]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
16 years agoThis commit was generated by cvs2svn to compensate for changes in r175296,
David E. O'Brien [Sun, 13 Jan 2008 20:35:16 +0000 (20:35 +0000)]
This commit was generated by cvs2svn to compensate for changes in r175296,
which included commits to RCS files with non-trunk default branches.

16 years agoImport file version 4.23 onto the vendor branch
David E. O'Brien [Sun, 13 Jan 2008 20:35:16 +0000 (20:35 +0000)]
Import file version 4.23 onto the vendor branch

16 years agoVersion bumping in order to reflect VOP_LOCK1(), VOP_UNLOCK() KPI
Attilio Rao [Sun, 13 Jan 2008 14:48:51 +0000 (14:48 +0000)]
Version bumping in order to reflect VOP_LOCK1(), VOP_UNLOCK() KPI
changing.

16 years agoVOP_LOCK1() (and so VOP_LOCK()) and VOP_UNLOCK() are only used in
Attilio Rao [Sun, 13 Jan 2008 14:44:15 +0000 (14:44 +0000)]
VOP_LOCK1() (and so VOP_LOCK()) and VOP_UNLOCK() are only used in
conjuction with 'thread' argument passing which is always curthread.
Remove the unuseful extra-argument and pass explicitly curthread to lower
layer functions, when necessary.

KPI results broken by this change, which should affect several ports, so
version bumping and manpage update will be further committed.

Tested by: kris, pho, Diego Sardina <siarodx at gmail dot com>

16 years agoAdd warning about this script dealing with untrusted data.
Simon L. B. Nielsen [Sun, 13 Jan 2008 14:27:53 +0000 (14:27 +0000)]
Add warning about this script dealing with untrusted data.

MFC after: 1 week

16 years agoUnbreak detection of cryptodev support for FreeBSD which was broken
Simon L. B. Nielsen [Sun, 13 Jan 2008 11:44:47 +0000 (11:44 +0000)]
Unbreak detection of cryptodev support for FreeBSD which was broken
with OpenSSL 0.9.8 import.

Note that this does not enable cryptodev by default, as it was the
case with OpenSSL 0.9.7 in FreeBSD base, but this change makes it
possible to enable cryptodev at all.

This has been submitted upstream as:
http://rt.openssl.org/Ticket/Display.html?id=1624

Submitted by: nork

16 years agoUse our standard verbose spelling of rules variables.
David E. O'Brien [Sun, 13 Jan 2008 09:45:53 +0000 (09:45 +0000)]
Use our standard verbose spelling of rules variables.
(as a nice side affect, this will make gnu/usr.bin/cvs/contrib/Makefile
have a later date than contrib/cvs/contrib/Makefile.in - which will help
the build break after the 1.11.22 CVS import...)

16 years agoDisable contrib - there is weirdness going on with it.
David E. O'Brien [Sun, 13 Jan 2008 08:58:25 +0000 (08:58 +0000)]
Disable contrib - there is weirdness going on with it.

16 years agoAttach the tr(1) regression tests to the build.
Giorgos Keramidas [Sun, 13 Jan 2008 08:33:58 +0000 (08:33 +0000)]
Attach the tr(1) regression tests to the build.

Approved by: das, dds

16 years agoAdd a new set of regression tests, for the tr(1) utility.
Giorgos Keramidas [Sun, 13 Jan 2008 08:33:20 +0000 (08:33 +0000)]
Add a new set of regression tests, for the tr(1) utility.

Approved by:    das, dds

16 years agoPrevent the `fix?' question from running together
Yaroslav Tykhiy [Sun, 13 Jan 2008 07:56:36 +0000 (07:56 +0000)]
Prevent the `fix?' question from running together
with the problem description on the terminal: use
conventional punctuation.

16 years agoReenable cvs with the build.
David E. O'Brien [Sun, 13 Jan 2008 06:56:31 +0000 (06:56 +0000)]
Reenable cvs with the build.

16 years agoUpdate for the 1.11.22 import.
David E. O'Brien [Sun, 13 Jan 2008 06:35:22 +0000 (06:35 +0000)]
Update for the 1.11.22 import.

Reviewed/begged-for by: peter

16 years agoThis commit was generated by cvs2svn to compensate for changes in r175283,
David E. O'Brien [Sun, 13 Jan 2008 06:24:49 +0000 (06:24 +0000)]
This commit was generated by cvs2svn to compensate for changes in r175283,
which included commits to RCS files with non-trunk default branches.

16 years agoWe use the stock versions of these files.
David E. O'Brien [Sun, 13 Jan 2008 06:20:11 +0000 (06:20 +0000)]
We use the stock versions of these files.

Reviewed/begged-for by: peter

16 years agoMerge what 1.11.17 -> 1.11.22 ../doc/cvs.1 changes I could.
David E. O'Brien [Sun, 13 Jan 2008 06:14:29 +0000 (06:14 +0000)]
Merge what 1.11.17 -> 1.11.22 ../doc/cvs.1 changes I could.
We never merged in the 1.11.15 -> 1.11.17 changes - due to the moving
of cvs.1 from man/ to doc/ and being auto generated.

