]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
17 years agoThere was a timeframe where crunchgen(1) was broken; force
Ruslan Ermilov [Mon, 27 Nov 2006 19:38:28 +0000 (19:38 +0000)]
There was a timeframe where crunchgen(1) was broken; force
an upgrade during the build so those who are affected can
build world again.

17 years agoFix for a bug caused by a race when 2 threads lookup the same
Mohan Srinivasan [Mon, 27 Nov 2006 19:06:43 +0000 (19:06 +0000)]
Fix for a bug caused by a race when 2 threads lookup the same
file. Leave the loser's lock(s) initialized, so the reclaim logic can
unconditionally destroy them when that race occurs (or if the vfs hash
insert happened to fail for some other reason). Thanks to ups@ for a
careful review of the code.
Reported by : Kris Kennaway

17 years agoRefine the previous change to only call bus_dmamap_sync() in case of
Marius Strobl [Mon, 27 Nov 2006 18:39:02 +0000 (18:39 +0000)]
Refine the previous change to only call bus_dmamap_sync() in case of
an URQ_REQUEST when DMA segments are passed to usbd_start_transfer();
when the request doesn't include the optional data buffer the size of
the transfer (xfer->length) is 0, in which case usbd_transfer() won't
create a DMA map but call usbd_start_transfer() with no DMA segments.
With the previous change this could result in the bus_dmamap_sync()
implementation dereferencing the NULL-pointer passed as the DMA map
argument.
While at it fix what appears to be a typo in usbd_start_transfer();
in order to determine wheter usbd_start_transfer() was called with
DMA segments check whether the number of segments is > 0 rather than
the pointer to them being > 0.

OK'ed by: imp

17 years agoOverhaul this manpage somewhat:
Christian Brueffer [Mon, 27 Nov 2006 18:37:45 +0000 (18:37 +0000)]
Overhaul this manpage somewhat:

- Move mentionings of required drivers to the SYNOPSIS where they belong
- Remove cruft (XiG, sheesh) from the EXAMPLES section and re-arrange things to make
  somewhat more sense
- Fix an Xref to manpage from ports to mention the corresponsing port
- Remove commented out HISTORY section header

MFC after: 3 days

17 years agosystat(1) reaches WARNS=6 on i386 and amd64. This is good
Yaroslav Tykhiy [Mon, 27 Nov 2006 17:54:28 +0000 (17:54 +0000)]
systat(1) reaches WARNS=6 on i386 and amd64.  This is good
for catching general regressions in future.  Unfortunately,
it still displays some problems at WARNS=6 on architectures
with stricter alignment requirements, e.g., ia64.

17 years agoThe logic of fetchnetstat_sysctl() isn't too complex: if idx is 0,
Yaroslav Tykhiy [Mon, 27 Nov 2006 17:34:40 +0000 (17:34 +0000)]
The logic of fetchnetstat_sysctl() isn't too complex: if idx is 0,
we set and use xtp; if idx is 1, we set and use xip; the other cases
are impossible.  However, GCC cannot see that xip and xtp are always
initialized before use because they are initialized and used in
different if/else blocks.  So setting them to NULL at the very
beginning won't hurt.

17 years ago+ WARNS=4 reminds that nlist.n_name isn't const.
Yaroslav Tykhiy [Mon, 27 Nov 2006 17:24:36 +0000 (17:24 +0000)]
+ WARNS=4 reminds that nlist.n_name isn't const.
+ Use C99 initializers to be WARNS-clean.
+ The last element in a namelist should have its n_name set to NULL,
  not to an empty string.

17 years agoDon't discard a const qualifier from constant strings.
Yaroslav Tykhiy [Mon, 27 Nov 2006 17:01:31 +0000 (17:01 +0000)]
Don't discard a const qualifier from constant strings.

17 years agoRemove a wrong "const" qualifier.
Yaroslav Tykhiy [Mon, 27 Nov 2006 16:59:08 +0000 (16:59 +0000)]
Remove a wrong "const" qualifier.

Spotted by: WARNS=6

17 years agoPush the detection of threading libs support down to ngctl/Makefile
Ruslan Ermilov [Mon, 27 Nov 2006 16:55:14 +0000 (16:55 +0000)]
Push the detection of threading libs support down to ngctl/Makefile
as the latter can be built without threading (with the loss of line
editing functionality).

