]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
18 years agoAlign output.
Ruslan Ermilov [Wed, 9 Nov 2005 15:12:57 +0000 (15:12 +0000)]
Align output.

18 years agoMake IFP2NG() usable as an lvalue.
Ruslan Ermilov [Wed, 9 Nov 2005 13:34:43 +0000 (13:34 +0000)]
Make IFP2NG() usable as an lvalue.

18 years agoUse sparse initializers for "struct domain" and "struct protosw",
Ruslan Ermilov [Wed, 9 Nov 2005 13:29:16 +0000 (13:29 +0000)]
Use sparse initializers for "struct domain" and "struct protosw",
so they are easier to follow for the human being.

18 years agoMFi386: Remove obsolete options.
Yoshihiro Takahashi [Wed, 9 Nov 2005 12:26:37 +0000 (12:26 +0000)]
MFi386: Remove obsolete options.

18 years agoFix pc98 build.
Yoshihiro Takahashi [Wed, 9 Nov 2005 12:22:26 +0000 (12:22 +0000)]
Fix pc98 build.

18 years agoNo longer needed: replaced by mmu_if.m/pmap_dispatch.c/mmu_oea.c
Peter Grehan [Wed, 9 Nov 2005 10:28:51 +0000 (10:28 +0000)]
No longer needed: replaced by mmu_if.m/pmap_dispatch.c/mmu_oea.c

18 years agoAdd new ports-net-im collection.
Pav Lucistnik [Wed, 9 Nov 2005 10:16:26 +0000 (10:16 +0000)]
Add new ports-net-im collection.

18 years agoApply r1.103 to correct place.
Peter Grehan [Wed, 9 Nov 2005 09:54:33 +0000 (09:54 +0000)]
Apply r1.103 to correct place.

pmap.c on PowerPC is now a combo of mmu_if.m, pmap_dispatch.c
and mmu_oea.c

(I forgot to delete pmap.c from CVS in last jumbo commit)

18 years agoUncomment em(4) as it's now working.
Pyun YongHyeon [Wed, 9 Nov 2005 08:49:46 +0000 (08:49 +0000)]
Uncomment em(4) as it's now working.

18 years agoConnect em(4) build on sparc64.
Pyun YongHyeon [Wed, 9 Nov 2005 08:46:02 +0000 (08:46 +0000)]
Connect em(4) build on sparc64.

18 years agoMake em(4) work on big-endian architectures.
Pyun YongHyeon [Wed, 9 Nov 2005 08:43:18 +0000 (08:43 +0000)]
Make em(4) work on big-endian architectures.
 - disable jumbo frame support on strict alignment architectures due
   to the limitation of hardware. The driver needs a fix-up code for
   RX side. The fix will show up in near future.
 - fix endian issue for 82544 on PCI-X bus. I couldn't test this as
   I don't have the NIC/hardware.
 - prefer PCIR_BAR to hardcoded EM_MMBA.
 - Properly checks for for 64bit BAR [1]
 - replace inl/outl with bus_space(9) [1]
 - fix endian issue on VLAN handling.
 - reorder header files and remove unnecessary one.

Reviewed by: cognet
No response from: pdeuskar, tackerman
Obtained from: OpenBSD [1]

18 years agoReimplement the reclamation of PV entries. Specifically, perform
Alan Cox [Wed, 9 Nov 2005 08:19:21 +0000 (08:19 +0000)]
Reimplement the reclamation of PV entries.  Specifically, perform
reclamation synchronously from get_pv_entry() instead of
asynchronously as part of the page daemon.  Additionally, limit the
reclamation to inactive pages unless allocation from the PV entry zone
or reclamation from the inactive queue fails.  Previously, reclamation
destroyed mappings to both inactive and active pages.  get_pv_entry()
still, however, wakes up the page daemon when reclamation occurs.  The
reason being that the page daemon may move some pages from the active
queue to the inactive queue, making some new pages available to future
reclamations.

Print the "reclaiming PV entries" message at most once per minute, but
don't stop printing it after the fifth time.  This way, we do not give
the impression that the problem has gone away.

Reviewed by: tegge

18 years agoWIFxxx macros requires an int type but p_xstat is short, convert it
David Xu [Wed, 9 Nov 2005 07:58:16 +0000 (07:58 +0000)]
WIFxxx macros requires an int type but p_xstat is short, convert it
to int before using the macros.

Bug reported by : Pyun YongHyeon pyunyh at gmail dot com

18 years agoKick off the suspend sequence from the keyboard in a SWI rather than
Warner Losh [Wed, 9 Nov 2005 07:32:01 +0000 (07:32 +0000)]
Kick off the suspend sequence from the keyboard in a SWI rather than
in the hardware interrupt context (even if it is likely just an
ithread).  We don't document that suspend/resume routines are run from
such a context and some of the things that happen in those routines
aren't interrupt safe.  Since there's no real need to run from that
context, this restores assumptions that suspend routines have made.

This fixes Thierry Herbelot's 'Trying to sleep while sleeping is
prohibited' problem.

