]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
17 years agoHook up gjournal bits to the build.
pjd [Tue, 31 Oct 2006 22:22:30 +0000 (22:22 +0000)]
Hook up gjournal bits to the build.

Sponsored by: home.pl

17 years agoAdd Makefile for geom_journal kernel module.
pjd [Tue, 31 Oct 2006 22:21:47 +0000 (22:21 +0000)]
Add Makefile for geom_journal kernel module.

Sponsored by: home.pl

17 years agoForgot to add file with gjournal specific fsck code.
pjd [Tue, 31 Oct 2006 22:21:01 +0000 (22:21 +0000)]
Forgot to add file with gjournal specific fsck code.

Sponsored by: home.pl

17 years agoAdd fs_unrefs field to the super block structure.
pjd [Tue, 31 Oct 2006 22:18:33 +0000 (22:18 +0000)]
Add fs_unrefs field to the super block structure.

Sponsored by: home.pl

17 years agoAdd recently added fs_unrefs field.
pjd [Tue, 31 Oct 2006 22:15:59 +0000 (22:15 +0000)]
Add recently added fs_unrefs field.

Sponsored by: home.pl

17 years agoUpdate after function renames.
pjd [Tue, 31 Oct 2006 22:07:29 +0000 (22:07 +0000)]
Update after function renames.

Sponsored by: home.pl

17 years agoImplements gjournal support. If file system has gjournal support enabled
pjd [Tue, 31 Oct 2006 22:06:56 +0000 (22:06 +0000)]
Implements gjournal support. If file system has gjournal support enabled
and -p flag was given perform fast file system checking (bascially only
garbage collecting of orphaned objects).

Rename bread() to blread() and bwrite() to blwrite() as we now link to
the libufs library, which also implement functions with that names.

Sponsored by: home.pl

17 years agoTeach about new fields (cg_unrefs and fs_unrefs) and new FS_GJOURNAL flag.
pjd [Tue, 31 Oct 2006 22:02:24 +0000 (22:02 +0000)]
Teach about new fields (cg_unrefs and fs_unrefs) and new FS_GJOURNAL flag.

Sponsored by: home.pl

17 years agoTeach mount(8) about MNT_GJOURNAL flag.
pjd [Tue, 31 Oct 2006 21:54:51 +0000 (21:54 +0000)]
Teach mount(8) about MNT_GJOURNAL flag.
MNT_GJOURNAL flag is not a mount-time flag, but it is needed to show
'gjournal' option in mount(8) output.

Sponsored by: home.pl

17 years agoAdd -J flag to both newfs(8) and tunefs(8) which allows to enable gjournal
pjd [Tue, 31 Oct 2006 21:52:28 +0000 (21:52 +0000)]
Add -J flag to both newfs(8) and tunefs(8) which allows to enable gjournal
support.
I left -j flag for UFS journal implementation which we may gain at some
point.

Sponsored by: home.pl

17 years agoAdd gjournal specific code to the UFS file system:
pjd [Tue, 31 Oct 2006 21:48:54 +0000 (21:48 +0000)]
Add gjournal specific code to the UFS file system:
- Add FS_GJOURNAL flag which enables gjournal support on a file system.
- Add cg_unrefs field to the cylinder group structure which holds
  number of unreferenced (orphaned) inodes in the given cylinder group.
- Add fs_unrefs field to the super block structure which holds
  total number of unreferenced (orphaned) inodes.
- When file or a directory is orphaned (last reference is removed, but
  object is still open), increase fs_unrefs and cg_unrefs fields,
  which is a hint for fsck in which cylinder groups looks for such
  (orphaned) objects.
- When file is last closed, decrease {fs,cg}_unrefs fields.
- Add VV_DELETED vnode flag which points at orphaned objects.

Sponsored by: home.pl

17 years agoFix buildworld.
glebius [Tue, 31 Oct 2006 21:44:46 +0000 (21:44 +0000)]
Fix buildworld.

17 years agoAdd MNT_GJOURNAL flag which indicates, that file system has gjournal
pjd [Tue, 31 Oct 2006 21:38:51 +0000 (21:38 +0000)]
Add MNT_GJOURNAL flag which indicates, that file system has gjournal
support enabled.
Add mnt_gjprovider field which keeps gjournal provider's name on which
file system is placed on. This allows to not place file system on gjournal
directly and allows gjournal class to pair gjournal provider with file
system.

Sponsored by: home.pl

17 years agoAdd userland control utility for gjournal GEOM class.
pjd [Tue, 31 Oct 2006 21:32:54 +0000 (21:32 +0000)]
Add userland control utility for gjournal GEOM class.

