ, SCSIRSTO);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
DELAY(1000);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCSISEQ, 0);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
/* Ensure we don't get a RSTI interrupt from this */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, CLRSINT1, CLRSCSIRSTI);
AHC_OUTB(ahc, CLRINT, CLRSCSIINT);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
}
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
Initialize SIMODE1, SCSIID, and SXFRCTL1 in ahc_init instead of at each reconnection/selection with a target.
1995-01-22 00:48:39 +00:00
/* Select Channel A */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SBLKCTL, 0);
Initialize SIMODE1, SCSIID, and SXFRCTL1 in ahc_init instead of at each reconnection/selection with a target.
1995-01-22 00:48:39 +00:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCSIID, ahc->our_id);
scsi_conf = AHC_INB(ahc, SCSICONF);
AHC_OUTB(ahc, SXFRCTL1, (scsi_conf & (ENSPCHK|STIMESEL))
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
| ENSTIMER|ACTNEGEN|STPWEN);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SIMODE1, ENSELTIMO|ENSCSIRST|ENSCSIPERR);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type & AHC_ULTRA)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SXFRCTL0, DFON|SPIOEN|ULTRAEN);
Initialize SXFRCTL0 with DFON|SPIOEN|ULTRAEN.
1995-10-29 05:57:48 +00:00
else
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SXFRCTL0, DFON|SPIOEN);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (scsi_conf & RESET_SCSI) {
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
/* Reset the bus */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bootverbose)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
printf("%s: Reseting Channel A\n", ahc_name(ahc));
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCSISEQ, SCSIRSTO);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
DELAY(1000);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCSISEQ, 0);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
/* Ensure we don't get a RSTI interrupt from this */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, CLRSINT1, CLRSCSIRSTI);
AHC_OUTB(ahc, CLRINT, CLRSCSIINT);
Honor the CHANNEL_B_PRIMARY bit of the BIOSCTRL register and probe channel B first as approriate. Only reset the SCSI bus if the RESET_SCSI bit of SCSICONF is set. This makes the aic7xxx driver honor all of the configuration settings availible in SCSI-Select or the ECU. Fix a benign bug in the reset code that caused us to always wait a full second after the chip reset. This should shave some time off the probe. Bug found by pedrosal@nce.ufrj.br (Pedro Salenbauch) It seems that only the top three sync rates are doubled when in ultra mode, so update the syncrates table as appropriate. Found by "Dan Willis" <dan@plutotech.com> and his SCSI bus analyzer
1996-05-10 16:21:05 +00:00
}
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
* Look at the information that board initialization or
* the board bios has left us. In the lower four bits of each
* target's scratch space any value other than 0 indicates
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
* that we should initiate syncronous transfers. If it's zero,
* the user or the BIOS has decided to disable syncronous
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
* negotiation to that target so we don't activate the needsdtr
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
* flag.
*/
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
ahc->needsdtr_orig = 0;
ahc->needwdtr_orig = 0;
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
/* Grab the disconnection disable table and invert it for our needs */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->flags & AHC_USEDEFAULTS) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
printf("%s: Host Adapter Bios disabled. Using default SCSI "
"device parameters\n", ahc_name(ahc));
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
ahc->discenable = 0xff;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc->discenable = ~((AHC_INB(ahc, DISC_DSB + 1) << 8)
| AHC_INB(ahc, DISC_DSB));
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (!(ahc->type & (AHC_WIDE|AHC_TWIN)))
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-05 23:52:03 +00:00
max_targ = 7;
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 13:54:41 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
for (i = 0; i <= max_targ; i++) {
u_int8_t target_settings;
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
if (ahc->flags & AHC_USEDEFAULTS) {
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
target_settings = 0; /* 10MHz */
ahc->needsdtr_orig |= (0x01 << i);
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
ahc->needwdtr_orig |= (0x01 << i);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else {
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
/* Take the settings leftover in scratch RAM. */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
target_settings = AHC_INB(ahc, TARG_SCRATCH + i);
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (target_settings & 0x0f) {
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
ahc->needsdtr_orig |= (0x01 << i);
/*Default to a asyncronous transfers(0 offset)*/
target_settings &= 0xf0;
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (target_settings & 0x80) {
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
ahc->needwdtr_orig |= (0x01 << i);
/*
* We'll set the Wide flag when we
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
* are successful with Wide negotiation.
* Turn it off for now so we aren't
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
* confused.
*/
target_settings &= 0x7f;
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->type & AHC_ULTRA) {
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 18:37:25 +00:00
/*
* Enable Ultra for any target that
* has a valid ultra syncrate setting.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t rate = target_settings & 0x70;
if (rate == 0x00 || rate == 0x10 ||
rate == 0x20 || rate == 0x40) {
if (rate == 0x40) {
Correct a botched commit from yesturday. It helps to bring over the right patch file.<sigh>
1996-05-23 15:02:18 +00:00
/* Treat 10MHz specially */
target_settings &= ~0x70;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else
Correct a botched commit from yesturday. It helps to bring over the right patch file.<sigh>
1996-05-23 15:02:18 +00:00
ultraenable |= (0x01 << i);
}
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 18:37:25 +00:00
}
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, TARG_SCRATCH+i,target_settings);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
}
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
/*
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
* If we are not a WIDE device, forget WDTR. This
* makes the driver work on some cards that don't
* leave these fields cleared when the BIOS is not
* installed.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if ((ahc->type & AHC_WIDE) == 0)
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->needwdtr_orig = 0;
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
ahc->needsdtr = ahc->needsdtr_orig;
Add $Ids. Use DMA for downloading SCBs to the sequencer. Implement SDTR, WDTR, and message reject handlers so they don't need to exist in the sequencer. All three of these cases are not on the critical path, so it makes little sense to use up precious sequencer ram for them.
1995-01-16 16:33:47 +00:00
ahc->needwdtr = ahc->needwdtr_orig;
Keep track of when we actually are awaiting a reply to an SDTR or WDTR message instead of relying on the fact that we are scheduled to send them. The old method worked 99.9% of the time, but someone reported some periferals that did MSG_REJECT at odd times (sometimes before we could send an SDTR or WDTR) that we would construe as the response to an SDTR or WDTR message. This also removes a possible race condition where after a bus reset (the result of a command time out not during intial probe time), we might queue two commands both requesting SDTR, WDTR or both.
1995-02-03 17:15:12 +00:00
ahc->sdtrpending = 0;
ahc->wdtrpending = 0;
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->tagenable = 0;
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 19:21:20 +00:00
ahc->orderedtag = 0;
Increase the DELAY used in ahc_poll. This is only used during intial probe of the device and shouldn't affect normal operation. It seems that some version of the aic7870 are extreamly slow in clearing their interrupt register so the added delay is required to ensure we don't poll a stale interrupt value. Clean up debugging printfs and stuff superflous output into #ifdef AHC_DEBUG. Ensure that sdtrpending and wdtrpending are cleared in all cases after we recieve an SDTR or WDTR message. With this commit, the 294x should work or at least be much closer to working properly.
