]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
24 years agoChange "atapi" (the old one) from "present but broken" to "not present".
Peter Wemm [Sun, 12 Dec 1999 20:35:47 +0000 (20:35 +0000)]
Change "atapi" (the old one) from "present but broken" to "not present".

24 years agoRemove a whole bunch of "CFLAGS+= -DFSNAME" cruft. It hasn't been
Peter Wemm [Sun, 12 Dec 1999 20:34:11 +0000 (20:34 +0000)]
Remove a whole bunch of "CFLAGS+= -DFSNAME" cruft.  It hasn't been
needed for ages, but keeps getting cut/pasted into new Makefiles.
(Once apon a time it was used to activate mount arguments in
 <sys/mount.h>, but that was killed with extreme prejudice long ago)

24 years agosd->da, wd->ad
Chris Piazza [Sun, 12 Dec 1999 19:12:45 +0000 (19:12 +0000)]
sd->da, wd->ad

24 years agoLook for disk devices rather than bdevs.
Poul-Henning Kamp [Sun, 12 Dec 1999 17:43:12 +0000 (17:43 +0000)]
Look for disk devices rather than bdevs.

Approved by: grog

24 years agoBring these more into line with other modules that have .h files generated
Peter Wemm [Sun, 12 Dec 1999 16:47:33 +0000 (16:47 +0000)]
Bring these more into line with other modules that have .h files generated
on the fly.

24 years agoUse the -c and -h args of vnode_if.sh to generate only the .h file that
Peter Wemm [Sun, 12 Dec 1999 16:45:58 +0000 (16:45 +0000)]
Use the -c and -h args of vnode_if.sh to generate only the .h file that
we use.  The .c half is statically compiled into the kernel.  It's kinda
silly to generate a .h file on the fly that has inlines to call the
.c stuff when the .c code is fixed.

Also, zap the special treatment for VFS_KLD modules.  This treatment
applies to lots of things, not just VFS's.

24 years ago Make the cvs-supfile cover the entire repository - add the doc hierarchy.
Matthew Dillon [Sun, 12 Dec 1999 16:44:26 +0000 (16:44 +0000)]
Make the cvs-supfile cover the entire repository - add the doc hierarchy.

Approved by: jkh

24 years agoUse a seperate -c and -h mode. The vnode_if.c file is compiled only into
Peter Wemm [Sun, 12 Dec 1999 16:43:05 +0000 (16:43 +0000)]
Use a seperate -c and -h mode.  The vnode_if.c file is compiled only into
the kernel while the vnode_if.h header is a bunch of inlines to call the
code that is in the kernel. Generating the .h file on the fly is kinda
bogus because it has to match the one compiled into the kernel.

IMHO we should have kern/vnode_if.c and sys/vnode_if.h committed in the
tree but that's another battle.

24 years agoPut on asbestos suit and put a splcam() around the 'Mounting root from..'
Peter Wemm [Sun, 12 Dec 1999 16:34:43 +0000 (16:34 +0000)]
Put on asbestos suit and put a splcam() around the 'Mounting root from..'
message to stop it splitting.  Every single scsi machine I've seen seems
to reliably collide with this and it's rather annoying.

24 years agoThe sysctl mod_xx hack is no longer required now that we have totally
Peter Wemm [Sun, 12 Dec 1999 16:30:34 +0000 (16:30 +0000)]
The sysctl mod_xx hack is no longer required now that we have totally
dynamic sysctl registration.

24 years agoRemove some horrible #if 0'd code that has no hope of working now. It
Peter Wemm [Sun, 12 Dec 1999 14:47:23 +0000 (14:47 +0000)]
Remove some horrible #if 0'd code that has no hope of working now.  It
used to edit the old-style isa_devtab config tables to insert a mapping
for a pci device into the isa tables so that the wdc driver could probe
it later.  This has been #if 0'd since April.

24 years agoTurn on warnings for the wd* driver with a loud pointer to ata*. This
Peter Wemm [Sun, 12 Dec 1999 14:31:40 +0000 (14:31 +0000)]
Turn on warnings for the wd* driver with a loud pointer to ata*.  This
doesn't break builds, but is difficult to miss..

24 years agoReplace the svr4_sys_getdents64() routine with a port of linux_getdents() --
Mark Newton [Sun, 12 Dec 1999 11:25:33 +0000 (11:25 +0000)]
Replace the svr4_sys_getdents64() routine with a port of linux_getdents() --
differences between the VFS interface between FreeBSD and NetBSD make
it easier to pick up the Linux one than to continue development with the
NetBSD port.

This patch fixes a bug which caused duplicate filenames to be seen by
callers to svr4_sys_getdents64(), leading to malformed directory listings
from Solaris client programs.

Obtained from: The Linuxulator, with a pointer from marcel

24 years agoAvoid excessive redundancy in svr4_sys_getmsg() and svr4_sys_putmsg():
Mark Newton [Sun, 12 Dec 1999 10:28:30 +0000 (10:28 +0000)]
Avoid excessive redundancy in svr4_sys_getmsg() and svr4_sys_putmsg():
Only look up the provided descriptor in fd_ofiles[] once.

