]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
18 years agoReduce diffs from i386/i386/machdep.c
Yoshihiro Takahashi [Sun, 16 Oct 2005 12:33:14 +0000 (12:33 +0000)]
Reduce diffs from i386/i386/machdep.c

18 years agoFor AMD processors, nullify CPUID.HTT. FreeBSD has no need for the
David E. O'Brien [Sun, 16 Oct 2005 08:58:27 +0000 (08:58 +0000)]
For AMD processors, nullify CPUID.HTT.  FreeBSD has no need for the
information it conveys, and it is only confusing people.
This fixes incorrect output in the previous commit.

18 years agoThe BCM5401 dspcode load on media changes also applies
Peter Grehan [Sun, 16 Oct 2005 05:29:14 +0000 (05:29 +0000)]
The BCM5401 dspcode load on media changes also applies
to the 100/1000 BCM5400 phy. This fixes the problem with
the GEM port not syncing up on Sawtooth G4's.

Obtained from:  NetBSD
Reported by:    Ben Rosengart <ben + freebsd org at narcissus net>

18 years agoForced commit to note that the previous commit referred to r1.67, not r1.66
Kris Kennaway [Sun, 16 Oct 2005 03:58:10 +0000 (03:58 +0000)]
Forced commit to note that the previous commit referred to r1.67, not r1.66

Pointed out by: bde

18 years agoMempage should be 1024 not 4096. This fixes the multiple 16-bit cards
Warner Losh [Sun, 16 Oct 2005 03:58:06 +0000 (03:58 +0000)]
Mempage should be 1024 not 4096.  This fixes the multiple 16-bit cards
failing to work in, eg, Soekris boxes.

This is a critcal fix for 6.0.

MFC After: ASAP

18 years agoAdd new option `q', which makes second stage loader quiet unless autoboot
Maxim Sobolev [Sun, 16 Oct 2005 01:55:35 +0000 (01:55 +0000)]
Add new option `q', which makes second stage loader quiet unless autoboot
is disabled or fails.

MFC after: 1 week

18 years agoFix compiling.
David Xu [Sat, 15 Oct 2005 22:57:39 +0000 (22:57 +0000)]
Fix compiling.

18 years agoadded a missing unlock
SUZUKI Shinsuke [Sat, 15 Oct 2005 08:49:49 +0000 (08:49 +0000)]
added a missing unlock

Submitted by: JINMEI Tatuya
MFC After: 1 day

18 years agoThe kernel accepts sockaddrs with len set to zero for sockaddr_in
Alfred Perlstein [Sat, 15 Oct 2005 06:17:29 +0000 (06:17 +0000)]
The kernel accepts sockaddrs with len set to zero for sockaddr_in
types, so refactor the code here to grab them when length is zero.

18 years agoregen after recvmsg, recvfrom, sendmsg
Paul Saab [Sat, 15 Oct 2005 05:57:34 +0000 (05:57 +0000)]
regen after recvmsg, recvfrom, sendmsg

18 years agoImplement the 32bit versions of recvmsg, recvfrom, sendmsg
Paul Saab [Sat, 15 Oct 2005 05:57:06 +0000 (05:57 +0000)]
Implement the 32bit versions of recvmsg, recvfrom, sendmsg

Partially obtained from: jhb

18 years agoregen for clock_gettime, clock_settime, clock_getres
Paul Saab [Sat, 15 Oct 2005 02:54:39 +0000 (02:54 +0000)]
regen for clock_gettime, clock_settime, clock_getres

18 years agoImplement 32bit wrappers for clock_gettime, clock_settime, and
Paul Saab [Sat, 15 Oct 2005 02:54:18 +0000 (02:54 +0000)]
Implement 32bit wrappers for clock_gettime, clock_settime, and
clock_getres.

18 years agoregen
Paul Saab [Sat, 15 Oct 2005 02:40:34 +0000 (02:40 +0000)]
regen

18 years agoCorrect the prototype for freebsd32_nanosleep and use the proper
Paul Saab [Sat, 15 Oct 2005 02:40:10 +0000 (02:40 +0000)]
Correct the prototype for freebsd32_nanosleep and use the proper
size when copying struct timespec32 in and out.

18 years agoFix compiling.
David Xu [Sat, 15 Oct 2005 02:09:43 +0000 (02:09 +0000)]
Fix compiling.

18 years agoCorrect few MSR addresses.
Jung-uk Kim [Sat, 15 Oct 2005 00:44:56 +0000 (00:44 +0000)]
Correct few MSR addresses.

PR: amd64/85852
Submitted by: Nate Eldredge <nge at cs dot hmc dot edu>

18 years agompsafevm has been stable and defaulted to 1 on sparc64 for over 6 months,
Kris Kennaway [Fri, 14 Oct 2005 23:56:13 +0000 (23:56 +0000)]
mpsafevm has been stable and defaulted to 1 on sparc64 for over 6 months,
so we are ready for mpsafevfs=1 by default on sparc64 too.  I have been
running this on all my sparc64 machines for over 6 months, and have not
encountered MD problems.

