]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
18 years agoContributions from XFS for FreeBSD project:
rodrigc [Mon, 12 Dec 2005 00:02:22 +0000 (00:02 +0000)]
Contributions from XFS for FreeBSD project:
- Implement cv_wait_unlock() method which has semantics compatible
  with the sv_wait() method in IRIX.  For cv_wait_unlock(), the lock
  must be held before entering the function, but is not held when the
  function is exited.

- Implement the existing cv_wait() function in terms of cv_wait_unlock().

Submitted by: kan
Feedback from: jhb, trhodes, Christoph Hellwig <hch at infradead dot org>

18 years agonote shuffle of commonly used programs in tools/tools/ath
sam [Sun, 11 Dec 2005 23:18:58 +0000 (23:18 +0000)]
note shuffle of commonly used programs in tools/tools/ath

18 years agothis didn't make it in the last commit
sam [Sun, 11 Dec 2005 23:15:24 +0000 (23:15 +0000)]
this didn't make it in the last commit

18 years agoo move programs to separate directories and use stock Makefiles;
sam [Sun, 11 Dec 2005 23:13:54 +0000 (23:13 +0000)]
o move programs to separate directories and use stock Makefiles;
  this simplifies including them in crunchgen images
o rename 80211* programs to wlan* as was originally intended

18 years agoremove cts burst extension stats
sam [Sun, 11 Dec 2005 22:46:41 +0000 (22:46 +0000)]
remove cts burst extension stats

18 years agoRemove unneeded calls to pmap_remove_all(). The given page is not mapped.
alc [Sun, 11 Dec 2005 22:06:57 +0000 (22:06 +0000)]
Remove unneeded calls to pmap_remove_all().  The given page is not mapped.

Reviewed by: tegge

18 years agoFix calculation of meminfo's swaptotal and swapfree on at least amd64.
mlaier [Sun, 11 Dec 2005 21:37:42 +0000 (21:37 +0000)]
Fix calculation of meminfo's swaptotal and swapfree on at least amd64.

MFC after: 3 days

18 years ago- Rename UQ_BROKEN_IPOD to UQ_NO_OPEN_CLEARSTALL since it's likely to be used
flz [Sun, 11 Dec 2005 20:14:38 +0000 (20:14 +0000)]
- Rename UQ_BROKEN_IPOD to UQ_NO_OPEN_CLEARSTALL since it's likely to be used
by more devices than iPods.

Proposed by: iedowse
Approved by: ssouhlal
MFC after: 3 days

18 years agoAdded comments about the magic behind
bde [Sun, 11 Dec 2005 19:51:30 +0000 (19:51 +0000)]
Added comments about the magic behind
    <cbrt(x) in bits> ~= <x in bits>/3 + BIAS.
Keep the large comments only in the double version as usual.

Fixed some style bugs (mainly grammar and spelling errors in comments).

18 years agoFixed the unexpectedly large maximum error after the previous commit.
bde [Sun, 11 Dec 2005 17:58:14 +0000 (17:58 +0000)]
Fixed the unexpectedly large maximum error after the previous commit.
It was because I forgot to translate the part of the double precision
algorithm that chops t so that t*t is exact.  Now the maximum error
is the same as for double precision (almost exactly 2.0/3 ulps).

18 years agoAdd /boot/firmware as iwi(4) now reads its firmware there.
delphij [Sun, 11 Dec 2005 15:21:18 +0000 (15:21 +0000)]
Add /boot/firmware as iwi(4) now reads its firmware there.

Reminded by: flz

18 years agoFixed all 502518670 errors of more than 1 ulp for cbrtf() on amd64.
bde [Sun, 11 Dec 2005 13:22:01 +0000 (13:22 +0000)]
Fixed all 502518670 errors of more than 1 ulp for cbrtf() on amd64.
The maximum error was 3.56 ulps.

The bug was another translation error.  The double precision version
has a comment saying "new cbrt to 23 bits, may be implemented in
precision".  This means exactly what it says -- that the 23 bit second
approximation for the double precision cbrt() may be implemented in
single (i.e., float) precision.  It doesn't mean what the translation
assumed -- that this approximation, when implemented in float precision,
is good enough for the the final approximation in float precision.
First, float precision needs a 24 bit approximation.  The "23 bit"
approximation is actually good to 24 bits on float precision args, but
only if it is evaluated in double precision.  Second, the algorithm
requires a cleanup step to ensure its error bound.

