]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
18 years agoTeach portsnap to parse the output of the host(1) in BIND 8 as well as
Colin Percival [Tue, 2 May 2006 05:27:30 +0000 (05:27 +0000)]
Teach portsnap to parse the output of the host(1) in BIND 8 as well as
the host(1) from BIND 9.  This doesn't matter for HEAD, but will help
people who install portsnap from the ports tree onto older versions of
FreeBSD.

PR: ports/93901
Sponsored by: FreeBSD security development fundraiser

18 years agoOnly print the driver name and version if bootverbose is set.
Scott Long [Tue, 2 May 2006 03:46:59 +0000 (03:46 +0000)]
Only print the driver name and version if bootverbose is set.

18 years agoType.
Pyun YongHyeon [Tue, 2 May 2006 02:12:42 +0000 (02:12 +0000)]
Type.

Submitted by: brad@OpenBSD

18 years agoAdd various constants for the PAT MSR and the PAT PTE and PDE flags.
John Baldwin [Mon, 1 May 2006 22:07:00 +0000 (22:07 +0000)]
Add various constants for the PAT MSR and the PAT PTE and PDE flags.
Initialize the PAT MSR during boot to map PAT type 2 to Write-Combining
(WC) instead of Uncached (UC-).

MFC after: 1 month

18 years agoBreak out socket access control and delivery logic from udp6_input()
Robert Watson [Mon, 1 May 2006 21:39:48 +0000 (21:39 +0000)]
Break out socket access control and delivery logic from udp6_input()
into its own function, udp6_append().  This mirrors a similar structure
in udp_input() and udp_append(), and makes the whole thing a lot more
readable.

While here, add missing inpcb locking in UDP6 input path.

Reviewed by: bz
MFC after: 3 months

18 years agoAdd a new 'pmap_invalidate_cache()' to flush the CPU caches via the
John Baldwin [Mon, 1 May 2006 21:36:47 +0000 (21:36 +0000)]
Add a new 'pmap_invalidate_cache()' to flush the CPU caches via the
wbinvd() instruction.  This includes a new IPI so that all CPU caches on
all CPUs are flushed for the SMP case.

MFC after: 1 month

18 years agoUsing an idea from Stephan Uphoff, use the empty pte's that correspond
Peter Wemm [Mon, 1 May 2006 21:22:38 +0000 (21:22 +0000)]
Using an idea from Stephan Uphoff, use the empty pte's that correspond
to the unused kva in the pv memory block to thread a freelist through.
This allows us to free pages that used to be used for pv entry chunks
since we can now track holes in the kva memory block.

Idea from:  ups

18 years agoo OpenBSD 3.9 added.
Maxim Konovalov [Mon, 1 May 2006 20:44:52 +0000 (20:44 +0000)]
o OpenBSD 3.9 added.

18 years agoFix missing changes required for the amd64->i386 conversion. Add the
Peter Wemm [Mon, 1 May 2006 19:57:00 +0000 (19:57 +0000)]
Fix missing changes required for the amd64->i386 conversion.  Add the
missing VM_ALLOC_WIRED flags to vm_page_alloc() calls I added.

Submitted by:  alc

18 years agoCorrect error messages.
Pawel Jakub Dawidek [Mon, 1 May 2006 12:05:45 +0000 (12:05 +0000)]
Correct error messages.

MFC after: 2 weeks

18 years agoCorrect the interrupt override information.
Christian Brueffer [Mon, 1 May 2006 11:56:18 +0000 (11:56 +0000)]
Correct the interrupt override information.

PR: 93429
Submitted by: Michaël Grünewald <michael.grunewald@laposte.net>
MFC after: 3 days

18 years agoAdd profile libs.
Alexander Leidinger [Mon, 1 May 2006 11:15:45 +0000 (11:15 +0000)]
Add profile libs.

Submitted by: Pawel Worach <pawel.worach@gmail.com>

18 years agoUpdate host.conf every time nsswitch.conf changes, instead of just creating
Dag-Erling Smørgrav [Mon, 1 May 2006 11:02:48 +0000 (11:02 +0000)]
Update host.conf every time nsswitch.conf changes, instead of just creating
it if it does not exist.

Submitted by: Rostislav Krasny <rosti.bsd@gmail.com>
MFC after: 2 weeks

18 years agoAdd inverted amplifier sense quirks for Sony VAIO VGN B1VP/B1XP.
Ariff Abdullah [Mon, 1 May 2006 09:28:28 +0000 (09:28 +0000)]
Add inverted amplifier sense quirks for Sony VAIO VGN B1VP/B1XP.

Reported by: Stan Behrens <s.behrens at kon.de>

18 years agoFor some time now, -i and -P options are mutually exclusive, there's even
Pawel Jakub Dawidek [Mon, 1 May 2006 09:12:52 +0000 (09:12 +0000)]
For some time now, -i and -P options are mutually exclusive, there's even
a regression test init-i-P.t which asserts this, but it looks I forgot to
update nokey.t regression test.

MFC after: 1 day

18 years agoFor the vmstat sub-display:
Bruce Evans [Mon, 1 May 2006 07:02:52 +0000 (07:02 +0000)]
For the vmstat sub-display:

vmstat.c:
Move totfr to be under daefr and prcfr since it logically belongs there.

