]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
16 years ago- add: Keep dependent packages too if -K is specified.
Florent Thoumie [Thu, 12 Jun 2008 15:21:13 +0000 (15:21 +0000)]
- add: Keep dependent packages too if -K is specified.
- updating: terminating '\n' is not part of the package origin.
- bump PKG_INSTALL_VERSION to 20080612.

PR: bin/119368 [1], bin/124459 [2]
Submitted by: gcooper [1], Beat Gatzi <beat@chruetertee.ch> [2]
MFC after: 3 days

16 years agoFix build of fstat after minor() changes.
Ed Schouten [Thu, 12 Jun 2008 10:15:14 +0000 (10:15 +0000)]
Fix build of fstat after minor() changes.

Even though I ran a `make universe' to see whether the changes to the
device minor number macro's broke the build, I was not expecting `make
universe' to silently continue if build errors occured, thus causing me
to overlook the build error.

Approved by: philip (mentor)
Pointyhat to: me

16 years agoRemove the $FreeBSD$ tag again, now I know fbsd:nokeywords exists.
Ed Schouten [Thu, 12 Jun 2008 08:53:54 +0000 (08:53 +0000)]
Remove the $FreeBSD$ tag again, now I know fbsd:nokeywords exists.

Requested by: pjd
Approved by: philip (mentor)

16 years agoTurn dev2unit(), minor(), unit2minor() and minor2unit() into macro's.
Ed Schouten [Thu, 12 Jun 2008 08:30:54 +0000 (08:30 +0000)]
Turn dev2unit(), minor(), unit2minor() and minor2unit() into macro's.

Now that we got rid of the minor-to-unit conversion and the constraints
on device minor numbers, we can convert the functions that operate on
minor and unit numbers to simple macro's. The unit2minor() and
minor2unit() macro's are now no-ops.

The ZFS code als defined a macro named `minor'. Change the ZFS code to
use umajor() and uminor() here, as it is the correct approach to do
this. Also add $FreeBSD$ to keep SVN happy.

Approved by: philip (mentor), pjd

16 years agoAdd the raw partition type to the XML.
Marcel Moolenaar [Thu, 12 Jun 2008 06:34:14 +0000 (06:34 +0000)]
Add the raw partition type to the XML.

16 years agoAdd the raw partition type to the XML.
Marcel Moolenaar [Thu, 12 Jun 2008 06:26:36 +0000 (06:26 +0000)]
Add the raw partition type to the XML.

16 years agoAdd the raw partition type to the XML.
Marcel Moolenaar [Thu, 12 Jun 2008 05:56:03 +0000 (05:56 +0000)]
Add the raw partition type to the XML.

16 years agoAdd the raw partiton type to the XML.
Marcel Moolenaar [Thu, 12 Jun 2008 05:28:47 +0000 (05:28 +0000)]
Add the raw partiton type to the XML.

16 years agoAdd the raw partition type to the XML.
Marcel Moolenaar [Thu, 12 Jun 2008 05:27:23 +0000 (05:27 +0000)]
Add the raw partition type to the XML.

16 years agoReverse the direction of pmap_promote_pde()'s traversal over the specified
Alan Cox [Thu, 12 Jun 2008 05:18:09 +0000 (05:18 +0000)]
Reverse the direction of pmap_promote_pde()'s traversal over the specified
page table page.  The direction of the traversal can matter if
pmap_promote_pde() has to remove write access (PG_RW) from a PTE that hasn't
been modified (PG_M).  In general, if there are two or more such PTEs to
choose among, it is better to write protect the one nearer the high end of
the page table page rather than the low end.  This is because most programs
access memory in an ascending direction.  The net result of this change is a
sometimes significant reduction in the number of failed promotion attempts
and the number of pages that are write protected by pmap_promote_pde().

16 years agoAdd the partition label and the raw partition type to the XML.
Marcel Moolenaar [Thu, 12 Jun 2008 04:43:34 +0000 (04:43 +0000)]
Add the partition label and the raw partition type to the XML.

16 years agoDefine APM_ENT_NAMELEN and APM_ENT_TYPELEN for general use.
Marcel Moolenaar [Thu, 12 Jun 2008 04:37:37 +0000 (04:37 +0000)]
Define APM_ENT_NAMELEN and APM_ENT_TYPELEN for general use.

16 years agoReturn an error code rather than ENXIO when both rman_init() and
Kevin Lo [Thu, 12 Jun 2008 02:28:58 +0000 (02:28 +0000)]
Return an error code rather than ENXIO when both rman_init() and
rman_manage_region() failed.

Reviewed by: marcel

16 years agoFix a typo: i80321_pci_probe -> i81342_pci_probe
Kevin Lo [Thu, 12 Jun 2008 01:46:06 +0000 (01:46 +0000)]
Fix a typo: i80321_pci_probe -> i81342_pci_probe

16 years agoMake sure we drain our taskqueues and stop our callouts in detach.
Benno Rice [Thu, 12 Jun 2008 00:38:54 +0000 (00:38 +0000)]
Make sure we drain our taskqueues and stop our callouts in detach.

