]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
17 years agoIn some particular cases (like in pccard and pccbb), the real device
piso [Thu, 31 May 2007 19:25:35 +0000 (19:25 +0000)]
In some particular cases (like in pccard and pccbb), the real device
handler is wrapped in a couple of functions - a filter wrapper and an
ithread wrapper. In this case (and just in this case), the filter
wrapper could ask the system to schedule the ithread and mask the
interrupt source if the wrapped handler is composed of just an ithread
handler: modify the "old" interrupt code to make it support
this situation, while the "new" interrupt code is already ok.

Discussed with: jhb

17 years agoLast major commit and updates for RELENG_7:
ariff [Thu, 31 May 2007 18:43:33 +0000 (18:43 +0000)]
Last major commit and updates for RELENG_7:

- Rework the entire pcm_channel structure:
  * Remove rarely used link placeholder, instead, make each pcm_channel
    as head/link of each own/each other. Unlock - Lock sequence due to
    sleep malloc has been reduced.
  * Implement "busy" queue which will contain list of busy/active
    channels. This greatly reduce locking contention for example while
    servicing interrupt for hardware with many channels or when virtual
    channels reach its 256 peak channels.

- So I heard you like v chan ... O RLY?
  Welcome to Virtual **Record** Channels (vrec, rec vchans, vchans for
  recording, Rec-Chan, you decide), the ultimate solutions for your
  nagging O_RDWR full-duplex wannabe (note: flash plugins) monopolizing
  single record channel causing EBUSY.  Vrec works exactly like Vchans
  (or, should I rename it to "Vplay" :) , except that it operates on the
  opposite direction (recording). Up to 256 vrecs (like vchans) are
  possible.

  Notes:
   * Relocate dev.pcm.%d.{vchans,vchanformat,vchanrate} to each of its
     respective node/direction:
       dev.pcm.%d.play.* for "play"   (cdev = dsp%d.vp%d)
       dev.pcm.%d.rec.*  for "record" (cdev = dsp%d.vr%d)
   * Don't expect that it will magically give you ability to split
     "recording source" (eg: 1 channel for cdrom, 1 channel for mic,
     etc). Just admit that you only have a *single* recording source /
     channel. Please bug your hardware vendor instead :)

- Bump maxautovchans from 4 to 16. For a full-fledged multimedia
  desktop/workstation with too many soundservers installed (esound,
  artsd, jackd, pulse/polypaudio, ding-dong pling plong mudkip fuh fuh,
  etc), 4 seems inadequate. There will be no memory penalty here, since
  virtual channels are allocate only by demand.

- Nuke/Rework the entire statically created cdev entries. Everything is
  clonable through snd own clone manager which designed to withstand many
  kind of abusive devfs droids such as:
      * while : ; do /bin/test -e /dev/dsp ; done
      * jot 16777216 0 | while read x ; do ls /dev/dsp0.$x ; done
      * hundreds (could be thousands) concurrent threads/process opening
"/dev/dsp" (previously, this might result EBUSY even with just
3 contesting threads/procs).
  o Reusable clone objects (instead of creating new one like there's no
    tomorrow) after certain expiration deadline. The clone allocator will
    decide whether to reuse, share, or creating new clone.
  o Automatic garbage collector.

- Dynamic unit magic allocator. Maximum attached soundcards can be tuned
  using tunable "hw.snd.maxunit" (Default to 512). Minimum is 16, and
  maximum is 2048.

- ..other fixes, mostly related to concurrency issues.

joel@ will do the manpage updates on sound(4).

Have fun.

17 years agoLast major commit and updates for RELENG_7:
ariff [Thu, 31 May 2007 18:35:24 +0000 (18:35 +0000)]
Last major commit and updates for RELENG_7:

Add few new files. The _real_ commit will follow shortly, so fasten
up your seatbelts, sit back and enjoy the ride..

17 years agoInstall omp.h file.
kan [Thu, 31 May 2007 13:21:30 +0000 (13:21 +0000)]
Install omp.h file.

Submitted by: Pieter de Goeje (pieter at degoejes dot nl)

17 years agoExport quad symbols. They were previously commented out. These symbols
deischen [Thu, 31 May 2007 13:07:37 +0000 (13:07 +0000)]
Export quad symbols.  They were previously commented out.  These symbols
really shouldn't be exported since they should be pulled from libgcc, but
the build of some applications is broken and they expect to see them in
libc.  glibc exports these symbols, although Solaris doesn't appear to,
so export them for compatibility's sake.

After discussion with: kan

17 years agoSome libc symbol map cleanups.
deischen [Thu, 31 May 2007 13:01:34 +0000 (13:01 +0000)]
Some libc symbol map cleanups.

  net: endhostdnsent is named _endhostdnsent and is
  private to netdb family of functions.

  posix1e: acl_size.c has been never compiled in,
  so there's no "acl_size".

  rpc: "getnetid" is a static function.

  stdtime: "gtime" is #ifdef'ed out in the source.

  some symbols are specific only to some architectures,
  e.g., ___tls_get_addr is only defined on i386.

