]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
21 years agoDOnt allow getting caps from other than master and slave device :)
Søren Schmidt [Mon, 5 May 2003 10:28:37 +0000 (10:28 +0000)]
DOnt allow getting caps from other than master and slave device :)

21 years agoDon't panic for FWXF_START state in fw_xfer_unload().
Hidetoshi Shimokawa [Mon, 5 May 2003 10:14:52 +0000 (10:14 +0000)]
Don't panic for FWXF_START state in fw_xfer_unload().

21 years agoAdd a missing ~ when clearing flags in close.
Søren Schmidt [Mon, 5 May 2003 10:11:17 +0000 (10:11 +0000)]
Add a missing ~ when clearing flags in close.

PR: 35392

21 years agoThings run thru the C preprocessor must use C-style comments.
David E. O'Brien [Mon, 5 May 2003 10:01:10 +0000 (10:01 +0000)]
Things run thru the C preprocessor must use C-style comments.

21 years agoThe vi(1) recovery script cannot send mail regarding interrupted
Mike Makonnen [Mon, 5 May 2003 09:53:32 +0000 (09:53 +0000)]
The vi(1) recovery script cannot send mail regarding interrupted
sessions if the mailer is dynamically linked. This happens because
on some systems the ldconfig script is run after the vi recovery
script.

I thought I had already fixed this with revision 1.5 of rc.d/ldconfig,
but apparently not. So, in addition I'm making ldconfig a requirement
of this script.

Approved by: markm (mentor)(implicit)
PR: bin/51767

21 years agoMake the __FBSDID usage consistent
David E. O'Brien [Mon, 5 May 2003 09:52:25 +0000 (09:52 +0000)]
Make the __FBSDID usage consistent
and protect copyright[] from Gcc 3.3's whining.

21 years agoChange the semantics of sysv shm emulation to take a additional
Martin Blapp [Mon, 5 May 2003 09:22:58 +0000 (09:22 +0000)]
Change the semantics of sysv shm emulation to take a additional
argument to the functions shm{at,ctl}1 and shm_find_segment_by_shmid{x}.
The BSD semantics didn't allow the usage of shared segment after
being marked for removal through IPC_RMID.

The patch involves the following functions:
  - shmat
  - shmctl
  - shm_find_segment_by_shmid
  - shm_find_segment_by_shmidx
  - linux_shmat
  - linux_shmctl

Submitted by: Orlando Bassotto <orlando.bassotto@ieo-research.it>
Reviewed by: marcel

21 years agoBack out rev 1.146 -- it broke the LINT build.
David E. O'Brien [Mon, 5 May 2003 09:15:50 +0000 (09:15 +0000)]
Back out rev 1.146 -- it broke the LINT build.
We are about to enter the 5.1 code freeze and things must be buildable.

21 years agoRevert rev 1.396 -- it broke the Alpha kernel build.
David E. O'Brien [Mon, 5 May 2003 09:09:16 +0000 (09:09 +0000)]
Revert rev 1.396 -- it broke the Alpha kernel build.

21 years agoAdd the dragon screen saver.
Murray Stokely [Mon, 5 May 2003 09:00:13 +0000 (09:00 +0000)]
Add the dragon screen saver.

PR: bin/51571
MFC After: 1 week

21 years agoRe-order the the initialization slightly to improve structure.
Poul-Henning Kamp [Mon, 5 May 2003 08:58:12 +0000 (08:58 +0000)]
Re-order the the initialization slightly to improve structure.

21 years agoNegate the logic of MAKE_KERBEROS5, and replace it with NO_KERBEROS.
Mark Murray [Mon, 5 May 2003 08:44:40 +0000 (08:44 +0000)]
Negate the logic of MAKE_KERBEROS5, and replace it with NO_KERBEROS.

21 years agoAdd sector size member to subdisk, plex and volume objects.
Greg Lehey [Mon, 5 May 2003 08:43:15 +0000 (08:43 +0000)]
Add sector size member to subdisk, plex and volume objects.

21 years agoRemove an accidental #include.
Greg Lehey [Mon, 5 May 2003 08:42:42 +0000 (08:42 +0000)]
Remove an accidental #include.

Maintain sector sizes for all objects, not just for drives.  Some of
this could do with improvement: in particular, we get an error if the
components of an object have different sector sizes.

Clean up some comments.

21 years agoImplement DIOCGSECTORSIZE and DIOCGMEDIASIZE ioctls. It appears that
Greg Lehey [Mon, 5 May 2003 08:41:53 +0000 (08:41 +0000)]
Implement DIOCGSECTORSIZE and DIOCGMEDIASIZE ioctls.  It appears that
they are now necessary to use newfs(8).