16 years agoChange include to reflect tcp_lro.h move.
Jack F Vogel [Wed, 11 Jun 2008 22:29:47 +0000 (22:29 +0000)]
Change include to reflect tcp_lro.h move.

16 years agoRemove the tcp_lro.[ch] files as they are now in netinet.
Jack F Vogel [Wed, 11 Jun 2008 22:26:27 +0000 (22:26 +0000)]
Remove the tcp_lro.[ch] files as they are now in netinet.

16 years agoRemove compile of tcp_lro since its now in netinet
Jack F Vogel [Wed, 11 Jun 2008 22:18:50 +0000 (22:18 +0000)]
Remove compile of tcp_lro since its now in netinet

16 years agoAdd generic TCP LOR into netinet
Jack F Vogel [Wed, 11 Jun 2008 22:12:50 +0000 (22:12 +0000)]
Add generic TCP LOR into netinet

16 years agoAdd LRO into kernel build
Jack F Vogel [Wed, 11 Jun 2008 22:10:10 +0000 (22:10 +0000)]
Add LRO into kernel build

16 years agoDuh, wrong directory, needed to be in netinet
Jack F Vogel [Wed, 11 Jun 2008 22:08:13 +0000 (22:08 +0000)]
Duh, wrong directory, needed to be in netinet

16 years agoAdd generic TCP LRO code, moved from the ixgbe driver into net
Jack F Vogel [Wed, 11 Jun 2008 22:00:29 +0000 (22:00 +0000)]
Add generic TCP LRO code, moved from the ixgbe driver into net

16 years agoFix a typo in a comment.
Wojciech A. Koszek [Wed, 11 Jun 2008 20:05:25 +0000 (20:05 +0000)]
Fix a typo in a comment.

16 years agoDon't enforce unique device minor number policy anymore.
Ed Schouten [Wed, 11 Jun 2008 18:55:19 +0000 (18:55 +0000)]
Don't enforce unique device minor number policy anymore.

Except for the case where we use the cloner library (clone_create() and
friends), there is no reason to enforce a unique device minor number
policy. There are various drivers in the source tree that allocate unr
pools and such to provide minor numbers, without using them themselves.

Because we still need to support unique device minor numbers for the
cloner library, introduce a new flag called D_NEEDMINOR. All cdevsw's
that are used in combination with the cloner library should be marked
with this flag to make the cloning work.

This means drivers can now freely use si_drv0 to store their own flags
and state, making it effectively the same as si_drv1 and si_drv2. We
still keep the minor() and dev2unit() routines around to make drivers
happy.

The NTFS code also used the minor number in its hash table. We should
not do this anymore. If the si_drv0 field would be changed, it would no
longer end up in the same list.

Approved by: philip (mentor)

16 years agoDocument the AUTO_UPGRADE (-U) knob for .mergemasterrc
Doug Barton [Wed, 11 Jun 2008 18:54:06 +0000 (18:54 +0000)]
Document the AUTO_UPGRADE (-U) knob for .mergemasterrc

Submitted by: mezz
No Cookie For: gordon  :)

16 years agoAdd man page for ixgbe driver
Jack F Vogel [Wed, 11 Jun 2008 18:49:55 +0000 (18:49 +0000)]
Add man page for ixgbe driver

16 years agoConvert ndis_spinlock to ndis_mtx and start using the sleepable
Coleman Kane [Wed, 11 Jun 2008 13:40:15 +0000 (13:40 +0000)]
Convert ndis_spinlock to ndis_mtx and start using the sleepable
mtx interface for NDIS_LOCK/UNLOCK. This should result in less
CPU utilization on behalf of the ndis driver. Additionally, this
commit also fixes a potential LOR in the ndis_tick code, by
not locking inside the ndis_tick function, but instead delegating
that work to the helpers called through IoQueueWorkItem. The
way that this is currently set up for NDIS prevents us from
simply implementing a callout_init_mtx mechanism.

However, the helper functions that handle the various timeout
cases implement fine-grained locking using the spinlocks provided
by the NDIS-compat layer, and using the mtx that is added with
this commit. This leaves the following ndis_softc members operated
on in ndis_tick in an unlocked context:

  * ndis_hang_timer - Only modified outside of ndis_tick once, before
                      the first callout_reset to schedule ndis_tick
  * ifp->if_oerrors - Only incremented in two places, which should be
                      an atomic op
  * ndis_tx_timer   - Assigned to 5 (when guaranteed to be 0) or 0
                      (in txeof), to indicate to ndis_tick what to
                      do. This is the only member of which I was
                      suspicious for needing the NDIS_LOCK here. My
                      testing (and another's) have been fine so far.
  * ndis_stat_callout - Only uses a simple set of callout routines,
                        callout_reset only called by ndis_tick after
                        the initial reset, and then callout_drain is
                        used exactly once in shutdown code.