16 years agoMerge rev 1.9 (update "-T" Template support) and rev 1.2 (-D'date'
David E. O'Brien [Sun, 13 Jan 2008 06:13:25 +0000 (06:13 +0000)]
Merge rev 1.9 (update "-T" Template support) and rev 1.2 (-D'date'
option with -r'branch' on update) into version 1.11.22.

16 years agoMerge rev 1.9 (new long flag to ignore the CVSROOT/passwd file)
David E. O'Brien [Sun, 13 Jan 2008 06:12:54 +0000 (06:12 +0000)]
Merge rev 1.9 (new long flag to ignore the CVSROOT/passwd file)
rev 1.2 (local tag/$Id$ keyword support)  into version 1.11.22.

16 years agoMerge rev 1.28 (CAN-2005-0753 / FreeBSD-SA-05:05.cvs fixes),
David E. O'Brien [Sun, 13 Jan 2008 06:12:27 +0000 (06:12 +0000)]
Merge rev 1.28 (CAN-2005-0753 / FreeBSD-SA-05:05.cvs fixes),
rev 1.27 ("iso8601" option keyword) revs 1.12/1.10/1.5/1.4 ($CVSHeader$
support) rev 1.2 ($CVS_LOCAL_BRANCH_NUM support for local commit
feature of cvsup)  into version 1.11.22.

Note rev 1.21 ("-D date" checkout bug relating to 1.1.1.1 vs 1.1
revisions), rev 1.13 (allow -D'date' with -r'branch' on a checkout),
rev 1.6 (use xstrdup rather than strdup)  are fixed in the vendor sources

16 years agoMerge rev 1.11 ("iso8601" option keyword) into version 1.11.22.
David E. O'Brien [Sun, 13 Jan 2008 06:11:36 +0000 (06:11 +0000)]
Merge rev 1.11 ("iso8601" option keyword)  into version 1.11.22.
(all our other local changes are OBE)

16 years agoMerge rev 1.2 (extensions to the $CVSROOT/config syntax) and
David E. O'Brien [Sun, 13 Jan 2008 06:10:58 +0000 (06:10 +0000)]
Merge rev 1.2 (extensions to the $CVSROOT/config syntax) and
rev 1.4 (1.12 CVSROOT/config keyword expansion forward compatability)
into version 1.11.22.

16 years agoMerge rev 1.25 ("iso8601" option keyword), rev 1.23 (blame),
David E. O'Brien [Sun, 13 Jan 2008 06:10:18 +0000 (06:10 +0000)]
Merge rev 1.25 ("iso8601" option keyword), rev 1.23 (blame),
rev 1.17 (environtmental var "CVS_OPTIONS"), rev 1.14 ('-g' option to
support shared-group access), rev 1.7 ('-R' read-only repository mode),
rev 1.6 (support for checking out from a read-only repository),
revs 1.4 & 1.5 ("tagexpand=")  into version 1.11.22.

16 years agoMerge rev 1.9 (put the blank line back in the beginning of constructed
David E. O'Brien [Sun, 13 Jan 2008 06:09:41 +0000 (06:09 +0000)]
Merge rev 1.9 (put the blank line back in the beginning of constructed
log messages being passed to $EDITOR)  into version 1.11.22.

16 years agoMerge rev 1.3 ('-n' option) & rev. 1.2 ("iso8601" option keyword)
David E. O'Brien [Sun, 13 Jan 2008 06:08:49 +0000 (06:08 +0000)]
Merge rev 1.3 ('-n' option) & rev. 1.2 ("iso8601" option keyword)
into version 1.11.22.

16 years agoMerge rev 1.3 (catch write-lock attempts immediately if running in
David E. O'Brien [Sun, 13 Jan 2008 06:08:17 +0000 (06:08 +0000)]
Merge rev 1.3 (catch write-lock attempts immediately if running in
read-only mode) & rev 1.2 (check out from read-only repository support)
into version 1.11.22.

16 years agoMerge revs 1.2 (comprehensive CVS/Template support) into version 1.11.22.
David E. O'Brien [Sun, 13 Jan 2008 06:07:20 +0000 (06:07 +0000)]
Merge revs 1.2 (comprehensive CVS/Template support)  into version 1.11.22.

16 years agoMerge rev 1.2 (do not check val-tags if the repository is read-only)
David E. O'Brien [Sun, 13 Jan 2008 06:06:41 +0000 (06:06 +0000)]
Merge rev 1.2 (do not check val-tags if the repository is read-only)
into version 1.11.22.

16 years agoToss all of our local changes - they are OBE.
David E. O'Brien [Sun, 13 Jan 2008 06:06:15 +0000 (06:06 +0000)]
Toss all of our local changes - they are OBE.
However, stay with version 1.9.26's buflen of BUFSIZ as we have in the
past vs. going with the 1.10+ limited length of 128.

