]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
20 years agoAdjust stack alignment so that when the 'call xxx' functions are
Peter Wemm [Sun, 21 Mar 2004 01:39:01 +0000 (01:39 +0000)]
Adjust stack alignment so that when the 'call xxx' functions are
gathered into the middle of the _init and _fini sections, they get
executed with their expected stack alignment.

20 years agoChange (yet again, sorry!) the path of the 32 bit ld-elf.so.1.
Peter Wemm [Sun, 21 Mar 2004 01:22:24 +0000 (01:22 +0000)]
Change (yet again, sorry!) the path of the 32 bit ld-elf.so.1.

20 years agoAdd initial support for compiling a special 32 bit version of
Peter Wemm [Sun, 21 Mar 2004 01:21:26 +0000 (01:21 +0000)]
Add initial support for compiling a special 32 bit version of
ld-elf.so.1 on 64 bit systems.  Most of this involves using alternate
paths, environment variables and diagnostic messages.

The build glue is seperate.

20 years agoFix another Intel 2200BG bug: don't schedule ndis_ticktask() on media
Bill Paul [Sun, 21 Mar 2004 00:06:56 +0000 (00:06 +0000)]
Fix another Intel 2200BG bug: don't schedule ndis_ticktask() on media
disconnect events if the link wasn't even up yet.

20 years ago- Rewrite the timer and event API routines in subr_ndis.c so that they
Bill Paul [Sat, 20 Mar 2004 23:39:43 +0000 (23:39 +0000)]
- Rewrite the timer and event API routines in subr_ndis.c so that they
  are actually layered on top of the KeTimer API in subr_ntoskrnl.c, just
  as it is in Windows. This reduces code duplication and more closely
  imitates the way things are done in Windows.

- Modify ndis_encode_parm() to deal with the case where we have
  a registry key expressed as a hex value ("0x1") which is being
  read via NdisReadConfiguration() as an int. Previously, we tried
  to decode things like "0x1" with strtol() using a base of 10, which
  would always yield 0. This is what was causing problems with the
  Intel 2200BG Centrino 802.11g driver: the .inf file that comes
  with it has a key called RadioEnable with a value of 0x1. We
  incorrectly decoded this value to '0' when it was queried, hence
  the driver thought we wanted the radio turned off.

- In if_ndis.c, most drivers don't accept NDIS_80211_AUTHMODE_AUTO,
  but NDIS_80211_AUTHMODE_SHARED may not be right in some cases,
  so for now always use NDIS_80211_AUTHMODE_OPEN.

NOTE: There is still one problem with the Intel 2200BG driver: it
happens that the kernel stack in Windows is larger than the kernel
stack in FreeBSD. The 2200BG driver sometimes eats up more than 2
pages of stack space, which can lead to a double fault panic.
For the moment, I got things to work by adding the following to
my kernel config file:

options         KSTACK_PAGES=8

I'm pretty sure 8 is too big; I just picked this value out of a hat
as a test, and it happened to work, so I left it. 4 pages might be
enough. Unfortunately, I don't think you can dynamically give a
thread a larger stack, so I'm not sure how to handle this short of
putting a note in the man page about it and dealing with the flood
of mail from people who never read man pages.

20 years agoRemove NetBSD's sort(1), which we stopped using two years ago.
Dag-Erling Smørgrav [Sat, 20 Mar 2004 23:27:42 +0000 (23:27 +0000)]
Remove NetBSD's sort(1), which we stopped using two years ago.

20 years agoDon't make having ${DESTDIR}/boot/device.hints a prerequisite to
Marcel Moolenaar [Sat, 20 Mar 2004 22:47:30 +0000 (22:47 +0000)]
Don't make having ${DESTDIR}/boot/device.hints a prerequisite to
installing a kernel on ia64.

20 years agoAdd uart_subr.c
Marcel Moolenaar [Sat, 20 Mar 2004 22:41:52 +0000 (22:41 +0000)]
Add uart_subr.c

20 years agoBug: Standard C still requires declarations to precede statements. <sigh>
Tim Kientzle [Sat, 20 Mar 2004 22:35:33 +0000 (22:35 +0000)]
Bug: Standard C still requires declarations to precede statements. <sigh>

Portability: Eliminate an accidental __unused, accomodate
  systems with non-POSIX strerror_r

20 years agoDon peril-sensitive sunglasses and add PCI Id's for two new cards. I've
Scott Long [Sat, 20 Mar 2004 21:07:36 +0000 (21:07 +0000)]
Don peril-sensitive sunglasses and add PCI Id's for two new cards.  I've
only done minimal testing on one of these cards and the firmware folks
have been extremely uncooperative in answering my qeustions about them, so
hopefully they will work ok for everyone.

20 years ago - Add uiomove_fromphys() implementations to alpha and ia64. These only
Alan Cox [Sat, 20 Mar 2004 21:06:20 +0000 (21:06 +0000)]
 - Add uiomove_fromphys() implementations to alpha and ia64.  These only
   differ trivially from amd64.
 - Correct a spelling error in a comment.