  __htonl, __htons, __ntohl and __ntohs are no longer
  functions, they are now (internal) defines in
  <machine/endian.h>.

Submitted by: ru

17 years ago(1) In tcp_usrclosed(), tp can never become NULL, so don't test for NULL
rwatson [Thu, 31 May 2007 12:06:02 +0000 (12:06 +0000)]
(1) In tcp_usrclosed(), tp can never become NULL, so don't test for NULL
    before handling the socket disconnection case.

(2) Clean up surrounding comments and formatting.

Found with: Coverity Prevent(tm) (1)
CID: 2203

17 years agoRevert UF_OPENING workaround for CURRENT.
kib [Thu, 31 May 2007 11:51:53 +0000 (11:51 +0000)]
Revert UF_OPENING workaround for CURRENT.
Change the VOP_OPEN(), vn_open() vnode operation and d_fdopen() cdev operation
argument from being file descriptor index into the pointer to struct file.

Proposed and reviewed by: jhb
Reviewed by: daichi (unionfs)
Approved by: re (kensmith)

17 years agoNow that sx(9) locks support an interruptible lock acquire primitive,
rwatson [Thu, 31 May 2007 11:51:22 +0000 (11:51 +0000)]
Now that sx(9) locks support an interruptible lock acquire primitive,
properly observe the SB_NOINTR flag in sblock.  This restores the
required behavior that lock acquisition be interruptible on the socket
buffer I/O serialization lock to allow threads waiting for I/O to be
signaled even if they aren't the thread currently holding the I/O lock.
With this change, the sblock regression test is again passed.

Reported by: alfred
sx(9) handiwork: attilio

17 years agoAdd CPUID2_PDCM
des [Thu, 31 May 2007 11:26:45 +0000 (11:26 +0000)]
Add CPUID2_PDCM

Requested by: jkim
MFC after: 3 days

17 years agoAdd functions sx_xlock_sig() and sx_slock_sig().
attilio [Thu, 31 May 2007 09:14:48 +0000 (09:14 +0000)]
Add functions sx_xlock_sig() and sx_slock_sig().
These functions are intended to do the same actions of sx_xlock() and
sx_slock() but with the difference to perform an interruptible sleep, so
that sleep can be interrupted by external events.
In order to support these new featueres, some code renstruction is needed,
but external API won't be affected at all.

Note: use "void" cast for "int" returning functions in order to avoid tools
like Coverity prevents to whine.

Requested by: rwatson
Tested by: rwatson
Reviewed by: jhb
Approved by: jeff (mentor)

17 years agoUse ACPICA defined value for notification rather than locally defined one.
takawata [Thu, 31 May 2007 08:49:51 +0000 (08:49 +0000)]
Use ACPICA defined value for notification rather than locally defined one.

17 years agoDiscard backlog on GDB port when connected.
simokawa [Thu, 31 May 2007 04:55:05 +0000 (04:55 +0000)]
Discard backlog on GDB port when connected.

MFC after: 3 days

17 years agoAdd information for FreeBSD 1.0.2 from November 1993. I have a
imp [Thu, 31 May 2007 03:40:29 +0000 (03:40 +0000)]
Add information for FreeBSD 1.0.2 from November 1993.  I have a
January 1994 Trans-Ameritech Release 1 CD-ROM that has this directory
named "update_bsd_1.0.2" dated November 14, 1993.

I didn't add it into the main-line sequence of releases because that
would disrupt things too much...

17 years agoRemove "acpi_bus_number: can't get _ADR" message. It usually appears as
njl [Thu, 31 May 2007 00:52:32 +0000 (00:52 +0000)]
Remove "acpi_bus_number: can't get _ADR" message.  It usually appears as
we traverse \_SB and \ in the namespace, which won't have _ADR anyway.
Use a proper extern instead of our own private copy.

MFC after: 1 week

17 years agoGiant is special. How do I love thee? Let me count the ways?
imp [Thu, 31 May 2007 00:05:59 +0000 (00:05 +0000)]
Giant is special.  How do I love thee?  Let me count the ways?

errr, I mean "Enumerate how the giant lock differs from other locks"

Please let me know if I missed any.  Or misrepresented any...

Reviewed by: ssouhlal@

17 years agoA few small but significant fixes:
jfv [Wed, 30 May 2007 23:32:21 +0000 (23:32 +0000)]
A few small but significant fixes:
 - Coverity Prevent(tm) CID 1906 a bogus use of bzero where unneeded.
 - ICH8 systems autoneg to 100 rather than 1000, this can also be
   seen in 82573, the logic was backwards.
 - On new 82575 quadports half duplex tx speed is slow... this was due
   to overwriting TCTL reg rather than adding bits.

17 years ago- Fixed (Apple) compiler warnings in sctp_input.c, sctputil.c, sctp_output.c
rrs [Wed, 30 May 2007 22:34:21 +0000 (22:34 +0000)]
- Fixed (Apple) compiler warnings in sctp_input.c, sctputil.c, sctp_output.c
- Fixed a LOR in handling a cookie. Turns out create lock is applied.
  And if we abort processing, this causes LOR. Changed to force the
  timer to clean up, that way create lock is released.

