]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
19 years agoDont limit all transfers to DEV_BSIZE (stale dbug code)
Søren Schmidt [Thu, 4 Aug 2005 18:39:36 +0000 (18:39 +0000)]
Dont limit all transfers to DEV_BSIZE (stale dbug code)
Hide the loaded/unloaded message behind bootverbose

19 years agodon't do mutex locking around kthread_create() call.
Hajimu UMEMOTO [Thu, 4 Aug 2005 18:34:26 +0000 (18:34 +0000)]
don't do mutex locking around kthread_create() call.

Reported by: Maxim Maximov <mcsi__at__mcsi.pp.ru>

19 years agoInitialize the if_addr mutex in if_alloc() rather than waiting until
John Baldwin [Thu, 4 Aug 2005 14:39:47 +0000 (14:39 +0000)]
Initialize the if_addr mutex in if_alloc() rather than waiting until
if_attach().  This allows ethernet drivers to use it in their routines
to program their MAC filters before ether_ifattach() is called (de(4) is
one such driver).  Also, the if_addr mutex is destroyed in if_free()
rather than if_detach(), so there was another potential bug in that a
driver that failed during attach and called if_free() without having
called ether_ifattach() would have tried to destroy an uninitialized mutex.

Reported by: Holm Tiffe holm at freibergnet dot de
Discussed with: rwatson

19 years agoThe check for invalid characters introduced in 1.9 incorrectly
Ollivier Robert [Thu, 4 Aug 2005 12:44:36 +0000 (12:44 +0000)]
The check for invalid characters introduced in 1.9 incorrectly
flags NIS entries as invalid.

MFC6 candidate

Submitted by: Bob Willcox <bob@immure.com>
PR: bin/82325 (but I used the patch from Bob).

19 years agoMake kernel build suceed when with "options CPU_DISABLE_SSE".
Anton Berezin [Thu, 4 Aug 2005 12:39:43 +0000 (12:39 +0000)]
Make kernel build suceed when with "options CPU_DISABLE_SSE".

PR: 84010
Submitted by: Sergey Gluschenko <deen@freebsd.org.ua>
MFC after: 1 week

19 years agoDefine LIBMEMSTAT so that vm_page.h won't perform a nested include of
Robert Watson [Thu, 4 Aug 2005 10:06:39 +0000 (10:06 +0000)]
Define LIBMEMSTAT so that vm_page.h won't perform a nested include of
opt_vmpage.h.

Remove definition of _KERNEL, it is no longer required in order to
include uma_int.h, as the sensitive parts of uma_int.h (a number of
inlines depending on kernel-only constants) are now protected by
_KERNEL.

19 years agoBug fix: a numeric flag specification in the substitute command would
Diomidis Spinellis [Thu, 4 Aug 2005 10:05:12 +0000 (10:05 +0000)]
Bug fix: a numeric flag specification in the substitute command would
cause the next substitute flag to be ignored.
While working at it, detect and report overflows.

Reported by: Jingsong Liu
MFC after: 1 week

19 years agoDon't perform a nested include of opt_vmpage.h if LIBMEMSTAT is defined,
Robert Watson [Thu, 4 Aug 2005 10:05:11 +0000 (10:05 +0000)]
Don't perform a nested include of opt_vmpage.h if LIBMEMSTAT is defined,
as opt_vmpage.h will not be available to user space library builds.  A
similar existing check is present for KLD_MODULE for similar reasons.

MFC after: 3 days

19 years agoWrap inlines in uma_int.h in #ifdef _KERNEL so that uma_int.h can be
Robert Watson [Thu, 4 Aug 2005 10:03:53 +0000 (10:03 +0000)]
Wrap inlines in uma_int.h in #ifdef _KERNEL so that uma_int.h can be
used from memstat_uma.c for the purposes of kvm access without lots
of additional unsafe includes.

MFC after: 3 days

19 years agoRemove gbde_swap_enable option which doesn't work and doesn't really have to
Pawel Jakub Dawidek [Thu, 4 Aug 2005 08:19:13 +0000 (08:19 +0000)]
Remove gbde_swap_enable option which doesn't work and doesn't really have to
work, as one still needs to put <device>.bde into /etc/fstab.

19 years agoCorrectly print system-wide PMC counts for short-lived command lines.
Joseph Koshy [Thu, 4 Aug 2005 03:12:55 +0000 (03:12 +0000)]
Correctly print system-wide PMC counts for short-lived command lines.

Reported by: cracauer
MFC after: 3 days

19 years agoFix two bugs that interacted to cause page faults in softclock()
Ian Dowse [Thu, 4 Aug 2005 02:21:37 +0000 (02:21 +0000)]
Fix two bugs that interacted to cause page faults in softclock()
when using mice containing a tilt movement: there was a missing
usb_callout_init() for the UMS_SPUR_BUT_UP quirk code, and UMS_T
was defined to the same flag value as UMS_SPUR_BUT_UP.