20 years agoFix loop termination condition for parsing resources in _PRS buffers.
Nate Lawson [Sat, 20 Mar 2004 20:47:08 +0000 (20:47 +0000)]
Fix loop termination condition for parsing resources in _PRS buffers.
This completes the effort to handle dependent functions, which are used
in some machines for irq link resources.  Also, clean up some nearby
comments while I'm at it.

20 years agoIntroduce the cpumask_t type. The purpose of the type is to create a
Marcel Moolenaar [Sat, 20 Mar 2004 20:41:40 +0000 (20:41 +0000)]
Introduce the cpumask_t type. The purpose of the type is to create a
level of abstraction for any and all CPU mask and CPU bitmap variables
so that platforms have the ability to break free from the hard limit
of 32 CPUs, simply because we don't have more bits in an u_int. Note
that the type is not supposed to solve massive parallelism, where
the number of CPUs can be larger than the width of the widest integral
type. As such, cpumask_t is not supposed to be a compound type. If
such would be necessary in the future, we can deal with the issues
then and there. For now, it can be assumed that the type is integral
and unsigned.

With this commit, all MD definitions start off as u_int. This allows
us to phase-in cpumask_t at our leasure without breaking anything.
Once cpumask_t is used consistently, platforms can switch to wider
(or smaller) types if such would be beneficial (or not; whatever :-)

Compile-tested on: i386

20 years agoLet ether_ifattach() announce our MAC address.
Matthew N. Dodd [Sat, 20 Mar 2004 20:12:13 +0000 (20:12 +0000)]
Let ether_ifattach() announce our MAC address.

Submitted by: Marius Strobl <marius@alchemy.franken.de>

20 years agoDon't announce MAC addresses twice.
Matthew N. Dodd [Sat, 20 Mar 2004 19:57:47 +0000 (19:57 +0000)]
Don't announce MAC addresses twice.
(ieee80211_ifattach() calls ether_ifattach().)

20 years agoIntroduce uiomove_fromphys(). This is a variant of uiomove() that takes
Alan Cox [Sat, 20 Mar 2004 19:36:29 +0000 (19:36 +0000)]
Introduce uiomove_fromphys().  This is a variant of uiomove() that takes
a collection of physical pages as the source.  On amd64 it is implemented
using the direct virtual-to-physical map.

20 years agoo Add user proxy.
Maxim Konovalov [Sat, 20 Mar 2004 19:13:06 +0000 (19:13 +0000)]
o Add user proxy.

20 years agoFix the ioctl types for two ioctls. I'm not sure if the switch was my
Scott Long [Sat, 20 Mar 2004 19:02:46 +0000 (19:02 +0000)]
Fix the ioctl types for two ioctls.  I'm not sure if the switch was my
fault or the vendor's fault when I brought in rev 1.5.  This allows
the 'storcon' utility to work again.

Sponsored by: freebsdsystems.com

20 years agoAdded entries for ibm151 and ibm3151-25.
Jens Schweikhardt [Sat, 20 Mar 2004 13:11:59 +0000 (13:11 +0000)]
Added entries for ibm151 and ibm3151-25.
Style:
o Use tabs instead of 8 spaces
o Space after # beginning comment
o Fix entries lacking ":" after tab
o Removed whitespace at EOL

PR: misc/62749 (the ibm additions)
Submitted by: Gerhard Gonter <gonter@wu-wien.ac.at>
MFC after: 3 days

20 years ago"CPU is an acronym, and as such should be spelled in all uppercase."
Mark Santcroos [Sat, 20 Mar 2004 13:05:11 +0000 (13:05 +0000)]
"CPU is an acronym, and as such should be spelled in all uppercase."

Pointed out by: ru

20 years agoParenthesize function names in masking macros for getc() etc. for the
Tim J. Robbins [Sat, 20 Mar 2004 11:45:52 +0000 (11:45 +0000)]
Parenthesize function names in masking macros for getc() etc. for the
benefit of obsolete C preprocessors.

20 years agoSync manpage with code for hw.acpi.toshiba.cpu_speed sysctl.
Mark Santcroos [Sat, 20 Mar 2004 10:25:36 +0000 (10:25 +0000)]
Sync manpage with code for hw.acpi.toshiba.cpu_speed sysctl.

Submitted by: Chia-liang Kao <clkao@clkao.org>
Approved by: njl (mentor)

20 years agoMention that funopen() uses fpos_t incorrectly in the BUGS section.
Tim J. Robbins [Sat, 20 Mar 2004 08:41:12 +0000 (08:41 +0000)]
Mention that funopen() uses fpos_t incorrectly in the BUGS section.

