]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
20 years agoChange u_intXX_t to uintXX_t. Change a couple of 'unsigned long's to
dfr [Sat, 22 May 2004 16:14:17 +0000 (16:14 +0000)]
Change u_intXX_t to uintXX_t. Change a couple of 'unsigned long's to
uint32_t where appropriate.

20 years agoPerform conversions straight from the stream buffer instead of scanning
tjr [Sat, 22 May 2004 15:41:03 +0000 (15:41 +0000)]
Perform conversions straight from the stream buffer instead of scanning
through byte by byte with mbrtowc(). In the usual case (buffer is big
enough to contain the multibyte character, character does not straddle
buffer boundary) this results in only one call to mbrtowc() for each
wide character read.

20 years agoAssociate a multibyte conversion state object with each stream. Reset it
tjr [Sat, 22 May 2004 15:19:41 +0000 (15:19 +0000)]
Associate a multibyte conversion state object with each stream. Reset it
to the initial state when a stream is opened or seeked upon. Use the
stream's conversion state object instead of a freshly-zeroed one in
fgetwc(), fputwc() and ungetwc().

This is only a performance improvement for now, but it would also be
required in order to support state-dependent encodings.

20 years agoAdd Intel PCI vendor ID.
le [Sat, 22 May 2004 14:18:05 +0000 (14:18 +0000)]
Add Intel PCI vendor ID.

20 years agoDon't forget to reset if_hwassist back to 0 when hardware checksumming
yar [Sat, 22 May 2004 13:59:17 +0000 (13:59 +0000)]
Don't forget to reset if_hwassist back to 0 when hardware checksumming
is being turned off, or else TCP/IP will keep assigning the job to us.

Drivers themselves should consult if_capenable, not if_hwassist--the
latter is for the TCP/IP stack.

20 years agoRemove two debugging printf().
mux [Sat, 22 May 2004 13:15:14 +0000 (13:15 +0000)]
Remove two debugging printf().

On behalf of: cognet

20 years agoGet rid of a lockmgr consumer by making agp(4) use a standard mutex,
mux [Sat, 22 May 2004 13:06:38 +0000 (13:06 +0000)]
Get rid of a lockmgr consumer by making agp(4) use a standard mutex,
since it's always acquiring the lock exclusively.  This was tested
with X on an SMP box, with and without WITNESS.

20 years agoAdd regression tests for geom_stripe and geom_nop.
pjd [Sat, 22 May 2004 10:58:53 +0000 (10:58 +0000)]
Add regression tests for geom_stripe and geom_nop.

Supported by: Wheel - Open Technologies - http://www.wheel.pl

20 years agoBe more precise.
pjd [Sat, 22 May 2004 10:53:06 +0000 (10:53 +0000)]
Be more precise.

20 years agostyle.Makefile(5).
pjd [Sat, 22 May 2004 10:33:18 +0000 (10:33 +0000)]
style.Makefile(5).

20 years agoFix typos in comments.
le [Sat, 22 May 2004 09:29:52 +0000 (09:29 +0000)]
Fix typos in comments.

Submitted by:   Gerhard Gonter <gonter@falbala.wu-wien.ac.at>

20 years agoAdd a "-r" flag to ktrdump(1) to print relative timestamps when used
rwatson [Sat, 22 May 2004 08:26:10 +0000 (08:26 +0000)]
Add a "-r" flag to ktrdump(1) to print relative timestamps when used
with "-t" rather than absolute timestamps.  This allows the reader
to get a better sense of latency between events, such as time to
schedule an interrupt thread from time the interrupt occurred.  Assert
a copyright on ktrdump.c since I seem to be modifying it more than I
thought.

20 years agoTo date, unwiring a fictitious page has produced a panic. The reason
alc [Sat, 22 May 2004 04:53:51 +0000 (04:53 +0000)]
To date, unwiring a fictitious page has produced a panic.  The reason
being that PHYS_TO_VM_PAGE() returns the wrong vm_page for fictitious
pages but unwiring uses PHYS_TO_VM_PAGE().  The resulting panic
reported an unexpected wired count.  Rather than attempting to fix
PHYS_TO_VM_PAGE(), this fix takes advantage of the properties of
fictitious pages.  Specifically, fictitious pages will never be
completely unwired.  Therefore, we can keep a fictitious page's wired
count forever set to one and thereby avoid the use of
PHYS_TO_VM_PAGE() when we know that we're working with a fictitious
page, just not which one.

