]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
19 years agoAn array was mistaken for a pointer in the previous commit.
nectar [Thu, 7 Apr 2005 19:26:35 +0000 (19:26 +0000)]
An array was mistaken for a pointer in the previous commit.

Noticed by: tinderbox, stefanf
Pointy hat to: nectar

19 years agoMake previous commit actually working by replacing TUNABLE_INT() with
sobomax [Thu, 7 Apr 2005 18:18:17 +0000 (18:18 +0000)]
Make previous commit actually working by replacing TUNABLE_INT() with
TUNABLE_INT_FETCH(). Apparently keyboard init is performed earlier
in the boot process than fetching all static tunables.

MFC after: 1 day

19 years agoRevert part of 1.19. We do want to set the count to 0 since otherwise
njl [Thu, 7 Apr 2005 17:33:22 +0000 (17:33 +0000)]
Revert part of 1.19.  We do want to set the count to 0 since otherwise
it would give false info to other parts of the driver.

19 years agoHook up new man page for devclass_get_drivers() to build
njl [Thu, 7 Apr 2005 17:27:14 +0000 (17:27 +0000)]
Hook up new man page for devclass_get_drivers() to build

Reminded by: ru

19 years agoProvide a new tunable hw.atkbdc.broken_kit_cmd, which if set to 1
sobomax [Thu, 7 Apr 2005 17:15:10 +0000 (17:15 +0000)]
Provide a new tunable hw.atkbdc.broken_kit_cmd, which if set to 1
instructs the driver to avoid using Keyboard Interface Test command.
This command causes problems with some non-compliant hardware, resulting
in machine being abruptly powered down early in the boot process.

Particularly it's known that HP ZV5000 and Compaq R3000Z notebooks
are affected by this problem.

Due to popularity of those models this patch is good MFC5.4 candidate.

PR: 67745
Submitted by: Jung-uk Kim jkim at niksun.com
MFC after: 1 days

19 years agoAllow classes to specify local source files.
pjd [Thu, 7 Apr 2005 15:57:38 +0000 (15:57 +0000)]
Allow classes to specify local source files.

MFC after: 3 weeks

19 years agos/foobar/mekmitasdigoat/g
glebius [Thu, 7 Apr 2005 15:28:14 +0000 (15:28 +0000)]
s/foobar/mekmitasdigoat/g

Requested by: ceri, des

19 years agoWhen Parse_Error is called after the top-level Makefile is closed we
harti [Thu, 7 Apr 2005 14:39:38 +0000 (14:39 +0000)]
When Parse_Error is called after the top-level Makefile is closed we
have no CURFILE anymore so we cannot print a file name or line number.

When ParseSkipLine() returns NULL (it does this when it has detected an
EOF in an .if block) try to pop the input stack and process the next line

19 years ago- Be more informative about advbase and advskew. [1]
glebius [Thu, 7 Apr 2005 14:20:34 +0000 (14:20 +0000)]
- Be more informative about advbase and advskew. [1]
- Fix ifconfig commands. Replace 'mekmitasdigoat' with
  'foobar'. While the former is more cool, the latter
  makes example lines shorter.

Wording by:     scottl
MFC after: 3 days

19 years agoHandle popping of the input stack in ParseReadLine() instead of
harti [Thu, 7 Apr 2005 11:29:39 +0000 (11:29 +0000)]
Handle popping of the input stack in ParseReadLine() instead of
Parse_File(). Remove a comment and a piece of code comming from
ancient times when the if-directive read like #if and not .if.
Correctly analyze the first character of a line.

19 years agoMostly stylistic changes: remove the noPLen field from the Shell
harti [Thu, 7 Apr 2005 11:26:15 +0000 (11:26 +0000)]
Mostly stylistic changes: remove the noPLen field from the Shell
structure - it is just the strlen() of noPrint. Inline JobCondPassSig()
in the only function JobPassSig() using it. Fix the argument types
of JobPrintCommand(). Better parsing for the .SHELL target keywords.

19 years agoGive msdosfs a unique inode number which is really the byteoffset of
phk [Thu, 7 Apr 2005 07:55:37 +0000 (07:55 +0000)]
Give msdosfs a unique inode number which is really the byteoffset of
the directory entry.

This solves the corruption problem I belive.