Pointed by: glebius

17 years agoAdd the Tera scale factor, which is an easy job now.
Yaroslav Tykhiy [Mon, 27 Nov 2006 16:33:44 +0000 (16:33 +0000)]
Add the Tera scale factor, which is an easy job now.
More scale factors would overflow the command line where
the help for "scale" were shown.

17 years agoImprove support for large ISOs:
Tim Kientzle [Mon, 27 Nov 2006 16:30:32 +0000 (16:30 +0000)]
Improve support for large ISOs:
   * Correct a signed/unsigned problem that broke handling of files >2G.
   * Implement "skip" support for much faster "tar -t".

Thanks to: Robert Sciuk for sending me a DVD that illustrated the first problem

17 years agoStop exposing things that can be private to convtbl.c.
Yaroslav Tykhiy [Mon, 27 Nov 2006 16:23:09 +0000 (16:23 +0000)]
Stop exposing things that can be private to convtbl.c.

17 years agoKeep all convtbl-related constants and strings in convtbl.[ch].
Yaroslav Tykhiy [Mon, 27 Nov 2006 16:14:32 +0000 (16:14 +0000)]
Keep all convtbl-related constants and strings in convtbl.[ch].

17 years agoAllow for large scale factors. C99 warrants that
Yaroslav Tykhiy [Mon, 27 Nov 2006 15:26:26 +0000 (15:26 +0000)]
Allow for large scale factors.  C99 warrants that
ULLONG_MAX is not less than 2^64-1; and uintmax_t
cannot be more narrow than unsigned long long.
This allows for scale factors up to Exa inclusively.

Use plain int for the scale index to be consistent
with ifcmds.c and enum.

17 years agoReturn back one initializer, it is needed.
Yaroslav Tykhiy [Mon, 27 Nov 2006 15:14:50 +0000 (15:14 +0000)]
Return back one initializer, it is needed.

17 years ago+ Drop useless initializers.
Yaroslav Tykhiy [Mon, 27 Nov 2006 15:11:30 +0000 (15:11 +0000)]
+ Drop useless initializers.
+ style(9).

17 years agoDrop useless #includes.
Yaroslav Tykhiy [Mon, 27 Nov 2006 14:48:46 +0000 (14:48 +0000)]
Drop useless #includes.

17 years agoUse C99 initializers so that we don't really have to worry
Yaroslav Tykhiy [Mon, 27 Nov 2006 14:18:54 +0000 (14:18 +0000)]
Use C99 initializers so that we don't really have to worry
about the order of related things at several places.

17 years agoC can enumerate things for us.
Yaroslav Tykhiy [Mon, 27 Nov 2006 14:16:40 +0000 (14:16 +0000)]
C can enumerate things for us.

17 years ago/usr/sbin/faithd is an INET6-specific tool, and now it won't be built
Yaroslav Tykhiy [Mon, 27 Nov 2006 13:33:31 +0000 (13:33 +0000)]
/usr/sbin/faithd is an INET6-specific tool, and now it won't be built
if MK_INET6 is "no", so add it to the MK_INET6 block.

PR:             bin/100805 (audit-trail)

17 years agoShorten the KERNCONF filename to the basename in the install target as
Nick Hibma [Mon, 27 Nov 2006 12:16:03 +0000 (12:16 +0000)]
Shorten the KERNCONF filename to the basename in the install target as
well as in the build target, although it is not used in this target.

17 years agore-enable tte hash resize, corruption was caused by a missing htole32 in mpt_cam.c
Kip Macy [Mon, 27 Nov 2006 06:51:51 +0000 (06:51 +0000)]
re-enable tte hash resize, corruption was caused by a missing htole32 in mpt_cam.c

17 years ago- LNV -> LENOVO
Ariff Abdullah [Mon, 27 Nov 2006 06:16:36 +0000 (06:16 +0000)]
- LNV -> LENOVO
- Fix support for Samsung Q1 UltraMobile PC (inverted EAPD).

Reported by: benno

17 years agoFix TX/RX checksum offload.
David E. O'Brien [Mon, 27 Nov 2006 04:47:27 +0000 (04:47 +0000)]
Fix TX/RX checksum offload.

Submitted by: Peer Chen at nVIDIA

17 years agotte hash resizing may be causing errors when building - disable for now
Kip Macy [Mon, 27 Nov 2006 02:17:33 +0000 (02:17 +0000)]
tte hash resizing may be causing errors when building - disable for now

17 years agoBack out half of my previous change to support parallel makes.
John Birrell [Mon, 27 Nov 2006 01:35:11 +0000 (01:35 +0000)]
Back out half of my previous change to support parallel makes.
The generated Makefile clashes with the src/bin/sh/Makefile, causing
it to try to use a rule to build something it doesn't need to.