MFC after: 1 week

18 years agoRevise the manpage to a certain extent, mostly with respect to
Yaroslav Tykhiy [Fri, 14 Oct 2005 23:54:20 +0000 (23:54 +0000)]
Revise the manpage to a certain extent, mostly with respect to
make's processing of top-level and included makefiles.  Point
out at make.conf(5) and __MAKE_CONF when telling about sys.mk.

Reviewed by: ru

18 years agoLet modules use the kernel's opt_*.h files if built along with
Yaroslav Tykhiy [Fri, 14 Oct 2005 23:30:17 +0000 (23:30 +0000)]
Let modules use the kernel's opt_*.h files if built along with
the kernel by wrapping all targets for fake opt_*.h files in
.if defined(KERNBUILDDIR).  Thus, such fake files won't be
created at all if modules are built with the kernel.

Some modules undergo cleanup like removing unused or unneeded
options or .h files, without which they wouldn't build this way
or the other.

Reviewed by: ru
Tested by: no binary changes in modules built alone
Tested on: i386 sparc64 amd64

18 years agoIn preparation for making the modules actually use opt_*.h files
Yaroslav Tykhiy [Fri, 14 Oct 2005 23:17:45 +0000 (23:17 +0000)]
In preparation for making the modules actually use opt_*.h files
provided in the kernel build directory, fix modules that were
failing to build this way due to not quite correct kernel option
usage.  In particular:

ng_mppc.c uses two complementary options, both of which are listed
in sys/conf/files.  Ideally, there should be a separate option for
including ng_mppc.c in kernel build, but now only
NETGRAPH_MPPC_ENCRYPTION is usable anyway, the other one requires
proprietary files.

nwfs and smbfs were trying to ensure they were built with proper
network components, but the check was rather questionable.

Discussed with: ru

18 years ago- Print number of physical/logical cores and more CPUID info.
Jung-uk Kim [Fri, 14 Oct 2005 22:52:01 +0000 (22:52 +0000)]
- Print number of physical/logical cores and more CPUID info.
- Add newer CPUID definitions for future use.

Many thanks to Mike Tancsa <mike at sentex dot net> for providing test
cases for Intel Pentium D and AMD Athlon 64 X2.

Approved by: anholt (mentor)

18 years agoClose a race in the ufs_lookup() code that handles the ISDOTDOT
Don Lewis [Fri, 14 Oct 2005 22:13:33 +0000 (22:13 +0000)]
Close a race in the ufs_lookup() code that handles the ISDOTDOT
case by saving the value of dp->i_ino before unlocking the vnode
for the current directory and passing the saved value to VFS_VGET().

Without this change, another thread can overwrite dp->i_ino after
the current directory is unlocked, causing  ufs_lookup() to lock
and return the wrong vnode in place of the vnode for its parent
directory.  A deadlock can occur if dp->i_ino was changed to a
subdirectory of the current directory because the root to leaf vnode
lock ordering will be violated.  A vnode lock can be leaked if
dp->i_ino was changed to point to the current directory, which
causes the current vnode lock for the current directory to be
recursed, which confuses lookup() into calling vrele() when it
should be calling vput().

The probability of this bug being triggered seems to be quite low
unless the sysctl variable debug.vfscache is set to 0.

Reviewed by: jhb
MFC after: 2 weeks

18 years agoMerge over the remaining changes from i386 of the ksiginfo_t changes so
John Baldwin [Fri, 14 Oct 2005 22:03:39 +0000 (22:03 +0000)]
Merge over the remaining changes from i386 of the ksiginfo_t changes so
that this compiles.

Pointy hat to: davidxu

18 years agoMake four more functions static that were missed in the last commit.
Andrew Thompson [Fri, 14 Oct 2005 20:57:02 +0000 (20:57 +0000)]
Make four more functions static that were missed in the last commit.

18 years agoAdd a default value for VM_BCACHE_SIZE_MAX of 400MB. This is copied from
Kris Kennaway [Fri, 14 Oct 2005 20:31:12 +0000 (20:31 +0000)]
Add a default value for VM_BCACHE_SIZE_MAX of 400MB.  This is copied from
amd64, and is a factor of 3 less than the value previously auto-sized on
a 12GB machine, which would cause an overflow in calculations involving the
maxbcache int, causing bufinit() to loop forever at boot.

Reviewed by: mlaier, peter

18 years agoThe signal code is now an int rather than a long, so update debug printfs.
John Baldwin [Fri, 14 Oct 2005 20:22:57 +0000 (20:22 +0000)]
The signal code is now an int rather than a long, so update debug printfs.

18 years agoPartially revert revision 1.66, which contained a change that did not
Kris Kennaway [Fri, 14 Oct 2005 19:15:10 +0000 (19:15 +0000)]
Partially revert revision 1.66, which contained a change that did not
correspond to the commit log.  It changed the maxswzone and maxbcache
parameters from int to long, without changing the extern definitions
in <sys/buf.h>.