16 years agoMerge rev 1.16: implement -j for cvs diff, allowing tag:date based diffs
David E. O'Brien [Sun, 13 Jan 2008 06:05:24 +0000 (06:05 +0000)]
Merge rev 1.16: implement -j for cvs diff, allowing tag:date based diffs
into version 1.11.22.

16 years agoChange default from rsh to ssh. (moved from client.c).
David E. O'Brien [Sun, 13 Jan 2008 06:04:43 +0000 (06:04 +0000)]
Change default from rsh to ssh. (moved from client.c).
Merge rev 1.14: comprehensive -T CVS/Template support, rev 1.9: new long
flag that causes cvs to ignore the CVSROOT/passwd file, rev 1.3: support
for checking out from a read-only repository, rev. 1.2: support for local
$Id$ keyword  into cvs 1.11.22.
Note that rev 1.4 (make verifymsg extra useful) is OBE.

16 years agoMerge rev 1.7: always upload new files, even if the timestamps match,
David E. O'Brien [Sun, 13 Jan 2008 06:02:28 +0000 (06:02 +0000)]
Merge rev 1.7: always upload new files, even if the timestamps match,
rev 1.4: flip the default for CVS_RSH to "ssh", rev 1.2: fix a problem
sometimes seen when doing checkouts from a local repo and committing
via remote cvs (a cvs -d override of the mismatched CVS/Root files was
missing)  into cvs 1.11.22.

16 years agoMerge our -T comprehensive CVS/Template support into cvs 1.11.22.
David E. O'Brien [Sun, 13 Jan 2008 06:00:42 +0000 (06:00 +0000)]
Merge our -T comprehensive CVS/Template support into cvs 1.11.22.

16 years agoUpdate for the 1.11.22 import.
David E. O'Brien [Sun, 13 Jan 2008 05:50:33 +0000 (05:50 +0000)]
Update for the 1.11.22 import.

16 years agoThis commit was generated by cvs2svn to compensate for changes in r175261,
David E. O'Brien [Sun, 13 Jan 2008 05:49:32 +0000 (05:49 +0000)]
This commit was generated by cvs2svn to compensate for changes in r175261,
which included commits to RCS files with non-trunk default branches.

16 years agoImport cvs-1.11.22 onto vendor branch.
David E. O'Brien [Sun, 13 Jan 2008 05:49:32 +0000 (05:49 +0000)]
Import cvs-1.11.22 onto vendor branch.

16 years agoDisable CVS from the build.
David E. O'Brien [Sun, 13 Jan 2008 05:49:03 +0000 (05:49 +0000)]
Disable CVS from the build.

16 years agoUse size_t to avoid overflow when sorting arrays larger than 2 GB.
David Schultz [Sun, 13 Jan 2008 02:11:10 +0000 (02:11 +0000)]
Use size_t to avoid overflow when sorting arrays larger than 2 GB.

PR: 111085
MFC after: 2 weeks

16 years ago- Add support for using swap backed md(4) devices for building the
Simon L. B. Nielsen [Sat, 12 Jan 2008 22:58:06 +0000 (22:58 +0000)]
- Add support for using swap backed md(4) devices for building the
  disk image.  In some cases this can be a significant speed-up, if
  most of the image can be kept in RAM while being populated.

  On the 2GB image I'm currently working with, the build time,
  excluding buildworld/buildkernel, goes from ~17 minutes to ~6
  minutes.

  This is not enabled by default, as it might have the opposite effect
  on low-memory systems.

- During the generation of the image file be a bit more verbose in the
  log file so it is possible to see what's being done.

- Add a NANO_DISKIMGDIR variable which makes it possibly to place the
  final images somewhere other than ${MAKEOBJDIRPREFIX}.  The default
  value for NANO_DISKIMGDIR is $MAKEOBJDIRPREFIX.

Go for it: phk

16 years agoThis commit was generated by cvs2svn to compensate for changes in r175256,
Nate Lawson [Sat, 12 Jan 2008 22:13:12 +0000 (22:13 +0000)]
This commit was generated by cvs2svn to compensate for changes in r175256,
which included commits to RCS files with non-trunk default branches.

16 years agoFix GPE livelock that occurs on HP/Compaq laptops, mostly in the thermal
Nate Lawson [Sat, 12 Jan 2008 22:13:12 +0000 (22:13 +0000)]
Fix GPE livelock that occurs on HP/Compaq laptops, mostly in the thermal
zone code.  The GPE handler method (i.e. _L00) generates various Notify
events that need to be run to completion before the GPE is re-enabled.
In ACPI-CA, we queue an asynch callback at the same priority as a Notify
so that it will only run after all Notify handlers have completed.  The
callback re-enables the GPE afterwards.  We also changed the priority of
Notifies to be the same as GPEs, given the possibility that another GPE
could arrive before the Notifies have completed and we don't want it to
get queued ahead of the rest.

The ACPI-CA change was submitted by Alexey Starikovskiy (SUSE) and will
appear in a later release.  Special thanks to him for helping track this
bug down.