17 years agoSync with the OpenBSD port of RSTP
Andrew Thompson [Mon, 27 Nov 2006 00:35:40 +0000 (00:35 +0000)]
Sync with the OpenBSD port of RSTP
 - use flags rather than sperate ioctls for edge, p2p
 - implement p2p and autop2p flags
 - define large pathcost constant as ULL
 - show bridgeid and rootid in ifconfig

Obtained from: Reyk Floeter <reyk@openbsd.org>

17 years agoReformat to KNF after revision 1.12.
David E. O'Brien [Mon, 27 Nov 2006 00:23:59 +0000 (00:23 +0000)]
Reformat to KNF after revision 1.12.

17 years agoFix the jumbo frame support.
David E. O'Brien [Mon, 27 Nov 2006 00:16:47 +0000 (00:16 +0000)]
Fix the jumbo frame support.

Submitted by: Peer Chen at nVIDIA (reworked by me)

17 years agoGrrr. new .vimrc file is really giving me trouble.. get the whitespace right.
David E. O'Brien [Mon, 27 Nov 2006 00:10:00 +0000 (00:10 +0000)]
Grrr. new .vimrc file is really giving me trouble.. get the whitespace right.

17 years agoFix editor fatfingering in rev 1.9.
David E. O'Brien [Mon, 27 Nov 2006 00:05:18 +0000 (00:05 +0000)]
Fix editor fatfingering in rev 1.9.

17 years agoFix module loading for sparc on machines with greater than 1GB of RAM
Kip Macy [Sun, 26 Nov 2006 23:16:46 +0000 (23:16 +0000)]
Fix module loading for sparc on machines with greater than 1GB of RAM

t1# uname -m
sparc64
t1# sysctl hw.physmem
hw.physmem: 17150509056
t1# kldload reiserfs
t1# kldstat
Id Refs Address            Size     Name
 1    3 0xc0000000 58cc38   kernel
 2    1 0xcbe70000 234000   xfs.ko
 3    1 0xcc0b6000 110000   reiserfs.ko

17 years agonfe_stop() requires lock to be held when calling.
David E. O'Brien [Sun, 26 Nov 2006 23:11:53 +0000 (23:11 +0000)]
nfe_stop() requires lock to be held when calling.

17 years agoDefault the rate sets for 802.11 operating modes so drivers aren't
Sam Leffler [Sun, 26 Nov 2006 22:48:03 +0000 (22:48 +0000)]
Default the rate sets for 802.11 operating modes so drivers aren't
required to.  Note this only happens when drivers don't set them
up before calling ieee80211_ifattach so this change is backwards
compatible.

MFC after: 1 month

17 years agoDeclare hypervisor system initiated reset function
Kip Macy [Sun, 26 Nov 2006 22:47:52 +0000 (22:47 +0000)]
Declare hypervisor system initiated reset function
as needed by the previous commit :-/

17 years agoFix "shutdown -r" and "shutdown -h"
Kip Macy [Sun, 26 Nov 2006 22:31:23 +0000 (22:31 +0000)]
Fix "shutdown -r" and "shutdown -h"
- "shutdown -r" will reset the system
- "shutdown -h" will power off the system
We don't drop into OFW as newer versions of solaris don't do this either

17 years agomodule glue for wlan_amrr
Sam Leffler [Sun, 26 Nov 2006 22:23:30 +0000 (22:23 +0000)]
module glue for wlan_amrr

MFC after: 1 month

17 years agoTeach find_processes() to use ${name}_chroot.
Florent Thoumie [Sun, 26 Nov 2006 22:02:09 +0000 (22:02 +0000)]
Teach find_processes() to use ${name}_chroot.

Obtained from: NetBSD
X-MFC after: 6.2-RELEASE

17 years agoClarify case body indention.
David E. O'Brien [Sun, 26 Nov 2006 21:22:49 +0000 (21:22 +0000)]
Clarify case body indention.