In float precision, any reasonable algorithm works for the cleanup
step.  Use the same algorithm as for double precision, although this
is much more than enough and is a significant pessimization, and don't
optimize or simplify anything using double precision to implement the
float case, so that the whole double precision algorithm can be verified
in float precision.  A maximum error of 0.667 ulps is claimed for cbrt()
and the max for cbrtf() using the same algorithm shouldn't be different,
but the actual max for cbrtf() on amd64 is now 0.9834 ulps.  (On i386
-O1 the max is 0.5006 (down from < 0.7) due to extra precision.)

18 years agoFixed some magic numbers.
bde [Sun, 11 Dec 2005 11:40:55 +0000 (11:40 +0000)]
Fixed some magic numbers.

The threshold for not being tiny was too small.  Use the usual 2**-12
threshold.  As for sinhf, use a different method (now the same as for
sinhf) to set the inexact flag for tiny nonzero x so that the larger
threshold works, although this method is imperfect.  As for sinhf,
this change is not just an optimization, since the general code that
we fell into has accuracy problems even for tiny x.  On amd64, avoiding
it fixes tanhf on 2*13495596 args with errors of between 1 and 1.3
ulps and thus reduces the total number of args with errors of >= 1 ulp
from 37533748 to 5271278; the maximum error is unchanged at 2.2 ulps.

The magic number 22 is log(DBL_MAX)/2 plus slop.  This is bogus for
float precision.  Use 9 (log(FLT_MAX)/2 plus less slop than for
double precision).  Unlike for coshf and tanhf, this is just an
optimization, and MAX isn't misspelled EPSILON in the commit log.

I started testing with nonstandard rounding modes, and verified that
the chosen thresholds work for all modes modulo problems not related
to thresholds.  The best thresholds are not very dependent on the mode,
at least for tanhf.

18 years agoMove the remaining entries from usbd.conf to devd.conf. This now
iedowse [Sun, 11 Dec 2005 00:18:28 +0000 (00:18 +0000)]
Move the remaining entries from usbd.conf to devd.conf. This now
makes usbd redundant.

PR: conf/73799
Submitted by: Anish Mistry

18 years agoDrop rcconf.sh now that it has been removed
dougb [Sat, 10 Dec 2005 23:23:09 +0000 (23:23 +0000)]
Drop rcconf.sh now that it has been removed

18 years agoRemove rcconf.sh from /etc/rc.d, and instead load the configuration
dougb [Sat, 10 Dec 2005 20:21:46 +0000 (20:21 +0000)]
Remove rcconf.sh from /etc/rc.d, and instead load the configuration
as part of rc. Doing this, and the sourcing of rc.subr after we have
determined if we are booting diskless (and correspondingly run
rc.initdiskless if necessary) are safe, and actually allow fewer files
to be needed on the diskless box. This also allows variables from
the configuration to be available to rc itself, such as ...

Add a variable to rc.conf, early_late_divider, which designates the
script which separates the early and late stages of the boot process.
Default this to mountcritlocal, and add text to etc/defaults/rc.conf,
rc.conf(5) and diskless(8) which describes how and why one might want
to change this.

Reviewed by: brooks

18 years agoSilence a warning about empty directories in all the places it might
dougb [Sat, 10 Dec 2005 20:19:08 +0000 (20:19 +0000)]
Silence a warning about empty directories in all the places it might
occur.

Reminded by: yar

18 years agoUse of REQUIRE is better than BEFORE for most scripts, and very
dougb [Sat, 10 Dec 2005 19:49:03 +0000 (19:49 +0000)]
Use of REQUIRE is better than BEFORE for most scripts, and very
few scripts should have no REQUIRE at all.

18 years agoHide the 4k mbuf clusters if the normal clusters are defined to be
andre [Sat, 10 Dec 2005 15:21:04 +0000 (15:21 +0000)]
Hide the 4k mbuf clusters if the normal clusters are defined to be
4k already.

This unbreaks tinderbox.

Submitted by: ru

18 years agoAdopt for modern FreeBSD.
ru [Sat, 10 Dec 2005 12:09:54 +0000 (12:09 +0000)]
Adopt for modern FreeBSD.