17 years agoRemove a KASSERT intended to help the developer, the condition is no longer
thompsa [Wed, 30 May 2007 19:39:55 +0000 (19:39 +0000)]
Remove a KASSERT intended to help the developer, the condition is no longer
valid since the span code was added.

PR: kern/113170
MFC after: 1 week

17 years ago- Fix a memory overwrite when the mapping array
rrs [Wed, 30 May 2007 17:39:45 +0000 (17:39 +0000)]
-  Fix a memory overwrite when the mapping array
   is expanded, size of expansion was not taken int consideration.
-  Fix so vtag hash is 1 bigger so that it modulo's out
   correctly, avoids a panic when restart with right modulo happens.
-  do not dereference stcb when control->do_not_ref_stcb is set
-  Fix up packet logging to not often use a lock and also to
   add to options.
-  Fix some logging option duplication in the sctputil.h

17 years agoDon't rely on pcm_channel index numbering while doing sequential channel
ariff [Wed, 30 May 2007 16:15:01 +0000 (16:15 +0000)]
Don't rely on pcm_channel index numbering while doing sequential channel
allocation.

17 years agoMFi386: revision 1.653.
nyan [Wed, 30 May 2007 14:36:20 +0000 (14:36 +0000)]
MFi386: revision 1.653.

17 years agoMFi386: PDCM, remove pointless message
des [Wed, 30 May 2007 14:23:26 +0000 (14:23 +0000)]
MFi386: PDCM, remove pointless message

MFC after: 3 days

17 years agoRemove unused !AUDIT audit_proc_*() prototypes: unlike in Mac OS X, we
rwatson [Wed, 30 May 2007 11:41:28 +0000 (11:41 +0000)]
Remove unused !AUDIT audit_proc_*() prototypes: unlike in Mac OS X, we
don't define or use these functions if AUDIT isn't configured.

Obtained from: TrustedBSD Project

17 years agoSynchronize white space to congruent user-space code in OpenBSM.
rwatson [Wed, 30 May 2007 09:48:37 +0000 (09:48 +0000)]
Synchronize white space to congruent user-space code in OpenBSM.

Obtained from: TrustedBSD Project

17 years agoRemove unused ar_subj_comm field from in-kernel audit record; we never
rwatson [Wed, 30 May 2007 09:14:14 +0000 (09:14 +0000)]
Remove unused ar_subj_comm field from in-kernel audit record; we never
export this via BSM, so don't pay space/time cost of maintaining it.

Obtained from: TrustedBSD Project

17 years ago- Sync with ng_bpf.h, rev. 1.7
glebius [Wed, 30 May 2007 09:05:49 +0000 (09:05 +0000)]
- Sync with ng_bpf.h, rev. 1.7
- Remove extra brace.

Submitted by: Anton Yuzhaninov <citrin rambler-co.ru>

17 years agoConsistent white space after .'s in comments.
rwatson [Wed, 30 May 2007 08:39:16 +0000 (08:39 +0000)]
Consistent white space after .'s in comments.

17 years agoRemove the volatile qualifier to apply to fxp_miibus_readreg().
kevlo [Wed, 30 May 2007 03:46:04 +0000 (03:46 +0000)]
Remove the volatile qualifier to apply to fxp_miibus_readreg().

17 years agoBow to incomplete GCC 4. constant propagation optimizations and
kan [Wed, 30 May 2007 03:03:06 +0000 (03:03 +0000)]
Bow to incomplete GCC 4. constant propagation optimizations and
initialize some of the local variables GCC claims are being used
uninitialized.

17 years agoIn ipsec6_output_tunnel() make sure that the SA contents do not change.
bz [Tue, 29 May 2007 22:44:24 +0000 (22:44 +0000)]
In ipsec6_output_tunnel() make sure that the SA contents do not change.

The same would apply to ipsec6_output_trans() but there is a larger patch
around which already corrected that case. Do not interfere with that one.

17 years agofix typo: s,applyed,applied,g
bz [Tue, 29 May 2007 22:34:58 +0000 (22:34 +0000)]
fix typo: s,applyed,applied,g

17 years agoImplement ICMPv6 support in ipsec6_get_ulp().
bz [Tue, 29 May 2007 22:32:12 +0000 (22:32 +0000)]
Implement ICMPv6 support in ipsec6_get_ulp().
This is needed to make security policies work correctly if ICMPv6 type
and/or code are given. See setkey(8) 'upperspec' para. for details.

17 years agoAdd missing
bz [Tue, 29 May 2007 22:18:44 +0000 (22:18 +0000)]
Add missing
break;
so when comparing AF_INET6 addresses, scope and ports we do not run into
the default case and return 'no match' instead of 'match'.

17 years agoRemove /usr/X11R6 from the path here too.
dougb [Tue, 29 May 2007 22:07:57 +0000 (22:07 +0000)]
Remove /usr/X11R6 from the path here too.