1995-03-17 23:58:27 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, ULTRA_ENB, ultraenable & 0xff);
AHC_OUTB(ahc, ULTRA_ENB + 1, (ultraenable >> 8) & 0xff);
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 18:37:25 +00:00
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
#ifdef AHC_DEBUG
/* How did we do? */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc_debug & AHC_SHOWMISC)
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
printf("NEEDSDTR == 0x%x\nNEEDWDTR == 0x%x\n"
"DISCENABLE == 0x%x\n", ahc->needsdtr,
ahc->needwdtr, ahc->discenable);
#endif
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
/*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* Set the number of availible hardware SCBs
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCBCOUNT, ahc->maxhscbs);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/*
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
* 2's compliment of maximum tag value
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
*/
i = ahc->maxscbs;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, COMP_SCBCOUNT, -i & 0xff);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/*
* Allocate enough "hardware scbs" to handle
* the maximum number of concurrent transactions
* we can have active. We have to use contigmalloc
* if this array crosses a page boundary since the
* sequencer depends on this array being physically
* contiguous.
*/
{
size_t array_size;
u_int32_t hscb_physaddr;
array_size = ahc->maxscbs * sizeof(struct hardware_scb);
if (array_size > PAGE_SIZE) {
ahc->hscbs = (struct hardware_scb *)
contigmalloc(array_size, M_DEVBUF,
M_NOWAIT, 0ul, 0xffffffff, PAGE_SIZE,
0x10000);
} else {
ahc->hscbs = (struct hardware_scb *)
malloc(array_size, M_DEVBUF, M_NOWAIT);
}
if (ahc->hscbs == NULL) {
printf("%s: unable to allocate hardware SCB array. "
"Failing attach\n");
return (-1);
}
/* Tell the sequencer where it can find the hscb array. */
hscb_physaddr = vtophys(ahc->hscbs);
AHC_OUTB(ahc, HSCB_ADDR0, hscb_physaddr & 0xFF);
AHC_OUTB(ahc, HSCB_ADDR1, (hscb_physaddr >> 8)& 0xFF);
AHC_OUTB(ahc, HSCB_ADDR2, (hscb_physaddr >> 16)& 0xFF);
AHC_OUTB(ahc, HSCB_ADDR3, (hscb_physaddr >> 24)& 0xFF);
}
/*
* Q-Full-Count. Some cards have more Q space
* then SCBs.
*/
if (ahc->type & AHC_AIC7770) {
ahc->qfullcount = 4;
ahc->qcntmask = 0x07;
} else if (ahc->type & AHC_AIC7850) {
ahc->qfullcount = 8;
ahc->qcntmask = 0x0f;
} else if (ahc->maxhscbs == 255) {
/* 7870/7880 with external SRAM */
ahc->qfullcount = 255;
ahc->qcntmask = 0xff;
} else {
/* 7870/7880 */
ahc->qfullcount = 16;
ahc->qcntmask = 0x1f;
}
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
/*
* QCount mask to deal with broken aic7850s that
* sporatically get garbage in the upper bits of
* their QCount registers.
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
*
* QFullCount to guard against overflowing the
* QINFIFO or QOUTFIFO when we are paging SCBs.
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, QCNTMASK, ahc->qcntmask);
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
AHC_OUTB(ahc, QFULLCNT, ahc->qfullcount);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Additions to support the WAITING_SCB list that the sequencer maintains. It is the kernel driver's responsibility to do the list manipulation whenever a selection timeout or a request sense occurs. Print out the interrupt type that the device has been set to. It seems that one of the Asus motherboards botches this and David thought a diagnostic would be nice. Fix a bug in my diagnostic code that David found. Reviewed by: Wcarchive and David Greenman
1995-04-27 17:47:17 +00:00
/* We don't have any waiting selections */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, WAITING_SCBH, SCB_LIST_NULL);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
/* Our disconnection list is empty too */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, DISCONNECTED_SCBH, SCB_LIST_NULL);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
/* Message out buffer starts empty */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, MSG_LEN, 0x00);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
/*
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
* Load the Sequencer program and Enable the adapter
* in "fast" mode.
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bootverbose)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
printf("%s: Downloading Sequencer Program...",
ahc_name(ahc));
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc_loadseq(ahc);
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bootverbose)
Clean up a few nits in the aic7xxx driver: 1) Make the driver "quiet" by sticking most boot messages behind bootverbose conditionals. This means that you won't see the sync and wide negotiation, but you will find out if they fail. 2) Add support to the 93cx6 serial eeprom code to read at an abitrary offset. This is needed so that we can access the second half of the eeprom on 3940 cards where the second channel's config is stored. 3) Add flags argument to ahcprobe(). This is used by the pci probe code to tell the generic driver that an adapter should be treated as a channel B device as well as notify it of the presence of external SCB SRAM. These are needed for some motherboard implementations of the aic7870 and for the 3940 controllers. 4) Print "Channel A"/"Channel B" instead of "Single Channel" for the two busses of the 3940. I received many reports of confusion about how the 3940 was probed since most people belived that only one ahc entry was needed. This will hopefully make it clearer. 5) Walk the SCBs to determine just how many their are if external SCB ram is detected. 6) Hard code that external SCB ram is present for the 3940 since it doesn't use the documented reporting facility for reporting the SRAM. :( 255 commands per channel are supported on the 3940. 7) Read the seeprom starting at addres 32 for the second channel of the 3940 so we get the right info for that channel. 8) Clean up printing of the "Disabling tagged queuing message". 9) Queue timeouts if they occur while we are handling a timeout. The code was totally unprotected in this scenario. Reviewed by: Timeout code reviewed by David Greenman <davidg>
1995-09-05 23:52:03 +00:00
printf("Done\n");
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SEQCTL, FASTMODE);
Compile cleanly -Wall. Remove some code that was an attempt to make the 294x probe more stable, but since the guess was wrong, go back to the clearer aproach I used the first time around.
1995-03-07 08:59:28 +00:00
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* Note that we are going and return (to probe)
*/
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
ahc->flags |= AHC_INIT;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
return (0);
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahcminphys(bp)
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
struct buf *bp;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
/*
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
* Even though the card can transfer up to 16megs per command
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
* we are limited by the number of segments in the dma segment
* list that we can hold. The worst case is that all pages are
* discontinuous physically, hense the "page per segment" limit
* enforced here.
