]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
17 years agoCraft the FreeBSD/ARM ABI by hacking the New ARM ABI source.
David E. O'Brien [Tue, 12 Sep 2006 07:03:58 +0000 (07:03 +0000)]
Craft the FreeBSD/ARM ABI by hacking the New ARM ABI source.

Submitted by: cognet

17 years agoDefault vector to big endian ARM if "TARGET_BIG_ENDIAN" is defined.
David E. O'Brien [Tue, 12 Sep 2006 07:02:15 +0000 (07:02 +0000)]
Default vector to big endian ARM if "TARGET_BIG_ENDIAN" is defined.

17 years agoBuild 'ld' supporting a big endian ARM if "TARGET_BIG_ENDIAN" is defined.
David E. O'Brien [Tue, 12 Sep 2006 06:59:39 +0000 (06:59 +0000)]
Build 'ld' supporting a big endian ARM if "TARGET_BIG_ENDIAN" is defined.

17 years agoConfigure for for big endian if TARGET_BYTES_BIG_ENDIAN is defined.
David E. O'Brien [Tue, 12 Sep 2006 06:39:27 +0000 (06:39 +0000)]
Configure for for big endian if TARGET_BYTES_BIG_ENDIAN is defined.

17 years agoWARNS=6 chocked on unitialized variable.
Philippe Charnier [Tue, 12 Sep 2006 05:08:36 +0000 (05:08 +0000)]
WARNS=6 chocked on unitialized variable.

17 years agoIntroduce a new entry point, mac_create_mbuf_from_firewall. This entry point
Christian S.J. Peron [Tue, 12 Sep 2006 04:25:13 +0000 (04:25 +0000)]
Introduce a new entry point, mac_create_mbuf_from_firewall. This entry point
exists to allow the mandatory access control policy to properly initialize
mbufs generated by the firewall. An example where this might happen is keep
alive packets, or ICMP error packets in response to other packets.

This takes care of kernel panics associated with un-initialize mbuf labels
when the firewall generates packets.

[1] I modified this patch from it's original version, the initial patch
    introduced a number of entry points which were programmatically
    equivalent. So I introduced only one. Instead, we should leverage
    mac_create_mbuf_netlayer() which is used for similar situations,
    an example being icmp_error()

    This will minimize the impact associated with the MFC

Submitted by: mlaier [1]
MFC after: 1 week

This is a RELENG_6 candidate

17 years ago- Revert making bus_generic_add_child() the default for BUS_ADD_CHILD().
John Baldwin [Mon, 11 Sep 2006 22:20:37 +0000 (22:20 +0000)]
- Revert making bus_generic_add_child() the default for BUS_ADD_CHILD().
  Instead, we want busses to explicitly specify an add_child routine if they
  want to support identify routines, but by default disallow having outside
  drivers add devices.
- Give smbus(4) an explicit bus_add_child() method.

Requested by: imp

17 years agoAdd note about identify routines needing to use BUS_ADD_CHILD rather
Warner Losh [Mon, 11 Sep 2006 22:17:18 +0000 (22:17 +0000)]
Add note about identify routines needing to use BUS_ADD_CHILD rather
than device_add_child.  The latter is only for busses adding children,
not children drivers telling a bus that they have an instance...

17 years agoFix issues found by Coverity (223392, 223393) due to TSO additions
Prafulla Deuskar [Mon, 11 Sep 2006 20:59:01 +0000 (20:59 +0000)]
Fix issues found by Coverity (223392, 223393) due to TSO additions