Sponsored by: home.pl

17 years agoAdd gjournal GEOM class (kernel side), which implements block level
pjd [Tue, 31 Oct 2006 21:31:00 +0000 (21:31 +0000)]
Add gjournal GEOM class (kernel side), which implements block level
journaling and can be tought about marking file system as clean before
doing journal switch, which easly allows to add journaling to file
systems that don't have this feature.

Sponsored by: home.pl

17 years agoImplement BIO_FLUSH handling by simply passing it down to the components.
pjd [Tue, 31 Oct 2006 21:23:51 +0000 (21:23 +0000)]
Implement BIO_FLUSH handling by simply passing it down to the components.

Sponsored by: home.pl

17 years agoImplement cgwrite1(3) function which stored a given cylinder group on disk.
pjd [Tue, 31 Oct 2006 21:21:48 +0000 (21:21 +0000)]
Implement cgwrite1(3) function which stored a given cylinder group on disk.

Sponsored by: home.pl

17 years agoImplement BIO_FLUSH handling for da(4), amr(4), ata(4) and ataraid(4).
pjd [Tue, 31 Oct 2006 21:19:25 +0000 (21:19 +0000)]
Implement BIO_FLUSH handling for da(4), amr(4), ata(4) and ataraid(4).

Sponsored by: home.pl

17 years agoAdd a new disk flag - DISKFLAG_CANFLUSHCACHE, which indicates that the disk
pjd [Tue, 31 Oct 2006 21:12:43 +0000 (21:12 +0000)]
Add a new disk flag - DISKFLAG_CANFLUSHCACHE, which indicates that the disk
can handle BIO_FLUSH requests.

Sponsored by: home.pl

17 years agoAdd a new I/O request - BIO_FLUSH, which basically tells providers below to
pjd [Tue, 31 Oct 2006 21:11:21 +0000 (21:11 +0000)]
Add a new I/O request - BIO_FLUSH, which basically tells providers below to
flush their caches. For now will mostly be used by disks to flush their
write cache.

Sponsored by: home.pl

17 years agoremove gdbserver on ARCH i386 as we now install it again..
jmg [Tue, 31 Oct 2006 21:05:35 +0000 (21:05 +0000)]
remove gdbserver on ARCH i386 as we now install it again..

Submitted by: Pawel Worach

17 years agoMake EWOULDBLOCK a recoverable error so that the request is retransmitted.
mohans [Tue, 31 Oct 2006 20:25:37 +0000 (20:25 +0000)]
Make EWOULDBLOCK a recoverable error so that the request is retransmitted.
This bug results in data corruption with NFS/TCP. Writes are silently dropped
on EWOULDBLOCK (because socket send buffer is full and sockbuf timer fires).

Reviewed by: ups@

17 years agoFix a typo resulting in truncated linux32 signal trampoline code copied
kib [Tue, 31 Oct 2006 17:53:02 +0000 (17:53 +0000)]
Fix a typo resulting in truncated linux32 signal trampoline code copied
to the usermode. Usually, signal handler segfaulted on return.

Reviewed by: jhb
MFC after: 3 days

17 years agoAllocate receive and transmit data structures during attach() and free them
jhb [Tue, 31 Oct 2006 17:21:14 +0000 (17:21 +0000)]
Allocate receive and transmit data structures during attach() and free them
during detach() similar to other NIC drivers rather than allocating them
during init() and freeing them during stop():
- Move creation of tx bus_dma tag amd maps and tx_buffer_area from
  em_setup_transmit_structures() to em_allocate_transmit_structures().
- Call em_allocate_xxx_structures() in em_attach().
- Only call em_free_xxx_structures() in em_detach().
- Change em_setup_xxx_structures() to free any existing tx or rx buffers
  and in the case of rx repopulate the ring with newer buffers.

Reviewed by: jfv

17 years ago- Use callout_init_mtx() to close various callout-related races.
jhb [Tue, 31 Oct 2006 17:05:02 +0000 (17:05 +0000)]
- Use callout_init_mtx() to close various callout-related races.
- Drain the two timers in detach.
- Check IFF_DRV_RUNNING in the link task and bail w/o doing anything if
  it is clear.

Reviewed by: jfv, scottl

17 years agoRework the transmit register handling. In em_encap() store index of
glebius [Tue, 31 Oct 2006 16:19:21 +0000 (16:19 +0000)]
Rework the transmit register handling. In em_encap() store index of
the EOP descriptor in the first descriptor of the packet. And then
in em_txeof() search for DD bits set only in the EOP descriptors,
embedding the cleanup of all packet's descriptors into inner loop.

This change is important for future chips, where DD bit is going
to be set only on the EOP descriptors.

Submitted by: jfv

17 years agoRemove unused softc pointer variable from the probe routine.
jhb [Tue, 31 Oct 2006 16:13:38 +0000 (16:13 +0000)]
Remove unused softc pointer variable from the probe routine.