18 years agoClarify panic message, I parsed the old one 'trying to sleep while sleeping'
Warner Losh [Wed, 9 Nov 2005 07:28:52 +0000 (07:28 +0000)]
Clarify panic message, I parsed the old one 'trying to sleep while sleeping'

18 years agoRemove obsolete options
Warner Losh [Wed, 9 Nov 2005 04:30:55 +0000 (04:30 +0000)]
Remove obsolete options

18 years agoAdd support for XBOX to the FreeBSD port. The xbox architecture is
Warner Losh [Wed, 9 Nov 2005 03:55:40 +0000 (03:55 +0000)]
Add support for XBOX to the FreeBSD port.  The xbox architecture is
nearly identical to wintel/ia32, with a couple of tweaks.  Since it is
so similar to ia32, it is optionally added to a i386 kernel.  This
port is preliminary, but seems to work well.  Further improvements
will improve the interaction with syscons(4), port Linux nforce driver
and future versions of the xbox.

This supports the 64MB and 128MB boxes.  You'll need the most recent
CVS version of Cromwell (the Linux BIOS for the XBOX) to boot.

Rink will be maintaining this port, and is interested in feedback.
He's setup a website http://xbox-bsd.nl to report the latest
developments.

Any silly mistakes are my fault.

Submitted by: Rink P.W. Springer rink at stack dot nl and
Ed Schouten ed at fxq dot nl

18 years agoImprove diagnostic message.
Warner Losh [Wed, 9 Nov 2005 03:37:52 +0000 (03:37 +0000)]
Improve diagnostic message.

18 years agoFor nmount(), allow a text string error message to be propagated back
Craig Rodrigues [Wed, 9 Nov 2005 02:26:38 +0000 (02:26 +0000)]
For nmount(), allow a text string error message to be propagated back
to user-space if a parameter named "errmsg" is passed into the iovec.
Used in conjunction with vfs_mount_error(), more useful error messages
than errno can be passed back to userspace when mounting a filesystem
fails.

Discussed with: phk, pjd

18 years agoAdded list of supported 9xxxSX models.
Vinod Kashyap [Wed, 9 Nov 2005 00:23:54 +0000 (00:23 +0000)]
Added list of supported 9xxxSX models.

18 years agoIn aio_waitcomplete, do not return EAGAIN if no other threads
David Xu [Tue, 8 Nov 2005 23:48:32 +0000 (23:48 +0000)]
In aio_waitcomplete, do not return EAGAIN if no other threads
have started aio, instead, initialize aio management structure
if it hasn't been done, the reason to adjust this behavior is
to make it a bit friendly for threaded program, consider two
threads, one submits aio_write, and another just calls
aio_waitcomplete to wait any I/O to be completed and recycle the
aio requests, before submitter doing any I/O, the recycler wants
to wait in kernel. This also fixes inconsistency with other aio
syscalls.

18 years agoMake sure pending SIGCHLD is removed from previous parent when process
David Xu [Tue, 8 Nov 2005 23:28:12 +0000 (23:28 +0000)]
Make sure pending SIGCHLD is removed from previous parent when process
is attached or detached.

18 years agotwa corresponding to the 9.3.0.1 release on the 3ware website. This driver has
Vinod Kashyap [Tue, 8 Nov 2005 22:51:43 +0000 (22:51 +0000)]
twa corresponding to the 9.3.0.1 release on the 3ware website.  This driver has
support for the 9xxxSX controllers, along with the earlier 9xxxS series
controllers.

18 years agoThe hptmv inherently believes that a 'long' can hold a physical address.
Scott Long [Tue, 8 Nov 2005 22:12:26 +0000 (22:12 +0000)]
The hptmv inherently believes that a 'long' can hold a physical address.
This hasn't been true on i386 for at least a decade, probably longer, but
I'm too lazy to look up the exact year that PAE support was introduced.
Thus, this driver doesn't work on PAE.

X-MFC After: now

18 years agoIntroduce 3 new options for pkg_create(1), -x for using basic
Kirill Ponomarev [Tue, 8 Nov 2005 20:48:26 +0000 (20:48 +0000)]
Introduce 3 new options for pkg_create(1), -x for using basic
regular expressions for pkg-name, -E for extended regexps and -G for
exact matching.

Submitted by: mux
MFC after: 3 days

18 years agoMove the cloned interface list management in to if_clone. For some drivers the
Andrew Thompson [Tue, 8 Nov 2005 20:08:34 +0000 (20:08 +0000)]
Move the cloned interface list management in to if_clone. For some drivers the
softc lists and associated mutex are now unused so these have been removed.

Calling if_clone_detach() will now destroy all the cloned interfaces for the
driver and in most cases is all thats needed to unload.

Idea by: brooks
Reviewed by: brooks

18 years agoVarious and sundry cleanups:
John Baldwin [Tue, 8 Nov 2005 17:43:05 +0000 (17:43 +0000)]
Various and sundry cleanups:
- Use curthread for calls to knlist_delete() and add a big comment
  explaining why as well as appropriate assertions.
- Use TAILQ_FOREACH and TAILQ_FOREACH_SAFE instead of handrolling them.
- Use fget() family of functions to lookup file objects instead of
  grovelling around in file descriptor tables.
- Destroy the aio_freeproc mutex if we are unloaded.