Reported by: flz
MFC after: 3 days

19 years agoFix commen typos.
Stefan Farfeleder [Wed, 3 Aug 2005 20:57:33 +0000 (20:57 +0000)]
Fix commen typos.

19 years ago- Remove duplicate setting from LIB32MAKE definition.
Ruslan Ermilov [Wed, 3 Aug 2005 20:27:06 +0000 (20:27 +0000)]
- Remove duplicate setting from LIB32MAKE definition.
- Install lib32 bits using the real install(1) program.

19 years agoAdd missing "not" in rev. 1.34.
Ruslan Ermilov [Wed, 3 Aug 2005 20:11:39 +0000 (20:11 +0000)]
Add missing "not" in rev. 1.34.

Submitted by: Jeremie Le Hen

19 years agoMake delete-old* and friends useable for cross-builds.
Ruslan Ermilov [Wed, 3 Aug 2005 20:09:59 +0000 (20:09 +0000)]
Make delete-old* and friends useable for cross-builds.

19 years agoIntroduce in_multi_mtx, which will protect IPv4-layer multicast address
Robert Watson [Wed, 3 Aug 2005 19:29:47 +0000 (19:29 +0000)]
Introduce in_multi_mtx, which will protect IPv4-layer multicast address
lists, as well as accessor macros.  For now, this is a recursive mutex
due code sequences where IPv4 multicast calls into IGMP calls into
ip_output(), which then tests for a multicast forwarding case.

For support macros in in_var.h to check multicast address lists, assert
that in_multi_mtx is held.

Acquire in_multi_mtx around iteration over the IPv4 multicast address
lists, such as in ip_input() and ip_output().

Acquire in_multi_mtx when manipulating the IPv4 layer multicast addresses,
as well as over the manipulation of ifnet multicast address lists in order
to keep the two layers in sync.

Lock down accesses to IPv4 multicast addresses in IGMP, or assert the
lock when performing IGMP join/leave events.

Eliminate spl's associated with IPv4 multicast addresses, portions of
IGMP that weren't previously expunged by IGMP locking.

Add in_multi_mtx, igmp_mtx, and if_addr_mtx lock order to hard-coded
lock order in WITNESS, in that order.

Problem reported by: Ed Maste <emaste at phaedrus dot sandvine dot ca>
MFC after: 10 days

19 years agoAdd a few more system call timing cases:
Robert Watson [Wed, 3 Aug 2005 17:33:46 +0000 (17:33 +0000)]
Add a few more system call timing cases:

Rename current socket creation test to socket_stream.
Add new socket_dgram test.
Add new socketpair_stream test.
Add new socketpair_dgram test.

19 years agoMinor style nits.
Warner Losh [Wed, 3 Aug 2005 15:07:14 +0000 (15:07 +0000)]
Minor style nits.

19 years agoCorrect attribution in clause three to address the correct copyright
Justin T. Gibbs [Wed, 3 Aug 2005 14:08:41 +0000 (14:08 +0000)]
Correct attribution in clause three to address the correct copyright
holders.  The license that was approved for my changes to this driver
originally came from LSI, but the changes to the driver core are not
owned by LSI.

MFC: 1 day

19 years agoAdd NO_PCVT
Poul-Henning Kamp [Wed, 3 Aug 2005 09:27:26 +0000 (09:27 +0000)]
Add NO_PCVT

19 years agoAdd NO_CALENDAR
Poul-Henning Kamp [Wed, 3 Aug 2005 09:27:01 +0000 (09:27 +0000)]
Add NO_CALENDAR

19 years agoDon't install includes if NO_TOOLCHAIN
Poul-Henning Kamp [Wed, 3 Aug 2005 09:26:01 +0000 (09:26 +0000)]
Don't install includes if NO_TOOLCHAIN

19 years agoAdd NO_SYSCONS.
Poul-Henning Kamp [Wed, 3 Aug 2005 09:24:39 +0000 (09:24 +0000)]
Add NO_SYSCONS.
Respect NO_MAN and NO_EXAMPLES
Make NO_GROFF prevet me installation.

19 years agoDon't install groff if NO_GROFF
Poul-Henning Kamp [Wed, 3 Aug 2005 09:22:42 +0000 (09:22 +0000)]
Don't install groff if NO_GROFF

19 years agoDon't install examples of NO_EXAMPLES
Poul-Henning Kamp [Wed, 3 Aug 2005 09:22:03 +0000 (09:22 +0000)]
Don't install examples of NO_EXAMPLES

19 years agoDon't install includes if NO_TOOLCHAIN
Poul-Henning Kamp [Wed, 3 Aug 2005 09:18:59 +0000 (09:18 +0000)]
Don't install includes if NO_TOOLCHAIN

19 years ago - Unlock before we call mac_destroy_vnode to prevent a lock order reversal.
Jeff Roberson [Wed, 3 Aug 2005 05:36:50 +0000 (05:36 +0000)]
 - Unlock before we call mac_destroy_vnode to prevent a lock order reversal.