CID: 1560
Found by: Coverity Prevent (tm)

17 years agoMerge new vendor release - 6.2.9.
glebius [Tue, 31 Oct 2006 15:00:14 +0000 (15:00 +0000)]
Merge new vendor release - 6.2.9.

Details:
o if_em.c changes:
  - Added several new PCI ids.
  - Check em_check_phy_reset_block() before doing SIOCSIFMEDIA ioctl.
  - Don't touch TARC registers, they are now handled in shared
    code in if_em_hw.c.
  - Move RDH and RDT setting to the end of
    em_initialize_receive_unit().
  - Declare em_read_pcie_cap_reg(), now empty.
o if_em_hw.c dropped in from vendor, then restored rev. 1.15.
o if_em_hw.h dropped in from vendor, then modified:
  - Added RX overrun interrupt flag to interrupt enable mask.
  - Remove declarations of em_io_read(), em_io_write().

Approved by: jfv

17 years agoBind to INADDR_ANY in the default configuration. This makes bsnmpd(1)
harti [Tue, 31 Oct 2006 10:23:28 +0000 (10:23 +0000)]
Bind to INADDR_ANY in the default configuration. This makes bsnmpd(1)
automatically work on multi-homed hosts and without explicite specification
of the hostname in the config file.

Submitted by: jmg

17 years agoDefine a base OID for the FreeBSD version as returned in sysObjectID
harti [Tue, 31 Oct 2006 10:09:10 +0000 (10:09 +0000)]
Define a base OID for the FreeBSD version as returned in sysObjectID
by bsnmpd(1). The actual OID is formed by appending the release numbers
to this base OID.

Idea by: phk

17 years agoThis commit was generated by cvs2svn to compensate for changes in r163820,
harti [Tue, 31 Oct 2006 09:00:35 +0000 (09:00 +0000)]
This commit was generated by cvs2svn to compensate for changes in r163820,
which included commits to RCS files with non-trunk default branches.

17 years agoVendor patch: synthesize the initial value for sysObjectId from the value
harti [Tue, 31 Oct 2006 09:00:35 +0000 (09:00 +0000)]
Vendor patch: synthesize the initial value for sysObjectId from the value
of uname -r in FreeBSD. This value can be overwritten in the configuration
file.

Suggested by: phk

17 years agoAvoid implict node generation in oids, people tell me a fair bit of
phk [Tue, 31 Oct 2006 08:18:02 +0000 (08:18 +0000)]
Avoid implict node generation in oids, people tell me a fair bit of
SNMP software has trouble with them.

17 years agoClaim my own subtree under freeBSDpeople
phk [Tue, 31 Oct 2006 07:44:28 +0000 (07:44 +0000)]
Claim my own subtree under freeBSDpeople

17 years agoFix Typo.
takawata [Tue, 31 Oct 2006 07:22:24 +0000 (07:22 +0000)]
Fix Typo.

Pointed out by: ru

17 years agoThe first of 3 major steps to move the CAM layer forward to using
mjacob [Tue, 31 Oct 2006 05:53:29 +0000 (05:53 +0000)]
The first of 3 major steps to move the CAM layer forward to using
the CAM_NEW_TRAN_CODE that has been in the tree for some years now.

This first step consists solely of adding to or correcting
CAM_NEW_TRAN_CODE pieces in the kernel source tree such
that a both a GENERIC (at least on i386) and a LINT build
with CAM_NEW_TRAN_CODE as an option will compile correctly
and run (at least with some the h/w I have).

After a short settle time, the other pieces (making
CAM_NEW_TRAN_CODE the default and updating libcam
and camcontrol) will be brought in.

This will be an incompatible change in that the size of structures
related to XPT_PATH_INQ and XPT_{GET,SET}_TRAN_SETTINGS change
in both size and content. However, basic system operation and
basic system utilities work well enough with this change.

Reviewed by: freebsd-scsi and specific stakeholders

17 years agoFix a typo in the device ID table that prevented 5708S chips from being
scottl [Tue, 31 Oct 2006 03:28:25 +0000 (03:28 +0000)]
Fix a typo in the device ID table that prevented 5708S chips from being
detected.

Submitted by: pyun

17 years agoAdd support for Option GT 3G/3G quad datacard in ubsa.
kevlo [Tue, 31 Oct 2006 02:27:24 +0000 (02:27 +0000)]
Add support for Option GT 3G/3G quad datacard in ubsa.

Approved by: cognet

17 years agoCorrect a security issue introduced in previous commit:
delphij [Tue, 31 Oct 2006 02:22:36 +0000 (02:22 +0000)]
Correct a security issue introduced in previous commit:
instead of removing the file and issue a warning about
the removal, do not do any operation at all in case -P
is specified when the dinode has hard links.

With -f and -P specified together, we assume that the
user wants rm to overwrite the contents of the file
and remove it (destroy the contents of file but leave
its hard links as is).

