]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
20 years agoFixed loss of setting of the RB_BOOTINFO flag in rev.1.43. Fixed wrong
Bruce Evans [Mon, 10 Nov 2003 19:06:09 +0000 (19:06 +0000)]
Fixed loss of setting of the RB_BOOTINFO flag in rev.1.43.  Fixed wrong
comment about this flag in rev.1.61.  It is not historical like the
comment said; it is the flag that says that most of what is laboriously
put in the bootinfo struct is actually there.  Newer kernels were
bootable by even the broken boot2 without losing anything except the
symbol table, but older kernels need at least the memory sizes.

Restoring the "|" with RB_BOOTINFO that was lost in rev.1.43 costs 5
bytes.  The fix can be done in only 4 bytes by fixing some code that
was removed in rev.1.61 (put RB_BOOTINFO back in in the initial value
of "opts" and fix RBX_MASK to not clobber it.)

20 years agoRemove a comment stating that -pthread isn't supported.
Daniel Eischen [Mon, 10 Nov 2003 18:52:57 +0000 (18:52 +0000)]
Remove a comment stating that -pthread isn't supported.

20 years agoMFCs noted: sendmail-8.12.10, tzdata2003d.
Bruce A. Mah [Mon, 10 Nov 2003 16:27:53 +0000 (16:27 +0000)]
MFCs noted:  sendmail-8.12.10, tzdata2003d.

20 years agoAdd a missing 'the'
Christian Brueffer [Mon, 10 Nov 2003 16:04:48 +0000 (16:04 +0000)]
Add a missing 'the'

20 years agoDon't establish the ISR in the sn_activate routine. I've had two
Warner Losh [Mon, 10 Nov 2003 16:04:11 +0000 (16:04 +0000)]
Don't establish the ISR in the sn_activate routine.  I've had two
crashes that had sn0 as the irq that's being serviced, when there was
no sn0 in the system.  This seems to prevent them.  Also, we want to
wait until after we've registered with the network layer before we
turn on the interrupt spigot to avoid races.

20 years agoUpdate a comment related to SMP and describe the NO_MIXED_MODE kernel
John Baldwin [Mon, 10 Nov 2003 15:54:32 +0000 (15:54 +0000)]
Update a comment related to SMP and describe the NO_MIXED_MODE kernel
option.

Requested by: bde

20 years agoUpdate a comment.
John Baldwin [Mon, 10 Nov 2003 15:48:30 +0000 (15:48 +0000)]
Update a comment.

Requested by: bde

20 years agoClarify the rule about structure typedefs being discouraged, following
Tony Finch [Mon, 10 Nov 2003 14:32:13 +0000 (14:32 +0000)]
Clarify the rule about structure typedefs being discouraged, following
a discussion on src-committers.

20 years agoAdd a describtion for the '-d' flag
Christian Brueffer [Mon, 10 Nov 2003 14:28:33 +0000 (14:28 +0000)]
Add a describtion for the '-d' flag

While I'm here, add a missing comma

PR: 41787
Obtained from: OpenBSD
MFC after: 5 days

20 years agoBased on an excellent suggestion from tanimura@ define I4BPRI and use it
Gary Jennejohn [Mon, 10 Nov 2003 14:20:34 +0000 (14:20 +0000)]
Based on an excellent suggestion from tanimura@ define I4BPRI and use it
in place of TTIPRI.

20 years agoReduce debug message.
Hidetoshi Shimokawa [Mon, 10 Nov 2003 14:04:37 +0000 (14:04 +0000)]
Reduce debug message.

20 years agoCorrect the include path so that the include files are found
Hartmut Brandt [Mon, 10 Nov 2003 14:04:34 +0000 (14:04 +0000)]
Correct the include path so that the include files are found
with include "...".

20 years agoFix breakage cuased by the selwakeuppri commit by defining TTIPRI for
Gary Jennejohn [Mon, 10 Nov 2003 14:02:22 +0000 (14:02 +0000)]
Fix breakage cuased by the selwakeuppri commit by defining TTIPRI for
recent versions of FreeBSD (based on __FreeBSD_version check).

20 years agoFix the build by adjusting the include directives. The problem was
Hartmut Brandt [Mon, 10 Nov 2003 13:43:26 +0000 (13:43 +0000)]
Fix the build by adjusting the include directives. The problem was
that the sources use "..." includes to get at include files that
later on reside in an include sub-directory.

20 years agoDon't compile with -g by default; there's a better way to build modules with
Ruslan Ermilov [Mon, 10 Nov 2003 12:54:54 +0000 (12:54 +0000)]
Don't compile with -g by default; there's a better way to build modules with
debug support.