In fact it's a good thing it did not, because other parts of the system
are not yet ready for this, and on large-memory sparc machines it causes
severe filesystem damage if you try.

The worst effect of the change was that the tunables controlling the
above variables stopped working.  These were necessary to allow such
large sparc64 machines (with >12GB RAM) to boot, since sparc64 did not
set a hard-coded upper limit on these parameters and they ended
up overflowing an int, causing an infinite loop at boot in bufinit().

Reviewed by: mlaier

18 years agoOnly allow the sk(4) driver to attach to revision 2 of the LinkSys EG1032
John Baldwin [Fri, 14 Oct 2005 18:51:30 +0000 (18:51 +0000)]
Only allow the sk(4) driver to attach to revision 2 of the LinkSys EG1032
cards and teach the re(4) driver to attach to revision 3 cards.

Submitted by: Fredrik Lindberg fli+freebsd-current at shapeshifter dot se
MFC after: 2 weeks
Reviewed by: imp, mdodd

18 years agoWhitespace.
John Baldwin [Fri, 14 Oct 2005 18:36:49 +0000 (18:36 +0000)]
Whitespace.

18 years agoRemove the sx(4) driver at the request of the author. The author
John Baldwin [Fri, 14 Oct 2005 18:24:58 +0000 (18:24 +0000)]
Remove the sx(4) driver at the request of the author.  The author
originally wrote it for 4.x and hasn't really had the time to fully update
it to 5.x and later.  Also, the author doesn't use the hardware anymore as
well.  If someone does need this driver they can always resurrect it from
the Attic.

Requested by: Frank Mayhar frank at exit dot com

18 years agoThe snd_via8233(4) support on amd64 isn't something new. Remove it.
Joel Dahl [Fri, 14 Oct 2005 18:12:18 +0000 (18:12 +0000)]
The snd_via8233(4) support on amd64 isn't something new.  Remove it.

Reviewed by: hrs

18 years agoChange the userland atomic operations on arm to use memory operands for
John Baldwin [Fri, 14 Oct 2005 18:07:45 +0000 (18:07 +0000)]
Change the userland atomic operations on arm to use memory operands for
the modified memory rather than using register operands that held a pointer
to the memory.  The biggest effect is that we now correctly tell the
compiler that these functions change the memory that these functions
modify.

Reviewed by: cognet

18 years agoSort ath_rate_* entries. Mark ath_rate_sample as the desired algorithm.
Ruslan Ermilov [Fri, 14 Oct 2005 17:22:28 +0000 (17:22 +0000)]
Sort ath_rate_* entries.  Mark ath_rate_sample as the desired algorithm.

Discussed with: sam

18 years agoFix installworld breakage. <sigh>
Tim Kientzle [Fri, 14 Oct 2005 16:32:50 +0000 (16:32 +0000)]
Fix installworld breakage. <sigh>

expr and printf are not available during installworld, so
use /bin/sh arithmetic expansion instead of expr and simply
give up on vanity formatting. ;-)

18 years agoBetter translation: language improvements, typos, etc.
Xin LI [Fri, 14 Oct 2005 16:25:53 +0000 (16:25 +0000)]
Better translation: language improvements, typos, etc.

Submitted by:   BSDunix at newsmth org,
carplee at newsmth org,
ThunderW at newsmth org (in part)
MFC After: 1 days

18 years agoFix a regression that prevented PERC3 hardware from working. Apparently they
Scott Long [Fri, 14 Oct 2005 16:22:45 +0000 (16:22 +0000)]
Fix a regression that prevented PERC3 hardware from working.  Apparently they
do not support the GETINFO immediate command, unlike just about every other
variant of the hardware.  Also document some magic values and fix some minor
nearby whitespace.

MFC After: 3 days

18 years agoStyle: move .PATH to where it belongs.
Ruslan Ermilov [Fri, 14 Oct 2005 15:50:22 +0000 (15:50 +0000)]
Style: move .PATH to where it belongs.

18 years agoRemove redundant (in FreeBSD) include.
Ruslan Ermilov [Fri, 14 Oct 2005 15:45:38 +0000 (15:45 +0000)]
Remove redundant (in FreeBSD) include.

18 years agoStyle: move .PATH to where it belongs.
Ruslan Ermilov [Fri, 14 Oct 2005 15:41:09 +0000 (15:41 +0000)]
Style: move .PATH to where it belongs.

18 years agoRemove redundant include.
Ruslan Ermilov [Fri, 14 Oct 2005 15:26:23 +0000 (15:26 +0000)]
Remove redundant include.

18 years agoGet rid of duplicate -I's in CFLAGS.
Ruslan Ermilov [Fri, 14 Oct 2005 15:13:36 +0000 (15:13 +0000)]
Get rid of duplicate -I's in CFLAGS.