Submitted by: Ville-Pertti Keinone <will@iki.fi>

24 years agofd_revoke() shouldn't panic if the descriptor provided is not a file or
Mark Newton [Sun, 12 Dec 1999 10:27:04 +0000 (10:27 +0000)]
fd_revoke() shouldn't panic if the descriptor provided is not a file or
socket.  Return EINVAL instead.

Submitted by: Ville-Pertti Keinone <will@iki.fi>

24 years ago Add blurb on massive improvements to NFS
Matthew Dillon [Sun, 12 Dec 1999 09:53:11 +0000 (09:53 +0000)]
Add blurb on massive improvements to NFS

Reviewed by: jkh

24 years agoAdd a 'warning' option for the files* files. This is intended to enable
Peter Wemm [Sun, 12 Dec 1999 09:38:48 +0000 (09:38 +0000)]
Add a 'warning' option for the files* files.  This is intended to enable
giving a dire warning about certain drivers going away in the future.

24 years ago-Wall fixes.
Bill Fumerola [Sun, 12 Dec 1999 07:25:14 +0000 (07:25 +0000)]
-Wall fixes.

24 years ago Fix a number of server-side issues related to aborting badly formed
Matthew Dillon [Sun, 12 Dec 1999 07:06:39 +0000 (07:06 +0000)]
Fix a number of server-side issues related to aborting badly formed
    NFS packets, mainly initializing structure pointers to NULL which
    are conditionally freed prior to return.

PR: kern/15249
Submitted by: Ian Dowse <iedowse@maths.tcd.ie>

24 years ago-Wall fixes.
Bill Fumerola [Sun, 12 Dec 1999 06:40:28 +0000 (06:40 +0000)]
-Wall fixes.

24 years ago-Wall fixes.
Bill Fumerola [Sun, 12 Dec 1999 06:30:46 +0000 (06:30 +0000)]
-Wall fixes.

24 years ago-Wall fixes.
Bill Fumerola [Sun, 12 Dec 1999 06:17:25 +0000 (06:17 +0000)]
-Wall fixes.

24 years ago Synopsis of problem being fixed: Dan Nelson originally reported that
Matthew Dillon [Sun, 12 Dec 1999 06:09:57 +0000 (06:09 +0000)]
Synopsis of problem being fixed:  Dan Nelson originally reported that
    blocks of zeros could wind up in a file written to over NFS by a client.
    The problem only occurs a few times per several gigabytes of data.   This
    problem turned out to be bug #3 below.

    bug #1:

        B_CLUSTEROK must be cleared when an NFS buffer is reverted from
        stage 2 (ready for commit rpc) to stage 1 (ready for write).
        Reversions can occur when a dirty NFS buffer is redirtied with new
        data.

        Otherwise the VFS/BIO system may end up thinking that a stage 1
        NFS buffer is clusterable.  Stage 1 NFS buffers are not clusterable.

    bug #2:

        B_CLUSTEROK was inappropriately set for a 'short' NFS buffer (short
        buffers only occur near the EOF of the file).  Change to only set
        when the buffer is a full biosize (usually 8K).  This bug has no
        effect but should be fixed in -current anyway.  It need not be
        backported.

    bug #3:

        B_NEEDCOMMIT was inappropriately set in nfs_flush() (which is
typically only called by the update daemon).  nfs_flush()
        does a multi-pass loop but due to the lack of vnode locking it
        is possible for new buffers to be added to the dirtyblkhd list
        while a flush operation is going on.  This may result in nfs_flush()
        setting B_NEEDCOMMIT on a buffer which has *NOT* yet gone through its
        stage 1 write, causing only the commit rpc to be made and thus
        causing the contents of the buffer to be thrown away (never sent to
        the server).

    The patch also contains some cleanup, which only applies to the commit
    into -current.

Reviewed by: dg, julian
Originally Reported by: Dan Nelson <dnelson@emsphone.com>

24 years agoBump local version number to 1.3.4.
Boris Popov [Sun, 12 Dec 1999 05:53:02 +0000 (05:53 +0000)]
Bump local version number to 1.3.4.

24 years agoThis is Bosko Milekic's mbuf allocation waiting code. Basically, this
Brian Feldman [Sun, 12 Dec 1999 05:52:51 +0000 (05:52 +0000)]
This is Bosko Milekic's mbuf allocation waiting code.  Basically, this
means that running out of mbuf space isn't a panic anymore, and code
which runs out of network memory will sleep to wait for it.

Submitted by: Bosko Milekic <bmilekic@dsuper.net>
Reviewed by: green, wollman

24 years agoUpdate to version 1.3.4 of ncplib. Cleanup header files.
Boris Popov [Sun, 12 Dec 1999 05:50:07 +0000 (05:50 +0000)]
Update to version 1.3.4 of ncplib. Cleanup header files.

24 years agoI shouldn't have incremented PART_OFF; it was wrong and broke label
Jordan K. Hubbard [Sun, 12 Dec 1999 04:58:02 +0000 (04:58 +0000)]
I shouldn't have incremented PART_OFF; it was wrong and broke label
display to boot.  Also fix some various warning fluff while I'm in
here cleaning up.