Submitted by: Matthew Jacob

17 years agoMinor overhaul of SMBus support:
John Baldwin [Mon, 11 Sep 2006 20:52:41 +0000 (20:52 +0000)]
Minor overhaul of SMBus support:
- Change smbus_callback() to pass a void * rather than caddr_t.
- Change smbus_bread() to pass a pointer to the count and have it be an
  in/out parameter.  The input is the size of the buffer (same as before),
  but on return it will contain the actual amount of data read back from
  the bus.  Note that this value may be larger than the input value.  It
  is up to the caller to treat this as an error if desired.
- Change the SMB_BREAD ioctl to write out the updated struct smbcmd which
  will contain the actual number of bytes read in the 'count' field.  To
  preserve the previous ABI, the old ioctl value is mapped to SMB_OLD_BREAD
  which doesn't copy the updated smbcmd back out to userland.  I doubt anyone
  actually used the old BREAD anyway as it was rediculous to do a bulk-read
  but not tell the using program how much data was actually read.
- Make the smbus driver and devclass public in the smbus module and
  push all the DRIVER_MODULE()'s for attaching the smbus driver to
  various foosmb drivers out into the foosmb modules.  This makes all
  the foosmb logic centralized and allows new foosmb modules to be
  self-contained w/o having to hack smbus.c everytime a new smbus driver
  is added.
- Add a new SMB_EINVAL error bit and use it in place of EINVAL to return
  an error for bad arguments (such as invalid counts for bread and bwrite).
- Map SMB bus error bits to EIO in smbus_error().
- Make the smbus driver call bus_generic_probe() and require child drivers
  such as smb(4) to create device_t's via identify routines.  Previously,
  smbus just created one anonymous device during attach, and if you had
  multiple drivers that could attach it was just random chance as to which
  driver got to probe for the sole device_t first.
- Add a mutex to the smbus(4) softc and use it in place of dummy splhigh()
  to protect the 'owner' field and perform necessary synchronization for
  smbus_request_bus() and smbus_release_bus().
- Change the bread() and bwrite() methods of alpm(4), amdpm(4), and
  viapm(4) to only perform a single transaction and not try to use a
  loop of multiple transactions for a large request.  The framing and
  commands to use for a large transaction depend on the upper-layer
  protocol (such as SSIF for IPMI over SMBus) from what I can tell, and the
  smb(4) driver never allowed bulk read/writes of more than 32-bytes
  anyway.  The other smb drivers only performed single transactions.
- Fix buffer overflows in the bread() methods of ichsmb(4), alpm(4),
  amdpm(4), amdsmb(4), intpm(4), and nfsmb(4).
- Use SMB_xxx errors in viapm(4).
- Destroy ichsmb(4)'s mutex after bus_generic_detach() to avoid problems
  from child devices making smb upcalls that would use the mutex during
  their detach methods.

MFC after: 1 week
Reviewed by: jmg (mostly)

17 years agoAdd a new ddb command 'show lapic' to dump details about the local APIC
John Baldwin [Mon, 11 Sep 2006 20:12:42 +0000 (20:12 +0000)]
Add a new ddb command 'show lapic' to dump details about the local APIC
registers for the current CPU.

MFC after: 3 days

17 years agoActually hook up the IPI_INVLCACHE IDT vectors backing
John Baldwin [Mon, 11 Sep 2006 20:10:42 +0000 (20:10 +0000)]
Actually hook up the IPI_INVLCACHE IDT vectors backing
pmap_invalidate_cache() in the SMP case so pmap_mapdev() in multiuser
doesn't panic with a trap 30.  I broke this many months ago when I
added pmap_invalidate_cache() as early parts of the PAT work.

Patience from: jmg
Pointy hat: jhb

17 years agoFix a NULL pointer dereference of ro->ro_rt->rt_flags by checking for the
Andre Oppermann [Mon, 11 Sep 2006 19:56:10 +0000 (19:56 +0000)]
Fix a NULL pointer dereference of ro->ro_rt->rt_flags by checking for the
validity of ro->ro_rt first.  This prevents crashing on any non-normally
routed IP packet.

Coverity CID: 162 (incorrectly, it was re-introduced by previous commit)

17 years agoAdd support for a few more Serverworks and lookalikes chips.
Søren Schmidt [Mon, 11 Sep 2006 19:48:30 +0000 (19:48 +0000)]
Add support for a few more Serverworks and lookalikes chips.

17 years agoUpdate to fit latest USB cleanup crusade.
Søren Schmidt [Mon, 11 Sep 2006 19:47:53 +0000 (19:47 +0000)]
Update to fit latest USB cleanup crusade.

17 years agoAdd prototype for bus_generic_add_child() missed in previous commit.
John Baldwin [Mon, 11 Sep 2006 19:42:27 +0000 (19:42 +0000)]
Add prototype for bus_generic_add_child() missed in previous commit.

17 years agoAdd a default method for BUS_ADD_CHILD() that just calls
John Baldwin [Mon, 11 Sep 2006 19:41:31 +0000 (19:41 +0000)]
Add a default method for BUS_ADD_CHILD() that just calls
device_add_child_ordered().  Previously, a device driver that wanted to
add a new child device in its identify routine had to know if the parent
driver had a custom bus_add_child method and use BUS_ADD_CHILD() in that
case, otherwise use device_add_child().  Getting it wrong in either
direction would result in panics or failure to add the child device.  Now,
BUS_ADD_CHILD() always works isolating child drivers from having to know
intimate details about the parent driver.

Discussed with: imp
MFC after: 1 week

17 years agoAdd myself as OpenSSL janitor/maintainer.
Simon L. B. Nielsen [Mon, 11 Sep 2006 19:39:46 +0000 (19:39 +0000)]
Add myself as OpenSSL janitor/maintainer.

17 years agoGive the ACPI I/O rman's unique description strings to make 'devinfo -u'
John Baldwin [Mon, 11 Sep 2006 19:32:54 +0000 (19:32 +0000)]
Give the ACPI I/O rman's unique description strings to make 'devinfo -u'
output less confusing.

