freebsd-skq/sys/dev/fdc/fdc.c

2024 lines
47 KiB
C
Raw Normal View History

/*-
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* Copyright (c) 2004 Poul-Henning Kamp
1993-06-12 14:58:17 +00:00
* Copyright (c) 1990 The Regents of the University of California.
* All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Don Ahn.
*
* Libretto PCMCIA floppy support by David Horwitt (dhorwitt@ucsd.edu)
* aided by the Linux floppy driver modifications from David Bateman
* (dbateman@eng.uts.edu.au).
*
* Copyright (c) 1993, 1994 by
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
* jc@irbs.UUCP (John Capo)
* vak@zebub.msk.su (Serge Vakulenko)
* ache@astral.msk.su (Andrew A. Chernov)
*
* Copyright (c) 1993, 1994, 1995 by
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
* joerg_wunsch@uriah.sax.de (Joerg Wunsch)
* dufault@hda.com (Peter Dufault)
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
*
* Copyright (c) 2001 Joerg Wunsch,
* joerg_wunsch@uriah.heep.sax.de (Joerg Wunsch)
*
1993-06-12 14:58:17 +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, 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.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
1993-06-12 14:58:17 +00:00
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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.
*
* from: @(#)fd.c 7.4 (Berkeley) 5/25/91
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
*
1993-06-12 14:58:17 +00:00
*/
2003-06-11 00:34:37 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_fdc.h"
1994-02-07 04:27:13 +00:00
#include <sys/param.h>
#include <sys/bio.h>
#include <sys/bus.h>
#include <sys/devicestat.h>
#include <sys/disk.h>
#include <sys/fcntl.h>
#include <sys/fdcio.h>
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
#include <sys/filio.h>
#include <sys/kernel.h>
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
#include <sys/kthread.h>
#include <sys/lock.h>
1994-02-07 04:27:13 +00:00
#include <sys/malloc.h>
#include <sys/module.h>
#include <sys/mutex.h>
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
#include <sys/proc.h>
#include <sys/rman.h>
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
#include <sys/sysctl.h>
#include <sys/systm.h>
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
#include <geom/geom.h>
#include <machine/bus.h>
#include <machine/clock.h>
#include <machine/stdarg.h>
#include <isa/isavar.h>
#include <isa/isareg.h>
#include <dev/fdc/fdcvar.h>
#include <isa/rtc.h>
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
#include <dev/ic/nec765.h>
/*
* Runtime configuration hints/flags
*/
/* configuration flags for fd */
#define FD_TYPEMASK 0x0f /* drive type, matches enum
* fd_drivetype; on i386 machines, if
* given as 0, use RTC type for fd0
* and fd1 */
#define FD_NO_PROBE 0x20 /* don't probe drive (seek test), just
* assume it is there */
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/*
* Things that could conceiveably considered parameters or tweakables
*/
/*
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* Maximal number of bytes in a cylinder.
* This is used for ISADMA bouncebuffer allocation and sets the max
* xfersize we support.
*
* 2.88M format has 2 x 36 x 512, allow for hacked up density.
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
#define MAX_BYTES_PER_CYL (2 * 40 * 512)
/*
* Timeout value for the PIO loops to wait until the FDC main status
* register matches our expectations (request for master, direction
* bit). This is supposed to be a number of microseconds, although
* timing might actually not be very accurate.
*
* Timeouts of 100 msec are believed to be required for some broken
* (old) hardware.
*/
#define FDSTS_TIMEOUT 100000
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
/*
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* After this many errors, stop whining. Close will reset this count.
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
*/
#define FDC_ERRMAX 100
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
/*
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* AutoDensity search lists for each drive type.
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static struct fd_type fd_searchlist_360k[] = {
{ FDF_5_360 },
{ 0 }
1993-06-12 14:58:17 +00:00
};
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
static struct fd_type fd_searchlist_12m[] = {
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
{ FDF_5_1200 | FL_AUTO },
{ FDF_5_360 | FL_2STEP | FL_AUTO},
{ 0 }
};
static struct fd_type fd_searchlist_720k[] = {
{ FDF_3_720 },
{ 0 }
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
};
static struct fd_type fd_searchlist_144m[] = {
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
{ FDF_3_1440 | FL_AUTO},
{ FDF_3_720 | FL_AUTO},
{ 0 }
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
};
static struct fd_type fd_searchlist_288m[] = {
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
{ FDF_3_1440 | FL_AUTO },
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
#if 0
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
{ FDF_3_2880 | FL_AUTO }, /* XXX: probably doesn't work */
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
#endif
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
{ FDF_3_720 | FL_AUTO},
{ 0 }
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
};
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/*
* Order must match enum fd_drivetype in <sys/fdcio.h>.
*/
static struct fd_type *fd_native_types[] = {
NULL, /* FDT_NONE */
fd_searchlist_360k, /* FDT_360K */
fd_searchlist_12m, /* FDT_12M */
fd_searchlist_720k, /* FDT_720K */
fd_searchlist_144m, /* FDT_144M */
fd_searchlist_288m, /* FDT_288M_1 (mapped to FDT_288M) */
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd_searchlist_288m, /* FDT_288M */
};
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/*
* Internals start here
*/
/* registers */
#define FDOUT 2 /* Digital Output Register (W) */
#define FDO_FDSEL 0x03 /* floppy device select */
#define FDO_FRST 0x04 /* floppy controller reset */
#define FDO_FDMAEN 0x08 /* enable floppy DMA and Interrupt */
#define FDO_MOEN0 0x10 /* motor enable drive 0 */
#define FDO_MOEN1 0x20 /* motor enable drive 1 */
#define FDO_MOEN2 0x40 /* motor enable drive 2 */
#define FDO_MOEN3 0x80 /* motor enable drive 3 */
#define FDSTS 4 /* NEC 765 Main Status Register (R) */
#define FDDATA 5 /* NEC 765 Data Register (R/W) */
#define FDCTL 7 /* Control Register (W) */
/*
* The YE-DATA PC Card floppies use PIO to read in the data rather
* than DMA due to the wild variability of DMA for the PC Card
* devices. DMA was deleted from the PC Card specification in version
* 7.2 of the standard, but that post-dates the YE-DATA devices by many
* years.
*
* In addition, if we cannot setup the DMA resources for the ISA
* attachment, we'll use this same offset for data transfer. However,
* that almost certainly won't work.
*
* For this mode, offset 0 and 1 must be used to setup the transfer
* for this floppy. This is OK for PC Card YE Data devices, but for
* ISA this is likely wrong. These registers are only available on
* those systems that map them to the floppy drive. Newer systems do
* not do this, and we should likely prohibit access to them (or
* disallow NODMA to be set).
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
*/
#define FDBCDR 0 /* And 1 */
#define FD_YE_DATAPORT 6 /* Drive Data port */
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
#define FDI_DCHG 0x80 /* diskette has been changed */
/* requires drive and motor being selected */
/* is cleared by any step pulse to drive */
/*
* We have three private BIO commands.
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
*/
#define BIO_PROBE BIO_CMD0
#define BIO_RDID BIO_CMD1
#define BIO_FMT BIO_CMD2
1993-06-12 14:58:17 +00:00
/*
* Per drive structure (softc).
*/
struct fd_data {
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
u_char *fd_ioptr; /* IO pointer */
u_int fd_iosize; /* Size of IO chunks */
u_int fd_iocount; /* Outstanding requests */
1994-02-07 04:27:13 +00:00
struct fdc_data *fdc; /* pointer to controller structure */
1993-06-12 14:58:17 +00:00
int fdsu; /* this units number on this controller */
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
enum fd_drivetype type; /* drive type */
struct fd_type *ft; /* pointer to current type descriptor */
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fd_type fts; /* type descriptors */
int sectorsize;
1993-06-12 14:58:17 +00:00
int flags;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
#define FD_WP (1<<0) /* Write protected */
#define FD_MOTOR (1<<1) /* motor should be on */
#define FD_MOTORWAIT (1<<2) /* motor should be on */
#define FD_EMPTY (1<<3) /* no media */
#define FD_NEWDISK (1<<4) /* media changed */
#define FD_ISADMA (1<<5) /* isa dma started */
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
int track; /* where we think the head is */
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
#define FD_NO_TRACK -2
int options; /* FDOPT_* */
struct callout toffhandle;
struct callout tohandle;
struct g_geom *fd_geom;
struct g_provider *fd_provider;
device_t dev;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct bio_queue_head fd_bq;
};
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
#define FD_NOT_VALID -2
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static driver_intr_t fdc_intr;
static void fdc_reset(struct fdc_data *);
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
SYSCTL_NODE(_debug, OID_AUTO, fdc, CTLFLAG_RW, 0, "fdc driver");
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static int fifo_threshold = 8;
SYSCTL_INT(_debug_fdc, OID_AUTO, fifo, CTLFLAG_RW, &fifo_threshold, 0,
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
"FIFO threshold setting");
1994-02-07 04:27:13 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static int debugflags = 0;
SYSCTL_INT(_debug_fdc, OID_AUTO, debugflags, CTLFLAG_RW, &debugflags, 0,
"Debug flags");
static int retries = 10;
SYSCTL_INT(_debug_fdc, OID_AUTO, retries, CTLFLAG_RW, &retries, 0,
"Number of retries to attempt");
static int spec1 = 0xaf;
SYSCTL_INT(_debug_fdc, OID_AUTO, spec1, CTLFLAG_RW, &spec1, 0,
"Specification byte one (step-rate + head unload)");
static int spec2 = 0x10;
SYSCTL_INT(_debug_fdc, OID_AUTO, spec2, CTLFLAG_RW, &spec2, 0,
"Specification byte two (head load time + no-dma)");
static int settle;
SYSCTL_INT(_debug_fdc, OID_AUTO, settle, CTLFLAG_RW, &settle, 0,
"Head settling time in sec/hz");
static void
fdprinttype(struct fd_type *ft)
{
printf("(%d,%d,%d,%d,%d,%d,%d,%d,%d,%d,%d,0x%x)",
ft->sectrac, ft->secsize, ft->datalen, ft->gap, ft->tracks,
ft->size, ft->trans, ft->heads, ft->f_gap, ft->f_inter,
ft->offset_side2, ft->flags);
}
static void
fdsettype(struct fd_data *fd, struct fd_type *ft)
{
fd->ft = ft;
ft->size = ft->sectrac * ft->heads * ft->tracks;
fd->sectorsize = 128 << fd->ft->secsize;
}
1993-06-12 14:58:17 +00:00
/*
* Bus space handling (access to low-level IO).