24 years agoWhen booting verbose, indicate if we are using manual termination
Justin T. Gibbs [Sun, 12 Dec 1999 04:54:14 +0000 (04:54 +0000)]
When booting verbose, indicate if we are using manual termination
settings for U2 cards.

Don't assume that all aic7859 cards are 2930CUs.

24 years agoMove the wakeup_one() prototype from proc.h to systm.h. It now hangs
Brian Feldman [Sun, 12 Dec 1999 04:21:27 +0000 (04:21 +0000)]
Move the wakeup_one() prototype from proc.h to systm.h.  It now hangs
out with it's sibling, wakeup().

24 years agoBuild a.out tools if MACHINE_ARCH == i386 and MACHINE == pc98.
Yoshihiro Takahashi [Sun, 12 Dec 1999 03:54:58 +0000 (03:54 +0000)]
Build a.out tools if MACHINE_ARCH == i386 and MACHINE == pc98.
The boot2 for pc98 is still a.out program.

I made the original patch, and many problems were fixed by Marcel Moolenaar.

24 years ago Remove accidental pollution unrelated to previous commit. The issue
Matthew Dillon [Sun, 12 Dec 1999 03:28:14 +0000 (03:28 +0000)]
Remove accidental pollution unrelated to previous commit.  The issue
    here is real but has not yet been discussed with Eivind.

24 years ago-Wall fixes.
Bill Fumerola [Sun, 12 Dec 1999 03:22:37 +0000 (03:22 +0000)]
-Wall fixes.

24 years ago Add MAP_NOSYNC feature to mmap(), and MADV_NOSYNC and MADV_AUTOSYNC to
Matthew Dillon [Sun, 12 Dec 1999 03:19:33 +0000 (03:19 +0000)]
Add MAP_NOSYNC feature to mmap(), and MADV_NOSYNC and MADV_AUTOSYNC to
    madvise().

    This feature prevents the update daemon from gratuitously flushing
    dirty pages associated with a mapped file-backed region of memory.  The
    system pager will still page the memory as necessary and the VM system
    will still be fully coherent with the filesystem.  Modifications made
    by other means to the same area of memory, for example by write(), are
    unaffected.  The feature works on a page-granularity basis.

    MAP_NOSYNC allows one to use mmap() to share memory between processes
    without incuring any significant filesystem overhead, putting it in
    the same performance category as SysV Shared memory and anonymous memory.

Reviewed by: julian, alc, dg

24 years ago-Wall fixes.
Bill Fumerola [Sun, 12 Dec 1999 03:04:17 +0000 (03:04 +0000)]
-Wall fixes.

24 years agoFinally give mk_cmds(1) a man page.
Chris Costello [Sun, 12 Dec 1999 02:43:07 +0000 (02:43 +0000)]
Finally give mk_cmds(1) a man page.

PR: docs/4691
Submitted by: Jeroen Ruigrok/Asmodai <asmodai@wxs.nl>

24 years agomake sb dependant on sbc
Cameron Grant [Sun, 12 Dec 1999 02:30:19 +0000 (02:30 +0000)]
make sb dependant on sbc
add support for non-pnp cards to sbc
move card identification to sbc
channel-swapping code is in sb now instead of dsp
vibra16x support is still broken, but will be fixed soon

note: sbc is now compulsory for sb cards

for pnp cards use:
device sbc0

for non-pnp cards eg:
device sbc0 at isa? port 0x240 irq 5 drq 3 flags 0x15
(hints as oldpcm)

both in addition to:
device pcm0

Reviewed by: tanimura,dfr
Said he liked it: peter

24 years ago-Wall cleanup.
Bill Fumerola [Sun, 12 Dec 1999 02:29:54 +0000 (02:29 +0000)]
-Wall cleanup.

24 years agomove channel-swapping support to the hardware driver since it knows the card
Cameron Grant [Sun, 12 Dec 1999 02:18:58 +0000 (02:18 +0000)]
move channel-swapping support to the hardware driver since it knows the card
state best

24 years agoincrease buffer size, reduce number of channels allocated since we only use
Cameron Grant [Sun, 12 Dec 1999 02:16:14 +0000 (02:16 +0000)]
increase buffer size, reduce number of channels allocated since we only use
1 at the moment

24 years agoSuport multiple ``ifconfig_*?="DHCP"'' configurations.
David E. O'Brien [Sun, 12 Dec 1999 01:58:30 +0000 (01:58 +0000)]
Suport multiple ``ifconfig_*?="DHCP"'' configurations.

Currently we have a problem in that `dhclient' bails when configuring the
second interface as port 68 is already in use (by the `dhclient' started
for the first interface).

PR: 14810
Submitted by: n_hibma

24 years agoDo not add routes for localhost to the `dhclient' controlled interface.
David E. O'Brien [Sun, 12 Dec 1999 01:04:36 +0000 (01:04 +0000)]
Do not add routes for localhost to the `dhclient' controlled interface.
Doing so is bogus if the loop-back interface was not configured.