MFC after: 3 days

17 years ago- Fix rman_manage_region() to be a lot more intelligent. It now checks
John Baldwin [Mon, 11 Sep 2006 19:31:52 +0000 (19:31 +0000)]
- Fix rman_manage_region() to be a lot more intelligent.  It now checks
  for overlaps, but more importantly, it collapses adjacent free regions.
  This is needed to cope with BIOSen that split up ports for system devices
  (like IPMI controllers) across multiple system resource entries.
- Now that rman_manage_region() is not so dumb, remove extra logic in the
  x86 nexus drivers to populate the IRQ rman that manually coalesced the
  regions.

MFC after: 1 week

17 years agoCF devices are ATA not ATAPI.
Søren Schmidt [Mon, 11 Sep 2006 18:33:59 +0000 (18:33 +0000)]
CF devices are ATA not ATAPI.

17 years agoDefine build 'as' TARGET_BYTES_BIG_ENDIAN if "TARGET_BIG_ENDIAN" is defined.
David E. O'Brien [Mon, 11 Sep 2006 18:07:09 +0000 (18:07 +0000)]
Define build 'as' TARGET_BYTES_BIG_ENDIAN if "TARGET_BIG_ENDIAN" is defined.
This is needed for bi-sex architectures.

17 years agonull commit to provide commit message to previous
Matt Jacob [Mon, 11 Sep 2006 17:57:23 +0000 (17:57 +0000)]
null commit to provide commit message to previous
at the request of Sam Leffler: The previous commit
established min and maxtags for VMware pseudo disks
to fix a submitted PR.

17 years agoPR: 103130
Matt Jacob [Mon, 11 Sep 2006 17:34:28 +0000 (17:34 +0000)]
PR: 103130
Submitted by: Shusuke Shinomiya
MFC after: 1 day

17 years agoRegen.
Ruslan Ermilov [Mon, 11 Sep 2006 13:56:28 +0000 (13:56 +0000)]
Regen.

17 years agoDocument the new know, WITHOUT_ASSERT_DEBUG.
Ruslan Ermilov [Mon, 11 Sep 2006 13:55:27 +0000 (13:55 +0000)]
Document the new know, WITHOUT_ASSERT_DEBUG.

17 years ago- Utilize the "showconfig" top-level target.
Ruslan Ermilov [Mon, 11 Sep 2006 13:39:44 +0000 (13:39 +0000)]
- Utilize the "showconfig" top-level target.
- Pass __MAKE_CONF=/dev/null to get a pristine output.

17 years agoMake it possible again to build -CURRENT snapshots on -STABLE.
Ruslan Ermilov [Mon, 11 Sep 2006 13:15:09 +0000 (13:15 +0000)]
Make it possible again to build -CURRENT snapshots on -STABLE.

Reported by: Angelo Turetta

17 years agoAdd support for MCP61 and MCP65 adaptors.
David E. O'Brien [Mon, 11 Sep 2006 07:31:55 +0000 (07:31 +0000)]
Add support for MCP61 and MCP65 adaptors.

Submitted by: Peer Chen at nVIDIA

17 years agoThe run_filter() procedure is a means of working around DMA engine bugs in
Scott Long [Mon, 11 Sep 2006 06:48:53 +0000 (06:48 +0000)]
The run_filter() procedure is a means of working around DMA engine bugs in
old/broken hardware.  Unfortunately, it adds cache pressure and possible
mispredicted branches to the fast path of the bus_dmamap_load collection of
functions.  Since it's meant for slow path exception processing, de-inline
it and allow its conditions to be pre-computed at tag_create time and thus
short-circuited at runtime.

While here, cut down on the size of _bus_dmamap_load_buffer() by pushing the
bounce page logic into a non-inlined function.  Again, this helps with
cache pressure and mispredicted branches.

According to the TSC, this shaves off a few cycles on average.  Unfortunately,
the data varies quite a bit due to interrupts and preemption, so it's hard to
get a good measurement.  Real world measurements of network PPS are welcomed.
A merge to amd64 and other arches is pending more testing.

17 years agoAdd a knob for compiling the tree -DNDEBUG. This turns off all the
Warner Losh [Mon, 11 Sep 2006 05:35:57 +0000 (05:35 +0000)]
Add a knob for compiling the tree -DNDEBUG.  This turns off all the
asserts and makes binaries smaller.  The binaries also become
repeatable again.  As it was, without this md5's of binaries built
with different paths differed.

# Where do I document this?

17 years agoFix three whitespace nits.
Warner Losh [Mon, 11 Sep 2006 05:12:12 +0000 (05:12 +0000)]
Fix three whitespace nits.

I think these are from kan@...