The benefit is that ndis_tick doesn't acquire NDIS_LOCK unless one of
the timeout conditions is flagged, and it still obeys the locking
order semantics that are dictated by the NDIS layer at the moment. I
have been investigating a more thorough s/spinlock/mtx/ of the NDIS
layer, but the simplest naive approach (replace KeAcquireSpinLock
with an mtx implementation) has anti-succeeded for me so far. This
is a good first step though.

Tested by: onemda@gmail.com
Reviewed by: current@, jhb, thompsa
Proposed by: jhb

16 years agoIn cd9660_readdir vop, always initialize the idp->uio_off member.
Konstantin Belousov [Wed, 11 Jun 2008 12:46:09 +0000 (12:46 +0000)]
In cd9660_readdir vop, always initialize the idp->uio_off member.

The while loop that is assumed to initialize the uio_off later, may
be not entered at all, causing uninitialized value to be returned in
uio->uio_offset.

PR:  122925
Submitted by: Jaakko Heinonen <jh saunalahti fi>
MFC after: 1 weeks

16 years agoWait up to 1S for the TFD data to signal un-busy before fetching the signature.
Søren Schmidt [Wed, 11 Jun 2008 08:48:25 +0000 (08:48 +0000)]
Wait up to 1S for the TFD data to signal un-busy before fetching the signature.
This at least helps a few slow devices out there.

Submitted by: Andrey V. Elsukov

16 years agofix a page fault that it occurred during ifp is NULL. This bug happens
Weongyo Jeong [Wed, 11 Jun 2008 07:55:07 +0000 (07:55 +0000)]
fix a page fault that it occurred during ifp is NULL.  This bug happens
when NDIS driver's initialization is failed and NDIS driver's trying to
call NdisWriteErrorLogEntry().

16 years agoSwitch to using a normal mutex instead of a spin mutex.
Benno Rice [Wed, 11 Jun 2008 07:26:02 +0000 (07:26 +0000)]
Switch to using a normal mutex instead of a spin mutex.

We still use the interrupt filter due to performance problems that show up if
we don't.  The main problem seen is that, due to the interrupt being edge
triggered, we occasionally miss interrupts which leads us to not notice that
we can transmit more packets.  Using the new approach, which just schedules
a task on a taskqueue, we are guaranteed to have the task run even if the
interrupt arrived while we were already executing.  If we were to use an
ithread the system would mask the interrupt while the handler is run and we'd
miss interrupts.

16 years agoConvert bus_space_{read,write}_* calls to bus_{read,write}_* calls.
Benno Rice [Wed, 11 Jun 2008 06:53:55 +0000 (06:53 +0000)]
Convert bus_space_{read,write}_* calls to bus_{read,write}_* calls.

Suggested by: jhb

16 years agoRearrange how to call dma.alloc() so that we have resources alloc'd when need but...
Søren Schmidt [Wed, 11 Jun 2008 06:44:58 +0000 (06:44 +0000)]
Rearrange how to call dma.alloc() so that we have resources alloc'd when need but also late enough to know how many to create.

16 years ago Keep proper track of nsegs counter: sem_free is called for all
Oleksandr Tymoshenko [Tue, 10 Jun 2008 20:55:10 +0000 (20:55 +0000)]
  Keep proper track of nsegs counter: sem_free is called for all
  allocated semaphores, so it's wrong to increase it conditionally,
  in this case for every over-the-limit semaphore nsegs is decreased
  without being previously increased.

  PR: kern/123685
  Approved by: cognet (mentor)

16 years agoWhen the file-system containing the audit log file is running low on
Simon L. B. Nielsen [Tue, 10 Jun 2008 20:05:32 +0000 (20:05 +0000)]
When the file-system containing the audit log file is running low on
disk space a warning is printed.  Make this warning a bit more
informative.

Approved by: rwatson

16 years agoAdd Belkin F5U257 to the mix.
Warner Losh [Tue, 10 Jun 2008 19:33:31 +0000 (19:33 +0000)]
Add Belkin F5U257 to the mix.

16 years agoAdd Belkin F5U257 to the mix.
Warner Losh [Tue, 10 Jun 2008 19:31:09 +0000 (19:31 +0000)]
Add Belkin F5U257 to the mix.

Submitted by: wilko@

16 years agoUse sysctl to fetch stats from the kernel instead of reading variables
John Baldwin [Tue, 10 Jun 2008 18:47:43 +0000 (18:47 +0000)]
Use sysctl to fetch stats from the kernel instead of reading variables
directly via libkvm.

PR: kern/122875, bin/123014
Tested by: Danny Braniss  danny cs.huji.ac.il
MFC after: 1 week

16 years agoMake tx(4) MPSAFE
John Baldwin [Tue, 10 Jun 2008 17:59:43 +0000 (17:59 +0000)]
Make tx(4) MPSAFE
- Add a mutex to the softc to protect the softc and device hardware.
- Use a private timer to implement a watchdog for tx timeouts and drive
  the timer for auto negotiation.
- Use bus_foo() rather than bus_space_foo() and remove the bus space
  tag & handle from the softc.
- Call bus_setup_intr() after ether_ifattach().