*/
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 18:37:25 +00:00
if (bp->b_bcount > ((AHC_NSEG - 1) * PAGE_SIZE)) {
bp->b_bcount = ((AHC_NSEG - 1) * PAGE_SIZE);
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#if defined(__NetBSD__)
minphys(bp);
#endif
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
/*
* start a scsi operation given the command and
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
* the data address, target, and lun all of which
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
* are stored in the scsi_xfer struct
*/
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
static int32_t
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc_scsi_cmd(xs)
struct scsi_xfer *xs;
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
struct scb *scb;
struct hardware_scb *hscb;
struct ahc_data *ahc;
u_int16_t mask;
int flags;
int s;
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
Use the new adapter_softc field in the scsi_link structure so that these drivers don't need to maintain an array of configured units. The bt driver still needs to because ISA interrupt handlers take a unit number. :(
1996-01-07 19:24:36 +00:00
ahc = (struct ahc_data *)xs->sc_link->adapter_softc;
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
mask = (0x01 << (xs->sc_link->target
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
| (IS_SCSIBUS_B(ahc, xs->sc_link) ? SELBUSB : 0)));
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
SC_DEBUG(xs->sc_link, SDEV_DB2, ("ahc_scsi_cmd\n"));
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
flags = xs->flags;
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
/*
* get an scb to use. If the transfer
* is from a buf (possibly from interrupt time)
* then we can't allow it to sleep
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if ((scb = ahc_get_scb(ahc, flags)) == NULL) {
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
xs->error = XS_DRIVER_STUFFUP;
return (TRY_AGAIN_LATER);
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
hscb = scb->hscb;
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
SC_DEBUG(xs->sc_link, SDEV_DB3, ("start scb(%p)\n", scb));
scb->xs = xs;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
/*
* Put all the arguments for the xfer in the scb
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (ahc->discenable & mask)
hscb->control |= DISCENB;
if (flags & SCSI_RESET) {
scb->flags |= SCB_DEVICE_RESET|SCB_IMMED;
hscb->control |= MK_MESSAGE;
} else if ((ahc->needwdtr & mask) && !(ahc->wdtrpending & mask)) {
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->wdtrpending |= mask;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
hscb->control |= MK_MESSAGE;
scb->flags |= SCB_MSGOUT_WDTR;
} else if((ahc->needsdtr & mask) && !(ahc->sdtrpending & mask)) {
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
ahc->sdtrpending |= mask;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
hscb->control |= MK_MESSAGE;
scb->flags |= SCB_MSGOUT_SDTR;
} else if (ahc->orderedtag & mask) {
/* XXX this should be handled by the upper SCSI layer */
printf("Ordered Tag sent\n");
hscb->control |= MSG_ORDERED_Q_TAG;
ahc->orderedtag &= ~mask;
} else if (hscb->control & DISCENB) {
if (ahc->tagenable & mask)
hscb->control |= TAG_ENB;
}
hscb->tcl = ((xs->sc_link->target << 4) & 0xF0)
| (IS_SCSIBUS_B(ahc,xs->sc_link)? SELBUSB : 0)
| (xs->sc_link->lun & 0x07);
hscb->cmdlen = xs->cmdlen;
hscb->cmdpointer = vtophys(xs->cmd);
Major overhaul of the aic7xxx driver: - catch the interrupt type (EDGE/LEVEL) before chip reset instead of guessing the right type. - Add pause variable to the ahc struct to better handle the different interrupt types and pausing the sequencer. - CLRINTSTAT -> CLRSCSIINT: This is a documented bit in the CLRINT register in newer Adaptec documentation, so use their name for it. - Report valid residual byte counts. - Don't mess with the target scratch areas > id 8 on single, narrow, channel devices. The BIOS does a checksum of this area and can flip out if we zero it out. - Initialize the sequencer FLAGS scratch ram variable in the single channel devices to 0. This was the cause of the annoying warning where we would get a cmdcmplt the first time we did any type of transfer negotiation with no valid scb. It also fixes the problem that looked like the INTSTAT register wasn't clearing fast enough. This only showed up on 294x cards, not motherboard aic7870s. - Add the AHC_AIC7870 type and use it as the superset of aic7870 based controllers. - clear the sync offset section of the targ scratch area so that we default to asyncronous transfers. This was only a problem for wide controllers because there was a scenario where the offset wouldn't get updated before a data(out/in) phase would occur. This required some change in the sequencer code since we were depending on this field to hold the rate to negotiate. - allow sync and wide negotiated commands to be tagged (the sequencer now handles this properly).
1995-03-31 13:54:41 +00:00
xs->resid = 0;
Merge RELENG_2_0_5 into HEAD
1995-06-11 19:33:05 +00:00
xs->status = 0;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/* Only use S/G if non-zero length */
if (xs->datalen) {
int seg;
u_int32_t datalen;
vm_offset_t vaddr;
u_int32_t paddr;
u_int32_t nextpaddr;
struct ahc_dma_seg *sg;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
seg = 0;
datalen = xs->datalen;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
vaddr = (vm_offset_t)xs->data;
paddr = vtophys(vaddr);
sg = scb->ahc_dma;
hscb->SG_list_pointer = vtophys(sg);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
while ((datalen > 0) && (seg < AHC_NSEG)) {
/* put in the base address and length */
sg->addr = paddr;
sg->len = 0;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
/* do it at least once */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
nextpaddr = paddr;
while ((datalen > 0) && (paddr == nextpaddr)) {
u_int32_t size;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
/*
* This page is contiguous (physically)
* with the the last, just extend the
* length
*/
/* how far to the end of the page */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
nextpaddr = (paddr & (~PAGE_MASK)) + PAGE_SIZE;
/*
* Compute the maximum size
*/
size = nextpaddr - paddr;
if (size > datalen)
size = datalen;
sg->len += size;
vaddr += size;
datalen -= size;
if (datalen > 0)
paddr = vtophys(vaddr);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
}
/*
* next page isn't contiguous, finish the seg
*/
seg++;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
sg++;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
hscb->SG_segment_count = seg;
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
/* Copy the first SG into the data pointer area */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
hscb->data = scb->ahc_dma->addr;
hscb->datalen = scb->ahc_dma->len | (SCB_LIST_NULL << 24);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
if (datalen) {
First pass cleanup of this driver. This pass does not include the sequencer optimizations I have been working on yet, but does bring in some bug fixes and performance improvments that were easy to regression test: Setup the data fifo threshold and bus off timing correctly for 27/284x cards. Users of these adapters with fast periferals (greater than 5MB/s) will notice a big performance difference. (Sometimes as large as going from 3.7->8.3MB/s). Fix handling of the active target flags. Some of the outbs where missing the base offset in the abort code. The abort code still needs lots of work. Support 3940 controllers, but only with 16 SCBs for now. Eventually I'll add support for all 255, but I need to find a tester for the code first since we have to enable the cards external SRAM to do this. Add Dan Eischen's serial eeprom reading facilities. This allows the 2940 adapters to pull additional information left over from SCSI-Select right out out of the configuration seeprom. If the BIOS is disabled on 274x controllers, reset all target parameters to there defaults since you can't rely on what is stored in scratch ram. Report motherboard controllers as such. Stick the first SG address and count into the SCB data and count areas for all transfers in preparation of a later sequencer optimization. Keep track of which targets can are allowed to have the disconnection priveledge since this will be handled by the kernel driver in the future. If a target issues a message reject in response to a tagged message, disable tagged queuing for that target. Some seagates say they can do tagged queuing, but lie, and its a shame to have to disable tagged queuing on all devices just because you have one that can't cope.
1995-07-04 21:14:45 +00:00
/* there's still data, must have run out of segs! */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
printf("%s: ahc_scsi_cmd: more than %d DMA segs\n",
ahc_name(ahc), AHC_NSEG);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
xs->error = XS_DRIVER_STUFFUP;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
ahc_free_scb(ahc, scb);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
return (COMPLETE);
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
#ifdef AHC_BROKEN_CACHE
if (ahc_broken_cache)
INVALIDATE_CACHE();
#endif
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else {
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
/*
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
* No data xfer, use non S/G values
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
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
hscb->SG_segment_count = 0;
hscb->SG_list_pointer = 0;
hscb->data = 0;
hscb->datalen = (SCB_LIST_NULL << 24);
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
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Add tagged queueing support. Right now, we only allow two tagged commands per target. I could have followed the route of the ncr driver and gone to great lengths to get the SCSI subsystem to support more, but I think I'll use the time saved to help Julian and Peter make tagged queuing a better handled generic feature. This also includes some comment and enum clean up and a possible fix for the hanging PCI controllers.