Found by: trhodes

19 years ago - Use lockmgr_printinfo rather than rolling our own. This introduces a
Jeff Roberson [Wed, 3 Aug 2005 05:02:08 +0000 (05:02 +0000)]
 - Use lockmgr_printinfo rather than rolling our own.  This introduces a
   slight problem by using printf instead of db_printf however
   'show lockedvnods' does the same so I believe it is ok for now.

19 years ago - Fix a problem that slipped through review; the stack member of the lockmgr
Jeff Roberson [Wed, 3 Aug 2005 04:59:07 +0000 (04:59 +0000)]
 - Fix a problem that slipped through review; the stack member of the lockmgr
   structure should have the lk_ prefix.
 - Add stack_print(lkp->lk_stack) to the information printed with
   lockmgr_printinfo().

19 years ago - Replace the series of DEBUG_LOCKS hacks which tried to save the vn_lock
Jeff Roberson [Wed, 3 Aug 2005 04:48:22 +0000 (04:48 +0000)]
 - Replace the series of DEBUG_LOCKS hacks which tried to save the vn_lock
   caller by saving the stack of the last locker/unlocker in lockmgr.  We
   also put the stack in KTR at the moment.

Contributed by: Antoine Brodin <antoine.brodin@laposte.net>

19 years ago- Add support for saving stack traces and displaying them via printf(9)
Jeff Roberson [Wed, 3 Aug 2005 04:33:48 +0000 (04:33 +0000)]
- Add support for saving stack traces and displaying them via printf(9)
  and KTR.

Contributed by: Antoine Brodin <antoine.brodin@laposte.net>
Concept code from: Neal Fachan <neal@isilon.com>

19 years ago - Add support for saving stack traces and displaying them via printf(9)
Jeff Roberson [Wed, 3 Aug 2005 04:27:40 +0000 (04:27 +0000)]
 - Add support for saving stack traces and displaying them via printf(9)
   and KTR.

Contributed by: Antoine Brodin <antoine.brodin@laposte.net>
Concept code from: Neal Fachan <neal@isilon.com>

19 years agosimplify use with non-ath devices; search the wlan sysctl nodes to
Sam Leffler [Wed, 3 Aug 2005 04:25:04 +0000 (04:25 +0000)]
simplify use with non-ath devices; search the wlan sysctl nodes to
locate the right one for the specified interface

MFC after: 3 days

19 years ago - Improve the definition of INKERNEL() to include the DMAP area and the
Jeff Roberson [Wed, 3 Aug 2005 04:21:51 +0000 (04:21 +0000)]
 - Improve the definition of INKERNEL() to include the DMAP area and the
   proper start of the kernel area.

Discussed with: peter

19 years agoDocument 'kernel-toolchain'.
David E. O'Brien [Wed, 3 Aug 2005 03:26:57 +0000 (03:26 +0000)]
Document 'kernel-toolchain'.

Submitted by: Jeremie Le Hen <jeremie@le-hen.org>

19 years agoAdd usleep to the map files.
Daniel Eischen [Wed, 3 Aug 2005 01:54:52 +0000 (01:54 +0000)]
Add usleep to the map files.

Noticed by: davidxu

19 years agoIn adjustrunqueue(), add code to handle thread migrating case for
David Xu [Wed, 3 Aug 2005 01:23:45 +0000 (01:23 +0000)]
In adjustrunqueue(), add code to handle thread migrating case for
ULE scheduler. In original code, local run queue of threaded ksegrp
is corrupted if adjustrunqueue() is called while thread is migrating.

19 years agoAdd a cancellation point for usleep().
Daniel Eischen [Wed, 3 Aug 2005 00:48:13 +0000 (00:48 +0000)]
Add a cancellation point for usleep().

19 years agoBack out change accidentally committed as Makefile:1.21 -- a local
Robert Watson [Wed, 3 Aug 2005 00:47:33 +0000 (00:47 +0000)]
Back out change accidentally committed as Makefile:1.21 -- a local
tweak to let the 6.x/7.x kernel series build on a 5.x userland.

Pointed out by: njl

19 years agoAdd a cancellation point for usleep().
Daniel Eischen [Wed, 3 Aug 2005 00:47:31 +0000 (00:47 +0000)]
Add a cancellation point for usleep().

While here, fix sleep() so that it is also a cancellation point (a
missing weak reference prevented that).

19 years agoMake syslog() use the internal (non-cancellation point) _usleep().
Daniel Eischen [Wed, 3 Aug 2005 00:45:58 +0000 (00:45 +0000)]
Make syslog() use the internal (non-cancellation point) _usleep().
Prior to this it was calling the cancellable usleep() while holding
a lock.

19 years agoMake usleep() overridable by the thread libraries so they can provide
Daniel Eischen [Wed, 3 Aug 2005 00:44:25 +0000 (00:44 +0000)]
Make usleep() overridable by the thread libraries so they can provide
cancellation points.

Noticed by: phk

19 years agoAdd namespace #defines for usleep.
Daniel Eischen [Wed, 3 Aug 2005 00:43:14 +0000 (00:43 +0000)]
Add namespace #defines for usleep.