18 years agoSort SUBDIR and surrounding definitions.
Ruslan Ermilov [Fri, 14 Oct 2005 15:12:45 +0000 (15:12 +0000)]
Sort SUBDIR and surrounding definitions.

18 years agoChange I-O data WB-B11/CFZ entry to ELSA XI330 as the former is a
Warner Losh [Fri, 14 Oct 2005 15:07:13 +0000 (15:07 +0000)]
Change I-O data WB-B11/CFZ entry to ELSA XI330 as the former is a
rebadged version of the latter.

18 years agoAdd ELSA XI330 product. This is rebadged and sold as SMC 2532W-B and
Warner Losh [Fri, 14 Oct 2005 15:06:16 +0000 (15:06 +0000)]
Add ELSA XI330 product.  This is rebadged and sold as SMC 2532W-B and
I/O Data also resells it.  Add an alternative airvast an100 id.

18 years agoAdd sigqueue function prototype.
David Xu [Fri, 14 Oct 2005 13:12:27 +0000 (13:12 +0000)]
Add sigqueue function prototype.

18 years agoRegen for sigqueue syscall.
David Xu [Fri, 14 Oct 2005 12:56:28 +0000 (12:56 +0000)]
Regen for sigqueue syscall.

18 years ago1. Change prototype of trapsignal and sendsig to use ksiginfo_t *, most
David Xu [Fri, 14 Oct 2005 12:43:47 +0000 (12:43 +0000)]
1. Change prototype of trapsignal and sendsig to use ksiginfo_t *, most
   changes in MD code are trivial, before this change, trapsignal and
   sendsig use discrete parameters, now they uses member fields of
   ksiginfo_t structure. For sendsig, this change allows us to pass
   POSIX realtime signal value to user code.

2. Remove cpu_thread_siginfo, it is no longer needed because we now always
   generate ksiginfo_t data and feed it to libpthread.

3. Add p_sigqueue to proc structure to hold shared signals which were
   blocked by all threads in the proc.

4. Add td_sigqueue to thread structure to hold all signals delivered to
   thread.

5. i386 and amd64 now return POSIX standard si_code, other arches will
   be fixed.

6. In this sigqueue implementation, pending signal set is kept as before,
   an extra siginfo list holds additional siginfo_t data for signals.
   kernel code uses psignal() still behavior as before, it won't be failed
   even under memory pressure, only exception is when deleting a signal,
   we should call sigqueue_delete to remove signal from sigqueue but
   not SIGDELSET. Current there is no kernel code will deliver a signal
   with additional data, so kernel should be as stable as before,
   a ksiginfo can carry more information, for example, allow signal to
   be delivered but throw away siginfo data if memory is not enough.
   SIGKILL and SIGSTOP have fast path in sigqueue_add, because they can
   not be caught or masked.
   The sigqueue() syscall allows user code to queue a signal to target
   process, if resource is unavailable, EAGAIN will be returned as
   specification said.
   Just before thread exits, signal queue memory will be freed by
   sigqueue_flush.
   Current, all signals are allowed to be queued, not only realtime signals.

Earlier patch reviewed by: jhb, deischen
Tested on: i386, amd64

18 years agoFix parsing of mdmfs(8) option "-w <user>:<group>" in case <user> or
Ralf S. Engelschall [Fri, 14 Oct 2005 11:21:21 +0000 (11:21 +0000)]
Fix parsing of mdmfs(8) option "-w <user>:<group>" in case <user> or
<group> is a numeric user/group ID instead of a user/group name (as
explicitly intended to be allowed by both the manual page and the
implementation).

Before this fix, mdmfs(8) aborted:

| # mdmfs -s 32m -w 0:0 md /var/tmp/foo
| Assertion failed: (mip->mi_have_uid), function extract_ugid, file /usr/src/sbin/mdmfs/mdmfs.c, line 555.
| Abort trap (core dumped)

The "mi_have_[ug]id" fields were only set in case a name lookup was
successful. Instead they also have to be set in case the string to
integer conversion was successful.

Additionally, as a result of this fix, two assertions at the end of
the function are now always true and hence can be just be removed. It
is guarrantied that both the UID and the GID are set when the function
returns regularily, else it would have been already bailed out with
usage()/exit(3) or errx(3) before.

Spotted by: Christoph Schug <chris@schug.net>
MFC after: 3 days

18 years agoFrom the PR:
Gleb Smirnoff [Fri, 14 Oct 2005 11:00:15 +0000 (11:00 +0000)]
From the PR:

  The receive function em_process_receive_interrupts() unlocks the
  adapter while ether_input() processes the packet, and then locks
  it back. In the meantime, em_init() may be called, either from
  em_watchdog() from softclock interrupt or from the ifconfig(8)
  program. The em_init() resets the card, in particular it sets
  adapter->next_rx_desc_to_check to 0 and resets hardware RX Head
  and Tail descriptor pointers. The loop in
  em_process_receive_interrupts() does not expect these things to
  change, and a mess may result.