*/
__inline static void
fdregwr(struct fdc_data *fdc, int reg, uint8_t v)
{
bus_space_write_1(fdc->iot, fdc->ioh[reg], fdc->ioff[reg], v);
}
__inline static uint8_t
fdregrd(struct fdc_data *fdc, int reg)
{
return bus_space_read_1(fdc->iot, fdc->ioh[reg], fdc->ioff[reg]);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static void
fdctl_wr(struct fdc_data *fdc, u_int8_t v)
{
fdregwr(fdc, FDCTL, v);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
}
static void
fdout_wr(struct fdc_data *fdc, u_int8_t v)
1999-11-11 08:48:40 +00:00
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdregwr(fdc, FDOUT, v);
1999-11-11 08:48:40 +00:00
}
static u_int8_t
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdsts_rd(struct fdc_data *fdc)
1999-11-11 08:48:40 +00:00
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
return fdregrd(fdc, FDSTS);
1999-11-11 08:48:40 +00:00
}
static void
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fddata_wr(struct fdc_data *fdc, u_int8_t v)
1999-11-11 08:48:40 +00:00
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdregwr(fdc, FDDATA, v);
1999-11-11 08:48:40 +00:00
}
static u_int8_t
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fddata_rd(struct fdc_data *fdc)
1999-11-11 08:48:40 +00:00
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
return fdregrd(fdc, FDDATA);
1999-11-11 08:48:40 +00:00
}
static u_int8_t
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdin_rd(struct fdc_data *fdc)
1999-11-11 08:48:40 +00:00
{
return fdregrd(fdc, FDCTL);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
}
/*
* Magic pseudo-DMA initialization for YE FDC. Sets count and
* direction.
*/
static void
fdbcdr_wr(struct fdc_data *fdc, int iswrite, uint16_t count)
{
fdregwr(fdc, FDBCDR, (count - 1) & 0xff);
fdregwr(fdc, FDBCDR + 1,
(iswrite ? 0x80 : 0) | (((count - 1) >> 8) & 0x7f));
}
static int
fdc_err(struct fdc_data *fdc, const char *s)
{
fdc->fdc_errs++;
if (s) {
if (fdc->fdc_errs < FDC_ERRMAX)
device_printf(fdc->fdc_dev, "%s", s);
else if (fdc->fdc_errs == FDC_ERRMAX)
device_printf(fdc->fdc_dev, "too many errors, not "
"logging any more\n");
}
1995-05-30 08:16:23 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
return (1);
}
/*
* FDC IO functions, take care of the main status register, timeout
* in case the desired status bits are never set.
*
* These PIO loops initially start out with short delays between
* each iteration in the expectation that the required condition
* is usually met quickly, so it can be handled immediately.
*/
static int
fdc_in(struct fdc_data *fdc, int *ptr)
{
int i, j, step;
step = 1;
for (j = 0; j < FDSTS_TIMEOUT; j += step) {
i = fdsts_rd(fdc) & (NE7_DIO | NE7_RQM);
if (i == (NE7_DIO|NE7_RQM)) {
i = fddata_rd(fdc);
if (ptr)
*ptr = i;
return (0);
}
if (i == NE7_RQM)
return (fdc_err(fdc, "ready for output in input\n"));
step += step;
DELAY(step);
}
return (fdc_err(fdc, bootverbose? "input ready timeout\n": 0));
}
static int
fdc_out(struct fdc_data *fdc, int x)
{
int i, j, step;
step = 1;
for (j = 0; j < FDSTS_TIMEOUT; j += step) {
i = fdsts_rd(fdc) & (NE7_DIO | NE7_RQM);
if (i == NE7_RQM) {
fddata_wr(fdc, x);
return (0);
}
if (i == (NE7_DIO|NE7_RQM))
return (fdc_err(fdc, "ready for input in output\n"));
step += step;
DELAY(step);
}
return (fdc_err(fdc, bootverbose? "output ready timeout\n": 0));
}
/*
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* fdc_cmd: Send a command to the chip.
* Takes a varargs with this structure:
* # of output bytes
* output bytes as int [...]
* # of input bytes
* input bytes as int* [...]
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static int
fdc_cmd(struct fdc_data *fdc, int n_out, ...)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
u_char cmd = 0;
int n_in;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
int n, i;
va_list ap;
va_start(ap, n_out);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
for (n = 0; n < n_out; n++) {
i = va_arg(ap, int);
if (n == 0)
cmd = i;
if (fdc_out(fdc, i) < 0) {
char msg[50];
snprintf(msg, sizeof(msg),
"cmd %x failed at out byte %d of %d\n",
cmd, n + 1, n_out);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc->flags |= FDC_NEEDS_RESET;
va_end(ap);
return fdc_err(fdc, msg);
}
}
n_in = va_arg(ap, int);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
for (n = 0; n < n_in; n++) {
int *ptr = va_arg(ap, int *);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (fdc_in(fdc, ptr) < 0) {
char msg[50];
snprintf(msg, sizeof(msg),
"cmd %02x failed at in byte %d of %d\n",
cmd, n + 1, n_in);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc->flags |= FDC_NEEDS_RESET;
va_end(ap);
return fdc_err(fdc, msg);
}
}
va_end(ap);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
return (0);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static void
fdc_reset(struct fdc_data *fdc)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
int i, r[10];
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Try a reset, keep motor on */
fdout_wr(fdc, fdc->fdout & ~(FDO_FRST|FDO_FDMAEN));
DELAY(100);
/* enable FDC, but defer interrupts a moment */
fdout_wr(fdc, fdc->fdout & ~FDO_FDMAEN);
DELAY(100);
fdout_wr(fdc, fdc->fdout);
/* XXX after a reset, silently believe the FDC will accept commands */
if (fdc_cmd(fdc, 3, NE7CMD_SPECIFY, spec1, spec2, 0))
device_printf(fdc->fdc_dev, " SPECIFY failed in reset\n");
if (fdc->fdct == FDC_ENHANCED) {
if (fdc_cmd(fdc, 4,
I8207X_CONFIGURE,
0,
0x40 | /* Enable Implied Seek */
0x10 | /* Polling disabled */
(fifo_threshold - 1), /* Fifo threshold */
0x00, /* Precomp track */
0))
device_printf(fdc->fdc_dev,
" CONFIGURE failed in reset\n");
if (debugflags & 1) {
if (fdc_cmd(fdc, 1,
0x0e, /* DUMPREG */
10, &r[0], &r[1], &r[2], &r[3], &r[4],
&r[5], &r[6], &r[7], &r[8], &r[9]))
device_printf(fdc->fdc_dev,
" DUMPREG failed in reset\n");
for (i = 0; i < 10; i++)
printf(" %02x", r[i]);
printf("\n");
}
}
}
1995-12-10 13:40:44 +00:00
static int
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc_sense_drive(struct fdc_data *fdc, int *st3p)
{
int st3;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (fdc_cmd(fdc, 2, NE7CMD_SENSED, fdc->fd->fdsu, 1, &st3))
return (fdc_err(fdc, "Sense Drive Status failed\n"));
if (st3p)
*st3p = st3;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
return (0);
}
1995-12-10 13:40:44 +00:00
static int
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc_sense_int(struct fdc_data *fdc, int *st0p, int *cylp)
{
int cyl, st0, ret;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
ret = fdc_cmd(fdc, 1, NE7CMD_SENSEI, 1, &st0);
if (ret) {
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
(void)fdc_err(fdc, "sense intr err reading stat reg 0\n");
return (ret);
}
if (st0p)
*st0p = st0;
if ((st0 & NE7_ST0_IC) == NE7_ST0_IC_IV) {
/*
* There doesn't seem to have been an interrupt.