20 years agoMdoc Janitor:
Hiten Pandya [Mon, 10 Nov 2003 12:45:10 +0000 (12:45 +0000)]
Mdoc Janitor:

  * Add MLINKS for:

-> Soundblaster emu10k1(4) Driver [points to pcm(4)]
-> Avance Logic ALS400 Driver [points to pcm(4)]

    We should not need separate manual page for each of these
    drivers; instead, linking them to pcm(4) manual page is
    simpler, and new device lists can be easily added to the
    said manual page.

  * While I am here, sort out mdoc(7) entries in ${MAN}.

20 years agoenable aes-xcbc-mac and aes-ctr, again.
Hajimu UMEMOTO [Mon, 10 Nov 2003 10:39:14 +0000 (10:39 +0000)]
enable aes-xcbc-mac and aes-ctr, again.

20 years agoRB_EMPTY() applies to red-black trees, not splay trees.
Dag-Erling Smørgrav [Mon, 10 Nov 2003 10:35:06 +0000 (10:35 +0000)]
RB_EMPTY() applies to red-black trees, not splay trees.

20 years agorijndael-alg-fst.[ch]:
Hajimu UMEMOTO [Mon, 10 Nov 2003 10:33:39 +0000 (10:33 +0000)]
rijndael-alg-fst.[ch]:
  - redo updating.

rijndael-api-fst.[ch]:
  - switch to use new low level rijndael api.
  - stop using u8, u16 and u32.
  - space cleanup.

Tested by: gbde(8) and phk's test program

20 years agoWhen the P flag is set (i.e. Overwrite regular files before deleting them),
Guido van Rooij [Mon, 10 Nov 2003 09:40:18 +0000 (09:40 +0000)]
When the P flag is set (i.e. Overwrite regular files before deleting them),
do only unlink the file if we could indeed overwrite the file.
Old behaviour: rm -P /tmp/foo (foo mode 0444) would NOT overwrite foo,
but still delete it (with a warning: rm: foo: Permission denied)
New behaviour: Just the EPERM warning, but no deletion

Reviewed by: bde

20 years agoLink the SNMP daemon for NgATM to the build.
Hartmut Brandt [Mon, 10 Nov 2003 09:23:09 +0000 (09:23 +0000)]
Link the SNMP daemon for NgATM to the build.

20 years agoAdd the build infrastructure for the SNMP daemon that is the base
Hartmut Brandt [Mon, 10 Nov 2003 09:22:40 +0000 (09:22 +0000)]
Add the build infrastructure for the SNMP daemon that is the base
for the NgATM ILMI daemon and for the tree parsing helper program.

20 years agoLink libbsnmp to the build.
Hartmut Brandt [Mon, 10 Nov 2003 09:18:04 +0000 (09:18 +0000)]
Link libbsnmp to the build.

20 years agoAdd build infrastructure for libbsnmp and the modules. Add the
Hartmut Brandt [Mon, 10 Nov 2003 09:17:34 +0000 (09:17 +0000)]
Add build infrastructure for libbsnmp and the modules. Add the
netgraph module for the SNMP daemon.

20 years agoAdd a define for libbsnmp.
Hartmut Brandt [Mon, 10 Nov 2003 09:07:30 +0000 (09:07 +0000)]
Add a define for libbsnmp.

20 years agoCreate sub-directories under share for bsnmp tree definition and MIB files.
Hartmut Brandt [Mon, 10 Nov 2003 09:06:38 +0000 (09:06 +0000)]
Create sub-directories under share for bsnmp tree definition and MIB files.

20 years agoCreate a subdirectory where bsnmp include files will be stuffed to.
Hartmut Brandt [Mon, 10 Nov 2003 09:04:24 +0000 (09:04 +0000)]
Create a subdirectory where bsnmp include files will be stuffed to.

20 years agoMake two directories under usr/share/bsnmp that are needed during
Hartmut Brandt [Mon, 10 Nov 2003 09:00:08 +0000 (09:00 +0000)]
Make two directories under usr/share/bsnmp that are needed during
library build to install tree definition files and MIBS.

Okay'ed by: ru

20 years agoAdd update instructions for bsnmp. Bsnmp is the base for the NgATM
Hartmut Brandt [Mon, 10 Nov 2003 08:56:56 +0000 (08:56 +0000)]
Add update instructions for bsnmp. Bsnmp is the base for the NgATM
ILMI daemon.