Requested by: az

18 years agoXref padlock(4), also fast_ipsec(4) and geli(8) as consumers of this device.
brueffer [Sat, 10 Dec 2005 10:51:55 +0000 (10:51 +0000)]
Xref padlock(4), also fast_ipsec(4) and geli(8) as consumers of this device.

MFC after: 3 days

18 years agoManpage for the padlock driver (VIA C3/Eden AES support and RNG).
brueffer [Sat, 10 Dec 2005 10:44:44 +0000 (10:44 +0000)]
Manpage for the padlock driver (VIA C3/Eden AES support and RNG).

MFC after: 3 days

18 years agoAllocate the jumbo rx frame buffer with busdma.
scottl [Sat, 10 Dec 2005 08:58:48 +0000 (08:58 +0000)]
Allocate the jumbo rx frame buffer with busdma.

18 years agoWhen we get a bogus hostname in an option, drop the option rather than
brooks [Sat, 10 Dec 2005 03:46:14 +0000 (03:46 +0000)]
When we get a bogus hostname in an option, drop the option rather than
refusing the lease.  This allow obtaining leases on misadministered
networks that use host names with underscores in them.

MFC After: 3 days

18 years agoif_ti has been operating with locks for a while, so remove the GIANT markers.
scottl [Sat, 10 Dec 2005 01:25:46 +0000 (01:25 +0000)]
if_ti has been operating with locks for a while, so remove the GIANT markers.
Also fix man potential locking problems in the cdev ioctl handler.

18 years agoThe if_ti Tigon I/II driver has moved to /sys/dev/ti
scottl [Sat, 10 Dec 2005 00:38:33 +0000 (00:38 +0000)]
The if_ti Tigon I/II driver has moved to /sys/dev/ti

18 years ago- Better use of the busdma API.
cognet [Fri, 9 Dec 2005 23:55:41 +0000 (23:55 +0000)]
- Better use of the busdma API.
- Use spin locks instead of sleep locks.

18 years agoFix a harmless bug in the way we allocate the early PTEs.
cognet [Fri, 9 Dec 2005 23:54:50 +0000 (23:54 +0000)]
Fix a harmless bug in the way we allocate the early PTEs.

18 years agoThe IQ80321 clock is 200MHz, but the IQ80321 is 198MHz, so add a kernel option
cognet [Fri, 9 Dec 2005 23:52:51 +0000 (23:52 +0000)]
The IQ80321 clock is 200MHz, but the IQ80321 is 198MHz, so add a kernel option
to override the frequency

18 years agoA #define is not enough, we need to cast from u_long * to uint32_t *.
cognet [Fri, 9 Dec 2005 22:58:07 +0000 (22:58 +0000)]
A #define is not enough, we need to cast from u_long * to uint32_t *.

18 years agoDefine atomic_whatever_long
cognet [Fri, 9 Dec 2005 22:33:20 +0000 (22:33 +0000)]
Define atomic_whatever_long

18 years agoIn copyout(), quad-align the source buffer, and use ldrd if possible.
cognet [Fri, 9 Dec 2005 15:31:02 +0000 (15:31 +0000)]
In copyout(), quad-align the source buffer, and use ldrd if possible.

18 years agoFiles are installed with mode 444 by default.
ru [Fri, 9 Dec 2005 15:19:31 +0000 (15:19 +0000)]
Files are installed with mode 444 by default.

18 years agoReally fix the relative timestamp bug. It was only incorrect for the ALQ
njl [Fri, 9 Dec 2005 14:27:03 +0000 (14:27 +0000)]
Really fix the relative timestamp bug.  It was only incorrect for the ALQ
case.  It seems entries are in reverse order when read from the kernel
memory but in the right order when read from a file (i.e. ALQ).  Handle
both cases.

MFC after: 1 day

18 years agoUnbreak hotplug support on the ICH6 and ICH7 chipsets.
sos [Fri, 9 Dec 2005 14:23:24 +0000 (14:23 +0000)]
Unbreak hotplug support on the ICH6 and ICH7 chipsets.

18 years agoSync with i386, fix compiling for non-SMP.
davidxu [Fri, 9 Dec 2005 13:30:34 +0000 (13:30 +0000)]
Sync with i386, fix compiling for non-SMP.

18 years agoFix compiling warning on 64 bits system.
davidxu [Fri, 9 Dec 2005 13:16:48 +0000 (13:16 +0000)]
Fix compiling warning on 64 bits system.