Tested by: Florian Smeets  flo of kasimir.com

16 years agoAdd support for the P212, P410, P410i, P411, and P812 HP Smart Array
Paul Saab [Tue, 10 Jun 2008 17:51:51 +0000 (17:51 +0000)]
Add support for the P212, P410, P410i, P411, and P812 HP Smart Array
controllers.

Submitted by:  Scott Benesh at HP

16 years agoIn the error path through base_alloc(), release base_mtx [1].
Jason Evans [Tue, 10 Jun 2008 15:46:18 +0000 (15:46 +0000)]
In the error path through base_alloc(), release base_mtx [1].

Fix bit vector initialization for run headers.

Submitted by: [1] Mike Schuster <schuster@adobe.com>

16 years agoAdd the pxa_teardown_intr() bus method function to de-associate the
Kevin Lo [Tue, 10 Jun 2008 06:06:15 +0000 (06:06 +0000)]
Add the pxa_teardown_intr() bus method function to de-associate the
interrupt handler

16 years agoPull all the code to deal with bus space methods into a shared set of
Kevin Lo [Tue, 10 Jun 2008 03:44:14 +0000 (03:44 +0000)]
Pull all the code to deal with bus space methods into a shared set of
routines.

16 years ago>From NetBSD:
Kevin Lo [Tue, 10 Jun 2008 02:41:36 +0000 (02:41 +0000)]
>From NetBSD:
Remove the code which disables port status change interrupts for 1s
when one occured -- this makes that events get lost or delayed until
the next change.

Obtained from: NetBSD

16 years ago- Fixed kern/123696 by increasing firmware timeout value from 100 to 1000.
David Christensen [Tue, 10 Jun 2008 02:19:11 +0000 (02:19 +0000)]
- Fixed kern/123696 by increasing firmware timeout value from 100 to 1000.
- Fixed a problem on i386 architecture when using split header/jumbo frame
  firmware caused by hardware alignment requirements.
- Added #define BCE_USE_SPLIT_HEADER to allow the feature to be enabled/
  disabled.  Enabled by default.

PR: kern/123696
MFC after: 2 weeks

16 years agoMark "highaddr" and "lowaddr" appropriately. Change "dma" to "DMA".
Wojciech A. Koszek [Mon, 9 Jun 2008 22:13:14 +0000 (22:13 +0000)]
Mark "highaddr" and "lowaddr" appropriately. Change "dma" to "DMA".

Fix spelling mistage found by rodrigc@.

Reviewed by: rodrigc, danger

16 years agoSince we create a DMA tag "mtag" for TX map with bus_dmamap_create(),
Wojciech A. Koszek [Mon, 9 Jun 2008 21:51:48 +0000 (21:51 +0000)]
Since we create a DMA tag "mtag" for TX map with bus_dmamap_create(),
we must synchronize such a map against "mtag" with bus_dmamap_sync(),
not the tag designated for RX map.

Fix it.

Approved by: cognet

16 years agoRemoval of gpt(8) and manpage.
Marcel Moolenaar [Mon, 9 Jun 2008 21:36:46 +0000 (21:36 +0000)]
Removal of gpt(8) and manpage.

16 years agoNote removal of gpt(8).
Marcel Moolenaar [Mon, 9 Jun 2008 21:33:57 +0000 (21:33 +0000)]
Note removal of gpt(8).

16 years agoRemove gpt(8). It's replaced by gpart(8).
Marcel Moolenaar [Mon, 9 Jun 2008 21:30:40 +0000 (21:30 +0000)]
Remove gpt(8). It's replaced by gpart(8).

16 years agoUse the -n flag to route(8) when calling "route get". Otherwise we hang
Brooks Davis [Mon, 9 Jun 2008 20:03:35 +0000 (20:03 +0000)]
Use the -n flag to route(8) when calling "route get".  Otherwise we hang
for a long time if we get a lease, but DNS isn't working.

MFC after: 1 week

16 years agoDisconnect gpt(8) from the build. gpt(8) is replaced by gpart(8).
Marcel Moolenaar [Mon, 9 Jun 2008 16:42:55 +0000 (16:42 +0000)]
Disconnect gpt(8) from the build. gpt(8) is replaced by gpart(8).

16 years agoDon't store a pointer in "next" if we're never going to use it.
Colin Percival [Mon, 9 Jun 2008 14:44:56 +0000 (14:44 +0000)]
Don't store a pointer in "next" if we're never going to use it.

Fix typo in checking results of strchr.