Regression test script by: silby

19 years agoFix bug in vfs_hash_rehash(): use correct bucket. This only affected
phk [Thu, 7 Apr 2005 07:54:08 +0000 (07:54 +0000)]
Fix bug in vfs_hash_rehash(): use correct bucket.  This only affected
msdosfs which is broken in other ways too.

19 years agoEmpty error buffer is not an error.
pjd [Thu, 7 Apr 2005 06:46:11 +0000 (06:46 +0000)]
Empty error buffer is not an error.

MFC after: 3 weeks

19 years agoAdjust hash function for smaller pthread structure size.
davidxu [Thu, 7 Apr 2005 06:09:17 +0000 (06:09 +0000)]
Adjust hash function for smaller pthread structure size.

19 years agoEliminate an unneeded instruction that is a vestige of mechanical
alc [Thu, 7 Apr 2005 05:46:46 +0000 (05:46 +0000)]
Eliminate an unneeded instruction that is a vestige of mechanical
translation from i386.

19 years agoFix strict-alias warnings by removing excessive (and wrong) casts.
peter [Thu, 7 Apr 2005 04:33:15 +0000 (04:33 +0000)]
Fix strict-alias warnings by removing excessive (and wrong) casts.

19 years agoAdd machine-specific, optimized implementations of bcopy, bzero, memcpy,
alc [Thu, 7 Apr 2005 03:56:03 +0000 (03:56 +0000)]
Add machine-specific, optimized implementations of bcopy, bzero, memcpy,
memmove, and memset.

PR: 73111
Submitted by: Ville-Pertti Keinonen <will@iki.fi> (taken from NetBSD)
MFC after: 3 weeks

19 years agoMFen 1.843
delphij [Thu, 7 Apr 2005 02:59:18 +0000 (02:59 +0000)]
MFen 1.843

Obtained from: The FreeBSD Simplified Chinese Project CVS

19 years agoCorrect typo that could cause FIFO overflow.
emax [Wed, 6 Apr 2005 22:09:32 +0000 (22:09 +0000)]
Correct typo that could cause FIFO overflow.

PR: kern/78431
MFC after: 3 days

19 years agoRemove PR_ATOMIC flag in ng_btsocket_protosw[] for BLUETOOTH_PROTO_RFCOMM
emax [Wed, 6 Apr 2005 20:54:05 +0000 (20:54 +0000)]
Remove PR_ATOMIC flag in ng_btsocket_protosw[] for BLUETOOTH_PROTO_RFCOMM
protocol. RFCOMM is a SOCK_STREAM protocol not SOCK_SEQPACKET. This was a
serious bug caused by cut-and-paste. I'm surprised it did not bite me before.
Dunce hat goes to me.

MFC after: 3 days

19 years agoNew release notes:
hrs [Wed, 6 Apr 2005 19:58:22 +0000 (19:58 +0000)]
New release notes:
debug.mpsafevm=1 by default on sparc64,
ixgb(4) MPSAFE,
core dump of natd(8) w/ -reverse fixed,
libarchive zip support improved, and
rexecd(8) root login problem fixed.

19 years agoIn ng_btsocket_rfcomm_receive_frame() correctly set length variable when
emax [Wed, 6 Apr 2005 18:55:58 +0000 (18:55 +0000)]
In ng_btsocket_rfcomm_receive_frame() correctly set length variable when
EA bit is set in hdr->length (16-bit length). This currently has no effect
on the rest of the code. It just fixes the debug message.

MFC After: 3 weeks

19 years agoBump &release.manpath.netbsd;.
hrs [Wed, 6 Apr 2005 18:30:34 +0000 (18:30 +0000)]
Bump &release.manpath.netbsd;.

19 years agoNew release notes: FreeBSD-SA-05:03.amd64.
hrs [Wed, 6 Apr 2005 18:28:54 +0000 (18:28 +0000)]
New release notes: FreeBSD-SA-05:03.amd64.

19 years agoInclude gettytab.h before extern.h so that the declarations of struct
stefanf [Wed, 6 Apr 2005 17:42:24 +0000 (17:42 +0000)]
Include gettytab.h before extern.h so that the declarations of struct
gettyflags, gettynums and gettystrs are available.