18 years agoCleanup _FreeBSD_version.
glebius [Fri, 9 Dec 2005 13:03:30 +0000 (13:03 +0000)]
Cleanup _FreeBSD_version.

18 years agoFix build breakage by fixing typo.
pjd [Fri, 9 Dec 2005 11:38:02 +0000 (11:38 +0000)]
Fix build breakage by fixing typo.

Reported by: glebius

18 years agoAdd a sysctl to force a process to sigexit if a trap signal is
davidxu [Fri, 9 Dec 2005 08:29:29 +0000 (08:29 +0000)]
Add a sysctl to force a process to sigexit if a trap signal is
being hold by current thread or ignored by current process,
otherwise, it is very possible the thread will enter an infinite loop
and lead to an administrator's nightmare.

18 years agoRemove itimers_event_hook, now it is a private function in kern_time.c.
davidxu [Fri, 9 Dec 2005 08:19:31 +0000 (08:19 +0000)]
Remove itimers_event_hook, now it is a private function in kern_time.c.

18 years agoFix memory leak.
ru [Fri, 9 Dec 2005 07:09:44 +0000 (07:09 +0000)]
Fix memory leak.

PR: kern/90113
Submitted by: Antoine Brodin

18 years agoconfig.5 was repocopied from share/man/man5/ to usr.sbin/config/.
ru [Fri, 9 Dec 2005 06:46:41 +0000 (06:46 +0000)]
config.5 was repocopied from share/man/man5/ to usr.sbin/config/.

Repocopied by: peter

18 years agoRegister itimers_event_hook as a kernel event handler, so I don't
davidxu [Fri, 9 Dec 2005 05:43:26 +0000 (05:43 +0000)]
Register itimers_event_hook as a kernel event handler, so I don't
have to duplicate code to call it in exec() and exit1().

18 years agoRemove the -B option from the directory-copying examples. The -B
kientzle [Fri, 9 Dec 2005 05:19:00 +0000 (05:19 +0000)]
Remove the -B option from the directory-copying examples.  The -B
option is undocumented because it does nothing.  It does nothing
because bsdtar never needs it.  It is accepted because gnutar does
sometimes need it and many scripts use it.

Reported by: Pawel Jakub Dawidek

18 years agofixed a kernel crash at the initialization time of PIM-SM register interface
suz [Fri, 9 Dec 2005 04:42:19 +0000 (04:42 +0000)]
fixed a kernel crash at the initialization time of PIM-SM register interface

MFC after: 2 days

18 years agoTweak -32 description and add -32 FILES.
obrien [Fri, 9 Dec 2005 03:12:25 +0000 (03:12 +0000)]
Tweak -32 description and add -32 FILES.

18 years agoComment out mqfs_create_link. Inline some small functions.
davidxu [Fri, 9 Dec 2005 02:38:29 +0000 (02:38 +0000)]
Comment out mqfs_create_link. Inline some small functions.

18 years agoNow SIGCHLD is always queued.
davidxu [Fri, 9 Dec 2005 02:27:55 +0000 (02:27 +0000)]
Now SIGCHLD is always queued.

18 years agoCleanup sigqueue sysctl.
davidxu [Fri, 9 Dec 2005 02:26:44 +0000 (02:26 +0000)]
Cleanup sigqueue sysctl.

18 years ago- Allow to specify the byte which will be used for filling read buffer.
pjd [Thu, 8 Dec 2005 23:06:59 +0000 (23:06 +0000)]
- Allow to specify the byte which will be used for filling read buffer.
- Improve sysctl description a bit.

Submitted by: Ivan Voras <ivoras@gmail.com>

18 years agoTeach NOP GEOM class how to gather the following statistics:
pjd [Thu, 8 Dec 2005 23:00:31 +0000 (23:00 +0000)]
Teach NOP GEOM class how to gather the following statistics:
- number of read I/O requests,
- number of write I/O requests,
- number of read bytes,
- number of written bytes.
Add 'reset' subcommand for resetting statistics.

18 years agoCareful measurement of the ST Labs card shows that the pulse width of
imp [Thu, 8 Dec 2005 22:29:42 +0000 (22:29 +0000)]
Careful measurement of the ST Labs card shows that the pulse width of
transmitted bits was between 8.6180us and 8.6200us when we used a RCLK
of 16.500MHz.  This is a little low (should be 8.6805us).  This error
is exactly the error one would expect if it actually had a 16.384MHz
watch oscillator (as suggested by garrett) instead of using the PCI
RCLK.  Assume that the pci clock therefore wasn't really used, but
instead the cheap 16.384MH watch quartz oscillator.  This gives bits
in the 8.6800us to 8.6810us ranage, which matches theoretical.