17 years agoFix a typo
Brian Somers [Sun, 10 Sep 2006 20:41:33 +0000 (20:41 +0000)]
Fix a typo

17 years agoCorrect incorrect PKCS#1 v1.5 padding validation in crypto(3).
Simon L. B. Nielsen [Sun, 10 Sep 2006 20:16:43 +0000 (20:16 +0000)]
Correct incorrect PKCS#1 v1.5 padding validation in crypto(3).

Obtained from: OpenSSL project
Security: FreeBSD-SA-06:19.openssl

17 years agoFix style(9) issues in the TSO specific changes.
Prafulla Deuskar [Sun, 10 Sep 2006 19:23:27 +0000 (19:23 +0000)]
Fix style(9) issues in the TSO specific changes.

Pointed out by: jmallett

17 years agomake use of the host route's mtu for processing. This means we can now
John-Mark Gurney [Sun, 10 Sep 2006 17:49:09 +0000 (17:49 +0000)]
make use of the host route's mtu for processing.  This means we can now
support a network w/ split mtu's by assigning each host route the correct
mtu.  an aspiring programmer could write a daemon to probe hosts and find
out if they support a larger mtu.

17 years agoNew sockets created by incoming connections into listen sockets should
Andre Oppermann [Sun, 10 Sep 2006 17:08:06 +0000 (17:08 +0000)]
New sockets created by incoming connections into listen sockets should
inherit all settings and options except listen specific options.

Add the missing send/receive timeouts and low watermarks.
Remove inheritance of the field so_timeo which is unused.

Noticed by: phk
Reviewed by: rwatson
Sponsored by: TCP/IP Optimization Fundraise 2005
MFC after: 3 days

17 years agoFix locking race in ttymodem(). The locking of the proctree happens too late
Martin Blapp [Sun, 10 Sep 2006 16:51:56 +0000 (16:51 +0000)]
Fix locking race in ttymodem(). The locking of the proctree happens too late
and opens a small race window before tp->t_session->s_leader is accessed. In case
tp->t_session has just been set to NULL elsewhere, we get a panic().

This fix is a bandaid until someone else fixes the whole locking in the tty subsystem.
Definitly more work needs to be done.

MFC after: 1 week
Reviewed by: mlaier
PR: kern/103101

17 years agoYank FreeBSD specific code out from under ifdef.
Tom Rhodes [Sun, 10 Sep 2006 15:20:39 +0000 (15:20 +0000)]
Yank FreeBSD specific code out from under ifdef.
Remove compatibility ifdefs.

Reviewed/Ok'ed by: imp

17 years agoThe Linux unlink syscall uses a different errno value when trying to unlink
Alexander Leidinger [Sun, 10 Sep 2006 13:47:56 +0000 (13:47 +0000)]
The Linux unlink syscall uses a different errno value when trying to unlink
a directory.

PR: 102897 [1]
Noticed by: Knut Anders Hatlen <kahatlen@gmail.com>, testrun with LTP [1]
Submitted by: Marcin Cieslak <saper@SYSTEM.PL>
Tested by: netchild (LTP test run)

17 years agoDelay an orphan event if provider has still in-flight I/O requests.
Pawel Jakub Dawidek [Sun, 10 Sep 2006 09:11:54 +0000 (09:11 +0000)]
Delay an orphan event if provider has still in-flight I/O requests.
This way GEOM classes can safely detach from provider when an orphan
event is received. This fixes 'detach with active requests' panic for
gstripe/gconcat under load.

PR: kern/102766
Submitted by: mjacob
OK'ed by: phk
MFC after: 1 week

17 years agoFix the manual build.
David E. O'Brien [Sun, 10 Sep 2006 08:58:47 +0000 (08:58 +0000)]
Fix the manual build.

17 years agoSync with NetBSD rev. 1.16 + 1.17
Martin Blapp [Sat, 9 Sep 2006 22:33:21 +0000 (22:33 +0000)]
Sync with NetBSD rev. 1.16 + 1.17

Coverity CID 2292: Plug memory leak.
Coverity CID 2291: Move function call before allocating storage to prevent
memory leak on error.

MFC after:    1 month

17 years agoSync with NetBSD rev. 1.29
Martin Blapp [Sat, 9 Sep 2006 22:32:07 +0000 (22:32 +0000)]
Sync with NetBSD rev. 1.29

Coverity CID 2293: Fix memory leak.

MFC after: 1 month

17 years agoSync with NetBSD rev. 1.6 + 1.7
Martin Blapp [Sat, 9 Sep 2006 22:29:16 +0000 (22:29 +0000)]
Sync with NetBSD rev. 1.6 + 1.7

Coverity CID 779 + 780: Avoid NULL pointer dereference.