21 years agoUse a dedicated malloc(9) bucket for sector storage.
Poul-Henning Kamp [Mon, 5 May 2003 08:37:07 +0000 (08:37 +0000)]
Use a dedicated malloc(9) bucket for sector storage.

21 years agovinum_start: Call vinum_read with correct number of parameters.
Greg Lehey [Mon, 5 May 2003 08:33:36 +0000 (08:33 +0000)]
vinum_start: Call vinum_read with correct number of parameters.

21 years agoAdd two KASSERTS which trigger if free(9) would drag the "memuse" statistic
Poul-Henning Kamp [Mon, 5 May 2003 08:32:53 +0000 (08:32 +0000)]
Add two KASSERTS which trigger if free(9) would drag the "memuse" statistic
for a malloc bucket under zero.  This typically happens if you malloc(9)
from one bucket and free to another.

21 years agoDon't warn if the rawoffset is zero, that is actually the best value it
Poul-Henning Kamp [Mon, 5 May 2003 08:30:59 +0000 (08:30 +0000)]
Don't warn if the rawoffset is zero, that is actually the best value it
could have.

21 years agoregen.
Murray Stokely [Mon, 5 May 2003 08:22:15 +0000 (08:22 +0000)]
regen.

21 years agoAdd KEYSPAN USA19QW, JRC PHONE, SANDISK SDDR75.
Murray Stokely [Mon, 5 May 2003 08:18:53 +0000 (08:18 +0000)]
Add KEYSPAN USA19QW, JRC PHONE, SANDISK SDDR75.

PR: misc/48141, bin/51265, kern/49053, kern/50144

21 years agoTurn MAKE_KERBEROS5 into NO_KERBEROS by negating the logic. Some extra
Mark Murray [Mon, 5 May 2003 07:58:44 +0000 (07:58 +0000)]
Turn MAKE_KERBEROS5 into NO_KERBEROS by negating the logic. Some extra
cleanups were necessary in release/Makefile, and the tinderbox code
was syntax checked, not run checked.

21 years agoCorrect failed checkin. Only one of the hunks in the last checkin
Greg Lehey [Mon, 5 May 2003 07:50:15 +0000 (07:50 +0000)]
Correct failed checkin.  Only one of the hunks in the last checkin
seems to have made it.

21 years agoSince we insist on loading the POS ACPI by default, give the poor user
David E. O'Brien [Mon, 5 May 2003 07:33:12 +0000 (07:33 +0000)]
Since we insist on loading the POS ACPI by default, give the poor user
instructions on the main help screen for disabling it.

21 years agoUse le32dec() instead of le32toh() because we are not guaranteed to have
Poul-Henning Kamp [Mon, 5 May 2003 07:22:35 +0000 (07:22 +0000)]
Use le32dec() instead of le32toh() because we are not guaranteed to have
a word aligned input.

21 years agoTurn the check that rawoffset == mbroffset into a warning instead.
Poul-Henning Kamp [Mon, 5 May 2003 06:46:49 +0000 (06:46 +0000)]
Turn the check that rawoffset == mbroffset into a warning instead.

21 years agoBackout last commit. It is redundant in -CURRENT.
Murray Stokely [Mon, 5 May 2003 06:25:03 +0000 (06:25 +0000)]
Backout last commit.  It is redundant in -CURRENT.

Pointed out by: David Schultz

21 years agoAdd Norwegian dvorak keymap.
Murray Stokely [Mon, 5 May 2003 06:06:55 +0000 (06:06 +0000)]
Add Norwegian dvorak keymap.

PR: conf/47556
Submitted by: Eirik Nygaard <eirikn@bluezone.no>

21 years agoAdd amnia, amniotic, and amniocentesis.
Murray Stokely [Mon, 5 May 2003 05:51:16 +0000 (05:51 +0000)]
Add amnia, amniotic, and amniocentesis.

PR: misc/47342
Submitted by: Gareth McCaughan <Gareth.McCaughan@pobox.com>
(a pedantic medical student?)
MFC After: 1 week

21 years agoMake the output more consistent.
Murray Stokely [Mon, 5 May 2003 05:36:20 +0000 (05:36 +0000)]
Make the output more consistent.

21 years agoClarify comments.
Greg Lehey [Mon, 5 May 2003 05:35:52 +0000 (05:35 +0000)]
Clarify comments.

21 years agoAdd "preferred" keyword.
Greg Lehey [Mon, 5 May 2003 05:35:28 +0000 (05:35 +0000)]
Add "preferred" keyword.