Submitted by: garrett

18 years agoRegen for futimes.
ambrisko [Thu, 8 Dec 2005 22:15:09 +0000 (22:15 +0000)]
Regen for futimes.

18 years agoAdd 32bit version of futimes so untar doesn't result in bad dates
ambrisko [Thu, 8 Dec 2005 22:14:25 +0000 (22:14 +0000)]
Add 32bit version of futimes so untar doesn't result in bad dates
(Jan 1, 1970) when run on amd64.

Reviewed by: ps

18 years ago- Document trap's -l option and the behaviour of a missing action or a single
stefanf [Thu, 8 Dec 2005 21:18:59 +0000 (21:18 +0000)]
- Document trap's -l option and the behaviour of a missing action or a single
  dash.
- Discourage the omission of the action.

PR: 70985 [1]
Submitted by: Martin Kammerhofer

18 years agoPrint empty quotes ('') when an empty string is passed to outqstr().
stefanf [Thu, 8 Dec 2005 21:00:39 +0000 (21:00 +0000)]
Print empty quotes ('') when an empty string is passed to outqstr().
This makes a difference for the trap builtin, where after "trap '' 0" we
printed "trap -- quit".  This is wrong, because an empty action means to reset
the action to the default.  A side effect of this commit is that empty
variables are now printed as "variable=''" instead of just "variable=".

18 years agoCorrectly quote the output when showing the installed trap actions.
stefanf [Thu, 8 Dec 2005 20:08:36 +0000 (20:08 +0000)]
Correctly quote the output when showing the installed trap actions.

PR: 74043
Submitted by: Jilles Tjoelker

18 years agoAdd support for 7320 and 915 PCIe chipsets.
rodrigc [Thu, 8 Dec 2005 18:55:15 +0000 (18:55 +0000)]
Add support for 7320 and 915 PCIe chipsets.

Submitted by: Gavin Atkinson <gavin.atkinson at ury dot york dot ac dot uk>
PR: kern/79139
Reviewed by: scottl

18 years agoWhitespace: reduce diffs with amd64.
jhb [Thu, 8 Dec 2005 18:33:48 +0000 (18:33 +0000)]
Whitespace: reduce diffs with amd64.

18 years agoMFi386:
jhb [Thu, 8 Dec 2005 18:33:30 +0000 (18:33 +0000)]
MFi386:
- Move PUSH_FRAME and POP_FRAME to asmacros.h and use PUSH_FRAME in
  atpic entry points.
- Move PCPU_* asm macros out of the middle of the asm profiling macros.
- Pass IRQ vector argument as an int rather than void * to reduce diffs
  with i386.
- EOI the lapic in C for the lapic timer handler.
- GC unused Xcpuast function.
- Split IPI_STOP handling code of ipi_nmi_handler() out into a
  cpustop_handler() function and call it from Xcpustop rather than
  duplicating all the logic in assembly.
- Fixup the list of symbols with interrupt frames in ddb traces.
  Xatpic_fastintr* have never existed on amd64, and the lapic timer
  handler and various IPI handlers were missing.
- Use trapframe instead of intrframe for interrupt entry points (on amd64
  the interrupt vector was already a separate argument, so the two frames
  were already identical) and GC intrframe.

Submitted by: peter (3)

18 years agoClarify that the echo builtin takes an arbitrary number of strings.
stefanf [Thu, 8 Dec 2005 17:59:54 +0000 (17:59 +0000)]
Clarify that the echo builtin takes an arbitrary number of strings.
Mention that spaces are printed between the strings.

18 years agoA big rewrite of receive Jumbo frame handling. Remove the local Jumbo
glebius [Thu, 8 Dec 2005 16:11:45 +0000 (16:11 +0000)]
A big rewrite of receive Jumbo frame handling. Remove the local Jumbo
cluster allocator, that wasn't MPSAFE. Instead, utilize our new generic
UMA jumbo cluster allocator. Since UMA gives us a 9k piece that is contigous
in virtual memory, but isn't contigous in physical memory we need to handle
a few segments. To deal with this we utilize Tigon chip feature - extended
RX descriptors, that can handle up to four DMA segments for one frame.

