]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
11 years ago - pctrie really only requires two byte alignment so that there is a single
jeff [Sun, 12 May 2013 20:44:28 +0000 (20:44 +0000)]
 - pctrie really only requires two byte alignment so that there is a single
   bit available for a flag in the pointer.  However, it felt more correct
   to enforce natural alignment of the key pointer.  Unfortunately on
   32bit architectures 64bit integers are not always naturally aligned.
   Change the assert to enforce only 32bit alignment of the 64bit key for
   now to fix the build.  A more correct fix would be to properly sort
   the struct buf fields which definitely suffer from bloat due to padding.

11 years agoRefactor vm_page_alloc()'s interactions with vm_reserv_alloc_page() and
alc [Sun, 12 May 2013 16:50:18 +0000 (16:50 +0000)]
Refactor vm_page_alloc()'s interactions with vm_reserv_alloc_page() and
vm_page_insert() so that (1) vm_radix_lookup_le() is never called while the
free page queues lock is held and (2) vm_radix_lookup_le() is called at most
once.  This change reduces the average time that the free page queues lock
is held by vm_page_alloc() as well as vm_page_alloc()'s average overall
running time.

Sponsored by: EMC / Isilon Storage Division

11 years agoFix several typos
eadler [Sun, 12 May 2013 16:43:26 +0000 (16:43 +0000)]
Fix several typos

PR: kern/176054
Submitted by: Christoph Mallon <christoph.mallon@gmx.de>
MFC after: 3 days

11 years agohead -n doesn't take negative arguments on FreeBSD, so instead use sed '$d'
markj [Sun, 12 May 2013 16:29:09 +0000 (16:29 +0000)]
head -n doesn't take negative arguments on FreeBSD, so instead use sed '$d'
to remove the last line of a file.

11 years agoBring back part of r249367 by adding DTrace's temporal option, which allows
markj [Sun, 12 May 2013 16:26:33 +0000 (16:26 +0000)]
Bring back part of r249367 by adding DTrace's temporal option, which allows
users to guarantee that the output of DTrace scripts will be time-ordered.
This option is enabled by adding the line

  #pragma D option temporal

to the beginning of a script, or by adding '-x temporal' to the arguments of
dtrace(1).

This change fixes a bug in the original port of the temporal option. This
bug was causing some assertions to fail, so they had been disabled; in this
revision the assertions are working properly and are enabled.

The DTrace version number has been bumped from 1.9.0 to 1.9.1 to reflect
the language change that's being introduced.

This change corresponds to part of illumos-gate commit e5803b76927480:
  3021 option for time-ordered output from dtrace(1M)

Reviewed by: pfg
Obtained from: illumos
MFC after: 1 month

11 years agoFix typo in UPDATING
eadler [Sun, 12 May 2013 16:07:23 +0000 (16:07 +0000)]
Fix typo in UPDATING

Submitted by: Rainer Hurling <rhurlin@gwdg.de>

11 years agofdescfs: Supply a real value for d_type in readdir.
jilles [Sun, 12 May 2013 15:44:49 +0000 (15:44 +0000)]
fdescfs: Supply a real value for d_type in readdir.

All the fdescfs nodes (except . and ..) appear as character devices to
stat(), so DT_CHR is correct.

11 years agoAdd support for "d" floating-point suffix, as defined by draft N1312
pfg [Sun, 12 May 2013 15:29:35 +0000 (15:29 +0000)]
Add support for "d" floating-point suffix, as defined by draft N1312
of TR 24732. Emit pedantic warning if the feature is being used.

Should solve GCC bug 39027.

Obtained from: OpenBSD
MFC after: 2 weeks

11 years agoMake newsyslog compress logs with xz instead of bzip2 to save space.
eadler [Sun, 12 May 2013 15:23:59 +0000 (15:23 +0000)]
Make newsyslog compress logs with xz instead of bzip2 to save space.

PR: conf/178504
Submitted by: ak
Reviewed by: smh

11 years agoRevert r250394, which disabled snmp_hast due to libl dependency issue --
trociny [Sun, 12 May 2013 13:42:49 +0000 (13:42 +0000)]
Revert r250394, which disabled snmp_hast due to libl dependency issue --
the dependency has been removed in r250503.

11 years agoSuppress error printing for "PREVENT ALLOW MEDIUM REMOVAL" on da open.
mav [Sun, 12 May 2013 09:33:33 +0000 (09:33 +0000)]
Suppress error printing for "PREVENT ALLOW MEDIUM REMOVAL" on da open.
Change at r250208 exposed more errors here, hidden before.  The same flag
is used in cd driver.