21 years agoWhite space fixes.
Greg Lehey [Mon, 5 May 2003 05:34:39 +0000 (05:34 +0000)]
White space fixes.

21 years agoRemove DIOCGDINFO ioctl. This almost makes newfs work again without
Greg Lehey [Mon, 5 May 2003 05:33:59 +0000 (05:33 +0000)]
Remove DIOCGDINFO ioctl.  This almost makes newfs work again without
the -v option, though it's not clear that it won't bite us elsewhere.

Forgotten by: phk

Implement setreadpol() function for the VINUM_READPOL ioctl.

Submitted by:  Allan Saddi <allan@saddi.com>

21 years agovinum_scandisk: Make sure the config is locked before we mess around
Greg Lehey [Mon, 5 May 2003 05:32:52 +0000 (05:32 +0000)]
vinum_scandisk: Make sure the config is locked before we mess around
with it.

Finally implement read policies.  The previous "implementation" didn't
work because it referred to plexes which were almost invariably when
referred to.  Instead, deprecate the "prefer" keyword for volumes
(though it's still there for the moment) and add a keyword "preferred"
to the plex definition.  The relationship is like this:

Old:

  vol foo ... prefer foo.p3

New:

  plex foo.p3 volume foo preferred

print_config: Print "preferred" where appropriate.
      No longer print "prefer" on volume config entries.

21 years agoFinally implement read policies. The previous "implementation" didn't
Greg Lehey [Mon, 5 May 2003 05:32:15 +0000 (05:32 +0000)]
Finally implement read policies.  The previous "implementation" didn't
work because it referred to plexes which were almost invariably when
referred to.  Instead, deprecate the "prefer" keyword for volumes
(though it's still there for the moment) and add a keyword "preferred"
to the plex definition.  The relationship is like this:

Old:

  vol foo ... prefer foo.p3

New:

  plex foo.p3 volume foo preferred

give_plex_to_volume: set preferred plex if specified on plex
definition entry.  This involves adding a parameter to the function to
specify the preferred plex.

config_plex: Implement preferred keyword.

21 years agoMove the setting of the BOOTABLE flag outside of .if MAKE_ISOS so that
Murray Stokely [Mon, 5 May 2003 05:29:10 +0000 (05:29 +0000)]
Move the setting of the BOOTABLE flag outside of .if MAKE_ISOS so that
bootable CDs can still be created if the iso.1 target is called
directly.

PR: misc/47191
Submitted by: Joost Bekkers <joost@bps.jodocus.org>
MFC after: 1 week

21 years agofree_vinum: Rearrange sequence of actions to avoid potential race
Greg Lehey [Mon, 5 May 2003 05:28:15 +0000 (05:28 +0000)]
free_vinum: Rearrange sequence of actions to avoid potential race
condition when shutting down.

21 years agovinum_readpol: Get this right.
Greg Lehey [Mon, 5 May 2003 05:27:19 +0000 (05:27 +0000)]
vinum_readpol: Get this right.

Submitted by: Allan Saddi <allan@saddi.com>

vinum_read: White space cleanup.

21 years agoFinally implement read policies. The previous "implementation" didn't
Greg Lehey [Mon, 5 May 2003 05:25:49 +0000 (05:25 +0000)]
Finally implement read policies.  The previous "implementation" didn't
work because it referred to plexes which were almost invariably when
referred to.  Instead, deprecate the "prefer" keyword for volumes
(though it's still there for the moment) and add a keyword "preferred"
to the plex definition.  The relationship is like this:

Old:

  vol foo ... prefer foo.p3

New:

  plex foo.p3 volume foo preferred

printconfig: Print out the "preferred" keyword for plexes where
     appropriate.

     Still print out "prefer" for volumes.  The kernel module
     continues to accept this version, but it's probably not
     going to live much longer.

21 years agocall dump_queues() only when DEBUG_THREAD_KERN is defined, save some
David Xu [Mon, 5 May 2003 05:01:19 +0000 (05:01 +0000)]
call dump_queues() only when DEBUG_THREAD_KERN is defined, save some
cpu cycles.

21 years agoCatch up with 1995(?) for modem speeds (use 57600 instead of 115200 for
Bruce Evans [Mon, 5 May 2003 03:26:50 +0000 (03:26 +0000)]
Catch up with 1995(?) for modem speeds (use 57600 instead of 115200 for
the example host to modem speed).