Details:

o Remove bge_alloc_jumbo_mem(), bge_free_jumbo_mem(),
  bge_jalloc(), bge_jfree() functions.
o Remove SLIST heads, bge_jumbo_tag, bge_jumbo_map from softc.
o Use extended RX BDs for Jumbo receive producer ring, and
  initialize it appropriately.
o New bge_newbuf_jumbo():
  - Allocate an mbuf with Jumbo cluster with help of m_cljget().
  - Load the cluster for DMA with help of bus_dmamap_load_mbuf_sg().
  - Assert that we got 3 segments in the DMA mapping.
  - Fill in these 3 segments into the extended RX descriptor.

18 years agoSuppress logging about unimplemented syscalls to one time per process. This
glebius [Thu, 8 Dec 2005 13:33:57 +0000 (13:33 +0000)]
Suppress logging about unimplemented syscalls to one time per process. This
prevents hard flood of the system console.

Reviewed by: bde

18 years agoAfter printing per-cpu cache contents, summarize total cached items and
rwatson [Thu, 8 Dec 2005 13:33:12 +0000 (13:33 +0000)]
After printing per-cpu cache contents, summarize total cached items and
entries across the caches.

Fix an off-by-one error in counting caches.

18 years ago1) fix tiny bug in bge_start_locked()
oleg [Thu, 8 Dec 2005 13:31:52 +0000 (13:31 +0000)]
1) fix tiny bug in bge_start_locked()
2) rework link state detection code & use it in POLLING mode
3) fix 2 bugs in link state detection code:
a) driver unable to detect link loss on bcm5721
b) on bcm570x chips (tested on bcm5700 bcm5701 bcm5702) driver fails
   to detect link loss with probability 1/6 (solved in brgphy.c)

Devices working in TBI mode should not be affected by this change.

Approved by: glebius (mentor)
MFC after: 1 month

18 years agoTeach umastat about the variable-length array of per-CPU caches at the end
rwatson [Thu, 8 Dec 2005 13:16:22 +0000 (13:16 +0000)]
Teach umastat about the variable-length array of per-CPU caches at the end
of struct uma_zone.  It is declared as an array of size [1], but then
sized at run-time by UMA to include room for mp_maxid+1 CPUs.  We have to
copy the uma_zone first at the declared structure size, then check to make
sure it's not an internal zone before copying the larger size (UMA
internal zones don't use per-CPU caches).  This fixes umastat for SMP.

18 years agoAdd an API for jumbo mbuf cluster allocation and also provide
andre [Thu, 8 Dec 2005 13:13:06 +0000 (13:13 +0000)]
Add an API for jumbo mbuf cluster allocation and also provide
4k clusters in addition to 9k and 16k ones.

 struct mbuf *m_getjcl(int how, short type, int flags, int size)
 void *m_cljget(struct mbuf *m, int how, int size)

m_getjcl() returns an mbuf with a cluster of the specified size attached
like m_getcl() does for 2k clusters.

m_cljget() is different from m_clget() as it can allocate clusters
without attaching them to an mbuf.  In that case the return value
is the pointer to the cluster of the requested size.  If an mbuf was
specified, it gets the cluster attached to it and the return value
can be safely ignored.

For size both take MCLBYTES, MJUM4BYTES, MJUM9BYTES, MJUM16BYTES.

Reviewed by: glebius
Tested by: glebius
Sponsored by: TCP/IP Optimization Fundraise 2005

18 years agoMACHINE is changed to "pc98" on FreeBSD/pc98.
nyan [Thu, 8 Dec 2005 12:43:20 +0000 (12:43 +0000)]
MACHINE is changed to "pc98" on FreeBSD/pc98.

18 years agoSwitch MACHINE to "pc98" on FreeBSD/pc98.
nyan [Thu, 8 Dec 2005 12:35:12 +0000 (12:35 +0000)]
Switch MACHINE to "pc98" on FreeBSD/pc98.
Add copyright.

Approved by: FreeBSD/pc98 development team.

18 years agoo Remove SI_KERNEL until I really implemented it.
davidxu [Thu, 8 Dec 2005 09:00:54 +0000 (09:00 +0000)]
o Remove SI_KERNEL until I really implemented it.
o Add definition SI_NOINFO for zero si_code.