20 years agoFix braino in previous commit: getenv() can return NULL.
Marcel Moolenaar [Sat, 20 Mar 2004 08:38:33 +0000 (08:38 +0000)]
Fix braino in previous commit: getenv() can return NULL.

20 years agoImprove documentation for fgetpos() and fsetpos(), and discourage
Tim J. Robbins [Sat, 20 Mar 2004 08:38:27 +0000 (08:38 +0000)]
Improve documentation for fgetpos() and fsetpos(), and discourage
users from assuming that fpos_t is an integral type.

20 years agoPut the event notification back where it was for freeBSD, after device creation.
Julian Elischer [Sat, 20 Mar 2004 07:31:11 +0000 (07:31 +0000)]
Put the event notification back where it was for freeBSD, after device creation.
Since NetBSD doesn't have devfs the order for them doesn't matter..
Reverses one part of 1.60->1.61 NetBSD diff reduction.

Obtained from:  Not NetBSD

20 years agoNew release note:
Hiroki Sato [Sat, 20 Mar 2004 07:17:12 +0000 (07:17 +0000)]
New release note:
ips(4) now supports the recent Adaptec flavors
of ServeRAID series SCSI controller cards.

20 years agoAdd a way for rc.d/devfs to set more than just the system devfs up.
Brian Feldman [Sat, 20 Mar 2004 07:10:16 +0000 (07:10 +0000)]
Add a way for rc.d/devfs to set more than just the system devfs up.
Yes, this means for stuff OTHER than jails, too.  Example usage:
#devfs_system_ruleset="root"
devfs_set_rulesets="/dev=root /etc/namedb/dev=named_devfs"

20 years agoMFC:
Hiroki Sato [Sat, 20 Mar 2004 06:19:38 +0000 (06:19 +0000)]
MFC:
xl(4) hardware TX checksum disabled.

20 years agoNew release note:
Hiroki Sato [Sat, 20 Mar 2004 06:14:55 +0000 (06:14 +0000)]
New release note:
xl(4) hardware TX checksum disabled,
uart(4) hw.uart.{console,dbgport} environment variable support.

20 years agoReplace uint64_t with unsigned long in struct dbreg.
Marcel Moolenaar [Sat, 20 Mar 2004 05:27:14 +0000 (05:27 +0000)]
Replace uint64_t with unsigned long in struct dbreg.

20 years agoActually program the list of recording devices in sv_mix_setrecsrc().
Marcel Moolenaar [Sat, 20 Mar 2004 04:38:21 +0000 (04:38 +0000)]
Actually program the list of recording devices in sv_mix_setrecsrc().
This change has not been tested.

This change was triggered by a gcc(1) warning on ia64 at -O2.  The
variable v was not used after being computed, which resulted in enough
dead code elimination (DCE) to confuse the compiler and emit a bogus
warning about the use of the variable i without prior definition. The
variable i is the loop variable.

Submitted by: des
Responsibility: marcel

20 years agoRemove the last traditional hints. These hints only served the purpose
Marcel Moolenaar [Sat, 20 Mar 2004 04:23:03 +0000 (04:23 +0000)]
Remove the last traditional hints. These hints only served the purpose
for uart(4) to figure out which device to use as console. Use this file
to define hw.uart.console instead so that we don't have to put it in
the default loader.conf, which makes it hard to override.

20 years agoForced commit to correct 1.167's truncated commit log:
Mike Silbersack [Sat, 20 Mar 2004 02:24:24 +0000 (02:24 +0000)]
Forced commit to correct 1.167's truncated commit log:

Disable hardware TX checksumming for 3c905 series chips, as we have
solid reports that it is buggy *and* that it slows down transmit speed.

20 years agoIntroduce the hw.uart.console and hw.uart.dbgport environment variables
Marcel Moolenaar [Sat, 20 Mar 2004 02:14:02 +0000 (02:14 +0000)]
Introduce the hw.uart.console and hw.uart.dbgport environment variables
to select a serial console and debug port (resp). On ia64 these replace
the use of hints completely and take precedence over hints on alpha,
amd64 and i386. On sparc64 these variables are not yet recognised.

The reasons for introducing these variables are:
1.  Hints have side-effects. They reserve the unit number for use by
    isa or acpi devices and therefore cannot be used to select a pci
    device. Also, the use of a unit number to select a device prior
    to bus enumeration is nonsense. The new variables have no side-
    effects and are not based on unit numbers.
2.  Hints don't have the expression power to allow the sysadmin to
    select UARTs that are not legacy PC devices and need the support
    of compile-time constants to give the sysadmin some level of
    flexibility.

The hw.uart.console and hw.uart.dbgport variables specify a list of
attributes. An attribute is a tag-value pair, seperated by a colon.
Attributes are seperated by a comma. Where possible, tags are the
same as those in /etc/remote (only br and pa in practice). Details
can be found in the manpage (not part of this commit).