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
return (FD_NOT_VALID);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (fdc_in(fdc, &cyl) < 0)
return fdc_err(fdc, "can't get cyl num\n");
if (cylp)
*cylp = cyl;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
return (0);
}
1995-12-10 13:40:44 +00:00
static int
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc_read_status(struct fdc_data *fdc)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
int i, ret, status;
for (i = ret = 0; i < 7; i++) {
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
ret = fdc_in(fdc, &status);
fdc->status[i] = status;
if (ret != 0)
break;
}
if (ret == 0)
fdc->flags |= FDC_STAT_VALID;
else
fdc->flags &= ~FDC_STAT_VALID;
return ret;
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/*
* Select this drive
*/
static void
fd_select(struct fd_data *fd)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fdc_data *fdc;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* XXX: lock controller */
fdc = fd->fdc;
fdc->fdout &= ~FDO_FDSEL;
fdc->fdout |= FDO_FDMAEN | FDO_FRST | fd->fdsu;
fdout_wr(fdc, fdc->fdout);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static void
fd_turnon(void *arg)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fd_data *fd;
struct bio *bp;
int once;
fd = arg;
mtx_lock(&fd->fdc->fdc_mtx);
fd->flags &= ~FD_MOTORWAIT;
fd->flags |= FD_MOTOR;
once = 0;
for (;;) {
bp = bioq_takefirst(&fd->fd_bq);
if (bp == NULL)
break;
bioq_disksort(&fd->fdc->head, bp);
once = 1;
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
mtx_unlock(&fd->fdc->fdc_mtx);
if (once)
wakeup(&fd->fdc->head);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static void
fd_motor(struct fd_data *fd, int turnon)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fdc_data *fdc;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc = fd->fdc;
/*
mtx_assert(&fdc->fdc_mtx, MA_OWNED);
*/
if (turnon) {
fd->flags |= FD_MOTORWAIT;
fdc->fdout |= (FDO_MOEN0 << fd->fdsu);
callout_reset(&fd->toffhandle, hz, fd_turnon, fd);
} else {
callout_drain(&fd->toffhandle);
fd->flags &= ~(FD_MOTOR|FD_MOTORWAIT);
fdc->fdout &= ~(FDO_MOEN0 << fd->fdsu);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdout_wr(fdc, fdc->fdout);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static void
fd_turnoff(void *xfd)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fd_data *fd = xfd;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
mtx_lock(&fd->fdc->fdc_mtx);
fd_motor(fd, 0);
mtx_unlock(&fd->fdc->fdc_mtx);
Make the Y-E Data PCMCIA floppy of the Toshiba Libretto work under -current. It doesn't work yet as stable as the 3.x/PAO version of the driver does, however, i get occasional `FDC direction bit not set' and other weird messages, but it basically works at least. The old (defunct) #ifdef FDC_YE stuff has been eliminated completely now, PCMCIA-FDC specific functions have been implemented differently where needed. Unfortunately, due to the fact that the traditional PeeCee FDC with its funny non-contiguous register space (one register for WD1003 harddisk controllers is interleaved into the FDC register set), and Peter's subsequent changes involving two different bus space handles for normal FDCs, the changes required for the Y-E stuff are more complex than i'd love them to be. I've done my best to keep the logic for normal FDCs intact. Since the Y-E FDC seems to lose interrupts after a FDC reset sometimes, i've also replaced the timeout logic in fd_turnoff() to generate an artificial pseudo interrupt in case of a timeout while the drive has still outstanding transfers waiting. This avoids the total starvation of the driver that could be observed with highly damaged media under 3.x/PAO. This part of the patch has been revied by bde previously. I've fixed a number of occasions where previous commits have been missing the encapuslation of ISA DMA related functions inside FDC_NODMA checks. I've added one call to SET_BCDR() during preparation of the format floppy operation. Floppy formatting has been totally broken before in 3.x/PAO (garbage ID fields have been written to the medium, causing `wrong cylinder' errors upon media reading). This is just black magic, i don't have the slightes idea _why_ this needs to be but just copied over the hack that has been used by the PAO folks in the normal read/write case anyway. The entired device_busy() stuff seems to be pointless to me. In any case, i had to add device_unbusy() calls symmetrical to the device_busy() calls, otherwise the PCMCIA floppy driver could never be deactivated. (As it used to be, it caused a `mark the device busier and busier' situation.) IMHO, all block device drivers should be marked busy based on active buffers still waiting for the driver, so the device_unbusy() calls should probably go to biodone(). Only one other driver (whose name escapes me at the moment) uses device_busy() calls at all, so i question the value of all this... I think this entire `device busy' logic simply doesn't fit for PCMCIA &al. It cannot be the decision of some piece of kernel software to declare a device `busy by now, you can't remove it', when the actual physical power of removing it is the user pulling the card. The kernel simply has to cope with the removal, however busy the device might have been by the time of the removal, period. Perhaps a force flag needs to be added? Upon inserting the card a second time, i get: WARNING: "fd" is usurping "fd"'s cdevsw[] WARNING: "fd" is usurping "fd"'s bmaj I suspect this is related to the XXX comment at the call to cdevsw_add(). Does anybody know what the correct way is to cleanup this?
2000-03-18 18:27:01 +00:00
}
1993-06-12 14:58:17 +00:00
/*
* fdc_intr - wake up the worker thread.
1993-06-12 14:58:17 +00:00
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static void
fdc_intr(void *arg)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
wakeup(arg);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/*
* fdc_pio(): perform programmed IO read/write for YE PCMCIA floppy.
*/
static void
fdc_pio(struct fdc_data *fdc)
{
u_char *cptr;
struct bio *bp;
u_int count;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
bp = fdc->bp;
cptr = fdc->fd->fd_ioptr;
count = fdc->fd->fd_iosize;
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (bp->bio_cmd == BIO_READ) {
fdbcdr_wr(fdc, 0, count);
bus_space_read_multi_1(fdc->iot, fdc->ioh[FD_YE_DATAPORT],
fdc->ioff[FD_YE_DATAPORT], cptr, count);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
} else {
bus_space_write_multi_1(fdc->iot, fdc->ioh[FD_YE_DATAPORT],
fdc->ioff[FD_YE_DATAPORT], cptr, count);
fdbcdr_wr(fdc, 0, count); /* needed? */
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
}
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static int
fdc_biodone(struct fdc_data *fdc, int error)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fd_data *fd;
struct bio *bp;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd = fdc->fd;
bp = fdc->bp;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
mtx_lock(&fdc->fdc_mtx);
if (--fd->fd_iocount == 0)
callout_reset(&fd->toffhandle, 4 * hz, fd_turnoff, fd);
fdc->bp = NULL;
fdc->fd = NULL;
mtx_unlock(&fdc->fdc_mtx);
if (bp->bio_to != NULL) {
if ((debugflags & 2) && fd->fdc->retry > 0)
printf("retries: %d\n", fd->fdc->retry);
g_io_deliver(bp, error);
return (0);
}
bp->bio_error = error;
bp->bio_flags |= BIO_DONE;
wakeup(bp);
return (0);
}
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static int retry_line;
static int
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc_worker(struct fdc_data *fdc)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fd_data *fd;
struct bio *bp;
int i, nsect;
int st0, st3, cyl, mfm, steptrac, cylinder, descyl, sec;
int head;
static int need_recal;
struct fdc_readid *idp;
struct fd_formb *finfo;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Have we exhausted our retries ? */
bp = fdc->bp;
fd = fdc->fd;
if (bp != NULL &&
(fdc->retry >= retries || (fd->options & FDOPT_NORETRY))) {
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if ((debugflags & 4))
printf("Too many retries (EIO)\n");
return (fdc_biodone(fdc, EIO));
}
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Disable ISADMA if we bailed while it was active */
if (fd != NULL && (fd->flags & FD_ISADMA)) {
mtx_lock(&Giant);
isa_dmadone(
bp->bio_cmd & BIO_READ ? ISADMA_READ : ISADMA_WRITE,
fd->fd_ioptr, fd->fd_iosize, fdc->dmachan);
mtx_unlock(&Giant);
mtx_lock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->flags &= ~FD_ISADMA;
mtx_unlock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Unwedge the controller ? */
if (fdc->flags & FDC_NEEDS_RESET) {
fdc->flags &= ~FDC_NEEDS_RESET;
fdc_reset(fdc);
msleep(fdc, NULL, PRIBIO, "fdcrst", hz);
/* Discard results */
for (i = 0; i < 4; i++)
fdc_sense_int(fdc, &st0, &cyl);
/* All drives must recal */
need_recal = 0xf;
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Pick up a request, if need be wait for it */
if (fdc->bp == NULL) {
mtx_lock(&fdc->fdc_mtx);
do {
fdc->bp = bioq_takefirst(&fdc->head);
if (fdc->bp == NULL)
msleep(&fdc->head, &fdc->fdc_mtx,
PRIBIO, "-", hz);
} while (fdc->bp == NULL &&
(fdc->flags & FDC_KTHREAD_EXIT) == 0);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
mtx_unlock(&fdc->fdc_mtx);
if (fdc->bp == NULL)
/*
* Nothing to do, worker thread has been
* requested to stop.