18 years agothe response NS to a DAD NS was not sent correctly due to the
ume [Thu, 8 Dec 2005 06:43:39 +0000 (06:43 +0000)]
the response NS to a DAD NS was not sent correctly due to the
invalid destination address.

Submitted by: JINMEI Tatuya <jinmei__at__isl.rdc.toshiba.co.jp>
MFC after: 1 day

18 years agoIn devfs_first(), set mp->mnt_opt to a valid empty list of mount options
rodrigc [Thu, 8 Dec 2005 04:27:53 +0000 (04:27 +0000)]
In devfs_first(), set mp->mnt_opt to a valid empty list of mount options
instead of leaving it NULL.  This eliminates a kernel panic
when trying to do a mount -o update of /dev.

Noticed by: cjsp
Reviewed by: phk

18 years agoAdd "errmsg" to list of global mount options.
rodrigc [Thu, 8 Dec 2005 04:09:29 +0000 (04:09 +0000)]
Add "errmsg" to list of global mount options.

18 years agoReorder the calling of the completion callback and the transfer
iedowse [Thu, 8 Dec 2005 03:08:17 +0000 (03:08 +0000)]
Reorder the calling of the completion callback and the transfer
"done" method so that for non-repeat operations we have completely
finished with the transfer by the time the callback is invoked.
This makes it possible to recycle a transfer from within the callback
routine for the same transfer. Previously this almost worked, but
with OHCI controllers calling the "done" method after the callback
would zero out some important fields needed by the recycled transfer.
Only some usb peripheral drivers such as ucom appear to rely on the
ability to reuse a transfer from its callback.

MFC after: 1 week

18 years agoDo not accept an empty bpf program.
jkim [Thu, 8 Dec 2005 00:05:03 +0000 (00:05 +0000)]
Do not accept an empty bpf program.

18 years agoUse cross-compile friendly spelling of CPUTYPE.
obrien [Wed, 7 Dec 2005 22:55:29 +0000 (22:55 +0000)]
Use cross-compile friendly spelling of CPUTYPE.

Submitted by: ru

18 years agoRemove the ports version of bsdiff - it is now in the base system.
obrien [Wed, 7 Dec 2005 22:25:12 +0000 (22:25 +0000)]
Remove the ports version of bsdiff - it is now in the base system.

18 years agoRemove the ports version of portsnap - it is now in the base system.
obrien [Wed, 7 Dec 2005 21:41:59 +0000 (21:41 +0000)]
Remove the ports version of portsnap - it is now in the base system.

18 years agoFix build without BPF_JITTER option.
jkim [Wed, 7 Dec 2005 21:41:45 +0000 (21:41 +0000)]
Fix build without BPF_JITTER option.

18 years agoAdd BPF Just-In-Time compiler support for ng_bpf(4).
jkim [Wed, 7 Dec 2005 21:30:47 +0000 (21:30 +0000)]
Add BPF Just-In-Time compiler support for ng_bpf(4).

The sysctl is changed from net.bpf.jitter.enable to net.bpf_jitter.enable
and this controls both bpf(4) and ng_bpf(4) now.

18 years agoAdd a paragraph to the COMMON ITEMS section that describes why
dougb [Wed, 7 Dec 2005 20:49:42 +0000 (20:49 +0000)]
Add a paragraph to the COMMON ITEMS section that describes why
upgrading to the latest code in one branch before trying a major
version upgrade is a good idea.

Fleshing out of my thoughts provided by: kris

18 years agoStyle, no functional changes.
ru [Wed, 7 Dec 2005 20:01:12 +0000 (20:01 +0000)]
Style, no functional changes.

18 years agoTeach rfcomm_sppd(8) to recognize "lan" (for LAN Access Using PPP) service
emax [Wed, 7 Dec 2005 19:41:58 +0000 (19:41 +0000)]
Teach rfcomm_sppd(8) to recognize "lan" (for LAN Access Using PPP) service
name in '-c' (RFCOMM channel) option.

MFC after: 3 days

18 years agoSSE2 comes thru the users choice of CPUTYPE.
obrien [Wed, 7 Dec 2005 17:42:27 +0000 (17:42 +0000)]
SSE2 comes thru the users choice of CPUTYPE.