20 years agoThis commit was generated by cvs2svn to compensate for changes in r122394,
Hartmut Brandt [Mon, 10 Nov 2003 08:53:38 +0000 (08:53 +0000)]
This commit was generated by cvs2svn to compensate for changes in r122394,
which included commits to RCS files with non-trunk default branches.

20 years agoVirgin import of bsnmp 1.4
Hartmut Brandt [Mon, 10 Nov 2003 08:53:38 +0000 (08:53 +0000)]
Virgin import of bsnmp 1.4

20 years agoThe current Perl 5.8 version is 5.8.1; 5.8.2 is right around the corner.
Dag-Erling Smørgrav [Mon, 10 Nov 2003 08:28:22 +0000 (08:28 +0000)]
The current Perl 5.8 version is 5.8.1; 5.8.2 is right around the corner.

20 years agoDocument EINVAL for len == 0.
Dag-Erling Smørgrav [Mon, 10 Nov 2003 08:02:05 +0000 (08:02 +0000)]
Document EINVAL for len == 0.

20 years agoWhen there are no free sem_undo structs available in semu_alloc(), only
Tim J. Robbins [Mon, 10 Nov 2003 07:22:41 +0000 (07:22 +0000)]
When there are no free sem_undo structs available in semu_alloc(), only
free one sem_undo with un_cnt == 0 instead of all of them. This is a
temporary workaround until the SLIST_FOREACH_PREVPTR loop gets fixed so
that it doesn't cause cycles in semu_list when removing multiple adjacent
items. It might be easier to just use (doubly-linked) LISTs here instead
of complicated SLIST code to achieve O(1) removals.

This bug manifested itself as a complete lockup under heavy semaphore use
by multiple processes with the SEM_UNDO flag set.

PR: 58984

20 years agoIn get_mcontext(), do not update bspstore and ndirty in the trapframe.
Marcel Moolenaar [Mon, 10 Nov 2003 05:28:05 +0000 (05:28 +0000)]
In get_mcontext(), do not update bspstore and ndirty in the trapframe.
Only update them in the newly created context to reflect the state
after copying the dirty registers onto the user stack. If we were to
update the trapframe, we lose the state at entry into the kernel. We
may need that after we create the context, such as for KSE upcalls.

We have to update the trapframe after writing the dirty registers to
the user stack for signal delivery to work. But this is best done in
sendsig() itself where it applies, not in get_mcontext() where it's
done unconditionally.

20 years agoDocument KTR_DROP.
Joseph Koshy [Mon, 10 Nov 2003 04:34:51 +0000 (04:34 +0000)]
Document KTR_DROP.

20 years ago* Improve sbp device probe in boot process.
Hidetoshi Shimokawa [Mon, 10 Nov 2003 03:51:23 +0000 (03:51 +0000)]
* Improve sbp device probe in boot process.
- Notify BUS RESET to CAM in sbp_attach().
- Use last bus reset time to determine login delay.

Tested by: imura

* Add some sysctl MIB and tunables.

20 years agoIf a thread masks all its signal, in cursig(), no signal will be exchanged
David Xu [Mon, 10 Nov 2003 03:11:08 +0000 (03:11 +0000)]
If a thread masks all its signal, in cursig(), no signal will be exchanged
with debugger, so testing P_TRACED in SIGPENDING is useless. This test also
is the culprit which causes lots of 'failed to set signal flags properly for
 ast()' to be printed on console which is just a false complaint.

20 years agoMove 'guessing' code from the probe into the identify routine where it more
Warner Losh [Mon, 10 Nov 2003 02:47:11 +0000 (02:47 +0000)]
Move 'guessing' code from the probe into the identify routine where it more
properly belongs.

20 years ago - The Open Group Base Specifications Issue 6 specifies that an munmap(2)
Alan Cox [Mon, 10 Nov 2003 01:37:40 +0000 (01:37 +0000)]
 - The Open Group Base Specifications Issue 6 specifies that an munmap(2)
   must return EINVAL if size is zero.  Submitted by: tegge
 - In order to avoid a race condition in multithreaded applications, the
   check and removal operations by munmap(2) must be in the same critical
   section.  To accomodate this, vm_map_check_protection() is modified to
   require its caller to obtain at least a read lock on the map.

20 years agoNFC: Update stale comments.
Jonathan Mini [Mon, 10 Nov 2003 00:44:00 +0000 (00:44 +0000)]
NFC: Update stale comments.