This fixes long wedges of em(4) interfaces receive part under high
load and IP fastforwarding enabled.

PR: kern/87418
Submitted by: Dmitrij Tejblum <tejblum yandex-team.ru>

18 years agoChange most of the bridge and stp funtions to static. This has highlighted
Andrew Thompson [Fri, 14 Oct 2005 10:38:12 +0000 (10:38 +0000)]
Change most of the bridge and stp funtions to static. This has highlighted
that the following funtions are not used, wrap in '#ifdef noused' for the
moment.

 bstp_enable_change_detection
 bstp_disable_change_detection
 bstp_set_bridge_priority
 bstp_set_port_priority
 bstp_set_path_cost

18 years agoCleanup from __FreeBSD_version.
Gleb Smirnoff [Fri, 14 Oct 2005 10:34:46 +0000 (10:34 +0000)]
Cleanup from __FreeBSD_version.

18 years agoNew release notes:
Yaroslav Tykhiy [Fri, 14 Oct 2005 09:57:08 +0000 (09:57 +0000)]
New release notes:
ifconfig(8) -tunnel parameter,
interface argument to ifconfig(8) -vlandev deprecated.

The changes already are in RELENG_6, too.

MFC after: 3 days

18 years agoMFP4: Minor tweaks.
Warner Losh [Fri, 14 Oct 2005 04:48:11 +0000 (04:48 +0000)]
MFP4: Minor tweaks.

18 years agoMerge in WIP from p4 for supporting ISA pccard bridges conforming to
Warner Losh [Fri, 14 Oct 2005 04:47:18 +0000 (04:47 +0000)]
Merge in WIP from p4 for supporting ISA pccard bridges conforming to
the ExCA spec, and close cousins:

o Write an activate routine that works.
o merge a couple of items from oldcard before they are lost
o write a deactivate routine

I suspect we're still a ways away from having this work, but maybe for
6.1/5.5?

18 years agoAdd ksiginfo_t which is a wrapper of siginfo_t but allows us to carry
David Xu [Fri, 14 Oct 2005 03:36:44 +0000 (03:36 +0000)]
Add ksiginfo_t which is a wrapper of siginfo_t but allows us to carry
more information which should not be in siginfo_t.

Reviewed by: jhb, deischen

18 years agoAdd POSIX siginfo_t's si_code, this is for upcoming POSIX realtime signal
David Xu [Fri, 14 Oct 2005 03:01:14 +0000 (03:01 +0000)]
Add POSIX siginfo_t's si_code, this is for upcoming POSIX realtime signal
support in kernel.

Earlier patch reviewed by: jhb, deischen

18 years agoFurther clean up the bridge hooks in if_ethersubr.c and ng_ether.c
Andrew Thompson [Fri, 14 Oct 2005 02:38:47 +0000 (02:38 +0000)]
Further clean up the bridge hooks in if_ethersubr.c and ng_ether.c

- move the function pointer definitions to if_bridgevar.h
- move most of the logic to the new BRIDGE_INPUT and BRIDGE_OUTPUT macros
- remove unneeded functions from if_bridgevar.h and sort a little.

18 years agoRestore the UP optimization to reduce the number of TLB invalidations. The
Alan Cox [Thu, 13 Oct 2005 23:42:12 +0000 (23:42 +0000)]
Restore the UP optimization to reduce the number of TLB invalidations.  The
previous revision only restored the MP optimization.

Describe the optimization strategy for TLB invalidations in a comment.

Reviewed by: ups@
MFC after: 3 days

18 years agoFrom 101 ways to panic your kernel.
Andrew Thompson [Thu, 13 Oct 2005 23:05:55 +0000 (23:05 +0000)]
From 101 ways to panic your kernel.

Use bridge_ifdetach() to notify the bridge that a member has been detached. The
bridge can then remove it from its interface list and not try to send out via a
dead pointer.

18 years agoIf we can't probe the RTL80x9 for some reason, fall back to probing it
Warner Losh [Thu, 13 Oct 2005 22:12:34 +0000 (22:12 +0000)]
If we can't probe the RTL80x9 for some reason, fall back to probing it
as a Novell NE-2000.  This is necessary for unpatched qemu working
correctly.  qemu claims to be a RTL8029, but doesn't implement the
RTL8029 specific registers at this time.  I've created patches for
that, but there's no reason we can't use qemu's emulation w/o these
patches.  This should make life easier for those folks that boot
FreeBSD via qemu.

18 years agoo Fix probing of rtl80x9 parts. We shouldn't be calling
Warner Losh [Thu, 13 Oct 2005 22:06:02 +0000 (22:06 +0000)]
o Fix probing of rtl80x9 parts.  We shouldn't be calling
  ed_probe_generic8390 where we're calling it.  It will be done as part
  of ed_probe_Novel_generic after things are setup in a way that
  ed_probe_generic8390 will grok.
o Fix operator precedence botch that causes a panic when setting the media
  type for 10baseT connections.
o Save the type of device so that it prints with the rest of the probe.