The reason of doing it this way is that, in case where
a hard link is created by a malicious user (currently
this is permitted even if the user has no access to the
file).  Losing the link can potentially mean that the
actual owner would lose control completely to the user
who wants to obtain access in a future day.

Discussed with: Peter Jermey

17 years ago- Add a 'verbose' switch -v
markus [Tue, 31 Oct 2006 00:26:58 +0000 (00:26 +0000)]
- Add a 'verbose' switch -v
- Only dump items that are being used for padding when being verbose. This
  brings bthidcontrol in line with the behaviour of usbhidctl(1).
- Update the manpage accordingly

Approved by: emax

17 years agoInclude <sys/types.h>, to get definition for uint32_t.
cognet [Mon, 30 Oct 2006 23:23:00 +0000 (23:23 +0000)]
Include <sys/types.h>, to get definition for uint32_t.

Submitted by: David Sharp

17 years agoAssign start to the value we were able to allocate and use that to
imp [Mon, 30 Oct 2006 22:46:33 +0000 (22:46 +0000)]
Assign start to the value we were able to allocate and use that to
write out the BAR.  Otherwise, we were trying to shift a 32-bit
quantity on 32-bit platforms.  Also, 'start' check sanity to where it
is known.

17 years agoIn the replacement text of the __bswapN_const() macros encapsulate the
marius [Mon, 30 Oct 2006 21:50:11 +0000 (21:50 +0000)]
In the replacement text of the __bswapN_const() macros encapsulate the
argument in parentheses so these macros are safe to use and invocations
with an expression as the argument like __bswap32_const(42 << 23 | 13)
work as expected. Additionally, mask all the individually shifted bytes
as appropriate so the bytes which exceed the width of the respective
__bswapN_const() macro in invocations like __bswap16_const(0xdead600d)
are ignored like it's the case with the corresponding __bswapN_var()
function.

MFC after: 3 days

17 years agoAdd configuration stubs for adding package derived files to the various
julian [Mon, 30 Oct 2006 19:50:01 +0000 (19:50 +0000)]
Add configuration stubs for adding package derived files to the various
sample configurations.
Submitted by Jeremie Le Hen and tested by Jean Milanez Melo.

17 years agoAdd some code to support including files ffrom packages in the image.
julian [Mon, 30 Oct 2006 19:47:05 +0000 (19:47 +0000)]
Add some code to support including files ffrom packages in the image.

Submitted by: Jeremie Le Hen and tested by Jean Milanez Melo.

17 years agoMore fully support 64-bit bars. Prior to this commit, we supported
imp [Mon, 30 Oct 2006 19:18:46 +0000 (19:18 +0000)]
More fully support 64-bit bars.  Prior to this commit, we supported
only those bars that had addresses assigned by the BIOS and where the
bridges were properly programmed.  Now even unprogrammed ones work.
This was needed for sun4v.  We still only implement up to 2GB memory
ranges, even for 64-bit bars.  PCI standards at least through 2.2 say
that this is the max (or 1GB is, I only know it is < 32bits).

o Always define pci_addr_t as uint64_t.  A pci address is always 64-bits,
  but some hosts can't address all of them.
o Preserve the upper half of the 64-bit word during resource probing.
o Test to make sure that 64-bit values can fit in a u_long (true on some
  platforms, but not others).  Don't use those that can't.
o minor pedantry about data sizes.
o Better bridge resource reporting in bootverbose case.
o Minor formatting changes to cope with different data types on different
  platforms.

Submitted by: jmg, with many changes by me to fully support 64-bit
addresses.

17 years ago- Handle timeouts from recv(2) properly.
pjd [Mon, 30 Oct 2006 18:29:24 +0000 (18:29 +0000)]
- Handle timeouts from recv(2) properly.
- Increase timeout to 8 seconds (should be made configurable).

Reported by: Ulrich Spoerlein <uspoerlein@gmail.com>
Reported by: Christian Laursen <xi@borderworlds.dk>
PR: kern/104829
MFC after: 1 week

17 years agoVendor patch: improve readability by using the IF_Mbps macro.
harti [Mon, 30 Oct 2006 16:56:38 +0000 (16:56 +0000)]
Vendor patch: improve readability by using the IF_Mbps macro.

Submitted by: glebius

17 years agoThis commit was generated by cvs2svn to compensate for changes in r163799,
harti [Mon, 30 Oct 2006 16:56:38 +0000 (16:56 +0000)]
This commit was generated by cvs2svn to compensate for changes in r163799,
which included commits to RCS files with non-trunk default branches.

17 years agoBelatedly bump manpage date for revision 1.35
keramida [Mon, 30 Oct 2006 16:43:42 +0000 (16:43 +0000)]
Belatedly bump manpage date for revision 1.35

Reminded by: ru

17 years agoForward declare struct cdev, since arguments of this type are used in
rwatson [Mon, 30 Oct 2006 15:20:49 +0000 (15:20 +0000)]
Forward declare struct cdev, since arguments of this type are used in
function prototypes.