11 years agoTypo.
pluknet [Sun, 12 May 2013 05:54:58 +0000 (05:54 +0000)]
Typo.

11 years agoAdd the letter 'x' to the font iso05-8x16.fnt
eadler [Sun, 12 May 2013 04:34:55 +0000 (04:34 +0000)]
Add the letter 'x' to the font iso05-8x16.fnt

PR: 169797
Submitted by: Takefu <takefu@airport.fm>
MFC after: 3 days

11 years ago - Add a new general purpose path-compressed radix trie which can be used
jeff [Sun, 12 May 2013 04:05:01 +0000 (04:05 +0000)]
 - Add a new general purpose path-compressed radix trie which can be used
   with any structure containing a uint64_t index.  The tree code
   auto-generates type safe wrappers.
 - Eliminate the buf splay and replace it with pctrie.  This is not only
   significantly faster with large files but also allows for the possibility
   of shared locking.

Reviewed by:    alc, attilio
Sponsored by:   EMC / Isilon Storage Division

11 years agoUpdate the gcc43 changelog.
pfg [Sun, 12 May 2013 03:36:28 +0000 (03:36 +0000)]
Update the gcc43 changelog.

The issue solved in r250392 actually originated upstream and was
fixed in upstream gcc43 branch in 2006/10/24 (under GPLv2).

Register it in the appropriate ChangeLog for reference.

MFC after: 3 days

11 years agoSome filesystems (NFS in particular) do not fill out the d_type field when
markj [Sun, 12 May 2013 01:58:04 +0000 (01:58 +0000)]
Some filesystems (NFS in particular) do not fill out the d_type field when
returning directory entries through readdir(3). In this case we need to
obtain the file type ourselves; otherwise newsyslog -t will not be able to
find archived log files and will fail to both delete old log files and to
do interval-based rotations properly.

Reported by: jilles
Reviewed by: jilles
MFC after: 2 weeks

11 years agoTidy up some CVS workarounds.
peter [Sun, 12 May 2013 01:53:47 +0000 (01:53 +0000)]
Tidy up some CVS workarounds.

11 years agoTurns out that there really isn't an RTC chip on this board, at all.
imp [Sun, 12 May 2013 01:29:18 +0000 (01:29 +0000)]
Turns out that there really isn't an RTC chip on this board, at all.

11 years agoComment.
dteske [Sun, 12 May 2013 00:50:18 +0000 (00:50 +0000)]
Comment.

11 years agoFix i18n violations in the package management module. A few words like
dteske [Sun, 12 May 2013 00:46:18 +0000 (00:46 +0000)]
Fix i18n violations in the package management module. A few words like
`packages', `installed', and `selected' were not internationalized.

11 years agoRemove duplicated string.
dteske [Sun, 12 May 2013 00:44:40 +0000 (00:44 +0000)]
Remove duplicated string.

11 years agoComments.
dteske [Sun, 12 May 2013 00:40:49 +0000 (00:40 +0000)]
Comments.

11 years agoWhitespace.
dteske [Sun, 12 May 2013 00:39:25 +0000 (00:39 +0000)]
Whitespace.

11 years agoUnconditionally install 210.backup-aliases as many MTAs other than
eadler [Sat, 11 May 2013 23:55:43 +0000 (23:55 +0000)]
Unconditionally install 210.backup-aliases as many MTAs other than
sendmail support the use of /etc/aliases.

PR: conf/176098
Submitted by: ak
MFC after: 2 weeks

11 years agoIntel's 320-series and 510-series SSDs advertise 512-byte sectors
eadler [Sat, 11 May 2013 23:13:49 +0000 (23:13 +0000)]
Intel's 320-series and 510-series SSDs advertise 512-byte sectors
sizes for both logical and physical. Add ADA_Q_4K quirks
for both.

PR: kern/178040
Submitted by: Jeremy Chadwick <jdc@koitsu.org>

11 years agoDon't use the old stack-walking code with
kientzle [Sat, 11 May 2013 22:32:43 +0000 (22:32 +0000)]
Don't use the old stack-walking code with
EABI ARM kernels or clang-compiled ARM kernels.

This fixes a crash seen in clang-compiled ARM
kernels that include WITNESS.

This code could be easily modified to walk the stack
for current clang-generated code (including EABI)
but Andrew Turner has raised concerns that the
stack frame currently emitted by clang isn't actually
required by EABI so such a change might cause problems
down the road.

In case anyone wants to experiment, the change
to support current clang-compiled kernels
involves simply setting FR_RFP=0 and FR_SCP=1.