# this should make it work with qemu again, but only if it has my patches
# to actually implement the RTL8029 specific registers.

18 years agoConsolidate two adjacent conditional blocks
Julian Elischer [Thu, 13 Oct 2005 21:48:27 +0000 (21:48 +0000)]
Consolidate two adjacent conditional blocks
I actually believe the code in question should be elsewhere (in the preceding
function).

MFC after: 1 week

18 years agoIn detach method, move if_free() after bus_teardown_intr().
Ruslan Ermilov [Thu, 13 Oct 2005 21:11:20 +0000 (21:11 +0000)]
In detach method, move if_free() after bus_teardown_intr().

18 years agoMinor tweaks and fixups:
John Baldwin [Thu, 13 Oct 2005 20:22:07 +0000 (20:22 +0000)]
Minor tweaks and fixups:
- Use device_printf() and if_printf() and remove nge_unit.
- Use callout_init_mtx() and remove nge_tick_locked() as nge_tick() is now
  always called with the driver lock held.
- Use M_ZERO to contigmalloc() when allocating nge_ldata.  It was possible
  for the random garbage to be used in certain cases otherwise.
- Cleanup attach error handling including no longer leaking nge_ldata.
- Add locking to the ifmedia callouts.
- Lock accesses to if_hwassist and if_capenable in nge_ioctl().

Submitted by: Yuriy N. Shkandybin jura at networks dot ru (1, 3, 4)
Tested by: Yuriy N. Shkandybin jura at networks dot ru
MFC after: 3 days

18 years agoAdd support for ATI IXP[234]00.
Joel Dahl [Thu, 13 Oct 2005 18:28:26 +0000 (18:28 +0000)]
Add support for ATI IXP[234]00.

18 years agofixed a crush when either -lh or -ls option is used.
Hajimu UMEMOTO [Thu, 13 Oct 2005 18:12:46 +0000 (18:12 +0000)]
fixed a crush when either -lh or -ls option is used.

Obtained from: KAME

18 years agosetkey(8) is not WARNS=2 compliant, yet.
Hajimu UMEMOTO [Thu, 13 Oct 2005 17:57:34 +0000 (17:57 +0000)]
setkey(8) is not WARNS=2 compliant, yet.

18 years agoRemove a stale comment.
Ruslan Ermilov [Thu, 13 Oct 2005 17:26:14 +0000 (17:26 +0000)]
Remove a stale comment.

18 years agoLanguage nits.
Xin LI [Thu, 13 Oct 2005 17:25:28 +0000 (17:25 +0000)]
Language nits.

Submitted by: nukq at newsmth org, me

18 years agoMention that hptmv(4) actually works for the amd64 platform as well.
Xin LI [Thu, 13 Oct 2005 17:20:06 +0000 (17:20 +0000)]
Mention that hptmv(4) actually works for the amd64 platform as well.

Suggested by: Mikhail Teterin [mi+mx aldan algebra com]
Submitted by: me
PR: docs/87302

18 years ago- Clarify when the softc is allocated. It is allocated when a device is
John Baldwin [Thu, 13 Oct 2005 16:57:53 +0000 (16:57 +0000)]
- Clarify when the softc is allocated.  It is allocated when a device is
  probed and attached, not on the first call to device_get_softc().
- Add a cross reference to DEVICE_PROBE regarding the caveats of using the
  softc in a driver's probe routine.
- Fix a grammar bogon.

PR: docs/87176 (1)
Submitted by: Devon H. O'Dell dodell at offmyserver dot com (1)
MFC after: 3 days

18 years ago- When bintime was added to this manpage, standalone .Nm changed from
John Baldwin [Thu, 13 Oct 2005 16:01:28 +0000 (16:01 +0000)]
- When bintime was added to this manpage, standalone .Nm changed from
  microtime to bintime.  However, one standaline .Nm wasn't changed, and as
  a result, the manpage claimed that bintime was added in both 5.0 and 3.0.
  Fix by listing microtime explicitly.
- Fix a grammar bogon.

PR: docs/87147 (1)
Submitted by: Matthew Luckie (1)
MFC after: 3 days

18 years ago- When flushing node input queue, check whether item has a callback. If
Gleb Smirnoff [Thu, 13 Oct 2005 11:55:50 +0000 (11:55 +0000)]
- When flushing node input queue, check whether item has a callback. If
  it does, then call it suppling ENOENT as error value.
- Add assert, that helped to catch the above error.

18 years agoClean up the if_bridge hooks a bit in if_ethersubr.c and ng_ether.c, move
Andrew Thompson [Thu, 13 Oct 2005 09:43:30 +0000 (09:43 +0000)]
Clean up the if_bridge hooks a bit in if_ethersubr.c and ng_ether.c, move
the broadcast/multicast test to bridge_input().