Not tested on: amd64, pc98

20 years agosolid reports that it is buggy *and* that it slows down transmit
Mike Silbersack [Fri, 19 Mar 2004 23:20:23 +0000 (23:20 +0000)]
solid reports that it is buggy *and* that it slows down transmit
speed.

Buggy report:           Matt Dillon & others
Slowness report:        I can't find the e-mail

MFC After: 1 minute

20 years agoMany fixes:
Tim Kientzle [Fri, 19 Mar 2004 22:37:06 +0000 (22:37 +0000)]
Many fixes:

* Disabled shared-library building, as some API breakage is
  still likely.  (I didn't realize it was turned on by default.)  If
  you have an existing /usr/lib/libarchive.so.2, I recommend deleting it.
* Pax interchange format now correctly stores and reads UTF8
  for extended attributes.  In particular, pax format can portably
  handle arbitrarily long pathnames containing arbitrary characters.
* Library compiles cleanly at -O2, -O3, and WARNS=6 on all
  FreeBSD-CURRENT platforms.
* Minor portability improvements inspired by Juergen Lock
  and Greg Lewis.  (Less reliance on stdint.h, isolating of
  various portability-challenged constructs.)
* archive_entry transparently converts multi-byte <-> wide character
  strings, allowing clients and format handlers to deal with either
  one, as appropriate.
* Support for reading 'L' and 'K' entries in standard tar archives
  for star compatibility.
* Recognize (but don't yet handle) ACL entries from Solaris tar.
* Pushed format-specific data for format readers down into
  format-specific storage and out of library-global storage.  This
  should make it easier to maintain individual formats without mucking
  with the core library management.
* Documentation updates to track the above changes.
* Updates to tar.5 to correct a few mistakes and add some additional
  information about GNU tar and Solaris tar formats.

Notes:
* The basic 'tar' reader is getting more general; there's not much
  point in keeping the 'gnutar' reader separate.  Merging the two
  would lose a bunch of duplicate code.
* The libc ACL support is looking increasingly inadequate for my needs
  here.  I might need to assemble some fairly significant code for
  parsing and building ACLs. <sigh>

20 years ago - Remove some unused #includes.
Alan Cox [Fri, 19 Mar 2004 21:19:15 +0000 (21:19 +0000)]
 - Remove some unused #includes.
 - Apply some style fixes to mdstart_swap().

20 years agoThis commit was generated by cvs2svn to compensate for changes in r127208,
Dag-Erling Smørgrav [Fri, 19 Mar 2004 20:14:23 +0000 (20:14 +0000)]
This commit was generated by cvs2svn to compensate for changes in r127208,
which included commits to RCS files with non-trunk default branches.

20 years agoSync with OpenBSD (two-year old bug fix)
Dag-Erling Smørgrav [Fri, 19 Mar 2004 20:14:23 +0000 (20:14 +0000)]
Sync with OpenBSD (two-year old bug fix)

20 years agoRemoved 3x2 dots I don't like. ;)
Ruslan Ermilov [Fri, 19 Mar 2004 17:57:07 +0000 (17:57 +0000)]
Removed 3x2 dots I don't like.  ;)

20 years agoDelete local junk in previous commit. Sorry.
David E. O'Brien [Fri, 19 Mar 2004 17:40:45 +0000 (17:40 +0000)]
Delete local junk in previous commit.  Sorry.

20 years agoAdd generic support for the recent Adaptec flavors of ServeRAID.
Scott Long [Fri, 19 Mar 2004 17:36:47 +0000 (17:36 +0000)]
Add generic support for the recent Adaptec flavors of ServeRAID.

20 years agoDepend on rev 1.40.
David E. O'Brien [Fri, 19 Mar 2004 16:36:29 +0000 (16:36 +0000)]
Depend on rev 1.40.

Submitted by: ru

20 years agoShave-off troff cycles by invoking .Fa only once.
Diomidis Spinellis [Fri, 19 Mar 2004 16:04:11 +0000 (16:04 +0000)]
Shave-off troff cycles by invoking .Fa only once.

Submitted by: ru
MFC after: 1 week

20 years agoCreate /usr/sbin in ${WORLDTMP}. I've accidentally removed it in
Ruslan Ermilov [Fri, 19 Mar 2004 15:08:58 +0000 (15:08 +0000)]
Create /usr/sbin in ${WORLDTMP}.  I've accidentally removed it in
revision 1.343, but it's needed for btxld(8), and this fix (along
with the --enable-64-bit-bfd configured BFD on i386) allows other
architectures to successfully cross-build the i386 world.

Tested on:  alpha

20 years agoWhen doing round-robin reads from a multi-plex volume, only switch to the
Lukas Ertl [Fri, 19 Mar 2004 10:28:34 +0000 (10:28 +0000)]
When doing round-robin reads from a multi-plex volume, only switch to the
next plex if the sector to be read isn't nearby the last read sector.