17 years agoInitialize the port info, this shouldnt have been removed in r1.28
Andrew Thompson [Sun, 26 Nov 2006 20:33:29 +0000 (20:33 +0000)]
Initialize the port info, this shouldnt have been removed in r1.28

17 years agoThere is no reason to use __inline here because we are rather far
Yaroslav Tykhiy [Sun, 26 Nov 2006 20:14:47 +0000 (20:14 +0000)]
There is no reason to use __inline here because we are rather far
from a path critical to performance.

17 years agoDon't overflow from the gigabyte scale to the bit scale if the
Yaroslav Tykhiy [Sun, 26 Nov 2006 20:08:40 +0000 (20:08 +0000)]
Don't overflow from the gigabyte scale to the bit scale if the
number to auto-scale is >= 1024 Gb.  Could be triggered on arches
where ifdata counters had 64 bits.

Reported by: Miroslav Slavkov on -net
MFC after: 3 days

17 years agoadd rate knob for net80211 rate control modules
Sam Leffler [Sun, 26 Nov 2006 19:56:25 +0000 (19:56 +0000)]
add rate knob for net80211 rate control modules

MFC after: 1 month

17 years agoadd rate control debug msg bit
Sam Leffler [Sun, 26 Nov 2006 19:55:57 +0000 (19:55 +0000)]
add rate control debug msg bit

MFC after: 1 month

17 years agodevice-independent implementation of AMRR tx rate control algorithm
Sam Leffler [Sun, 26 Nov 2006 19:55:26 +0000 (19:55 +0000)]
device-independent implementation of AMRR tx rate control algorithm

Obtained from: openbsd (w/ mods)
MFC after: 1 month

17 years agoRemove redundant setting of port state.
Andrew Thompson [Sun, 26 Nov 2006 19:49:34 +0000 (19:49 +0000)]
Remove redundant setting of port state.

17 years agoThe real contents of this file were repeated twice.
Yaroslav Tykhiy [Sun, 26 Nov 2006 19:25:10 +0000 (19:25 +0000)]
The real contents of this file were repeated twice.
The repetition was harmless due to a usual #ifndef _FOO_H_ wrapper.
Fortunately, nobody started to hack the second copy,
so just remove it from the file.

MFC after: 3 days

17 years agoBump the libarchive version number, correct the shell hackery to break
Tim Kientzle [Sun, 26 Nov 2006 19:09:42 +0000 (19:09 +0000)]
Bump the libarchive version number, correct the shell hackery to break
the full version down into major/minor values.

17 years agoRemove leading dollar sign in rcvar command output.
Florent Thoumie [Sun, 26 Nov 2006 19:03:19 +0000 (19:03 +0000)]
Remove leading dollar sign in rcvar command output.
There's no dollar use in variable assignment in sh.
Assuming this is can be expected behavior for some
people, this change won't be MFC'ed to RELENG_6.

Discussed with: yar on -rc

17 years agoWrite-blocking cleanup, largely thanks to Colin Percival (cperciva@).
Tim Kientzle [Sun, 26 Nov 2006 19:00:50 +0000 (19:00 +0000)]
Write-blocking cleanup, largely thanks to Colin Percival (cperciva@).
   * If write block size is zero, don't block at all.
     This supports the unusual requirement of applications
     that need "no-delay" writes.
   * Expose _write_finish_entry() to give such applications more
     control over write boundaries.  (Normal applications do not
     need this, as entries are completed automatically.)
   * Correct the type of write callbacks; this is a minor API
     change that does not affect the ABI.
   * Correct the error handling in _write_next_header() around
     completing the previous entry.
   * Correct the documentation for block-size markers:  Remove
     docs for the long-defunct _read_set_block_size(); document
     all of the write block size manipulators.

MFC after: 14 days

17 years agoo From the submitter: dos2unixchr will convert to lower case if
Maxim Konovalov [Sun, 26 Nov 2006 18:49:44 +0000 (18:49 +0000)]
o From the submitter: dos2unixchr will convert to lower case if
LCASE_BASE or LCASE_EXT or both are set.  But dos2unixfn uses
dos2unixchr separately for the basename and the extension.  So if
either LCASE_BASE or LCASE_EXT is set, dos2unixfn will convert both
the basename and extension to lowercase because it is blindly
passing in the state of both flags to dos2unixchr.  The bit masks I
used ensure that only the state of LCASE_BASE gets passed to
dos2unixchr when the basename is converted, and only the state of
LCASE_EXT is passed in when the extension is converted.