Move all the count fields (wire, act, inact, cache and free) to near
the bottom of the sub-display (after all the rate fields) to reduce
competition with adjoining sub-displays.

systat.1:
Move things as above.

Attempt to improve missing and poor wording in the description of the
fields.  The long sentence was hard to parse and didn't say anything
about the different units.

Increment .Dd.

18 years agoDon't attach to Marvell 88E805X (Yukon-II) by default, the driver is not
Maxim Sobolev [Mon, 1 May 2006 05:16:36 +0000 (05:16 +0000)]
Don't attach to Marvell 88E805X (Yukon-II) by default, the driver is not
functional yet.

Requested by: bz

18 years agoIn the examples, use "tar.gz" as the extension for
Tim Kientzle [Mon, 1 May 2006 01:34:21 +0000 (01:34 +0000)]
In the examples, use "tar.gz" as the extension for
gzip-compressed tar archives.

Thanks to: Mike Hunter

18 years agoSimplify some of the wide-character handling, inspired
Tim Kientzle [Mon, 1 May 2006 01:02:19 +0000 (01:02 +0000)]
Simplify some of the wide-character handling, inspired
in part by OpenBSD's not-quite-standard-compliant
standard libraries.  (No loss of functionality,
just minor recoding to not rely on certain "standard"
facilities that weren't actually needed.)

18 years agoWhitespace cleanup.
Tim Kientzle [Mon, 1 May 2006 00:55:44 +0000 (00:55 +0000)]
Whitespace cleanup.

18 years agoUnbreak the support for 24-row terminals in the vmstat display. The
Bruce Evans [Mon, 1 May 2006 00:26:43 +0000 (00:26 +0000)]
Unbreak the support for 24-row terminals in the vmstat display.  The
part that handled the 17th and 18th rows of the vmstat-proper subdisplay
was deleted in rev.1.10 when these rows stopped being used and was not
restored when the 17th row was used again.  For such terminals, we now
lose the `buf' field instead of making a mess with it.  Terminals with
fewer than 24 rows have never been supported.

The problem is not avoided by using curses since we use the last line
for data entry and don't use a separate subwindow for this line.
Some other things in the vmstat display could be handled better using
subwindows.

18 years agoSort the ex-extended vmstat fields into their documented order in the
Bruce Evans [Sun, 30 Apr 2006 23:52:16 +0000 (23:52 +0000)]
Sort the ex-extended vmstat fields into their documented order in the
output too.

Fine tune all coordinates and most field widths in the vmstat (sub)display
for this and previous changes now that we have to change almost all of them
just to move the ex-extended fields:
- change VMSTATROW back to 7.  It was 6 due to a hack in the extended vm
  stats changes.
- reduce the maximum field width that we try for from 9 to 8.  4 or 5 is
  enough for most fields but we try to use the same width for all fields.
  8 is enough to display everything without changing units memory sizes
  exceed 100GB.

Fix some unrelated coordinates and field widths in comments.

18 years agoEliminate the "extended" vm stats. Move all fields in the extended
Bruce Evans [Sun, 30 Apr 2006 22:34:54 +0000 (22:34 +0000)]
Eliminate the "extended" vm stats.  Move all fields in the extended
vm stats to the normal vm stats.  Sort them into the normal stats
according to the man page only in the source code so that diffs are
almost readable.  Reduce style bugs in printing the value of %ozfod.

18 years agoWe shouldn't lock the topology here - we will panic on assertion inside
Pawel Jakub Dawidek [Sun, 30 Apr 2006 22:14:17 +0000 (22:14 +0000)]
We shouldn't lock the topology here - we will panic on assertion inside
g_raid3_bump_syncid().

Reported by: Bradley W. Dutton <brad-fbsd-stable@duttonbros.com>
MFC after: 1 day

18 years agoAdd missed SYNOPSIS flag for auto upgrade.
Gordon Tetlow [Sun, 30 Apr 2006 22:09:47 +0000 (22:09 +0000)]
Add missed SYNOPSIS flag for auto upgrade.

Submitted by: marck at rinet dot ru

18 years agoUpdate for moving some fields to the new vnstat display.
Bruce Evans [Sun, 30 Apr 2006 22:05:02 +0000 (22:05 +0000)]
Update for moving some fields to the new vnstat display.

Don't say that `cache' is for the buffer cache.

Describe the uselessnes of `buf'.

18 years agoo Sort .Xrs.
Maxim Konovalov [Sun, 30 Apr 2006 21:31:52 +0000 (21:31 +0000)]
o Sort .Xrs.

18 years agoo Reformat FILES section.
Maxim Konovalov [Sun, 30 Apr 2006 21:30:25 +0000 (21:30 +0000)]
o Reformat FILES section.

18 years ago`cache' couldn't co-exist with `compat'.
Hajimu UMEMOTO [Sun, 30 Apr 2006 21:25:00 +0000 (21:25 +0000)]
`cache' couldn't co-exist with `compat'.

Submitted by: "Michael Bushkov" <bushman__at__rsu.ru>

18 years agoo Add missed commas.
Maxim Konovalov [Sun, 30 Apr 2006 21:22:01 +0000 (21:22 +0000)]
o Add missed commas.