Submitted by:  Vsevolod Lobko <seva@ip.net.ua> via ru@
Approved by:   grog (mentor)

20 years agoDo not redundantly set the stream orientation in getc(), putc(), and
Tim J. Robbins [Fri, 19 Mar 2004 09:04:56 +0000 (09:04 +0000)]
Do not redundantly set the stream orientation in getc(), putc(), and
related functions - __sgetc() and __sputc() will set it when necessary.

20 years agoDiff reduction to NetBSD
Julian Elischer [Fri, 19 Mar 2004 08:19:52 +0000 (08:19 +0000)]
Diff reduction to NetBSD
Bring over sundry small fixes from NetBSD

Obtained from: NetBSD
MFC after: 1 week

20 years agoIsolate PCB-specific ethertalk DDP functions in ddp_pcb.c, removing them
Robert Watson [Fri, 19 Mar 2004 07:21:22 +0000 (07:21 +0000)]
Isolate PCB-specific ethertalk DDP functions in ddp_pcb.c, removing them
from ddp_usrreq.c.  Functions moved are:

  at_pcballoc()
  at_pcbconnect()
  at_pcbdetach()
  at_pcbdisconnect()
  at_pcbsetaddr()
  at_sockaddr()

Also moved are ddp_ports and ddpcb, global variables associated with DDP
pcbs.  This makes PCB implementation more parallel to inet, inet6, and
ipx.

20 years agoDiff reduction to NetBSD
Julian Elischer [Fri, 19 Mar 2004 07:14:23 +0000 (07:14 +0000)]
Diff reduction to NetBSD
Trying to figure out why this only works with SOME EHCI  controllers.

Obtained from: NetBSD
MFC after: 1 week

20 years agoMove the poweroff handler to a separate function. Make sure it is run
Nate Lawson [Fri, 19 Mar 2004 07:05:01 +0000 (07:05 +0000)]
Move the poweroff handler to a separate function.  Make sure it is run
on the boot processor (cpuid == 0).  Some chipsets do not power off the
system if the shutdown handler runs on an AP.

20 years agoRe-enable detach events after adding a bugfix from NetBSD
Julian Elischer [Fri, 19 Mar 2004 06:15:45 +0000 (06:15 +0000)]
Re-enable detach events after adding a bugfix from NetBSD
that unbreaks them.

Submitted by: dillon
Obtained from: NetBSD
MFC after: 2 days

20 years ago'vi' got away from me in rev. 1.13.
David E. O'Brien [Fri, 19 Mar 2004 03:28:38 +0000 (03:28 +0000)]
'vi' got away from me in rev. 1.13.

20 years agoForced commit to recognize repo-copy of ddb_pcb.c from ddp_usrreq.c.
Robert Watson [Fri, 19 Mar 2004 02:24:05 +0000 (02:24 +0000)]
Forced commit to recognize repo-copy of ddb_pcb.c from ddp_usrreq.c.
Thanks Peter!

20 years agoPrevent the strange situation that after each load/unload of a ppbus
Guido van Rooij [Thu, 18 Mar 2004 21:10:11 +0000 (21:10 +0000)]
Prevent the strange situation that after each load/unload of a ppbus
device, the device is probed multiple times (so each device is
detected N times after unloading/loading the module N-1 times).

The real fix is (quote Doug and Warner):
> : In an ideal world, there should be some kind of BUS_UNIDENTIFY method
> : which a driver could use to delete the devices it created in
> : BUS_IDENTIFY.
>
> Or the bus would have a driver deleted routine that got called and it
> would remove all instances of the devclass attached to it.

Reviewed by: Doug Rabson & Warner Losh

20 years agoFixed a nasty old bug where a visual bell in the currently active
Ruslan Ermilov [Thu, 18 Mar 2004 21:07:54 +0000 (21:07 +0000)]
Fixed a nasty old bug where a visual bell in the currently active
VTY prevented waking up processes waiting for the output queue to
get free on other VTYs.

In collaboration with: Vsevolod Lobko
MFC after: 1 week

20 years agoWhen mmap-ing a file from a noexec mount, be sure not to grant the right
Guido van Rooij [Thu, 18 Mar 2004 20:58:51 +0000 (20:58 +0000)]
When mmap-ing a file from a noexec mount, be sure not to grant the right
to mmap it PROT_EXEC. This also depends on the architecture, as some
architextures (e.g. i386) do not distinguish between read and exec pages

Inspired by:  http://linux.bkbits.net:8080/linux-2.4/cset@1.1267.1.85
Reviewed by: alc

20 years agoGet rid of a spurious "the".
Daniel Harris [Thu, 18 Mar 2004 20:36:14 +0000 (20:36 +0000)]
Get rid of a spurious "the".

PR: 64081
Submitted by: Chris Pepper <pepper@reppep.com>
MFC after: 1 day

20 years agoAdd tunables for disabling serialized method execution and disabling the
Nate Lawson [Thu, 18 Mar 2004 18:42:22 +0000 (18:42 +0000)]
Add tunables for disabling serialized method execution and disabling the
new _OSI method.  These can be used if these new features end up causing
regression for users.