19 years ago- we are no longer shareing any resources to be locked between
ume [Wed, 6 Apr 2005 15:36:34 +0000 (15:36 +0000)]
- we are no longer shareing any resources to be locked between
  getaddrinfo(3) and getipnodeby*(3).
- use definitions in reentrant.h.
- remove obsolete comment.

19 years agoMUTEX_INITIALIZER should be used instead of
ume [Wed, 6 Apr 2005 15:16:04 +0000 (15:16 +0000)]
MUTEX_INITIALIZER should be used instead of
PTHREAD_MUTEX_INITIALIZER, here.

19 years agoWhen a packet has been reinjected into ipfw(4) after dummynet(4) processing
glebius [Wed, 6 Apr 2005 14:00:33 +0000 (14:00 +0000)]
When a packet has been reinjected into ipfw(4) after dummynet(4) processing
we have a non-NULL args.rule. If the same packet later is subject to "tee"
rule, its original is sent again into ipfw_chk() and it reenters at the same
rule. This leads to infinite loop and frozen router.

Assign args.rule to NULL, any time we are going to send packet back to
ipfw_chk() after a tee rule. This is a temporary workaround, which we
will leave for RELENG_5. In HEAD we are going to make divert(4) save
next rule the same way as dummynet(4) does.

PR: kern/79546
Submitted by: Oleg Bulyzhin
Reviewed by: maxim, andre
MFC after: 3 days

19 years agoRemove unique id field which is no longer used by debugger.
davidxu [Wed, 6 Apr 2005 13:57:31 +0000 (13:57 +0000)]
Remove unique id field which is no longer used by debugger.

19 years agoRemove debug symbol from installed library, one can always
davidxu [Wed, 6 Apr 2005 13:48:11 +0000 (13:48 +0000)]
Remove debug symbol from installed library, one can always
use compiled version in libthr directory.

19 years agoseparate gai_strerror(3) from getaddrinfo.c.
ume [Wed, 6 Apr 2005 12:45:51 +0000 (12:45 +0000)]
separate gai_strerror(3) from getaddrinfo.c.

Requested by: phantom

19 years agoAdd support for controllers that doesn't have the usual taskfile
sos [Wed, 6 Apr 2005 10:22:56 +0000 (10:22 +0000)]
Add support for controllers that doesn't have the usual taskfile
layout.  No functional changes.

19 years agoConstify hexdump() harder.
phk [Wed, 6 Apr 2005 10:14:13 +0000 (10:14 +0000)]
Constify hexdump() harder.

19 years ago - Remove dead code.
jeff [Wed, 6 Apr 2005 10:11:14 +0000 (10:11 +0000)]
 - Remove dead code.

19 years ago - Assert that the bufobj matches in flushbuflists. I still haven't gotten
jeff [Wed, 6 Apr 2005 06:49:46 +0000 (06:49 +0000)]
 - Assert that the bufobj matches in flushbuflists.  I still haven't gotten
   to root cause on exactly how this happens.
 - If the assert is disabled, we presently try to handle this case, but the
   BUF_UNLOCK was missing.  Thus, if this condition ever hit we would leak
   a buf lock.

Many thanks to Peter Holm for all his help in finding this bug.  He really
put more effort into it than I did.

19 years agoA number of improvements to ZIP support.
kientzle [Wed, 6 Apr 2005 04:19:30 +0000 (04:19 +0000)]
A number of improvements to ZIP support.
  * Handles entries with compressed size >2GB (signed/unsigned cleanup)
  * Handles entries with compressed size >4GB ("ZIP64" extension)
  * Handles Unix extensions (ctime, atime, mtime, mode, uid, etc)
  * Format-specific "skip data" override allows ZIP reader to skip
    entries without decompressing them, which makes "tar -t"
    a lot faster.
  * Handles "length-at-end" entries generated by, e.g., "zip -r - foo"

Many thanks to: Dan Nelson, who contributed the code and test files for
   the first three items above and suggested the fourth.

19 years agoMerge from RELENG_5: invert WITH_LIB32 into NO_LIB32 so that we build the
peter [Wed, 6 Apr 2005 01:55:43 +0000 (01:55 +0000)]
Merge from RELENG_5: invert WITH_LIB32 into NO_LIB32 so that we build the
32 bit libraries by default.  Define NO_LIB32 in make.conf if you do not
want this.