Reviewed by: alc

20 years agoMFNetBSD:
Josef Karthauser [Mon, 10 Nov 2003 00:20:52 +0000 (00:20 +0000)]
MFNetBSD:
    date: 2003/10/18 04:50:35;  author: simonb
    Remove assigned-to but otherwise unused variables.
    Remove unreachable break after return statements.

20 years agoMFNetBSD:
Josef Karthauser [Mon, 10 Nov 2003 00:16:36 +0000 (00:16 +0000)]
MFNetBSD:
    revision 1.142
    date: 2003/10/11 03:04:26;  author: toshii
    Fix a done list handling bug which exhibits under high shared
    interrupt rate and bus traffic.  As the interrupt register is
    read after checking hcca_done_head, there was a small chance
    of dropping a done list.  Ignore OHCI_WDH interrupt bit if
    hcca_done_head is zero so that OHCI_WDH is processed later.

20 years agoMFNetBSD:
Josef Karthauser [Mon, 10 Nov 2003 00:12:39 +0000 (00:12 +0000)]
MFNetBSD:
    revision 1.141
    date: 2003/09/10 20:08:29;  author: mycroft;
    Update actlen even in the case where a TD returns an error --
    this is critical for the umass bulk-only STALL case.

20 years agoMFNetBSD:
Josef Karthauser [Mon, 10 Nov 2003 00:08:41 +0000 (00:08 +0000)]
MFNetBSD:
    revision 1.176
    date: 2003/11/04 19:11:21;  author: mycroft;
    Ignore a CRCTO error on a SETUP transaction in combination with
    STALLED or NAK.  This fixes problems with the GL641.

20 years agoMFNetBSD:
Josef Karthauser [Sun, 9 Nov 2003 23:59:53 +0000 (23:59 +0000)]
MFNetBSD:
    date: 2003/09/12 16:18:08;  author: mycroft;
    Tweak a debugging printf().

20 years agoRevise the NetBSD revision control strings.
Josef Karthauser [Sun, 9 Nov 2003 23:56:19 +0000 (23:56 +0000)]
Revise the NetBSD revision control strings.

20 years agoMFNetBSD:
Josef Karthauser [Sun, 9 Nov 2003 23:54:21 +0000 (23:54 +0000)]
MFNetBSD:
- remove the unnecessary elm arg from SIMPLEQ_REMOVE_HEAD().
  this mirrors the functionality of SLIST_REMOVE_HEAD() (the other
  singly-linked list type) and FreeBSD's STAILQ_REMOVE_HEAD()

20 years agoReorganize and trim devices from the aic(4), aha(4), ahb(4), and
Bruce A. Mah [Sun, 9 Nov 2003 23:42:51 +0000 (23:42 +0000)]
Reorganize and trim devices from the aic(4), aha(4), ahb(4), and
ahc(4) sections.

Trim devices from sis(4), de(4), and ex(4).

Don't bother saying what standards are supported by the ATM stack;
this list is kind of misplaced here (and may not even be accurate
anymore).

Delete mention of pca; this driver has been removed.

20 years agofix typo that broke AP scanning by BSSID
Sam Leffler [Sun, 9 Nov 2003 23:36:46 +0000 (23:36 +0000)]
fix typo that broke AP scanning by BSSID

Submitted by: Len Widra

20 years agoWhen a thread is being swapped-out, save the high FP registers. We
Marcel Moolenaar [Sun, 9 Nov 2003 23:13:23 +0000 (23:13 +0000)]
When a thread is being swapped-out, save the high FP registers. We
have a pointer in the PCPU to the PCB of the thread that currently
has its high FP registers loaded.

20 years agoTrim device entries for sym(4)/ncr(4).
Simon L. B. Nielsen [Sun, 9 Nov 2003 23:08:05 +0000 (23:08 +0000)]
Trim device entries for sym(4)/ncr(4).

20 years agoAdd some pc98-specific devices from the hardware notes.
Bruce A. Mah [Sun, 9 Nov 2003 22:53:58 +0000 (22:53 +0000)]
Add some pc98-specific devices from the hardware notes.

20 years agoAdd two PC-98 devices, from the hardware notes, which was missed in
Simon L. B. Nielsen [Sun, 9 Nov 2003 22:36:04 +0000 (22:36 +0000)]
Add two PC-98 devices, from the hardware notes, which was missed in
the last commit.

MFC after: 1 week

20 years ago- Sync the list of supported device with the hardware notes and the
Simon L. B. Nielsen [Sun, 9 Nov 2003 22:21:21 +0000 (22:21 +0000)]
- Sync the list of supported device with the hardware notes and the
  driver source code.
- Use a mdoc list for the list of supported devices.