20 years agoGrammar fix
Christian Brueffer [Thu, 18 Mar 2004 18:27:04 +0000 (18:27 +0000)]
Grammar fix

MFC after: 3 days

20 years agoUtilize sf_buf_alloc() and sf_buf_free() to implement the ephemeral
Alan Cox [Thu, 18 Mar 2004 18:23:37 +0000 (18:23 +0000)]
Utilize sf_buf_alloc() and sf_buf_free() to implement the ephemeral
mappings required by mdstart_swap().  On i386, if the ephemeral mapping
is already in the sf_buf mapping cache, a swap-backed md performs
similarly to a malloc-backed md.  Even if the ephemeral mapping is not
cached, this implementation is still faster.  On 64-bit platforms, this
change has the effect of using the direct virtual-to-physical mapping,
avoiding ephemeral mapping overheads, such as TLB shootdowns on SMPs.

On a 2.4GHz, 400MHz FSB P4 Xeon configured with 64K sf_bufs and
"mdmfs -S -o async -s 128m md /mnt"

before:
dd if=/dev/md0 of=/dev/null bs=64k
134217728 bytes transferred in 0.430923 secs (311465697 bytes/sec)

after with cold sf_buf cache:
dd if=/dev/md0 of=/dev/null bs=64k
134217728 bytes transferred in 0.367948 secs (364773576 bytes/sec)

after with warm sf_buf cache:
dd if=/dev/md0 of=/dev/null bs=64k
134217728 bytes transferred in 0.252826 secs (530870010 bytes/sec)

malloc-backed md:
dd if=/dev/md0 of=/dev/null bs=64k
134217728 bytes transferred in 0.253126 secs (530240978 bytes/sec)

20 years agoAdd the missing Giant when doing anything with VFS -- in this case,
Brian Feldman [Thu, 18 Mar 2004 18:15:58 +0000 (18:15 +0000)]
Add the missing Giant when doing anything with VFS -- in this case,
releasing the ktrace vnode.

20 years agoFor AMD64, lets officially support a -O2 kernel build!
David E. O'Brien [Thu, 18 Mar 2004 18:05:08 +0000 (18:05 +0000)]
For AMD64, lets officially support a -O2 kernel build!
I've added -fno-strict-aliasing for now so we can ease into this.
I wanted to shoot for -O3, but the inlining caused problems due to GCC's
size heuristics; so also add -frename-registers, which is one of the things
-O3 would have given us.

20 years agoNew release note:
Hiroki Sato [Thu, 18 Mar 2004 17:51:55 +0000 (17:51 +0000)]
New release note:
Intel ACPI-CA 20040311 import.

20 years agoCheck in files off the vendor branch as well as files with local patches.
Nate Lawson [Thu, 18 Mar 2004 17:46:27 +0000 (17:46 +0000)]
Check in files off the vendor branch as well as files with local patches.

20 years agoThis commit was generated by cvs2svn to compensate for changes in r127175,
Nate Lawson [Thu, 18 Mar 2004 17:42:14 +0000 (17:42 +0000)]
This commit was generated by cvs2svn to compensate for changes in r127175,
which included commits to RCS files with non-trunk default branches.

20 years agoImport of Intel ACPI-CA 20040311.
Nate Lawson [Thu, 18 Mar 2004 17:42:14 +0000 (17:42 +0000)]
Import of Intel ACPI-CA 20040311.

20 years agoNew release note:
Hiroki Sato [Thu, 18 Mar 2004 16:39:19 +0000 (16:39 +0000)]
New release note:
SA-04:05.openssl
/etc/rc.d/mixer
/etc/rc.d/gbde_swap (wording fix)

20 years agoVerify more bits of the ELF header: the program header table
Jacques Vidrine [Thu, 18 Mar 2004 16:33:05 +0000 (16:33 +0000)]
Verify more bits of the ELF header: the program header table
entry size and the ELF version.  Also, avoid a potential integer
overflow when determining whether the ELF header fits entirely
within the first page.

Reviewed by: jdp

A panic when attempting to execute an ELF binary with a bogus program
header table entry size was

Reported by: Christer Öberg <christer.oberg@texonet.com>

20 years agoEcho and pass dumpdev as the device argument to savecore.
Chad David [Thu, 18 Mar 2004 15:18:20 +0000 (15:18 +0000)]
Echo and pass dumpdev as the device argument to savecore.

PR: bin/51655

20 years agoCorrect a bug introduced with the recent clone API chang: when the clone
Robert Watson [Thu, 18 Mar 2004 14:18:51 +0000 (14:18 +0000)]
Correct a bug introduced with the recent clone API chang: when the clone
event handler for if_tap fails, make sure to clean up clone state to
prevent a clone memory leak.