18 years agoo Silence mdoc(7) warnings: remove an empty line, add missed .El macro.
Maxim Konovalov [Sun, 30 Apr 2006 21:14:48 +0000 (21:14 +0000)]
o Silence mdoc(7) warnings: remove an empty line, add missed .El macro.
New sentence - new line.

18 years agoReduce the namei (sub)display by 5 columns to make enough space for a
Bruce Evans [Sun, 30 Apr 2006 20:31:00 +0000 (20:31 +0000)]
Reduce the namei (sub)display by 5 columns to make enough space for a
new vnstat display to the right of the namei display.

Move the non-vmstat fields {des,num,fre}vn from the vmstat display to a
new vnstat display.  Move the dtbuf field there too.  The buf and dtbuf
fields are non-vmstat and non-vnstat, so there is no good place to
display them.  I need to move at least 1 of them out of the vm stats
for further cleanups of the vm stats, and there is only space for 1
of them in the vn stats.  (The best place for the current buf field
is actually /dev/null, since it has been completely broken for about
10 years and broken for longer.  It gives an uninteresting virtual
memory count where an interesting real memory count is wanted.)

18 years agoSpecify default values for positive-policy and negative-policy .
Matteo Riondato [Sun, 30 Apr 2006 18:34:48 +0000 (18:34 +0000)]
Specify default values for positive-policy and negative-policy .

18 years agoRemove header pollution.
Scott Long [Sun, 30 Apr 2006 16:46:11 +0000 (16:46 +0000)]
Remove header pollution.

18 years agoRemove some header polution.
Scott Long [Sun, 30 Apr 2006 16:44:41 +0000 (16:44 +0000)]
Remove some header polution.

18 years agoTypo.
Ceri Davies [Sun, 30 Apr 2006 12:40:30 +0000 (12:40 +0000)]
Typo.

18 years agoRemoved the description of the nonexistent want_fd command. want_fd existed
Bruce Evans [Sun, 30 Apr 2006 09:23:11 +0000 (09:23 +0000)]
Removed the description of the nonexistent want_fd command.  want_fd existed
for only 2 weeks in 1998-1999.  It was replaced by general commands to
select the set of disk drives displayed.

18 years agoDon't redraw the disk names on every update. This was apparently done
Bruce Evans [Sun, 30 Apr 2006 09:13:59 +0000 (09:13 +0000)]
Don't redraw the disk names on every update.  This was apparently done
to handle changes to the set of disks selected, but it is unnecessary
for that since the whole screen is redrawn when this set is changed.
It was also buggy:
- MAXDRIVES*6 = 42 was hard-coded as only 30 spaces in a string literal,
  the last 2 disk names were not cleared as intended
- when the extended vmstats are active, clearing of even 30 columns
  overruns the ozfod value field by 3 columns.  This was harmless because
  the field is much wider than necessary.

18 years agoBackout my last changes. ISO-8859-15 does not specify ACS graphics.
Matteo Riondato [Sun, 30 Apr 2006 09:05:56 +0000 (09:05 +0000)]
Backout my last changes. ISO-8859-15 does not specify ACS graphics.
Requested by: ache

18 years agoFix "slow (on-the-fly) zero fills percentage (`%slo-z')" some more. The
Bruce Evans [Sun, 30 Apr 2006 07:27:23 +0000 (07:27 +0000)]
Fix "slow (on-the-fly) zero fills percentage (`%slo-z')" some more.  The
value printed is actually the optimized (i.e., the non-slow, not-on-the-fly
zero fills percentage) except in overflow cases.  Describe it as %ozfod
in the display.  Move the field descriptor 1 to the left so that there
is space for 5 characters after the % sign (this leaves no space between
the number and the descriptor but the % character serves well as a
separator).

Fixed integer overflow at z.ozfod = UINT_MAX/100 in the calculation of
%ozfod.  This value can be reached just a few hours or minutes after
booting, so %ozfod was usually garbage in boot mode.  Now %ozfod is
correct in boot mode for a few days or hours.

Print a non-dummy %ozfod when the division for it isn't division by 0
instead of when the result will be less than 100%.  A result of 100%
may be correct, though a result of more than 100% indicates overflow
of one or both counters.

18 years agoGo with a different version of the previous patch so to preserve errno.
Joe Marcus Clarke [Sun, 30 Apr 2006 07:08:43 +0000 (07:08 +0000)]
Go with a different version of the previous patch so to preserve errno.

Approved by: scottl (implicit)

18 years agoFix a file descriptor leak in cam_lookup_pass() when the ioctl to find
Joe Marcus Clarke [Sun, 30 Apr 2006 07:02:40 +0000 (07:02 +0000)]
Fix a file descriptor leak in cam_lookup_pass() when the ioctl to find
the passthru device fails.

Approved by: scottl
MFC after: 1 day

18 years agoDocument sysinstall's limitation on operating on mounted disks.
Scott Long [Sun, 30 Apr 2006 06:28:29 +0000 (06:28 +0000)]
Document sysinstall's limitation on operating on mounted disks.