*/
return (0);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
bp = fdc->bp;
fd = fdc->fd = bp->bio_driver1;
fdc->retry = 0;
fd->fd_ioptr = bp->bio_data;
if (bp->bio_cmd & BIO_FMT) {
i = offsetof(struct fd_formb, fd_formb_cylno(0));
fd->fd_ioptr += i;
fd->fd_iosize = bp->bio_length - i;
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
}
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Select drive, setup params */
fd_select(fd);
fdctl_wr(fdc, fd->ft->trans);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (bp->bio_cmd & BIO_PROBE) {
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (!(fdin_rd(fdc) & FDI_DCHG) && !(fd->flags & FD_EMPTY))
return (fdc_biodone(fdc, 0));
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/*
* Try to find out if we have a disk in the drive
*
* First recal, then seek to cyl#1, this clears the
* old condition on the disk change line so we can
* examine it for current status
*/
if (debugflags & 0x40)
printf("New disk in probe\n");
mtx_lock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->flags |= FD_NEWDISK;
mtx_unlock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
retry_line = __LINE__;
if (fdc_cmd(fdc, 2, NE7CMD_RECAL, fd->fdsu, 0))
return (1);
msleep(fdc, NULL, PRIBIO, "fdrecal", hz);
retry_line = __LINE__;
if (fdc_sense_int(fdc, &st0, &cyl) == FD_NOT_VALID)
return (1); /* XXX */
retry_line = __LINE__;
if ((st0 & 0xc0) || cyl != 0)
return (1);
/* Seek to track 1 */
retry_line = __LINE__;
if (fdc_cmd(fdc, 3, NE7CMD_SEEK, fd->fdsu, 1, 0))
return (1);
msleep(fdc, NULL, PRIBIO, "fdseek", hz);
retry_line = __LINE__;
if (fdc_sense_int(fdc, &st0, &cyl) == FD_NOT_VALID)
return (1); /* XXX */
need_recal |= (1 << fd->fdsu);
if (fdin_rd(fdc) & FDI_DCHG) {
if (debugflags & 0x40)
printf("Empty in probe\n");
mtx_lock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->flags |= FD_EMPTY;
mtx_unlock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
} else {
if (debugflags & 0x40)
printf("Got disk in probe\n");
mtx_lock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->flags &= ~FD_EMPTY;
mtx_unlock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
retry_line = __LINE__;
if(fdc_sense_drive(fdc, &st3) != 0)
return (1);
mtx_lock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if(st3 & NE7_ST3_WP)
fd->flags |= FD_WP;
else
fd->flags &= ~FD_WP;
mtx_unlock(&fdc->fdc_mtx);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
return (fdc_biodone(fdc, 0));
}
1995-05-30 08:16:23 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/*
* If we are dead just flush the requests
*/
if (fd->flags & FD_EMPTY)
return (fdc_biodone(fdc, ENXIO));
/* Check if we lost our media */
if (fdin_rd(fdc) & FDI_DCHG) {
if (debugflags & 0x40)
printf("Lost disk\n");
mtx_lock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->flags |= FD_EMPTY;
fd->flags |= FD_NEWDISK;
mtx_unlock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
g_topology_lock();
g_orphan_provider(fd->fd_provider, EXDEV);
fd->fd_provider->flags |= G_PF_WITHER;
fd->fd_provider =
g_new_providerf(fd->fd_geom, fd->fd_geom->name);
g_error_provider(fd->fd_provider, 0);
g_topology_unlock();
return (fdc_biodone(fdc, ENXIO));
}
1995-05-30 08:16:23 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Check if the floppy is write-protected */
if(bp->bio_cmd & (BIO_FMT | BIO_WRITE)) {
retry_line = __LINE__;
if(fdc_sense_drive(fdc, &st3) != 0)
return (1);
if(st3 & NE7_ST3_WP)
return (fdc_biodone(fdc, EROFS));
}
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
mfm = (fd->ft->flags & FL_MFM)? NE7CMD_MFM: 0;
steptrac = (fd->ft->flags & FL_2STEP)? 2: 1;
i = fd->ft->sectrac * fd->ft->heads;
cylinder = bp->bio_pblkno / i;
descyl = cylinder * steptrac;
sec = bp->bio_pblkno % i;
nsect = i - sec;
head = sec / fd->ft->sectrac;
sec = sec % fd->ft->sectrac + 1;
/* If everything is going swimmingly, use multisector xfer */
if (fdc->retry == 0 && bp->bio_cmd & (BIO_READ|BIO_WRITE)) {
fd->fd_iosize = imin(nsect * fd->sectorsize, bp->bio_resid);
nsect = fd->fd_iosize / fd->sectorsize;
} else if (bp->bio_cmd & (BIO_READ|BIO_WRITE)) {
fd->fd_iosize = fd->sectorsize;
nsect = 1;
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Do RECAL if we need to or are going to track zero anyway */
if ((need_recal & (1 << fd->fdsu)) ||
(cylinder == 0 && fd->track != 0) ||
fdc->retry > 2) {
retry_line = __LINE__;
if (fdc_cmd(fdc, 2, NE7CMD_RECAL, fd->fdsu, 0))
return (1);
msleep(fdc, NULL, PRIBIO, "fdrecal", hz);
retry_line = __LINE__;
if (fdc_sense_int(fdc, &st0, &cyl) == FD_NOT_VALID)
return (1); /* XXX */
retry_line = __LINE__;
if ((st0 & 0xc0) || cyl != 0)
return (1);
need_recal &= ~(1 << fd->fdsu);
fd->track = 0;
/* let the heads settle */
if (settle)
msleep(fdc->fd, NULL, PRIBIO, "fdhdstl", settle);
}
/*
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* SEEK to where we want to be
*
* Enhanced controllers do implied seeks for read&write as long as
* we do not need multiple steps per track.
*/
if (cylinder != fd->track && (
fdc->fdct != FDC_ENHANCED ||
descyl != cylinder ||
(bp->bio_cmd & (BIO_RDID|BIO_FMT)))) {
retry_line = __LINE__;
if (fdc_cmd(fdc, 3, NE7CMD_SEEK, fd->fdsu, descyl, 0))
return (1);
msleep(fdc, NULL, PRIBIO, "fdseek", hz);
retry_line = __LINE__;
if (fdc_sense_int(fdc, &st0, &cyl) == FD_NOT_VALID)
return (1); /* XXX */
retry_line = __LINE__;
if ((st0 & 0xc0) || cyl != descyl) {
need_recal |= (1 << fd->fdsu);
return (1);
}
/* let the heads settle */
if (settle)
msleep(fdc->fd, NULL, PRIBIO, "fdhdstl", settle);
}
fd->track = cylinder;
if (debugflags & 8)
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
printf("op %x bn %ju siz %u ptr %p retry %d\n",
bp->bio_cmd, bp->bio_pblkno, fd->fd_iosize,
fd->fd_ioptr, fdc->retry);
/* Setup ISADMA if we need it and have it */
if ((bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_FMT))
&& !(fdc->flags & FDC_NODMA)) {
mtx_lock(&Giant);
isa_dmastart(
bp->bio_cmd & BIO_READ ? ISADMA_READ : ISADMA_WRITE,
fd->fd_ioptr, fd->fd_iosize, fdc->dmachan);
mtx_unlock(&Giant);
mtx_lock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->flags |= FD_ISADMA;
mtx_unlock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
}
/* Do PIO if we have to */
if (fdc->flags & FDC_NODMA) {
if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_FMT))
fdbcdr_wr(fdc, 1, fd->fd_iosize);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (bp->bio_cmd & (BIO_WRITE|BIO_FMT))
fdc_pio(fdc);
}
switch(bp->bio_cmd) {
case BIO_FMT:
/* formatting */
finfo = (struct fd_formb *)bp->bio_data;
retry_line = __LINE__;
if (fdc_cmd(fdc, 6,
NE7CMD_FORMAT | mfm,
head << 2 | fd->fdsu,
finfo->fd_formb_secshift,
finfo->fd_formb_nsecs,
finfo->fd_formb_gaplen,
finfo->fd_formb_fillbyte, 0))
return (1);
break;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
case BIO_RDID:
retry_line = __LINE__;
if (fdc_cmd(fdc, 2,
NE7CMD_READID | mfm,
head << 2 | fd->fdsu, 0))
return (1);
break;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
case BIO_READ:
retry_line = __LINE__;
if (fdc_cmd(fdc, 9,
NE7CMD_READ | NE7CMD_SK | mfm | NE7CMD_MT,
head << 2 | fd->fdsu, /* head & unit */
fd->track, /* track */
head, /* head */
sec, /* sector + 1 */
fd->ft->secsize, /* sector size */
fd->ft->sectrac, /* sectors/track */
fd->ft->gap, /* gap size */
fd->ft->datalen, /* data length */
0))
return (1);
break;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
case BIO_WRITE:
retry_line = __LINE__;
if (fdc_cmd(fdc, 9,
NE7CMD_WRITE | mfm | NE7CMD_MT,
head << 2 | fd->fdsu, /* head & unit */
fd->track, /* track */
head, /* head */
sec, /* sector + 1 */
fd->ft->secsize, /* sector size */
fd->ft->sectrac, /* sectors/track */
fd->ft->gap, /* gap size */
fd->ft->datalen, /* data length */
0))
return (1);
break;
default:
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
KASSERT(0 == 1, ("Wrong bio_cmd %x\n", bp->bio_cmd));
}
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Wait for interrupt */
i = msleep(fdc, NULL, PRIBIO, "fddata", hz);
/* PIO if the read looks good */
if (i == 0 && (fdc->flags & FDC_NODMA) && (bp->bio_cmd & BIO_READ))
fdc_pio(fdc);
/* Finish DMA */
if (fd->flags & FD_ISADMA) {
mtx_lock(&Giant);
isa_dmadone(
bp->bio_cmd & BIO_READ ? ISADMA_READ : ISADMA_WRITE,
fd->fd_ioptr, fd->fd_iosize, fdc->dmachan);
mtx_unlock(&Giant);
mtx_lock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->flags &= ~FD_ISADMA;
mtx_unlock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (i != 0) {
/*
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* Timeout.