Obtained from: TrustedBSD Project

17 years agoMention that we do not support route labels in the BUGS section.
mlaier [Mon, 30 Oct 2006 15:15:37 +0000 (15:15 +0000)]
Mention that we do not support route labels in the BUGS section.

PR: docs/93590
Reported by: Niki Denev

17 years agoThe first (optional) argument of the "trace" command is either the
ru [Mon, 30 Oct 2006 12:55:06 +0000 (12:55 +0000)]
The first (optional) argument of the "trace" command is either the
process or thread ID.

PR: docs/61859

17 years agoSquash a typo.
keramida [Mon, 30 Oct 2006 12:01:57 +0000 (12:01 +0000)]
Squash a typo.

Submitted by: Miltiadis Margaronis <mmargaron@gmail.com>

17 years agoHook acpi_aiboost module to build.
takawata [Mon, 30 Oct 2006 06:05:10 +0000 (06:05 +0000)]
Hook acpi_aiboost module to build.

17 years agoAdd conf file entries for acpi_aiboost drivers.
takawata [Mon, 30 Oct 2006 05:51:54 +0000 (05:51 +0000)]
Add conf file entries for acpi_aiboost drivers.

17 years agoDriver for some ASUS desktop motherboard extras.
takawata [Mon, 30 Oct 2006 05:38:48 +0000 (05:38 +0000)]
Driver for some ASUS desktop motherboard extras.
Though it is named after overclocking tool for ASUS motherboards,
it is not capable to change clock ratio or CPU core voltage.

This driver exports Templature, Power output voltage, Fan RPM under
dev.acpi_aiboost.0.*.

Descriptions for these values are set to sysctl describe, which can be
get by sysctl -d.

17 years agoBe more reasonable when overwrite mode is specified while there
delphij [Mon, 30 Oct 2006 03:32:09 +0000 (03:32 +0000)]
Be more reasonable when overwrite mode is specified while there
is hard links.  Overwritting when links > 1 would cause data
loss, which is usually undesired.

Inspired by: discussion on -hackers@
Suggested by: elessar at bsdforen de
Obtained from: OpenBSD

17 years agoForced commit to denote that the third item of the previous commit
marius [Mon, 30 Oct 2006 01:58:50 +0000 (01:58 +0000)]
Forced commit to denote that the third item of the previous commit
message should have read:

- Remove the hw.dc_quick SYSCTL, which allowed to turn off the above
  mentioned optimization, as like the equivalent and already removed
  hw.sis_quick it existed for testing purposes only.

17 years agoAdd the padding fields to 'struct proc' for the !KSE case that I missed.
jb [Sun, 29 Oct 2006 21:02:22 +0000 (21:02 +0000)]
Add the padding fields to 'struct proc' for the !KSE case that I missed.

Noticed by: pjd

17 years ago- Wrap code optimized for architectures without alignment constraints
marius [Sun, 29 Oct 2006 20:24:27 +0000 (20:24 +0000)]
- Wrap code optimized for architectures without alignment constraints
  in #ifdef __NO_STRICT_ALIGNMENT rather than #ifdef __i386__. This
  means that amd64 now also uses the optimized code. [1]
  While at it, fix a nearby style(9) bug.
- Remove the hw.dc_quick SYSCTL, which allowed to turn off the above
  mentioned optimization, as like the equivalent and already removed
- In dc_setcfg() suppress printing a warning when forcing the receiver
  and transceiver to idle state times out for chips where the status
  bits in question just never change (observed in detail with DM9102A)
  and therefore the warning would be highly likely false positive. [2]
- In dc_ifmedia_sts() add a missing DC_UNLOCK().

Tested by: Hans-Joerg Sirtl on amd64 [1]
PR: 82681 [2]
Obtained from: NetBSD tlp(4) [2]
MFC after: 1 week

17 years agoWrap code optimized for architectures without alignment constraints
marius [Sun, 29 Oct 2006 20:19:41 +0000 (20:19 +0000)]
Wrap code optimized for architectures without alignment constraints
in #ifdef __NO_STRICT_ALIGNMENT rather than #if defined(__i386__) ||
defined(__amd64__). Currently this change is cosmetic only though.
While at it, fix a nearby style(9) bug and remove a no longer used
header.

17 years agoRetire the failed OBSOLETE_IN_6 experiment code.
imp [Sun, 29 Oct 2006 20:03:55 +0000 (20:03 +0000)]
Retire the failed OBSOLETE_IN_6 experiment code.

17 years agoAdd missing headers.
marius [Sun, 29 Oct 2006 17:23:54 +0000 (17:23 +0000)]
Add missing headers.