11 years agoAdd simple testcases for fcntl(F_DUP2FD_CLOEXEC).
jilles [Sat, 11 May 2013 22:13:24 +0000 (22:13 +0000)]
Add simple testcases for fcntl(F_DUP2FD_CLOEXEC).

11 years agoTeach wpa_supplicant to properly compute wireless signal strength.
eadler [Sat, 11 May 2013 20:55:14 +0000 (20:55 +0000)]
Teach wpa_supplicant to properly compute wireless signal strength.

PR: bin/170631
Submitted by: Daan Vreeken <Daan@vitsch.nl>
Reviewed by: adrian

11 years agosh: Remove linked list of stack marks.
jilles [Sat, 11 May 2013 20:51:00 +0000 (20:51 +0000)]
sh: Remove linked list of stack marks.

The linked list of stack marks may cause problems if the allocation stack is
used between an exception and a higher-level popstackmark(), as it may then
touch a stack mark that is local to a function which has returned.

Also, the adjustment compares to a pointer passed to realloc(), which is
undefined behaviour.

Instead of adjusting stack marks when reallocating stack blocks, ensure that
such an adjustment is never necessary by fixing a small piece of memory in
place at a stack mark. This also simplifies the code.

To avoid the problems reported in bin/175922, it remains necessary to call
setstackmark() after popstackmark() if the stack mark remains in use.

11 years agoAdd IFF_MONITOR support to gre(4).
hrs [Sat, 11 May 2013 19:05:38 +0000 (19:05 +0000)]
Add IFF_MONITOR support to gre(4).

Tested by: Chip Marshall
MFC after: 1 week

11 years agoReturn meaningful error code from pf_state_key_attach() and
glebius [Sat, 11 May 2013 18:06:51 +0000 (18:06 +0000)]
Return meaningful error code from pf_state_key_attach() and
pf_state_insert().

11 years agoBetter debug message.
glebius [Sat, 11 May 2013 18:03:36 +0000 (18:03 +0000)]
Better debug message.

11 years agoTo reduce the amount of arithmetic performed in the various radix tree
alc [Sat, 11 May 2013 18:01:41 +0000 (18:01 +0000)]
To reduce the amount of arithmetic performed in the various radix tree
functions, reverse the numbering scheme for the levels.  The highest
numbered level in the tree now appears near the root instead of the leaves.

Sponsored by: EMC / Isilon Storage Division

11 years agoFix DIOCADDSTATE operation.
glebius [Sat, 11 May 2013 17:58:26 +0000 (17:58 +0000)]
Fix DIOCADDSTATE operation.

11 years agoInvalid creatorid is always EINVAL, not only when we are in verbose mode.
glebius [Sat, 11 May 2013 17:57:52 +0000 (17:57 +0000)]
Invalid creatorid is always EINVAL, not only when we are in verbose mode.

11 years agoAdd simple testcases for fcntl(F_DUPFD_CLOEXEC).
jilles [Sat, 11 May 2013 16:31:41 +0000 (16:31 +0000)]
Add simple testcases for fcntl(F_DUPFD_CLOEXEC).

11 years agoAdd missing argument to fcntl(F_DUPFD) in regression test.
jilles [Sat, 11 May 2013 15:45:44 +0000 (15:45 +0000)]
Add missing argument to fcntl(F_DUPFD) in regression test.

11 years agoAdd some missing DPADD.
antoine [Sat, 11 May 2013 13:46:05 +0000 (13:46 +0000)]
Add some missing DPADD.

11 years agoAdd compile_et to kerberos bootstrap tools.
antoine [Sat, 11 May 2013 13:44:11 +0000 (13:44 +0000)]
Add compile_et to kerberos bootstrap tools.
This makes it possible to build a regular world on a host without kerberos.

PR: conf/178421
Submitted by: Garrett Cooper
MFC after: 1 month

11 years agoThe M_VESA_MODE_MAX is latest valid mode - not the first invalid.
eadler [Sat, 11 May 2013 13:32:38 +0000 (13:32 +0000)]
The M_VESA_MODE_MAX is latest valid mode - not the first invalid.
Permit its use in vidcontrol -i

PR: bin/119483
Submitted by: Dan Lukes <dan@obluda.cz>
Arrival-Date: Wed Jan 09 10:30:01 UTC 2008
Reviewed by: ed
MFC after: 3 days

11 years agoDisable sending Early R_OK on SiI3726/SiI3826 port multipliers.
mav [Sat, 11 May 2013 13:21:31 +0000 (13:21 +0000)]
Disable sending Early R_OK on SiI3726/SiI3826 port multipliers.
With "cached read" HDD testing and multiple ports busy on a SATA
host controller, 3726/3826 PMP will very rarely drop a deferred
R_OK that was intended for the host. Symptom will be all 5 drives
under test will timeout, get reset, and recover.