PR: kern/86655
Submitted by: Micah Lieske
MFC after: 3 weeks

17 years agouse two stage creation of stp ports, this means that the stp variables can be
Andrew Thompson [Sun, 26 Nov 2006 18:43:48 +0000 (18:43 +0000)]
use two stage creation of stp ports, this means that the stp variables can be
set before the port is marked STP and they will no longer be overwrittten

17 years agoo Add cam to a list of cscope dirs.
Maxim Konovalov [Sun, 26 Nov 2006 18:27:16 +0000 (18:27 +0000)]
o Add cam to a list of cscope dirs.

17 years agoo Style: remove a w/space before a comma.
Maxim Konovalov [Sun, 26 Nov 2006 17:18:31 +0000 (17:18 +0000)]
o Style: remove a w/space before a comma.

17 years agoo Style(9): trim eol w/spaces, w/spaces to tabs, kill an empty line.
Maxim Konovalov [Sun, 26 Nov 2006 17:15:54 +0000 (17:15 +0000)]
o Style(9): trim eol w/spaces, w/spaces to tabs, kill an empty line.

17 years agoo Add a quirk for X-Micro Flash Disk.
Maxim Konovalov [Sun, 26 Nov 2006 17:13:24 +0000 (17:13 +0000)]
o Add a quirk for X-Micro Flash Disk.

PR: usb/96901
Submitted by: Szalai Andras
Reviewed by: nate (a half of year ago)
MFC after: 3 weeks

17 years agoAdd some old includes to remove.
Alexander Leidinger [Sun, 26 Nov 2006 15:04:30 +0000 (15:04 +0000)]
Add some old includes to remove.

Reported by: kris

17 years agoAdd libusb.h and libusbhid.h.
Alexander Leidinger [Sun, 26 Nov 2006 14:43:45 +0000 (14:43 +0000)]
Add libusb.h and libusbhid.h.

17 years agoFix "make checkdpadd".
Ruslan Ermilov [Sun, 26 Nov 2006 14:43:03 +0000 (14:43 +0000)]
Fix "make checkdpadd".

17 years agoah_if.m was removed a while ago
Alexander Leidinger [Sun, 26 Nov 2006 14:41:17 +0000 (14:41 +0000)]
ah_if.m was removed a while ago

17 years ago- When building world WITHOUT_LIBPTHREAD, link libthr to libpthread.
Ruslan Ermilov [Sun, 26 Nov 2006 14:36:34 +0000 (14:36 +0000)]
- When building world WITHOUT_LIBPTHREAD, link libthr to libpthread.
- Don't build ngctl(8) and cached(8) if threading libs aren't built.
- Fix various issues in a cached(8) makefile.

17 years ago- Update the Runtime Configuration, FILES and DIAGNOSTICS sections to
Joel Dahl [Sun, 26 Nov 2006 12:27:02 +0000 (12:27 +0000)]
-  Update the Runtime Configuration, FILES and DIAGNOSTICS sections to
   the new world order.
-  Use our standard section 4 SYNOPSIS.
-  Minor nits.

Reviewed by: brueffer, ariff

17 years agoWelcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
Ariff Abdullah [Sun, 26 Nov 2006 12:24:06 +0000 (12:24 +0000)]
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.

General
-------

- Multichannel safe, endian safe, format safe
   * Large part of critical pcm filters such as vchan.c, feeder_rate.c,
     feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
     using them does not cause the pcm data to be converted to 16bit little
     endian.
   * Macrosses for accessing pcm data safely are defined within sound.h in
     the form of PCM_READ_* / PCM_WRITE_*
   * Currently, most of them are probably limited for mono/stereo handling,
     but the future addition of true multichannel will be much easier.

- Low latency operation
  * Well, this require lot more works to do not just within sound driver,
    but we're heading towards right direction. Buffer/block sizing within
    channel.c is rewritten to calculate precise allocation for various
    combination of sample/data/rate size. As a result, applying correct
    SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
    to what commercial 4front driver do.
  * Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
    result long delay.
  * Eliminate sound truncation if the sound data is too small.
    DIY:
      1) Download / extract
         http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
      2) Do a comparison between "cat state*.au > /dev/dsp" and
         "for x in state*.au ; do cat $x > /dev/dsp ; done"
         - there should be no "perceivable" differences.
    Double close for PR kern/31445.