MFC after: 1 week

20 years agoUse get_mcontext() to construct the signal context in sendsig() and
Marcel Moolenaar [Sun, 9 Nov 2003 22:17:36 +0000 (22:17 +0000)]
Use get_mcontext() to construct the signal context in sendsig() and
use set_mcontext() to restore the context in sigreturn(). Since we
put the syscall number and the syscall arguments in the trapframe
(we don't save the scratch registers for syscalls, which allows us
to reuse the space to our advantage), create a MD specific flag so
that we save the scratch registers even for syscalls. We would not
be able to restart a syscall otherwise.

The signal trampoline does not need to flush the regiters anymore,
because get_mcontext() already handles that. In fact, if we set up
the context correctly, we do not need to have a trampoline at all.
This change however only minimally changes the trampoline code. In
follow-up commits this can be further optimized.

Note that normally we preserve cfm and iip in the trapframe created
by the EPC syscall path when we restore a context in set_mcontext()
because those fields are not normally set for a synchronuous context.
The kernel puts the return address and frame info of the syscall
stub in there. By preserving these fields we hide this detail from
userland which allows us to use setcontext(2) for user created
contexts. However, sigreturn() is commonly called from the trampoline,
which means that if we preserve cfm and iip in all cases, we would
return to the trampoline after the sigreturn(), which means we hit
the safety net: we call exit(2). So, we do not preserve cfm and iip
when we have a synchronous context that also has scratch registers
(the uncommon context created by sendsig() only), under the assumption
that if such a context is created in userland, something special is
going on and the use of cfm and iip is then just another quirk. All
this is invisible in the common case.

20 years ago - Remove Giant from msync(2). Giant is still acquired by the lower layers
Alan Cox [Sun, 9 Nov 2003 22:09:04 +0000 (22:09 +0000)]
 - Remove Giant from msync(2).  Giant is still acquired by the lower layers
   if we drop into the pmap or vnode layers.
 - Migrate the handling of zero-length msync(2)s into vm_map_sync() so that
   multithread applications can't change the map between implementing the
   zero-length hack in msync(2) and reacquiring the map lock in
   vm_map_sync().

Reviewed by: tegge

20 years agoFree the ata_request struct after we are done using it. This should stop
Scott Long [Sun, 9 Nov 2003 20:46:08 +0000 (20:46 +0000)]
Free the ata_request struct after we are done using it.  This should stop
the memory leak seen when using ATAPICAM.

20 years agomake minimum column size for interrupt name be the heading (depends upon
John-Mark Gurney [Sun, 9 Nov 2003 20:39:56 +0000 (20:39 +0000)]
make minimum column size for interrupt name be the heading (depends upon
Total being shorter than interrupt)..

Pointed out by: bde

20 years agoChange the clear_ret argument of get_mcontext() to be a flags argument.
Marcel Moolenaar [Sun, 9 Nov 2003 20:31:04 +0000 (20:31 +0000)]
Change the clear_ret argument of get_mcontext() to be a flags argument.
Since all callers either passed 0 or 1 for clear_ret, define bit 0 in
the flags for use as clear_ret. Reserve bits 1, 2 and 3 for use by MI
code for possible (but unlikely) future use. The remaining bits are for
use by MD code.

This change is triggered by a need on ia64 to have another knob for
get_mcontext().

20 years agoTweak the front end driver file list
Warner Losh [Sun, 9 Nov 2003 20:02:47 +0000 (20:02 +0000)]
Tweak the front end driver file list
Also some minor whitespace nits.

20 years agoAdd my copyright to aha_isa.c to make it crystal clear its license
Warner Losh [Sun, 9 Nov 2003 20:01:36 +0000 (20:01 +0000)]
Add my copyright to aha_isa.c to make it crystal clear its license

20 years agoLet bus space manage softc.
Warner Losh [Sun, 9 Nov 2003 19:51:16 +0000 (19:51 +0000)]
Let bus space manage softc.

20 years agoMinor tweaks to make it behave better:
Warner Losh [Sun, 9 Nov 2003 17:16:39 +0000 (17:16 +0000)]
Minor tweaks to make it behave better:
o When we're resetting the board, make sure that we error out the pending
  CCBs first.  Otherwise the aha_cmd won't accept further commands, such
  as those that are used to reset the card (AOP_INITIALIZE_MBOX).  This
  appears to cause a cascade failure where no more commands are possible
  to the card.
o Reduce from 10s down to 1s the amount of time we're willing to tolerate
  the card being awol.  This helps the above case.
o Add some error checking to two commands issued in the probe.