Reminded by: Niclas Zeising <niclas.zeising@gmail.com>

17 years agoAdd a placeholder document to be filled out by Nate which will describe
mjacob [Tue, 29 May 2007 20:07:22 +0000 (20:07 +0000)]
Add a placeholder document to be filled out by Nate which will describe
in detail the procedure for getting device quirks into the CAM subsystem.

17 years agoAdd more devices from the hps tree. These appear to have come from
imp [Tue, 29 May 2007 20:05:13 +0000 (20:05 +0000)]
Add more devices from the hps tree.  These appear to have come from
OpenBSD's if_ral.c.

I didn't make the LINKSYS4 -> CISCOLINKSYS name change, nor did I
include the RALINK RT2573 that's supported by the rum(4) driver.  I
didn't merge any code changes either.

17 years agoBack out 1.212 at Nate's request in order to go through the
mjacob [Tue, 29 May 2007 20:04:01 +0000 (20:04 +0000)]
Back out 1.212 at Nate's request in order to go through the
actual formal process he's trying to get established for quirks..

17 years agostyle(9) fixes for sx locks.
attilio [Tue, 29 May 2007 19:46:37 +0000 (19:46 +0000)]
style(9) fixes for sx locks.

Approved by: jeff (mentor)

17 years agoAdd descriptive comment to PDCM entry.
des [Tue, 29 May 2007 19:39:18 +0000 (19:39 +0000)]
Add descriptive comment to PDCM entry.

17 years agoAdd a small fix for lock profiling in sx locks.
attilio [Tue, 29 May 2007 19:34:32 +0000 (19:34 +0000)]
Add a small fix for lock profiling in sx locks.
"0" cannot be a correct value since when the function is entered at least
one shared holder must be present and since we want the last one "1" is
the correct value.
Note that lock_profiling for sx locks is far from being perfect.
Expect further fixes for that.

Approved by: jeff (mentor)

17 years agoRemove a pointless bootverbose message.
des [Tue, 29 May 2007 19:25:50 +0000 (19:25 +0000)]
Remove a pointless bootverbose message.

MFC after: 3 days

17 years agoAdd feature name for features2 bit 15.
des [Tue, 29 May 2007 19:21:53 +0000 (19:21 +0000)]
Add feature name for features2 bit 15.

PR: i386/113133
Submitted by: Pankov Pavel <pankov_p@mail.ru>
MFC after: 3 days

17 years agoFix some problems introduced with the last descriptors tables locking
attilio [Tue, 29 May 2007 18:55:41 +0000 (18:55 +0000)]
Fix some problems introduced with the last descriptors tables locking
patch:
- Do the correct test for ldt allocation
- Drop dt_lock just before to call kmem_free (since it acquires blocking
  locks inside)
- Solve a deadlock with smp_rendezvous() where other CPU will wait
  undefinitively for dt_lock acquisition.
- Add dt_lock in the WITNESS list of spinlocks

While applying these modifies, change the requirement for user_ldt_free()
making that returning without dt_lock held.

Tested by: marcus, tegge
Reviewed by: tegge
Approved by: jeff (mentor)

17 years agoRemove the hardcoded IXP425_UART?_VBASE values in the
jhay [Tue, 29 May 2007 18:10:42 +0000 (18:10 +0000)]
Remove the hardcoded IXP425_UART?_VBASE values in the
uart_ixp425_probe() and uart_cpu_getdev(). Change
uart_cpu_getdev() to use hints to find the console.

Reviewed by: marcel

17 years agoFix the path to 7.0 online manual pages.
bmah [Tue, 29 May 2007 16:33:24 +0000 (16:33 +0000)]
Fix the path to 7.0 online manual pages.

17 years agoNew release note: SA-07:04.file.
bmah [Tue, 29 May 2007 16:30:29 +0000 (16:30 +0000)]
New release note:  SA-07:04.file.

Update release note:  FILE 4.21.

17 years agoAdd a few sentences about updating. Some older text on this subject
bmah [Tue, 29 May 2007 16:14:06 +0000 (16:14 +0000)]
Add a few sentences about updating.  Some older text on this subject
went away around the time of 5.0, when doing binary upgrades from
within sysinstall (at least from 4.X) was deemed hazardous due to the
huge differences between these two major versions.

It's much less risky now, and it's also high time that we mentioned
freebsd-update(8) too.

17 years agoDrain task q items when transitioning to INIT state; this closes a
sam [Tue, 29 May 2007 16:13:59 +0000 (16:13 +0000)]
Drain task q items when transitioning to INIT state; this closes a
race seen on smp laptops when suspending where the rx task can be
entered after the interface is detach'd.

NB: use of taskqueue_drain while holding the softc mutex is problematic

Submitted by: ambrisko
MFC after: 1 month

17 years agoAdd LOADER_FIREWIRE_SUPPORT knob.
simokawa [Tue, 29 May 2007 15:19:55 +0000 (15:19 +0000)]
Add LOADER_FIREWIRE_SUPPORT knob.