18 years agoDefault to producing 'k8' COMPAT_32BIT bits, but allow override.
obrien [Wed, 7 Dec 2005 17:41:10 +0000 (17:41 +0000)]
Default to producing 'k8' COMPAT_32BIT bits, but allow override.

18 years agoAdd support for the nVidia nForce MCP12 & MCP13 Networking Adapters.
obrien [Wed, 7 Dec 2005 17:38:03 +0000 (17:38 +0000)]
Add support for the nVidia nForce MCP12 & MCP13 Networking Adapters.

18 years agoCatch up to the nvenetlib 1.0-0310 import.
obrien [Wed, 7 Dec 2005 17:36:53 +0000 (17:36 +0000)]
Catch up to the nvenetlib 1.0-0310 import.

18 years agoThis commit was generated by cvs2svn to compensate for changes in r153200,
obrien [Wed, 7 Dec 2005 17:32:13 +0000 (17:32 +0000)]
This commit was generated by cvs2svn to compensate for changes in r153200,
which included commits to RCS files with non-trunk default branches.

18 years agoUpdate the nForce MCP NIC bits. This is version 1.0-0310 23-Nov-2005.
obrien [Wed, 7 Dec 2005 17:32:13 +0000 (17:32 +0000)]
Update the nForce MCP NIC bits.  This is version 1.0-0310 23-Nov-2005.

18 years agoWording tweaks.
des [Wed, 7 Dec 2005 12:33:52 +0000 (12:33 +0000)]
Wording tweaks.

PR: not quite docs/84620
MFC after: 2 weeks

18 years agoAdd a -0 option for interoperability with xargs -0.
des [Wed, 7 Dec 2005 12:22:46 +0000 (12:22 +0000)]
Add a -0 option for interoperability with xargs -0.

PR: bin/56558
Submitted by: Eugene Grosbein <eugen@grosbein.pp.ru>
MFC after: 2 weeks

18 years agoAdd -r option for GNU compatibility.
des [Wed, 7 Dec 2005 12:20:08 +0000 (12:20 +0000)]
Add -r option for GNU compatibility.

MFC after: 2 weeks

18 years agoUse __DEVOLATILE to cast conspeed.
imp [Wed, 7 Dec 2005 07:23:53 +0000 (07:23 +0000)]
Use __DEVOLATILE to cast conspeed.

18 years agostyle(9) nits
obrien [Wed, 7 Dec 2005 03:41:12 +0000 (03:41 +0000)]
style(9) nits

18 years agoAdd Sparc TLS relocation definitions.
obrien [Wed, 7 Dec 2005 03:39:37 +0000 (03:39 +0000)]
Add Sparc TLS relocation definitions.

18 years agoChanges imported from XFS for FreeBSD project:
rodrigc [Wed, 7 Dec 2005 03:39:08 +0000 (03:39 +0000)]
Changes imported from XFS for FreeBSD project:
- add fields to struct buf (needed by XFS)
    - 3 private fields: b_fsprivate1, b_fsprivate2, b_fsprivate3
    - b_pin_count, count of pinned buffer

- add new B_MANAGED flag
- add breada() function to initiate asynchronous I/O on read-ahead blocks.
- add bufdone_finish(), bpin(), bunpin_wait() functions

Patches provided by: kan
Reviewed by: phk
Silence on: arch@

18 years ago- The geom(8) utility only uses three types of arguments: string (char *),
pjd [Wed, 7 Dec 2005 01:38:27 +0000 (01:38 +0000)]
- The geom(8) utility only uses three types of arguments: string (char *),
  value (intmax_t) and boolean (int).
  Based on that provide three functions:
        - gctl_get_ascii()
        - gctl_get_int()
        - gctl_get_intmax()
- Hide gctl_get_param() function, as it is only used internally in
  subr.c.
- Allow to provide argument name as (fmt, ...).
- Assert geom(8) bugs (missing argument is a geom(8) bug).

- Clean-up and simplify the code by using new functions and assumtions
  (no more checking for missing argument).

Tested by: regression tests

18 years agoConvert gstripe(8) regression tests to the new framework.
pjd [Wed, 7 Dec 2005 01:30:44 +0000 (01:30 +0000)]
Convert gstripe(8) regression tests to the new framework.

18 years agoConvert gshsec(8) regression tests to the new framework.
pjd [Wed, 7 Dec 2005 01:29:50 +0000 (01:29 +0000)]
Convert gshsec(8) regression tests to the new framework.