19 years agoModify device drivers supporting multicast addresses to lock if_addr_mtx
Robert Watson [Wed, 3 Aug 2005 00:18:35 +0000 (00:18 +0000)]
Modify device drivers supporting multicast addresses to lock if_addr_mtx
over iteration of their multicast address lists when synchronizing the
hardware address filter with the network stack-maintained list.

Problem reported by: Ed Maste (emaste at phaedrus dot sandvine dot ca>
MFC after: 1 week

19 years agoModify network protocol consumers of the ifnet multicast address lists
Robert Watson [Tue, 2 Aug 2005 23:51:22 +0000 (23:51 +0000)]
Modify network protocol consumers of the ifnet multicast address lists
to lock if_addr_mtx.

Problem reported by: Ed Maste <emaste at phaedrus dot sandvine dot ca>
MFC after: 1 week

19 years agoProtect link layer network interface multicast address list manipulation
Robert Watson [Tue, 2 Aug 2005 23:23:26 +0000 (23:23 +0000)]
Protect link layer network interface multicast address list manipulation
using ifp->if_addr_mtx:

- Initialize if_addr_mtx when ifnet is initialized.

- Destroy if_addr_mtx when ifnet is torn down.

- Rename ifmaof_ifpforaddr() to if_findmulti(); assert if_addr_mtx.
  Staticize.

- Extract ifmultiaddr allocation and initialization into if_allocmulti();
  accept a 'mflags' argument to indicate whether or not sleeping is
  permitted.  This centralizes error handling and address duplication.

- Extract ifmultiaddr tear-down and deallocation in if_freemulti().

- Re-structure if_addmulti() to hold if_addr_mtx around manipulation of
  the ifnet multicast address list and reference count manipulation.
  Make use of non-sleeping allocations.  Annotate the fact that we only
  generate routing socket events for explicit address addition, not
  implicit link layer address addition.

- Re-structure if_delmulti() to hold if_addr_mtx around manipulation of
  the ifnet multicast address list and reference count manipulation.
  Annotate the lack of a routing socket event for implicit link layer
  address removal.

- De-spl all and sundry.

Problem reported by: Ed Maste <emaste at phaedrus dot sandvine dot ca>
MFC after: 1 week

19 years agoCross-reference pw(8) into chpass(1), passwd(1), and vipw(8).
Sean Chittenden [Tue, 2 Aug 2005 21:38:03 +0000 (21:38 +0000)]
Cross-reference pw(8) into chpass(1), passwd(1), and vipw(8).

19 years agoInclude a SYSUNINIT() to destroy the mutex in MTX_SYSINIT. This makes
John Baldwin [Tue, 2 Aug 2005 20:50:45 +0000 (20:50 +0000)]
Include a SYSUNINIT() to destroy the mutex in MTX_SYSINIT.  This makes
MTX_SYSINIT mutexes play well with modules that can be unloaded.

Reported by: sam
MFC after: 3 days

19 years agoAdd missing check for the NO_CRYPT build option.
Ruslan Ermilov [Tue, 2 Aug 2005 20:12:30 +0000 (20:12 +0000)]
Add missing check for the NO_CRYPT build option.

Reported by: Alexander Polakov

19 years agopolish various bits.
Poul-Henning Kamp [Tue, 2 Aug 2005 20:09:45 +0000 (20:09 +0000)]
polish various bits.

19 years agoIgnore CVS directories
Poul-Henning Kamp [Tue, 2 Aug 2005 20:09:24 +0000 (20:09 +0000)]
Ignore CVS directories

19 years agoFix up the comment.
Ruslan Ermilov [Tue, 2 Aug 2005 20:06:48 +0000 (20:06 +0000)]
Fix up the comment.

19 years agoFixed parsing of unsigned integers.
Ruslan Ermilov [Tue, 2 Aug 2005 20:05:37 +0000 (20:05 +0000)]
Fixed parsing of unsigned integers.

19 years agoLong overdue, keep up with mbuf.h,v 1.148.
Ruslan Ermilov [Tue, 2 Aug 2005 20:03:23 +0000 (20:03 +0000)]
Long overdue, keep up with mbuf.h,v 1.148.

19 years agoWireless suport fixups: add a bunch of stuff that's been in the
Sam Leffler [Tue, 2 Aug 2005 19:13:13 +0000 (19:13 +0000)]
Wireless suport fixups: add a bunch of stuff that's been in the
code but not documented (e.g. wme, mac acl) and correct some
information, etc.

Reviewed by: brueffer

19 years agoeliminate the regex used to match ethernet and 802.11 devices;
Sam Leffler [Tue, 2 Aug 2005 18:28:31 +0000 (18:28 +0000)]
eliminate the regex used to match ethernet and 802.11 devices;
instead use the interface's media-type

Reviewed by: imp
MFC after: 1 week

19 years agoWhen allocating link layer ifnet address list entries in
Robert Watson [Tue, 2 Aug 2005 17:52:52 +0000 (17:52 +0000)]
When allocating link layer ifnet address list entries in
ifp->if_resolvemulti(), do so with M_NOWAIT rather than M_WAITOK, so
that a mutex can be held over the call.  In the FDDI code, add a
missing M_ZERO.  Consumers are already aware that if_resolvemulti()
can fail.