Typically ``network_interfaces="auto"'' will return the list of
interfaces such that "lo0" is not first.  Thus there are times when
`dhclient' configs an interface before "lo0" is configured.

Under BSD4.4, there's no need to add the above route, as it will be
automatically generated by the kernel.

PR: conf/14098, misc/15183
Submitted by: luoqi (true identification of the problem)

24 years agoQuirk all Pioneer changers as changers up front, instead of waiting for the
Kenneth D. Merry [Sat, 11 Dec 1999 23:00:44 +0000 (23:00 +0000)]
Quirk all Pioneer changers as changers up front, instead of waiting for the
second LUN to show up.

mjacob's change (which is correct) in rev 1.21 of cam_periph.c to elminiate
infinite retries of the SCSI busy status bit seems to have broken probing
of Pioneer changers that aren't already quirked.

The right way to fix this is probably to change things around so we can
guarantee 100% sequential probing of LUN-based changers even if they aren't
quirked.  This should fix things for now, though.

24 years agoTrim to 24 lines. Yeah, I know, looks ugly.
Daniel C. Sobral [Sat, 11 Dec 1999 22:31:50 +0000 (22:31 +0000)]
Trim to 24 lines. Yeah, I know, looks ugly.

24 years agoAdd reference to hier(7). This gets motd just over 25 lines. I'm
Daniel C. Sobral [Sat, 11 Dec 1999 21:43:38 +0000 (21:43 +0000)]
Add reference to hier(7). This gets motd just over 25 lines. I'm
don't like it, but I think this can greatly ease the learning curve
of Linux-immigrants.

Approved by: jkh

24 years agoreactivate cvs subdir build
Peter Wemm [Sat, 11 Dec 1999 17:32:07 +0000 (17:32 +0000)]
reactivate cvs subdir build

24 years agoDon't add MACHINE_ARCH to MAKEOBJDIRPREFIX when not cross-building.
Marcel Moolenaar [Sat, 11 Dec 1999 17:05:36 +0000 (17:05 +0000)]
Don't add MACHINE_ARCH to MAKEOBJDIRPREFIX when not cross-building.
This should fix make release.

Reported by: jhay, phk

24 years agoLock reporting and assertion changes.
Eivind Eklund [Sat, 11 Dec 1999 16:13:02 +0000 (16:13 +0000)]
Lock reporting and assertion changes.
* lockstatus() and VOP_ISLOCKED() gets a new process argument and a new
  return value: LK_EXCLOTHER, when the lock is held exclusively by another
  process.
* The ASSERT_VOP_(UN)LOCKED family is extended to use what this gives them
* Extend the vnode_if.src format to allow more exact specification than
  locked/unlocked.

This commit should not do any semantic changes unless you are using
DEBUG_VFS_LOCKS.

Discussed with: grog, mch, peter, phk
Reviewed by: peter

24 years agoUpdate files generated by configure.
Peter Wemm [Sat, 11 Dec 1999 15:12:37 +0000 (15:12 +0000)]
Update files generated by configure.

24 years agoSomething went wrong with an earlier commit and these files ended up
John Polstra [Sat, 11 Dec 1999 15:12:31 +0000 (15:12 +0000)]
Something went wrong with an earlier commit and these files ended up
empty.  Fix that with help from grog.

24 years agoUnmangle cvs's MD5* calls.
Peter Wemm [Sat, 11 Dec 1999 15:10:02 +0000 (15:10 +0000)]
Unmangle cvs's MD5* calls.

24 years agoTake a shot at using mkstemp() since we have a __warn_references() on
Peter Wemm [Sat, 11 Dec 1999 14:58:02 +0000 (14:58 +0000)]
Take a shot at using mkstemp() since we have a __warn_references() on
the other temporary file creation functions..

24 years agoWhile comparing this with OpenBSD (ie: trying to figure out what mkstemps()
Peter Wemm [Sat, 11 Dec 1999 14:48:24 +0000 (14:48 +0000)]
While comparing this with OpenBSD (ie: trying to figure out what mkstemps()
is good for... :-)), I discovered that part of the change when mkstemps()
was brought in was missed - it was missing the termination case to make
sure it doesn't walk into the suffix.  This isn't the same code OpenBSD
has, I think this is a little better as we terminate the loop in a better
spot.

24 years agoCheck in correct versions of files. Somehow null files got committed
Greg Lehey [Sat, 11 Dec 1999 14:27:12 +0000 (14:27 +0000)]
Check in correct versions of files.  Somehow null files got committed
last time.

There's a possibility that there are still things wrong with the
commit.  I wasn't able to build the version I committed, and I won't
be for another few days.  If any committer finds further omissions,
the original source archive is on freefall:~grog/picobsd.tar.gz.  Feel
free to correct.

Reported-by: "Louis A. Mamakos" <louie@TransSys.COM>
24 years agoInstall include files with mode 444.
Marcel Moolenaar [Sat, 11 Dec 1999 13:38:04 +0000 (13:38 +0000)]
Install include files with mode 444.