MFC after: 1 week
Tested by: jhb, Yousif Hassan <yousif / alumni.jmu.edu>

16 years agoFix GPE livelock that occurs on HP/Compaq laptops, mostly in the thermal
Nate Lawson [Sat, 12 Jan 2008 22:13:12 +0000 (22:13 +0000)]
Fix GPE livelock that occurs on HP/Compaq laptops, mostly in the thermal
zone code.  The GPE handler method (i.e. _L00) generates various Notify
events that need to be run to completion before the GPE is re-enabled.
In ACPI-CA, we queue an asynch callback at the same priority as a Notify
so that it will only run after all Notify handlers have completed.  The
callback re-enables the GPE afterwards.  We also changed the priority of
Notifies to be the same as GPEs, given the possibility that another GPE
could arrive before the Notifies have completed and we don't want it to
get queued ahead of the rest.

The ACPI-CA change was submitted by Alexey Starikovskiy (SUSE) and will
appear in a later release.  Special thanks to him for helping track this
bug down.

MFC after: 1 week
Tested by: jhb, Yousif Hassan <yousif / alumni.jmu.edu>

16 years agoBack when I committed the arm port, I've been asked to move
Olivier Houchard [Sat, 12 Jan 2008 21:11:43 +0000 (21:11 +0000)]
Back when I committed the arm port, I've been asked to move
memcpy/memset/memcmp and friends from libkern/arm to arm/arm/support.S, and so
I did, but in the process, I didn't add the appropriate copyrights.
This is a major oversight from me, and I apology to the NetBSD people for it.

MFC After: 1 day

16 years agoo inetd(8) requires wait/nowait column in inetd.conf for
Maxim Konovalov [Sat, 12 Jan 2008 21:09:48 +0000 (21:09 +0000)]
o inetd(8) requires wait/nowait column in inetd.conf for
ONC services as well.

PR: bin/119203
Submitted by: Peter Jeremy
MFC atfer: 1 week

16 years agoo .Dd.year++.
Maxim Konovalov [Sat, 12 Jan 2008 21:07:03 +0000 (21:07 +0000)]
o .Dd.year++.

16 years agoo Fix a typo. errx() -> err().
Maxim Konovalov [Sat, 12 Jan 2008 20:57:19 +0000 (20:57 +0000)]
o Fix a typo.  errx() -> err().

Submitted by: das

16 years agoo From the Problem Report: the TCP_DROP_SYNFIN kernel option is now
Maxim Konovalov [Sat, 12 Jan 2008 20:52:30 +0000 (20:52 +0000)]
o From the Problem Report: the TCP_DROP_SYNFIN kernel option is now
included in the kernel by default.  Remove reference to this option
from defaults/rc.conf and rc.conf(5).

PR: conf/119098
Submitted by: Beat Gaetzi
MFC after: 1 week

16 years agoo Plug a memory leak: do not forget to free asprintf(3) allocated
Maxim Konovalov [Sat, 12 Jan 2008 20:37:06 +0000 (20:37 +0000)]
o Plug a memory leak: do not forget to free asprintf(3) allocated
memory.

PR: bin/119608
Submitted by: peter.schuller
MFC after: 1 week

16 years agoMFp4 multiple queue support
Kip Macy [Sat, 12 Jan 2008 20:34:56 +0000 (20:34 +0000)]
MFp4 multiple queue support

16 years agoo Allow setkey(8) to recognize esp as a protocoal name for spdadd.
Maxim Konovalov [Sat, 12 Jan 2008 20:00:23 +0000 (20:00 +0000)]
o Allow setkey(8) to recognize esp as a protocoal name for spdadd.

PR: bin/107392
Submitted by: Eugene Grosbein
MFC after: 1 month

16 years agoo Do not claim vlan(4) uses mbuf_tags(9).
Maxim Konovalov [Sat, 12 Jan 2008 19:37:18 +0000 (19:37 +0000)]
o Do not claim vlan(4) uses mbuf_tags(9).

PR: docs/11959
Submitted by: Alexander Chernikov
MFC after: 1 week

16 years agoo Correct EasyMP3 EM732X usb 2.0 flash mp3 player revision:
Maxim Konovalov [Sat, 12 Jan 2008 19:24:55 +0000 (19:24 +0000)]
o Correct EasyMP3 EM732X usb 2.0 flash mp3 player revision:
it's 1.00 not 1.0.

PR: usb/96546
Submitted by: Eugene Grosbein
MFC after: 1 month

16 years agoo Add Moxa Technologies CP-104EL PCI Express 4 port Serial card.
Maxim Konovalov [Sat, 12 Jan 2008 19:14:29 +0000 (19:14 +0000)]
o Add Moxa Technologies CP-104EL PCI Express 4 port Serial card.

PR: kern/119515
Submitted by: Gavin Stone-Tolcher
MFC after: 1 month

16 years agoo Correct an info about "Firewalls and Internet Security" book: name,
Maxim Konovalov [Sat, 12 Jan 2008 19:02:09 +0000 (19:02 +0000)]
o Correct an info about "Firewalls and Internet Security" book: name,
authors list, ISBN, URLs.