In collaboration with: green@, tegge@
PR: kern/29915

20 years agoSpelling and style fixes.
marius [Sat, 22 May 2004 01:56:18 +0000 (01:56 +0000)]
Spelling and style fixes.

Obtained from: NetBSD

20 years agoUse unsigned types for the arguments of the atomic(9) operations,
marius [Sat, 22 May 2004 00:52:16 +0000 (00:52 +0000)]
Use unsigned types for the arguments of the atomic(9) operations,
like described in the man page and done on all other architectures.

OK'ed by: tmm

20 years agoSwitch from BSD-style u_intXX_t to ISO C99 uintXX_t.
marius [Sat, 22 May 2004 00:47:26 +0000 (00:47 +0000)]
Switch from BSD-style u_intXX_t to ISO C99 uintXX_t.

20 years agoPlug three lock leaks.
mux [Sat, 22 May 2004 00:44:08 +0000 (00:44 +0000)]
Plug three lock leaks.

20 years agoRemove gconcat(8) utility. It was replaced by geom(8).
pjd [Fri, 21 May 2004 22:38:49 +0000 (22:38 +0000)]
Remove gconcat(8) utility. It was replaced by geom(8).

20 years ago- More clear example description.
pjd [Fri, 21 May 2004 22:35:18 +0000 (22:35 +0000)]
- More clear example description.
- Fix copy&paste bug.

20 years agoForce commit to note, that those manual pages were...
pjd [Fri, 21 May 2004 22:18:05 +0000 (22:18 +0000)]
Force commit to note, that those manual pages were...

Reviewed by: simon

20 years agoAdd manual pages for gconcat(8), gstripe(8) and gnop(8) utilities.
pjd [Fri, 21 May 2004 22:12:24 +0000 (22:12 +0000)]
Add manual pages for gconcat(8), gstripe(8) and gnop(8) utilities.

Supported by: Wheel - Open Technologies - http://www.wheel.pl

20 years agoSome wordsmithing and mdoc(7) cleanup.
josef [Fri, 21 May 2004 21:38:17 +0000 (21:38 +0000)]
Some wordsmithing and mdoc(7) cleanup.

Submitted by:   Michel Lavondès <fox@vader.aacc.cc.md.us>
PR:             docs/66823
Reviewed by:    simon

20 years agoAdd a quiet mode to ktrdump(1): if the "-q" flag is used, don't print
rwatson [Fri, 21 May 2004 21:24:58 +0000 (21:24 +0000)]
Add a quiet mode to ktrdump(1): if the "-q" flag is used, don't print
the pretty text header on top of the output.  Simplifies feeding the
results of tracing into a script for mechanical processing.

20 years agoWhen the 'f' flag is passed to ktrdump(1), use 40 characters for the
rwatson [Fri, 21 May 2004 21:15:48 +0000 (21:15 +0000)]
When the 'f' flag is passed to ktrdump(1), use 40 characters for the
"file and line" field consistently; previously, a 32-character field
length was used for the table header, which resulted in the header
not lining up with the table.

20 years agoCosmetic:
yar [Fri, 21 May 2004 20:34:04 +0000 (20:34 +0000)]
Cosmetic:
Set capability bits in a consistent way.
Add a comment on why the VLAN_MTU stuff comes after ether_ifattach().

20 years agoEnter the 1990's and assume that the computer knows what time it is.
imp [Fri, 21 May 2004 20:13:33 +0000 (20:13 +0000)]
Enter the 1990's and assume that the computer knows what time it is.
Print the ETA of dump being finished, rather than a cryptic delta
time.  Also, if we have written more blocks than the tapesize, assume
that we are 99.99% done and that we'll be finished 'soon'.

20 years agoFix cutNpasto in last commit.
imp [Fri, 21 May 2004 19:47:55 +0000 (19:47 +0000)]
Fix cutNpasto in last commit.