24 years agoMerge error. rcs_lockfile is freed after unlock.
Peter Wemm [Sat, 11 Dec 1999 13:19:17 +0000 (13:19 +0000)]
Merge error. rcs_lockfile is freed after unlock.

24 years agoUpdate for 1.10.7 update.
Peter Wemm [Sat, 11 Dec 1999 13:11:47 +0000 (13:11 +0000)]
Update for 1.10.7 update.

24 years agoMerge cyclic changes from 1.10.7 into our mainline. I did this seperately
Peter Wemm [Sat, 11 Dec 1999 13:00:18 +0000 (13:00 +0000)]
Merge cyclic changes from 1.10.7 into our mainline.  I did this seperately
as cvs update -j had kittens over the whole thing and I ended up merging
it by hand.

24 years agoMerge cyclic changes for 1.10.7 only our mainline.
Peter Wemm [Sat, 11 Dec 1999 12:50:10 +0000 (12:50 +0000)]
Merge cyclic changes for 1.10.7 only our mainline.

24 years agoRevert to vendor version. Sigh, this left the vendor branch because
Peter Wemm [Sat, 11 Dec 1999 12:42:49 +0000 (12:42 +0000)]
Revert to vendor version.  Sigh, this left the vendor branch because
of a fix for a y2k non-problem. :-(

24 years agoThis commit was generated by cvs2svn to compensate for changes in r54427,
Peter Wemm [Sat, 11 Dec 1999 12:24:21 +0000 (12:24 +0000)]
This commit was generated by cvs2svn to compensate for changes in r54427,
which included commits to RCS files with non-trunk default branches.

24 years agoImport cvs-1.10.7. There are a number of nasty bugs that have been fixed.
Peter Wemm [Sat, 11 Dec 1999 12:24:21 +0000 (12:24 +0000)]
Import cvs-1.10.7.  There are a number of nasty bugs that have been fixed.

Obtained from:  cyclic.com

24 years agoTurn off cvs for a moment pending import of a newer version..
Peter Wemm [Sat, 11 Dec 1999 12:12:54 +0000 (12:12 +0000)]
Turn off cvs for a moment pending import of a newer version..

24 years agoReclaim UPAGES_HOLE (8k) that was chopped out of process address space.
Peter Wemm [Sat, 11 Dec 1999 10:54:06 +0000 (10:54 +0000)]
Reclaim UPAGES_HOLE (8k) that was chopped out of process address space.
The UPAGES have not been there since Jan '96, but the hole was preserved
for BSD/OS binary compatability.  This has been fixed other ways (%ebx
now has a pointer to PS_STRINGS), and the stack is nowhere near where
it used to be so this hack isn't required anymore.

24 years agoDon't simulate a pseudo address-space beyond VM_MAXUSER_ADDRESS that
Peter Wemm [Sat, 11 Dec 1999 10:21:34 +0000 (10:21 +0000)]
Don't simulate a pseudo address-space beyond VM_MAXUSER_ADDRESS that
maps onto the upages.  We used to use this extensively, particularly
for ps and gdb.  Both of these have been "fixed".  ps gets the p_stats
via eproc along with all the other stats, and gdb uses the regs, fpregs
etc files.

Once apon a time the UPAGES were mapped here, but that changed back
in January '96.  This essentially kills my revisions 1.16 and 1.17.
The 2-page "hole" above the stack can be reclaimed now.

24 years agoAllow kernel accesses to a small region of the user stack which is used
Doug Rabson [Sat, 11 Dec 1999 09:58:06 +0000 (09:58 +0000)]
Allow kernel accesses to a small region of the user stack which is used
by the Linux emulator (and other emulators) for syscall argument
translation. The x86 port currently seems to allow unrestricted kernel
accesses to user memory.

Reviewed by: alc, gallatin

24 years agobktr driver 2.06 changelog
Roger Hardiman [Sat, 11 Dec 1999 07:33:36 +0000 (07:33 +0000)]
bktr driver 2.06 changelog

24 years agoAccording to RFC 793, a reset should be honored if the sequence number
Jonathan Lemon [Sat, 11 Dec 1999 04:05:52 +0000 (04:05 +0000)]
According to RFC 793, a reset should be honored if the sequence number
is within the receive window.  Follow this behavior, instead of only
allowing resets at last_ack_sent.

Pointed out by: jayanth@yahoo-inc.com

24 years agoRetire error(1). This is no longer useful as it applies only to the 4.xBSD
Kris Kennaway [Sat, 11 Dec 1999 03:33:33 +0000 (03:33 +0000)]
Retire error(1). This is no longer useful as it applies only to the 4.xBSD
toolchain.

Reviewed by: current@freebsd.org
Axe kindly supplied by: Institute of Danish Surgeons

24 years agoMajor update to the Mylex DAC960 driver adding new hardware support
Mike Smith [Sat, 11 Dec 1999 00:00:13 +0000 (00:00 +0000)]
Major update to the Mylex DAC960 driver adding new hardware support
and fixing some major bugs.

 - Add support for the v5 firmware interface, used by the DAC1164P
   (tested) and AcceleRAID 352 (untested but should work).  We now cover
   all of the Mylex family's protocols except for v2 (used by EISA and
   Alpha-compatible cards).