MFC after:    1 month

17 years agoSync with NetBSD rev. 1.20 + 1.21
Martin Blapp [Sat, 9 Sep 2006 22:26:47 +0000 (22:26 +0000)]
Sync with NetBSD rev. 1.20 + 1.21

Coverity CID 2284: Fix multiple memory leaks.
Coverity CID 710: Remove unreachable code.

MFC after:    1 month

17 years agoSync with NetBSD rev. 1.10
Martin Blapp [Sat, 9 Sep 2006 22:22:39 +0000 (22:22 +0000)]
Sync with NetBSD rev. 1.10

Coverity CID 2276: Don't leak memory on error.

MFC after:    1 month

17 years agoSync with NetBSD rev. 1.15
Martin Blapp [Sat, 9 Sep 2006 22:21:15 +0000 (22:21 +0000)]
Sync with NetBSD rev. 1.15

Coverity CID 2275: Avoid memory leak on error.

MFC after:    1 month

17 years agoSync part of NetBSD rev. 1.14
Martin Blapp [Sat, 9 Sep 2006 22:18:57 +0000 (22:18 +0000)]
Sync part of NetBSD rev. 1.14

Fix unpaired sigblock which possibly leaves the process with all
signals blocked.

MFC after: 2 weeks

17 years agoSync with NetBSD rev. 1.16 + 1.17
Martin Blapp [Sat, 9 Sep 2006 22:14:42 +0000 (22:14 +0000)]
Sync with NetBSD rev. 1.16 + 1.17

Coverity CID 3651: Don't leak ifaddrs on getaddrinfo failure.
Coverity CID 2283: Don't leak sys_auth on error.

MFC after:    1 month

17 years agomove created/detected/activated under debug level 1 to quiet the common case..
John-Mark Gurney [Sat, 9 Sep 2006 21:45:37 +0000 (21:45 +0000)]
move created/detected/activated under debug level 1 to quiet the common case..

add count of active and total components to the launched line so you can
see at a glance if your mirror/raid3 is complete...

now:
GEOM_MIRROR: Device mirror/sam launched (2/2).

Reviewed by: pjd

17 years agoSecond attempt at fixing module build
Prafulla Deuskar [Sat, 9 Sep 2006 20:05:24 +0000 (20:05 +0000)]
Second attempt at fixing module build

Pointyhat: pdeuskar

17 years agoFix build breakage while compiling em as a module.
Prafulla Deuskar [Sat, 9 Sep 2006 19:55:13 +0000 (19:55 +0000)]
Fix build breakage while compiling em as a module.

17 years agoBuild linprocfs and linsysfs as modules on amd64.
Alexander Leidinger [Sat, 9 Sep 2006 16:58:22 +0000 (16:58 +0000)]
Build linprocfs and linsysfs as modules on amd64.

Sponsored by: Google SoC 2006
Submitted by: rdivacky

17 years ago- Extend the coverage of PROC_LOCK to cover wakeup(&p->p_emuldata);
Alexander Leidinger [Sat, 9 Sep 2006 16:55:55 +0000 (16:55 +0000)]
- Extend the coverage of PROC_LOCK to cover wakeup(&p->p_emuldata);
- Lock the emuldata in a case when we just created it.

Sponsored by: Google SoC 2006
Submitted by: rdivacky
Suggested by: jhb

17 years agoAdd struct msg to the forwarded declared data structures in mac_policy.h.
Robert Watson [Sat, 9 Sep 2006 16:35:44 +0000 (16:35 +0000)]
Add struct msg to the forwarded declared data structures in mac_policy.h.

Obtained from: TrustedBSD Project

17 years agoChange futex lock from mutex to sx. Make futex_get atomic (protected by the
Alexander Leidinger [Sat, 9 Sep 2006 16:25:25 +0000 (16:25 +0000)]
Change futex lock from mutex to sx. Make futex_get atomic (protected by the
futex lock).

Sponsored by: Google SoC 2006
Submitted by: rdivacky
Suggested by: jhb

17 years agoRemove slightly oddly placed suser() call from the KTR/ALQ setup sysctl:
Robert Watson [Sat, 9 Sep 2006 16:09:01 +0000 (16:09 +0000)]
Remove slightly oddly placed suser() call from the KTR/ALQ setup sysctl:
it was present only in the enable path, not the disable path, which one
presumes would be equally of interest.  Either way, it was not needed,
as the sysctl framework already calls suser() if the operation is a
write operation, which configuration requests are.

Sponsored by: nCircle Network Security, Inc.

17 years agoFix the check where we want to use the end of the supported range if the
Alexander Leidinger [Sat, 9 Sep 2006 14:43:03 +0000 (14:43 +0000)]
Fix the check where we want to use the end of the supported range if the
value is out of the supported range.