19 years agoFix the lib32 build for amd64 after v1.238 of src/include/Makefile (imp).
peter [Wed, 6 Apr 2005 01:41:08 +0000 (01:41 +0000)]
Fix the lib32 build for amd64 after v1.238 of src/include/Makefile (imp).
That change exposed the fact that I'd forgotten to set $MACHINE here.

19 years agoFully initialize the required TSS fields so that the io permission
cperciva [Wed, 6 Apr 2005 01:05:51 +0000 (01:05 +0000)]
Fully initialize the required TSS fields so that the io permission
bitmap is set correctly.

Patch from: peter
Security: FreeBSD-SA-05:03.amd64

19 years agoUse ACTION_PTR(r) instead of (r->cmd + r->act_ofs).
brooks [Wed, 6 Apr 2005 00:26:08 +0000 (00:26 +0000)]
Use ACTION_PTR(r) instead of (r->cmd + r->act_ofs).

Reviewed by: md5

19 years agoNew release note: SA-05:02.sendfile (+MFC).
bmah [Wed, 6 Apr 2005 00:10:19 +0000 (00:10 +0000)]
New release note:  SA-05:02.sendfile (+MFC).

19 years agoMake dummynet_flush() match its prototype.
brooks [Tue, 5 Apr 2005 23:38:16 +0000 (23:38 +0000)]
Make dummynet_flush() match its prototype.

19 years agoFix a long-standing bug in k_rem_pio2(), which led to large errors when
das [Tue, 5 Apr 2005 23:27:47 +0000 (23:27 +0000)]
Fix a long-standing bug in k_rem_pio2(), which led to large errors when
tanf() was called with big arguments close to multiples of pi/2.

Reported by: ucbtest via bde

19 years agoFix 32 bit signals on amd64. It turns out that I was sign extending
peter [Tue, 5 Apr 2005 22:41:49 +0000 (22:41 +0000)]
Fix 32 bit signals on amd64.  It turns out that I was sign extending
the register values coming back from sigreturn(2).  Normally this wouldn't
matter because the 32 bit environment would truncate the upper 32 bits
and re-save the truncated values at the next trap.  However, if we got
a fast second signal and it was pending while we were returning from
sigreturn(2) in the signal trampoline, we'd never have had a chance to
truncate the bogus values in 32 bit mode, and the new sendsig would get
an EFAULT when trying to write to the bogus user stack address.

19 years agoDocument the configuration files used.
jcamou [Tue, 5 Apr 2005 22:29:47 +0000 (22:29 +0000)]
Document the configuration files used.

PR: docs/79572
Submitted by: Josh Paetzel <josh@tcbug.org>
Approved by:  trhodes (mentor)

19 years agoBe more specific when complaining about bit masks.
brooks [Tue, 5 Apr 2005 22:26:02 +0000 (22:26 +0000)]
Be more specific when complaining about bit masks.

19 years agoAdd support for FTDI's FT2232C twin channel chip.
ticso [Tue, 5 Apr 2005 22:09:18 +0000 (22:09 +0000)]
Add support for FTDI's FT2232C twin channel chip.

Tested by: Mike Tancsa <mike@sentex.net>

19 years agoremove usbd_errstr(err) from printf.
ticso [Tue, 5 Apr 2005 22:03:07 +0000 (22:03 +0000)]
remove usbd_errstr(err) from printf.
err has nothing to say about this specific error.

19 years agoDon't free the _PSS buffer until after we check to see if we have no valid
jhb [Tue, 5 Apr 2005 19:39:44 +0000 (19:39 +0000)]
Don't free the _PSS buffer until after we check to see if we have no valid
states as otherwise we will try to free the buffer twice.

19 years ago'dumpon' can run before 'initrandom' so make it.
obrien [Tue, 5 Apr 2005 18:59:24 +0000 (18:59 +0000)]
'dumpon' can run before 'initrandom' so make it.
This gives a better chance of debugging /dev/random related panics.

19 years agoCorrect type mismatch introduced in last commit.
nectar [Tue, 5 Apr 2005 18:25:27 +0000 (18:25 +0000)]
Correct type mismatch introduced in last commit.

Noticed by: Steve Kargl <sgk@troutmask.apl.washington.edu>

19 years agomake yp stuff re-entrant.
ume [Tue, 5 Apr 2005 18:25:23 +0000 (18:25 +0000)]
make yp stuff re-entrant.