 - Fix an accounting bug which resulted in endless 'poll still busy'
   messages.  In situations of high controller load the count of poll
   commands could be incremented without actually successfully launching
   a command.  This totally removes the accounting for status poll
   commnads; it was its own worst enemy.

 - Add some simple reentry prevention locks to processing of the waiting
   and completed command queues to prevent races which could result in
   I/O being done or completed twice (both are fatal).  This highlights
   a need for simple locking primitives in both the UP and SMP kernels.

 - Streamline the handling of command completion to reduce the amount of
   redundant work being done.  Remove the code which tests for commands
   that have gone missing in action; nobody has ever seen one of these
   and it wouldn't have worked properly anyhow.

 - Handle disconnection of drives from the controller in the detach,
   not shutdown method.  This avoids problems flushing the cache in
   a panic when a drive is mounted.

 - Don't call bus_generic_detach when disconnecting drives; it doesn't
   actually do anything useful.

 - Increment the log message index regardless of whether we actually
   retrieved one or not.  If we run into a message that we can't fetch,
   we don't want to spin endlessly complaining about the fact.

 - Don't assume that interrupts will work when we're flushing the
   controller.  We may think they are enabled, but in eg. a panic
   situation the controller may not be able to deliver an interrupt.

24 years agoAdd 'custom' directory with significantly restructured build (now
Greg Lehey [Fri, 10 Dec 1999 21:52:18 +0000 (21:52 +0000)]
Add 'custom' directory with significantly restructured build (now
using make instead of custom scripts) and two floppies instead of
one.  The resultant floppy can do everything that the individual
floppies (dial, net, install, isp, router) could do, modulo some bit
rot that has occurred since PicoBSD last compiled.  It also includes
all the programs on the fixit floppy, which could thus also die.

/bin currently contains the following files:

-sh             dump            ln              ns              sps
[               ed              login           ping            stty
badsect         ex              ls              ps              swapon
cat             expr            mkdir           pwd             sync
chgrp           fdisk           mknod           pwd_mkdb        sysctl
chmod           find            more            rdump         syslogd
chown           fsck            mount           reboot          tar
chroot          ftp             mount_cd9660    restore         telnet
clri            getty           mount_msdos     rlogin                telnetd
cp              grep            mount_nfs       rm              test
date            gunzip          mount_std       rmdir         traceroute
dd              gzip            msg             route           umount
dev_mkdb        hostname        mt              routed          vi
df              ifconfig        mv              rrestore        view
dhclient        inetd           natd            rsh             vm
dhclient-script init            netstat         sed             w
disklabel       kget            newfs           sh              zcat
dmesg           kill            nfs             sleep

Structure is in place for using the same build for the other
directories, but I'm no longer sure we need this.  The current first
floppy will run fine by itself, but the size of a compressed kernel
has increased by nearly 50% since 3.2, and there's not much space for
anything useful on the remainder of the floppy.  The current method
creates a larger mfs and can read as many floppies as the user can
stand.  The footprint appears to be round 14 MB.

24 years agoAdd 'custom' directory with significantly restructured build (now
Greg Lehey [Fri, 10 Dec 1999 21:43:10 +0000 (21:43 +0000)]
Add 'custom' directory with significantly restructured build (now
using make instead of custom scripts) and two floppies instead of
one.  The resultant floppy can do everything that the individual
floppies (dial, net, install, isp, router) could do, modulo some bit
rot that has occurred since PicoBSD last compiled.  It also includes
all the programs on the fixit floppy, which could thus also die.

/bin currently contains the following files:

-sh             dump            ln              ns              sps
[               ed              login           ping            stty
badsect         ex              ls              ps              swapon
cat             expr            mkdir           pwd             sync
chgrp           fdisk           mknod           pwd_mkdb        sysctl
chmod           find            more            rdump syslogd
chown           fsck            mount           reboot          tar
chroot          ftp             mount_cd9660    restore         telnet
clri            getty           mount_msdos     rlogin telnetd
cp              grep            mount_nfs       rm              test
date            gunzip          mount_std       rmdir traceroute
dd              gzip            msg             route           umount
dev_mkdb        hostname        mt              routed          vi
df              ifconfig        mv              rrestore        view
dhclient        inetd           natd            rsh             vm
dhclient-script init            netstat         sed             w
disklabel       kget            newfs           sh              zcat
dmesg           kill            nfs             sleep

Structure is in place for using the same build for the other
directories, but I'm no longer sure we need this.  The current first
floppy will run fine by itself, but the size of a compressed kernel
has increased by nearly 50% since 3.2, and there's not much space for
anything useful on the remainder of the floppy.  The current method
creates a larger mfs and can read as many floppies as the user can
stand.  The footprint appears to be round 14 MB.

Work-sponsored-by: Sitara Networks Inc.
24 years ago1. Don't overwrite scorefiles if they already exist.
Bill Fumerola [Fri, 10 Dec 1999 21:13:42 +0000 (21:13 +0000)]
1. Don't overwrite scorefiles if they already exist.