Found by: Coverity Prevent

16 years agoMake one-bit fields unsigned instead of signed. This has no effect,
Colin Percival [Mon, 9 Jun 2008 14:41:28 +0000 (14:41 +0000)]
Make one-bit fields unsigned instead of signed.  This has no effect,
since they are only tested for zero/nonzero; but it's arguably a bad
idea to set a {-1, 0} variable to 1 (as happens in this code).

Found by: Coverity Prevent

16 years agoRework code to avoid using a pointer after freeing it. Aside from the
Colin Percival [Mon, 9 Jun 2008 14:03:55 +0000 (14:03 +0000)]
Rework code to avoid using a pointer after freeing it.  Aside from the
possibility of memory becoming undereferenceable when it is freed, this
change should have no effect on bsdtar behaviour.

Found by: Coverity Prevent

16 years agoProvide the mutual exclusion between the nfs export list modifications
Konstantin Belousov [Mon, 9 Jun 2008 10:31:38 +0000 (10:31 +0000)]
Provide the mutual exclusion between the nfs export list modifications
and nfs requests processing. Lockmgr lock provides the shared locking for
nfs requests, while exclusive mode is used for modifications. The writer
starvation is handled by lockmgr too.

Reported by: kris, pho, many
Based on the submission by: mohan
Tested by: pho
MFC after: 2 weeks

16 years agoDocument the misleading nature of the REQUIRE line. The patch in
Mike Makonnen [Mon, 9 Jun 2008 09:07:58 +0000 (09:07 +0000)]
Document the misleading nature of the REQUIRE line. The patch in
the PR has been heavily edited for style(9) and clarity. Mistakes are
mine.

PR: bin/124251

16 years agoRemove sicontrol(8)'s "ttystat".
Ed Schouten [Mon, 9 Jun 2008 08:43:27 +0000 (08:43 +0000)]
Remove sicontrol(8)'s "ttystat".

In the FreeBSD base system, there are only two utilities that use struct
tty, namely pstat and sicontrol. The sicontrol utility calls the
TCSI_TTY ioctl(), which copies struct tty back to userspace.

sicontrol should not have this functionality. The same data is already
provided by pstat. If we really want to be able to export these numbers
through a file descriptor to userspace, we can export struct xtty, which
should provide a better abstraction. The ttystat option was only used as
a debugging aid.

This makes sicontrol compile in the mpsafetty branch.

Reviewed by: peter
Approved by: philip (mentor)

16 years agoChange the CTF conversion makefile code to use a new line to avoid
John Birrell [Mon, 9 Jun 2008 06:33:26 +0000 (06:33 +0000)]
Change the CTF conversion makefile code to use a new line to avoid
spawning another shell.

Requested by: Ed Schouten

M    config/mkmakefile.c

16 years agoRemove some sparc-specific stuff from my earlier sun4v work in p4.
John Birrell [Mon, 9 Jun 2008 06:31:17 +0000 (06:31 +0000)]
Remove some sparc-specific stuff from my earlier sun4v work in p4.
It never belonged in current.

Pointed out by: marius

16 years agoRemove sa1_cache_clean_addr
Kevin Lo [Mon, 9 Jun 2008 05:53:04 +0000 (05:53 +0000)]
Remove sa1_cache_clean_addr

16 years agoUnify arminit() and clean up
Kevin Lo [Mon, 9 Jun 2008 05:50:42 +0000 (05:50 +0000)]
Unify arminit() and clean up

16 years agoMake pthread_cleanup_push() and pthread_cleanup_pop() as a pair of macros,
David Xu [Mon, 9 Jun 2008 01:14:10 +0000 (01:14 +0000)]
Make pthread_cleanup_push() and pthread_cleanup_pop() as a pair of macros,
use stack space to keep cleanup information, this eliminates overhead of
calling malloc() and free() in thread library.

Discussed on: thread@

16 years agoChange header file references from <filename>.h to
Wojciech A. Koszek [Sun, 8 Jun 2008 21:08:20 +0000 (21:08 +0000)]
Change header file references from <filename>.h to
path relative to /usr/include. It looks much better anyway.

Instead of referencing "socket", which is bogus, reference
'sys/socket.h', which is what should be placed here from
the beggining.

Suggested by: maxim

16 years agoFix the way the date must be specified (leading 0 is redundant).
Wojciech A. Koszek [Sun, 8 Jun 2008 21:01:39 +0000 (21:01 +0000)]
Fix the way the date must be specified (leading 0 is redundant).
Fix several spelling mistakes brought by my earlier commit.
Trim whitespace.

Submitted by: maxim

16 years agoRemove checks against DDB, which isn't used in this file.
Wojciech A. Koszek [Sun, 8 Jun 2008 20:43:27 +0000 (20:43 +0000)]
Remove checks against DDB, which isn't used in this file.

My intention is to bring no functional change.

Discussion on: IRC
Reviewed by: ed, kan, rink,

16 years agoI missed some "register"s in non-dot-C files.
David Malone [Sun, 8 Jun 2008 19:59:15 +0000 (19:59 +0000)]
I missed some "register"s in non-dot-C files.