Tested on: i386

18 years agoGiant clean up for exit(2)
Christian S.J. Peron [Tue, 8 Nov 2005 17:11:03 +0000 (17:11 +0000)]
Giant clean up for exit(2)

-Change unconditional aquisition of Giant to only pickup Giant if the vnode
 for the controlling tty resides on a non-mpsafe file system.
-Pickup Giant around executable vnode reference counting operations only if
 the executable resides on a non-mpsafe file system.
-If this process is being traced, pickup Giant for trace file reference count
 operations only if it resides on a non-mpsafe file system.

Discussed with: jhb
Tested by: kris

18 years agoFix standalone module build for viapm. Note that by default it doesn't
John Baldwin [Tue, 8 Nov 2005 17:03:09 +0000 (17:03 +0000)]
Fix standalone module build for viapm.  Note that by default it doesn't
include the ISA bus support even though it probably should.

18 years agoMollify the whitespace police
Warner Losh [Tue, 8 Nov 2005 15:42:12 +0000 (15:42 +0000)]
Mollify the whitespace police

18 years agoFix support for multiple RocketPort cards in the same machine by including
John Baldwin [Tue, 8 Nov 2005 15:33:39 +0000 (15:33 +0000)]
Fix support for multiple RocketPort cards in the same machine by including
the RocketPort unit number in the name of the devices.  This means that
unit 0 device names will change from ttyR0 .. ttyRf to ttyR00 .. ttyR0f.

Reviewed by: phk

18 years agoThere's no need to include <machine/asmacros.h> here.
Olivier Houchard [Tue, 8 Nov 2005 13:01:29 +0000 (13:01 +0000)]
There's no need to include <machine/asmacros.h> here.

18 years agoRework ARP retransmission algorythm so that ARP requests are
Gleb Smirnoff [Tue, 8 Nov 2005 12:05:57 +0000 (12:05 +0000)]
Rework ARP retransmission algorythm so that ARP requests are
retransmitted without suppression, while there is demand for
such ARP entry. As before, retransmission is rate limited to
one packet per second. Details:
  - Remove net.link.ether.inet.host_down_time
  - Do not set/clear RTF_REJECT flag on route, to
    avoid rt_check() returning error. We will generate error
    ourselves.
  - Return EWOULDBLOCK on first arp_maxtries failed
    requests , and return EHOSTDOWN/EHOSTUNREACH
    on further requests.
  - Retransmit ARP request always, independently from return
    code. Ratelimit to 1 pps.

18 years agoFix minor white space nit introduced in 1.102: use spaces, not tabs.
Robert Watson [Tue, 8 Nov 2005 09:53:28 +0000 (09:53 +0000)]
Fix minor white space nit introduced in 1.102: use spaces, not tabs.

18 years agoAdd code to test queued SIGCHLD.
David Xu [Tue, 8 Nov 2005 09:19:44 +0000 (09:19 +0000)]
Add code to test queued SIGCHLD.

18 years agoAdd support for queueing SIGCHLD same as other UNIX systems did.
David Xu [Tue, 8 Nov 2005 09:09:26 +0000 (09:09 +0000)]
Add support for queueing SIGCHLD same as other UNIX systems did.

For each child process whose status has been changed, a SIGCHLD instance
is queued, if the signal is stilling pending, and process changed status
several times, signal information is updated to reflect latest process
status. If wait() returns because the status of a child process is
available, pending SIGCHLD signal associated with the child process is
discarded. Any other pending SIGCHLD signals remain pending.

The signal information is allocated at the same time when proc structure
is allocated, if process signal queue is fully filled or there is a memory
shortage, it can still send the signal to process.

There is a booting time tunable kern.sigqueue.queue_sigchild which
can control the behavior, setting it to zero disables the SIGCHLD queueing
feature, the tunable will be removed if the function is proved that it is
stable enough.

Tested on: i386 (SMP and UP)

18 years agoSimplify setting the link-level address.
Ruslan Ermilov [Tue, 8 Nov 2005 09:03:06 +0000 (09:03 +0000)]
Simplify setting the link-level address.

18 years agoBump the maximum number of archive formats that can be
Tim Kientzle [Tue, 8 Nov 2005 07:44:39 +0000 (07:44 +0000)]
Bump the maximum number of archive formats that can be
enabled at one time from 4 to 8.

18 years agoCorrectly clean up if gzip format gets mis-identified as compress format.
Tim Kientzle [Tue, 8 Nov 2005 07:42:42 +0000 (07:42 +0000)]
Correctly clean up if gzip format gets mis-identified as compress format.
(This can only happen in the pathalogical case where the client is
providing single-byte blocks.)