21 years agoRevert rev 1.78 of dokern.sh. The ncv, stg, and nsp drivers aren't
Scott Long [Mon, 5 May 2003 02:04:55 +0000 (02:04 +0000)]
Revert rev 1.78 of dokern.sh.  The ncv, stg, and nsp drivers aren't
specific to pc98, and it looks like space on the kernel floppy isn't
as tight as it was.

21 years agoBack out accidentally committed debugging
Warner Losh [Sun, 4 May 2003 23:58:37 +0000 (23:58 +0000)]
Back out accidentally committed debugging

21 years agoWhen resetting to defaults, reset drainwait to the current default pf
Bruce Evans [Sun, 4 May 2003 22:46:33 +0000 (22:46 +0000)]
When resetting to defaults, reset drainwait to the current default pf
`sysctl -n kern.drainwait` instead of to the boot-time default of 300.

21 years agoReplace use of a spinlock with a mutex.
Daniel Eischen [Sun, 4 May 2003 22:36:46 +0000 (22:36 +0000)]
Replace use of a spinlock with a mutex.

21 years agoProtect against a race between granting a lock and accessing
Daniel Eischen [Sun, 4 May 2003 22:29:09 +0000 (22:29 +0000)]
Protect against a race between granting a lock and accessing
other parts of the lock.

Submitted by: davidxu

21 years agoStyle fixes.
Alexander Kabaev [Sun, 4 May 2003 22:13:04 +0000 (22:13 +0000)]
Style fixes.
Remove DBL_DIG, DBL_MIN, DBL_MAX and their FLT_ counterparts, they
were marked for deprecation ever since SUSv1 at least.
Only define ULLONG_MIN/MAX and LLONG_MAX if long long type is
supported.
Restore a lost comment in MI _limits.h file and remove it from
sys/limits.h where it does not belong.

21 years agoA few grammatical improvements.
Murray Stokely [Sun, 4 May 2003 22:13:00 +0000 (22:13 +0000)]
A few grammatical improvements.

PR: docs/40911
MFC After: 3 days

21 years agoCleanup hostname matching in syslogd:
Gregory Neil Shapiro [Sun, 4 May 2003 22:05:40 +0000 (22:05 +0000)]
Cleanup hostname matching in syslogd:

1. Hostnames were not treated case insensitively in all cases.

2. The method for stripping hostnames when reading the syslog.conf
   differed from that when finding the hostname of an incoming request.
   This lead to a broken match check.  In my case, it meant I had to
   have '@scooter.smi.example.com.example.com' to have 'logger.example.com'
   properly save messages from 'scooter.smi.sendmail.com'.

3. Add paranoia to cfline() such that it doesn't try to access memory
   outside of the bounds of the f_host string.

4. While I am here, get rid of an outdated comment, argv[{0,1,2}] are now
   checked for NULL after the strdup() calls.

Reviewed by: dwmalone
MFC after: 1 week

21 years agoNote that the idletime setting is not enforced.
Murray Stokely [Sun, 4 May 2003 21:35:41 +0000 (21:35 +0000)]
Note that the idletime setting is not enforced.

PR: docs/40952
MFC After: 3 days

21 years agoAdd more detail on CPUs to AS4100 section based on some experimentation.
Wilko Bulte [Sun, 4 May 2003 20:25:53 +0000 (20:25 +0000)]
Add more detail on CPUs to AS4100 section based on some experimentation.
Minor misc. wordsmithing

21 years agoAdd transparent handling of mbroffset for backwards compatibility.
Poul-Henning Kamp [Sun, 4 May 2003 19:27:22 +0000 (19:27 +0000)]
Add transparent handling of mbroffset for backwards compatibility.

21 years agoOnly accept a rawoffset if it is identical to the mbroffset.
Poul-Henning Kamp [Sun, 4 May 2003 19:26:31 +0000 (19:26 +0000)]
Only accept a rawoffset if it is identical to the mbroffset.

21 years agoAdd a way to read the current mbroffset from a BSD label class.
Poul-Henning Kamp [Sun, 4 May 2003 19:25:48 +0000 (19:25 +0000)]
Add a way to read the current mbroffset from a BSD label class.

21 years agoAdd gctl_set_param() function.
Poul-Henning Kamp [Sun, 4 May 2003 19:24:34 +0000 (19:24 +0000)]
Add gctl_set_param() function.

21 years agoExtend the scope of the vm_object lock in vm_object_terminate().
Alan Cox [Sun, 4 May 2003 19:23:40 +0000 (19:23 +0000)]
Extend the scope of the vm_object lock in vm_object_terminate().