Noticed by: Ed Schouten <ed@fxq.nl>
Reviewed by: Kazuhito HONDA <kazuhito@ph.noda.tus.ac.jp>

17 years ago- don't wake every sleeper just the first one [1]
Alexander Leidinger [Sat, 9 Sep 2006 13:04:28 +0000 (13:04 +0000)]
- don't wake every sleeper just the first one [1]
- remove debuging printf [2]

Submitted by: intron <mag@intron.ac> [1], rdivacky [2]

17 years agoAdd a BSM conversion switch case for AUE_GETCWD, so that a console
Robert Watson [Sat, 9 Sep 2006 10:23:44 +0000 (10:23 +0000)]
Add a BSM conversion switch case for AUE_GETCWD, so that a console
warning isn't generated when __getcwd() is invoked.

MFC after: 3 days
Obtained from: TrustedBSD Project

17 years agoSmall style cleanup.
Robert Watson [Sat, 9 Sep 2006 10:23:00 +0000 (10:23 +0000)]
Small style cleanup.

MFC after: 3 days

17 years agoAudit sysarch() operation argument.
Robert Watson [Sat, 9 Sep 2006 10:20:31 +0000 (10:20 +0000)]
Audit sysarch() operation argument.

MFC after: 3 days

17 years agoCatch up to xchat and samba package changes
Kris Kennaway [Sat, 9 Sep 2006 07:25:40 +0000 (07:25 +0000)]
Catch up to xchat and samba package changes

17 years agoAdd support for TSO. Thanks to Andre for adding support in the stack
Prafulla Deuskar [Sat, 9 Sep 2006 06:19:20 +0000 (06:19 +0000)]
Add support for TSO. Thanks to Andre for adding support in the stack
and Jack Vogel for driver changes.

Submitted by: Jack Vogel

17 years agoBump __FreeBSD_version for rev. 1.117 of libexec/rtld-elf/rtld.c.
Konstantin Belousov [Sat, 9 Sep 2006 04:41:40 +0000 (04:41 +0000)]
Bump __FreeBSD_version for rev. 1.117 of libexec/rtld-elf/rtld.c.

Requested by: jkim
Approved by: kan (mentor)

17 years agoAdd support to bge(4) to not break IPMI support when the driver attaches
Doug Ambrisko [Sat, 9 Sep 2006 03:36:57 +0000 (03:36 +0000)]
Add support to bge(4) to not break IPMI support when the driver attaches
to it.  Try to co-operate with the IPMI/ASF firmware accessing the PHY.
One we get link we don't mess with the PHY.  If we do then over time
the NIC will go off line.  It would be nice if we could tell if IPMI
was enabled on the chip but I can't figure out a reliable way to do
that.  The scheme I tried worked on a Dell PE850 but not on an HP machine.
So we assume any NIC that has ASF capability needs to deal with it.

The code was inspired by the support in Linux from kernel.org and Broadcom.
Broadcom did give me some info. but it is rather limited and is mostly
just what is in the Linux driver.  Thanks to the numerous people that
helped debug the many prior versions and that I didn't break other
bge(4) HW.

Reviewed by: several people
Tested by: even more

17 years agoThe following functions need not to be reimplemented, reuse 64bit
David Xu [Sat, 9 Sep 2006 01:22:13 +0000 (01:22 +0000)]
The following functions need not to be reimplemented, reuse 64bit
syscalls instead:
sigqueue, thr_set_name, thr_setscheduler, thr_getscheduler,
thr_setschedparam.

17 years agoFix style nits. No md5 changes in .o's. ;-)
Jung-uk Kim [Fri, 8 Sep 2006 21:46:01 +0000 (21:46 +0000)]
Fix style nits.  No md5 changes in .o's. ;-)

17 years agoStop cutting the tarball names to eight symbols. This makes
Ruslan Ermilov [Fri, 8 Sep 2006 19:54:37 +0000 (19:54 +0000)]
Stop cutting the tarball names to eight symbols.  This makes
it possible to e.g. distribute kernels with config names larger
than eight symbols, without the clash.  Previousy, LALALALA
and LALALALA-SMP would be the same tarball.  (I think this
comes from the old days where tarballs were put on the MS-DOS
formatted diskettes.)

MFC after: 3 days

17 years agoChange the way base allocation is done for internal malloc data
Jason Evans [Fri, 8 Sep 2006 17:52:15 +0000 (17:52 +0000)]
Change the way base allocation is done for internal malloc data
structures, in order to avoid the possibility of attempted recursive
lock acquisition for chunks_mtx.

Reported by: Slawa Olhovchenkov <slw@zxy.spb.ru>

17 years agoMF6: Attach if_bce.ko to the build.
Ruslan Ermilov [Fri, 8 Sep 2006 15:40:36 +0000 (15:40 +0000)]
MF6: Attach if_bce.ko to the build.