Requested by: ru

17 years agoUse the same ARG_MAX for arm than for the other platforms. I don't see any
cognet [Tue, 29 May 2007 15:14:46 +0000 (15:14 +0000)]
Use the same ARG_MAX for arm than for the other platforms. I don't see any
reason to special case it.

Spotted out by: marcel

17 years agoMFp4: add FireWire/dcons support in loader for i386/amd64.
simokawa [Tue, 29 May 2007 14:35:57 +0000 (14:35 +0000)]
MFp4: add FireWire/dcons support in loader for i386/amd64.

It is disabled by default. You need to put
LOADER_FIREWIRE_SUPPORT=yes in /etc/make.conf
and rebuild loader to enable it.
(cd /sys/boot/i386 && make clean && make && make install)

You can find a short introduction of dcons at
http://wiki.freebsd.org/DebugWithDcons

17 years agoAdds gcc attribute to prevent inlining of a function. If
rrs [Tue, 29 May 2007 14:17:47 +0000 (14:17 +0000)]
Adds gcc attribute to prevent inlining of a function. If
it goes inline we may well blow the stack if witness and
such are enabled.

17 years agoSync ether_ioctl() with ioctl(2) and ifnet.if_ioctl
yar [Tue, 29 May 2007 12:40:45 +0000 (12:40 +0000)]
Sync ether_ioctl() with ioctl(2) and ifnet.if_ioctl
as to the type of the command argument: int -> u_long.
These types have different widths in the 64-bit world.

Add a note to UPDATING because the change breaks KBI
on 64-bit platforms.

Discussed on: -net, -current
Reviewed by: bms, ru

17 years agoAdd DDB "show unpcb" command, allowing DDB to print out many pertinent
rwatson [Tue, 29 May 2007 12:36:00 +0000 (12:36 +0000)]
Add DDB "show unpcb" command, allowing DDB to print out many pertinent
details from UNIX domain socket protocol layer state.

17 years ago- Fix spelling errors in comments per Ruslan (.. thanks... )
rrs [Tue, 29 May 2007 11:53:27 +0000 (11:53 +0000)]
- Fix spelling errors in comments per Ruslan (.. thanks... )

17 years agoWhere I previously removed calls to kdb_enter(), now remove include of
rwatson [Tue, 29 May 2007 11:28:28 +0000 (11:28 +0000)]
Where I previously removed calls to kdb_enter(), now remove include of
kdb.h.

Pointed out by: bde

17 years ago- Fixes so we won't try to start a timer when we
rrs [Tue, 29 May 2007 09:29:03 +0000 (09:29 +0000)]
- Fixes so we won't try to start a timer when we
  hold a wq lock for the iterator. Panda uses a
  silly recursive lock they hold through the timer.
- Add poor mans wireshark compile option..
- Allocate and start using SCTP_M_XXX for all SCTP_MALLOC() calls.
- sysctl now will get back the refcnt for viewing by onlookers.

Reviewed by: gnn

17 years agoMerge the relevant part of rev.1.14 of s_cbrt.c (a micro-optimization
bde [Tue, 29 May 2007 07:13:07 +0000 (07:13 +0000)]
Merge the relevant part of rev.1.14 of s_cbrt.c (a micro-optimization
involving moving the check for x == 0).  The savings in cycles are
smaller for cbrtf() than for cbrt(), and positive in all measured cases
with gcc-3.4.4, but still very machine/compiler-dependent.

17 years agoRemove more vestiges of /usr/X11R6, but leave mtree for portmgr.
dougb [Tue, 29 May 2007 06:37:58 +0000 (06:37 +0000)]
Remove more vestiges of /usr/X11R6, but leave mtree for portmgr.

17 years agoRemove X11R6 from the default PATH to join the new world order.
dougb [Tue, 29 May 2007 06:33:10 +0000 (06:33 +0000)]
Remove X11R6 from the default PATH to join the new world order.

While I'm here, make the default PATH match that in the csh profile,
and login.conf.

17 years agoHonor maxsegsz of less than a page size in a DMA tag. Previously it
yongari [Tue, 29 May 2007 06:30:26 +0000 (06:30 +0000)]
Honor maxsegsz of less than a page size in a DMA tag. Previously it
used to return PAGE_SIZE without respect to restrictions of a DMA tag.
This affected all of the busdma load functions that use
_bus_dmamap_loader_buffer() as their back-end.

Reviewed by: scottl

17 years agoNow that a separate /usr/X11R6 directory is no longer in fashion,
dougb [Tue, 29 May 2007 06:22:14 +0000 (06:22 +0000)]
Now that a separate /usr/X11R6 directory is no longer in fashion,
stop looking there for things like rc.d and periodic. This avoids
duplicating effort when /usr/X11R6 is a symlink to /usr/local,
which it is by default now.

It is not anticipated at this time that we will MFC this change, since
we'd like to avoid breaking legacy systems. However, there is a fix for
/etc/rc.subr in the works to avoid running any rc.d scripts twice which
we should be able to MFC.