18 years agoFine-tune the format detection for CPIO and ISO9660 sub-types.
Tim Kientzle [Tue, 8 Nov 2005 07:41:03 +0000 (07:41 +0000)]
Fine-tune the format detection for CPIO and ISO9660 sub-types.
This has no impact on the actual operation, it just fixes some
inaccuracies in the format code and description reported back to the caller.

18 years agoName change from pmap_* to moea_* to fit into the new order of
Peter Grehan [Tue, 8 Nov 2005 06:49:45 +0000 (06:49 +0000)]
Name change from pmap_* to moea_* to fit into the new order of
mmu implementation.

This code handles the 32-bit 'OEA' MMU found on G2/G3/G4 PPC cores.

18 years agoInsert a layer of indirection to the pmap code, using a kobj for
Peter Grehan [Tue, 8 Nov 2005 06:48:08 +0000 (06:48 +0000)]
Insert a layer of indirection to the pmap code, using a kobj for
the interface. This allows run-time selection of MMU code, based
on CPU-type detection, or tunable-overrides when testing new code.

Pre-requisite for G5 support.

conf/files.powerpc
  - remove pmap.c
  - add mmu_if.h, mmu_oea.c, pmap_dispatch.c

powerpc/include/mmuvar.h
  - definitions for MMU implementations

powerpc/include/pmap.h
  - remove pmap_pte_spill declaration
  - add pmap_mmu_install declaration
  - size the phys_avail array
  - pmap_bootstrapped is now global-scope

powerpc/powerpc/machdep.c
  - call kobj_machdep_init early in the boot sequence to allow
    kobj usage prior to SI_SUB_LOCK
  - install the OEA pmap code. This will be moved to CPU-specific
    init code in the future.

powerpc/powerpc/mmu_if.m
  - Kobj MMU interface definitions

powerpc/powerpc/pmap_dispatch.c
  - central dispatch for pmap calls
  - contains the global mmu kobj and the routine to locate the
   the mmu implementation and init the kobj

18 years agoIf a physical page is mapped by two or more virtual addresses, transmitted
Alan Cox [Tue, 8 Nov 2005 06:33:21 +0000 (06:33 +0000)]
If a physical page is mapped by two or more virtual addresses, transmitted
by the zero-copy sockets method, and written to before the transmission
completes, we need to destroy all of the existing mappings to the page,
not just the one that we fault on.  Otherwise, the mappings will no longer
be to the same page and changes made through one of the mappings will not
be visible through the others.

Observed by: tegge

18 years agoForced commit to note that this file has been repo-copied
Peter Grehan [Tue, 8 Nov 2005 04:20:49 +0000 (04:20 +0000)]
Forced commit to note that this file has been repo-copied
from pmap.c

18 years agoAdd utility function to propagate mount errors as text string messages.
Craig Rodrigues [Tue, 8 Nov 2005 04:13:39 +0000 (04:13 +0000)]
Add utility function to propagate mount errors as text string messages.

Discussed with: phk

18 years agoAdd #nclude <dev/pci/pcireg.h> to pick up definitions for PCIR_BAR and
Craig Rodrigues [Tue, 8 Nov 2005 04:11:50 +0000 (04:11 +0000)]
Add #nclude <dev/pci/pcireg.h> to pick up definitions for PCIR_BAR and
PCIR_SUBVEND_0.

18 years agoPortability: Use some autoconf magic to include the
Tim Kientzle [Tue, 8 Nov 2005 03:52:42 +0000 (03:52 +0000)]
Portability:  Use some autoconf magic to include the
correct headers for major()/minor()/makedev() on various
platforms.

Thanks to: Darin Broady

18 years agoUse PCIR_xxx constants for PCI config space header registers rather than
John Baldwin [Mon, 7 Nov 2005 21:53:58 +0000 (21:53 +0000)]
Use PCIR_xxx constants for PCI config space header registers rather than
magic numbers.

18 years ago*sigh* Revert stuff that wasn't supposed to be committed. The
John Baldwin [Mon, 7 Nov 2005 21:52:06 +0000 (21:52 +0000)]
*sigh*  Revert stuff that wasn't supposed to be committed.  The
acpi_resource change was a minor nit offered as an early candidate for
the recent ACPICA import problem and the acpi.c change is one I need to
test still that makes the ordered probing of system devices actually work
as advertised (probe devices in order based on the type of device rather
than in the order we encounter them in the device tree).

18 years agoWork around at least one busted BIOS. If we get a source index in a _PRT
John Baldwin [Mon, 7 Nov 2005 21:48:45 +0000 (21:48 +0000)]
Work around at least one busted BIOS.  If we get a source index in a _PRT
entry that is not zero, assume that it is really a hard-wired IRQ (commonly
used for APIC routing) and not a source index.  In practice, we've only
ever seen source indices of 0 for legitimate non-hard-wired _PRT entries.

Reviewed by: njl
Tested by: Alex Lyashkov shadow at psoft dot net
MFC after: 2 weeks

18 years agoFix malloc size (visible on amd64, with "kldconfig -r").
Ruslan Ermilov [Mon, 7 Nov 2005 19:22:20 +0000 (19:22 +0000)]
Fix malloc size (visible on amd64, with "kldconfig -r").