MFC after: 1 week

19 years agoAdd if_addr_mtx to struct ifnet, a mutex to protect ifnet-related address
Robert Watson [Tue, 2 Aug 2005 17:43:35 +0000 (17:43 +0000)]
Add if_addr_mtx to struct ifnet, a mutex to protect ifnet-related address
lists.  Add accessor macros.

This changes the size of struct ifnet, but ideally, all ifnet consumers
are now using if_alloc() to allocate these structures rather than
embedding them into device driver softc's, so this won't modify the
network device driver ABI.

MFC after: 1 week

19 years agoDo not lock an to check gone flag. Only need to hold the lock to set
Maksim Yevmenkin [Tue, 2 Aug 2005 16:03:51 +0000 (16:03 +0000)]
Do not lock an to check gone flag. Only need to hold the lock to set
the gone flag.

Reviewed by: imp
MFC after: 1 day

19 years agoMention that the softc is zeroed after allocation.
Christian Brueffer [Tue, 2 Aug 2005 13:54:52 +0000 (13:54 +0000)]
Mention that the softc is zeroed after allocation.

Noticed by: sam
MFC after: 3 days

19 years ago- add missing ether_poll_deregister
Christian Brueffer [Tue, 2 Aug 2005 09:27:42 +0000 (09:27 +0000)]
- add missing ether_poll_deregister
- add per-device polling
- move the setting of if_capenable further down so that the
  VLAN caps are honored as well

Approved by: ru
MFC after: 2 weeks

19 years agoAdd missing ether_poll_deregister(). This is still not enough to
Ruslan Ermilov [Tue, 2 Aug 2005 08:44:45 +0000 (08:44 +0000)]
Add missing ether_poll_deregister().  This is still not enough to
kldunload/kldload without a panic.  The same (but worse) problem
is also present in ixgb(4).

19 years agoGenerate default fake "device" and "inode" numbers for entries
Tim Kientzle [Tue, 2 Aug 2005 03:17:57 +0000 (03:17 +0000)]
Generate default fake "device" and "inode" numbers for entries
extracted from tar archives.  Otherwise, converting tar archives to
cpio format (with "bsdtar -cf out.cpio @in.tar") convert every entry
into a hard link to a single file.  This simple logic breaks hard
links, but that's better than the alternative.

MFC after: 7 days

19 years agoWhen copying time values from the main entry header to be used in the
Tim Kientzle [Tue, 2 Aug 2005 03:13:42 +0000 (03:13 +0000)]
When copying time values from the main entry header to be used in the
header of the pax extension entry, clip them to ustar limits.  In particular,
this prevents an internal panic for very old files.

Thanks to: Chris Spiegel
MFC after: 7 days

19 years agoCorrect a few minor mis-statements (libarchive does support reading
Tim Kientzle [Tue, 2 Aug 2005 03:10:52 +0000 (03:10 +0000)]
Correct a few minor mis-statements (libarchive does support reading
GNU tar sparse files, people have extended cpio) and clarify an
important detail about pax format (that ustar-compliant archivers
can mostly read pax archives correctly).

MFC after: 7 days

19 years agoThe bsdtar_warnc() reporting function requires the program name to be
Tim Kientzle [Tue, 2 Aug 2005 03:02:55 +0000 (03:02 +0000)]
The bsdtar_warnc() reporting function requires the program name to be
set up before it is called, so move the progname initialization before
the first possible call to bsdtar_warnc().

Thanks to: Stanislav Sedov
PR: bin/83366
MFC after: 7 days

19 years ago - Document network_interfaces=auto (the default!).
Brooks Davis [Tue, 2 Aug 2005 02:24:47 +0000 (02:24 +0000)]
 - Document network_interfaces=auto (the default!).
 - Remove documentation of pccard_ifconfig as it is now gone.
 - Document pccard_ifconfig's replacement ifconfig_DEFAULT.

19 years agoTeach vmstat's domemstat_zone() to use memstat_kvm_uma() when the kvm
Robert Watson [Mon, 1 Aug 2005 21:33:17 +0000 (21:33 +0000)]
Teach vmstat's domemstat_zone() to use memstat_kvm_uma() when the kvm
descriptor is non-NULL, restoring vmstat -z support for core dumps and
kmem access.  These were broken with the introduction of UMA.

19 years agoMake getsockopt(..., SOL_SOCKET, SO_ACCEPTCONN, ...) work per IEEE Std
Kelly Yancey [Mon, 1 Aug 2005 21:15:09 +0000 (21:15 +0000)]
Make getsockopt(..., SOL_SOCKET, SO_ACCEPTCONN, ...) work per IEEE Std
1003.1 (POSIX).