  CAVEAT: Low latency come with (unbearable) price especially for poorly
          written applications. Applications that trying to act smarter
  by requesting (wrong) blocksize/blockcount will suffer the most.
  Fixup samples/patches can be found at:
  http://people.freebsd.org/~ariff/ports/

- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
  due to closer compatibility with 4front driver.
  Discussed with: marcus@ (long time ago?)

- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
  moved to their own dev sysctl nodes, notably:
  hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
  Bump __FreeBSD_version.

Driver specific
---------------

- Ditto for sysctls.

- snd_atiixp, snd_es137x, snd_via8233, snd_hda
  * Numerous cleanups and fixes.
  * _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
   This was intended for pure debugging and latency measurement, but proven
   good enough in few unexpected and rare cases (such as problematic shared
   IRQ with GIANT devices - USB). Polling can be enabled/disabled through
   dev.pcm.0.polling. Disabled by default.

- snd_ich
  * Fix possible overflow during speed calibration. Delay final
    initialization (pcm_setstatus) after calibration finished.
    PR: kern/100169
    Tested by: Kevin Overman <oberman@es.net>
  * Inverted EAPD for few Nec VersaPro.
    PR: kern/104715
    Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>

Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.

Joel Dahl will do the manpage update.

17 years agoMFP4 (109713):
Alexander Leidinger [Sun, 26 Nov 2006 11:55:48 +0000 (11:55 +0000)]
MFP4 (109713):
Add OSS_GETVERSION ioctl for compatibility.

Some ports expect this and fail to compile on -current ATM.

Submitted by: ryanb

17 years agoAdd entry and no-UHID quirk for I-tuner networks USB-LCD 2x20 as found
Poul-Henning Kamp [Sun, 26 Nov 2006 11:07:37 +0000 (11:07 +0000)]
Add entry and no-UHID quirk for I-tuner networks USB-LCD 2x20 as found
in http://www.mini-box.com/Mini-Box-M200-LCD

17 years ago- remove dead code
Kip Macy [Sun, 26 Nov 2006 07:54:44 +0000 (07:54 +0000)]
- remove dead code
- revert a previous change to pmap_enter where we
  could skip invalidates on unmanaged pages

17 years agodevsw.9 removed
Warner Losh [Sun, 26 Nov 2006 06:52:54 +0000 (06:52 +0000)]
devsw.9 removed

17 years agoUnbreak libarchive on arm. Two parts of libarchive relied on a
Tim Kientzle [Sun, 26 Nov 2006 05:39:28 +0000 (05:39 +0000)]
Unbreak libarchive on arm.  Two parts of libarchive relied on a
traditional shortcut of defining on-disk layouts using structures of
character arrays. Unfortunately, as recently discussed on cvs-all@,
this usage is not actually sanctioned by the standards and
specifically fails on GCC/arm (unless your data structures happen to
be "naturally aligned").

The new code defines offsets/sizes for data fields and accesses
them using explicit pointer arithmetic, instead of casting to
a structure and accessing structure fields.  In particular,
the new code is now clean with WARNS=6 on arm.

MFC after: 14 days

17 years agoadd interrupt cookie hypervisor functions
Kip Macy [Sun, 26 Nov 2006 04:37:49 +0000 (04:37 +0000)]
add interrupt cookie hypervisor functions

17 years agoTweak the comment about mapping a kernel using large pages.
Ruslan Ermilov [Sat, 25 Nov 2006 23:00:46 +0000 (23:00 +0000)]
Tweak the comment about mapping a kernel using large pages.

17 years agodevsw() was removed by phk:
Warner Losh [Sat, 25 Nov 2006 22:36:09 +0000 (22:36 +0000)]
devsw() was removed by phk:

revision 1.199
date: 2004/09/24 08:30:57;  author: phk;  state: Exp;  lines: +0 -1
Remove the cdevsw() function which is now unused.

(the log is wrong, it was really devsw that was removed).

# we really need to actually document the functions in sys/conf.h as well
# as things like d_open...

17 years agoClarify what non-zero return values mean.
Warner Losh [Sat, 25 Nov 2006 22:34:01 +0000 (22:34 +0000)]
Clarify what non-zero return values mean.

17 years ago- In nd6_rtrequest(), when caching an rtentry, don't forget
Ruslan Ermilov [Sat, 25 Nov 2006 20:38:56 +0000 (20:38 +0000)]
- In nd6_rtrequest(), when caching an rtentry, don't forget
  to add a reference to it; otherwise, we could later access
  a freed memory.  This is believed to fix panics some users
  were observing when running route6d(8), and is similar to
  the fix in sys/netinet/if_ether.c,v 1.139 by glebius@.