18 years agoMinor comment tweak to prevent gcc from being upset about the substring
Robert Watson [Mon, 7 Nov 2005 18:45:46 +0000 (18:45 +0000)]
Minor comment tweak to prevent gcc from being upset about the substring
/* appearing in a comment.

18 years agoSlightly reorganize to reduce duplicated code.
Xin LI [Mon, 7 Nov 2005 18:25:23 +0000 (18:25 +0000)]
Slightly reorganize to reduce duplicated code.

Reviewed by: rwatson

18 years agoAdd some rationale about when to bump and not bump the config version.
Peter Wemm [Mon, 7 Nov 2005 17:37:27 +0000 (17:37 +0000)]
Add some rationale about when to bump and not bump the config version.
Clarify that it is not like the shlib versions, and not like param.h's
__FreeBSD_version/osreldate either.

When config(8) was actively changing a while back, the interface between
config and the build system (eg: /sys/conf/files.* and Makefile.*) was
changing rapidly.  configvers is a version number of that interface.

User specified config files do not have a version number.  The decision
about whether a user supplied config file is syntactically valid or not
belongs to the parser and sanity checks, not an arbitary number.

18 years agoChange the definition for EXT_NDIS to EXT_NET_DRV. Since the latest
Bill Paul [Mon, 7 Nov 2005 16:57:14 +0000 (16:57 +0000)]
Change the definition for EXT_NDIS to EXT_NET_DRV. Since the latest
mbuf code changes, MEXTADD() can be used to add an external buffer with
arbitrary type, but mb_ext_free() won't let you free it.

18 years agoAdd tests for -t option with short tty name - pkill(1) should accept both
Pawel Jakub Dawidek [Mon, 7 Nov 2005 16:56:16 +0000 (16:56 +0000)]
Add tests for -t option with short tty name - pkill(1) should accept both
(eg. "ttyv0" and "v0").

18 years agoFinish the removal of threads support in ../config.mk,v 1.15.
Ruslan Ermilov [Mon, 7 Nov 2005 15:22:35 +0000 (15:22 +0000)]
Finish the removal of threads support in ../config.mk,v 1.15.

18 years agoRequire DESTDIR to be set when installing for a different architecture.
Ruslan Ermilov [Mon, 7 Nov 2005 15:03:04 +0000 (15:03 +0000)]
Require DESTDIR to be set when installing for a different architecture.

18 years agoAdd sigqueue test code.
David Xu [Mon, 7 Nov 2005 14:10:33 +0000 (14:10 +0000)]
Add sigqueue test code.

18 years agoEliminate tinderbox errors.
Craig Rodrigues [Mon, 7 Nov 2005 13:10:45 +0000 (13:10 +0000)]
Eliminate tinderbox errors.

18 years agoFix recording device selection based on ALS4000 datasheet.
Ariff Abdullah [Mon, 7 Nov 2005 09:26:17 +0000 (09:26 +0000)]
Fix recording device selection based on ALS4000 datasheet.
- http://www.alsa-project.org/alsa/ftp/manuals/avance_logic/ALS4000a.PDF

Submitted by: Watanabe Kazuhiro <CQG00620@nifty.ne.jp>

18 years agoFix kernel panic caused by double mss_unlock().
Ariff Abdullah [Mon, 7 Nov 2005 09:25:15 +0000 (09:25 +0000)]
Fix kernel panic caused by double mss_unlock().

Noticed by: Watanabe Kazuhiro <CQG00620@nifty.ne.jp>

18 years agoRemove unecessary include file. machine/rpb.h is very alpha specific
Warner Losh [Mon, 7 Nov 2005 07:41:17 +0000 (07:41 +0000)]
Remove unecessary include file.  machine/rpb.h is very alpha specific
and is not needed for this font, which isn't alpha speciifc.

18 years agoFinally (!?) get to the bottom of the mysterious G3 boot-time panics.
Peter Grehan [Mon, 7 Nov 2005 06:55:48 +0000 (06:55 +0000)]
Finally (!?) get to the bottom of the mysterious G3 boot-time panics.
After a number of tests using nop's to change the alignment, it was
confirmed that the mtibat instructions should be cache-aligned.
FreeScale app note AN2540 indicates that the isync before and after
the mtdbat is the right thing to do, but sync/isync isn't required
before the mtibat so it has been removed.

Fix by using a ".balign 32" to pull the code in question to the correct
alignment.

MFC after: 3 days

18 years agoPortability: timegm() isn't standard, so check for timegm() in
Tim Kientzle [Sun, 6 Nov 2005 23:38:01 +0000 (23:38 +0000)]
Portability:  timegm() isn't standard, so check for timegm() in
the configure script and substitute mktime() when necessary.

Thanks to:  Darin Broady

18 years agoEdit pathnames for -x but not for -t. Otherwise, people get confused
Tim Kientzle [Sun, 6 Nov 2005 22:53:51 +0000 (22:53 +0000)]
Edit pathnames for -x but not for -t.  Otherwise, people get confused
when list the archive contents, then try to extract selected files
(file selection always works against unedited pathnames).  With this change,
-t always shows the pathnames as they appear in the archive.