20 years agoAdd --dereference as a synonym for -H
kientzle [Fri, 21 May 2004 18:40:32 +0000 (18:40 +0000)]
Add --dereference as a synonym for -H

Pointed out by: Kris Kennaway (unbreaks nspr port build)

20 years agoThe driver fxp(4) has reception of large frames enabled hardcodedly,
yar [Fri, 21 May 2004 18:11:38 +0000 (18:11 +0000)]
The driver fxp(4) has reception of large frames enabled hardcodedly,
so let VLAN_MTU be marked in if_capenable from the beginning.

20 years agoMark the capability of this driver to receive VLAN frames >1500 bytes
yar [Fri, 21 May 2004 17:11:34 +0000 (17:11 +0000)]
Mark the capability of this driver to receive VLAN frames >1500 bytes
as initially active in if_capenable since it is always on.

Reviewed by: simokawa

20 years agoMFi386 (1.103 and 1.104: fixed some problems in high resolution profiling
bde [Fri, 21 May 2004 16:50:57 +0000 (16:50 +0000)]
MFi386 (1.103 and 1.104: fixed some problems in high resolution profiling
and improved some comments).  Also, made the documented {f,s}uword()
functions the standard entry points and the undocumented {f,s}uword64()
functions alternative entry points, like {f,s}uword32() for i386's.  The
bitrot in the comments was a little larger here -- there are new undocumented
32-bit sub-word functions, not just renaming of 16-bit functions from
documented ones to undocumented ones.

20 years agoUpdated and reorganized the comments for the fetch and store families of
bde [Fri, 21 May 2004 16:08:26 +0000 (16:08 +0000)]
Updated and reorganized the comments for the fetch and store families of
functions.

20 years agoFixed high resoultion profiling of fuword32() and suword32(). Use the
bde [Fri, 21 May 2004 16:01:54 +0000 (16:01 +0000)]
Fixed high resoultion profiling of fuword32() and suword32().  Use the
standard macro ALTENTRY() instead of a home made incomplete version
of it.

20 years ago- Change command name from 'config' to 'configure'.
pjd [Fri, 21 May 2004 15:23:48 +0000 (15:23 +0000)]
- Change command name from 'config' to 'configure'.
- Bump version number.

20 years agoClear KSE thread flags after KSE thread mode is ended. The side effect
davidxu [Fri, 21 May 2004 14:50:23 +0000 (14:50 +0000)]
Clear KSE thread flags after KSE thread mode is ended. The side effect
of not clearing the flags for execv() syscall will result that a new
program runs in KSE thread mode without enabling it.

Submitted by: tjr
Modified by: davidxu

20 years agoifdef writing to registers that the base pci standard says are
imp [Fri, 21 May 2004 14:41:02 +0000 (14:41 +0000)]
ifdef writing to registers that the base pci standard says are
read-only on D3->D0 power state transition.  Add a define to enable
them, but include a comment to contact me if there's a problem.

20 years agoUpon further review it was decided this piece of the msync(2)
kensmith [Fri, 21 May 2004 12:05:48 +0000 (12:05 +0000)]
Upon further review it was decided this piece of the msync(2)
fixes was applicable to HEAD, originally it was thought this
should only be done in RELENG_4.  Implement IO_INVAL in the vnode
op for writing by marking the buffer as "no cache".  This fix
has already been applied to RELENG_4 as Rev. 1.65.2.15 of
ufs/ufs/ufs_readwrite.c.

Reviewed by: alc, tegge

20 years agoAdd ru_RU.KOI8-R relnotes and build infrastructure
den [Fri, 21 May 2004 11:59:56 +0000 (11:59 +0000)]
Add ru_RU.KOI8-R relnotes and build infrastructure

Obtained from: The FreeBSD Russian Documentation Project

20 years agoFixed some style bugs in tdsigwakeup().
bde [Fri, 21 May 2004 10:02:24 +0000 (10:02 +0000)]
Fixed some style bugs in tdsigwakeup().

20 years agoFix spelling.
dfr [Fri, 21 May 2004 09:12:07 +0000 (09:12 +0000)]
Fix spelling.

20 years agoCorrect parsing of Solaris default ACLs.
kientzle [Fri, 21 May 2004 09:01:13 +0000 (09:01 +0000)]
Correct parsing of Solaris default ACLs.

20 years agoAdded dependency on the miibus module.
ru [Fri, 21 May 2004 08:43:38 +0000 (08:43 +0000)]
Added dependency on the miibus module.