18 years agoIf DEBUG is defined, then fill numeric fields with asterisks instead of
Bruce Evans [Sun, 30 Apr 2006 05:54:21 +0000 (05:54 +0000)]
If DEBUG is defined, then fill numeric fields with asterisks instead of
spaces and numbers for temporary(?) debugging.

18 years agoNew release notes:
Hiroki Sato [Sun, 30 Apr 2006 05:22:43 +0000 (05:22 +0000)]
New release notes:
bfe(4) + >1GB RAM problem fixed.

18 years agoRemove trailing whitespace.
Bruce Evans [Sun, 30 Apr 2006 04:47:23 +0000 (04:47 +0000)]
Remove trailing whitespace.

Submitted by: Se=E1n Farley <sean-freebsd at farley dot org>
PR: bin/81874

18 years agoShow the load average in the tcp display (it was already shown, perhaps
Bruce Evans [Sun, 30 Apr 2006 04:26:46 +0000 (04:26 +0000)]
Show the load average in the tcp display (it was already shown, perhaps
not very usefully, in all other displays).  This was the original point
of the PR.

Move the load average up by 2 so that it starts in row 0 for all windows
(2 lines above it were wasted for all other windows except vmstat).
Move everything below it up by 2 or 3 (3 for icmp and icmp6 which had
an extra blank line due from not compensating for the foot-shooting in
note (3); only ip and ip6 compensated).  Reduce the magic numbers related
to this.

Notes by the submitter:
%%%
1. All the subwin() calls are identical using #define MAINWIN_ROW 3
    (systat.h).
2. The load average is at the top of the window.
3. Each display starts on the fourth line.  I made changes to those
    displays that shifted the start line (i.e., icmp).  This entailed a
    lot of changes within the comments at the top of those displays.
4. For ip6, I shifted the "Input next-header histogram" column down one
    row to separate it from "IPv6 Output".  I raised "bad scope packets"
    and "address selection failed" up one row to stay with "IPv6 Input"
    (valid?).  They were down one row to probably line up at the bottom,
    but I think they should stick with their fellow items in a column.
5. I condensed ifstat a bit.  It had a lot of empty rows.
%%%

Submitted by: Se=E1n Farley <sean-freebsd at farley dot org>
PR: bin/81874

18 years agoReword a sentence modified in my previous commit. This new one is probably
Tom Rhodes [Sun, 30 Apr 2006 01:53:57 +0000 (01:53 +0000)]
Reword a sentence modified in my previous commit.  This new one is probably
better.

Hinted by: ceri

18 years agoEdit the interrupt name strings to shorten them. This is believed to
Bruce Evans [Sun, 30 Apr 2006 01:39:46 +0000 (01:39 +0000)]
Edit the interrupt name strings to shorten them.  This is believed to
only affect amd64 and i386.  alpha uses "intr N" instead of "irqN" and
mostly has no device names.  ia64 uses only device names.

- Edit interrupt names once after they are read from the kernel and not
  every time they are displayed.
- Discard bogus trailing spaces so that the next step doesn't move things
  to oblivion.
- If an interrupt name starts with "irqN:" (as it usually does in on
  amd64 and i386), then move "irqN" to the end and strip ":", since we
  have no space for the ":" and don't want to start descriptions with
  "N" after stripping "irq" in the next step (since "N" would look like
  a count).  This step may need reworking for interrupt names containing
  several device names -- then moving the irq number to the end would
  lose it instead of losing some device names.
- Remove "irq" from an interrupt name if and only if the original name is
  too long to display.

18 years agoBacked out rev.1.49 since it had buffer overruns and only worked
Bruce Evans [Sun, 30 Apr 2006 00:50:08 +0000 (00:50 +0000)]
Backed out rev.1.49 since it had buffer overruns and only worked
accidentally.

Read buffer overruns:
The size of the target array (TSOTTA == 10) is a wrong limit to use for
scanning the source string.

Write buffer overruns:
TSOTTA is also a wrong limit to use for copying to the target buffer,
since we want to add a NUL terminator afterwards.  TSOTTA was also 1
too small for holding both the desired number of visible characters
and the NUL.

Worked accidentally:
There is error in the algorithm that tends to result in the space saved
by stripping "irq" not actually being used, but some cases worked
accidentally provided "irqN" is near the end of the source string and
"N" is only 1 digit.

Starting with 5.mumble-CURRENT, "irqN" is at the beginning of the
string on all (?) arches that have it and the accidents don't happen.
E.g. on i386's, the keyboard irq is now named
"irq1: atkbd0<bogus blank padding>" by the kernel, and this name was
converted to "1: atkb" -- not only the device number but part of the
device name has been lost --, while before 5.mumble the kernel name
was "atkbd0 irq1" and systat accidentally preserved the irq number to
give "atkbd0 1".  The ":" in the string wastes precious space, and
stripping "irq" results in descriptions starting with numbers which
makes them look too much like counts.  This commit just fixes the last
problem.

18 years agovn_start_write()/vn_finished_write() is not needed here, because
Pawel Jakub Dawidek [Sat, 29 Apr 2006 21:57:38 +0000 (21:57 +0000)]
vn_start_write()/vn_finished_write() is not needed here, because
vn_start_write() is always called earlier in the code path and calling
the function recursively may lead to a deadlock.