PR: conf/119590
MFC after: 1 week

16 years agoPlug memory leaks that is observed when argbuf or argspc is used in the
Xin LI [Sat, 12 Jan 2008 00:54:47 +0000 (00:54 +0000)]
Plug memory leaks that is observed when argbuf or argspc is used in the
context.

Submitted by: Michal Vranek <michal.vranek seznam cz>
PR: bin/118380
MFC after: 1 month

16 years ago - Handle the case where interface from "middle" is missing by
Xin LI [Sat, 12 Jan 2008 00:11:26 +0000 (00:11 +0000)]
 - Handle the case where interface from "middle" is missing by
   more carefully inspecting the return value from sysctl(3). [1]
 - Use calloc instead of malloc+memset of zero.

Submitted by: Alexander Chernikov <admin su29 net> [1]
PR: bin/119581
MFC after: 2 weeks

16 years agoThe previous revision broke the case of reconnecting to a TCP NFS server
John Baldwin [Fri, 11 Jan 2008 23:57:39 +0000 (23:57 +0000)]
The previous revision broke the case of reconnecting to a TCP NFS server
via a new socket during an NFS operation as that reconnect takes place in
the context of an arbitrary thread with an arbitrary credential.  Ideally
we would like to use the mount point's credential for the entire process
of setting up the socket to connect to the NFS server.  Since some of the
APIs (sobind(), etc.) only take a thread pointer and infer the credential
from that instead of a direct credential, work around the problem by
temporarily changing the current thread's credential to that of the mount
point while connecting the socket and then reverting back to the original
credential when we are done.

Reviewed by: rwatson
Tested on: UDP, TCP, TCP with forced reconnect

16 years agoRemove from the default empty zone list zones that, unlike the others,
Doug Barton [Fri, 11 Jan 2008 22:41:21 +0000 (22:41 +0000)]
Remove from the default empty zone list zones that, unlike the others,
could theoretically be allocated one day.

16 years agoMFamd64 (everything possible up to 1.19; mainly the amd64 implementations
Bruce Evans [Fri, 11 Jan 2008 18:59:35 +0000 (18:59 +0000)]
MFamd64 (everything possible up to 1.19; mainly the amd64 implementations
of fpget*() and fpset*()).

The i386 fpget*() were efficient but a bit obfuscated (using macros
and a case statement to demultiplex them through a single inline).
The demultiplexing mainly gave smaller source code.

The i386 fpset*() were obfuscated in the same way and were very
inefficient due to the case statement not having enough cases or
complexity so all cases used the FP environment.

This also fixes a harmless bug in rev.1.12.  fpsetmask() extracted the
old value from the bit-field twice, but the doubled shift was harmless
since the shift count is 0.

All fp*() interfaces are now inline functions on i386.  They used to
be macros that call (a different set of) inline functions.  This is a
small ABI change which shouldn't cause problems since cases where
inlining fails (mainly -O0) only give (working) static functions.

16 years agoSeparate fpresetsticky() from the other fpset functions so that the
Bruce Evans [Fri, 11 Jan 2008 18:27:01 +0000 (18:27 +0000)]
Separate fpresetsticky() from the other fpset functions so that the
others can be replaced cleanly by the amd64 versions.   There is no
current amd64 version to merge, but there is an old one which is
similar.

Fix the following bugs in fpresetsticky():
- garbage args clobbered non-sticky bits in the status register
- the return value was usually garbage since it was masked with the
  arg instead of with the field selector.

Optimize fpresetsticky() to avoid using the environment as in
feclearexcept() (use only fnclex() if possible) and also to avoid
using fnclex() for null changes.  The second of these optimizations
might not be so good since its branch might cost more than it saves.

16 years agoMFamd64 1.15-1.18 (cosmetic changes, mainly to comments). The inline
Bruce Evans [Fri, 11 Jan 2008 17:54:20 +0000 (17:54 +0000)]
MFamd64 1.15-1.18 (cosmetic changes, mainly to comments).  The inline
functions haven't been cleaned up here because the amd64 cleanups
don't apply directly and the functions here will be merged or rewritten
later.

16 years agoFix fpset*() to not trap if there is a currently unmasked exception.
Bruce Evans [Fri, 11 Jan 2008 17:11:32 +0000 (17:11 +0000)]
Fix fpset*() to not trap if there is a currently unmasked exception.
Unmasked exceptions (which can be fixed up using fpset*() before they
trap) are very rare, especially on amd64 since SSE exceptions trap
synchronously, but I want to merge the faster amd64 implementations of
fpset*() back to i386 without introducing the bug on i386.

The i386 implementation has always avoided the trap automatically by
changing things using load/store of the FP environment, but this is
very slow.  Most changes only affect the control word, so they can
usually be done much more efficiently, and amd64 has always done this,
but loading the control word can trap.

This version use the fast method only in the usual case where it will
not trap.  This only costs a couple of integer instructions (including
one branch which I haven't optimized carefully yet) in the usual case,
but bloats the inlines a lot.  The inlines were already a bit too large
to handle both the FPU and SSE.