*
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* Due to IBM's brain-dead design, the FDC has a faked ready
* signal, hardwired to ready == true. Thus, any command
* issued if there's no diskette in the drive will _never_
* complete, and must be aborted by resetting the FDC.
* Many thanks, Big Blue!
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
retry_line = __LINE__;
fdc->flags |= FDC_NEEDS_RESET;
return (1);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
retry_line = __LINE__;
if (fdc_read_status(fdc))
return (1);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (debugflags & 0x10)
printf(" -> %x %x %x %x\n",
fdc->status[0], fdc->status[1],
fdc->status[2], fdc->status[3]);
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
st0 = fdc->status[0] & NE7_ST0_IC;
if (st0 != 0) {
retry_line = __LINE__;
if (st0 == NE7_ST0_IC_AT && fdc->status[1] & NE7_ST1_OR) {
/*
* DMA overrun. Someone hogged the bus and
* didn't release it in time for the next
* FDC transfer.
*/
return (1);
}
retry_line = __LINE__;
if(st0 == NE7_ST0_IC_IV) {
fdc->flags |= FDC_NEEDS_RESET;
return (1);
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
retry_line = __LINE__;
if(st0 == NE7_ST0_IC_AT && fdc->status[2] & NE7_ST2_WC) {
need_recal |= (1 << fd->fdsu);
return (1);
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (debugflags & 0x20) {
printf("status %02x %02x %02x %02x %02x %02x\n",
fdc->status[0], fdc->status[1], fdc->status[2],
fdc->status[3], fdc->status[4], fdc->status[5]);
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
retry_line = __LINE__;
return (1);
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* All OK */
switch(bp->bio_cmd) {
case BIO_RDID:
/* copy out ID field contents */
idp = (struct fdc_readid *)bp->bio_data;
idp->cyl = fdc->status[3];
idp->head = fdc->status[4];
idp->sec = fdc->status[5];
idp->secshift = fdc->status[6];
if (debugflags & 0x40)
printf("c %d h %d s %d z %d\n",
idp->cyl, idp->head, idp->sec, idp->secshift);
break;
case BIO_READ:
case BIO_WRITE:
bp->bio_pblkno += nsect;
bp->bio_resid -= fd->fd_iosize;
bp->bio_completed += fd->fd_iosize;
fd->fd_ioptr += fd->fd_iosize;
/* Since we managed to get something done, reset the retry */
fdc->retry = 0;
if (bp->bio_resid > 0)
return (0);
break;
case BIO_FMT:
break;
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
return (fdc_biodone(fdc, 0));
}
static void
fdc_thread(void *arg)
{
struct fdc_data *fdc;
fdc = arg;
int i;
mtx_lock(&fdc->fdc_mtx);
fdc->flags |= FDC_KTHREAD_ALIVE;
while ((fdc->flags & FDC_KTHREAD_EXIT) == 0) {
mtx_unlock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
i = fdc_worker(fdc);
if (i && debugflags & 0x20) {
if (fdc->bp != NULL) {
g_print_bio(fdc->bp);
printf("\n");
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
printf("Retry line %d\n", retry_line);
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc->retry += i;
mtx_lock(&fdc->fdc_mtx);
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
}
fdc->flags &= ~(FDC_KTHREAD_EXIT | FDC_KTHREAD_ALIVE);
wakeup(&fdc->fdc_thread);
mtx_unlock(&fdc->fdc_mtx);
kthread_exit(0);
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
}
/*
* Enqueue a request.
*/
static void
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd_enqueue(struct fd_data *fd, struct bio *bp)
1993-06-12 14:58:17 +00:00
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fdc_data *fdc;
int call;
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
call = 0;
fdc = fd->fdc;
mtx_lock(&fdc->fdc_mtx);
/* If we go from idle, cancel motor turnoff */
if (fd->fd_iocount++ == 0)
callout_drain(&fd->toffhandle);
if (fd->flags & FD_MOTOR) {
/* The motor is on, send it directly to the controller */
bioq_disksort(&fdc->head, bp);
wakeup(&fdc->head);
} else {
/* Queue it on the drive until the motor has started */
bioq_insert_tail(&fd->fd_bq, bp);
if (!(fd->flags & FD_MOTORWAIT))
fd_motor(fd, 1);
1993-06-12 14:58:17 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
mtx_unlock(&fdc->fdc_mtx);
1993-06-12 14:58:17 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static int
fdmisccmd(struct fd_data *fd, u_int cmd, void *data)
1993-06-12 14:58:17 +00:00
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct bio *bp;
struct fd_formb *finfo;
struct fdc_readid *idfield;
int error;
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
bp = malloc(sizeof(struct bio), M_TEMP, M_WAITOK | M_ZERO);
1993-06-12 14:58:17 +00:00
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
/*
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* Set up a bio request for fdstrategy(). bio_offset is faked
* so that fdstrategy() will seek to the the requested
* cylinder, and use the desired head.
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
bp->bio_cmd = cmd;
if (cmd == BIO_FMT) {
finfo = (struct fd_formb *)data;
bp->bio_pblkno =
(finfo->cyl * fd->ft->heads + finfo->head) *
fd->ft->sectrac;
bp->bio_length = sizeof *finfo;
} else if (cmd == BIO_RDID) {
idfield = (struct fdc_readid *)data;
bp->bio_pblkno =
(idfield->cyl * fd->ft->heads + idfield->head) *
fd->ft->sectrac;
bp->bio_length = sizeof(struct fdc_readid);
} else if (cmd == BIO_PROBE) {
/* nothing */
} else
panic("wrong cmd in fdmisccmd()");
bp->bio_offset = bp->bio_pblkno * fd->sectorsize;
bp->bio_data = data;
bp->bio_driver1 = fd;
bp->bio_flags = 0;
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd_enqueue(fd, bp);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
do {
msleep(bp, NULL, PRIBIO, "fdwait", hz);
} while (!(bp->bio_flags & BIO_DONE));
error = bp->bio_error;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
free(bp, M_TEMP);
return (error);
}
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
/*
* Try figuring out the density of the media present in our device.
*/
static int
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdautoselect(struct fd_data *fd)
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
{
struct fd_type *fdtp;
struct fdc_readid id;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
int oopts, rv;
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (!(fd->ft->flags & FL_AUTO))
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
return (0);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdtp = fd_native_types[fd->type];
fdsettype(fd, fdtp);
if (!(fd->ft->flags & FL_AUTO))
return (0);
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
/*
* Try reading sector ID fields, first at cylinder 0, head 0,
* then at cylinder 2, head N. We don't probe cylinder 1,
* since for 5.25in DD media in a HD drive, there are no data
* to read (2 step pulses per media cylinder required). For
* two-sided media, the second probe always goes to head 1, so
* we can tell them apart from single-sided media. As a
* side-effect this means that single-sided media should be
* mentioned in the search list after two-sided media of an
* otherwise identical density. Media with a different number
* of sectors per track but otherwise identical parameters
* cannot be distinguished at all.
*
* If we successfully read an ID field on both cylinders where
* the recorded values match our expectation, we are done.
* Otherwise, we try the next density entry from the table.
*
* Stepping to cylinder 2 has the side-effect of clearing the
* unit attention bit.
*/
oopts = fd->options;
fd->options |= FDOPT_NOERRLOG | FDOPT_NORETRY;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
for (; fdtp->heads; fdtp++) {
fdsettype(fd, fdtp);
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
id.cyl = id.head = 0;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
rv = fdmisccmd(fd, BIO_RDID, &id);
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
if (rv != 0)
continue;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (id.cyl != 0 || id.head != 0 || id.secshift != fdtp->secsize)
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
continue;
id.cyl = 2;
id.head = fd->ft->heads - 1;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
rv = fdmisccmd(fd, BIO_RDID, &id);
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
if (id.cyl != 2 || id.head != fdtp->heads - 1 ||
id.secshift != fdtp->secsize)
continue;
if (rv == 0)
break;
}
fd->options = oopts;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (fdtp->heads == 0) {
if (debugflags & 0x40)
device_printf(fd->dev, "autoselection failed\n");
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdsettype(fd, fd_native_types[fd->type]);
return (0);
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
} else {
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (debugflags & 0x40) {
device_printf(fd->dev,
"autoselected %d KB medium\n", fd->ft->size / 2);
fdprinttype(fd->ft);
}
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
return (0);
}
}
/*
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
* GEOM class implementation
*/
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static g_access_t fd_access;
static g_start_t fd_start;
static g_ioctl_t fd_ioctl;
1993-06-12 14:58:17 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct g_class g_fd_class = {
.name = "FD",
.version = G_VERSION,
.start = fd_start,
.access = fd_access,
.ioctl = fd_ioctl,
};
Make the Y-E Data PCMCIA floppy of the Toshiba Libretto work under -current. It doesn't work yet as stable as the 3.x/PAO version of the driver does, however, i get occasional `FDC direction bit not set' and other weird messages, but it basically works at least. The old (defunct) #ifdef FDC_YE stuff has been eliminated completely now, PCMCIA-FDC specific functions have been implemented differently where needed. Unfortunately, due to the fact that the traditional PeeCee FDC with its funny non-contiguous register space (one register for WD1003 harddisk controllers is interleaved into the FDC register set), and Peter's subsequent changes involving two different bus space handles for normal FDCs, the changes required for the Y-E stuff are more complex than i'd love them to be. I've done my best to keep the logic for normal FDCs intact. Since the Y-E FDC seems to lose interrupts after a FDC reset sometimes, i've also replaced the timeout logic in fd_turnoff() to generate an artificial pseudo interrupt in case of a timeout while the drive has still outstanding transfers waiting. This avoids the total starvation of the driver that could be observed with highly damaged media under 3.x/PAO. This part of the patch has been revied by bde previously. I've fixed a number of occasions where previous commits have been missing the encapuslation of ISA DMA related functions inside FDC_NODMA checks. I've added one call to SET_BCDR() during preparation of the format floppy operation. Floppy formatting has been totally broken before in 3.x/PAO (garbage ID fields have been written to the medium, causing `wrong cylinder' errors upon media reading). This is just black magic, i don't have the slightes idea _why_ this needs to be but just copied over the hack that has been used by the PAO folks in the normal read/write case anyway. The entired device_busy() stuff seems to be pointless to me. In any case, i had to add device_unbusy() calls symmetrical to the device_busy() calls, otherwise the PCMCIA floppy driver could never be deactivated. (As it used to be, it caused a `mark the device busier and busier' situation.) IMHO, all block device drivers should be marked busy based on active buffers still waiting for the driver, so the device_unbusy() calls should probably go to biodone(). Only one other driver (whose name escapes me at the moment) uses device_busy() calls at all, so i question the value of all this... I think this entire `device busy' logic simply doesn't fit for PCMCIA &al. It cannot be the decision of some piece of kernel software to declare a device `busy by now, you can't remove it', when the actual physical power of removing it is the user pulling the card. The kernel simply has to cope with the removal, however busy the device might have been by the time of the removal, period. Perhaps a force flag needs to be added? Upon inserting the card a second time, i get: WARNING: "fd" is usurping "fd"'s cdevsw[] WARNING: "fd" is usurping "fd"'s bmaj I suspect this is related to the XXX comment at the call to cdevsw_add(). Does anybody know what the correct way is to cleanup this?