I have a dim memory of gibbs@ trying to tell me about this problem a
few years ago, so pointy hat to imp@ for sitting on it so long.

20 years agoMinor comment smithing
Warner Losh [Sun, 9 Nov 2003 17:05:55 +0000 (17:05 +0000)]
Minor comment smithing

20 years agoUse kern_sendit rather than sendit for the Linux send* syscalls.
David Malone [Sun, 9 Nov 2003 17:04:04 +0000 (17:04 +0000)]
Use kern_sendit rather than sendit for the Linux send* syscalls.
This means we can avoid using the stack gap for most send* syscalls
now (it is still used in the IP_HDRINCL case).

20 years agoSync the sym(4) manual page with the hardware notes, and change the
Simon L. B. Nielsen [Sun, 9 Nov 2003 14:58:28 +0000 (14:58 +0000)]
Sync the sym(4) manual page with the hardware notes, and change the
list of supported controllers into a list.

Note that the 53C875A has not been included in the list of supported
devices, since this controller does not seem to be supported by the
version of the sym(4) driver currently in FreeBSD.

PR: docs/55557
Submitted by: Lukas Ertl <l.ertl@univie.ac.at> (original version)
MFC after: 1 week
53C875A omission reviewed by: silence from -scsi and groudier

20 years agoIncrease MAXDEV up to 10.
Hidetoshi Shimokawa [Sun, 9 Nov 2003 14:49:43 +0000 (14:49 +0000)]
Increase MAXDEV up to 10.

20 years agoQuick fix for scaling of statclock ticks in the SMP case. As explained
Bruce Evans [Sun, 9 Nov 2003 13:45:54 +0000 (13:45 +0000)]
Quick fix for scaling of statclock ticks in the SMP case.  As explained
in the log message for kern_sched.c 1.83 (which should have been
repo-copied to preserve history for this file), the (4BSD) scheduler
algorithm only works right if stathz is nearly 128 Hz.  The old
commit lock said 64 Hz; the scheduler actually wants nearly 16 Hz
but there was a scale factor of 4 to give the requirement of 64 Hz,
and rev.1.83 changed the scale factor so that the requirement became
128 Hz.  The change of the scale factor was incomplete in the SMP
case.  Then scheduling ticks are provided by smp_ncpu CPUs, and the
scheduler cannot tell the difference between this and 1 CPU providing
scheduling ticks smp_ncpu times faster, so we need another scale
factor of smp_ncp or an algorithm change.

This quick fix uses the scale factor without even trying to optimize
the runtime divisions required for this as is done for the other
scale factor.

The main algorithmic problem is the clamp on the scheduling tick counts.
This was 295; it is now approximately 295 * smp_ncpu.  When the limit
is reached, threads get free timeslices and scheduling becomes very
unfair to the threads that don't hit the limit.  The limit can be
reached and maintained in the worst case if the load average is larger
than (limit / effective_stathz - 1) / 2 = 0.65 now (was just 0.08 with
2 CPUs before this change), so there are algorithmic problems even for
a load average of 1.  Fortunately, the worst case isn't common enough
for the problem to be very noticeable (it is mainly for niced CPU hogs
competing with less nice CPU hogs).

20 years agoFix typo in breaking up requests to size limit.
Søren Schmidt [Sun, 9 Nov 2003 10:11:15 +0000 (10:11 +0000)]
Fix typo in breaking up requests to size limit.

Found by: Peter Edwards <pmedwards@eircom.net>

20 years agoAdd the 'package-recursive' option
Christian Brueffer [Sun, 9 Nov 2003 09:30:39 +0000 (09:30 +0000)]
Add the 'package-recursive' option

PR: 57838
Submitted by: Alexander Haderer <alexander.haderer@charite.de>
MFC after: 5 days

20 years ago- Implement selwakeuppri() which allows raising the priority of a
Seigo Tanimura [Sun, 9 Nov 2003 09:17:26 +0000 (09:17 +0000)]
- Implement selwakeuppri() which allows raising the priority of a
  thread being waken up.  The thread waken up can run at a priority as
  high as after tsleep().

- Replace selwakeup()s with selwakeuppri()s and pass appropriate
  priorities.

- Add cv_broadcastpri() which raises the priority of the broadcast
  threads.  Used by selwakeuppri() if collision occurs.

Not objected in: -arch, -current

20 years agoImplement PAL_HALT_LIGHT now that the kernel halts the processor
Marcel Moolenaar [Sun, 9 Nov 2003 07:42:16 +0000 (07:42 +0000)]
Implement PAL_HALT_LIGHT now that the kernel halts the processor
when idle. All we have to do is return.