Obtained from: NetBSD

19 years agoprotect _yp_domain with mutex lock.
ume [Tue, 5 Apr 2005 18:07:59 +0000 (18:07 +0000)]
protect _yp_domain with mutex lock.

Inspired by: NetBSD

19 years agoOops, revert a commit that snuck in with the earlier critical section
jhb [Tue, 5 Apr 2005 18:06:28 +0000 (18:06 +0000)]
Oops, revert a commit that snuck in with the earlier critical section
changes.

Noticed by: bde

19 years agoRelease the IRQ resource on detach. This fixes repeatedly loading and
marius [Tue, 5 Apr 2005 17:51:56 +0000 (17:51 +0000)]
Release the IRQ resource on detach. This fixes repeatedly loading and
unloading the hme(4) module in case it attaches to sbus(4).

MFC after: 1 week

19 years ago- Update and expand description of node operation.
glebius [Tue, 5 Apr 2005 17:27:36 +0000 (17:27 +0000)]
- Update and expand description of node operation.
- Cut section IMPLELEMENTATION NOTES. It doesn't apply to
  modern FreeBSD.

Sponsored by: Rambler

19 years agoMajor overhaul and cleanup of ng_source node.
glebius [Tue, 5 Apr 2005 17:22:05 +0000 (17:22 +0000)]
Major overhaul and cleanup of ng_source node.

Functional changes:
- Cut struct source_hookinfo. Just use hook_p pointer.
- Remove "start_now" command. "start" command now requires number of
  packets to send as argument. "start" command actually starts sending.
  Move the code that actually starts sending from ng_source_rcvmsg()
  to ng_source_start().
- Remove check for NG_SOURCE_ACTIVE in ng_source_stop(). We can be called
  with flag cleared (see begin of ng_source_intr()).
- If NG_SEND_DATA_ONLY() use log(LOG_DEBUG) instead of printf(). Otherwise
  we will *flood* console.
- Add ng_connect_t method, which sends NGM_ETHER_GET_IFNAME command
  to "output" hook. Cut ng_source_request_output_ifp(). Refactor
  ng_source_store_output_ifp() to use ifunit() and don't muck through
  interface list.
- Add "setiface" command, which gives ability to configure interface
  in case when ng_source_connect() failed. This happens, when we are not
  connected directly to ng_ether(4) node.
- Remove KASSERTs, which can never fire.
- Don't check for M_PKTHDR in rcvdata method. netgraph(4) does this
  for us.

Style:
- Assign sc_p = NG_NODE_PRIVATE(node) in declaration, to be
  consistent with style of other nodes.
- Sort variables.
- u_intXX -> uintXX.
- Dots at ends of comments.

Sponsored by:   Rambler

19 years agoadd missing mutex unlock.
ume [Tue, 5 Apr 2005 17:13:28 +0000 (17:13 +0000)]
add missing mutex unlock.

19 years agoFix a change in a debug printf I missed in an earlier commit.
jhb [Tue, 5 Apr 2005 15:28:06 +0000 (15:28 +0000)]
Fix a change in a debug printf I missed in an earlier commit.

19 years agoFix a use-after-free problem in atapi_cb(). Add some necessary synchronization
scottl [Tue, 5 Apr 2005 15:08:19 +0000 (15:08 +0000)]
Fix a use-after-free problem in atapi_cb().  Add some necessary synchronization
to the XPT_PATH_INQ op.  Don't leak locks on failure in XPT_SCSIIO.  Correctly
fix the CAMDEBUG message.

19 years agoDES pointed out that the PAM layer may change the target user name
nectar [Tue, 5 Apr 2005 14:55:33 +0000 (14:55 +0000)]
DES pointed out that the PAM layer may change the target user name
during authentication.  Thus we need to call getpwnam *after* the user
has been authenticated.  Colin mentioned that we should also move the
check for root in that case.

19 years agoAdd firstshot support for Acer Labs Inc SATA parts
sos [Tue, 5 Apr 2005 14:51:43 +0000 (14:51 +0000)]
Add firstshot support for Acer Labs Inc SATA parts
Update ALi PATA support to handle ATA133 as well

19 years agonatd core dumps when -reverse switch is used because of a bug in
phk [Tue, 5 Apr 2005 13:04:35 +0000 (13:04 +0000)]
natd core dumps when -reverse switch is used because of a bug in
libalias.