2000-03-18 18:27:01 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
static int
fd_access(struct g_provider *pp, int r, int w, int e)
{
struct fd_data *fd;
struct fdc_data *fdc;
int ar, aw, ae;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd = pp->geom->softc;
fdc = fd->fdc;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/*
* If our provider is withering, we can only get negative requests
* and we don't want to even see them
*/
if (pp->flags & G_PF_WITHER)
return (0);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
ar = r + pp->acr;
aw = w + pp->acw;
ae = e + pp->ace;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (ar == 0 && aw == 0 && ae == 0) {
device_unbusy(fd->dev);
return (0);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) {
if (fdmisccmd(fd, BIO_PROBE, NULL))
return (ENXIO);
if (fd->flags & FD_EMPTY)
return (ENXIO);
if (fd->flags & FD_NEWDISK) {
fdautoselect(fd);
mtx_lock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->flags &= ~FD_NEWDISK;
mtx_unlock(&fdc->fdc_mtx);
1993-06-12 14:58:17 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
device_busy(fd->dev);
1993-06-12 14:58:17 +00:00
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (w > 0 && (fd->flags & FD_WP))
return (EROFS);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
pp->sectorsize = fd->sectorsize;
pp->stripesize = fd->ft->heads * fd->ft->sectrac * fd->sectorsize;
pp->mediasize = pp->stripesize * fd->ft->tracks;
return (0);
1993-06-12 14:58:17 +00:00
}
static void
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd_start(struct bio *bp)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fdc_data * fdc;
struct fd_data * fd;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd = bp->bio_to->geom->softc;
fdc = fd->fdc;
bp->bio_driver1 = fd;
if (bp->bio_cmd & BIO_GETATTR) {
if (g_handleattr_int(bp, "GEOM::fwsectors", fd->ft->sectrac))
return;
if (g_handleattr_int(bp, "GEOM::fwheads", fd->ft->heads))
return;
g_io_deliver(bp, ENOIOCTL);
return;
}
if (!(bp->bio_cmd & (BIO_READ|BIO_WRITE))) {
g_io_deliver(bp, EOPNOTSUPP);
return;
}
bp->bio_pblkno = bp->bio_offset / fd->sectorsize;
bp->bio_resid = bp->bio_length;
fd_enqueue(fd, bp);
return;
}
static int
fd_ioctl(struct g_provider *pp, u_long cmd, void *data, int fflag, struct thread *td)
{
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fd_data *fd;
struct fdc_status *fsp;
struct fdc_readid *rid;
int error;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd = pp->geom->softc;
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
switch (cmd) {
case FD_GTYPE: /* get drive type */
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
*(struct fd_type *)data = *fd->ft;
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
return (0);
case FD_STYPE: /* set drive type */
2005-02-28 13:49:06 +00:00
if (!(fflag & FWRITE))
return (EPERM);
/*
* Allow setting drive type temporarily iff
* currently unset. Used for fdformat so any
* user can set it, and then start formatting.
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->fts = *(struct fd_type *)data;
if (fd->fts.sectrac) {
/* XXX: check for rubbish */
fdsettype(fd, &fd->fts);
} else {
fdsettype(fd, fd_native_types[fd->type]);
}
if (debugflags & 0x40)
fdprinttype(fd->ft);
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
return (0);
case FD_GOPTS: /* get drive options */
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
*(int *)data = fd->options;
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
return (0);
case FD_SOPTS: /* set drive options */
2005-02-28 13:49:06 +00:00
if (!(fflag & FWRITE))
return (EPERM);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->options = *(int *)data;
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
return (0);
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
case FD_CLRERR:
if (suser(td) != 0)
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
return (EPERM);
fd->fdc->fdc_errs = 0;
return (0);
case FD_GSTAT:
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fsp = (struct fdc_status *)data;
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
if ((fd->fdc->flags & FDC_STAT_VALID) == 0)
return (EINVAL);
memcpy(fsp->status, fd->fdc->status, 7 * sizeof(u_int));
return (0);
case FD_GDTYPE:
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
*(enum fd_drivetype *)data = fd->type;
Long promised major enhancement set for the floppy disk driver: . The main device node now supports automatic density selection for commonly used media densities. So you can stuff your 1.44 MB and 720 KB media into your drive and just access /dev/fd0, no questions asked. It's all that easy, isn't it? :) . Device density handling has been completely overhauled. The old way of hardwired kernel density knowledge is no longer there. Instead, the kernel now implements 16 subdevices per drive. The first subdevice uses automatic density selection, while the remaining 15 devices are freely programmable. They can be assigned an arbitrary name of the form /dev/fd[:digit]+.[:digit:]{1,4}, where the second number is meant to either implement device names that are mnemonic for their raw capacity (as it used to be), or they can alternatively be created as "anonymous" devices like fd0.1 through fd0.15, depending on the taste of the administrator. After creating a subdevice, it is initialized to the maximal native density of the respective drive type, so it needs to be customized for other densities by using fdcontrol(8). Pseudo-partition devices (fd0a through fd0h) are still supported as symlinks. . The old hack to use flags 0x1 to always assume drive 0 were there is no longer supported; this is now supposed to be done by wiring the devices down from the loader via device flags. On IA32 architectures, the first two drives are looked up in the CMOS configuration records though. On PCMCIA (i. e., the Y-E Data controller of the Toshiba Libretto), a single drive is always assumed. . Other specialities like disabling the FIFO and not probing the drive at boot-time are selected by per-controller or per-drive flags, too. . Unit attentions (media has been changed) are supposed to be detected now; density autoselection only occurs after a unit attention. (Can be turned off by a per-drive flag, this will cause each Fdopen() to perform the autoselection.) . FM floppies can be handled now (on controllers that actually support it -- not all do these days). . Fdopen() can be told to avoid density selection by setting O_NONBLOCK; this leaves the descriptor in a half-opened state where only a few ioctls are accepted. This is necessary to run fdformat on a device that uses automatic density selection (since you cannot autoselect on an unformatted medium, obviously). . Just differentiate between a plain old NE765 and the enhanced chips, but don't try more; the existing code was wrong and only misdetected the chips anyway. BUGS and TODOs: . All documentation update still needs to be done. . Formatting not-so-standard format yields unpredictable results; i have yet to figure out why this happens. "Standard" formats like 720 and 1440 KB do work, however. . rc scripts are needed to setup device nodes with nonstandard densities (like the old /dev/fdN.MMM we used to have). . Obtaining device flags from the kernel environment doesn't work yet, thus currently only drives that are present in (IA32) CMOS are really detected. Someone who knows the odds and ends about device flags is needed here, i can't figure out what i'm doing wrong. . 2.88 MB still needs to be done.