Submitted by: Rich Futyma <rich.futyma@sanmina.com>
MFC after: 2 weeks

11 years ago- Fix nullfs vnode reference leak in nullfs_reclaim_lowervp(). The
kib [Sat, 11 May 2013 11:17:44 +0000 (11:17 +0000)]
- Fix nullfs vnode reference leak in nullfs_reclaim_lowervp().  The
  null_hashget() obtains the reference on the nullfs vnode, which must
  be dropped.

- Fix a wart which existed from the introduction of the nullfs
  caching, do not unlock lower vnode in the nullfs_reclaim_lowervp().
  It should be innocent, but now it is also formally safe.  Inform the
  nullfs_reclaim() about this using the NULLV_NOUNLOCK flag set on
  nullfs inode.

- Add a callback to the upper filesystems for the lower vnode
  unlinking. When inactivating a nullfs vnode, check if the lower
  vnode was unlinked, indicated by nullfs flag NULLV_DROP or VV_NOSYNC
  on the lower vnode, and reclaim upper vnode if so.  This allows
  nullfs to purge cached vnodes for the unlinked lower vnode, avoiding
  excessive caching.

Reported by: G??ran L??wkrantz <goran.lowkrantz@ismobile.com>
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks

11 years agoRate limit the number of remotely triggered ARP log messages
glebius [Sat, 11 May 2013 10:51:32 +0000 (10:51 +0000)]
Rate limit the number of remotely triggered ARP log messages
to 1 log message per second.

11 years agoGet rid of libl dependency. We needed it only to provide yywrap. But
trociny [Sat, 11 May 2013 09:51:44 +0000 (09:51 +0000)]
Get rid of libl dependency.  We needed it only to provide yywrap.  But
yywrap is not necessary when parsing a single hast.conf file.

Suggested by: kib
Reviewed by: pjd

11 years agoFix several standard extended feature bits.
rpaulo [Sat, 11 May 2013 01:31:51 +0000 (01:31 +0000)]
Fix several standard extended feature bits.

Submitted by: Oliver Pinter <oliver.pntr at gmail.com>

11 years agoAdding a detach method to p4tcc driver.
hiren [Fri, 10 May 2013 22:43:27 +0000 (22:43 +0000)]
Adding a detach method to p4tcc driver.

PR: 118739
Submitted by: Dan Lukes <dan@obluda.cz> (earlier version)
Reviewed by: jhb
Approved by: sbruno (mentor)
MFC after: 1 week

11 years ago REST IN PEACE
delphij [Fri, 10 May 2013 20:02:15 +0000 (20:02 +0000)]
  REST IN PEACE
     Alexander Botero-Lowry

Born June 2, 1986 in Austin, Texas
Died August 24, 2012 in San Francisco, California

Thank you for your contributions, you will be
greatly missed.

11 years agoMake dd's signal handler async safe.
eadler [Fri, 10 May 2013 18:43:36 +0000 (18:43 +0000)]
Make dd's signal handler async safe.

PR: bin/75258
Submitted by: "Oleg V. Nauman" <oleg@reis.zp.ua>
Arrival Date: Sun Dec 19 14:50:21 GMT 2004
Reviewed by: mjg, jhb
Reviewed by: jilles (earlier version)
MFC after: 1 week

11 years agoHonor the net.inet6.ip6.v6only sysctl variable and the IPV6_V6ONLY
tuexen [Fri, 10 May 2013 18:09:38 +0000 (18:09 +0000)]
Honor the net.inet6.ip6.v6only sysctl variable and the IPV6_V6ONLY
socket option for SCTP sockets in the same way as for UDP or TCP
sockets.

MFC after: 2 weeks

11 years agoPermit relative paths with mergemaster -m
eadler [Fri, 10 May 2013 17:42:13 +0000 (17:42 +0000)]
Permit relative paths with mergemaster -m

Reviewed by: dim, brooks
MFC after: 3 days

11 years agoAdd option to lockf to avoid creating a file if it does not exist.
eadler [Fri, 10 May 2013 17:30:29 +0000 (17:30 +0000)]
Add option to lockf to avoid creating a file if it does not exist.

PR: bin/170775
Submitted by: Matthew Story <matthewstory@gmail.com>
Reviewed by: scottl
MFC after: 1 week

11 years agoFxi a bunch of typos.
eadler [Fri, 10 May 2013 16:41:26 +0000 (16:41 +0000)]
Fxi a bunch of typos.