Requested by: glebius

18 years agoMinor cleanups, no functional changes
Søren Schmidt [Thu, 13 Oct 2005 08:05:21 +0000 (08:05 +0000)]
Minor cleanups, no functional changes

18 years ago1) Use GNU libtool to build shared libraries on non-FreeBSD
Tim Kientzle [Thu, 13 Oct 2005 05:51:38 +0000 (05:51 +0000)]
1) Use GNU libtool to build shared libraries on non-FreeBSD
   systems (or on FreeBSD systems when using ports).

2) Overhaul the versioning logic.  In particular,
   SHLIB_MAJOR number is now computed as "major+minor",
   which ensures library versions are the same for
   the FreeBSD build system and the portable
   libtool/autoconf/automake build system.

18 years agoRestore optimizations to reduce TLB shootdowns.
Stephan Uphoff [Thu, 13 Oct 2005 03:55:25 +0000 (03:55 +0000)]
Restore optimizations to reduce TLB shootdowns.
Alan Cox pointed out that they are really useful for
sendfile().

MFC after: 3 days

18 years agoFix tinderbox box by removing incomplete/bad spl usage. Proper giant free
Doug Ambrisko [Wed, 12 Oct 2005 22:33:22 +0000 (22:33 +0000)]
Fix tinderbox box by removing incomplete/bad spl usage.  Proper giant free
locking is required in for aio.

Pointed out by: imp

18 years agoFirst start rc.d/ipsec and then rc.d/mountcritremote, so we can mount
Pawel Jakub Dawidek [Wed, 12 Oct 2005 22:14:44 +0000 (22:14 +0000)]
First start rc.d/ipsec and then rc.d/mountcritremote, so we can mount
NFS file system over IPsec.

Suggested by: Tomasz Pi³at <tomasz.pilat@axelspringer.pl>

18 years agosetkey(8) was moved to /sbin/.
Pawel Jakub Dawidek [Wed, 12 Oct 2005 21:43:47 +0000 (21:43 +0000)]
setkey(8) was moved to /sbin/.

18 years agosetkey(8) was repo-copied from usr.sbin/ to sbin/.
Pawel Jakub Dawidek [Wed, 12 Oct 2005 21:40:41 +0000 (21:40 +0000)]
setkey(8) was repo-copied from usr.sbin/ to sbin/.
This will allow for NFS mount of /usr over IPsec.

Discussed on: arch@

18 years agoMFen:
Xin LI [Wed, 12 Oct 2005 20:30:39 +0000 (20:30 +0000)]
MFen:
errata/article.sgml: 1.72 -> 1.73.2.1
hardware/amd64/proc-amd64.sgml: 1.3.4.1 -> 1.5.2.1 [dummy merge]
hardware/common/artheader.sgml: 1.6 -> 1.7
hardware/common/dev.sgml: 1.277 -> 1.282.2.3
hardware/common/intro.sgml: 1.6 -> 1.7
installation/common/artheader.sgml: 1.5 -> 1.6
installation/common/trouble.sgml: 1.17 -> 1.19
installation/common/upgrade.sgml: 1.12 -> 1.13
readme/article.sgml:        1.31.2.2 -> 1.37
relnotes/common/new.sgml: 1.849 -> 1.883.2.7
Makefile: 1.3 -> 1.7

Follow the English revision to remove early adopter guide.

This is intended for a RELENG_6 and RELENG_6_0 merge.

Obtained from: The FreeBSD Simplified Chinese Project (delphij)

18 years agoAdd support for the ATI IXP[234]00 series chipsets.
Søren Schmidt [Wed, 12 Oct 2005 20:00:26 +0000 (20:00 +0000)]
Add support for the ATI IXP[234]00 series chipsets.

HW donated by: sentex

18 years agoChange the reference counting to count the number of cloned interfaces for each
Andrew Thompson [Wed, 12 Oct 2005 19:52:16 +0000 (19:52 +0000)]
Change the reference counting to count the number of cloned interfaces for each
cloner. This ensures that ifc->ifc_units is not prematurely freed in
if_clone_detach() before the clones are destroyed, resulting in memory modified
after free. This could be triggered with if_vlan.

Assert that all cloners have been destroyed when freeing the memory.

Change all simple cloners to destroy their clones with ifc_simple_destroy() on
module unload so the reference count is properly updated. This also cleans up
the interface destroy routines and allows future optimisation.

Discussed with: brooks, pjd, -current
Reviewed by: brooks

18 years agoBe pedantic here: We're converting from network byte order to host
Warner Losh [Wed, 12 Oct 2005 19:12:46 +0000 (19:12 +0000)]
Be pedantic here: We're converting from network byte order to host
byte order in these cases.  This is a nop in terms of the generated
code, but is logically incorrect.