1995-02-22 01:43:25 +00:00
#ifdef AHC_DEBUG
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if((ahc_debug & AHC_SHOWSCBS) && (xs->sc_link->target == DEBUGTARGET))
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
ahc_print_scb(scb);
#endif
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
s = splbio();
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
STAILQ_INSERT_TAIL(&ahc->waiting_scbs, scb, links);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
scb->flags |= SCB_ACTIVE;
ahc_run_waiting_queue(ahc);
if ((flags & SCSI_NOMASK) == 0) {
timeout(ahc_timeout, (caddr_t)scb, (xs->timeout * hz) / 1000);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
splx(s);
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
return (SUCCESSFULLY_QUEUED);
}
/*
* If we can't use interrupts, poll for completion
*/
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
SC_DEBUG(xs->sc_link, SDEV_DB3, ("cmd_poll\n"));
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
do {
if (ahc_poll(ahc, xs->timeout)) {
if (!(xs->flags & SCSI_SILENT))
printf("cmd fail\n");
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc_timeout(scb);
break;
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} while ((xs->flags & ITSDONE) == 0); /* a non command complete intr */
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
splx(s);
return (COMPLETE);
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/*
* Look for space in the QINFIFO and queue as many SCBs in the waiting
* queue as possible. Assumes that it is called at splbio().
*/
static void
ahc_run_waiting_queue(ahc)
struct ahc_data *ahc;
{
struct scb *scb;
/*
* On aic78X0 chips, we rely on Auto Access Pause (AAP)
* instead of doing an explicit pause/unpause.
*/
if ((ahc->type & AHC_AIC78X0) == 0)
pause_sequencer(ahc);
while ((scb = ahc->waiting_scbs.stqh_first) != NULL) {
if (ahc->curqincnt >= ahc->qfullcount) {
ahc->curqincnt = AHC_INB(ahc, QINCNT) & ahc->qcntmask;
if (ahc->curqincnt >= ahc->qfullcount)
/* Still no space */
break;
}
STAILQ_REMOVE_HEAD(&ahc->waiting_scbs, links);
AHC_OUTB(ahc, QINFIFO, scb->hscb->tag);
if ((ahc->flags & AHC_PAGESCBS) != 0)
/*
* We only care about this statistic when paging
* since it is impossible to overflow the qinfifo
* in the non-paging case.
*/
ahc->curqincnt++;
}
if ((ahc->type & AHC_AIC78X0) == 0)
unpause_sequencer(ahc, /*Unpause always*/FALSE);
}
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
* A scb (and hence an scb entry on the board is put onto the
* free list.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
ahc_free_scb(ahc, scb)
struct ahc_data *ahc;
struct scb *scb;
{
struct hardware_scb *hscb;
int opri;
hscb = scb->hscb;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 22:04:58 +00:00
opri = splbio();
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
/* Clean up for the next user */
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
scb->flags = SCB_FREE;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
hscb->control = 0;
hscb->status = 0;
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
STAILQ_INSERT_HEAD(&ahc->free_scbs, scb, links);
if (scb->links.stqe_next == NULL) {
/*
* If there were no SCBs availible, wake anybody waiting
* for one to come free.
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
wakeup((caddr_t)&ahc->free_scbs);
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
}
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
#ifdef AHC_DEBUG
ahc->activescbs--;
#endif
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 22:04:58 +00:00
splx(opri);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
/*
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
* Get a free scb, either one already assigned to a hardware slot
* on the adapter or one that will require an SCB to be paged out before
* use. If there are none, see if we can allocate a new SCB. Otherwise
* either return an error or sleep.
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static struct scb *
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc_get_scb(ahc, flags)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
struct ahc_data *ahc;
u_int32_t flags;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
struct scb *scbp;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
int opri;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 22:04:58 +00:00
opri = splbio();
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
/*
* If we can and have to, sleep waiting for one to come free
* but only if we can't allocate a new one.
*/
while (1) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if ((scbp = ahc->free_scbs.stqh_first)) {
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
STAILQ_REMOVE_HEAD(&ahc->free_scbs, links);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else if(ahc->numscbs < ahc->maxscbs) {
scbp = ahc_alloc_scb(ahc);
if (scbp == NULL)
printf("%s: Can't malloc SCB\n", ahc_name(ahc));
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
else if ((flags & SCSI_NOSLEEP) == 0) {
tsleep((caddr_t)&ahc->free_scbs, PRIBIO,
"ahcscb", 0);
continue;
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
}
break;
}
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 22:04:58 +00:00
#ifdef AHC_DEBUG
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
if (scbp) {
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 22:04:58 +00:00
ahc->activescbs++;
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
if((ahc_debug & AHC_SHOWSCBCNT)
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
&& (ahc->activescbs == ahc->maxhscbs))
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
printf("%s: Max SCBs active\n", ahc_name(ahc));
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
}
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
#endif
Don't arbitrarily set SCSI_NOSLEEP. It is now handled correctly by the higher level scsi code. Spls should never be conditionalized, so don't do so here. Restructure the get_scb routine so that we can't get into an infinite loop if the ccbs are exhausted and we are are called with SCSI_NOSLEEP set. Other driver maintainer's that based their scb allocation routines on Julian's code should look at these changes and implement them for their driver. The aic7xxx driver inspired these changes because early revs of the aic7770 chips have so few SCBs that you can actually run out. If you have a rev C or aic7770 (as is reported by the driver probe) and had more than 2 drives, you could get into an infinite loop when using up all of the SCBs. Since the driver will only allow two SCBs per device and I only had two devices, I never saw this problem on my Rev C card. Bzero only 19 bytes of the scb instead of 2k (ack!). This was a hold over from when a struct SCB only contained the information downloaded to the board, but we now store kernel driver data in there as well. This greatly lowers the overhead for small transactions (I get ~1MB/sec for dds with a 512 byte block size). Submitted by: John Dyson with the aic7xxx specific optimization by me
1995-04-23 22:04:58 +00:00
splx(opri);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
return (scbp);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
static struct scb *
ahc_alloc_scb(ahc)
struct ahc_data *ahc;
{
static struct ahc_dma_seg *next_sg_array = NULL;
static int sg_arrays_free;
struct scb *newscb;
newscb = (struct scb *) malloc(sizeof(struct scb), M_DEVBUF, M_NOWAIT);
if (newscb != NULL) {
bzero(newscb, sizeof(struct scb));
if (next_sg_array == NULL) {
size_t alloc_size = sizeof(struct ahc_dma_seg)
* AHC_NSEG;
sg_arrays_free = PAGE_SIZE / alloc_size;
/* Don't ever allocate more than we may need */
sg_arrays_free = MIN(ahc->maxscbs - ahc->numscbs,
sg_arrays_free);
alloc_size *= sg_arrays_free;
if (alloc_size == 0)
panic("%s: SG list doesn't fit in a page",
ahc_name(ahc));
next_sg_array = (struct ahc_dma_seg *)
malloc(alloc_size, M_DEVBUF, M_NOWAIT);
}
if (next_sg_array != NULL) {
struct hardware_scb *hscb;
newscb->ahc_dma = next_sg_array;
sg_arrays_free--;
if (sg_arrays_free == 0)
next_sg_array = NULL;
else
next_sg_array = &next_sg_array[AHC_NSEG];
hscb = &ahc->hscbs[ahc->numscbs];
newscb->hscb = hscb;
hscb->control = 0;
hscb->status = 0;
hscb->tag = ahc->numscbs;
hscb->residual_data_count[2] = 0;
hscb->residual_data_count[1] = 0;
hscb->residual_data_count[0] = 0;
hscb->residual_SG_segment_count = 0;
ahc->numscbs++;
/*
* Place in the scbarray
* Never is removed.