PR: misc/174625
Submitted by: Jeremy Chadwick <jdc@koitsu.org>

11 years agoAdd the letter 'q' to the font iso05-8x16.fnt
eadler [Fri, 10 May 2013 16:30:39 +0000 (16:30 +0000)]
Add the letter 'q' to the font iso05-8x16.fnt

PR: 72585
PR: 169797
Submitted by: Takefu <takefu@airport.fm>
Submitted by: Satoshi KImura <satosi.kimura@nifty.ne.jp>

11 years agoBring /etc/protocols up to date.
eadler [Fri, 10 May 2013 13:57:44 +0000 (13:57 +0000)]
Bring /etc/protocols up to date.

PR: conf/175397
Submitted by: ak

11 years agoBump .Dd for recent content change.
eadler [Fri, 10 May 2013 12:49:16 +0000 (12:49 +0000)]
Bump .Dd for recent content change.

Reported by: delphij

11 years agoMake sure the holding descriptor and link pointer are both freed during
adrian [Fri, 10 May 2013 10:06:45 +0000 (10:06 +0000)]
Make sure the holding descriptor and link pointer are both freed during
a non-loss reset.

When the drain functions are called, the holding descriptor and link pointers
are NULLed out.

But when the processq function is called during a non-loss reset, this
doesn't occur.  So the next time a DMA occurs, it's chained to a descriptor
that no longer exists and the hardware gets angry.

Tested:

* AR5416, STA mode; use sysctl dev.ath.X.forcebstuck=1 to force a non-loss
  reset.

TODO:

* Further AR9380 testing just to check that the behaviour for the EDMA
  chips is sane.

PR: kern/178477

11 years agoCheck the return value of sbuf_finish().
jh [Fri, 10 May 2013 09:58:32 +0000 (09:58 +0000)]
Check the return value of sbuf_finish().

MFC after: 1 week

11 years agoFix a VAP BSS node reference in the HT code to actually take a reference
adrian [Fri, 10 May 2013 09:37:58 +0000 (09:37 +0000)]
Fix a VAP BSS node reference in the HT code to actually take a reference
before using said node.

The "blessed" way here is to take a node reference before referencing
anything inside the node, otherwise the node can be freed between
the time the pointer is copied/dereferenced and the time the node contents
are used.

This mirrors fixes that I've done elsewhere in the net80211/driver
stack.

PR: kern/178470

11 years agoanother minor bugfix in the memory allocator, this time in the free routine.
luigi [Fri, 10 May 2013 08:46:10 +0000 (08:46 +0000)]
another minor bugfix in the memory allocator, this time in the free routine.

11 years agoMake the read/write routines default to the newer DS1337 part, but keep
imp [Fri, 10 May 2013 06:28:01 +0000 (06:28 +0000)]
Make the read/write routines default to the newer DS1337 part, but keep
the supported bits board specific.

11 years agoMove 'compatible' line out of the common am335x.dtsi and into
kientzle [Fri, 10 May 2013 05:34:08 +0000 (05:34 +0000)]
Move 'compatible' line out of the common am335x.dtsi and into
the beaglebone-specific .dts file.

Add a new .dts for the BeagleBone Black with more memory,
slightly different pinmux initialization, and with mmchs1
configured (though the latter doesn't quite work yet).

11 years agoAs requested, move this back to opt-in and list my boards.
imp [Fri, 10 May 2013 04:49:40 +0000 (04:49 +0000)]
As requested, move this back to opt-in and list my boards.

11 years agoRemove the commented out code I just committed. If we need it, I'll
imp [Fri, 10 May 2013 04:30:14 +0000 (04:30 +0000)]
Remove the commented out code I just committed. If we need it, I'll
bring it back uncommented our, or rewrite it.

11 years agoImplement 'split -d' which allows a numeric suffix instead of an
eadler [Fri, 10 May 2013 04:23:03 +0000 (04:23 +0000)]
Implement 'split -d' which allows a numeric suffix instead of an
alphabetic one.

PR: bin/116209
Submitted by: Marcin Gryszkalis <mg@fork.pl> (adapted from)
Reviewed by: will
MFC after: 1 week

11 years agoStyle(9) changes fo xargs.c
eadler [Fri, 10 May 2013 03:49:05 +0000 (03:49 +0000)]
Style(9) changes fo xargs.c

11 years agoAdd support for 'dmesg -c' which clears the dmesg buffer after it has
eadler [Fri, 10 May 2013 03:42:48 +0000 (03:42 +0000)]
Add support for 'dmesg -c' which clears the dmesg buffer after it has
been printed.  This provides compatibility with other *nix systems
(including Linux).