16 years agoThe wakeup() line from the rev. 1.319 is wrong and reintroduces
Konstantin Belousov [Fri, 11 Jan 2008 16:50:52 +0000 (16:50 +0000)]
The wakeup() line from the rev. 1.319 is wrong and reintroduces
a panic race on module unload. The wakeup() is internal to
kproc_exit/kthread_exit. The correct fix is to fix the msleep() in
detach to sleep on fdc->fdc_thread instead of &fdc->fdc_thread.

Noted and reviewed by: jhb
Pointy hat to: kib
MFC after: 1 week

16 years agolockmgr() function will return successfully when trying to work under
Attilio Rao [Fri, 11 Jan 2008 16:38:12 +0000 (16:38 +0000)]
lockmgr() function will return successfully when trying to work under
panic but it won't actually lock anything.
This can lead some paths to reach lockmgr_disown() with inconsistent
lock which will let trigger the relative assertions.

Fix those in order to recognize panic situation and to not trigger.

Reported by: pho
Submitted by: kib

16 years agoFix some style bugs:
Bruce Evans [Fri, 11 Jan 2008 14:11:46 +0000 (14:11 +0000)]
Fix some style bugs:
- fix a previous style fix: shifts should be in the correct direction even
  if they are null.
- restore a comment about namespace pollution from floatingpoint.h 1.12 and
  update it.
- remove unused namespace pollution FP_*REG.
- improve some comments.
- sort macro definitions for entry points.
- don't use underscores for macro args.

16 years agoAdd entries for manpages for pmc_x86_get_msr(3), kthread_create(9),
John Baldwin [Fri, 11 Jan 2008 13:41:31 +0000 (13:41 +0000)]
Add entries for manpages for pmc_x86_get_msr(3), kthread_create(9),
shm_open(3), and shm_unlink(3).

16 years agoFix unload of the fdc.ko:
Konstantin Belousov [Fri, 11 Jan 2008 11:53:04 +0000 (11:53 +0000)]
Fix unload of the fdc.ko:

Wakeup the thread doing the fdc_detach() when the fdc worker thread exits [1].
Write access to the write-protected floppy shall call device_unbusy() to
pair the device_busy() in the fd_access() [2].

PR: 116537 [1], 116539 [2]
MFC after: 1 week

16 years ago- Correct the range check in the double version to catch negative values
David Schultz [Fri, 11 Jan 2008 04:18:25 +0000 (04:18 +0000)]
- Correct the range check in the double version to catch negative values
  that would overflow.
- Style fixes and improved handling of NaNs suggested by bde.

16 years agoBe more aggressive about tx cleaning - when multiples streams were running the tx
Kip Macy [Fri, 11 Jan 2008 02:26:54 +0000 (02:26 +0000)]
Be more aggressive about tx cleaning - when multiples streams were running the tx
queue could fill up and stop getting cleaned.

16 years agoIf we're not running with multiqueue enabled we need to wait to acquire the
Kip Macy [Thu, 10 Jan 2008 23:51:34 +0000 (23:51 +0000)]
If we're not running with multiqueue enabled we need to wait to acquire the
rspq lock. Not doing so was causing us to skip re-enabling the interrupt.

- remove duplicate credits sysctl
- add support for dumping hardware context of the txq
- decrement budget_left when we break out of the process_responses loop

16 years agoWork around problems with the ppbus(4)'s interesting way of managing
John Baldwin [Thu, 10 Jan 2008 23:43:47 +0000 (23:43 +0000)]
Work around problems with the ppbus(4)'s interesting way of managing
interrupt handlers for child devices by adding a dummy handler that is
always present so that the underlying interrupt thread is always around
avoiding panics from stray interrupts.

MFC after: 3 days

16 years agoPass curthread to various socket routines (socreate(), sobind(), and
John Baldwin [Thu, 10 Jan 2008 23:36:00 +0000 (23:36 +0000)]
Pass curthread to various socket routines (socreate(), sobind(), and
soconnect()) instead of &thread0 when establishing a connection to the NFS
server.  Otherwise inconsistent credentials may be used when setting up
the NFS socket.

MFC after: 1 week
Reviewed by: rwatson

16 years agoAdd a feature_present(3) function which checks to see if a named kernel
John Baldwin [Thu, 10 Jan 2008 22:11:21 +0000 (22:11 +0000)]
Add a feature_present(3) function which checks to see if a named kernel
feature is present by checking the kern.features sysctl MIB.

MFC after: 1 week

16 years agoDon't zero td_runtime when billing thread CPU usage to the process;
Robert Watson [Thu, 10 Jan 2008 22:11:20 +0000 (22:11 +0000)]
Don't zero td_runtime when billing thread CPU usage to the process;
maintain a separate td_incruntime to hold unbilled CPU usage for
the thread that has the previous properties of td_runtime.

When thread information is requested using the thread monitoring
sysctls, export thread td_runtime instead of process rusage runtime
in kinfo_proc.