19 years agoAdd memstat_kvm_uma(), an implementation of a libmemstat(3) query routine
Robert Watson [Mon, 1 Aug 2005 19:07:39 +0000 (19:07 +0000)]
Add memstat_kvm_uma(), an implementation of a libmemstat(3) query routine
that knows how to extract UMA(9) allocator statistics from a core dump or
live memory image using kvm(3).  The caller is expected to provide the
necessary kvm_t handle, which is then used by libmemstat(3).

With these changes, it is trivially straight forward to re-introduce
vmstat -z support on core dumps, which was lost when UMA was introduced.

In the short term, this requires including vm/ include files that are not
intended for extra-kernel use, requiring in turn some ugliness.

19 years agoFix a buffer aging problem in new r300 code that could lead to hangs with some
Eric Anholt [Mon, 1 Aug 2005 17:50:19 +0000 (17:50 +0000)]
Fix a buffer aging problem in new r300 code that could lead to hangs with some
apps.

Obtained from: DRM CVS

19 years agoPropagate error code of kern_execve() to the caller properly.
Maxim Sobolev [Mon, 1 Aug 2005 17:35:48 +0000 (17:35 +0000)]
Propagate error code of kern_execve() to the caller properly.

PR: 81670
Submitted by: Andrew Bliznak <andriko.b@gmail.com>
Pointy hat to: sobomax

19 years agoacm and ack are specified per-ac
Sam Leffler [Mon, 1 Aug 2005 16:34:41 +0000 (16:34 +0000)]
acm and ack are specified per-ac

MFC after: 3 days

19 years agoFixing compilation error by:
Tai-hwa Liang [Mon, 1 Aug 2005 13:51:52 +0000 (13:51 +0000)]
Fixing compilation error by:

- Conforming to the latest ether_ifattach() change;
- Moving PCCARD_API_LEVEL to the right place.

Reported and Tested by: Vladimir Grebenschikov <vova at fbsd dot ru>
MFC after: 3 days

19 years agoCorrect two libmemstat(3) bugs:
Robert Watson [Mon, 1 Aug 2005 13:18:21 +0000 (13:18 +0000)]
Correct two libmemstat(3) bugs:

- Move memory_type_list flushing logic from memstat_mtl_free() to
  _memstat_mtl_empty(), a libmemstat-internal function that can
  be called from other parts of the library.  Invoke
  _memstat_mtl_empty() from memstat_mtl_free(), which also frees
  the containing list structure.

  Invoke _memstat_mtl_empty() instead of memstat_mtl_free() in
  various error cases in memstat_malloc.c and memstat_uma.c, which
  previously resulted in the list being freed prematurely.

- Reverse the order of updating the mt_kegfree and mt_free fields
  of the memory_type in memstat_uma.c, otherwise keg free items
  won't be counted properly for non-secondary zones.

MFC after: 3 days

19 years agoThe Berkeley pascal implementation for the VAX 11/780 has only
Poul-Henning Kamp [Mon, 1 Aug 2005 12:14:53 +0000 (12:14 +0000)]
The Berkeley pascal implementation for the VAX 11/780 has only
historical relevance these days, stow it in the attic instead of
on millions of FreeBSD computers.

19 years agoRemove the \*(DY at the end of the document, it makes the installed
Poul-Henning Kamp [Mon, 1 Aug 2005 12:10:14 +0000 (12:10 +0000)]
Remove the \*(DY at the end of the document, it makes the installed
file differ for each install.

19 years agorecover the line which was wrongly disappeared during scope cleanup.
Hajimu UMEMOTO [Mon, 1 Aug 2005 12:08:49 +0000 (12:08 +0000)]
recover the line which was wrongly disappeared during scope cleanup.
tcpdrop(8) should work for IPv6, again.

19 years agotypo
Poul-Henning Kamp [Mon, 1 Aug 2005 08:24:03 +0000 (08:24 +0000)]
typo

19 years agoAdd support for IPv6 over GRE [1]. PR kern/80340 includes the
Bjoern A. Zeeb [Mon, 1 Aug 2005 08:14:21 +0000 (08:14 +0000)]
Add support for IPv6 over GRE [1]. PR kern/80340 includes the
  FreeBSD specific ip_newid() changes NetBSD does not have.
Correct handling of non AF_INET packets passed to bpf [2].

PR: kern/80340[1], NetBSD PRs 29150[1], 30844[2]
Obtained from: NetBSD ip_gre.c rev. 1.34,1.35, if_gre.c rev. 1.56
Submitted by: Gert Doering <gert at greenie.muc.de>[2]
MFC after: 4 days

19 years agoMake the eisa probe messages just like all the others in the system.
Warner Losh [Mon, 1 Aug 2005 07:09:15 +0000 (07:09 +0000)]
Make the eisa probe messages just like all the others in the system.
Make the eisa no match printf closer to pci.