20 years agoMake the vaule of PTT_RELAY_SID match the RFC. This should help PPPoE
David Malone [Thu, 18 Mar 2004 12:34:14 +0000 (12:34 +0000)]
Make the vaule of PTT_RELAY_SID match the RFC. This should help PPPoE
users that are working with relayed PPPoE.

Submitted by: Bodo Rüskamp <jordbaer@mac.com>
PR: 44936
Approved by: julian
MFC after: 1 week

20 years agos/enable/enables/ in a comment
Christian Brueffer [Thu, 18 Mar 2004 12:22:31 +0000 (12:22 +0000)]
s/enable/enables/ in a comment

20 years agoInstall /bin/sh safely. This allows a shell script to be used
Ruslan Ermilov [Thu, 18 Mar 2004 10:55:47 +0000 (10:55 +0000)]
Install /bin/sh safely.  This allows a shell script to be used
to strip binaries by specifying it in the STRIPBIN environment
variable honoured by install(1).

MFC after: 3 days

20 years agoComment various stages of buildkernel, symmetrical to buildworld.
Ruslan Ermilov [Thu, 18 Mar 2004 10:17:03 +0000 (10:17 +0000)]
Comment various stages of buildkernel, symmetrical to buildworld.

20 years agosAdd a comment indicating why there continues to be a race condition in
Robert Watson [Thu, 18 Mar 2004 09:55:11 +0000 (09:55 +0000)]
sAdd a comment indicating why there continues to be a race condition in
the tap driver, even with Giant over the cdev operation vector, due to
a non-atomic test-and-set of the si_drv1 field in the dev_t.  This bug
exists with Giant under high memory pressure, as malloc() may sleep
in tapcreate(), but is less likely to occur.  The resolution will
probably be to cover si_drv1 using the global tapmtx since no softc is
available, but I need to think about this problem more generally
across a range of drivers using si_drv1 in combination with SI_CHEAPCLONE
to defer expensive allocation to open().

Correct what appears to be a bug in the original if_tap implementation,
in which tapopen() will panic if a tap device instance is opened more
than once due to an incorrect assertion -- only triggered if INVARIANTS
is compiled in (i.e., when built into a kernel).  Return EBUSY instead.

Expand mtx_lock() coverage using tp->tap_mtx to include tp->ether_addr.

20 years agoGenerate opt_inet.h.
Max Khon [Thu, 18 Mar 2004 09:45:17 +0000 (09:45 +0000)]
Generate opt_inet.h.
Add opt_inet.h to SRCS.

20 years agoMore cleanups:
Ruslan Ermilov [Thu, 18 Mar 2004 07:21:31 +0000 (07:21 +0000)]
More cleanups:

- 0 should have been -1 in previous commit (just to stay consistent),
- Spell null pointers as NULL, not 0,
- Fixed the comment about pr_usesysctl to not confuse it with boolean.

Pointed by: bde

20 years agoMove "is consumer attached?" check before G_VALID_PROVIDER() check,
Pawel Jakub Dawidek [Thu, 18 Mar 2004 07:17:10 +0000 (07:17 +0000)]
Move "is consumer attached?" check before G_VALID_PROVIDER() check,
because if consumer is not attached, its provider never will be valid,
so we never reach this check.

Approved by: phk

20 years agoConfigure as --enable-64-bit-bfd.
David E. O'Brien [Thu, 18 Mar 2004 04:45:29 +0000 (04:45 +0000)]
Configure as --enable-64-bit-bfd.

20 years agoBraino in previous commit.
Dag-Erling Smørgrav [Thu, 18 Mar 2004 03:01:28 +0000 (03:01 +0000)]
Braino in previous commit.

20 years agoRemoved a vestige of the stl driver.
Bruce Evans [Thu, 18 Mar 2004 02:54:42 +0000 (02:54 +0000)]
Removed a vestige of the stl driver.

20 years agoDocument machdep.hlt_cpus.
David E. O'Brien [Thu, 18 Mar 2004 02:53:38 +0000 (02:53 +0000)]
Document machdep.hlt_cpus.

Submitted by: Craig Rodrigues <rodrigc@crodrigues.org>

20 years agoBack out code for auto-gdb detection that accidentally leaked into the
Nate Lawson [Thu, 18 Mar 2004 02:36:41 +0000 (02:36 +0000)]
Back out code for auto-gdb detection that accidentally leaked into the
bus_alloc_resource_any commit.

Submitted by: bde
Pointy-hat: njl

20 years agoSupport the DPF (start dependent function) resource type in parsing _PRS.
Nate Lawson [Thu, 18 Mar 2004 02:33:58 +0000 (02:33 +0000)]
Support the DPF (start dependent function) resource type in parsing _PRS.
This should fix this error people get attaching cardbus controllers:

    pcib0: _PRS resource entry has unsupported type 2

20 years agoFix 'ps -p proclist' and 'ps -u userlist' so the command returns non-zero
Garance A Drosehn [Thu, 18 Mar 2004 01:28:23 +0000 (01:28 +0000)]
Fix 'ps -p proclist' and 'ps -u userlist' so the command returns non-zero
if no processes were matched.  Also sorts the list of 'int's in main, as
long as I had to add another one...