16 years ago- Display '-C' option in usage().
Stanislav Sedov [Sun, 8 Jun 2008 19:48:41 +0000 (19:48 +0000)]
- Display '-C' option in usage().

Approved by: kib

16 years agoThe change to add subversion ID has two problems. The first is that when
Doug Barton [Sun, 8 Jun 2008 19:46:23 +0000 (19:46 +0000)]
The change to add subversion ID has two problems. The first is that when
newvers.sh is run pwd is actually the obj directory, so "../../.svn"
doesn't exist and the test always fails. The second is that buildkernel
is executed with a restrictive PATH, so unless you have svnversion in
/bin or /usr/bin it can't run.

Fix this by looking for svnversion in /bin, /usr/bin, and /usr/local/bin
in that order. If found, store the location and derive the value of the
source directory. Then run svnversion in the appropriate directory.

There is one possible refinement which would be to add a test for
LOCALBASE!=/usr/local if we don't find svnversion the first time, but
IMO that's not necessary at this time.

16 years agoFixed the output grammar to properly speak non-terminal dits.
Sean Farley [Sun, 8 Jun 2008 19:35:47 +0000 (19:35 +0000)]
Fixed the output grammar to properly speak non-terminal dits.

Updated Lyndon Nerenberg's radio callsign and E-mail address.

PR: bin/7868
Submitted by: Lyndon Nerenberg <lyndon@orthanc.ca>
MFC after: 1 week

16 years agoFix typo in comment.
Rui Paulo [Sun, 8 Jun 2008 14:42:43 +0000 (14:42 +0000)]
Fix typo in comment.

16 years agoMake usage() 'static'.
Wojciech A. Koszek [Sun, 8 Jun 2008 12:43:02 +0000 (12:43 +0000)]
Make usage() 'static'.

16 years agod_ino member of linux_dirent structure should be unsigned long.
Roman Divacky [Sun, 8 Jun 2008 11:09:25 +0000 (11:09 +0000)]
d_ino member of linux_dirent structure should be unsigned long.

Submitted by: Chagin Dmitry <chagin.dmitry@gmail.com>
Approved by: kib (mentor)

16 years agoBring more commands to the ddb(4) manual page. This should be considered
Wojciech A. Koszek [Sun, 8 Jun 2008 09:09:08 +0000 (09:09 +0000)]
Bring more commands to the ddb(4) manual page. This should be considered
"complete" list, but some commands might be still missing.

Reviewed by: julian

16 years agoFix spelling of "virtual".
Wojciech A. Koszek [Sun, 8 Jun 2008 08:56:46 +0000 (08:56 +0000)]
Fix spelling of "virtual".

There should be no visible change.

Reviewed by: rink

16 years agoEthernet hardware address stored in DC_AL_PAR0/DC_AL_PAR1 register
Pyun YongHyeon [Sun, 8 Jun 2008 02:52:26 +0000 (02:52 +0000)]
Ethernet hardware address stored in DC_AL_PAR0/DC_AL_PAR1 register
is in little endian form. Likewise setting DC_AL_PAR0/DC_AL_PAR1
register expect the address to be in little endian form. For big
endian architectures the address should be swapped to get correct
one.
Change setting/getting ethernet hardware address to big endian
architecture frendly.

Reported by: Robert Murillo ( billypilgrim782001 at yahoo dot com )
Tested by: Robert Murillo ( billypilgrim782001 at yahoo dot com )

16 years agoMove bm(4) from the sys/conf/NOTES to sys/powerpc/conf/NOTES.
Marcel Moolenaar [Sun, 8 Jun 2008 01:58:11 +0000 (01:58 +0000)]
Move bm(4) from the sys/conf/NOTES to sys/powerpc/conf/NOTES.
The driver applies to PowerPC only.

16 years agoAdd support for the Apple Big Mac (BMAC) Ethernet controller,
Marcel Moolenaar [Sat, 7 Jun 2008 22:58:32 +0000 (22:58 +0000)]
Add support for the Apple Big Mac (BMAC) Ethernet controller,
found on various Apple G3 models.

Submitted by: Nathan Whitehorn

16 years agoAdd support for Apple's Descriptor-Based DMA (DBDMA) engine. The DMA
Marcel Moolenaar [Sat, 7 Jun 2008 21:56:48 +0000 (21:56 +0000)]
Add support for Apple's Descriptor-Based DMA (DBDMA) engine. The DMA
engine is usful to various  existing drivers, such as ata(4) and scc(4),
and is used bhy the soon to be added bm(4).