Confirmed by: tegge
MFC after: 2 weeks

18 years agoAbbreviate long field descriptors at write time so that they don't get
Bruce Evans [Sat, 29 Apr 2006 21:30:23 +0000 (21:30 +0000)]
Abbreviate long field descriptors at write time so that they don't get
clobbered at runtime:
    dirtybuf -> dtbuf
    desiredvnodes -> desvn
    numvnodes -> numvn
    freevnodes -> frevn
The vmstats column has only 5 characters available for descriptors, but up
to 13 were used.  The extras get clobbered at runtime by interrupt values
and/or descriptors on systems with more than 12 interrupt sources.
    %slo-z -> %sloz
This one is in the "extended" vmstats area and doesn't get clobbered now.

Removed stale documentation of desvn.

Changed a descriptor:
    tfree -> totfr
so that it is consistent with the abbreviations for other free counts
(daefr and prcfr) and thus almost decodeable.

Fixed missing documentation of tfree/totfr.  This and everything else
in the extended vmstats area is misdocumented as being in a certain
place in the vmstats column.

18 years agoAdd auto upgrade capability to mergemaster.
Gordon Tetlow [Sat, 29 Apr 2006 18:21:43 +0000 (18:21 +0000)]
Add auto upgrade capability to mergemaster.

An mtree description of all non-zero files that make
distribution installs (only size and md5) is built from the
temproot.  When the user completes a mergemaster run, the
mtree description file gets installed into /var/db for
safe-keeping.

When the user then decides to do a subsequent upgrade (with
the -U flag), the existing mtree description from /var/db
is called into service looking for files that are different in
DESTDIR. This is stashed away until a file that would normally
end up prompting the user to look at changes is encountered.
Since there are no user modified changes, the new file is
installed without bothering the user.

Looked at by: dougb
MFC after: 6 weeks

18 years agoFix the acpi.ko path for boot.flp.
Yoshihiro Takahashi [Sat, 29 Apr 2006 14:03:17 +0000 (14:03 +0000)]
Fix the acpi.ko path for boot.flp.

PR: misc/96466
Submitted by: Nobuyuki Koganemaru
MFC after: 1 day

18 years agoPOSIX demands that set's output (when invoked without arguments) should be
Stefan Farfeleder [Sat, 29 Apr 2006 12:57:53 +0000 (12:57 +0000)]
POSIX demands that set's output (when invoked without arguments) should be
sorted.  Sort the variables before printing.

PR: 96415

18 years agoAdd curses ACS line graphics support for iso15 fonts
Matteo Riondato [Sat, 29 Apr 2006 12:34:19 +0000 (12:34 +0000)]
Add curses ACS line graphics support for iso15 fonts
Now ncurses-based programs such as sysinstall and mc will display the
correct font for graphical lines instead of "-" and "+" characters.

Correct two special characters for cons25l1 in termcap: use real
arrows instead of ">>" and "<<".
Add a lot of additional symbols for line drawing which are taken from
the CP437 font.
Almost all of the ACS symbols are now implemented.

PR: conf/90082
Submitted by: Oliver Fromme <olli@secnetix.de>
MFC after: 1 week

18 years agoCheck the buffer size when copying the line returned by el_gets() into our
Stefan Farfeleder [Sat, 29 Apr 2006 10:29:10 +0000 (10:29 +0000)]
Check the buffer size when copying the line returned by el_gets() into our
own buffer.  Interactively typing in long lines (>1023 characters)
previously overflowed the buffer.  Unlike the NetBSD people I don't see the
need to subtract 8 from BUFSIZ, so I just used BUFSIZ-1.

Obtained from: NetBSD
PR: 91110

18 years agoLock giant when assigning ni_vp and keep vfslocked state valid.
Kris Kennaway [Sat, 29 Apr 2006 07:13:49 +0000 (07:13 +0000)]
Lock giant when assigning ni_vp and keep vfslocked state valid.

Committed for: jeff

18 years agoOops, services should be compat by default for backward compatibility.
Hajimu UMEMOTO [Sat, 29 Apr 2006 06:39:07 +0000 (06:39 +0000)]
Oops, services should be compat by default for backward compatibility.

18 years agoAdd support for fragmenting ipv4 packets.
Andrew Thompson [Sat, 29 Apr 2006 05:37:25 +0000 (05:37 +0000)]
Add support for fragmenting ipv4 packets.

The packet filter may reassemble the ip fragments and return a packet that is
larger than the MTU of the sending interface. There is no check for DF or icmp
replies as we can only get a large packet to fragment by reassembling a
previous fragment, and this only happens after a call to pfil(9).

Obtained from: OpenBSD (mostly)
Glanced at by: mlaier
MFC after: 1 month

18 years agoAdd newly supported databases such as services, protocols and rpc
Hajimu UMEMOTO [Sat, 29 Apr 2006 04:49:19 +0000 (04:49 +0000)]
Add newly supported databases such as services, protocols and rpc
into generated nsswitch.conf.

18 years agoEnglish grammar fix in comment.
Hajimu UMEMOTO [Sat, 29 Apr 2006 04:26:16 +0000 (04:26 +0000)]
English grammar fix in comment.