Thanks to: Robert Watson

18 years agoRefactor the PCI probe code a bit.
Scott Long [Sun, 6 Nov 2005 22:52:52 +0000 (22:52 +0000)]
Refactor the PCI probe code a bit.

18 years agoBe more consistent with the rest of the manual page.
Pawel Jakub Dawidek [Sun, 6 Nov 2005 20:01:47 +0000 (20:01 +0000)]
Be more consistent with the rest of the manual page.

18 years ago- Do not raise IFF_DRV_OACTIVE flag in vlan_start, because this
Gleb Smirnoff [Sun, 6 Nov 2005 19:43:04 +0000 (19:43 +0000)]
- Do not raise IFF_DRV_OACTIVE flag in vlan_start, because this
  can lead to stalled interface
- Explain this fact in a comment.

Reviewed by: rwatson, thompsa, yar

18 years agoThe latest version of the Intel 2200BG/2915ABG driver (9.0.0.3-9) from
Bill Paul [Sun, 6 Nov 2005 19:38:34 +0000 (19:38 +0000)]
The latest version of the Intel 2200BG/2915ABG driver (9.0.0.3-9) from
Intel's web site requires some minor tweaks to get it to work:

- The driver seems to have been released with full WMI tracing enabled,
  and makes references to some WMI APIs, namely IoWMIRegistrationControl(),
  WmiQueryTraceInformation() and WmiTraceMessage(). Only the first
  one is ever called (during intialization). These have been implemented
  as do-nothing stubs for now. Also added a definition for STATUS_NOT_FOUND
  to ntoskrnl_var.h, which is used as a return code for one of the WMI
  routines.

- The driver references KeRaiseIrqlToDpcLevel() and KeLowerIrql()
  (the latter as a function, which is unusual because normally
  KeLowerIrql() is a macro in the Windows DDK that calls KfLowewIrql()).
  I'm not sure why these are being called since they're not really
  part of WDM. Presumeably they're being used for backwards
  compatibility with old versions of Windows. These have been
  implemented in subr_hal.c. (Note that they're _stdcall routines
  instead of _fastcall.)

- When querying the OID_802_11_BSSID_LIST OID to get a BSSID list,
  you don't know ahead of time how many networks the NIC has found
  during scanning, so you're allowed to pass 0 as the list length.
  This should cause the driver to return an 'insufficient resources'
  error and set the length to indicate how many bytes are actually
  needed. However for some reason, the Intel driver does not honor
  this convention: if you give it a length of 0, it returns some
  other error and doesn't tell you how much space is really needed.
  To get around this, if using a length of 0 yields anything besides
  the expected error case, we arbitrarily assume a length of 64K.
  This is similar to the hack that wpa_supplicant uses when doing
  a BSSID list query.

18 years agoCopy out the number of iovecs in freebsd32_recvmsg, not the length
Paul Saab [Sun, 6 Nov 2005 18:12:43 +0000 (18:12 +0000)]
Copy out the number of iovecs in freebsd32_recvmsg, not the length
of a single iovec.

18 years agoDetach k_rem_pio2f.c from the build since it is now unused. It is a libm
Bruce Evans [Sun, 6 Nov 2005 17:59:40 +0000 (17:59 +0000)]
Detach k_rem_pio2f.c from the build since it is now unused.  It is a libm
internal so this shouldn't cause version problems.

18 years agoUse a 53-bit approximation to pi/2 instead of a 33+53 bit one for the
Bruce Evans [Sun, 6 Nov 2005 17:48:02 +0000 (17:48 +0000)]
Use a 53-bit approximation to pi/2 instead of a 33+53 bit one for the
special case pi/4 <= |x| < 3*pi/4.  This gives a tiny optimization (it
saves 2 subtractions, which are scheduled well so they take a whole 1
cycle extra on an AthlonXP), and simplifies the code so that the
following optimization is not so ugly.

Optimize for the range 3*pi/4 < |x| < 9*Pi/2 in the same way.  On
Athlon{XP,64} systems, this gives a 25-40% optimization (depending a
lot on CFLAGS) for the cosf() and sinf() consumers on this range.
Relative to i387 hardware fcos and fsin, it makes the software versions
faster in most cases instead of slower in most cases.  The relative
optimization is smaller for tanf() the inefficient part is elsewhere.

The 53-bit approximation to pi/2 is good enough for pi/4 <= |x| <
3*pi/4 because after losing up to 24 bits to subtraction, we still
have 29 bits of precision and only need 25 bits.  Even with only 5
extra bits, it is possible to get perfectly rounded results starting
with the reduced x, since if x is nearly a multiple of pi/2 then x is
not near a half-way case and if x is not nearly a multiple of pi/2
then we don't lose many bits.  With our intentionally imperfect rounding
we get the same results for cosf(), sinf() and tanf() as without this
optimization.