Submitted by: Nathan Whitehorn

16 years agoChange the calling convention for ic_node_alloc to deal with
Sam Leffler [Sat, 7 Jun 2008 18:38:02 +0000 (18:38 +0000)]
Change the calling convention for ic_node_alloc to deal with
some longstanding issues:
o pass the vap since it's now the "coin of the realm" and required
  to do things like set initial tx parameters in private node
  state for use prior to association
o pass the mac address as cards that maintain outboard station
  tables require this to create an entry (e.g. in ibss mode)
o remove the node table reference, we only have one node table
  and it's unlikely this will change so this is not needed to
  find the com structure

16 years agoSplit ieee80211_notify_erp into locked and unlocked variants
Sam Leffler [Sat, 7 Jun 2008 17:51:41 +0000 (17:51 +0000)]
Split ieee80211_notify_erp into locked and unlocked variants
and use the locked version to handle the hostap input path
case where the com lock is not already held.

Noticed by: Jared Go

16 years agoFix node cleanup issues when the last reference is held by
Sam Leffler [Sat, 7 Jun 2008 17:50:24 +0000 (17:50 +0000)]
Fix node cleanup issues when the last reference is held by
the driver (e.g. in the tx q); ni_vap may not be valid, we
must carefully check before using it.

16 years agocorrect code returned in AsssocResp; the previous code is spec'd for use
Sam Leffler [Sat, 7 Jun 2008 17:43:41 +0000 (17:43 +0000)]
correct code returned in AsssocResp; the previous code is spec'd for use
only in deauth+disassoc frames

Submitted by: Chris Zimmermann
MFC after: 1 month

16 years agouse 'const' for the parameters of the two static functions unalias() and hashalias()
Ralf S. Engelschall [Sat, 7 Jun 2008 16:28:20 +0000 (16:28 +0000)]
use 'const' for the parameters of the two static functions unalias() and hashalias()

16 years agoremove an unnecessary include
Ralf S. Engelschall [Sat, 7 Jun 2008 16:19:28 +0000 (16:19 +0000)]
remove an unnecessary include

16 years agoIf we can find it, include SVN version number in kernel version strings.
Poul-Henning Kamp [Sat, 7 Jun 2008 09:49:57 +0000 (09:49 +0000)]
If we can find it, include SVN version number in kernel version strings.

See also: http://www.bikeshed.org/

16 years agoUnbreak last commit to ln for amd64.
Giorgos Keramidas [Sat, 7 Jun 2008 09:37:30 +0000 (09:37 +0000)]
Unbreak last commit to ln for amd64.

Cast string precision to `int'.  amd64 systems warn about the
field precision being `long int' if we don't, and pathnames are
normally short enough to fit in an `int'.

Noticed by: pav

16 years agoFix the incorrect calculation of a block address within a single indirect
Konstantin Belousov [Sat, 7 Jun 2008 05:49:24 +0000 (05:49 +0000)]
Fix the incorrect calculation of a block address within a single indirect
block.

PR: 108215
Submitted by: Yuichiro Goto, y7goto gmail com
MFC after: 2 weeks

16 years agoWorkaround a bug in the BIOS of Dell R900 machines. Specifically, each
John Baldwin [Sat, 7 Jun 2008 03:07:32 +0000 (03:07 +0000)]
Workaround a bug in the BIOS of Dell R900 machines.  Specifically, each
entry in the SMAP is a 20 byte structure and they are queried from the
BIOS via sucessive BIOS calls.  Due to an apparent bug in the R900's
BIOS, for some SMAP requests the BIOS overflows the 20 byte buffer
trashing a few bytes of memory immediately after the SMAP structure.  As
a workaround, add 8 bytes of padding after the SMAP structure used in
the loader for SMAP queries.

PR: i386/122668
Submitted by: Mike Hibler  mike flux.utah.edu, silby
MFC after: 3 days

16 years agoDocument the new -p and -i options for writing bootstrap code into
Marcel Moolenaar [Sat, 7 Jun 2008 00:14:06 +0000 (00:14 +0000)]
Document the new -p and -i options for writing bootstrap code into
a partition. Make it clear that the -b option embeds boot code in
the meta-data.

16 years agoEnhance the bootcode command to also allow bootcode to be written
Marcel Moolenaar [Fri, 6 Jun 2008 23:58:29 +0000 (23:58 +0000)]
Enhance the bootcode command to also allow bootcode to be written
to a partition. This avoids that users need to use dd(1) to install
boot code (as is needed for VTOC8 and booting GPT on PCs).

16 years agoAdd two support functions:
Marcel Moolenaar [Fri, 6 Jun 2008 22:44:03 +0000 (22:44 +0000)]
Add two support functions:
o  gctl_delete_param() -- intended for parameters that are consumed
   by geom(8) itself and which should not be passed to the kernel.
o  gctl_has_param() -- intended to check if optional parameters are
   present.

Both are needed by gpart(8) to process the optional parameter for
writing bootcode to a partition (as part of the bootcode verb).
However, the kernel is itself not involved in this matter and the
parameter needs to be removed from the request destined for the
kernel.