*/
ahc->scbarray[hscb->tag] = newscb;
} else {
free(newscb, M_DEVBUF);
newscb = NULL;
}
}
return newscb;
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
static void ahc_loadseq(ahc)
struct ahc_data *ahc;
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
{
Fix problem with scb flag handing that crept in with the SCB paging support. This only affected userland initiated device resets (using the reset command from cdplay for instance). Convert some spaces to tabs.
1996-06-23 20:02:37 +00:00
static u_char seqprog[] = {
Generic support for the entire aic7x70 line of adaptors. No one else came up with a directory place ment more me, so now we have i386/scsi.
1995-01-13 02:24:31 +00:00
# include "aic7xxx_seq.h"
Fix support for the aic7850 by looking only at the relavent bits of the QINCNT. The 7850 puts random garbage in the high bits and all my attempts to determine the cause of this failed. This approach does seem to work around the problem. Go back to relying on the SCSIPERR interrupt instead of having the sequencer interrupt at the beginning of ITloop after a parity error occured. Determine the number of SCBs on a card automatically and base the qcntmask on the number of SCBs. Add entries for 11.4MHz, 8.8MHz, 8.0MHz, and 7.2MHz to ULTRA portion of the syncrate table. They seem to work fine on the 2940UW I have here and will allow more non-ultra devices (like my tape drive) to run sync while the adapter is in ULTRA mode. Return XS_SELTIMEOUT instead of XS_TIMEOUT for selection timeouts. I was getting sick of waiting for the SCSI code to retry each non-existant unit multiple times during boot and XS_SELTIMEOUT bypasses all retries. Use new SLIST queue macros. This was inspired by NetBSD using TAILQs in their SCSI drivers. For optimum cache hits, the free scb list should be LIFO which is what the old and new code does. NetBSD implemented a FIFO queue for some reason. Spaces -> tabs.
1996-03-31 03:15:31 +00:00
};
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SEQCTL, PERRORDIS|SEQRESET|LOADRAM);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTSB(ahc, SEQRAM, seqprog, sizeof(seqprog));
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
Bring back the loop in RESTART_SEQUENCER. It seems to be necessary for the aic7850. Go back to autoATN on parity errors.
1996-06-08 06:55:01 +00:00
do {
AHC_OUTB(ahc, SEQCTL, SEQRESET|FASTMODE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} while ((AHC_INB(ahc, SEQADDR0) != 0)
|| (AHC_INB(ahc, SEQADDR1) != 0));
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
/*
* Function to poll for command completion when
* interrupts are disabled (crash dumps)
*/
static int
ahc_poll(ahc, wait)
struct ahc_data *ahc;
int wait; /* in msec */
{
while (--wait) {
DELAY(1000);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
if (AHC_INB(ahc, INTSTAT) & INT_PEND)
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
break;
} if (wait == 0) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
printf("%s: board is not responding\n", ahc_name(ahc));
Changes to support Performance enhancements in the sequencer. Bring back the polling code (for dumps mostly), but protect it with splbio() to make it safe for reprobes.
1996-01-29 03:17:39 +00:00
return (EIO);
}
ahc_intr((void *)ahc);
return (0);
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc_timeout(arg)
void *arg;
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
{
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
struct scb *scb = (struct scb *)arg;
struct ahc_data *ahc;
Enable/Disable Ultra mode on a per target basis. This allows the driver to use the full range of settings from 3.6-20MHz on any target. Remove all Ultra settings except for the top three that are documented to work. This fixes some problem reports with the last revision of the driver since at least the 5.7MHz entry doesn't work in Ultra mode.
1996-05-21 18:37:25 +00:00
int s, found;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
u_char bus_state;
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
s = splbio();
if (!(scb->flags & SCB_ACTIVE)) {
/* Previous timeout took care of me already */
splx(s);
return;
}
ahc = (struct ahc_data *)scb->xs->sc_link->adapter_softc;
if (ahc->in_timeout) {
/*
* Some other SCB has started a recovery operation
* and is still working on cleaning things up.
*/
if (scb->flags & SCB_TIMEDOUT) {
/*
* This SCB has been here before and is not the
* recovery SCB. Cut our losses and panic. Its
* better to do this than trash a filesystem.
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
panic("%s: Timed-out command times out "
"again\n", ahc_name(ahc));
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else if ((scb->flags & (SCB_ABORTED | SCB_DEVICE_RESET
| SCB_SENTORDEREDTAG)) == 0) {
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
/*
* This is not the SCB that started this timeout
* processing. Give this scb another lifetime so
* that it can continue once we deal with the
* timeout.
*/
scb->flags |= SCB_TIMEDOUT;
timeout(ahc_timeout, (caddr_t)scb,
(scb->xs->timeout * hz) / 1000);
splx(s);
return;
}
}
ahc->in_timeout = TRUE;
/*
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
* Ensure that the card doesn't do anything
* behind our back.
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
*/
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
pause_sequencer(ahc);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
sc_print_addr(scb->xs->sc_link);
printf("timed out ");
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
/*
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* Take a snapshot of the bus state and print out
* some information so we can track down driver bugs.
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
bus_state = AHC_INB(ahc, LASTPHASE);
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
switch(bus_state & PHASE_MASK)
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
case P_DATAOUT:
printf("in dataout phase");
break;
case P_DATAIN:
printf("in datain phase");
break;
case P_COMMAND:
printf("in command phase");
break;
case P_MESGOUT:
printf("in message out phase");
break;
case P_STATUS:
printf("in status phase");
break;
case P_MESGIN:
printf("in message in phase");
break;
default:
printf("while idle, LASTPHASE == 0x%x",
bus_state);
/*
* We aren't in a valid phase, so assume we're
* idle.
*/
bus_state = 0;
break;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
printf(", SCSISIGI == 0x%x\n", AHC_INB(ahc, SCSISIGI));
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
/* Decide our course of action */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (scb->flags & SCB_ABORTED) {
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
/*
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
* Been down this road before.
* Do a full bus reset.