19 years agoAdd pnp and location info for the ISA bus. The pnp info is the
Warner Losh [Mon, 1 Aug 2005 07:03:10 +0000 (07:03 +0000)]
Add pnp and location info for the ISA bus.  The pnp info is the
primary vendor id for this device.  The location is empty because ISA
doesn't give one a way to generally locate a card.  PNP BIOS entries
do provide a way to locate cards, as do isa pnp cards.  These
locations will be added as soon as the code to remember them is
written.

19 years agoAT_MAKE_TAGID needs an instance as the 2nd arg- not just a 0.
Matt Jacob [Sun, 31 Jul 2005 23:21:19 +0000 (23:21 +0000)]
AT_MAKE_TAGID needs an instance as the 2nd arg- not just a 0.

19 years agoDelay freeing disk space for file system blocks until all dirty buffers
Stephan Uphoff [Sun, 31 Jul 2005 20:24:14 +0000 (20:24 +0000)]
Delay freeing disk space for file system blocks until all dirty buffers
are safely released. This fixes softdep problems on truncation (deletion)
of files with dirty buffers.

Reviewed by: jeff@, mckusick@, ps@, tegge@
Tested by:  glebius@, ps@
MFC after: 3 weeks

19 years agoStyle fix.
Alexander Leidinger [Sun, 31 Jul 2005 18:59:47 +0000 (18:59 +0000)]
Style fix.

Noticed by: njl

19 years agoWhats New:
Alexander Leidinger [Sun, 31 Jul 2005 16:16:22 +0000 (16:16 +0000)]
Whats New:
1. Support wide range sampling rate, as low as 1hz up to int32 max
   (which is, insane) through new feeder_rate, multiple precisions
   choice (32/64 bit converter). This is indeed, quite insane, but it
   does give us more room and flexibility. Plenty sysctl options to
   adjust resampling characteristics.
2. Support 24/32 bit pcm format conversion through new, much improved,
   simplified and optimized feeder_fmt.

Changes:
1. buffer.c / dsp.c / sound.h
   * Support for 24/32 AFMT.
2. feeder_rate.c
   * New implementation of sampling rate conversion with 32/64 bit
     precision, 1 - int32max hz (which is, ridiculous, yet very
     addictive).  Much improved / smarter buffer management to not
     cause any missing samples at the end of conversion process
   * Tunable sysctls for various aspect:
       hw.snd.feeder_rate_ratemin - minimum allowable sampling rate
       (default to 4000)
       hw.snd.feeder_rate_ratemax - maximum allowable sampling rate
       (default to 1102500)
       hw.snd.feeder_rate_buffersize - conversion buffer size
       (default to 8192)
       hw.snd.feeder_rate_scaling - scaling / conversion method
       (please refer to the source for explaination). Default to
       previous implementation type.
3. feeder_fmt.c / sound.h
   * New implementation, support for 24/32bit conversion, optimized,
     and simplified. Few routines has been removed (8 to xlaw, 16 to
     8). It just doesn't make sense.
4. channel.c
   * Support for 24/32 AFMT
   * Fix wrong xruns increment, causing incorrect underruns statistic
     while using vchans.
5. vchan.c
   * Support for 24/32 AFMT
   * Proper speed / rate detection especially for fixed rate ac97.
     User can override it using kernel hint:
     hint.pcm.<unit>.vchanrate="xxxx".

Notes / Issues:
        * Virtual Channels (vchans)
          Enabling vchans can really, really help to solve overrun
          issues.  This is quite understandable, because it operates
          entirely within its own buffering system without relying on
          hardware interrupt / state. Even if you don't need vchan,
          just enable single channel can help much. Few soundcards
          (notably via8233x, sblive, possibly others) have their own
          hardware multi channel, and this is unfortunately beyond
          vchan reachability.
        * The arrival of 24/32 also come with a price. Applications
          that can do 24/32bit playback need to be recompiled (notably
          mplayer).  Use (recompiled) mplayer to experiment / test /
          debug this various format using -af format=fmt. Note that
          24bit seeking in mplayer is a little bit broken, sometimes
          can cause silence or loud static noise. Pausing / seeking
          few times can solve this problem.
          You don't have to rebuild world entirely for this. Simply
          copy /usr/src/sys/sys/soundcard.h to
          /usr/include/sys/soundcard.h would suffice. Few drivers also
          need recompilation, and this can be done via
          /usr/src/sys/modules/sound/.
          Support for 24bit hardware playback is beyond the scope of
          this changes. That would require spessific hardware driver
          changes.
        * Don't expect playing 9999999999hz is a wise decision. Be
          reasonable. The new feeder_rate implemention provide
          flexibility, not insanity. You can easily chew up your CPU
          with this kind of mind instability. Please use proper
          mosquito repellent device for this obvious cracked brain
          attempt. As for testing purposes, you can use (again)
          mplayer to generate / play with different sampling rate. Use
          something like "mplayer -af resample=192000:0:0 <files>".

Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Tested by: multimedia@

19 years agoOSS defines AFMT_x32_xE as:
Alexander Leidinger [Sun, 31 Jul 2005 16:08:03 +0000 (16:08 +0000)]
OSS defines AFMT_x32_xE as:
        1) 32bit data, packed within 32bit (4bytes) boundary.
        2) 24bit data, packed within 32bit (4bytes) boundary where the data
           is stored in the 24 most significant bits and least significant 8
           bits are not used and should be set to 0.

While this might hold true in few cases, lots of applications (notably
mplayer, sweep) really deal / produce 24bit as what they should meant
to be: 24bit data / 3bytes per sample.
To handle this "true" 24bit pcm format add AFMT_x24_xE, so the in-kernel
conversion space did not confuse itself with 32bit variant.

You need to rebuild mplayer after installing this change (this header and
the upcomming kernel changes), if you want to use this new feature.

Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Tested by: multimedia@

19 years agoAdd stuff to HTML output.
Poul-Henning Kamp [Sun, 31 Jul 2005 15:42:25 +0000 (15:42 +0000)]
Add stuff to HTML output.

19 years agoIf a thread was removed from system run queue, kse_assign shouldn't
David Xu [Sun, 31 Jul 2005 15:11:21 +0000 (15:11 +0000)]
If a thread was removed from system run queue, kse_assign shouldn't
add it again.

19 years ago* Slightly new method to detect mixer capabilities and resolution.
Alexander Leidinger [Sun, 31 Jul 2005 14:28:31 +0000 (14:28 +0000)]
* Slightly new method to detect mixer capabilities and resolution.
  This mostly to help CT4730, but apparently it does help other
  cards too (especially via8233x). This probably need further test
  and confirmation from other people with ac97 cards other than via
  / es137x.
* Aggresive dac power wake up call, again, to help CT4730 (and
  probably others).

Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Tested by: multimedia@

19 years agoThe beginnings of a tool to survey the build options (NO_THIS, NO_THAT etc)
Poul-Henning Kamp [Sun, 31 Jul 2005 14:16:13 +0000 (14:16 +0000)]
The beginnings of a tool to survey the build options (NO_THIS, NO_THAT etc)
for effect.

19 years agoCorrect the description of the TAPE environment variable.
Ceri Davies [Sun, 31 Jul 2005 13:56:19 +0000 (13:56 +0000)]
Correct the description of the TAPE environment variable.
Based on:

PR: docs/84200
Submitted by: Gary W. Swearingen <garys at opusnet dot com>
MFC after: 1 week

19 years ago- Fixup the locking.
Alexander Leidinger [Sun, 31 Jul 2005 13:53:53 +0000 (13:53 +0000)]
- Fixup the locking.
- Don't mark MPSAFE (yet).
- DSP_CMD_DMAEXIT_8 doesn't work on old cards, use sb_reset_dsp() instead.

Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>

19 years ago- Fixup the locking.
Alexander Leidinger [Sun, 31 Jul 2005 13:51:04 +0000 (13:51 +0000)]
- Fixup the locking.
- Don't mark MPSAFE (yet).

Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>

19 years agoAdd another ID.
Alexander Leidinger [Sun, 31 Jul 2005 13:49:47 +0000 (13:49 +0000)]
Add another ID.

Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>

19 years ago* Add locking / MPSAFE
Alexander Leidinger [Sun, 31 Jul 2005 13:43:42 +0000 (13:43 +0000)]
* Add locking / MPSAFE
* Add kernel hint option to disable DXS channels entirely. Report
  from several skype users / Pav Lucistnik indicate that disabling
  DXS may fix lots of pop / crackling noise. To disable DXS add
  hint.pcm.<unit>.via_dxs_disabled="1" to /boot/device.hints.
  Further investigation of the issues regarding DXS showed, that
  the problem is in another (more generic) place, but until the
  right fix is tested/reviewed this may help a little bit.

Added sysctl's to aid testing/debugging:
hint.pcm.<unit>.via_dxs_disabled=X - Disable / Enable DXS channels entirely
hint.pcm.<unit>.via_dxs_channels=X - Limit DXS channels up to X
hint.pcm.<unit>.via_sgd_channels=X - Limit SGD channels up to X
hint.pcm.<unit>.via_dxs_src=X      - Enable / Disable DXS sample rate
                                     converter.

Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Tested by: multimedia@

19 years agoMention the default value of PATH and that it may be overridden.
Christian Brueffer [Sun, 31 Jul 2005 13:41:23 +0000 (13:41 +0000)]
Mention the default value of PATH and that it may be overridden.

Submitted by: Roman Divacky
MFC after: 3 days

19 years agoDocument 2 new sysctl's.
Alexander Leidinger [Sun, 31 Jul 2005 13:27:52 +0000 (13:27 +0000)]
Document 2 new sysctl's.

Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Submitted by: Jonathan Noack <noackjr@alumni.rice.edu> (implicit)

19 years agoFix a LOR introduced with the last commit (some hours ago), I had the
Alexander Leidinger [Sun, 31 Jul 2005 13:24:47 +0000 (13:24 +0000)]
Fix a LOR introduced with the last commit (some hours ago), I had the
wrong version of the patch.

Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Pointy hat to: netchild