21 years agoRemove debugging printfs which should not have been committed.
Poul-Henning Kamp [Sun, 4 May 2003 18:48:25 +0000 (18:48 +0000)]
Remove debugging printfs which should not have been committed.

21 years agoThis is now Gcc 3.3 WARNS 6 clean.
David E. O'Brien [Sun, 4 May 2003 18:34:00 +0000 (18:34 +0000)]
This is now Gcc 3.3 WARNS 6 clean.

21 years agoSet abitag __unused.
David E. O'Brien [Sun, 4 May 2003 18:33:26 +0000 (18:33 +0000)]
Set abitag __unused.

21 years agoChange the way loadbalancing works on RAID1's.
Søren Schmidt [Sun, 4 May 2003 16:17:54 +0000 (16:17 +0000)]
Change the way loadbalancing works on RAID1's.
Based on code partially by me and by <Tor.Egge@cvsup.no.freebsd.org>.

21 years agoFix suspend and resume.
Daniel Eischen [Sun, 4 May 2003 16:17:01 +0000 (16:17 +0000)]
Fix suspend and resume.

Submitted (in part) by: Kazuaki Oda <kaakun@highway.ne.jp>

21 years agofix typo in TI1515 ID.
Warner Losh [Sun, 4 May 2003 12:49:37 +0000 (12:49 +0000)]
fix typo in TI1515 ID.

21 years agoImplement dump function for ATA RAID's. Minor fixes by me...
Søren Schmidt [Sun, 4 May 2003 12:16:47 +0000 (12:16 +0000)]
Implement dump function for ATA RAID's. Minor fixes by me...

Submitted by: Tor Egge <Tor.Egge@cvsup.no.freebsd.org>

21 years agoAdd support for using the CDROM env var for device to use;
Søren Schmidt [Sun, 4 May 2003 10:09:57 +0000 (10:09 +0000)]
Add support for using the CDROM env var for device to use;

PR: 38178

21 years agoFix filename printing when using -l option.
Søren Schmidt [Sun, 4 May 2003 10:03:15 +0000 (10:03 +0000)]
Fix filename printing when using -l option.
PR: 43574

21 years agoDont fixate when doing dvdrw or test writing.
Søren Schmidt [Sun, 4 May 2003 09:55:46 +0000 (09:55 +0000)]
Dont fixate when doing dvdrw or test writing.

PR: 47195

21 years agoDefault speed is 4 not 1.
Søren Schmidt [Sun, 4 May 2003 09:53:05 +0000 (09:53 +0000)]
Default speed is 4 not 1.
PR: 50474

21 years agoAdded more descriptive error outputs.
Søren Schmidt [Sun, 4 May 2003 09:51:06 +0000 (09:51 +0000)]
Added more descriptive error outputs.

PR: 46096

21 years agoGrap the ATA lock on all channels before suspend, this makes certain
Søren Schmidt [Sun, 4 May 2003 09:34:14 +0000 (09:34 +0000)]
Grap the ATA lock on all channels before suspend, this makes certain
that we have no outstanding ops in transit, which would cause problems
on resume.

21 years agoOn getting status print the stripesize on RAID0's.
Søren Schmidt [Sun, 4 May 2003 09:28:43 +0000 (09:28 +0000)]
On getting status print the stripesize on RAID0's.

21 years agoFix a bunch of typos and grammatical errors.
Murray Stokely [Sun, 4 May 2003 08:23:24 +0000 (08:23 +0000)]
Fix a bunch of typos and grammatical errors.

PR: docs/40234
Submitted by: Chris Pepper <pepper@rockefeller.edu>  (mostly)
MFC After: 3 days

21 years agoEliminate the separate malloc type for the sparing table.
Scott Long [Sun, 4 May 2003 07:41:07 +0000 (07:41 +0000)]
Eliminate the separate malloc type for the sparing table.

21 years agoAdd a missing __inline. Strange that gcc never complained about it.
Scott Long [Sun, 4 May 2003 07:40:29 +0000 (07:40 +0000)]
Add a missing __inline.  Strange that gcc never complained about it.
Implement udf_readlblks() in terms of RDSECTOR.

21 years agoCorrectly calculate the size of the extent that should be read in
Scott Long [Sun, 4 May 2003 07:39:11 +0000 (07:39 +0000)]
Correctly calculate the size of the extent that should be read in
udf_readatoffset().  This should fixe problems with reading udf filesystems
created with mkisofs.

21 years agoForgot to update string and signal tables when some of the trap types
Jake Burkholder [Sun, 4 May 2003 07:21:04 +0000 (07:21 +0000)]
Forgot to update string and signal tables when some of the trap types
changed.