Submitted by: keramida

18 years agoFollow FreeBSD mdoc(7) conventions:
Tom Rhodes [Sat, 29 Apr 2006 01:34:03 +0000 (01:34 +0000)]
Follow FreeBSD mdoc(7) conventions:
- Remove hard sentence breaks;
- Avoid using double negatives or "sexist" language;
- Expand contractions;
- Remove a blank line;
- Some grammar changes.

Usually we do not "hard code" requests to submit bugs to the author, but
I will leave this go for now.

18 years agoCorrect sentence, forwarding non IP packets is now an option.
Andrew Thompson [Sat, 29 Apr 2006 01:19:08 +0000 (01:19 +0000)]
Correct sentence, forwarding non IP packets is now an option.

18 years agoEliminate unnecessary, recursive acquisitions and releases of the page
Alan Cox [Sat, 29 Apr 2006 00:59:15 +0000 (00:59 +0000)]
Eliminate unnecessary, recursive acquisitions and releases of the page
queues lock by free_pv_entry() and pmap_remove_pages().

Reduce the scope of the page queues lock in pmap_remove_pages().

18 years agoTypo.
Giorgos Keramidas [Sat, 29 Apr 2006 00:43:23 +0000 (00:43 +0000)]
Typo.

18 years agoFix a couple of typos and other minor nits
Giorgos Keramidas [Fri, 28 Apr 2006 23:21:36 +0000 (23:21 +0000)]
Fix a couple of typos and other minor nits

Reviewed by: rwatson

18 years agoMinor improvements after last revision.
Giorgos Keramidas [Fri, 28 Apr 2006 22:37:16 +0000 (22:37 +0000)]
Minor improvements after last revision.

Reviewed by: thompsa

18 years agovn_start_write() is called only when v_type != VCHR, so corresponding
Pawel Jakub Dawidek [Fri, 28 Apr 2006 21:54:05 +0000 (21:54 +0000)]
vn_start_write() is called only when v_type != VCHR, so corresponding
vn_finished_write() should also be called only then.

BTW. I fixed two functions here: vn_rdwr() and vn_write(). The latter seems
to be unused.

MFC after: 3 weeks

18 years agoo Borrow from if_bridge.4: note that the bridge interface need to be upped.
Maxim Konovalov [Fri, 28 Apr 2006 21:45:08 +0000 (21:45 +0000)]
o Borrow from if_bridge.4: note that the bridge interface need to be upped.

PR: docs/94545
Submitted by: thompsa

18 years agoAlso check use_pty in the ptmx clone lookup; this means that when ptmx
Robert Watson [Fri, 28 Apr 2006 21:39:57 +0000 (21:39 +0000)]
Also check use_pty in the ptmx clone lookup; this means that when ptmx
support is turned off using the sysctl, we no longer even allow the
ptmx device to be looked up.

Foot provided by: peter

18 years agoRemove the puc-specific hacks. The puc(4) driver now properly uses
Marcel Moolenaar [Fri, 28 Apr 2006 21:23:09 +0000 (21:23 +0000)]
Remove the puc-specific hacks. The puc(4) driver now properly uses
the rman(9) interface.

18 years agoRewrite of puc(4). Significant changes are:
Marcel Moolenaar [Fri, 28 Apr 2006 21:21:53 +0000 (21:21 +0000)]
Rewrite of puc(4). Significant changes are:
o  Properly use rman(9) to manage resources. This eliminates the
   need to puc-specific hacks to rman. It also allows devinfo(8)
   to be used to find out the specific assignment of resources to
   serial/parallel ports.
o  Compress the PCI device "database" by optimizing for the common
   case and to use a procedural interface to handle the exceptions.
   The procedural interface also generalizes the need to setup the
   hardware (program chipsets, program clock frequencies).
o  Eliminate the need for PUC_FASTINTR. Serdev devices are fast by
   default and non-serdev devices are handled by the bus.
o  Use the serdev I/F to collect interrupt status and to handle
   interrupts across ports in priority order.
o  Sync the PCI device configuration to include devices found in
   NetBSD and not yet merged to FreeBSD.
o  Add support for Quatech 2, 4 and 8 port UARTs.
o  Add support for a couple dozen Timedia serial cards as found
   in Linux.

18 years agoThe nvidia binary blob sometimes defers tx completion notification to the
John Baldwin [Fri, 28 Apr 2006 20:08:16 +0000 (20:08 +0000)]
The nvidia binary blob sometimes defers tx completion notification to the
OS dependent layer.  Thus, the watchdog timer can go off when the tx
engine is working fine but the OS dependent layer just hasn't been called
to cleanup finished tx transactions.  To workaround this, when the watchdog
fires, poke the binary blob to force it to flush any pending tx
completions.  If this drops the pending tx count to zero then just return
without logging a message or resetting the chip.

This reportedly fixes the 'device timeout()' errors with at least several
NF4 nve(4) parts.

Submitted by: Nathan Alexander Whitehorn <nathanw@uchicago.edu> (code)
Submitted by: dg (inspiration for comment and explanation)
MFC after: 1 week

18 years agoFix typo.
Robert Watson [Fri, 28 Apr 2006 19:43:44 +0000 (19:43 +0000)]
Fix typo.