While here use stdbool booleans for 'all'.

PR: bin/178295
Submitted by: Levent Serinol <lserinol@gmail.com>
Reviewed by: will

11 years agoAdd commented out OCTEON_VENDOR_GEFES
imp [Fri, 10 May 2013 03:05:44 +0000 (03:05 +0000)]
Add commented out OCTEON_VENDOR_GEFES

11 years agoAdd support from GE Intelligent Platform Cavium Octeon boards. Add
imp [Fri, 10 May 2013 03:04:45 +0000 (03:04 +0000)]
Add support from GE Intelligent Platform Cavium Octeon boards. Add
options OCTEON_VENDOR_GEFES to enable support for these boards, to
match changes that GE publishes to the Octeon Simple Executive. Since
board types overlap with other boards, it is unlikely that we will
properly boot on other Octeon boards with OCTEON_VENDOR_GEFES enabled.
Tested extensively on the WANIC 6354, but I retained support for all
the other models. Some features need changes in the base kernel, and
those are in progress.

11 years agoSupport array-type of stats in bhyve.
neel [Fri, 10 May 2013 02:59:49 +0000 (02:59 +0000)]
Support array-type of stats in bhyve.

An array-type stat in vmm.ko is defined as follows:
VMM_STAT_ARRAY(IPIS_SENT, VM_MAXCPU, "ipis sent to vcpu");

It is incremented as follows:
vmm_stat_array_incr(vm, vcpuid, IPIS_SENT, array_index, 1);

And output of 'bhyvectl --get-stats' looks like:
ipis sent to vcpu[0]     3114
ipis sent to vcpu[1]     0

Reviewed by: grehan
Obtained from: NetApp

11 years agoBoth my EBH5200 and GE WANIC 6354 have a RTC as well. It looks from
imp [Fri, 10 May 2013 02:57:46 +0000 (02:57 +0000)]
Both my EBH5200 and GE WANIC 6354 have a RTC as well. It looks from
the Linux tree that they always include this chip in their FDT, so
make support for the ds1337 opt-out rather than opt-in. Now my boards
boot with the correct time.

11 years agoRetire write-only PCB_GS32BIT pcb flag on amd64.
dchagin [Thu, 9 May 2013 21:42:43 +0000 (21:42 +0000)]
Retire write-only PCB_GS32BIT pcb flag on amd64.

11 years agoFix two typos
eadler [Thu, 9 May 2013 21:29:16 +0000 (21:29 +0000)]
Fix two typos

Reviewed by: jilles

11 years agoposix_spawn_file_actions_adddup2(3): Document difference with dup2().
jilles [Thu, 9 May 2013 21:29:13 +0000 (21:29 +0000)]
posix_spawn_file_actions_adddup2(3): Document difference with dup2().

The ability to clear a file descriptor's close-on-exec flag via
posix_spawn_file_actions_adddup2() is in fact proposed in Austin Group issue
#411.

MFC after: 1 week

11 years agoMFp4 CH=227440:
bz [Thu, 9 May 2013 20:13:43 +0000 (20:13 +0000)]
MFp4 CH=227440:

  Convert the structures to C99 style initialisation, which makes it
  a lot easier to check that all of them are set and to generate a
  derived template from them.

Sponsored by: DARPA, AFRL
MFC after: 2 weeks

11 years agoAdd some missing arge MDIO hints.
adrian [Thu, 9 May 2013 19:37:00 +0000 (19:37 +0000)]
Add some missing arge MDIO hints.

Submitted by: Luiz Otavio O Souza <loos.br@gmail.com>

11 years agoRevision 233677 broke certain machines. Specifically, if the firmware/BIOS
jhb [Thu, 9 May 2013 19:24:50 +0000 (19:24 +0000)]
Revision 233677 broke certain machines.  Specifically, if the firmware/BIOS
assigned conflicting ranges to BARs then leaving the BARs alone could
result in one device stealing mmio accesses intended to go to a second
device.  Prior to 233677 the PCI bus driver attempted to handle this case
by clearing the BAR to 0 depending on BARs based at 0 not decoding (which
is not guaranteed to be true).  Now when a conflicting BAR is detected the
following steps are taken:

 1) If hw.pci.realloc_bars (a new tunable) is enabled (default is enabled),
    then ignore the current BAR setting from the firmware and attempt to
    allocate a fresh resource range for the BAR.