21 years agoAvoid a lock-order reversal and implement vm_object locking
Alan Cox [Sun, 4 May 2003 06:56:27 +0000 (06:56 +0000)]
Avoid a lock-order reversal and implement vm_object locking
in vm_pageout_page_free().

21 years agoDocument the login-backoff and login-retries capabilities.
Murray Stokely [Sun, 4 May 2003 06:43:24 +0000 (06:43 +0000)]
Document the login-backoff and login-retries capabilities.

PR: docs/51397
MFC After: 3 days

21 years agoDocument a workaround for getting past the groff infinite loop problem
Warner Losh [Sun, 4 May 2003 06:26:30 +0000 (06:26 +0000)]
Document a workaround for getting past the groff infinite loop problem
when your userland is older tahn your kernel and Makefile.inc1 therefore
thinks that you don't need to rebuild groff, when in fact you do.  This
workaround is to set OSRELDATE to 500110 before building, but only if you
have a new enough Makefile.inc1.

I've confirmed this on my laptop, which was suffering from this problem.
Since it only hits power users, we'll naively assume that such users are
smart enough to read UPDATING.

21 years agoRemove man page reference for gated, as it isn't in the base system
Murray Stokely [Sun, 4 May 2003 06:17:38 +0000 (06:17 +0000)]
Remove man page reference for gated, as it isn't in the base system
nor is it available from the ports tree.

21 years agoNote that gated is a commercial routing product now, and remove
Murray Stokely [Sun, 4 May 2003 06:13:25 +0000 (06:13 +0000)]
Note that gated is a commercial routing product now, and remove
references to it's man page, which is almost never installed on a
FreeBSD system.  The information about using this command with gated
has been retained.  I have just made it clear that gated is not a part
of FreeBSD.

PR: docs/51407

21 years agoGrab some of the latest TI cardbus bridge IDs. Inspired by a similar
Warner Losh [Sun, 4 May 2003 06:10:05 +0000 (06:10 +0000)]
Grab some of the latest TI cardbus bridge IDs.  Inspired by a similar
commit to NetBSD.  I'll add these to various tables soon.

21 years agoMy previous commit broke builds for restricted namespaces. Add
Martin Blapp [Sun, 4 May 2003 05:57:50 +0000 (05:57 +0000)]
My previous commit broke builds for restricted namespaces. Add
a underscore to fix this.

Reviewed by: peter

21 years agoForced commit. Grr: I used the wrong text before. Here's the correct
Greg Lehey [Sun, 4 May 2003 05:20:56 +0000 (05:20 +0000)]
Forced commit.  Grr: I used the wrong text before.  Here's the correct
information.

Rewrite of startup code:

vinum_scandisk:

  1.  Hand a string with device names, not a table and length.  This
      is the way we get it at system startup time.  This also means
      that we don't insist on a /dev in the names.
  2.  Allocate drive list dynamically.
  3.  If passed a null drive list, use kern.disks as a list to search.

21 years agoUpdate declaration of vinum_scandisk (only one parameter).
Greg Lehey [Sun, 4 May 2003 05:19:48 +0000 (05:19 +0000)]
Update declaration of vinum_scandisk (only one parameter).

21 years agoRewrite of startup code:
Greg Lehey [Sun, 4 May 2003 05:16:02 +0000 (05:16 +0000)]
Rewrite of startup code:

Add ioctl VINUM_READCONFIG which implements both the "read" and
"start" commands in vinum(8).  Aim for marginally better error
messages when something goes wrong.

21 years agoRewrite of startup code:
Greg Lehey [Sun, 4 May 2003 05:15:32 +0000 (05:15 +0000)]
Rewrite of startup code:

Add ioctl VINUM_READCONFIG which implements both the "read" and
"start" commands in vinum(8).

21 years agoRewrite of startup code:
Greg Lehey [Sun, 4 May 2003 05:15:07 +0000 (05:15 +0000)]
Rewrite of startup code:

parse_config: Remove kw_read case.  It's not used any more.

parse_user_config: Remove check for failed read.

21 years agoRewrite of startup code:
Greg Lehey [Sun, 4 May 2003 05:13:45 +0000 (05:13 +0000)]
Rewrite of startup code:

vinumattach: Move the startup code to vinum_scandisk.

21 years agoRewrite of startup code:
Greg Lehey [Sun, 4 May 2003 05:13:17 +0000 (05:13 +0000)]
Rewrite of startup code:

vinum_read:
  Accept 0 parameters, hand an empty string to the kernel if so.
  Use new ioctl VINUM_READCONFIG, not VINUM_CREATE, to read config.

vinum_start: When passed 0 parameters, drop the devstat kludge and
     call vinum_read instead.  It's no longer necessary to distinguish
     between "start" and "read" if no parameters are passed, and
     probably one of them should go away.