16 years agoUpdate description text
Peter Wemm [Fri, 6 Jun 2008 21:32:01 +0000 (21:32 +0000)]
Update description text

16 years ago- Store the device_t of the smbX device in the softc.
John Baldwin [Fri, 6 Jun 2008 18:45:32 +0000 (18:45 +0000)]
- Store the device_t of the smbX device in the softc.
- Store the softc of the device in the 'si_drv1' of the cdev.
- Lookup the softc via 'si_drv1' in cdev methods rather than using the
  minor number as a unit for devclass_get_softc().
- Lookup the device_t via the softc field in cdev methods rather than
  using the minor number as a unit for devclass_get_device().
- Add a mutex to the softc to protect 'sc_opened'.
- Remove D_NEEDGIANT as all the smbus drivers are now MPSAFE and this driver
  is now MPSAFE.
- Remove some checks for NULL softc pointers that can't happen and don't
  bzero the softc during attach.

16 years agoExplicitly lock Giant in smbus_if methods in the bktr_i2c and iicsmb
John Baldwin [Fri, 6 Jun 2008 18:40:38 +0000 (18:40 +0000)]
Explicitly lock Giant in smbus_if methods in the bktr_i2c and iicsmb
drivers for now.  This can be replaced with driver locks when these
drivers are locked.

16 years agoEssentially, neither madvise(..., MADV_DONTNEED) nor madvise(..., MADV_FREE)
Alan Cox [Fri, 6 Jun 2008 18:38:43 +0000 (18:38 +0000)]
Essentially, neither madvise(..., MADV_DONTNEED) nor madvise(..., MADV_FREE)
work.  (Moreover, I don't believe that they have ever worked as intended.)
The explanation is fairly simple.  Both MADV_DONTNEED and MADV_FREE perform
vm_page_dontneed() on each page within the range given to madvise().  This
function moves the page to the inactive queue.  Specifically, if the page is
clean, it is moved to the head of the inactive queue where it is first in
line for processing by the page daemon.  On the other hand, if it is dirty,
it is placed at the tail.  Let's further examine the case in which the page
is clean.  Recall that the page is at the head of the line for processing by
the page daemon.  The expectation of vm_page_dontneed()'s author was that
the page would be transferred from the inactive queue to the cache queue by
the page daemon.  (Once the page is in the cache queue, it is, in effect,
free, that is, it can be reallocated to a new vm object by vm_page_alloc()
if it isn't reactivated quickly enough by a user of the old vm object.)  The
trouble is that nowhere in the execution of either MADV_DONTNEED or
MADV_FREE is either the machine-independent reference flag (PG_REFERENCED)
or the reference bit in any page table entry (PTE) mapping the page cleared.
Consequently, the immediate reaction of the page daemon is to reactivate the
page because it is referenced.  In effect, the madvise() was for naught.
The case in which the page was dirty is not too different.  Instead of being
laundered, the page is reactivated.

Note: The essential difference between MADV_DONTNEED and MADV_FREE is
that MADV_FREE clears a page's dirty field.  So, MADV_FREE is always
executing the clean case above.

This revision changes vm_page_dontneed() to clear both the machine-
independent reference flag (PG_REFERENCED) and the reference bit in all PTEs
mapping the page.

MFC after: 6 weeks

16 years ago- Use bus_foo() rather than bus_space_foo() and remove bus space tag/handle
John Baldwin [Fri, 6 Jun 2008 18:29:56 +0000 (18:29 +0000)]
- Use bus_foo() rather than bus_space_foo() and remove bus space tag/handle
  from softc.
- Mark interrupt handlers MPSAFE as these drivers have been locked for a
  while.

16 years agoTweak a comment and a constant to restore old 30ms upper bound.
Warner Losh [Fri, 6 Jun 2008 17:27:19 +0000 (17:27 +0000)]
Tweak a comment and a constant to restore old 30ms upper bound.
cs_readreg takes ~2us not ~1us to run.

16 years agoo Remove unused cs_debug tunable. I think I added it and then nothing with
Warner Losh [Fri, 6 Jun 2008 17:22:07 +0000 (17:22 +0000)]
o Remove unused cs_debug tunable.  I think I added it and then nothing with
  it.  Bad imp.  Removing us dips us under 10,000 in size too.
o Replace an unconditional 30ms DELAY (yes, busy wait) with a check of the
  SIBUSY bit in the SelfST register before accessing the eeprom.  This changes
  the time to read the EEPROM from 2 * 20 * 30ms (1.2s) to < 20*25us (.0005s)
  and make the attach of the card tolerable when ethernet media is present.
  Include data from the datasheet about why this works.  While this is a 2500x
  speed increase, it doesn't really matter at all once the card is probed...
o set dev earlier in softc.

16 years agoOptimize the CVS bug #17168 fix by relaxing the conditions for update
David E. O'Brien [Fri, 6 Jun 2008 16:16:44 +0000 (16:16 +0000)]
Optimize the CVS bug #17168 fix by relaxing the conditions for update
with respect to sticky options.  (r179561 was WRT sticky and branch tags)

Consulted with: Mark Baushke <mdb@gnu.org>