20 years agoDo not strip skiload when installed. The stripped binary does not load
Marcel Moolenaar [Sun, 9 Nov 2003 06:53:37 +0000 (06:53 +0000)]
Do not strip skiload when installed. The stripped binary does not load
in the simulator.

20 years ago - Rename vm_map_clean() to vm_map_sync(). This better reflects the fact
Alan Cox [Sun, 9 Nov 2003 05:25:35 +0000 (05:25 +0000)]
 - Rename vm_map_clean() to vm_map_sync().  This better reflects the fact
   that msync(2) is its only caller.
 - Migrate the parts of the old vm_map_clean() that examined the internals
   of a vm object to a new function vm_object_sync() that is implemented in
   vm_object.c.  At the same, introduce the necessary vm object locking so
   that vm_map_sync() and vm_object_sync() can be called without Giant.

Reviewed by: tegge

20 years agoPort truss(1) to 64-bit architectures:
Marcel Moolenaar [Sun, 9 Nov 2003 03:48:13 +0000 (03:48 +0000)]
Port truss(1) to 64-bit architectures:
o  Syscall return values do not fit in int on 64-bit architectures.
   Change the type of retval in <arch>_syscall_exit() to long and
   change the prototype of said function to return a long as well.
o  Change the prototype of print_syscall_ret() to take a long for
   the return address and change the format string accordingly.
o  Replace the code sequence
tmp = malloc(X);
sprintf(tmp, format, ...);
   with X by definition too small on 64-bit platforms by
        asprintf(&tmp, format, ...);

With these changes the output makes sense again, although it does
mess up the tabulation on ia64. Go widescreen...

Not tested on: alpha, sparc64.

20 years agoSprinkle GIANT_REQUIRED asserts around the xpt layer to aid with locking the
Scott Long [Sun, 9 Nov 2003 02:22:33 +0000 (02:22 +0000)]
Sprinkle GIANT_REQUIRED asserts around the xpt layer to aid with locking the
SCSI drivers.

20 years agoUpdate list of supported devices for ahb
Warner Losh [Sun, 9 Nov 2003 01:52:13 +0000 (01:52 +0000)]
Update list of supported devices for ahb

20 years agoComplete the variants of 1522/1520 cards
Warner Losh [Sun, 9 Nov 2003 01:46:28 +0000 (01:46 +0000)]
Complete the variants of 1522/1520 cards

20 years agoAHA-1505 also alegedly supported
Warner Losh [Sun, 9 Nov 2003 01:44:16 +0000 (01:44 +0000)]
AHA-1505 also alegedly supported

20 years agoFlesh out the list of supported devices for this driver.
Warner Losh [Sun, 9 Nov 2003 01:41:11 +0000 (01:41 +0000)]
Flesh out the list of supported devices for this driver.
Add bugs that describe the current state of this driver.

# Note: Don't ask me about the driver: I can't get it to work well either :-)

20 years agoo Update to closer to the official names for these cards.
Warner Losh [Sun, 9 Nov 2003 01:30:46 +0000 (01:30 +0000)]
o Update to closer to the official names for these cards.
o Remove entries for 1510, 152x and 1535.  These are supported, for some value
  of supported, by the aic driver.
o Add notes about 1542-CP being plug and play, but it can still conflict with
  other resources because all the resources it uses are set with the onboard
  BIOS.

20 years agoAlternate version of rev 1.20.
David E. O'Brien [Sun, 9 Nov 2003 01:22:27 +0000 (01:22 +0000)]
Alternate version of rev 1.20.
Comment out rather than totally remove the ipfilter pieces that we need
reconnected some day.  This is now only ipnat as it is for configuring NAT.
ipfstat is meant for reporting statistics/filter lists.  For /rescue it is
enough to configure lists but not view the installed ones.

20 years agoMake this driver a little more style(9) compliant
Warner Losh [Sun, 9 Nov 2003 00:51:52 +0000 (00:51 +0000)]
Make this driver a little more style(9) compliant

20 years agoSynchronize list of supported cards with hardware notes, reformat
Bruce A. Mah [Sun, 9 Nov 2003 00:50:55 +0000 (00:50 +0000)]
Synchronize list of supported cards with hardware notes, reformat
as a bullet list.

20 years agoIf a thread in critical region got a synchronous signal, according current
David Xu [Sun, 9 Nov 2003 00:37:14 +0000 (00:37 +0000)]
If a thread in critical region got a synchronous signal, according current
signal handling mode, there is no chance to handle the signal, something
must be wrong in the library, just call kse_thr_interrupt to dump its core.
I have the code for a long time, but forgot to commit it.