PR: kern/93910, kern/105437
Testing by: Wojciech Puchar (still ongoing)

- Add rtentry locking to nd6_output() similar to rt_check().

MFC after: 4 days

17 years agoGrr, fix typo introduced in the previous commit.
Christian Brueffer [Sat, 25 Nov 2006 12:12:11 +0000 (12:12 +0000)]
Grr, fix typo introduced in the previous commit.

17 years agoThree minor bug fixes:
Colin Percival [Sat, 25 Nov 2006 07:30:12 +0000 (07:30 +0000)]
Three minor bug fixes:
1. When downloading metadata files, make sure we only download each
file once; without this fix, "freebsd-update fetch" will fail the first
time it is run if there have been no updates yet for the installed
release.
2. If the FOO kernel is installed in /boot/kernel instead of /boot/FOO
and the /boot/FOO directory does not exist, don't try to update
/boot/FOO.  This is an issue only where an update involves adding a new
kernel module.
3. When removing files and directories, operate in reverse
lexographical order, in order to ensure that files are removed before
the directory which contains them.

MFC after: 3 days

17 years agoThe mountroot prompt will drop into ddb if we don't recognize error codes from
Kip Macy [Sat, 25 Nov 2006 06:29:46 +0000 (06:29 +0000)]
The mountroot prompt will drop into ddb if we don't recognize error codes from
getchar correctly - we also need to check for HUP and BREAK

17 years agomark tx/rx descriptors COHERENT; we do not sync changes so on
Sam Leffler [Fri, 24 Nov 2006 22:45:26 +0000 (22:45 +0000)]
mark tx/rx descriptors COHERENT; we do not sync changes so on
architectures like arm this is necessary

MFC after: 1 month

17 years agoMake sure we do not sleep while locks are held. Change the malloc(9)
Christian S.J. Peron [Fri, 24 Nov 2006 22:14:37 +0000 (22:14 +0000)]
Make sure we do not sleep while locks are held. Change the malloc(9)
flags from M_WAITOK to M_NOWAIT. This should not cause any problems
since the calling code appears to properly handle failed allocations.

Discussed with: kmacy

17 years agokernel will not compile without genclock, thus move to DEFAULTS
Kip Macy [Fri, 24 Nov 2006 20:56:43 +0000 (20:56 +0000)]
kernel will not compile without genclock, thus move to DEFAULTS

17 years ago- implement remaining pci functions
Kip Macy [Fri, 24 Nov 2006 20:47:29 +0000 (20:47 +0000)]
- implement remaining pci functions
- fix build errors

17 years agoAdd instructions for creating a ramdisk root image. These likely need
Warner Losh [Fri, 24 Nov 2006 18:53:45 +0000 (18:53 +0000)]
Add instructions for creating a ramdisk root image.  These likely need
to be augmented with instructions for those platforms that have
/boot/loader.

17 years agoImplement mmu functions and cpu_mondo_send
Kip Macy [Fri, 24 Nov 2006 18:50:56 +0000 (18:50 +0000)]
Implement mmu functions and cpu_mondo_send
fix some more kernel compile fallout

17 years agoremove unused reference to tsb pa
Kip Macy [Fri, 24 Nov 2006 18:36:04 +0000 (18:36 +0000)]
remove unused reference to tsb pa

17 years agoConnect four new files to the build.
Tim Kientzle [Fri, 24 Nov 2006 16:35:54 +0000 (16:35 +0000)]
Connect four new files to the build.
PR: bin/86742

17 years agoRemove reference to a man page of pib, an obscure third-party tool, whose port
Pav Lucistnik [Fri, 24 Nov 2006 14:47:25 +0000 (14:47 +0000)]
Remove reference to a man page of pib, an obscure third-party tool, whose port
does not install any manpages.

MFC after: 3 days

17 years agoMake it clear that the device name may contain '/' characters to place
Warner Losh [Fri, 24 Nov 2006 13:10:30 +0000 (13:10 +0000)]
Make it clear that the device name may contain '/' characters to place
a device in a subdirectory.  Also explicitly state that the name is
constructed using fmt like printf would print it.

MFC After: 3 days

17 years agoo Xr sendfile(2).
Maxim Konovalov [Fri, 24 Nov 2006 12:00:02 +0000 (12:00 +0000)]
o Xr sendfile(2).