In /usr/src/lib/libalias/alias.c, the functions LibAliasIn and
LibAliasOutTry call the legacy PacketAliasIn/PacketAliasOut instead
of LibAliasIn/LibAliasOut when the PKT_ALIAS_REVERSE option is set.
In this case, the context variable "la" gets lost because the legacy
compatibility routines expect "la" to be global.  This was obviously
an oversight when rewriting the PacketAlias* functions to the
LibAlias* functions.

The fix (as shown in the patch below) is to remove the legacy
subroutine calls and replace with the new ones using the "la" struct
as the first arg.

Submitted by: Gil Kloepfer <fgil@kloepfer.org>
Confirmed by: <nicolai@catpipe.net>
PR: 76839
MFC after: 3 days

19 years agoThere is not much sense in having an else after Punt() and _exit() which
harti [Tue, 5 Apr 2005 12:33:54 +0000 (12:33 +0000)]
There is not much sense in having an else after Punt() and _exit() which
never return. Also use pid_t for variables holding PIDs.

19 years agoPrint thread address.
davidxu [Tue, 5 Apr 2005 11:40:58 +0000 (11:40 +0000)]
Print thread address.

PR: threads/76821

19 years agoExport thread address.
davidxu [Tue, 5 Apr 2005 11:38:30 +0000 (11:38 +0000)]
Export thread address.

19 years agoPush some faulty debugging aside so LINT compiles.
phk [Tue, 5 Apr 2005 09:42:52 +0000 (09:42 +0000)]
Push some faulty debugging aside so LINT compiles.

19 years ago - Move NDFREE() from vfs_subr to vfs_lookup where namei() is.
jeff [Tue, 5 Apr 2005 08:58:49 +0000 (08:58 +0000)]
 - Move NDFREE() from vfs_subr to vfs_lookup where namei() is.

19 years ago - Use taskqueue_thread rather than taskqueue_swi since our task is going
jeff [Tue, 5 Apr 2005 08:51:45 +0000 (08:51 +0000)]
 - Use taskqueue_thread rather than taskqueue_swi since our task is going
   to vrele, which may vop lock.  This is not safe in a software interrupt
   context.

19 years ago - Consistently call 'vp' vp rather than ovp sometimes in ffs_truncate().
jeff [Tue, 5 Apr 2005 08:49:41 +0000 (08:49 +0000)]
 - Consistently call 'vp' vp rather than ovp sometimes in ffs_truncate().
   Do the same for oip.

Pointed out by: glebius

19 years agoPut parantheses into 'if (foo && bar & flag)'. While they are not strictly
harti [Tue, 5 Apr 2005 08:14:40 +0000 (08:14 +0000)]
Put parantheses into 'if (foo && bar & flag)'. While they are not strictly
needed, they are useful for the human reader.

19 years agoRemove lots if irritating parantheses and fix two comments.
harti [Tue, 5 Apr 2005 08:09:48 +0000 (08:09 +0000)]
Remove lots if irritating parantheses and fix two comments.

Checked by: md5

19 years agoRemove the definition for STATIC and just use static.
harti [Tue, 5 Apr 2005 07:44:11 +0000 (07:44 +0000)]
Remove the definition for STATIC and just use static.

19 years agoRemove unneccessary local prototypes.
harti [Tue, 5 Apr 2005 07:43:02 +0000 (07:43 +0000)]
Remove unneccessary local prototypes.

19 years agoThe static tfile is used only in one single function so move it into that
harti [Tue, 5 Apr 2005 07:40:40 +0000 (07:40 +0000)]
The static tfile is used only in one single function so move it into that
function.

19 years agoThe tfile field of struct Job isn't used anymore since revision 1.22 so
harti [Tue, 5 Apr 2005 07:32:12 +0000 (07:32 +0000)]
The tfile field of struct Job isn't used anymore since revision 1.22 so
remove it now.

19 years agoRun em_local_timer() once per second instead of running it once per 2 seconds.
glebius [Tue, 5 Apr 2005 07:06:47 +0000 (07:06 +0000)]
Run em_local_timer() once per second instead of running it once per 2 seconds.
This makes gathering of error stats more precise, and netstat(1) output look
right.