MFC after: 1 month
Submitted by: Antoine Brodin <antoine dot brodin@laposte dot net>

18 years agoEnsure outbound data packets in hostap mode are delivered only to
Sam Leffler [Fri, 28 Apr 2006 19:06:15 +0000 (19:06 +0000)]
Ensure outbound data packets in hostap mode are delivered only to
stations that are associated by making ieee80211_find_txnode return
NULL when a unicast frame is to be delivered to an unassociated
station.  This will be handled differently in the future but for
now putting the check here allows all drivers to immediately do
the right thing.

Reviewed by: avatar
MFC after: 1 week

18 years agoInterim fix for pmap problems I introduced with my last commit.
Peter Wemm [Fri, 28 Apr 2006 19:05:08 +0000 (19:05 +0000)]
Interim fix for pmap problems I introduced with my last commit.
Remove the code to dyanmically change the pv_entry limits.  Go back
to a single fixed kva reservation for pv entries, like was done
before when using the uma zone.  Go back to never freeing pages
back to the free pool after they are no longer used, just like
before.

This stops the lock order reversal due to aquiring the kernel map
lock while pmap was locked.

This fixes the recursive panic if invariants are enabled.

The problem was that allocating/freeing kva causes vm_map_entry
nodes to be allocated/freed.  That can recurse back into pmap as
new pages are hooked up to kvm and hence all the problem.
Allocating/freeing kva indirectly allocate/frees memory.

So, by going back to a single fixed size kva block and an index,
we avoid the recursion panics and the LOR.

The problem is that now with a linear block of kva, we have no
mechanism to track holes once pages are freed.  UMA has the same
problem when using custom object for a zone and a fixed reservation
of kva.  Simple solutions like having a bitmap would work, but would
be very inefficient when there are hundreds of thousands of bits
in the map.  A first-free pointer is similarly flawed because pages
can be freed at random and the first-free pointer would be rewinding
huge amounts.  If we could allocate memory for tree strucures or
an external freelist, that would work.  Except we cannot allocate/free
memory here because we cannot allocate/free address space to use
it in.  Anyway, my change here reverts back to the UMA behavior of
not freeing pages for now, thereby avoiding holes in the map.

ups@ had a truely evil idea that I'll investigate.  It should allow
freeing unused pages again by giving us a no-cost way to track the
holes in the kva block.  But in the meantime,  this should get people
booting with witness and/or invariants again.

Footnote: amd64 doesn't have this problem because of the direct map
access method.  I'd done all my witness/invariants testing there.  I'd
never considered that the harmless-looking kmem_alloc/kmem_free calls
would cause such a problem and it didn't show up on the boot test.

18 years agoImplement the ipend() method of the serdev I/F.
Marcel Moolenaar [Fri, 28 Apr 2006 18:29:23 +0000 (18:29 +0000)]
Implement the ipend() method of the serdev I/F.

18 years agoMinor wording and formatting fixes.
Christian Brueffer [Fri, 28 Apr 2006 13:12:40 +0000 (13:12 +0000)]
Minor wording and formatting fixes.

18 years ago- Don't hold the device sx lock when going to sleep.
Pawel Jakub Dawidek [Fri, 28 Apr 2006 12:18:03 +0000 (12:18 +0000)]
- Don't hold the device sx lock when going to sleep.
- Prevent possible live-lock in case of memory problems by freeing
  already completed requests first.

Reported and tested by: markus, Bradley W. Dutton <brad-fbsd-stable@duttonbros.com>
MFC after: 1 day

18 years ago- Remove dead code.
Pawel Jakub Dawidek [Fri, 28 Apr 2006 12:13:49 +0000 (12:13 +0000)]
- Remove dead code.
- Comment possible event miss, which isn't critical, but probably can be
  fixed by replacing the event lock usage with the queue lock.

MFC after: 2 weeks

18 years ago- Extend the nsswitch to support Services, Protocols and Rpc
Hajimu UMEMOTO [Fri, 28 Apr 2006 12:03:38 +0000 (12:03 +0000)]
- Extend the nsswitch to support Services, Protocols and Rpc
  databases.
- Make nsswitch support caching.

Submitted by: Michael Bushkov <bushman__at__rsu.ru>
Sponsored by: Google Summer of Code 2005

18 years agoBe sure to not destroy device twice. This is not possible in theory, but
Pawel Jakub Dawidek [Fri, 28 Apr 2006 11:52:45 +0000 (11:52 +0000)]
Be sure to not destroy device twice. This is not possible in theory, but
with this change there is even no theoretical race.

MFC after: 2 weeks

18 years ago - use ath(4) in the wireless examples rather than the aging wi(4)
Andrew Thompson [Fri, 28 Apr 2006 11:48:53 +0000 (11:48 +0000)]
 - use ath(4) in the wireless examples rather than the aging wi(4)
 - make the packet filtering its own section and clarify a few points
 - note that the interfaces need to be upped [1]

PR: docs/94545 [1]

18 years agoBe sure to not destroy device twice. This is not possible in theory, but
Pawel Jakub Dawidek [Fri, 28 Apr 2006 11:47:28 +0000 (11:47 +0000)]
Be sure to not destroy device twice. This is not possible in theory, but
with this change there is even no theoretical race.