17 years agoPush Giant a bit further off the NFS server in a number of straight
Robert Watson [Fri, 24 Nov 2006 11:53:16 +0000 (11:53 +0000)]
Push Giant a bit further off the NFS server in a number of straight
forward cases by converting from unconditional acquisition of Giant
around vnode operations to conditional acquisition:

- Remove nfsrv_access_withgiant(), and cause nfsrv_access() to now
  assert that Giant will be held if it is required for the vnode.

- Add nfsrv_fhtovp_locked(), which will drop the NFS server lock if
  required, and modify nfsrv_fhtovp() to conditionally acquire
  Giant if required.

- In the VOP's not dealing with more than one vnode at a time (i.e.,
  not involving a lookup), conditionally acquire Giant.

This removes Giant use for MPSAFE file systems for a number of quite
important RPCs, including getattr, read, write.  It leaves
unconditional Giant acquisitions in vnode operations that interact
with the name space or more than one vnode at a time as these
require further work.

Tested by: kris
Reviewed by: kib

17 years agoNote that, thanks to the work by Alan Cox et al, some arch'es
Yaroslav Tykhiy [Fri, 24 Nov 2006 11:44:19 +0000 (11:44 +0000)]
Note that, thanks to the work by Alan Cox et al, some arch'es
don't need sendfile() buffers any more.

The report on the work referenced can be found at
http://usenix.org/events/usenix05/tech/general/elmeleegy.html

MFC after: 1 week

17 years agoEliminate atomic operations in thread cancellation functions, it should
David Xu [Fri, 24 Nov 2006 09:57:38 +0000 (09:57 +0000)]
Eliminate atomic operations in thread cancellation functions, it should
reduce overheads of cancellation points.

17 years agoMake use of mk/ files from a fresh source tree, rather than
Xin LI [Fri, 24 Nov 2006 09:07:26 +0000 (09:07 +0000)]
Make use of mk/ files from a fresh source tree, rather than
the currently installed ones.  This makes it possible to use
the script against Makefiles that uses bsd.own.mk on an old
system.

Submitted by: bushman

17 years agocomment all remaining documented hypervisor functions except for msi
Kip Macy [Fri, 24 Nov 2006 07:49:15 +0000 (07:49 +0000)]
comment all remaining documented hypervisor functions except for msi
implement performance counter functions

17 years agodocument and comment all functions outside of MMU and MSI services
Kip Macy [Fri, 24 Nov 2006 07:11:24 +0000 (07:11 +0000)]
document and comment all functions outside of MMU and MSI services
from those, implement all those whose arguments don't require save/restore

17 years agoI mistakenly committed the wrong version of my patch (sorry).
Doug Barton [Fri, 24 Nov 2006 06:51:49 +0000 (06:51 +0000)]
I mistakenly committed the wrong version of my patch (sorry).

s/O2/O2Micro, as that's how they seem to prefer it, and remove
what is now one blank line too many.

17 years ago- style: use `=' instead of `+=' for initial assignments.
Ruslan Ermilov [Fri, 24 Nov 2006 06:38:11 +0000 (06:38 +0000)]
- style: use `=' instead of `+=' for initial assignments.

- don't add generated sources' objects to OBJS explicitly;
  GENSRCS is part of SRCS so they were already in OBJS.

17 years agoFill in a couple of missing casts: clarify one narrowing conversion
Tim Kientzle [Fri, 24 Nov 2006 05:48:04 +0000 (05:48 +0000)]
Fill in a couple of missing casts: clarify one narrowing conversion
and correct the use of unary minus with an unsigned value.  (The unary
minus here is actually being used as a bitwise operation, which is
unusual enough to deserve a clarifying cast.)

17 years agoA few minor clarifications and corrections.
Tim Kientzle [Fri, 24 Nov 2006 05:41:16 +0000 (05:41 +0000)]
A few minor clarifications and corrections.

17 years agoA few minor corrections to the libarchive.3 overview page.
Tim Kientzle [Fri, 24 Nov 2006 05:37:38 +0000 (05:37 +0000)]
A few minor corrections to the libarchive.3 overview page.

17 years agoDocument the new _open_FILE() and _open_memory() interfaces.
Tim Kientzle [Fri, 24 Nov 2006 05:34:23 +0000 (05:34 +0000)]
Document the new _open_FILE() and _open_memory() interfaces.

PR: bin/86742