Noticed by: Nate Lawson
MFC after: 10 days

20 years agoRegen after 1.169 of usbdevs.
Maxim Sobolev [Thu, 18 Mar 2004 01:06:28 +0000 (01:06 +0000)]
Regen after 1.169 of usbdevs.

20 years agoAdd support for Crystalfontz CFA-632, CFA-633 and CFA-634, all of them
Maxim Sobolev [Thu, 18 Mar 2004 01:02:46 +0000 (01:02 +0000)]
Add support for Crystalfontz CFA-632, CFA-633 and CFA-634, all of them
are based on the same USB->COM bridge, but have different product IDs.

PR:
Submitted by:
Reviewed by:
Approved by:
Obtained from:  http://www.tnpi.biz/computing/freebsd/crystalfontz.shtml
MFC after:      3 days

20 years agoMention the user account for 'pf' is "proxy".
David E. O'Brien [Thu, 18 Mar 2004 00:50:40 +0000 (00:50 +0000)]
Mention the user account for 'pf' is "proxy".

20 years agoCleanup hints, given that no hammer machine have (nor ever will have)
David E. O'Brien [Thu, 18 Mar 2004 00:18:45 +0000 (00:18 +0000)]
Cleanup hints, given that no hammer machine have (nor ever will have)
ISA slots.

Submitted by: Peter

20 years agoRevise socow_iodone() in light of recent sf_buf changes. Specifically,
Alan Cox [Wed, 17 Mar 2004 23:25:04 +0000 (23:25 +0000)]
Revise socow_iodone() in light of recent sf_buf changes.  Specifically,
use sf_buf_free() instead of sf_buf_mext() to consolidate all actions
that require the page queues lock in one critical section.  While I'm
here remove unnecessary splvm() and splx() calls.

20 years agoImprovements to 'ps -p <x>'. If <x> is a process id that does not exist, then
Garance A Drosehn [Wed, 17 Mar 2004 22:46:58 +0000 (22:46 +0000)]
Improvements to 'ps -p <x>'.  If <x> is a process id that does not exist, then
just print the header (if any) and exit, thus matching the behavior on -stable
and other OS's.

Also adds support for <x> being a comma-separated list of processes, and does
a much better checking for invalid-values of <x>, such as 'ps -p someword'.

Reviewed by: mentioned on freebsd-current
MFC after: 10 days

20 years agoEliminate bogus usage of WI_RSSI_TO_DBM(). Not only does it bogusly
Brian Feldman [Wed, 17 Mar 2004 21:54:52 +0000 (21:54 +0000)]
Eliminate bogus usage of WI_RSSI_TO_DBM().  Not only does it bogusly
clip/destroy the dB value contained in the wi(4)'s receive frames,
it doesn't match with the flag set in the radiotap header
(unperturbed dB versus dBm).

20 years agoFix border error to allow systems that specify 100 for latency also use
Nate Lawson [Wed, 17 Mar 2004 21:49:47 +0000 (21:49 +0000)]
Fix border error to allow systems that specify 100 for latency also use
C2 and 1000 to use C3.

Submitted by: Bruno Ducrot <ducrot@poupinou.org>
Tested by: Scott Lambert <lambert@lambertfam.org>

20 years agosync comment with i386's isa.c.. This removes a comment that is YEARS
John-Mark Gurney [Wed, 17 Mar 2004 21:45:55 +0000 (21:45 +0000)]
sync comment with i386's isa.c..  This removes a comment that is YEARS
old...

20 years agoStyle(9) round for the pf kernel parts. Mostly #if defined() -> #ifdef
Max Laier [Wed, 17 Mar 2004 21:11:02 +0000 (21:11 +0000)]
Style(9) round for the pf kernel parts. Mostly #if defined() -> #ifdef

Also set HOOK_HACK to true (remove the related #ifdef's) as we have the
hooks in the kernel this was missed during the merge from the port.

Noticed by: Amir S. (for the HOOK_HACK part)
Approved by: bms(mentor)

20 years agoAdd a comment with an explanation why we don't report EPIPE errors on
Pawel Jakub Dawidek [Wed, 17 Mar 2004 21:10:20 +0000 (21:10 +0000)]
Add a comment with an explanation why we don't report EPIPE errors on
nfs sockets.

Requested by: ru

20 years agoNULL -> 0.
Ruslan Ermilov [Wed, 17 Mar 2004 20:50:25 +0000 (20:50 +0000)]
NULL -> 0.

20 years agoNULL -> 0.
Ruslan Ermilov [Wed, 17 Mar 2004 20:19:45 +0000 (20:19 +0000)]
NULL -> 0.