2001-12-15 19:09:04 +00:00
return (0);
case FD_FORM:
2005-02-28 13:49:06 +00:00
if (!(fflag & FWRITE))
return (EPERM);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (((struct fd_formb *)data)->format_version !=
FD_FORMAT_VERSION)
return (EINVAL); /* wrong version of formatting prog */
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
error = fdmisccmd(fd, BIO_FMT, data);
mtx_lock(&fd->fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->flags |= FD_NEWDISK;
mtx_unlock(&fd->fdc->fdc_mtx);
break;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
case FD_READID:
rid = (struct fdc_readid *)data;
if (rid->cyl > 85 || rid->head > 1)
return (EINVAL);
error = fdmisccmd(fd, BIO_RDID, data);
break;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
case FIONBIO:
case FIOASYNC:
/* For backwards compat with old fd*(8) tools */
error = 0;
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
break;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
default:
if (debugflags & 0x80)
printf("Unknown ioctl %lx\n", cmd);
error = ENOIOCTL;
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
break;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
}
return (error);
};
1995-05-30 08:16:23 +00:00
Updated driver to the 1.1.5 version: date: 1994/05/22 12:35:38; author: joerg; state: Exp; lines: +6 -6 First round of floppy changes. Try making `fd' more robust. New features: . ioctl command for setting the drive type (density etc.); restricted to the super-user . ioctl for getting/seting `drive options'; currently only option is FDOPT_NORETRY: inhibit the usual retries; used when verifying a newly formatted track Fixes: . function prototypes . made all internal functions `static' . cleaned up & corrected .h files . restructured, to make the chaotic function sequence more rational . compiled with -Wall, and cleared all warnings . introduced a mirror for the (write-only) `digital output register', to avoid the current kludge . device probing completed by seeking/recalibrating, and looking for track 0 being found . holding the controller down in reset state while it is idle (and thus saving allot of headaches) . make requests fail that are not a multiple of the (physical) sector size . removed the fixed physical sector size (512 bytes), allowing for any size the controller could handle (128/256/512/1024 bytes) . replaced some silly messages . fixed the TRACE* macro usage, debugging reports should be complete now again (debugging output is HUGE! though) . removed fd_timeout for SEEK command; seeks are always reported by the controller to succeed, since the `success' only refers to the controller's idea of success - there is no hardware line to tell about the seek end (other than the `track 0' line) . catch SENSEI's that report about a `terminated due to READY changed' status - could happen after a controller reset . converted ``hz / <something>'' divide operations to divisors that are powers of two, so gcc can optimize them into shifts . write/format operations are checked against a write-protected medium now *prior* starting the operation . error reports of `invalid command' and `wrong cylinder' will cause shortcuts in the retrier() now . fixed a bug in the retrier() causing bogus block numbers to be reported . fdformat() does care for errors now Known Bugs: . no attempts have been made (yet) to improve the performance . sometimes, bogus ``seek/recalib failed'' messages are logged; this is still a bug in the driver, but it's not harmful since it's usually caught by the retrier() Reviewed by: Submitted by: Obtained from:
1994-09-17 16:56:10 +00:00
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/*
* Configuration/initialization stuff, per controller.
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
devclass_t fdc_devclass;
static devclass_t fd_devclass;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
struct fdc_ivars {
int fdunit;
int fdtype;
};
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
void
fdc_release_resources(struct fdc_data *fdc)
{
device_t dev;
struct resource *last;
int i;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
dev = fdc->fdc_dev;
if (fdc->fdc_intr)
bus_teardown_intr(dev, fdc->res_irq, fdc->fdc_intr);
fdc->fdc_intr = NULL;
if (fdc->res_irq != NULL)
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
bus_release_resource(dev, SYS_RES_IRQ, fdc->rid_irq,
fdc->res_irq);
fdc->res_irq = NULL;
last = NULL;
for (i = 0; i < FDC_MAXREG; i++) {
if (fdc->resio[i] != NULL && fdc->resio[i] != last) {
bus_release_resource(dev, SYS_RES_IOPORT,
fdc->ridio[i], fdc->resio[i]);
last = fdc->resio[i];
fdc->resio[i] = NULL;
}
}
if (fdc->res_drq != NULL)
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
bus_release_resource(dev, SYS_RES_DRQ, fdc->rid_drq,
fdc->res_drq);
fdc->res_drq = NULL;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
}
int
fdc_read_ivar(device_t dev, device_t child, int which, uintptr_t *result)
{
struct fdc_ivars *ivars = device_get_ivars(child);
switch (which) {
case FDC_IVAR_FDUNIT:
*result = ivars->fdunit;
break;
case FDC_IVAR_FDTYPE:
*result = ivars->fdtype;
break;
default:
return (ENOENT);
}
return (0);
}
int
fdc_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
{
struct fdc_ivars *ivars = device_get_ivars(child);
switch (which) {
case FDC_IVAR_FDUNIT:
ivars->fdunit = value;
break;
case FDC_IVAR_FDTYPE:
ivars->fdtype = value;
break;
default:
return (ENOENT);
}
return (0);
}
int
fdc_initial_reset(device_t dev, struct fdc_data *fdc)
{
int ic_type, part_id;
/*
* A status value of 0xff is very unlikely, but not theoretically
* impossible, but it is far more likely to indicate an empty bus.
*/
if (fdsts_rd(fdc) == 0xff)
return (ENXIO);
/*
* Assert a reset to the floppy controller and check that the status
* register goes to zero.
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdout_wr(fdc, 0);
fdout_wr(fdc, 0);
if (fdsts_rd(fdc) != 0)
return (ENXIO);
/*
* Clear the reset and see it come ready.
*/
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdout_wr(fdc, FDO_FRST);
DELAY(100);
if (fdsts_rd(fdc) != 0x80)
return (ENXIO);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* Then, see if it can handle a command. */
if (fdc_cmd(fdc, 3, NE7CMD_SPECIFY, 0xaf, 0x1e, 0))
return (ENXIO);
/*
* Try to identify the chip.
*
* The i8272 datasheet documents that unknown commands
* will return ST0 as 0x80. The i8272 is supposedly identical
* to the NEC765.
* The i82077SL datasheet says 0x90 for the VERSION command,
* and several "superio" chips emulate this.
*/
if (fdc_cmd(fdc, 1, NE7CMD_VERSION, 1, &ic_type))
return (ENXIO);
if (fdc_cmd(fdc, 1, 0x18, 1, &part_id))
return (ENXIO);
2004-08-27 17:08:24 +00:00
if (bootverbose)
device_printf(dev,
"ic_type %02x part_id %02x\n", ic_type, part_id);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
switch (ic_type & 0xff) {
case 0x80:
device_set_desc(dev, "NEC 765 or clone");
fdc->fdct = FDC_NE765;
break;
case 0x81:
case 0x90:
device_set_desc(dev,
"Enhanced floppy controller");
fdc->fdct = FDC_ENHANCED;
break;
default:
device_set_desc(dev, "Generic floppy controller");
fdc->fdct = FDC_UNKNOWN;
break;
}
return (0);
}
int
fdc_detach(device_t dev)
{
struct fdc_data *fdc;
int error;
fdc = device_get_softc(dev);
/* have our children detached first */
if ((error = bus_generic_detach(dev)))
return (error);
/* kill worker thread */
fdc->flags |= FDC_KTHREAD_EXIT;
mtx_lock(&fdc->fdc_mtx);
wakeup(&fdc->head);
while ((fdc->flags & FDC_KTHREAD_ALIVE) != 0)
msleep(&fdc->fdc_thread, &fdc->fdc_mtx, PRIBIO, "fdcdet", 0);
mtx_unlock(&fdc->fdc_mtx);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
/* reset controller, turn motor off */
fdout_wr(fdc, 0);
if (!(fdc->flags & FDC_NODMA))
isa_dma_release(fdc->dmachan);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc_release_resources(fdc);
mtx_destroy(&fdc->fdc_mtx);
return (0);
}
/*
* Add a child device to the fdc controller. It will then be probed etc.
*/
device_t
fdc_add_child(device_t dev, const char *name, int unit)
{
struct fdc_ivars *ivar;
device_t child;
ivar = malloc(sizeof *ivar, M_DEVBUF /* XXX */, M_NOWAIT | M_ZERO);
if (ivar == NULL)
return (NULL);
child = device_add_child(dev, name, unit);
if (child == NULL) {
free(ivar, M_DEVBUF);
return (NULL);
}
device_set_ivars(child, ivar);
ivar->fdunit = unit;
ivar->fdtype = FDT_NONE;
if (resource_disabled(name, unit))
device_disable(child);
return (child);
}
int
fdc_attach(device_t dev)
{
struct fdc_data *fdc;
int error;
fdc = device_get_softc(dev);
fdc->fdc_dev = dev;
error = fdc_initial_reset(dev, fdc);
if (error) {
device_printf(dev, "does not respond\n");
return (error);
}
error = bus_setup_intr(dev, fdc->res_irq,
INTR_TYPE_BIO | INTR_ENTROPY | INTR_MPSAFE |
((fdc->flags & FDC_NOFAST) ? 0 : INTR_FAST),
fdc_intr, fdc, &fdc->fdc_intr);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (error) {
device_printf(dev, "cannot setup interrupt\n");
return (error);
}
if (!(fdc->flags & FDC_NODMA)) {
error = isa_dma_acquire(fdc->dmachan);
if (!error) {
error = isa_dma_init(fdc->dmachan,
MAX_BYTES_PER_CYL, M_WAITOK);
if (error)
isa_dma_release(fdc->dmachan);
}
if (error)
return (error);
}
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fdc->fdcu = device_get_unit(dev);
fdc->flags |= FDC_NEEDS_RESET;
mtx_init(&fdc->fdc_mtx, "fdc lock", NULL, MTX_DEF);
/* reset controller, turn motor off, clear fdout mirror reg */
fdout_wr(fdc, fdc->fdout = 0);
bioq_init(&fdc->head);
kthread_create(fdc_thread, fdc, &fdc->fdc_thread, 0, 0,
"fdc%d", device_get_unit(dev));
settle = hz / 8;
return (0);
}
int
fdc_hints_probe(device_t dev)
{
const char *name, *dname;
int i, error, dunit;
/*
* Probe and attach any children. We should probably detect
* devices from the BIOS unless overridden.