17 years agobsdtar 2.2.3:
kientzle [Tue, 29 May 2007 05:39:10 +0000 (05:39 +0000)]
bsdtar 2.2.3:
  * Implement --use-compress-program using new libarchive feature.
  * Minor portability improvement by adjusting casts used to
    print out uids, gids, and device numbers.

Thanks to: Joerg Sonnenberger for the --use-compress-program implementation.
MFC after: 15 days

17 years agoFix case of setting OACTIVE erroneously
kmacy [Tue, 29 May 2007 04:18:21 +0000 (04:18 +0000)]
Fix case of setting OACTIVE erroneously

17 years ago"make delete-old" needs more than 64KB worth of arguments. Both
marcel [Tue, 29 May 2007 03:49:00 +0000 (03:49 +0000)]
"make delete-old" needs more than 64KB worth of arguments. Both
arm and powerpc have 64KB as the maximum argument size, so one
cannot run "make delete-old" on arm or powerpc anymore. Stop
special-casing powerpc and give it 256KB of arguments like all
other platforms, but keep arm on 64KB for now. There may be a
purpose to it that doesn't exist for powerpc.

17 years agoFix interrupt setup for the non-MSI-X case
kmacy [Tue, 29 May 2007 03:13:53 +0000 (03:13 +0000)]
Fix interrupt setup for the non-MSI-X case

17 years agoUse the Hitachi 2.5" HD, 120 GB as a USB device.
mjacob [Tue, 29 May 2007 01:23:38 +0000 (01:23 +0000)]
Use the Hitachi 2.5" HD, 120 GB as a USB device.

Submitted by: Joakim Bentholm`
MFC after: 3 days

17 years agolibarchive 2.2.3
kientzle [Tue, 29 May 2007 01:00:21 +0000 (01:00 +0000)]
libarchive 2.2.3
  * "compression_program" support uses an external program
  * Portability: no longer uses "struct stat" as a primary
    data interchange structure internally
  * Part of the above: refactor archive_entry to separate
    out copy_stat() and stat() functions
  * More complete tests for archive_entry
  * Finish archive_entry_clone()
  * Isolate major()/minor()/makedev() in archive_entry; remove
    these from everywhere else.
  * Bug fix: properly handle decompression look-ahead at end-of-data
  * Bug fixes to 'ar' support
  * Fix memory leak in ZIP reader
  * Portability: better timegm() emulation in iso9660 reader
  * New write_disk flags to suppress auto dir creation and not
    overwrite newer files (for future cpio front-end)
  * Simplify trailing-'/' fixup when writing tar and pax
  * Test enhancements:  fix various compiler warnings, improve
    portability, add lots of new tests.
  * Documentation: document new functions, first draft of
    libarchive_internals.3

MFC after: 14 days
Thanks to: Joerg Sonnenberger (compression_program)
Thanks to: Kai Wang (ar)
Thanks to: Colin Percival (many small fixes)
Thanks to: Many others who sent me various patches and problem reports.

17 years agoMake log messages more verbose and simpler to understand for non-experts.
andre [Mon, 28 May 2007 23:27:44 +0000 (23:27 +0000)]
Make log messages more verbose and simpler to understand for non-experts.
Update comments to be more conscious, verbose and fully reflect reality.

17 years agoDisable shared libgcc use for profiled builds. GNU linker pulls symbols
kan [Mon, 28 May 2007 23:02:56 +0000 (23:02 +0000)]
Disable shared libgcc use for profiled builds. GNU linker pulls symbols
out of libgcc_s.so.1 somehow and fails to record DT_NEEED dependency
when linking with profiled libraries despite --ld-as-needed option.

17 years agoWhen building cxgb as a module make include paths relative to the driver's root.
kmacy [Mon, 28 May 2007 22:57:27 +0000 (22:57 +0000)]
When building cxgb as a module make include paths relative to the driver's root.
This will make it possible to build the module out of tree against an older src tree.

MFC after: 3 days

17 years agoRevert 1.197 and instead avoid calling kdb_enter() if the KDB_UNATTENDED
emaste [Mon, 28 May 2007 21:50:54 +0000 (21:50 +0000)]
Revert 1.197 and instead avoid calling kdb_enter() if the KDB_UNATTENDED
option is in use.

17 years agoEliminate some unused definitions that came from NetBSD.
alc [Mon, 28 May 2007 21:04:22 +0000 (21:04 +0000)]
Eliminate some unused definitions that came from NetBSD.

17 years agoSimplify the kernel configuration file return code.
imp [Mon, 28 May 2007 20:41:10 +0000 (20:41 +0000)]
Simplify the kernel configuration file return code.

Reviewed by: wkoszek

17 years agoEliminate explicit kdb_enter in the software watchdog handler (which
emaste [Mon, 28 May 2007 19:51:12 +0000 (19:51 +0000)]
Eliminate explicit kdb_enter in the software watchdog handler (which
produced incorrect behaviour with the KDB_UNATTENDED option) and call
panic in both the KDB and non-KDB cases.  This change is consistent
with rwatson's current kdb/ddb work.