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
char channel = (scb->hscb->tcl & SELBUSB)
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
? 'B': 'A';
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
found = ahc_reset_channel(ahc, channel, scb,
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
XS_TIMEOUT, /*Initiate Reset*/TRUE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
printf("%s: Issued Channel %c Bus Reset. "
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
"%d SCBs aborted\n", ahc_name(ahc), channel, found);
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc->in_timeout = FALSE;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else if ((scb->hscb->control & TAG_ENB) != 0
&& (scb->flags & SCB_SENTORDEREDTAG) == 0) {
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 19:21:20 +00:00
/*
* We could be starving this command
* try sending an ordered tag command
* to the target we come from.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
scb->flags |= SCB_SENTORDEREDTAG;
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 19:21:20 +00:00
ahc->orderedtag |= 0xFF;
timeout(ahc_timeout, (caddr_t)scb, (5 * hz));
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
unpause_sequencer(ahc, /*unpause_always*/FALSE);
Fix the abort code in the BUS DEVICE RESET case. Needed to set cmdlen to 0. Ensure that queued commands are not touched by the abort code by setting the SCB status to indicate what queue it is in. Fix deadlocks when using SCB paging by using SCBs from the assigned_scbs queue or an SCB that completed during the same interrupt if needed. Don't ever use insl to pull SCBs from any of the controllers. You can only do 8bit PIO reads. This only affected SCB paging. With this checkin, SCB paging works quite a bit better, but I still have some problems with it that may be caused by a firmware problem in my PD1800s. It seems that using a tag number higher than the maximum number of tags allowed by the device, confuses it. For example, if I queue two commands, tagged 3 and 36, it never reconnects for tag 36.
1996-04-28 19:21:20 +00:00
printf("Ordered Tag queued\n");
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else {
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
/*
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* Send an Abort Message:
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* The target that is holding up the bus may not
* be the same as the one that triggered this timeout
* (different commands have different timeout lengths).
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* Our strategy here is to queue an abort message
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* to the timed out target if it is disconnected.
* Otherwise, if we have an active target we stuff the
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
* message buffer with an abort message and assert ATN
* in the hopes that the target will let go of the bus
* and go to the mesgout phase. If this fails, we'll
* get another timeout 2 seconds later which will attempt
* a bus reset.
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t saved_scbptr;
u_int8_t active_scb_index;
struct scb *active_scb;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
saved_scbptr = AHC_INB(ahc, SCBPTR);
active_scb_index = AHC_INB(ahc, SCB_TAG);
active_scb = ahc->scbarray[active_scb_index];
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (bus_state != 0) {
/* Send the abort to the active SCB */
AHC_OUTB(ahc, MSG_LEN, 1);
AHC_OUTB(ahc, MSG0,
(active_scb->hscb->control & TAG_ENB) == 0 ?
MSG_ABORT : MSG_ABORT_TAG);
AHC_OUTB(ahc, SCSISIGO, bus_state|ATNO);
sc_print_addr(active_scb->xs->sc_link);
printf("abort message in message buffer\n");
active_scb->flags |= SCB_ABORTED;
if (active_scb != scb) {
untimeout(ahc_timeout,
(caddr_t)active_scb);
/* Give scb a new lease on life */
timeout(ahc_timeout, (caddr_t)scb,
(scb->xs->timeout * hz) / 1000);
}
timeout(ahc_timeout, (caddr_t)active_scb, (2 * hz));
unpause_sequencer(ahc, /*unpause_always*/FALSE);
} else {
u_int8_t hscb_index;
int disconnected;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
disconnected = FALSE;
hscb_index = find_scb(ahc, scb);
if (hscb_index == SCB_LIST_NULL)
disconnected = TRUE;
else {
AHC_OUTB(ahc, SCBPTR, hscb_index);
if (AHC_INB(ahc, SCB_CONTROL) & DISCONNECTED)
disconnected = TRUE;
Implement SCB paging which allows up to 255 active commands on aic7770 (Rev E or greater), aic7850, aic7860, aic7870, and aic7880 controllers. SCB paging is enabled with the option "AHC_SCBPAGING_ENABLE". Full comments on the algorithm are at the top of i386/scsi/aic7xxx.c. options "AHC_TAGENABLE" and "AHC_QUEUE_FULL" have been removed. The default is 4 tags without SCB paging, 8 with. Clear the SCSIRSTI bit after throwing a bus reset. Some cards seem to get confused otherwise. Handle SCSIRSTI interrupts before checking to see if there is a valid SCB in use since this can happen. (Clears PR# i386/1123) Clean up the way we determine the number of SCBs on the card (courtesy of Dan Eischen). Guard against attempts to negotiate wide to a narrow controller. Fix some comments. Update my copyrights.
1996-04-20 21:29:27 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
scb->flags |= SCB_ABORTED;
if (disconnected) {
/* Simply set the ABORT_SCB control bit */
scb->hscb->control |= ABORT_SCB;
if (hscb_index != SCB_LIST_NULL)
AHC_OUTB(ahc, SCB_CONTROL, ABORT_SCB);
timeout(ahc_timeout, (caddr_t)scb, (2 * hz));
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
AHC_OUTB(ahc, SCBPTR, saved_scbptr);
unpause_sequencer(ahc, /*unpause_always*/FALSE);
if (!disconnected)
/* Go "immediatly" to the bus reset */
timeout(ahc_timeout, (caddr_t)scb, hz / 2);
Handle both channels of Twin channeled devices. Respect the configuration values for syncronous negotiation. The 284x series adaptors can now be supported without the Bios being enabled. If you disable the Bios on the 274x series adaptors, all configuration parameters revert to the default since there is no way to retrieve them.
1994-12-31 19:31:56 +00:00
}
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
Properly deal with the Ultra series of adapters. We should now understand the new seeprom format and negotiate up to 20MHz sync if set in SCSI-Select. Reduce the complexity of the timeout code by running it at splhigh(). Fix a bug that caused rescheduled timeouts at 0 clock ticks in the future causing an infinite loop. Obtained from: Timeout bug noticed by David Greenman and wcarchive.
1995-10-26 23:57:18 +00:00
splx(s);
My port of Julian's 1742 driver to work with the Linux aic7770 sequencer code.
1994-11-17 20:22:31 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
/*
* Look through the SCB array of the card and attempt to find the
* hardware SCB that corresponds to the passed in SCB. Return
* SCB_LIST_NULL if unsuccessful. This routine assumes that the
* card is already paused.
*/
static u_int8_t
find_scb(ahc, scb)
struct ahc_data *ahc;
struct scb *scb;
{
u_int8_t saved_scbptr;
u_int8_t curindex;
saved_scbptr = AHC_INB(ahc, SCBPTR);
curindex = 0;
for (curindex = 0; curindex < ahc->maxhscbs; curindex++) {
AHC_OUTB(ahc, SCBPTR, curindex);
if (AHC_INB(ahc, SCB_TAG) == scb->hscb->tag)
break;
}
AHC_OUTB(ahc, SCBPTR, saved_scbptr);
if (curindex > ahc->maxhscbs)
curindex = SCB_LIST_NULL;
return curindex;
}
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
/*
* The device at the given target/channel has been reset. Abort
* all active and queued scbs for that target/channel.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static int
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
ahc_reset_device(ahc, target, channel, timedout_scb, xs_error)
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
struct ahc_data *ahc;
int target;
char channel;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
struct scb *timedout_scb;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
u_int32_t xs_error;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
{
struct scb *scbp;
u_char active_scb;
int i = 0;
int found = 0;
/* restore this when we're done */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
active_scb = AHC_INB(ahc, SCBPTR);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
/*
* Search the QINFIFO.