17 years agoBecause the BTX mini-kernel now uses flat memory mode and clients
ru [Sun, 29 Oct 2006 14:50:58 +0000 (14:50 +0000)]
Because the BTX mini-kernel now uses flat memory mode and clients
are no longer limited to a virtual address space of 16 megabytes,
only mask high two bits of a virtual address.  This allows to load
larger kernels (up to 1 gigabyte).  Not masking addresses at all
was a bad idea on machines with less than >3G of memory -- kernels
are linked at 0xc0xxxxxx, and that would attempt to load a kernel
at above 3G.  By masking only two highest bits we stay within the
safe limits while still allowing to boot larger kernels.

(This is a safer reimplmentation of sys/boot/i386/boot2/boot.2.c
rev. 1.71.)

Prodded by: jhb
Tested by: nyan (pc98)

17 years agoregen after linux_io_* backout
netchild [Sun, 29 Oct 2006 14:12:44 +0000 (14:12 +0000)]
regen after linux_io_* backout

17 years agoBackout the linux aio stuff. Several problems where identified and the
netchild [Sun, 29 Oct 2006 14:02:39 +0000 (14:02 +0000)]
Backout the linux aio stuff. Several problems where identified and the
dynamic nature (if no native aio code is available, the linux part
returns ENOSYS because of missing requisites) should be solved differently
than it is.

All this will be done in P4.

Not included in this commit is a backout of the changes to the native aio
code (removing static in some places). Those changes (and some more) will
also be needed when the reworked linux aio stuff will reenter the tree.

Requested by: rwatson
Discussed with: rwatson

17 years agoDo not try to rtsol on pflog or pfsync devices.
mlaier [Sun, 29 Oct 2006 13:29:49 +0000 (13:29 +0000)]
Do not try to rtsol on pflog or pfsync devices.

17 years ago- Use non-recursive mutex. MTX_RECURSE is unnecessary since rev. 1.70
oleg [Sun, 29 Oct 2006 12:09:24 +0000 (12:09 +0000)]
- Use non-recursive mutex. MTX_RECURSE is unnecessary since rev. 1.70
- Pay respect to net.isr.direct: use netisr_dispatch() instead of ip_input()

Reviewed by: glebius, rwatson

- purge_flow_set():
    - Do not leak memory while purging queues which are not bound to pipe.
    - style(9) cleanup

MFC after: 2 months

17 years agostyle(9)
netchild [Sun, 29 Oct 2006 09:50:55 +0000 (09:50 +0000)]
style(9)

Noticed by: rwatson

17 years agoRemoved some SMP ifdefs so that using the TSC as a cputime clock is
bde [Sun, 29 Oct 2006 09:48:44 +0000 (09:48 +0000)]
Removed some SMP ifdefs so that using the TSC as a cputime clock is
not completely decided at config time.  Just don't default to using
the TSC if there are multiple active CPUs.  Also, don't default to
using the TSC if it is broken.  SMP ifdefs are still used to disallow
using perfmon since perfmon is always broken if SMP is just configured.

This only helps much for SMP kernels running on 1 CPU.  The overheads
for using the i8254 cputime clock were a bit too high on 486/33's, and
now on multi-GHz CPUs they are usually in the 99-99.9% range.  Switching
from the old default of an i8254 clock to the TSC works poorly because
the overheads are not recalibrated.

Use the same condition for declaring perfmon stuff as for using it.

17 years agoRefactor vfs_setdirty(), creating vfs_setdirty_locked_object().
alc [Sun, 29 Oct 2006 00:04:39 +0000 (00:04 +0000)]
Refactor vfs_setdirty(), creating vfs_setdirty_locked_object().

Call vfs_setdirty_locked_object() from vfs_busy_pages() instead of
vfs_setdirty(), thereby eliminating a second acquisition and release
of the same vm object lock.

17 years agoGive rc.firewall a polish and a new method.
phk [Sat, 28 Oct 2006 20:08:12 +0000 (20:08 +0000)]
Give rc.firewall a polish and a new method.

Factor out the loopback setup

Use "me" instead of hardcoded $ip where possible.

Add "workstation" which protects just this machine with stateful
    firewalling.  Put the variables for this in rc.conf.

Submitted by: Flemming Jacobsen <fj@batmule.dk>
Reviewed by: cperciva

17 years agoIn bufdone_finish() restrict the acquisition and release of the page
alc [Sat, 28 Oct 2006 19:16:57 +0000 (19:16 +0000)]
In bufdone_finish() restrict the acquisition and release of the page
queues lock to BIO_READ operations.  Recent changes to the implementation
of the per-page flags have eliminated the need for the page queues lock
in the other cases.

17 years agoFix style(9).
netchild [Sat, 28 Oct 2006 16:47:38 +0000 (16:47 +0000)]
Fix style(9).