Reviewed by: tackerman

19 years agoAdd locking and enable INTR_MPSAFE. Eliminate spl calls.
alc [Tue, 5 Apr 2005 05:05:29 +0000 (05:05 +0000)]
Add locking and enable INTR_MPSAFE.  Eliminate spl calls.

Submitted by: Paul Willmann willmann at rice dot edu
MFC After: 3 weeks

19 years agoBuild exp2(), exp2f(), and related documentation.
das [Tue, 5 Apr 2005 02:57:39 +0000 (02:57 +0000)]
Build exp2(), exp2f(), and related documentation.

19 years agoDocument exp2() and exp2f(), and make other minor tweaks and updates.
das [Tue, 5 Apr 2005 02:57:28 +0000 (02:57 +0000)]
Document exp2() and exp2f(), and make other minor tweaks and updates.

19 years agoImplement exp2() and exp2f().
das [Tue, 5 Apr 2005 02:57:15 +0000 (02:57 +0000)]
Implement exp2() and exp2f().

19 years agoRe-enable the atapicam driver in NOTES.
scottl [Tue, 5 Apr 2005 02:05:38 +0000 (02:05 +0000)]
Re-enable the atapicam driver in NOTES.

19 years agoConnect the atapicam module to the build.
scottl [Tue, 5 Apr 2005 02:05:01 +0000 (02:05 +0000)]
Connect the atapicam module to the build.

19 years agoAdd a Makefile for atapi-cam.
scottl [Tue, 5 Apr 2005 02:04:30 +0000 (02:04 +0000)]
Add a Makefile for atapi-cam.

19 years agoCommit ATAPI-CAM for ATAmkIII. Improvements and changes:
scottl [Tue, 5 Apr 2005 02:03:31 +0000 (02:03 +0000)]
Commit ATAPI-CAM for ATAmkIII.  Improvements and changes:

- newbus plumbing.  Each atapicam bus is a child off of a parent ata channel
  bus.  This is somewhat of a hack, but allows the ata core to be completely
  free of atapicam knowledge.
- No more global lists of softc's and no more groping around in internal ata
  structures on each command.
- Giant-free operation of the completion handler.
- Per-bus mutex for protecting the busy list and synchronizing detach.
- Lots of streamlining and dead code elimination, better adherence to the
  CAM locking protocol.

This feature still requires that the appropriate atapi-* driver be present
for each atapi device that you want to talk to (i.e. atapi-cd for cdroms).
It does work both compiled into the kernel and as a loadable module.

Reviewed by: thomas, sos

19 years agoAssert that the vnode is locked. This is meant to catch bugs or
csjp [Tue, 5 Apr 2005 01:11:43 +0000 (01:11 +0000)]
Assert that the vnode is locked. This is meant to catch bugs or
mis-use of the vnode API in conditions where IO_NODELOCKED has been
used without the vnode actually being locked.

19 years agoPass exact number of threads.
davidxu [Mon, 4 Apr 2005 23:43:53 +0000 (23:43 +0000)]
Pass exact number of threads.

19 years agoWhen reloading rules via rc.d/pf, flush everything but existing state
seanc [Mon, 4 Apr 2005 23:06:10 +0000 (23:06 +0000)]
When reloading rules via rc.d/pf, flush everything but existing state
entries that way when rules are read in, it doesn't break established
connections.

Approved by: mlaier
Reviewed by: rc
MFC after: 3 weeks

19 years agoDivorce critical sections from spinlocks. Critical sections as denoted by
jhb [Mon, 4 Apr 2005 21:53:56 +0000 (21:53 +0000)]
Divorce critical sections from spinlocks.  Critical sections as denoted by
critical_enter() and critical_exit() are now solely a mechanism for
deferring kernel preemptions.  They no longer have any affect on
interrupts.  This means that standalone critical sections are now very
cheap as they are simply unlocked integer increments and decrements for the
common case.