PR: bin/13068
Submitted by: Dominic Mitchell <Dom.Mitchell@palmerharvey.co.uk>

2. Sprinkle $FreeBSD$ around.

24 years agoFix a '&&' that should have been a '&'.
Archie Cobbs [Fri, 10 Dec 1999 20:04:53 +0000 (20:04 +0000)]
Fix a '&&' that should have been a '&'.

Submitted by: Erik Salander <erik@whistle.com>

24 years agoAdd reference to RFC 1702, after learning how to have "T. Li"
Archie Cobbs [Fri, 10 Dec 1999 19:29:43 +0000 (19:29 +0000)]
Add reference to RFC 1702, after learning how to have "T. Li"
as an author without nroff blowing up.

Problem solved by: fenner

24 years agoMake the usb and ide/ata device identification a little saner. Rather than
Peter Wemm [Fri, 10 Dec 1999 17:44:22 +0000 (17:44 +0000)]
Make the usb and ide/ata device identification a little saner.  Rather than
attaching to the device via chip*, use the newbus nomatch method to report
the device.  This leaves them unattached so that a driver can be easily
loaded to grab them later.

24 years agoZap c_index() and c_rindex(). Bruce prefers these to implicitly convert
Peter Wemm [Fri, 10 Dec 1999 17:38:41 +0000 (17:38 +0000)]
Zap c_index() and c_rindex().  Bruce prefers these to implicitly convert
a const into a non-const as they do in libc.  I feel that defeating the
type checking like that quite evil, but that's the way it is.

24 years agoAdd some gross ad-hock hacks to increase stability of if_detach:
Warner Losh [Fri, 10 Dec 1999 16:31:25 +0000 (16:31 +0000)]
Add some gross ad-hock hacks to increase stability of if_detach:
o be more careful about clearing addresses (this isn't a kludge)
o For AF_INET interfaces, call SIOCDIFFADDR to remove last(?) bit
  of cruft.

Special cases for AF_INET shouldn't be here, but I didn't see a good
generic way of doing this.  If I missed something, please let me know.

This gross hack makes pccard ejection stable for ethernet cards.

Submitted by: Atushi Onoe-san

24 years ago-Wall cleanup.
Bill Fumerola [Fri, 10 Dec 1999 16:21:50 +0000 (16:21 +0000)]
-Wall cleanup.

24 years agoo Don't make games/fortune/strfile if games does not exist or NOGAMES
Marcel Moolenaar [Fri, 10 Dec 1999 16:13:41 +0000 (16:13 +0000)]
o  Don't make games/fortune/strfile if games does not exist or NOGAMES
   has been defined.
o  Make libraries before making depend.

24 years agoMerge from sys/isa/sio.c rev 1.279.
KATO Takenori [Fri, 10 Dec 1999 14:03:47 +0000 (14:03 +0000)]
Merge from sys/isa/sio.c rev 1.279.

24 years agoRemove ze and zp drivers.
KATO Takenori [Fri, 10 Dec 1999 14:02:48 +0000 (14:02 +0000)]
Remove ze and zp drivers.

24 years agoMerge from sys/dev/syscons/syscons.c rev 1.329.
KATO Takenori [Fri, 10 Dec 1999 13:55:47 +0000 (13:55 +0000)]
Merge from sys/dev/syscons/syscons.c rev 1.329.

24 years agofix the same typo as in uniq
Andrey A. Chernov [Fri, 10 Dec 1999 13:27:02 +0000 (13:27 +0000)]
fix the same typo as in uniq

24 years agofix fatal typo
Andrey A. Chernov [Fri, 10 Dec 1999 13:06:53 +0000 (13:06 +0000)]
fix fatal typo

24 years agoRemove unused includes.
Marcel Moolenaar [Fri, 10 Dec 1999 12:36:20 +0000 (12:36 +0000)]
Remove unused includes.

Found by: phk-scan

24 years agoRemove installworld related settings that I had as a safety-net and
Marcel Moolenaar [Fri, 10 Dec 1999 11:24:06 +0000 (11:24 +0000)]
Remove installworld related settings that I had as a safety-net and
for development. :-/

24 years agoRemove references to ze and zp drivers.
Poul-Henning Kamp [Fri, 10 Dec 1999 10:55:27 +0000 (10:55 +0000)]
Remove references to ze and zp drivers.

24 years agoNull commit:
Sheldon Hearn [Fri, 10 Dec 1999 10:54:59 +0000 (10:54 +0000)]
Null commit:

The commit message for rev 1.7 was incorrect with respect to the %e
and %l specifiers, which _can_ be fixed by correcting excessive
whitespace-gobbling.

24 years agoNull commit:
Sheldon Hearn [Fri, 10 Dec 1999 10:52:54 +0000 (10:52 +0000)]
Null commit:

The commit message for rev 1.16 was incorrect with respect to the %e
and %l specifiers, which _can_ be fixed by correcting excessive
whitespace-gobbling.

24 years agoRemove reference to ze and zp drivers.
Poul-Henning Kamp [Fri, 10 Dec 1999 10:49:51 +0000 (10:49 +0000)]
Remove reference to ze and zp drivers.