20 years agoCompeletely rewrite the description of hw.pci.do_powerstate to sound
imp [Fri, 21 May 2004 07:06:54 +0000 (07:06 +0000)]
Compeletely rewrite the description of hw.pci.do_powerstate to sound
better.

20 years agoImprove the English somewhat.
imp [Fri, 21 May 2004 07:03:07 +0000 (07:03 +0000)]
Improve the English somewhat.

Prodded by: ru@

20 years agoOoops, forgot to commit the updated definition for hw.pci.do_powerstate
imp [Fri, 21 May 2004 06:43:46 +0000 (06:43 +0000)]
Ooops, forgot to commit the updated definition for hw.pci.do_powerstate
when I committed code that changed its meaning.

20 years agoMFp4:
imp [Fri, 21 May 2004 06:41:15 +0000 (06:41 +0000)]
MFp4:

Split the baby.  For idepci devices, now both legacy mode bits need
not be set.  We can run an idepci in a split mode.  However, it only
works better than before, not works.  It works better in that when one
device is legacy and the other isn't and disabled, we now operate
correctly.

sos submitted a version of this patch.

20 years agoMove pci_do_powerstate up a level. Now it just means 'do not turn devices
imp [Fri, 21 May 2004 06:39:09 +0000 (06:39 +0000)]
Move pci_do_powerstate up a level.  Now it just means 'do not turn devices
off into d3 state when there's no driver for the device'.  This should
help suspend/resume in the default case.

20 years agoMFp4: o save/restore subvendor, subdevice, vendor, device, baseclass,
imp [Fri, 21 May 2004 06:36:36 +0000 (06:36 +0000)]
MFp4: o save/restore subvendor, subdevice, vendor, device, baseclass,
subclass, progif and revid.  While these are typically read
only fields, they aren't always read-only.  progif is writable
for ata devices, for example.  It does no harm when they are
read only, and helps when they aren't.

20 years agoWhen attaching pccard and cardbus children, there's no need to set the
imp [Fri, 21 May 2004 06:11:47 +0000 (06:11 +0000)]
When attaching pccard and cardbus children, there's no need to set the
device == NULL on failure.  A warning should suffice.

# I wrote this back before I understood the unattached but loosely bound
# newbus concept...

20 years agoNow that we have the resource allocation code in current, the kludge
imp [Fri, 21 May 2004 06:10:13 +0000 (06:10 +0000)]
Now that we have the resource allocation code in current, the kludge
to try to allocate things on my parent can be taken out.  It duplicates code.

Also, add comment about why the power state stuff is here (type 2
devices don't participate in the power state save/restore due to
larger Bx issues).

20 years agomake the pci power state and resource code a lot less chatty. The
imp [Fri, 21 May 2004 06:03:26 +0000 (06:03 +0000)]
make the pci power state and resource code a lot less chatty.  The
chattiness was left in for debugging, but now that nearly all of the
problems relating to the changes have been fixed, it is only annoying.  It
is still available via bootverbose.

Prodded by: jhb

20 years agoSync to 1.178 of usbdevs
sanpei [Fri, 21 May 2004 01:39:38 +0000 (01:39 +0000)]
Sync to 1.178 of usbdevs

20 years agoadd support Kyocera AH-K3001V (cellular phone in Japan)
sanpei [Fri, 21 May 2004 01:36:48 +0000 (01:36 +0000)]
add support Kyocera AH-K3001V (cellular phone in Japan)

PR: kern/66779
Submitted by: Togawa Satoshi <toga@puyo.org>
MFC after: 1 week

20 years agoDelete history.
markm [Thu, 20 May 2004 20:52:04 +0000 (20:52 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:52:00 +0000 (20:52 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:51:56 +0000 (20:51 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:48:23 +0000 (20:48 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:48:19 +0000 (20:48 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:30:59 +0000 (20:30 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:30:51 +0000 (20:30 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:30:48 +0000 (20:30 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:24:18 +0000 (20:24 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:24:00 +0000 (20:24 +0000)]
Delete history.

20 years agoDelete history.
markm [Thu, 20 May 2004 20:23:26 +0000 (20:23 +0000)]
Delete history.