21 years agoMany developers run with userland != to kernel. While this isn't
Warner Losh [Sun, 4 May 2003 04:14:10 +0000 (04:14 +0000)]
Many developers run with userland != to kernel.  While this isn't
supported, it usually works for months at a time.  Allow these people
to override the OSRELDATE of their installed world when things don't
match and the exact OSRELDATE matters and is different than the
kernel.  Now that Makefile.inc1 depends more and more about which date
you have to optimize the pieces it builds, it may be necessary to
pessimize things if its guesses are wrong.

If OSRELDATE is already set, we won't fork the sysctl to find out what
the kernel's date is.

Developers on IRC suggested that they run mismatches all the time as
well.

Reviewed by: obrien

21 years agoImplement the node cache as a hash table.
Scott Long [Sun, 4 May 2003 03:40:11 +0000 (03:40 +0000)]
Implement the node cache as a hash table.

21 years agoUse __FBSDID vs. rcsid[]. Also protect sccs[] and copyright[] from GCC 3.3.
David E. O'Brien [Sun, 4 May 2003 02:54:49 +0000 (02:54 +0000)]
Use __FBSDID vs. rcsid[].  Also protect sccs[] and copyright[] from GCC 3.3.

21 years agoUse __FBSDID vs. rcsid[].
David E. O'Brien [Sun, 4 May 2003 02:51:42 +0000 (02:51 +0000)]
Use __FBSDID vs. rcsid[].

21 years agoSomething I missed in my tree.
Paul Saab [Sun, 4 May 2003 02:03:20 +0000 (02:03 +0000)]
Something I missed in my tree.
If the media changes, poke the DSP on the 5401/5411 Broadcom PHY's.

Obtained from: NetBSD

21 years agoKSE is useful for pc98.
Yoshihiro Takahashi [Sun, 4 May 2003 02:02:48 +0000 (02:02 +0000)]
KSE is useful for pc98.

21 years agoUse __FBSDID vs. rcsid[].
David E. O'Brien [Sun, 4 May 2003 00:59:13 +0000 (00:59 +0000)]
Use __FBSDID vs. rcsid[].

21 years agoFix signed/unsigned comparison warnings.
David E. O'Brien [Sun, 4 May 2003 00:56:00 +0000 (00:56 +0000)]
Fix signed/unsigned comparison warnings.

21 years agoFix a sign/unsigned comparison.
David E. O'Brien [Sun, 4 May 2003 00:43:39 +0000 (00:43 +0000)]
Fix a sign/unsigned comparison.

21 years agoCorrect spelling errors.
Justin T. Gibbs [Sun, 4 May 2003 00:20:07 +0000 (00:20 +0000)]
Correct spelling errors.

Switch to handling bad SCSI status as a sequencer interrupt
instead of having the kernel proccess these failures via
the completion queue.  This is done because:

 o The old scheme required us to pause the sequencer and clear
   critical sections for each SCB.  It seems that these pause
   actions, if coincident with a sequencer FIFO interrupt, would
   result in a FIFO interrupt getting lost or directing to the
   wrong FIFO.  This caused hangs when the driver was stressed
   under high "queue full" loads.
 o The completion code assumed that it was always called with
   the sequencer running.  This may not be the case in timeout
   processing where completions occur manually via
   ahd_pause_and_flushwork().
 o With this scheme, the extra expense of clearing critical
   sections is avoided since the sequencer will only self pause
   once all pending selections have cleared and it is not in
   a critical section.

  aic79xx.c
Add code to handle the new BAD_SCB_STATUS sequencer
interrupt code.  This just redirects the SCB through
the already existing ahd_complete_scb() code path.
Remove code in ahd_handle_scsi_status() that paused
the sequencer, made sure that no selections where
pending, and cleared critical sections.  Bad
status SCBs are now only processed when all of these
conditions are true.

  aic79xx.reg:
Add the BAD_SCB_STATUS sequencer interrupt code.

  aic79xx.seq:
When completing an SCB upload to the host, if
we are doing this because the SCB contains non-zero
SCSI status, defer completing the SCB until there
are no pending selection events.  When completing
these SCBs, use the new BAD_SCB_STATUS sequencer
interrupt.  For all other uploaded SCBs (currently
only for underruns), the SCB is completed via the
normal done queue.  Additionally, keep the SCB that
is currently being uploaded on the COMPLETE_DMA_SCB
list until the dma is completed, not just until the
DMA is started.  This ensures that the DMA is restarted
properly should the host disable the DMA transfer for
some reason.