*/
name = device_get_nameunit(dev);
i = 0;
while ((resource_find_match(&i, &dname, &dunit, "at", name)) == 0) {
resource_int_value(dname, dunit, "drive", &dunit);
fdc_add_child(dev, dname, dunit);
}
if ((error = bus_generic_attach(dev)) != 0)
return (error);
return (0);
}
int
fdc_print_child(device_t me, device_t child)
{
int retval = 0, flags;
retval += bus_print_child_header(me, child);
retval += printf(" on %s drive %d", device_get_nameunit(me),
fdc_get_fdunit(child));
if ((flags = device_get_flags(me)) != 0)
retval += printf(" flags %#x", flags);
retval += printf("\n");
return (retval);
}
/*
* Configuration/initialization, per drive.
*/
static int
fd_probe(device_t dev)
{
int i, unit;
u_int st0, st3;
struct fd_data *fd;
struct fdc_data *fdc;
int fdsu;
int flags, type;
fdsu = fdc_get_fdunit(dev);
fd = device_get_softc(dev);
fdc = device_get_softc(device_get_parent(dev));
flags = device_get_flags(dev);
fd->dev = dev;
fd->fdc = fdc;
fd->fdsu = fdsu;
unit = device_get_unit(dev);
/* Auto-probe if fdinfo is present, but always allow override. */
type = flags & FD_TYPEMASK;
if (type == FDT_NONE && (type = fdc_get_fdtype(dev)) != FDT_NONE) {
fd->type = type;
goto done;
} else {
/* make sure fdautoselect() will be called */
fd->flags = FD_EMPTY;
fd->type = type;
}
#if (defined(__i386__) && !defined(PC98)) || defined(__amd64__)
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
if (fd->type == FDT_NONE && (unit == 0 || unit == 1)) {
/* Look up what the BIOS thinks we have. */
if (unit == 0)
fd->type = (rtcin(RTC_FDISKETTE) & 0xf0) >> 4;
else
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->type = rtcin(RTC_FDISKETTE) & 0x0f;
if (fd->type == FDT_288M_1)
fd->type = FDT_288M;
}
#endif /* __i386__ || __amd64__ */
/* is there a unit? */
if (fd->type == FDT_NONE)
return (ENXIO);
/*
mtx_lock(&fdc->fdc_mtx);
*/
/* select it */
fd_select(fd);
fd_motor(fd, 1);
fdc->fd = fd;
fdc_reset(fdc); /* XXX reset, then unreset, etc. */
DELAY(1000000); /* 1 sec */
if ((flags & FD_NO_PROBE) == 0) {
/* If we're at track 0 first seek inwards. */
if ((fdc_sense_drive(fdc, &st3) == 0) &&
(st3 & NE7_ST3_T0)) {
/* Seek some steps... */
if (fdc_cmd(fdc, 3, NE7CMD_SEEK, fdsu, 10, 0) == 0) {
/* ...wait a moment... */
DELAY(300000);
/* make ctrlr happy: */
fdc_sense_int(fdc, NULL, NULL);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
}
}
for (i = 0; i < 2; i++) {
/*
* we must recalibrate twice, just in case the
* heads have been beyond cylinder 76, since
* most FDCs still barf when attempting to
* recalibrate more than 77 steps
*/
/* go back to 0: */
if (fdc_cmd(fdc, 2, NE7CMD_RECAL, fdsu, 0) == 0) {
/* a second being enough for full stroke seek*/
DELAY(i == 0 ? 1000000 : 300000);
/* anything responding? */
if (fdc_sense_int(fdc, &st0, NULL) == 0 &&
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
(st0 & NE7_ST0_EC) == 0)
break; /* already probed succesfully */
}
}
}
fd_motor(fd, 0);
fdc->fd = NULL;
/*
mtx_unlock(&fdc->fdc_mtx);
*/
if ((flags & FD_NO_PROBE) == 0 &&
(st0 & NE7_ST0_EC) != 0) /* no track 0 -> no drive present */
return (ENXIO);
done:
switch (fd->type) {
case FDT_12M:
device_set_desc(dev, "1200-KB 5.25\" drive");
break;
case FDT_144M:
device_set_desc(dev, "1440-KB 3.5\" drive");
break;
case FDT_288M:
device_set_desc(dev, "2880-KB 3.5\" drive (in 1440-KB mode)");
break;
case FDT_360K:
device_set_desc(dev, "360-KB 5.25\" drive");
break;
case FDT_720K:
device_set_desc(dev, "720-KB 3.5\" drive");
break;
default:
return (ENXIO);
}
fd->track = FD_NO_TRACK;
fd->fdc = fdc;
fd->fdsu = fdsu;
fd->options = 0;
callout_init(&fd->toffhandle, 1);
callout_init(&fd->tohandle, 1);
/* initialize densities for subdevices */
fdsettype(fd, fd_native_types[fd->type]);
return (0);
}
/*
* We have to do this in a geom event because GEOM is not running
* when fd_attach() is.
* XXX: move fd_attach after geom like ata/scsi disks
*/
static void
fd_attach2(void *arg, int flag)
{
struct fd_data *fd;
fd = arg;
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->fd_geom = g_new_geomf(&g_fd_class,
"fd%d", device_get_unit(fd->dev));
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
fd->fd_provider = g_new_providerf(fd->fd_geom, fd->fd_geom->name);
fd->fd_geom->softc = fd;
g_error_provider(fd->fd_provider, 0);
}
static int
fd_attach(device_t dev)
{
struct fd_data *fd;
fd = device_get_softc(dev);
g_post_event(fd_attach2, fd, M_WAITOK, NULL);
fd->flags |= FD_EMPTY;
bioq_init(&fd->fd_bq);
return (0);
}
static int
fd_detach(device_t dev)
{
struct fd_data *fd;
fd = device_get_softc(dev);
g_topology_lock();
g_wither_geom(fd->fd_geom, ENXIO);
g_topology_unlock();
while (device_get_state(dev) == DS_BUSY)
tsleep(fd, PZERO, "fdd", hz/10);
Rewrite of the floppy driver to make it MPsafe & GEOM friendly: Centralize the fdctl_wr() function by adding the offset in the resource to the softc structure. Bugfix: Read the drive-change signal from the correct place: same place as the ctl register. Remove the cdevsw{} related code and implement a GEOM class. Ditch the state-engine and park a thread on each controller to service the queue. Make the interrupt FAST & MPSAFE since it is just a simple wakeup(9) call. Rely on a per controller mutex to protect the bioqueues. Grab GEOMs topology lock when we have to and Giant when ISADMA needs it. Since all access to the hardware is isolated in the per controller thread, the rest of the driver is lock & Giant free. Create a per-drive queue where requests are parked while the motor spins up. When the motor is running the requests are purged to the per controller queue. This allows requests to other drives to be serviced during spin-up. Only setup the motor-off timeout when we finish the last request on the queue and cancel it when a new request arrives. This fixes the bug in the old code where the motor turned off while we were still retrying a request. Make the "drive-change" work reliably. Probe the drive on first opens. Probe with a recal and a seek to cyl=1 to reset the drive change line and check again to see if we have a media. When we see the media disappear we destroy the geom provider, create a new one, and flag that autodetection should happen next time we see a media (unless a specific format is configured). Add sysctl tunables for a lot of drive related parameters. If you spend a lot of time waiting for floppies you can grab the i82078 pdf from Intels web-page and try tuning these. Add sysctl debug.fdc.debugflags which will enable various kinds of debugging printfs. Add central definitions of our well known floppy formats. Simplify datastructures for autoselection of format and call the code at the right times. Bugfix: Remove at least one piece of code which would have made 2.88M floppies not work. Use implied seeks on enhanced controllers. Use multisector transfers on all controllers. Increase ISADMA bounce buffers accordingly. Fall back to single sector when retrying. Reset retry count on every successful transaction. Sort functions in a more sensible order and generally tidy up a fair bit here and there. Assorted related fixes and adjustments in userland utilities. WORKAROUNDS: Do allow r/w opens of r/o media but refuse actual write operations. This is necessary until the p4::phk_bufwork branch gets integrated (This problem relates to remounting not reopening devices, see sys/*/*/${fs}_vfsops.c for details). Keep PC98's private copy of the old floppy driver compiling and presumably working (see below). TODO (planned) Move probing of drives until after interrupts/timeouts work (like for ATA/SCSI drives). TODO (unplanned) This driver should be made to work on PC98 as well. Test on YE-DATA PCMCIA floppy drive. Fix 2.88M media. This is a MT5 candidate (depends on the bioq_takefirst() addition).
2004-08-20 15:14:25 +00:00
callout_drain(&fd->toffhandle);
return (0);
}
static device_method_t fd_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, fd_probe),
DEVMETHOD(device_attach, fd_attach),
DEVMETHOD(device_detach, fd_detach),
DEVMETHOD(device_shutdown, bus_generic_shutdown),
DEVMETHOD(device_suspend, bus_generic_suspend), /* XXX */
DEVMETHOD(device_resume, bus_generic_resume), /* XXX */
{ 0, 0 }
};
static driver_t fd_driver = {
"fd",
fd_methods,
sizeof(struct fd_data)
};
static int
fdc_modevent(module_t mod, int type, void *data)
{
g_modevent(NULL, type, &g_fd_class);
return (0);
}
DRIVER_MODULE(fd, fdc, fd_driver, fd_devclass, fdc_modevent, 0);