20 years agoIn tdsigwakeup(), use TD_ON_SLEEPQ() rather than TD_IS_SLEEPING() to see if
jhb [Thu, 20 May 2004 20:17:28 +0000 (20:17 +0000)]
In tdsigwakeup(), use TD_ON_SLEEPQ() rather than TD_IS_SLEEPING() to see if
a thread is on a sleep queue and should have it's sleep aborted.

Reported by: Thierry Herbelot thierry at herbelot dot com

20 years agoVarious style.Makefile(5) improvements.
pjd [Thu, 20 May 2004 20:12:17 +0000 (20:12 +0000)]
Various style.Makefile(5) improvements.

Inspired by: ru

20 years agoVarious style.Makefile(5) improvements.
pjd [Thu, 20 May 2004 20:05:05 +0000 (20:05 +0000)]
Various style.Makefile(5) improvements.

Provoked by: ru

20 years agoUpdate the date.
pjd [Thu, 20 May 2004 19:32:45 +0000 (19:32 +0000)]
Update the date.

Requested by: ru

20 years agoRemove unneeded lseek(2) hack to position past the 2GB point,
hmp [Thu, 20 May 2004 19:25:27 +0000 (19:25 +0000)]
Remove unneeded lseek(2) hack to position past the 2GB point,
use fseeko(3) instead.

This commit fixes breakage when `lastcomm matchstring` is run.

PR:           bin/66765, bin/64568
Submitted by: Dan Nelson <dnelson at allantgroup.com>

20 years agoUpdate the document date.
alc [Thu, 20 May 2004 18:47:15 +0000 (18:47 +0000)]
Update the document date.

Reminded by: ru@

20 years agoIn cpu_sched_exit(), we must check vm_refcnt against 0, not 1, since
tmm [Thu, 20 May 2004 18:41:07 +0000 (18:41 +0000)]
In cpu_sched_exit(), we must check vm_refcnt against 0, not 1, since
exit1() decrements the reference count before calling this function.

20 years agoAdded "rescue".
ru [Thu, 20 May 2004 18:02:24 +0000 (18:02 +0000)]
Added "rescue".

Submitted by: Kent Hauser <kent.hauser@verizon.net>

20 years agoFixed printf format errors which helped break GUPROF for arches with
bde [Thu, 20 May 2004 16:48:17 +0000 (16:48 +0000)]
Fixed printf format errors which helped break GUPROF for arches with
64-bit function pointers.

20 years agoInitialize the history counter type field in struct gmonparam as
bde [Thu, 20 May 2004 16:42:39 +0000 (16:42 +0000)]
Initialize the history counter type field in struct gmonparam as
threatened in rev.1.10 of usr.sbin/kgmon/kgmon.c more than 2 years ago.
kgmon has been recovering from the missing initialization for too
long, but the fixup there is ifdefed for i386's and shouldn't be
needed for other arches.

20 years agoClean up language.
ru [Thu, 20 May 2004 16:40:06 +0000 (16:40 +0000)]
Clean up language.

Reviewed by: rik

20 years agoMFi386 (1.37: GUPROF calibration macros; only routine adjustments needed).
bde [Thu, 20 May 2004 16:22:57 +0000 (16:22 +0000)]
MFi386 (1.37: GUPROF calibration macros; only routine adjustments needed).

20 years agoMoved i386 asms to an i386 header. The asms are for calibration of
bde [Thu, 20 May 2004 16:12:19 +0000 (16:12 +0000)]
Moved i386 asms to an i386 header.  The asms are for calibration of
high resolution kernel profiling (options GUPROF.  "U" in GUPROF stands
for microseconds resolution, but the resolution is now smaller than 1
nanosecond on multi-GHz machines and the accuracy is heading towards
1 nanosecond too).  Arches that support GUPROF must now provide certain
macros for the calibration.  GUPROF is now only supported for i386's,
so the absence of the new macros for other arches doesn't break anything
that wasn't already broken.  amd64's have uncommitted support for
GUPROF, and sparc64's have support that seems to be complete except
here (there was an #error for non-i386 cases; now there are undefined
macros).

Changed the asms a little:
- declare them as __volatile.  They must not be moved, and exporting a
  label across asms is technically incorrect, so try harder to stop gcc
  moving them.
- don't put the non-clobbered register "bx" in the clobber list.  The
  clobber lists are still more conservative than necessary.
- drop the non-support for gcc-1.  It just gave a better error message,
  and this is not useful since compiling with gcc-1 would cause thousands
  of worse error messages.
- drop the support for aout.