 2) If 1) failed (or was disabled), disable decoding for the relevant
    BAR type (e.g. disable mem decoding for a memory BAR) and emit a
    warning if booting verbose.

Tested by: Alex Keda <admin@lissyara.su>
MFC after: 1 week

11 years agoCorrect the type for the literal used on the left side of the shift up
kib [Thu, 9 May 2013 17:25:29 +0000 (17:25 +0000)]
Correct the type for the literal used on the left side of the shift up
to 63 bit positions.

Do not fill the save area and do not set the saved bit in the xstate
bit vector for the state which is not marked as enabled in xsave_mask.

Reported and tested by: Jim Ohlstein <jim@ohlste.in>
MFC after: 3 days

11 years agoif_lem.c: make sure that lem_rxeof() can drain the entire rx queue
luigi [Thu, 9 May 2013 17:07:30 +0000 (17:07 +0000)]
if_lem.c:  make sure that lem_rxeof() can drain the entire rx queue
irrespective of the setting of lem_rx_process_limit, while
giving a chance to the taskqueue scheduler to act after
each chunk.
This makes lem_rxeof similar to the one in if_em.c and if_igb.c .

if_lem.c and if_em.c: add a sysctl to manually configure the
'itr' moderation register.

Approved by: Jack Vogel

11 years agosimplify the code to initialize the RDT while in netmap mode.
luigi [Thu, 9 May 2013 16:57:02 +0000 (16:57 +0000)]
simplify the code to initialize the RDT while in netmap mode.

11 years agoposix_spawn_file_actions_addopen(3): Correct error for bad file descriptor.
jilles [Thu, 9 May 2013 16:42:01 +0000 (16:42 +0000)]
posix_spawn_file_actions_addopen(3): Correct error for bad file descriptor.

As per POSIX.1-2008, posix_spawn_file_actions_add* return [EBADF] if a file
descriptor is negative, not [EINVAL]. The bug was only in the manual page;
the code is correct.

MFC after: 1 week

11 years agoAdd option WITNESS_NO_VNODE to suppress printing LORs between VNODE
marcel [Thu, 9 May 2013 16:28:18 +0000 (16:28 +0000)]
Add option WITNESS_NO_VNODE to suppress printing LORs between VNODE
locks. To support this, VNODE locks are created with the LK_IS_VNODE
flag. This flag is propagated down using the LO_IS_VNODE flag.

Note that WITNESS still records the LOR. Only the printing and the
optional entering into the kernel debugger is bypassed with the
WITNESS_NO_VNODE option.

11 years agoFix a bug that would cause the category menu to display the wrong value
dteske [Thu, 9 May 2013 16:09:39 +0000 (16:09 +0000)]
Fix a bug that would cause the category menu to display the wrong value
for number of packages available in the "All" category. Problem caused by
re-using a variable that was still needed; fixed by variable name change.

11 years agoItem 1 in r248830 causes earlier exits from the sendfile(2), before
kib [Thu, 9 May 2013 16:05:51 +0000 (16:05 +0000)]
Item 1 in r248830 causes earlier exits from the sendfile(2), before
all requested data was sent.  The reason is that xfsize <= 0 condition
must not be tested at all if space == loopbytes.  Otherwise, the done
is set to 1, and sendfile(2) is aborted too early.

Instead of moving the condition to exiting the inner loop after the
xfersize check, directly check for the completed transfer before the
testing of the available space in the socket buffer, and revert item 1
of r248830.  It is arguably another bug to sleep waiting for socket
buffer space (or return EAGAIN for non-blocking socket) if all bytes
are already transferred.

Reported by: pho
Discussed with: scottl, gibbs
Tested by: scottl (stable/9 backport), pho

11 years agoUpdate the holding buffer locking for EDMA.
adrian [Thu, 9 May 2013 15:57:55 +0000 (15:57 +0000)]
Update the holding buffer locking for EDMA.

11 years agoRemove duplicated copies of various NetBSD compatibility shims used by
brooks [Thu, 9 May 2013 14:43:36 +0000 (14:43 +0000)]
Remove duplicated copies of various NetBSD compatibility shims used by
makefs and use libnetbsd, contrib/mtree, and contrib/mknod instead.

Sponsored by: DARPA, AFRLo
MFC after: 1 month

11 years agowordexp(): Simplify code by deferring work to sh.
jilles [Thu, 9 May 2013 11:43:45 +0000 (11:43 +0000)]
wordexp(): Simplify code by deferring work to sh.

11 years agoMove snmp_hast manual to section 3, where all other manual pages for
trociny [Thu, 9 May 2013 11:10:46 +0000 (11:10 +0000)]
Move snmp_hast manual to section 3, where all other manual pages for
bsnmp modules are located.