Noticed by: davidch

17 years agoTemporarily put ugly workaround in place to make sparc64 build happy.
Maksim Yevmenkin [Fri, 8 Sep 2006 15:05:38 +0000 (15:05 +0000)]
Temporarily put ugly workaround in place to make sparc64 build happy.
panther is down at the moment, so better fix will follow later.

17 years agoWhen looking up the symbol by dlsym, look it not only in the object
Konstantin Belousov [Fri, 8 Sep 2006 14:59:54 +0000 (14:59 +0000)]
When looking up the symbol by dlsym, look it not only in the object
given as dso handle, but also in the implicit dependencies of that dso.

Also, const-ify the read-only parameter objlist of symlook_list.

Reported by: "Simon 'corecode' Schubert" <corecode at fs ei tum de>
Approved by: kan (mentor)
X-MFC-After: 6.2

17 years agoDocument that bits are numbered from 1 in the %b format argument.
Ruslan Ermilov [Fri, 8 Sep 2006 14:05:03 +0000 (14:05 +0000)]
Document that bits are numbered from 1 in the %b format argument.

17 years agoRe-word the description of the "async" flag.
Daniel Gerzo [Fri, 8 Sep 2006 13:47:39 +0000 (13:47 +0000)]
Re-word the description of the "async" flag.

Suggested by: Milos Vyletel (mv@rulez.sk)
Reviewed by: ru
Approved by: keramida (mentor), trhodes (mentor)

17 years agoFix format character.
Pawel Jakub Dawidek [Fri, 8 Sep 2006 13:46:18 +0000 (13:46 +0000)]
Fix format character.

Reported by: andre

17 years agoFix octal representation of TSO4 and TSO6 bits in interface capabilities
Andre Oppermann [Fri, 8 Sep 2006 13:29:38 +0000 (13:29 +0000)]
Fix octal representation of TSO4 and TSO6 bits in interface capabilities
description.

17 years agoAdd a sysctl net.inet.tcp.nolocaltimewait that allows to suppress
Gleb Smirnoff [Fri, 8 Sep 2006 13:09:15 +0000 (13:09 +0000)]
Add a sysctl net.inet.tcp.nolocaltimewait that allows to suppress
creating a compress TIME WAIT states, if both connection endpoints
are local. Default is off.

17 years agoBump copyright year.
Pawel Jakub Dawidek [Fri, 8 Sep 2006 10:20:44 +0000 (10:20 +0000)]
Bump copyright year.

17 years agoUse __FBSDID in .c files.
Pawel Jakub Dawidek [Fri, 8 Sep 2006 10:19:24 +0000 (10:19 +0000)]
Use __FBSDID in .c files.

17 years agoIn "make universe", utilize simplified cross-build synopsis,
Ruslan Ermilov [Fri, 8 Sep 2006 10:13:15 +0000 (10:13 +0000)]
In "make universe", utilize simplified cross-build synopsis,
"make TARGET=<machine> build*", now that it actually works.

Prompted by: imp

17 years agoPrevious revision wasn't enough for "make TARGET=<machine> build*"
Ruslan Ermilov [Fri, 8 Sep 2006 10:09:02 +0000 (10:09 +0000)]
Previous revision wasn't enough for "make TARGET=<machine> build*"
to work.  Some build-tools that are part of cross-tools (cc_tools)
also need to know the proper values of TARGET_ARCH and TARGET, so
tell them.

17 years agoUpdate some of the bulgarian holidays since it no longer
Erwin Lansing [Fri, 8 Sep 2006 09:45:06 +0000 (09:45 +0000)]
Update some of the bulgarian holidays since it no longer
is a socialist state.

Submitted by: Shteryana Shopova <shteryana@FreeBSD.org>

17 years agoUse return value of _thr_umutex_lock instead of using zero.
David Xu [Fri, 8 Sep 2006 09:29:14 +0000 (09:29 +0000)]
Use return value of _thr_umutex_lock instead of using zero.

17 years ago- Split failure probability configuration into read failure probability and
Pawel Jakub Dawidek [Fri, 8 Sep 2006 09:21:21 +0000 (09:21 +0000)]
- Split failure probability configuration into read failure probability and
  write failure probability.
- Allow to specify an error number to return of failure.

MFC after: 3 days

17 years agoMinor comment fix.
Thomas Quinot [Fri, 8 Sep 2006 08:14:32 +0000 (08:14 +0000)]
Minor comment fix.

17 years agoSupport for PCI-Express 4Gb Cards.
Matt Jacob [Fri, 8 Sep 2006 05:27:04 +0000 (05:27 +0000)]
Support for PCI-Express 4Gb Cards.