20 years agoReally remove the return statement this time. Thanks to
mtm [Thu, 20 May 2004 15:53:49 +0000 (15:53 +0000)]
Really remove the return statement this time. Thanks to
ru for noticing.

20 years agoWhitespace cleanup
des [Thu, 20 May 2004 15:09:41 +0000 (15:09 +0000)]
Whitespace cleanup

20 years agoWhitespace cleanup.
des [Thu, 20 May 2004 15:01:26 +0000 (15:01 +0000)]
Whitespace cleanup.

20 years agoFix supposed compilation problem, using LIBDIR here can confuse core/
pjd [Thu, 20 May 2004 14:59:41 +0000 (14:59 +0000)]
Fix supposed compilation problem, using LIBDIR here can confuse core/
compilation.

20 years agoWhitespace cleanup.
des [Thu, 20 May 2004 14:49:12 +0000 (14:49 +0000)]
Whitespace cleanup.

20 years agoThe err routine requires two arguments: an exit value and a string.
mtm [Thu, 20 May 2004 14:16:05 +0000 (14:16 +0000)]
The err routine requires two arguments: an exit value and a string.
Additionaly, it will exit the script so an return statements after
calling it are superflous.

20 years ago- Add missing 'the'.
pjd [Thu, 20 May 2004 13:38:48 +0000 (13:38 +0000)]
- Add missing 'the'.
- Document /usr/include/geom/ directory and its subdirectories.

Requested by: ru

20 years agoMerge updated patch from vendor branch (1.1.1.17)
des [Thu, 20 May 2004 13:17:16 +0000 (13:17 +0000)]
Merge updated patch from vendor branch (1.1.1.17)

20 years agoAdd myself as maintainer for geom_nop, geom_stripe and geom(8).
pjd [Thu, 20 May 2004 12:44:23 +0000 (12:44 +0000)]
Add myself as maintainer for geom_nop, geom_stripe and geom(8).

20 years agoMake libthr async-signal-safe without costly signal masking. The guidlines I
mtm [Thu, 20 May 2004 12:06:16 +0000 (12:06 +0000)]
Make libthr async-signal-safe without costly signal masking. The guidlines I
followed are: Only 3 functions (pthread_cancel, pthread_setcancelstate,
pthread_setcanceltype) are required to be async-signal-safe by POSIX. None of
the rest of the pthread api is required to be async-signal-safe. This means
that only the three mentioned functions are safe to use from inside
signal handlers.
However, there are certain system/libc calls that are
cancellation points that a caller may call from within a signal handler,
and since they are cancellation points calls have to be made into libthr
to test for cancellation and exit the thread if necessary. So, the
cancellation test and thread exit code paths must be async-signal-safe
as well. A summary of the changes follows:

o Almost all of the code paths that masked signals, as well as locking the
  pthread structure now lock only the pthread structure.
o Signals are masked (and left that way) as soon as a thread enters
  pthread_exit().
o The active and dead threads locks now explicitly require that signals
  are masked.
o Access to the isdead field of the pthread structure is protected by both
  the active and dead list locks for writing. Either one is sufficient for
  reading.
o The thread state and type fields have been combined into one three-state
  switch to make it easier to read without requiring a lock. It doesn't need
  a lock for writing (and therefore for reading either) because only the
  current thread can write to it and it is an integer value.
o The thread state field of the pthread structure has been eliminated. It
  was an unnecessary field that mostly duplicated the flags field, but
  required additional locking that would make a lot more code paths require
  signal masking. Any truly unique values (such as PS_DEAD) have been
  reborn as separate members of the pthread structure.
o Since the mutex and condvar pthread functions are not async-signal-safe
  there is no need to muck about with the wait queues when handling
  a signal ...
o ... which also removes the need for wrapping signal handlers and sigaction(2).
o The condvar and mutex async-cancellation code had to be revised as a result
  of some of these changes, which resulted in semi-unrelated changes which
  would have been difficult to work on as a separate commit, so they are
  included as well.

The only part of the changes I am worried about is related to locking for
the pthread joining fields. But, I will take a closer look at them once this
mega-patch is committed.