18 years agoFix two minor typos that caused schedgraph to exit with an exception
Robert Watson [Sun, 6 Nov 2005 17:43:25 +0000 (17:43 +0000)]
Fix two minor typos that caused schedgraph to exit with an exception
when running on traces referencing >2 CPUs.

18 years agoFix panic string in last revision.
Gleb Smirnoff [Sun, 6 Nov 2005 16:47:59 +0000 (16:47 +0000)]
Fix panic string in last revision.

18 years agoMFi386 rev 1.536 (sort of)
Olivier Houchard [Sun, 6 Nov 2005 16:10:28 +0000 (16:10 +0000)]
MFi386 rev 1.536 (sort of)
Move what can be moved (UMA zones creation, pv_entry_* initialization) from
pmap_init2() to pmap_init().
Create a new function, pmap_postinit(), called from cpu_startup(), to do the
L1 tables allocation.
pmap_init2() is now empty for arm as well.

18 years agoAvoid trouble with PUC_FASTINTR if it is already defined.
Poul-Henning Kamp [Sun, 6 Nov 2005 15:55:45 +0000 (15:55 +0000)]
Avoid trouble with PUC_FASTINTR if it is already defined.

18 years agoFix the path to aac_ioctl.h in the comment and style includes.
Maxim Konovalov [Sun, 6 Nov 2005 15:47:06 +0000 (15:47 +0000)]
Fix the path to aac_ioctl.h in the comment and style includes.

MFC after: 1 week

18 years agoFix include: aac_ioctl.h was moved from sys/dev/aac/ to sys/sys.
Maxim Konovalov [Sun, 6 Nov 2005 15:40:59 +0000 (15:40 +0000)]
Fix include: aac_ioctl.h was moved from sys/dev/aac/ to sys/sys.

PR: misc/88549
Submitted by: Sokolov Alexey
MFC after: 1 week

18 years agoRemove spl markers from AMR.
Scott Long [Sun, 6 Nov 2005 15:13:42 +0000 (15:13 +0000)]
Remove spl markers from AMR.

MFC After: 3 days

18 years agoThe logb() functions are not just ieee754 "test" functions, but are
Bruce Evans [Sun, 6 Nov 2005 12:18:27 +0000 (12:18 +0000)]
The logb() functions are not just ieee754 "test" functions, but are
standard in C99 and POSIX.1-2001+.  They are also not deprecated, since
apart from being standard they can handle special args slightly better
than the ilogb() functions.

Move their documentation to ilogb.3.  Try to use consistent and improved
wording for both sets of functions.  All of ieee854, C99 and POSIX
have better wording and more details for special args.

Add history for the logb() functions and ilogbl().  Fix history for
ilogb().

18 years agoChange author's email address.
Philip Paeps [Sun, 6 Nov 2005 00:41:41 +0000 (00:41 +0000)]
Change author's email address.

Submitted by: Sean Bullington shegget -at- gmail.com

18 years agoUn-break processing of device major/minor values with fstat -n. We do
Christian S.J. Peron [Sat, 5 Nov 2005 23:38:08 +0000 (23:38 +0000)]
Un-break processing of device major/minor values with fstat -n. We do
this by accessing the cdev_priv element of the cdev structure. Looking
forward we need a better way to handle this, as this structure shouldn't
be frobbed by userspace.

Submitted by: Doug Steinwand
PR: bin/88203
MFC after: 1 week
Discussed with: phk

18 years agoNow that fast interrupts can be shared we can use them in puc.
Poul-Henning Kamp [Sat, 5 Nov 2005 21:04:53 +0000 (21:04 +0000)]
Now that fast interrupts can be shared we can use them in puc.

18 years agoAdd yet another entry to the list.
Poul-Henning Kamp [Sat, 5 Nov 2005 21:04:26 +0000 (21:04 +0000)]
Add yet another entry to the list.

18 years agoAdd uart(4). When both sio(4) and uart(4) can handle a serial port,
Marcel Moolenaar [Sat, 5 Nov 2005 19:48:53 +0000 (19:48 +0000)]
Add uart(4). When both sio(4) and uart(4) can handle a serial port,
sio(4) will claim it. This change therefore only affects how ports
are handled when they are not claimed by sio(4), and in principle
will improve hardware support.

MFC after: 2 months

18 years agoFree only those mbuf+clusters back to the packet zone that were allocated
Andre Oppermann [Sat, 5 Nov 2005 19:43:55 +0000 (19:43 +0000)]
Free only those mbuf+clusters back to the packet zone that were allocated
from there.  All others get broken up and free'd individually to the mbuf
and cluster zones.

The packet zone is a secondary zone to the mbuf zone.  There is currently
a limitation in UMA which prevents decreasing the packet zone stock when
the mbuf and cluster zone are drained and all their members are part of
packets.  When this is fixed this change may be reverted.

18 years agoRemove spl.
Bernd Walter [Sat, 5 Nov 2005 17:17:57 +0000 (17:17 +0000)]
Remove spl.
Restart request on USBD_IOERROR.