This restores the display of individual ithread and other kernel
thread CPU usage since inception in ps -H and top -SH, as well for
libthr user threads, valuable debugging information lost with the
move to try kthreads since they are no longer independent processes.

There is universal agreement that we should rewrite the process and
thread export sysctls, but this commit gets things going a bit
better in the mean time.  Likewise, there are resevations about the
continued validity of statclock given the speed of modern processors.

Reviewed by: attilio, emaste, jhb, julian

16 years agoAdd IFT_BRIDGE to the Ethernet section so l2 addresses are formatted correctly.
Andrew Thompson [Thu, 10 Jan 2008 20:53:13 +0000 (20:53 +0000)]
Add IFT_BRIDGE to the Ethernet section so l2 addresses are formatted correctly.

PR: bin/119542
Submitted by: Niki Denev

16 years agoRemoved a contraction.
Christian Brueffer [Thu, 10 Jan 2008 19:11:37 +0000 (19:11 +0000)]
Removed a contraction.

16 years agoFix non-SMP kernel regression introduced in the previous commit.
Jung-uk Kim [Thu, 10 Jan 2008 16:31:14 +0000 (16:31 +0000)]
Fix non-SMP kernel regression introduced in the previous commit.

Reviewed by: jkoshy

16 years agoDisable the check for errata AE18. On MacBooks (1,1 version) there's
Rui Paulo [Thu, 10 Jan 2008 16:09:22 +0000 (16:09 +0000)]
Disable the check for errata AE18.  On MacBooks (1,1 version) there's
no problem with coretemp and C3 state.

Approved by: njl (mentor)

16 years agoA real gem from freebsd-current@
Dag-Erling Smørgrav [Thu, 10 Jan 2008 14:51:24 +0000 (14:51 +0000)]
A real gem from freebsd-current@

16 years agoRemove "lock pushdown" todo item in comment -- I did that for 7.0.
Robert Watson [Thu, 10 Jan 2008 12:38:17 +0000 (12:38 +0000)]
Remove "lock pushdown" todo item in comment -- I did that for 7.0.

MFC after: 3 weeks

16 years agoCorrect typos in comments.
Robert Watson [Thu, 10 Jan 2008 12:29:12 +0000 (12:29 +0000)]
Correct typos in comments.

MFC after: 3 weeks

16 years agoWhen one tries to allocate memory with the M_WAITOK flag and we are short in
Pawel Jakub Dawidek [Thu, 10 Jan 2008 08:36:38 +0000 (08:36 +0000)]
When one tries to allocate memory with the M_WAITOK flag and we are short in
address space in kmem map call vm_lowmem event in a loop and wait a bit for
subsystems to reclaim some memory which in turn will reclaim address space as
well.

Note, this is a work-around.

Reviewed by: alc
Approved by: alc
MFC after: 3 days

16 years agoAdd support for selectively dumping the state of the hardware response queue.
Kip Macy [Thu, 10 Jan 2008 06:54:20 +0000 (06:54 +0000)]
Add support for selectively dumping the state of the hardware response queue.
Change ordering of a couple of types.

16 years agoshould always free when refcount is 1
Kip Macy [Thu, 10 Jan 2008 06:52:48 +0000 (06:52 +0000)]
should always free when refcount is 1

16 years agoquiet compiler complaint about unused parameters
Sam Leffler [Thu, 10 Jan 2008 04:28:26 +0000 (04:28 +0000)]
quiet compiler complaint about unused parameters

16 years agoquiet compiler complaint about comparing &v against NULL
Sam Leffler [Thu, 10 Jan 2008 04:26:44 +0000 (04:26 +0000)]
quiet compiler complaint about comparing &v against NULL

16 years agoMake sure that grog is linked to me, and that I (in *-ports*) am
Edwin Groothuis [Thu, 10 Jan 2008 01:25:40 +0000 (01:25 +0000)]
Make sure that grog is linked to me, and that I (in *-ports*) am
linked to the people I helped out, and add them to the overview
too.

Approved by: grog@

16 years agoBump FreeBSD_version in order to reflect vn_lock() prototype changes.
Attilio Rao [Thu, 10 Jan 2008 01:25:23 +0000 (01:25 +0000)]
Bump FreeBSD_version in order to reflect vn_lock() prototype changes.

16 years agoUpdate vn_lock() prototype accordingly with the 'thread' argument removal.
Attilio Rao [Thu, 10 Jan 2008 01:23:59 +0000 (01:23 +0000)]
Update vn_lock() prototype accordingly with the 'thread' argument removal.

16 years agovn_lock() is currently only used with the 'curthread' passed as argument.
Attilio Rao [Thu, 10 Jan 2008 01:10:58 +0000 (01:10 +0000)]
vn_lock() is currently only used with the 'curthread' passed as argument.
Remove this argument and pass curthread directly to underlying
VOP_LOCK1() VFS method. This modify makes the code cleaner and in
particular remove an annoying dependence helping next lockmgr() cleanup.
KPI results, obviously, changed.

Manpage and FreeBSD_version will be updated through further commits.