*/
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t saved_queue[AHC_SCB_MAX];
u_int8_t queued = AHC_INB(ahc, QINCNT) & ahc->qcntmask;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
for (i = 0; i < (queued - found); i++) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
saved_queue[i] = AHC_INB(ahc, QINFIFO);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
scbp = ahc->scbarray[saved_queue[i]];
if (ahc_match_scb (scbp, target, channel)) {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
/*
* We found an scb that needs to be aborted.
*/
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
scbp->flags = SCB_ABORTED|SCB_QUEUED_FOR_DONE;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
scbp->xs->error |= xs_error;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if(scbp != timedout_scb)
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
untimeout(ahc_timeout, (caddr_t)scbp);
i--;
found++;
}
}
/* Now put the saved scbs back. */
for (queued = 0; queued < i; queued++) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, QINFIFO, saved_queue[queued]);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
}
/*
* Search waiting for selection list.
*/
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t next, prev;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
next = AHC_INB(ahc, WAITING_SCBH); /* Start at head of list. */
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
prev = SCB_LIST_NULL;
while (next != SCB_LIST_NULL) {
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCBPTR, next);
scbp = ahc->scbarray[AHC_INB(ahc, SCB_TAG)];
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
if (ahc_match_scb(scbp, target, channel)) {
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
next = ahc_abort_wscb(ahc, scbp, next, prev,
timedout_scb, xs_error);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
found++;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
prev = next;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
next = AHC_INB(ahc, SCB_NEXT);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
}
}
/*
* Go through the entire SCB array now and look for
* commands for this target that are active. These
* are other (most likely tagged) commands that
* were disconnected when the reset occured.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
for (i = 0; i < ahc->numscbs; i++) {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
scbp = ahc->scbarray[i];
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if ((scbp->flags & SCB_ACTIVE)
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
&& ahc_match_scb(scbp, target, channel)) {
/* Ensure the target is "free" */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc_unbusy_target(ahc, target, channel);
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
scbp->flags = SCB_ABORTED|SCB_QUEUED_FOR_DONE;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
scbp->xs->error |= xs_error;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (scbp != timedout_scb)
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
untimeout(ahc_timeout, (caddr_t)scbp);
found++;
}
}
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCBPTR, active_scb);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
return found;
}
/*
* Manipulate the waiting for selection list and return the
* scb that follows the one that we remove.
*/
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static u_char
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
ahc_abort_wscb (ahc, scbp, scbpos, prev, timedout_scb, xs_error)
struct ahc_data *ahc;
struct scb *scbp;
u_int8_t scbpos;
u_int8_t prev;
struct scb *timedout_scb;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
u_int32_t xs_error;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t curscb, next;
int target = ((scbp->hscb->tcl >> 4) & 0x0f);
char channel = (scbp->hscb->tcl & SELBUSB) ? 'B' : 'A';
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
/*
* Select the SCB we want to abort and
* pull the next pointer out of it.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
curscb = AHC_INB(ahc, SCBPTR);
AHC_OUTB(ahc, SCBPTR, scbpos);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
next = AHC_INB(ahc, SCB_NEXT);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
/* Clear the necessary fields */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCB_CONTROL, 0);
AHC_OUTB(ahc, SCB_NEXT, SCB_LIST_NULL);
ahc_unbusy_target(ahc, target, channel);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
/* update the waiting list */
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (prev == SCB_LIST_NULL)
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
/* First in the list */
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, WAITING_SCBH, next);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
else {
/*
* Select the scb that pointed to us
* and update its next pointer.
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCBPTR, prev);
AHC_OUTB(ahc, SCB_NEXT, next);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
/*
* Point us back at the original scb position
* and inform the SCSI system that the command
* has been aborted.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
AHC_OUTB(ahc, SCBPTR, curscb);
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
scbp->flags = SCB_ABORTED|SCB_QUEUED_FOR_DONE;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
scbp->xs->error |= xs_error;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (scbp != timedout_scb)
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
untimeout(ahc_timeout, (caddr_t)scbp);
return next;
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc_unbusy_target(ahc, target, channel)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
struct ahc_data *ahc;
int target;
char channel;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t active_scb;
u_int8_t info_scb;
u_int32_t scb_offset;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
info_scb = target / 4;
if (channel == 'B')
info_scb += 2;
active_scb = AHC_INB(ahc, SCBPTR);
AHC_OUTB(ahc, SCBPTR, info_scb);
scb_offset = SCB_ACTIVE0 + (target & 0x03);
AHC_OUTB(ahc, scb_offset, SCB_LIST_NULL);
AHC_OUTB(ahc, SCBPTR, active_scb);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static void
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc_reset_current_bus(ahc)
struct ahc_data *ahc;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
{
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCSISEQ, SCSIRSTO);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
DELAY(1000);
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SCSISEQ, 0);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static int
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc_reset_channel(ahc, channel, timedout_scb, xs_error, initiate_reset)
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
struct ahc_data *ahc;
char channel;
struct scb *timedout_scb;
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
u_int32_t xs_error;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
int initiate_reset;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t sblkctl;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
char cur_channel;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int32_t offset, offset_max;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
int found;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
int target;
int maxtarget;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
maxtarget = 8;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
/*
* Clean up all the state information for the
* pending transactions on this bus.
*/
The long awaited stability patch set for the aic7xxx driver: Simplify the initialization of adapters by pulling all card specific initialization to the card specific modules. Update comments and fix formating. Pass struct ahc_data*'s to functions instead of unit numbers. Take advantage of the quad word alignment of SCB fields. Adapt to new sequencer changes: 1) Waiting scb list no longer has a tail. 2) Fill the message buffer as appropriate during a parity error. 3) Count all of the SGs involved in a residual instead of just the current one. The reset/abort code still needs a lot of work. Reviewed by: David Greenman <davidg@FreeBSd.org>
1996-01-03 06:32:12 +00:00
found = ahc_reset_device(ahc, ALL_TARGETS, channel,
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
timedout_scb, xs_error);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (channel == 'B') {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
ahc->needsdtr |= (ahc->needsdtr_orig & 0xff00);
ahc->sdtrpending &= 0x00ff;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
offset = TARG_SCRATCH + 8;
offset_max = TARG_SCRATCH + 16;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else if (ahc->type & AHC_WIDE){
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
ahc->needsdtr = ahc->needsdtr_orig;
ahc->needwdtr = ahc->needwdtr_orig;
ahc->sdtrpending = 0;
ahc->wdtrpending = 0;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
maxtarget = 16;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
offset = TARG_SCRATCH;
offset_max = TARG_SCRATCH + 16;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
ahc->needsdtr |= (ahc->needsdtr_orig & 0x00ff);
ahc->sdtrpending &= 0xff00;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
offset = TARG_SCRATCH;
offset_max = TARG_SCRATCH + 8;
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
for (target = 0; target < maxtarget; target++)
ahc_unbusy_target(ahc, target, channel);
for (; offset < offset_max; offset++) {
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
/*
* Revert to async/narrow transfers
* until we renegotiate.
*/
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
u_int8_t targ_scratch;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
targ_scratch = AHC_INB(ahc, offset);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
targ_scratch &= SXFR;
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, offset, targ_scratch);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
/*
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
* Reset the bus if we are initiating this reset and
* restart/unpause the sequencer
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
sblkctl = AHC_INB(ahc, SBLKCTL);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
cur_channel = (sblkctl & SELBUSB) ? 'B' : 'A';
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (cur_channel != channel) {
/* Case 1: Command for another bus is active
* Stealthily reset the other bus without
* upsetting the current bus.