20 years agoForced commit for rev. 1.26
mtm [Thu, 20 May 2004 12:04:17 +0000 (12:04 +0000)]
Forced commit for rev. 1.26

Bugfix: recursive mutex reference counting.

Noticed by:Michael Bretterklieber <mbretter@inode.at>
Partl Submitted by: deischen

20 years ago\16
mtm [Thu, 20 May 2004 11:55:04 +0000 (11:55 +0000)]
\16

20 years agoStylistic changes around the previous commit:
yar [Thu, 20 May 2004 11:04:09 +0000 (11:04 +0000)]
Stylistic changes around the previous commit:

- since the number of supported capabilities is growing,
  set bits in if_cap* in a consistent way;

- unexpand(1) leading SPACE characters.

20 years agoInform hier(7) manual page about /lib/geom/ directory.
pjd [Thu, 20 May 2004 11:02:58 +0000 (11:02 +0000)]
Inform hier(7) manual page about /lib/geom/ directory.

Reminded by: ru

20 years agoSet the VLAN bits in if_capenable as well as in if_capabilities
yar [Thu, 20 May 2004 10:57:33 +0000 (10:57 +0000)]
Set the VLAN bits in if_capenable as well as in if_capabilities
because VLAN hardware features are enabled in em(4) by default.

Note: Currently vlan(4) has a bug that it consults
if_capabilities, not if_capenable.  This will be fixed
after all the network drivers set VLAN bits in
if_capenable properly.

20 years ago- Teach CONCAT class how to talk with geom(8).
pjd [Thu, 20 May 2004 10:40:18 +0000 (10:40 +0000)]
- Teach CONCAT class how to talk with geom(8).
- Remove provider if any disk was lost.
- Dump CONCAT version.

Supported by: Wheel - Open Technologies - http://www.wheel.pl

20 years ago- Connect geom(8) and its libraries to the build.
pjd [Thu, 20 May 2004 10:37:13 +0000 (10:37 +0000)]
- Connect geom(8) and its libraries to the build.
- Connect geom_stripe and geom_nop modules to the build.
- Connect STRIPE and NOP classes to the LINT build.
- Disconnect gconcat(8) from the build.

Supported by: Wheel - Open Technologies - http://www.wheel.pl

20 years ago- Install includes used by STRIPE and NOP GEOM classes.
pjd [Thu, 20 May 2004 10:29:26 +0000 (10:29 +0000)]
- Install includes used by STRIPE and NOP GEOM classes.
- Create needed directories.

Supported by: Wheel - Open Technologies - http://www.wheel.pl

20 years agoModules Makefiles for geom_stripe and geom_nop.
pjd [Thu, 20 May 2004 10:27:20 +0000 (10:27 +0000)]
Modules Makefiles for geom_stripe and geom_nop.

Supported by: Wheel - Open Technologies - http://www.wheel.pl

20 years agoIntroduce geom(8)-specific shared libraries for CONCAT, STRIPE and NOP
pjd [Thu, 20 May 2004 10:24:23 +0000 (10:24 +0000)]
Introduce geom(8)-specific shared libraries for CONCAT, STRIPE and NOP
GEOM classes. CONCAT should be 100% compatible with existing gconcat(8)
utility, which is going to be removed.

Supported by: Wheel - Open Technologies - http://www.wheel.pl

20 years agoIntroduce STRIPE GEOM class. It implements RAID0 transformation and it
pjd [Thu, 20 May 2004 10:20:49 +0000 (10:20 +0000)]
Introduce STRIPE GEOM class. It implements RAID0 transformation and it
is intend to be fast. Just like CONCAT class it provides manual and
auto configuration methods.

Supported by: Wheel - Open Technologies - http://www.wheel.pl

20 years agoUpdate copyright years.
den [Thu, 20 May 2004 10:16:19 +0000 (10:16 +0000)]
Update copyright years.

20 years agoIntroduce NOP GEOM class. This is totally transparent GEOM class, but
pjd [Thu, 20 May 2004 10:15:53 +0000 (10:15 +0000)]
Introduce NOP GEOM class. This is totally transparent GEOM class, but
it is very useful for tests. One is able to destroy its provider
forcibly if wants to test how other class handle such events.
One is also able to specify failure probability to check how other
classes handle I/O errors.

Supported by: Wheel - Open Technologies - http://www.wheel.pl