17 years agoSo there is where that handbook paragraph came from. Kill it here too.
Tom Rhodes [Fri, 8 Sep 2006 04:56:21 +0000 (04:56 +0000)]
So there is where that handbook paragraph came from.  Kill it here too.
Remove a paragraph about over building security, it's a bit off.

Discussed with: des, FreeBSD-security

17 years agoMake 8139C+ work again which was broken since rev 1.68.
Pyun YongHyeon [Fri, 8 Sep 2006 00:58:02 +0000 (00:58 +0000)]
Make 8139C+ work again which was broken since rev 1.68.
Ever since rev 1.68 re(4) checks the validity of link in re_start.
But rlphy(4) got a garbled data due to a different bit layout used on
8139C+ and it couldn't report correct link state. To fix it, ignore
BMCR_LOOP and BMCR_ISO bits which have different meanings on 8139C+.
I think this also make dhclient(8) work on 8139C+.

Reported by: Gerrit Kuehn <gerrit AT pmp DOT uni-hannover DOT de>
Tested by: Gerrit Kuehn <gerrit AT pmp DOT uni-hannover DOT de>

17 years agoSigh. Fix name/path to SA-06:20.
Bruce A. Mah [Thu, 7 Sep 2006 23:49:26 +0000 (23:49 +0000)]
Sigh.  Fix name/path to SA-06:20.

17 years agoMFC noted: lukemftpd from NetBSD snapshot.
Bruce A. Mah [Thu, 7 Sep 2006 23:41:11 +0000 (23:41 +0000)]
MFC noted:  lukemftpd from NetBSD snapshot.

17 years agos/USBDEVNAME/device_get_nameunit/g
Maksim Yevmenkin [Thu, 7 Sep 2006 23:38:09 +0000 (23:38 +0000)]
s/USBDEVNAME/device_get_nameunit/g
s/USBBASEDEVICE/device_t/g

17 years agoCreate a 'ready' handler for each personality. The purpose of this handler
Matt Jacob [Thu, 7 Sep 2006 23:08:21 +0000 (23:08 +0000)]
Create a 'ready' handler for each personality. The purpose of this handler
is to able to be called after *all* attach and enable events are done.

We establish a SYSINIT hook to call this handler. The current usage for it
is to add scsi target resources *after* all enables are done. There seems
to be some dependencies between different halves of a dual-port with respect
to target mode.

Put in more meaningful event messages for some events- in particular
QUEUE FULL events so we can see what the queue depth was when the
IOC sent us this message.

MFC after: 1 week

17 years agoMerge from DRM upstream:
Eric Anholt [Thu, 7 Sep 2006 23:04:47 +0000 (23:04 +0000)]
Merge from DRM upstream:
- Add support for Intel 965 Express chipsets.
- Add support for R200 vertex programs, along with minor bugfixes.
- Add support for vblank synchronization to pipe B of Intel hardware
  (laptop screens).

17 years agoAdd bthidd(8) rc(8) script
Maksim Yevmenkin [Thu, 7 Sep 2006 22:25:08 +0000 (22:25 +0000)]
Add bthidd(8) rc(8) script

MFC after: 1 month

17 years agoMFp4: first cut at getting I2C transfers working (generically). I'm
Warner Losh [Thu, 7 Sep 2006 21:53:28 +0000 (21:53 +0000)]
MFp4: first cut at getting I2C transfers working (generically).  I'm
unsure if this driver correctly implements all the start/stop junk
right (but it did or didn't before I made this commit).

17 years agoMFp4: berndt pointed me at an errata that shows that the stat register
Warner Losh [Thu, 7 Sep 2006 21:50:01 +0000 (21:50 +0000)]
MFp4: berndt pointed me at an errata that shows that the stat register
offsets were originally documented incorrectly.  This fixes that.  It
shouldn't affect anything other than error stat reporting.

17 years agoUpdate bthidd(8) code and hook it up to the build.
Maksim Yevmenkin [Thu, 7 Sep 2006 21:47:49 +0000 (21:47 +0000)]
Update bthidd(8) code and hook it up to the build.
bthidd(8) now was integrated with vkbd(4) and supports
multiple keyboards via vkbd(4)/kbdmux(4).

The code was tested with Apple Bluetooth keyboard and
SE k700i cell phone (remote control feature).

MFC after: 1 month

17 years agoWhile experimenting with cross-building by specifying just TARGET,
Ruslan Ermilov [Thu, 7 Sep 2006 20:27:56 +0000 (20:27 +0000)]
While experimenting with cross-building by specifying just TARGET,
I found one bug.  Pass our idea of TARGET_ARCH and TARGET down to
XMAKE, the cross-tools make.  Previously it worked because usually
TARGET_ARCH was specified on the initial make's command line.
This should also allow us to simplify the "universe" target, which
I'm currently testing.