Noticed by: rwatson

17 years agoBacked out rev.1.10 (a special case for amd64). The amd64 mcount code
bde [Sat, 28 Oct 2006 13:34:35 +0000 (13:34 +0000)]
Backed out rev.1.10 (a special case for amd64).  The amd64 mcount code
never needed any frame pointers for itself.

17 years agoIn the userland .mcount():
bde [Sat, 28 Oct 2006 13:12:06 +0000 (13:12 +0000)]
In the userland .mcount():
- Don't use a frame pointer.  Our callers need a frame pointer, but we
  could only use one to support things that aren't supported.  (These
  things are:
  - profiling of profiling
  - debugging of profiling.  The core ENTRY() macro doesn't support
    forcing a frame pointer for debugging, so don't do more here.)
- Ensure that we are in the text section and have normal alignment.
- Use the normal syntax for `.type'.

17 years agoAdd missing markup bits.
ru [Sat, 28 Oct 2006 13:05:10 +0000 (13:05 +0000)]
Add missing markup bits.

Submitted by: Emil Mikulic

17 years agoregen (prctl addition)
netchild [Sat, 28 Oct 2006 11:24:38 +0000 (11:24 +0000)]
regen (prctl addition)

17 years agoi386/include/profile.h:
bde [Sat, 28 Oct 2006 11:03:03 +0000 (11:03 +0000)]
i386/include/profile.h:
Fixed a syntax error for the (!__KERNEL && !__GNUCLIKE_ASM) case in
rev.1.36.  Apparently, this case has never been reached even by lint.

Submitted by: stefanf

{amd64,i386}/include/profile.h:
In case the above case is actually reached, break it properly by
providing null support that will fail at link time instead of a stub
that gives wrong (null) profiling at runtime.

17 years agoMFP4:
netchild [Sat, 28 Oct 2006 10:59:59 +0000 (10:59 +0000)]
MFP4:
Implement prctl().

Submitted by: rdivacky
Tested with: LTP

17 years agoo Fix incorrect function name (kobj_free -> kobj_delete).
maxim [Sat, 28 Oct 2006 10:57:35 +0000 (10:57 +0000)]
o Fix incorrect function name (kobj_free -> kobj_delete).

Obtained from: DragonFlyBSD
MFC after: 1 week

17 years agoo Fix function name.
maxim [Sat, 28 Oct 2006 10:53:39 +0000 (10:53 +0000)]
o Fix function name.

Obtained from: DragonFlyBSD
MFC after: 1 week

17 years agoMFP4:
netchild [Sat, 28 Oct 2006 09:51:10 +0000 (09:51 +0000)]
MFP4:
Rename MAX_SAMPLE_RATES macro to OSS_MAX_SAMPLE_RATES.  The old
macro clashed with those used in other applications and libaries
(ex: RtAudio).  4Front responded by updating their spec, so we
will follow suit.

Submitted by: ryanb
Noticed by: pointyhat/kris

17 years agoBackout bogus checkin to HEAD
jfv [Sat, 28 Oct 2006 08:11:07 +0000 (08:11 +0000)]
Backout bogus checkin to HEAD
Approved by: scottl

17 years agoIn MCOUNT_OVERHEAD(label), actually use the `label' parameter. We were
bde [Sat, 28 Oct 2006 07:59:11 +0000 (07:59 +0000)]
In MCOUNT_OVERHEAD(label), actually use the `label' parameter.  We were
still using the global label named "profil", and this worked accidentally
because all callers use the same name.

17 years agoDocument Soekris net48xx support.
brueffer [Sat, 28 Oct 2006 07:26:50 +0000 (07:26 +0000)]
Document Soekris net48xx support.

PR: docs/104865
Submitted by: Henrik Brix Andersen <henrik@brixandersen.dk>

17 years agoCleaned up includes. <machine/profile.h> was unused. <machine/timerreg.h>
bde [Sat, 28 Oct 2006 06:38:51 +0000 (06:38 +0000)]
Cleaned up includes.  <machine/profile.h> was unused.  <machine/timerreg.h>
was only used in the GUPROF case, so the messes to get its i386 prerequisites
included shouldn't have been needed.

Fixed some style bugs. Quote #error contents, and don't repeat an #error
directive on amd64.

17 years agoRemoved all traces of HIDENAME() in amd64 and i386 kernel code. Using
bde [Sat, 28 Oct 2006 06:04:29 +0000 (06:04 +0000)]
Removed all traces of HIDENAME() in amd64 and i386 kernel code.  Using
this used to be slightly cleaner than using ifdefs in a few places to
support both a.out and elf, but using it now just causes messes and
unportabilities.  It seems to be impossible to implement the elf
HIDENAME() portably in cpp (since token pasting of "." and <name> is
invalid).