MFC after: 2 weeks

18 years agoAutogenerate hardware notes for rr232x(4).
Christian Brueffer [Fri, 28 Apr 2006 10:45:28 +0000 (10:45 +0000)]
Autogenerate hardware notes for rr232x(4).

18 years agoAdd a basic man page for the sysctl(9) macro interfaces. Previously man
Robert Watson [Fri, 28 Apr 2006 10:45:27 +0000 (10:45 +0000)]
Add a basic man page for the sysctl(9) macro interfaces.  Previously man
pages existed only for the dynamic sysctl interfaces.  There's probably
more complete and accurate content, better advice, etc, that could be added
here.

Per scottl's suggest, add a small piece of moralizing text regarding the
fact that sysctl names quickly get embedded in system configuration files,
libraries, third party applications, and even books, so renaming and
removing names after they've been published is a tricky issue.

MFC after: 1 month

18 years agoAdd a section about locking protocol for busdma.
Scott Long [Fri, 28 Apr 2006 08:36:49 +0000 (08:36 +0000)]
Add a section about locking protocol for busdma.

18 years agoOops, forgot to remove the old entry in the last revision.
Christian Brueffer [Fri, 28 Apr 2006 06:23:42 +0000 (06:23 +0000)]
Oops, forgot to remove the old entry in the last revision.

18 years agoInstall rr232x.4 only on i386 and amd64.
Christian Brueffer [Fri, 28 Apr 2006 06:13:19 +0000 (06:13 +0000)]
Install rr232x.4 only on i386 and amd64.

18 years agoTweak the DMA limit from rev 1.33, it was off by one byte.
Mike Silbersack [Fri, 28 Apr 2006 05:38:12 +0000 (05:38 +0000)]
Tweak the DMA limit from rev 1.33, it was off by one byte.

Submitted by: scottl

18 years agoSwitch all bus_dmamap_sync calls that used PREREAD to PREWRITE and all
Mike Silbersack [Fri, 28 Apr 2006 05:27:27 +0000 (05:27 +0000)]
Switch all bus_dmamap_sync calls that used PREREAD to PREWRITE and all
POSTWRITE to POSTREAD.

No guarantee that all busdma is usage is perfect, but this change (in
addition to scott's last two commits) makes if_bfe work with > 1GB of
memory in my laptop.

18 years agoEnable the rr232x driver for amd64.
Scott Long [Fri, 28 Apr 2006 05:23:10 +0000 (05:23 +0000)]
Enable the rr232x driver for amd64.

18 years agoFix a typo.
Scott Long [Fri, 28 Apr 2006 04:39:50 +0000 (04:39 +0000)]
Fix a typo.

18 years agoAdd a manpage for the rr232x driver.
Scott Long [Fri, 28 Apr 2006 04:26:00 +0000 (04:26 +0000)]
Add a manpage for the rr232x driver.

18 years agoFix a typo that slipped in right before commit.
Scott Long [Fri, 28 Apr 2006 04:00:50 +0000 (04:00 +0000)]
Fix a typo that slipped in right before commit.

18 years agoUnbreak pc98. Sorry...
Maxim Sobolev [Fri, 28 Apr 2006 03:38:23 +0000 (03:38 +0000)]
Unbreak pc98. Sorry...

18 years agoAdd some incomplete support for Marvell Yukon EC controllers based on
Maxim Sobolev [Fri, 28 Apr 2006 03:17:37 +0000 (03:17 +0000)]
Add some incomplete support for Marvell Yukon EC controllers based on
OpenBSD changes. With these changes, PHY part of the driver becomes
functional (it senses media changes and negotiates speed just fine),
previously it just hang with no PHY message, but no data goes through
interface (error message is "can not stop transfer of Tx/Rx descriptor).

Hopefully somebody with more clue/free time will be able to pick up
after me.

18 years ago - Add a BO_NEEDSGIANT flag to the bufobj. This flag forces all child
Jeff Roberson [Fri, 28 Apr 2006 01:05:31 +0000 (01:05 +0000)]
 - Add a BO_NEEDSGIANT flag to the bufobj.  This flag forces all child
   buffers to go on the buf daemon's DIRTYGIANT queue.
 - Set BO_NEEDSGIANT on ffs's devvp since the ffs_copyonwrite handler
   runs in the context of the buf daemon and may require Giant.

18 years ago - Consistently track ni_dvp and ni_vp with dvfslocked and vfslocked rather
Jeff Roberson [Fri, 28 Apr 2006 00:59:48 +0000 (00:59 +0000)]
 - Consistently track ni_dvp and ni_vp with dvfslocked and vfslocked rather
   than trying to optimize it into a single lock.  This adds more calls to
   lock giant with non smpsafe filesystems but is the only way to reliably
   hold the correct lock.
 - Remove an invalid assert in the mountedhere case in lookup and fix the
   code to properly deal with the scenario.  We can actually have a lookup
   that returns dp == dvp with mountedhere set with certain unmount races.

Tested by: kris
Reported by: kris/mohans