As a side note, would be valuable to say that next commits will address
a similar cleanup about VFS methods, in particular vop_lock1 and
vop_unlock.

Tested by: Diego Sardina <siarodx at gmail dot com>,
Andrea Di Pasquale <whyx dot it at gmail dot com>

16 years agohttp://www.freebsd.org/doc/en_US.ISO8859-1/articles/committers-guide/conventions...
Edwin Groothuis [Thu, 10 Jan 2008 01:10:31 +0000 (01:10 +0000)]
http://www.freebsd.org/doc/en_US.ISO8859-1/articles/committers-guide/conventions.html

4.1 point 6:

Add an entry for yourself to src/share/misc/committers-repository.dot,
where repository is either doc, ports or src, depending on the
commit privileges you obtained.

Approved by: grog@

16 years ago- make 9k clusters the default unless a tunable is set
Kip Macy [Thu, 10 Jan 2008 00:55:42 +0000 (00:55 +0000)]
- make 9k clusters the default unless a tunable is set
- return the error from cxgb_tx_common so that when an error is hit we dont
  spin forever in the taskq thread
- remove unused rxsd_ref
- simplify header_offset calculation for embedded mbuf headers
- fix memory leak by making sure that mbuf header initialization took place
- disable printf's for stalled queue, don't do offload/ctrl queue restart
  when tunnel queue is restarted
- add more diagnostic information about the txq state
- add facility to dump the actual contents of the hardware queue using sysctl

16 years agoRefine textdump comments slightly.
Robert Watson [Thu, 10 Jan 2008 00:26:47 +0000 (00:26 +0000)]
Refine textdump comments slightly.

MFC after: 3 months

16 years agoFix a buffer overflow in the processing of various SCSI commands. This likely
Scott Long [Wed, 9 Jan 2008 20:02:56 +0000 (20:02 +0000)]
Fix a buffer overflow in the processing of various SCSI commands.  This likely
fixes a an great number of weird problems that have been reported with this
driver.

PR: 119114
Submitted by: Jens Rehsack

16 years agoImprove -u (limit uid lookups) behavior.
David E. O'Brien [Wed, 9 Jan 2008 18:06:24 +0000 (18:06 +0000)]
Improve -u (limit uid lookups) behavior.

Submitted by: David Frascone <dave@frascone.com>
PR: 119490

16 years agoBring in fix for Binutils PR other/16240: Check for a failure return from
David E. O'Brien [Wed, 9 Jan 2008 16:46:14 +0000 (16:46 +0000)]
Bring in fix for Binutils PR other/16240: Check for a failure return from
cplus_demangle_type.  This is the rev 1.50-1.51 change.

Our addr2line, etc.. would crash if used on C++ code that contains
certain symbol types.  One example is
_ZN13PatternDriver23StringScalarDeleteValueC1ERKNS_25ConflateStringScalarValueERKNS_25AbstractStringScalarValueERKNS_12TemplateEnumINS_12pdcomplementELZNS_16complement_namesEELZNS_14COMPLEMENTENUMEEEE

16 years agoSimplify the ifdefs:
Bruce Evans [Wed, 9 Jan 2008 15:03:03 +0000 (15:03 +0000)]
Simplify the ifdefs:
- fix this to compile with C++ by casting ints to enums in a few places
  and by using the correct parameter type for _fpsetprec().  Remove
  __cplusplus ifdefs which disabled the buggy code.
- remove __CC_SUPPORTS___INLINE ifdefs.  `__inline' vs `inline', and either
  of these #defined away, are supposed to be handled by very old ifdefs
  in <sys/cdefs.h>.  Thus the __CC_SUPPORTS___INLINE macro is not needed
  here (or anywhere else that it used).  It is less needed here than in
  most places, since this file is userland-only and userland is far from
  supporting INTEL_COMPILER.  The __CC_SUPPORTS___INLINE__ macro which
  was used here is even less needed.  It is to support spelling `inline'
  as `__inline__' instead of the usual spelling `__inline'.

Fix some style bugs that I missed in the previous commit (remove unused
asms and sort more variables).

16 years agoFix some style bugs (mainly, use explicit shifts when accessing bit-fields
Bruce Evans [Wed, 9 Jan 2008 13:35:31 +0000 (13:35 +0000)]
Fix some style bugs (mainly, use explicit shifts when accessing bit-fields
even if the shift count happens to be 0, sort declarations, and spell
__inline normally).

16 years agoImprove some comments.
Bruce Evans [Wed, 9 Jan 2008 10:42:47 +0000 (10:42 +0000)]
Improve some comments.

16 years agoRevise the TECHNICAL DETAILS section.
Ruslan Ermilov [Wed, 9 Jan 2008 09:38:53 +0000 (09:38 +0000)]
Revise the TECHNICAL DETAILS section.

Reviewed by: skv

16 years agomake nqsets a uint32_t so that sysctl will work
Kip Macy [Wed, 9 Jan 2008 08:12:24 +0000 (08:12 +0000)]
make nqsets a uint32_t so that sysctl will work
add 2 fields for allowing queue dumping