20 years agoMention Olicom OC2220, which is listed in the hardware notes and
Bruce A. Mah [Sun, 9 Nov 2003 00:31:41 +0000 (00:31 +0000)]
Mention Olicom OC2220, which is listed in the hardware notes and
driver source.

20 years agoCapitalize "Fast Ethernet" and "Ethernet" consistently.
Bruce A. Mah [Sun, 9 Nov 2003 00:27:45 +0000 (00:27 +0000)]
Capitalize "Fast Ethernet" and "Ethernet" consistently.

20 years agoSynchronize manual page with hardware notes and reformat list of
Bruce A. Mah [Sun, 9 Nov 2003 00:26:55 +0000 (00:26 +0000)]
Synchronize manual page with hardware notes and reformat list of
supported devices.

20 years agoreplace explicit changes to rt_refcnt by RT_ADDREF and RT_REMREF
Sam Leffler [Sat, 8 Nov 2003 23:36:32 +0000 (23:36 +0000)]
replace explicit changes to rt_refcnt by RT_ADDREF and RT_REMREF
macros that expand to include assertions when the system is built
with INVARIANTS

Supported by: FreeBSD Foundation

20 years agoRemove the atkbd, psm, sc and vga devices. Most ia64 boxes out there
Marcel Moolenaar [Sat, 8 Nov 2003 23:19:13 +0000 (23:19 +0000)]
Remove the atkbd, psm, sc and vga devices. Most ia64 boxes out there
are zx1 based machines and they don't particularly like it when we
poke at them with PC legacy code. The atkbd and psm devices were
disabled in the hints file so that one could enable them on machines
that support legacy devices, but that's not really something you can
expect from a first-time installer. This still leaves syscons (sc)
and the vga device, which were enabled by default and wrecking havoc
anyway. We could disable them by default like the atkbd and psm
devices, but there's really no point in pretending we're in a better
shape that way.

20 years agoTrim device entries for an(4), ray(4), and awi(4).
Bruce A. Mah [Sat, 8 Nov 2003 23:16:43 +0000 (23:16 +0000)]
Trim device entries for an(4), ray(4), and awi(4).

20 years agodivert socket fixups:
Sam Leffler [Sat, 8 Nov 2003 23:09:42 +0000 (23:09 +0000)]
divert socket fixups:

o pickup Giant in divert_packet to protect sbappendaddr since it
  can be entered through MPSAFE callouts or through ip_input when
  mpsafenet is 1
o add missing locking on output
o add locking to abort and shutdown
o add a ctlinput handler to invalidate held routing table references
  on an ICMP redirect (may not be needed)

Supported by: FreeBSD Foundation

20 years agoassert optional inpcb is passed in locked
Sam Leffler [Sat, 8 Nov 2003 23:03:29 +0000 (23:03 +0000)]
assert optional inpcb is passed in locked

Supported by: FreeBSD Foundation

20 years agoadd locking assertions
Sam Leffler [Sat, 8 Nov 2003 23:02:36 +0000 (23:02 +0000)]
add locking assertions

Supported by: FreeBSD Foundation

20 years agoassert inpcb is locked in udp_output
Sam Leffler [Sat, 8 Nov 2003 23:00:48 +0000 (23:00 +0000)]
assert inpcb is locked in udp_output

Supported by: FreeBSD Foundation

20 years agoo correct locking problem: the inpcb must be held across tcp_respond
Sam Leffler [Sat, 8 Nov 2003 22:59:22 +0000 (22:59 +0000)]
o correct locking problem: the inpcb must be held across tcp_respond
o add assertions in tcp_respond to validate inpcb locking assumptions
o use local variable instead of chasing pointers in tcp_respond

Supported by: FreeBSD Foundation

20 years agouse local values instead of chasing pointers
Sam Leffler [Sat, 8 Nov 2003 22:57:13 +0000 (22:57 +0000)]
use local values instead of chasing pointers

Supported by: FreeBSD Foundation

20 years agoreplace mtx_assert by INP_LOCK_ASSERT
Sam Leffler [Sat, 8 Nov 2003 22:55:52 +0000 (22:55 +0000)]
replace mtx_assert by INP_LOCK_ASSERT

Supported by: FreeBSD Foundation

20 years agoadd some missing locking
Sam Leffler [Sat, 8 Nov 2003 22:53:41 +0000 (22:53 +0000)]
add some missing locking

Supported by: FreeBSD Foundation