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
*/
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
AHC_OUTB(ahc, SBLKCTL, sblkctl ^ SELBUSB);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (initiate_reset)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc_reset_current_bus(ahc);
AHC_OUTB(ahc, CLRSINT1, CLRSCSIRSTI|CLRSELTIMEO);
AHC_OUTB(ahc, CLRINT, CLRSCSIINT);
AHC_OUTB(ahc, SBLKCTL, sblkctl);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
unpause_sequencer(ahc, /*unpause_always*/TRUE);
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
} else {
/* Case 2: A command from this bus is active or we're idle */
if (initiate_reset)
Merge in changes for NetBSD/OpenBSD. Add a panic for attempts to page in a non paged out SCB. Re-order some of the interrupt routine for better performance. NetBSD/OpenBSD support Submitted by:Noriyuki Soda <soda@sra.co.jp>, Pete Bentley <pete@demon.net>, Charles M. Hannum <mycroft@mit.edu>, Theo de Raadt <deraadt@theos.com>
1996-05-30 07:19:59 +00:00
ahc_reset_current_bus(ahc);
AHC_OUTB(ahc, CLRSINT1, CLRSCSIRSTI|CLRSELTIMEO);
AHC_OUTB(ahc, CLRINT, CLRSCSIINT);
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
restart_sequencer(ahc);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
}
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc_run_done_queue(ahc);
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
return found;
}
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
void
ahc_run_done_queue(ahc)
struct ahc_data *ahc;
{
int i;
struct scb *scbp;
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
for (i = 0; i < ahc->numscbs; i++) {
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
scbp = ahc->scbarray[i];
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
if (scbp->flags & SCB_QUEUED_FOR_DONE)
bt.c, aic7xxx.c: Cleanse the SCSI subsystem of its internally defined types u_int32, u_int16, u_int8, int32, int16, int8. Use the system defined *_t types instead. aic7xxx.c: Fix the reset code. Instead of queing up all of the SCBs that timeout during timeout processing, we take the first and have it champion the effort. Any other scbs that timeout during timeout handling are given another lifetime to complete in the hopes that once timeout handing is finished, they will complete normally. If one of these SCBs times out a second time, we panic and Justin tries again. The other major change is to queue flag aborted SCBs during timeout handling, and "ahc_done" them all at once as soon as we have the controller back into a sane state. Calling ahc_done any earlier will cause the SCSI subsystem to toss the command right back at us and the attempt to queue the command will conflict with what the timeout routine is trying to accomplish. The aic7xxx driver will now respond to bus resets initiated by other devices.
1996-03-10 07:11:45 +00:00
ahc_done(ahc, scbp);
}
}
Cleanup, make things static.
1995-10-31 18:41:49 +00:00
static int
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
ahc_match_scb (scb, target, channel)
struct scb *scb;
int target;
char channel;
{
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
int targ = (scb->hscb->tcl >> 4) & 0x0f;
char chan = (scb->hscb->tcl & SELBUSB) ? 'B' : 'A';
Long overdue, more complete, reset code. These changes implement a BUS DEVICE RESET followed by BUS RESET failure recovery strategy including the necesary renegotiation of sync/wide transfers after recovery completes. Clean up debugging code to make it more finely selectable. Reset code debugging is enabled for now so I can get more feedback on how this code behaves in real life.
1995-07-31 08:25:36 +00:00
if (target == ALL_TARGETS)
return (chan == channel);
else
return ((chan == channel) && (targ == target));
}
Bring aic7xxx driver bug fixes from 'SCSI' into current.
1996-10-06 16:38:45 +00:00
static void
ahc_construct_sdtr(ahc, start_byte, period, offset)
struct ahc_data *ahc;
int start_byte;
u_int8_t period;
u_int8_t offset;
{
AHC_OUTB(ahc, MSG0 + start_byte, MSG_EXTENDED);
AHC_OUTB(ahc, MSG1 + start_byte, MSG_EXT_SDTR_LEN);
AHC_OUTB(ahc, MSG2 + start_byte, MSG_EXT_SDTR);
AHC_OUTB(ahc, MSG3 + start_byte, period);
AHC_OUTB(ahc, MSG4 + start_byte, offset);
AHC_OUTB(ahc, MSG_LEN, start_byte + 5);
}
static void
ahc_construct_wdtr(ahc, start_byte, bus_width)
struct ahc_data *ahc;
int start_byte;
u_int8_t bus_width;
{
AHC_OUTB(ahc, MSG0 + start_byte, MSG_EXTENDED);
AHC_OUTB(ahc, MSG1 + start_byte, MSG_EXT_WDTR_LEN);
AHC_OUTB(ahc, MSG2 + start_byte, MSG_EXT_WDTR);
AHC_OUTB(ahc, MSG3 + start_byte, bus_width);
AHC_OUTB(ahc, MSG_LEN, start_byte + 4);
}
- KNF cleanup. - Add support for memory mapped I/O. - Use DMA to get SCBs down to the adapters. - Remove old paging code. - Be much smarter about how we allocate SCB space. The old, simple method wasted almost half a page per SCB. Ooops. - Make command complete interrupt processing more efficient. - Break the monolithic ahc_intr into sub-routines. The sub-routines handle rare, special case events so the function call is not a penalty and the removal of the code from the main routine most likely improves performance instruction prefech will work better and less code is pushed into the cache. - Never, ever allow tagged queueing if a device has disconnection disabled. - Clean up and simplify timeout code. Many of the changes are to handle the new DMA scheme.
1996-10-25 06:42:53 +00:00
static void
ahc_calc_residual(scb)
struct scb *scb;
{
struct scsi_xfer *xs;
struct hardware_scb *hscb;
int resid_sgs;
xs = scb->xs;
hscb = scb->hscb;
if ((scb->flags & SCB_SENSE) == 0) {
/*
* Remainder of the SG where the transfer
* stopped.
*/
xs->resid = (hscb->residual_data_count[2] <<16) |
(hscb->residual_data_count[1] <<8) |
(hscb->residual_data_count[0]);
/*
* Add up the contents of all residual
* SG segments that are after the SG where
* the transfer stopped.
*/
resid_sgs = hscb->residual_SG_segment_count - 1;
while (resid_sgs > 0) {
int sg;
sg = hscb->SG_segment_count - resid_sgs;
xs->resid += scb->ahc_dma[sg].len;
resid_sgs--;
}
#if defined(__FreeBSD__)
xs->flags |= SCSI_RESID_VALID;
#elif defined(__NetBSD__)
/* XXX - Update to do this right */
#endif
}
/*
* Clean out the residual information in this SCB for the
* next consumer of this SCB.
*/
hscb->residual_data_count[2] = 0;
hscb->residual_data_count[1] = 0;
hscb->residual_data_count[0] = 0;
hscb->residual_SG_segment_count = 0;
#ifdef AHC_DEBUG
if (ahc_debug & AHC_SHOWMISC) {
sc_print_addr(xs->sc_link);
printf("Handled Residual of %ld bytes\n" ,xs->resid);
}
#endif
}
Reference in New Issue Copy Permalink