PR: 73852

18 years agoDocument that changes to nsswitch.conf might be needed when
Bjoern A. Zeeb [Wed, 12 Oct 2005 19:05:22 +0000 (19:05 +0000)]
Document that changes to nsswitch.conf might be needed when
compiling with NO_NIS.

PR: bin/87221
Reviewed by: simon
MFC after: 4 weeks

18 years agoo INP_ONESBCAST is inpcb.inp_vflag flag not inp_flags. The confusion
Maxim Konovalov [Wed, 12 Oct 2005 18:13:25 +0000 (18:13 +0000)]
o INP_ONESBCAST is inpcb.inp_vflag flag not inp_flags.  The confusion
with IP_PORTRANGE_HIGH leads to the incorrect checksum calculation.

PR: kern/87306
Submitted by: Rickard Lind
Reviewed by: bms
MFC after: 2 weeks

18 years agoUse a better EVFILT_LIO description!
Doug Ambrisko [Wed, 12 Oct 2005 18:07:13 +0000 (18:07 +0000)]
Use a better EVFILT_LIO description!

Submitted by: alc

18 years agoThis test can run now.
Doug Ambrisko [Wed, 12 Oct 2005 18:04:18 +0000 (18:04 +0000)]
This test can run now.

18 years agoAdd in kqueue support to LIO event notification and fix how it handled
Doug Ambrisko [Wed, 12 Oct 2005 17:51:31 +0000 (17:51 +0000)]
Add in kqueue support to LIO event notification and fix how it handled
notifications when LIO operations completed.  These were the problems
with LIO event complete notification:
      - Move all LIO/AIO event notification into one general function
so we don't have bugs in different data paths.  This unification
got rid of several notification bugs one of which if kqueue was
used a SIGILL could get sent to the process.
      - Change the LIO event accounting to count all AIO request that
could have been split across the fast path and daemon mode.
The prior accounting only kept track of AIO op's in that
mode and not the entire list of operations.  This could cause
a bogus LIO event complete notification to occur when all of
the fast path AIO op's completed and not the AIO op's that
ended up queued for the daemon.

Suggestions from: alc

18 years agoMake some purely internal symbols static to reduce link pollution.
Tim Kientzle [Wed, 12 Oct 2005 15:38:45 +0000 (15:38 +0000)]
Make some purely internal symbols static to reduce link pollution.

18 years agoWhen performing a VOP_LOOKUP() as part of UFS1 extended attribute
Robert Watson [Wed, 12 Oct 2005 14:18:58 +0000 (14:18 +0000)]
When performing a VOP_LOOKUP() as part of UFS1 extended attribute
auto-start, set cnp.cn_lkflags to LK_EXCLUSIVE.  This flag must now
be set so that lockmgr knows what kind of lock to acquire, and it
will panic if not specified.  This resulted in a panic when using
extended attributes on UFS1 as of locking work present in the 6.x
branch.

This is a RELENG_6_0 merge candidate.

Reported by: lofi
MFC after: 3 days

18 years agoNew release notes: snd_mss(4) manual page added.
Joel Dahl [Wed, 12 Oct 2005 12:24:37 +0000 (12:24 +0000)]
New release notes: snd_mss(4) manual page added.

18 years agoAfter rev. 1.103 the oitem and ierror are no longer needed, remove them.
Gleb Smirnoff [Wed, 12 Oct 2005 10:18:44 +0000 (10:18 +0000)]
After rev. 1.103 the oitem and ierror are no longer needed, remove them.

18 years ago__MAKE_CONF doesn't really belong here because it is
Yaroslav Tykhiy [Wed, 12 Oct 2005 10:09:36 +0000 (10:09 +0000)]
__MAKE_CONF doesn't really belong here because it is
a FreeBSD extension of sys.mk.  A xref to make.conf(5)
will be enough here.

Requested by: ru

18 years agoUnbreak the net.inet6.tcp6.getcred sysctl.
Philip Paeps [Wed, 12 Oct 2005 09:24:18 +0000 (09:24 +0000)]
Unbreak the net.inet6.tcp6.getcred sysctl.

This makes inetd/auth work again in IPv6 setups.

Pointy hat to: ume/KAME

18 years agoAES counter mode uses 8byte IV, not 16 bytes.
Hajimu UMEMOTO [Wed, 12 Oct 2005 09:13:48 +0000 (09:13 +0000)]
AES counter mode uses 8byte IV, not 16 bytes.

Obtained from: NetBSD

18 years agoMove execve's access time update functionality into a new
Diomidis Spinellis [Wed, 12 Oct 2005 06:56:00 +0000 (06:56 +0000)]
Move execve's access time update functionality into a new
vfs_mark_atime() function, and use the new function for
performing efficient atime updates in mmap().

Reviewed by: bde
MFC after: 2 weeks

18 years agoMinor style nit: tab instead of space after #define
Tim Kientzle [Wed, 12 Oct 2005 03:28:38 +0000 (03:28 +0000)]
Minor style nit:  tab instead of space after #define