MFC after: 2 weeks

18 years agoUnbreak !SMP kernels
Poul-Henning Kamp [Sat, 5 Nov 2005 10:42:40 +0000 (10:42 +0000)]
Unbreak !SMP kernels

18 years agoDefine M_IOAPIC the same as i386
Peter Wemm [Fri, 4 Nov 2005 23:02:28 +0000 (23:02 +0000)]
Define M_IOAPIC the same as i386

18 years agoMFamd64: indent with tabs instead of spaces.
Peter Wemm [Fri, 4 Nov 2005 22:53:44 +0000 (22:53 +0000)]
MFamd64: indent with tabs instead of spaces.

18 years agoThis commit was generated by cvs2svn to compensate for changes in r152069,
Jung-uk Kim [Fri, 4 Nov 2005 21:29:41 +0000 (21:29 +0000)]
This commit was generated by cvs2svn to compensate for changes in r152069,
which included commits to RCS files with non-trunk default branches.

18 years ago- Fix more resource parsing problems. The previous commit was imcomplete.
Jung-uk Kim [Fri, 4 Nov 2005 21:29:41 +0000 (21:29 +0000)]
- Fix more resource parsing problems.  The previous commit was imcomplete.
- Fix a typo in rsmisc.c and a style change for consistency.

This patch will also appear in future ACPI-CA release.

Submitted by: Robert Moore <robert dot moore at intel dot com>
Tested by: ru

18 years agoCatch up with the recent <sys/signal.h> change and make this compile.
Ruslan Ermilov [Fri, 4 Nov 2005 20:32:26 +0000 (20:32 +0000)]
Catch up with the recent <sys/signal.h> change and make this compile.

18 years agoAccount for the minimum resource size when parsing the end tag resource
Nate Lawson [Fri, 4 Nov 2005 20:15:09 +0000 (20:15 +0000)]
Account for the minimum resource size when parsing the end tag resource
descriptor.  This should fix the "memory modified after free" panics.  This
patch will appear in a future acpi-ca distribution.

Submitted by: Robert Moore <robert.moore / intel.com>
Tested by: Peter Holm

18 years agoThis commit was generated by cvs2svn to compensate for changes in r152058,
Nate Lawson [Fri, 4 Nov 2005 20:15:09 +0000 (20:15 +0000)]
This commit was generated by cvs2svn to compensate for changes in r152058,
which included commits to RCS files with non-trunk default branches.

18 years agoBegin and end the initialization of pvzone in pmap_init().
Alan Cox [Fri, 4 Nov 2005 18:03:24 +0000 (18:03 +0000)]
Begin and end the initialization of pvzone in pmap_init().
Previously, pvzone's initialization was split between pmap_init() and
pmap_init2().  This split initialization was the underlying cause of
some UMA panics during initialization.  Specifically, if the UMA boot
pages was exhausted before the pvzone was fully initialized, then UMA,
through no fault of its own, would use an inappropriate back-end
allocator leading to a panic.  (Previously, as a workaround, we have
increased the UMA boot pages.)  Fortunately, there is no longer any
reason that pvzone's initialization cannot be completed in
pmap_init().

Eliminate a check for whether pv_entry_high_water has been initialized
or not from get_pv_entry().  Since pvzone's initialization is
completed in pmap_init(), this check is no longer needed.

Use cnt.v_page_count, the actual count of available physical pages,
instead of vm_page_array_size to compute the maximum number of pv
entries.

Introduce the vm.pmap.pv_entries tunable on alpha and ia64.

Eliminate some unnecessary white space.

Discussed with: tegge (item #1)
Tested by: marcel (ia64)

18 years agoFix a logic error introduced with mandatory mbuf cluster refcounting and
Andre Oppermann [Fri, 4 Nov 2005 17:20:53 +0000 (17:20 +0000)]
Fix a logic error introduced with mandatory mbuf cluster refcounting and
freeing of mbufs+clusters back to the packet zone.

18 years agoUse if_setlladdr() to reprogram the link-level addresses; it causes
Ruslan Ermilov [Fri, 4 Nov 2005 16:29:51 +0000 (16:29 +0000)]
Use if_setlladdr() to reprogram the link-level addresses; it causes
a synchronous reprogramming of hardware MAC filters if the physical
interface are up and running.  Previously, MAC filters would be
reconfigured only when the fec interface was brought up.

18 years agoBring some level of stability to this driver:
Ruslan Ermilov [Fri, 4 Nov 2005 15:42:01 +0000 (15:42 +0000)]
Bring some level of stability to this driver:

- Disallow bundle reconfiguration when virtual
  interface is running; otherwise, removing a
  port from a running configuration will cause
  a panic in the start() method on the next packet
  on an assumption that a bundle has an even
  number of ports (2 or 4).

- Disallow bringing of virtual interface to a
  running state when a bundle size is 0; otherwise,
  adding and then removing the port will similarly
  cause a panic.

- Add missing initialization of fec_ifstat when
  adding a new port and fix media status reporting
  when virtual interface isn't yet up (check for
  fec_status of 1 rather than != 0).