17 years agoWe do not need to get the irq out of ivars in ixp425_setup_intr(). By
jhay [Mon, 28 May 2007 18:54:08 +0000 (18:54 +0000)]
We do not need to get the irq out of ivars in ixp425_setup_intr(). By
this time they have already been set. In fact trying to set it here too
breaks irqs for pci devices.

17 years agoOptimize a bit more, both the Avila and Pronghorn Metro boards work with
jhay [Mon, 28 May 2007 18:45:16 +0000 (18:45 +0000)]
Optimize a bit more, both the Avila and Pronghorn Metro boards work with
GPIO_TYPE_EDG_RISING.

Reviewed by: sam

17 years agoDon't save SYSCTL_ADD_*() results in a local variable just to throw them
rwatson [Mon, 28 May 2007 18:20:15 +0000 (18:20 +0000)]
Don't save SYSCTL_ADD_*() results in a local variable just to throw them
away; preserve the ones that are needed for further calls in the init
function and ignore the rest entirely.

Found with: Coverity Prevent(tm)
CID: 563

17 years agoIn kern_kevent(), unconditionally fdrop() fp once fget() has succeeded,
rwatson [Mon, 28 May 2007 17:15:05 +0000 (17:15 +0000)]
In kern_kevent(), unconditionally fdrop() fp once fget() has succeeded,
as we never have an opportunity to set it to NULL.

Found with: Coverity Prevent(tm)
CID: 2161

17 years ago- Enable soft pcm volume flag early to ensure it not being clobbered
ariff [Mon, 28 May 2007 16:22:07 +0000 (16:22 +0000)]
- Enable soft pcm volume flag early to ensure it not being clobbered
  by the subsequent mix_setdevs() and friends.
- Minor style(9) declaration arrangement nit.

Requested by: joeld
Submitted by: pluknet <pluknet@gmail.com>

17 years agoFix misspelling: s/maintanence/maintenance/
bmah [Mon, 28 May 2007 16:00:21 +0000 (16:00 +0000)]
Fix misspelling:  s/maintanence/maintenance/

Submitted by: brueffer

17 years agoAdd AK4396 to the list of supported codecs.
joel [Mon, 28 May 2007 16:00:08 +0000 (16:00 +0000)]
Add AK4396 to the list of supported codecs.

17 years agoDocument support for the following sound cards:
joel [Mon, 28 May 2007 15:57:22 +0000 (15:57 +0000)]
Document support for the following sound cards:

- AudioTrak Prodigy HD2
- AudioTrak Prodigy 7.1 XT
- ESI Juli@

Reminded by: brueffer

17 years agoEnable fwip and dcons in GENERIC. They seem fairly stable.
simokawa [Mon, 28 May 2007 14:38:43 +0000 (14:38 +0000)]
Enable fwip and dcons in GENERIC. They seem fairly stable.

Note on dcons:
To enable dcons in kernel, put the following lines in /boot/loader.conf.
You may also want to enable dcons in /etc/ttys.

boot_multicons="YES"
#Force dcons to be the high-level console if a firewire bus presents.
#hw.firewire.dcons_crom.force_console=1

FireWire/dcons support in loader will come shortly.
(i386/amd64 only)

17 years agoFix broken "rec" and "igain" introduced by previous commit. Convert
ariff [Mon, 28 May 2007 14:09:06 +0000 (14:09 +0000)]
Fix broken "rec" and "igain" introduced by previous commit. Convert
reg to a full blown int since there's not much gain compacting it,
and we do need its signess.

17 years agoCorrect spelling errors in comments.
rwatson [Mon, 28 May 2007 11:36:43 +0000 (11:36 +0000)]
Correct spelling errors in comments.

17 years agoFix indentation of the syncache_expand() section in tcp_input().
andre [Mon, 28 May 2007 11:35:40 +0000 (11:35 +0000)]
Fix indentation of the syncache_expand() section in tcp_input().

17 years agoPrecede symbol names consistently with tabs rather than spaces.
rwatson [Mon, 28 May 2007 11:33:44 +0000 (11:33 +0000)]
Precede symbol names consistently with tabs rather than spaces.