In our RevA workaround for Maxtor drives, guard against
the host pausing us while trying to pause I/O until the
first data-valid REQ by clearing the current snapshot
so that we can tell if the transfer has completed prior
to us noticing the REQINIT status.

In cfg4data_intr, shave off an instruction before getting
the data path running by adding an entrypoint to the
overrun handler to also increment the FIFO use count.

In the overrun handler, be sure to clear our LONGJMP
address in both exit paths.

Perform a few sequencer optimizations.

  aic79xx.c:
Print the full path from the SCB when a packetized
status overrun occurs.

Remove references to LONGJMP_SCB which is being
removed from firmware usage.

Print the new SCB_FIFO_USE_COUNT field in the
per-SCB section of ahd_dump_card_state().  The
SCB_TAG field is now re-used by the sequencer,
so it no longer makes sense to reference this
field in the kernel driver.

  aic79xx.h:
Re-arrange fields in the hardware SCB from largest
size type to smallest.  This makes it easier to
move fields without changing field alignment.

The hardware scb tag field is now down near the
"spare" portion of the SCB to facilitate reuse
by the sequencer.

  aic79xx.reg:
Remove LONGJMP_ADDR.

Rearrange SCB fields to match aic79xx.h.
Add SCB_FIFO_USE_COUNT as the first byte
of the SCB_TAG field.

  aic79xx.seq:
Add a per-SCB "Fifos in use count" field and use
it to determine when it is safe (all data posted)
to deliver status back to the host.  The old method
involved polling one or both FIFOs to verify that
the current task did not have pending data.  This
makes running down the GSFIFO very cheap, so we
will empty the GSFIFO in one idle loop pass in
all cases.

Use this simplification of the completion process
to prune down the data FIFO teardown sequencer for
packetized transfers.  Much more code is now shared
between the data residual and transfer complete cases.

Correct some issues in the packetized status handler.
It used to be possible to CLRCHN our FIFO before status
had fully transferred to the host.  We also failed to
handle NONPACKREQ phases that could occur should a CRC
error occur during transmission of the status data packet.

Correct a few big endian issues:

  aic79xx.c:
  aic79xx_inline.h:
  aic79xx_pci.c:
  aic79xx_osm.c:
o Always get the SCB's tag via the SCB_GET_TAG acccessor
o Add missing use of byte swapping macros when touching
  hscb fields.
o Don't double swap SEEPROM data when it is printed.
  Correct a big-endian bug.  We cannot assign a
o When assigning a 32bit LE variable to a 64bit LE
  variable, we must be explict about how the words
  of the 64bit LE variable are initialized.  Cast to
  (uint32_t*) to do this.

aic79xx.c:
In ahd_clear_critical_section(), hit CRLSCSIINT
after restoring the interrupt masks to avoid what
appears to be a glitch on SCSIINT.  Any real SCSIINT
status will be persistent and will immidiately
reset SCSIINT.  This clear should only get rid of
spurious SCSIINTs.

This glitch was the cause of the "Unexpected PKT busfree"
status that occurred under high queue full loads

Call ahd_fini_scbdata() after shutdown so that
any ahd_chip_init() routine that might access
SCB data will not access free'd memory.

Reset the bus on an IOERR since the chip doesn't
seem to reset to the new voltage level without
this.

Change offset calculation for scatter gather maps
so that the calculation is correct if an integral
multiple of sg lists does not fit in the allocation
size.

Adjust bus dma tag for data buffers based on 39BIT
addressing flag in our softc.

Use the QFREEZE count to simplify ahd_pause_and_flushworkd().
We can thus rely on the sequencer eventually clearing ENSELO.

In ahd_abort_scbs(), fix a bug that could potentially
corrupt sequencer state.  The saved SCB was being
restored in the SCSI mode instead of the saved mode.
It turns out that the SCB did not need to be saved at all
as the scbptr is already restored by all subroutines
called during this function that modify that register.

aic79xx.c:
aic79xx.h:
aic79xx_pci.c:
Add support for parsing the seeprom vital product
data.  The VPD data are currently unused.

aic79xx.h:
aic79xx.seq:
aic79xx_pci.c:
Add a firmware workaround to make the LED blink
brighter during packetized operations on the H2A.

aic79xx_inline.h:
The host does not use timer interrupts, so don't
gate our decision on whether or not to unpause
the sequencer on whether or not a timer interrupt
is pending.