Section 3 (Library Functions) looks wrong for this manual page, which
contains only module description, that is why initially it was located
to section 8 (System Manager's Manual).  On the other hand manual
pages for all other bsnmpd modules are already located in the section
3, and having all pages in one section looks more consistent.  Also,
currently, similarly to manuals for other modules, snmp_hast manual
contains LIBRARY section, which is not good style for section 8.

Requested by: joel
X-MFC with: r250379

11 years agoFix return value for setcontext and swapcontext.
davidxu [Thu, 9 May 2013 04:41:03 +0000 (04:41 +0000)]
Fix return value for setcontext and swapcontext.

11 years agoStart sentences on new lines.
eadler [Thu, 9 May 2013 02:04:28 +0000 (02:04 +0000)]
Start sentences on new lines.

11 years agoGeneralize the bitset operations, present in cpuset and offer a KPI to
attilio [Thu, 9 May 2013 00:04:59 +0000 (00:04 +0000)]
Generalize the bitset operations, present in cpuset and offer a KPI to
redefine such operations for different consumers.
This will be used when NUMA support will be finished and numaset
will need to be used.

Sponsored by: EMC / Isilon storage division
Obtained from: jeff
Reviewed by: alc

11 years agoUnbreak build - this can only build on i386 as-is due to use of libl in
peter [Wed, 8 May 2013 23:30:24 +0000 (23:30 +0000)]
Unbreak build - this can only build on i386 as-is due to use of libl in
a .so file, and we currently (intentionally) only build libl.a.

11 years agoWhen gcc 4.2.0 was imported, the 3DNow intrinsics header <mm3dnow.h> was
dim [Wed, 8 May 2013 22:54:53 +0000 (22:54 +0000)]
When gcc 4.2.0 was imported, the 3DNow intrinsics header <mm3dnow.h> was
not added to the list of intrinsics headers to install, so belatedly (5
years :) add it.

MFC after: 1 week

11 years agoFor some reason, the gcc intrinsics header tmmintrin.h was imported with
dim [Wed, 8 May 2013 22:50:36 +0000 (22:50 +0000)]
For some reason, the gcc intrinsics header tmmintrin.h was imported with
two copies of itself pasted together.  Remove the extraneous copy.

MFC after: 3 days

11 years agoFix the holding descriptor logic to actually be "right" (for values
adrian [Wed, 8 May 2013 21:23:51 +0000 (21:23 +0000)]
Fix the holding descriptor logic to actually be "right" (for values
of "right".)

Flip back on the "always continue TX DMA using the holding descriptor"
code - by always setting ATH_BUF_BUSY and never setting axq_link to NULL.

Since the holding descriptor is accessed via txq->axq_link and _that_
is done behind the TXQ lock rather than the TX path lock, the holding
descriptor stuff itself needs to be behind the TXQ lock.

So, do the mental gymnastics needed to do this.

I've not seen any of the hardware failures that I was seeing when
I last tried to do this.

Tested:

* AR5416, STA mode

11 years agoMinor improvements.
joel [Wed, 8 May 2013 21:07:11 +0000 (21:07 +0000)]
Minor improvements.

11 years agoRevert r250339 as apparently it is more clutter than help.
attilio [Wed, 8 May 2013 21:06:47 +0000 (21:06 +0000)]
Revert r250339 as apparently it is more clutter than help.

Sponsored by: EMC / Isilon storage division
Requested by: jhb

11 years agoRemove EOL whitespace.
joel [Wed, 8 May 2013 21:04:19 +0000 (21:04 +0000)]
Remove EOL whitespace.

11 years agomdoc: new sentence, new line.
joel [Wed, 8 May 2013 21:02:48 +0000 (21:02 +0000)]
mdoc: new sentence, new line.

11 years agoBring in a basic ethernet switch driver for the IP17x series of
adrian [Wed, 8 May 2013 20:58:41 +0000 (20:58 +0000)]
Bring in a basic ethernet switch driver for the IP17x series of
switches.

These are notably found on some AR71xx based Mikrotik boards.

Submitted by: Luiz Otavio O Souza <loos.br@gmail.com>
Reviewed by: ray

11 years agoModify the routerstation config to use ukswitch for now.
adrian [Wed, 8 May 2013 20:57:37 +0000 (20:57 +0000)]
Modify the routerstation config to use ukswitch for now.

Until an ADM6996 driver shows up, this allows for the two switch
ports to be used.

Submitted by: Luiz Otavio O Souza <loos.br@gmail.com>
Reviewed by: ray