17 years ago- fixed autclose to not allow setting on 1-2-1 model.
rrs [Mon, 28 May 2007 11:17:24 +0000 (11:17 +0000)]
- fixed autclose to not allow setting on 1-2-1 model.
- bounded cookie-life to 1 second minimum in socket option set.
- Delayed_ack_time becomes delayed_ack per new socket api document.
- Improve port number selection, we now use low/high bounds and
  no chance of a endless loop. Only one call to random per bind
  as well.
- fixes so set_peer_primary pre-screens addresses to be
  valid to this host.
- maxseg did not allow setting on an assoc basis. We needed
  to thus track and use an association value instead of a inp value.
- Fixed ep get of HB status to report back properly.
- use settings flag to tell if assoc level hb is on off not
  the timer.. since the timer may still run if unconf address
  are present.
- check for crazy ENABLE/DISABLE conditions.
- set and get of pmtud (fixed path mtu) not always taking into account ovh.
- Getting PMTU info on stcb only needs to return PMTUD_ENABLED if
  any net is doing PMTU discovery.
- Panic or warning fixed to not do so when a valid ip frag is
  taking place.
- sndrcvinfo appearing in both inp and stcb was full size, instead
  of the non-pad version. This saves about 92 bytes from each struct
  by carefully converting to use the smaller version.
- one-2-one model get(maxseg) would always get ep value, never the
  tcb's value.
- The delayed ack time could be under a tick, this fixes so
  it bounds it to at least 1 tick for platforms whos tick
  is more than a ms.
- Fragment interleave level set to wrong default value.
- Fragment interleave could not set level 0.
- Defered stream reset was broken due to a guard check and ntohl issue.
- Found two lock order reversals and fixed.
- Tighten up address checking, if the user gives an address the sa_len
  had better be set properly.
- Get asoc by assoc-id would return a locked tcb when it was asked
  not to if the tcb was in the restart hash.
- sysctl to dig down and get more association details

Reviewed by: gnn

17 years agoRefactor and rewrite in parts the SYN handling code on listen sockets
andre [Mon, 28 May 2007 11:03:53 +0000 (11:03 +0000)]
Refactor and rewrite in parts the SYN handling code on listen sockets
in tcp_input():

 o tighten the checks on allowed TCP flags to be RFC793 and
   tcp-secure conform
 o log check failures to syslog at LOG_DEBUG level
 o rearrange the code flow to be easier to follow
 o add KASSERTs to validate assumptions of the code flow

Add sysctl net.inet.tcp.syncache.rst_on_sock_fail defaulting to enable
that controls the behavior on socket creation failure for a otherwise
successful 3-way handshake.  The socket creation can fail due to global
memory shortage, listen queue limits and file descriptor limits.  The
sysctl allows to chose between two options to deal with this.  One is
to send a reset to the other endpoint to notify it about the failure
(default).  The other one is to ignore and treat the failure as a
transient error and have the other endpoint retransmit for another try.

Reviewed by: rwatson (in general)

17 years agoCheck fdopen return value.
kevlo [Mon, 28 May 2007 09:48:25 +0000 (09:48 +0000)]
Check fdopen return value.

Reviewed by: phk

17 years agoAdd a '-k' flag which is similar with its bzip2(1) counterpart,
delphij [Mon, 28 May 2007 08:20:46 +0000 (08:20 +0000)]
Add a '-k' flag which is similar with its bzip2(1) counterpart,
meaning that the user wants the input file to be left intact.

Feature request: Ighighi <ighighi gmail.com>
PR:  bin/103006
MFC after:  1 month

17 years agoFix probably copy&paste from chmod(1) - we change file flags here, not mode.
pjd [Mon, 28 May 2007 04:23:09 +0000 (04:23 +0000)]
Fix probably copy&paste from chmod(1) - we change file flags here, not mode.

17 years agoAdjust va_mask for setattr. FreeBSD doesn't have va_mask, so we initialize it
pjd [Mon, 28 May 2007 02:37:43 +0000 (02:37 +0000)]
Adjust va_mask for setattr. FreeBSD doesn't have va_mask, so we initialize it
based on individual fields beeing set. This doesn't work for setattr replay,
because va_type is set there, so we add AT_TYPE flag to va_mask, which won't
be accepted by zfs_setattr().

Reported by: kris

17 years agoWe should better ignore a break on gdb port if gdb is not enabled.
simokawa [Mon, 28 May 2007 02:20:40 +0000 (02:20 +0000)]
We should better ignore a break on gdb port if gdb is not enabled.

17 years ago- Remove unnecessary vnode internal locking - v_vflag is protect by vnode's
pjd [Mon, 28 May 2007 00:28:15 +0000 (00:28 +0000)]
- Remove unnecessary vnode internal locking - v_vflag is protect by vnode's
  lock (not vnode's interlock).
- Simplify code a bit.

17 years agoBecause we allocate componentname structures on stack, bzero() them before
pjd [Mon, 28 May 2007 00:26:20 +0000 (00:26 +0000)]
Because we allocate componentname structures on stack, bzero() them before
use just in case.

17 years agoFix compiles when user chooses to disable both ObjC and C++ support in
kan [Mon, 28 May 2007 00:25:07 +0000 (00:25 +0000)]
Fix compiles when user chooses to disable both ObjC and C++ support in
GCC.

Reported by: bz

17 years agoTuning for small packet handling
kmacy [Sun, 27 May 2007 22:07:47 +0000 (22:07 +0000)]
Tuning for small packet handling
- Double the number of descriptors that a single call to send can use
- Quadruple the number of descriptors that can be reclaimed per pass
- only run reclaim twice per second
- increase coalesce timer from 3.5us to 5us

fix printf warning on 64-bit platforms