Spin mutexes now use a separate KPI implemented in MD code: spinlock_enter()
and spinlock_exit().  This KPI is responsible for providing whatever MD
guarantees are needed to ensure that a thread holding a spin lock won't
be preempted by any other code that will try to lock the same lock.  For
now all archs continue to block interrupts in a "spinlock section" as they
did formerly in all critical sections.  Note that I've also taken this
opportunity to push a few things into MD code rather than MI.  For example,
critical_fork_exit() no longer exists.  Instead, MD code ensures that new
threads have the correct state when they are created.  Also, we no longer
try to fixup the idlethreads for APs in MI code.  Instead, each arch sets
the initial curthread and adjusts the state of the idle thread it borrows
in order to perform the initial context switch.

This change is largely a big NOP, but the cleaner separation it provides
will allow for more efficient alternative locking schemes in other parts
of the kernel (bare critical sections rather than per-CPU spin mutexes
for per-CPU data for example).

Reviewed by: grehan, cognet, arch@, others
Tested on: i386, alpha, sparc64, powerpc, arm, possibly more

19 years agoInclude <string.h> and <strings.h> for various functions.
stefanf [Mon, 4 Apr 2005 21:19:48 +0000 (21:19 +0000)]
Include <string.h> and <strings.h> for various functions.

19 years agoFlip the switch and turn mpsafevm on by default for sparc64.
jhb [Mon, 4 Apr 2005 20:59:02 +0000 (20:59 +0000)]
Flip the switch and turn mpsafevm on by default for sparc64.

Approved by: alc

19 years agomake _files_getaddrinfo() re-entrant.
ume [Mon, 4 Apr 2005 19:45:27 +0000 (19:45 +0000)]
make _files_getaddrinfo() re-entrant.

Obtained from: NetBSD

19 years agoDon't leak mutex on open failure.
phk [Mon, 4 Apr 2005 17:37:35 +0000 (17:37 +0000)]
Don't leak mutex on open failure.

19 years agoAdd support for _PDC/_OSC by advertising that we support direct access to
njl [Mon, 4 Apr 2005 15:51:13 +0000 (15:51 +0000)]
Add support for _PDC/_OSC by advertising that we support direct access to
the PERF_CTL/STS MSRs via the new acpi_get_features() method.  This should
allow newer systems to use SpeedStep.

19 years agoAdd the acpi_get_features() method. This method is called on child drivers
njl [Mon, 4 Apr 2005 15:46:57 +0000 (15:46 +0000)]
Add the acpi_get_features() method.  This method is called on child drivers
to see what features they may support before calling identify/probe/attach.
This is necessary because the ACPI 3.0 spec requires driver support be
advertised before running any methods.  For now, the flags are as specified
in for the _PDC and _OSC methods but we can support private flags as needed.

Add an implementation of this for acpi_cpu.  It checks all its children
(notably cpufreq drivers) and calls the _PDC method to report the results.

19 years agoDocument that devclass_get_maxunit(9) returns one greater than the current
njl [Mon, 4 Apr 2005 15:37:59 +0000 (15:37 +0000)]
Document that devclass_get_maxunit(9) returns one greater than the current
highest unit.

Reviewed by: dfr
MFC after: 2 weeks

19 years agoAdd devclass_get_drivers(9) which provides an array of pointers to driver
njl [Mon, 4 Apr 2005 15:26:51 +0000 (15:26 +0000)]
Add devclass_get_drivers(9) which provides an array of pointers to driver
instances in a given devclass.  This is useful for systems that want to
call code in driver static methods, similar to device_identify().

Reviewed by: dfr
MFC after: 2 weeks

19 years ago - Add a missing unlock of the vnode_free_list_mtx.
jeff [Mon, 4 Apr 2005 12:07:16 +0000 (12:07 +0000)]
 - Add a missing unlock of the vnode_free_list_mtx.

Spotted by: Antoine Brodin

19 years ago - Instead of waiting forever to get a vnode in getnewvnode() wait for
jeff [Mon, 4 Apr 2005 11:43:44 +0000 (11:43 +0000)]
 - Instead of waiting forever to get a vnode in getnewvnode() wait for
   one to become available for one second and then return ENFILE.  We
   can run out of vnodes, and there must be a hard limit because without
   one we can quickly run out of KVA on x86.  Presently the system can
   deadlock if there are maxvnodes directories in the namecache.  The
   original 4.x BSD behavior was to return ENFILE if we reached the max,
   but 4.x BSD did not have the vnlru proc so it was less profitable to
   wait.