*/prof_machdep.c:
- Removed all uses of CNAME().  CNAME() is easy enough to use in pure
  asm code, but using it in inline asm requires messy quoting.  The
  core pure asm code has been hacked on more and all uses of CNAME() in
  it have already gone away.  Just assume the elf convention here too.
- Removed now-uneeded include of <machine/asmacros.h>.
- Removed the workaround for a namespace conflict with this include.

17 years agoThis is the merge of the Intel 6.2.9 driver. It provides all new shared code,
jfv [Sat, 28 Oct 2006 00:47:55 +0000 (00:47 +0000)]
This is the merge of the Intel 6.2.9 driver. It provides all new shared code,
new device support, and it is hoped a more stable driver for 6.2. RELEASE.
This checkin was discussed and approved today by RE, scottl, jhb, and pdeuskar

17 years agoDon't call mexitcount or provide a stub mexitcount to call when
bde [Fri, 27 Oct 2006 14:17:50 +0000 (14:17 +0000)]
Don't call mexitcount or provide a stub mexitcount to call when
profiling is configured but high resolution profiling is not configured.
Only functions in *.[Ss] called the stub, so efficiency was not
significantly affected.

17 years ago- Convert
oleg [Fri, 27 Oct 2006 13:05:37 +0000 (13:05 +0000)]
- Convert
net.inet.ip.dummynet.curr_time
net.inet.ip.dummynet.searches
net.inet.ip.dummynet.search_steps
  to SYSCTL_LONG nodes. It will prevent frequent wrap around on 64bit archs.

- Implement simple mechanics for dummynet(4) internal time correction.
  Under certain circumstances (system high load, dummynet lock contention, etc)
  dummynet's tick counter can be significantly slower than it should be.
  (I've observed up to 25% difference on one of my production servers).
  Since this counter used for packet scheduling, it's accuracy is vital for
  precise bandwidth limitation.

  Introduce new sysctl nodes:
  net.inet.ip.dummynet.
    tick_lost - number of ticks coalesced by taskqueue thread.
    tick_adjustment - number of time corrections done.
    tick_diff - adjusted vs non-adjusted tick counter difference
    tick_delta - last vs 'standard' tick differnece (usec).
    tick_delta_sum - accumulated (and not corrected yet) time
     difference (usec).

Reviewed by: glebius
MFC after: 2 month

17 years agoUse separate thread for servicing dummynet(4).
oleg [Fri, 27 Oct 2006 11:16:58 +0000 (11:16 +0000)]
Use separate thread for servicing dummynet(4).
Utilize taskqueue(9) API.

Submitted by: glebius
MFC after: 2 month

17 years agostyle(9) cleanup.
oleg [Fri, 27 Oct 2006 10:52:32 +0000 (10:52 +0000)]
style(9) cleanup.

MFC after: 2 month

17 years agoWording nits.
trhodes [Fri, 27 Oct 2006 08:26:24 +0000 (08:26 +0000)]
Wording nits.

17 years agoRemove member p_procscopegrp which is no longer used by libthr.
davidxu [Fri, 27 Oct 2006 05:45:44 +0000 (05:45 +0000)]
Remove member p_procscopegrp which is no longer used by libthr.

17 years agoAdd MPEG2-TS/HDV support to fwcontrol.
imp [Thu, 26 Oct 2006 22:33:38 +0000 (22:33 +0000)]
Add MPEG2-TS/HDV support to fwcontrol.

Submitted by: Petr Holub" <hopet@ics.muni.cz>
Clean up by: mi@
Pr: 98134

17 years agoRemove the KSE option now that it's in DEFAULTS on these arches/machines.
jb [Thu, 26 Oct 2006 22:11:35 +0000 (22:11 +0000)]
Remove the KSE option now that it's in DEFAULTS on these arches/machines.

The 'nooption' kernel config entry has to be used to turn KSE off now.
This isn't my preferred way of dealing with this, but I'll defer to
scottl's experience with the io/mem kernel option change and the grief
experienced over that.

Submitted by: scottl@

17 years agoAdd 'options KSE' to the kernel config DEFAULTS on all arches/machines
jb [Thu, 26 Oct 2006 22:05:25 +0000 (22:05 +0000)]
Add 'options KSE' to the kernel config DEFAULTS on all arches/machines
except sun4v.

This change makes the transition from a default to an option more
transparent and is an attempt to head off all the compliants that are
likely from people who don't read UPDATING, based on experience with
the io/mem change.

Submitted by: scottl@

17 years agoMake KSE a kernel option, turned on by default in all GENERIC
jb [Thu, 26 Oct 2006 21:42:22 +0000 (21:42 +0000)]
Make KSE a kernel option, turned on by default in all GENERIC
kernel configs except sun4v (which doesn't process signals properly
with KSE).

Reviewed by: davidxu@