24 years agoMore cross-building related changes:
Marcel Moolenaar [Fri, 10 Dec 1999 10:48:20 +0000 (10:48 +0000)]
More cross-building related changes:
o  Build tools before doing anything in or with the object tree.
o  Tools don't use the object tree any more, but have there object
   tree located in the temp. world.
o  Use the proper make env. for cleaning and building the object tree.
o  Don't create kernel include subdirectories in the temp. world. These
   are removed later on and replaced by symlinks.
o  Change the layout of the object tree:

The temp. world now is /usr/obj/${MACHINE_ARCH}${.CURDIR}/${BUILD_ARCH}.
/usr/obj can be set/changed by using MAKEOBJDIRPREFIX, and {.CURDIR}
obviously depends on where the source tree is located. MACHINE_ARCH
is the arch. for which the world is to be build and BUILD_ARCH is the
arch. on which we are building.

The object tree now is /usr/obj/${MACHINE_ARCH}${.CURDIR}.

This allows concurrent cross-builds and allows the object tree to be
shared on different archs., each doing the same cross-build. This of
course assumes that the output on Alpha (for example) is the same as
the output of an Alpha cross-build on i386 (for example).

The use of NOCLEAN is is still dangerous, but should be usable in many
more situations than before. It should now be possible to safely
restart an interrupted build with NOCLEAN without side-effects. Because
the tools don't share the object tree with the normal (cross-build), no
tools have to be rebuild.

24 years agoRemove the if_ze and if_zp drivers.
Poul-Henning Kamp [Fri, 10 Dec 1999 10:45:11 +0000 (10:45 +0000)]
Remove the if_ze and if_zp drivers.

These drivers were cloned from the ed and ep drivers back in 1994
when PCMCIA cards were a very new thing and we had no other support
for such devices.  They treated the PCIC (the chip which controls the
PCCARD slot) as part of their device and generally hacked their way
to success.  They have significantly bit-rotted relative to their
ancestor drivers (ed & ep) and they were a dead-end on the evolution
path to proper PCCARD support in FreeBSD.

They have been terminally broken since August 18 where mdodd forgot
them and nobody seems to have missed them enough to fix them since.

I found no outstanding PRs against these drivers.

24 years agoUse install -C -p instead of cp -p.
Marcel Moolenaar [Fri, 10 Dec 1999 10:14:06 +0000 (10:14 +0000)]
Use install -C -p instead of cp -p.

Pointed out by: bde

24 years agoRemove the B_BAD buffer flag, it is no longer used.
Poul-Henning Kamp [Fri, 10 Dec 1999 09:40:29 +0000 (09:40 +0000)]
Remove the B_BAD buffer flag, it is no longer used.

24 years agoFix memory leak.
Kazutaka YOKOTA [Fri, 10 Dec 1999 09:36:05 +0000 (09:36 +0000)]
Fix memory leak.

PR: kern/15363
Submitted by: Oliver Fromme

24 years agoo Make pccard work at all by including card.h
Warner Losh [Fri, 10 Dec 1999 08:19:19 +0000 (08:19 +0000)]
o Make pccard work at all by including card.h
o fix return type of sio_pccard_detach
o don't free softc in deatch, since that is done by newbus
o disconnect interrupt we used to have.  Add cookie to com so that we can
  tear down the interrupt on unload
o Set gone earlier, but likely doesn't matter

This makes sio pccards work again.  Cards that are active when ejects may
not work (but they might, softc goes away quickly).

These changes are unreviewed by bde.  I'll make any style changes he wants.

24 years agokill debug printf that says detach
Warner Losh [Fri, 10 Dec 1999 07:24:39 +0000 (07:24 +0000)]
kill debug printf that says detach

24 years agoFix pccard ed driver, I think.
Warner Losh [Fri, 10 Dec 1999 07:22:53 +0000 (07:22 +0000)]
Fix pccard ed driver, I think.
o Expose ed_stop and call it early to shutdown the hardware.
o When releasing the interrupt, pass the cookie for the irq, not
  a pointer to the cookie (this is the base problem).
o Release other resources used, just like the ep driver

24 years agoMove the turning on of the interrupts for the card at the bridge from
Warner Losh [Fri, 10 Dec 1999 07:02:41 +0000 (07:02 +0000)]
Move the turning on of the interrupts for the card at the bridge from
the activate method to the setup_intr, and turn it off to
teardown_intr.

This makes the ed driver not enter its interrupt routine during the
probe.  Apparently, an interrupt happens when you disable the
interrupts.  There are other problems with ed still.

24 years agoAdd support new keys: lshifta, rshifta, lctrla, rctrla, lalta, and
Kazutaka YOKOTA [Fri, 10 Dec 1999 04:31:33 +0000 (04:31 +0000)]
Add support new keys: lshifta, rshifta, lctrla, rctrla, lalta, and
ralta.  These keys combine shift/ctrl/alt function and the AltLock
function.  When these keys pressed together with another key, they act
just like the ordinary shift/ctrl/alt keys.  When these keys are
pressed and released alone, Alt lock state is toggled.

PR: kern/12475