Matthew Dillon [Thu, 10 Sep 2009 04:52:01 +0000 (21:52 -0700)]
IPV6 - Attempt to fix tcp46 compatibility listen sockets
* When creating the accepted socket via the syncache, if the
listen socket is INET6 but allows compatibility connections
the inp created via sonewconn() will probably set IN6P_IPV6_V6ONLY.
This flag must be cleared if the incoming connection is
actually IPV4, otherwise tcp6_soport() will not properly
fall through to tcp_soport().
* NOTE: Even though the connection is IPV4 the socket's protocol
is copied from the listen socket which is INET6. This seems
to work ok but it isn't ideal.
* When connecting make sure that IPV6 connections use cpu 0 for
now (we don't have a proper hash for IPV6 addresses yet).
Reported-by: corecode
Matthew Dillon [Wed, 9 Sep 2009 16:34:13 +0000 (09:34 -0700)]
Kernel - Enable the TCP inflight limiter by default, but with generous values.
The inflight limiter is now turned on by default, but with generous values
which should allow maximal bandwidth. In particular, the slop defaults to
50 (5 packets), meaning that the TCP connection will allow up to N+5 packets
in-flight instead of just N, where N is the number of packets calculated
for the bandwidth-delay product of the connection.
The reason for doing this is that the limiter is the only mechanism we
have which seems to do a really good job preventing receiver RX rings
on network interfaces from getting blown out. Even though GigE/10GigE
is supposed to flow control it looks like either it doesn't actually
do it or Open Source drivers do not properly enable it. Large packet
bursts appear to be able to blow out the receiver's rx ring.
People using the limiter to reduce bottlenecks on slower WAN connections
should set the slop to 20 (2 packets).
Matthew Dillon [Wed, 9 Sep 2009 16:17:57 +0000 (09:17 -0700)]
MBUF - Do not limit an objcache common pool to 10 magazines.
* An objcache's common magazine pool was limited to 10 in order to return
memory to the system sooner rather then later.
This created a situation with the mbuf allocator where high-rate
mbuf producers (GigE, 10GigE) would sometimes not be able to allocate
a large enough batch of mbufs and drop packets. 10 magazines x 64 is
only 640 mbufs.
* Increase the limit from 10 to 1000.
Tested-with: 10GigE setup from Aggelos Economopoulos <aoiko@cc.ece.ntua.gr>
Matthew Dillon [Wed, 9 Sep 2009 16:17:18 +0000 (09:17 -0700)]
MBUF - Account for failed allocations so netstat -m reports properly.
Matthew Dillon [Wed, 9 Sep 2009 16:13:22 +0000 (09:13 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Wed, 9 Sep 2009 16:10:58 +0000 (09:10 -0700)]
NFE - Change default RX ring size from 128 -> 256, Adjust moderation timer.
* NFE does not seem to be able to flow control the phy. TCP reception
using window scaling and large socket bufs can blow out the receiver ring
before the interrupt has a chance to drain it.
This results in the receiver dropping a packet and the transmitter
seeing bursts of duplicate TCP acks and having to go into recovery.
* Shorten the moderation timer from 500uS to 250uS, increasing its
frequency from 2000Hz to 4000Hz. Note that the moderation timer
only kicks in when the discrete interrupt rate exceeds the frequency.
* The problem still occurs with the larger RX ring and shorter moderation
timer, but not nearly as often.
Simon Schubert [Wed, 9 Sep 2009 15:23:39 +0000 (17:23 +0200)]
gcc44/gmp: make directories before populating them
The snapshot build seems to have uncovered a race in the depend stage.
This is a tentative, unverified fix.
Alexander Polakov [Wed, 9 Sep 2009 09:43:08 +0000 (13:43 +0400)]
Revert "Move isonum_ declaration to iso.h"
This reverts commit
f5ff88dca23b1b8f4b9287fb7c59959eb976a808. It
broke sysutils/hal and maybe something else.
Requested-by: Hasso Tepper
Matthew Dillon [Wed, 9 Sep 2009 04:03:05 +0000 (21:03 -0700)]
powernow - only build for i386, BIOS macros currently broken for 64-bit.
Matthew Dillon [Tue, 8 Sep 2009 23:52:33 +0000 (16:52 -0700)]
USB - Reintroduce OHCI fixes without the larger polling infrastructure fixes
* Fix the OHCI interrupt livelock issue.
* Don't fix races with turning polling on and off yet as there appear
to be other issues with OHCI that break whene we do.
Matthew Dillon [Tue, 8 Sep 2009 21:31:20 +0000 (14:31 -0700)]
NFS - Fix panic if the readdir base offset is beyond the directory EOF
* A calculation could cause an signed underflow into unsigned if
the base offset in a getdirentries call exceeds the directory
EOF offset.
Matthew Dillon [Tue, 8 Sep 2009 21:00:11 +0000 (14:00 -0700)]
SILI - Merge Alex's ATA pass-through ioctls from AHCI driver into SILI driver.
* Basically an exact copy from the AHCI code.
Alex Hornung [Tue, 8 Sep 2009 07:48:30 +0000 (08:48 +0100)]
ahci(4) - Add support for ATA passthrough
* Add support for ATA_PASS_{12,16} cam commands to pass through
an ATA command directly to the device.
Incidentally, this will make smartmontools work again when used
with device type 'sat'.
Dragonfly-bug: http://bugs.dragonflybsd.org/issue1412
Reported-By: Hasso Tepper
Sascha Wildner [Tue, 8 Sep 2009 16:39:51 +0000 (18:39 +0200)]
iscontrol.8: A little cleanup.
Matthew Dillon [Tue, 8 Sep 2009 18:36:25 +0000 (11:36 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Tue, 8 Sep 2009 18:28:14 +0000 (11:28 -0700)]
PIPE - Fix a panic introduced by the last commit.
* Forgot to release a token on the Re-tests.
Reported-by: Tobias Weingartner <weingart@tepid.org>
Alexander Polakov [Tue, 8 Sep 2009 18:15:36 +0000 (22:15 +0400)]
powernow: AMD K8 frequency control.
Antonio Huete [Fri, 4 Sep 2009 22:18:21 +0000 (00:18 +0200)]
ISCSI - Minor module handling
- Rename iscsi module name to iscsi_initiator
- iscontrol now loads automatically iscsi_initiator module
Sascha Wildner [Tue, 8 Sep 2009 14:19:30 +0000 (16:19 +0200)]
Sync zoneinfo database with tzdata2009m from elsie.
asia: 8.36 -> 8.40
australasia: 8.12 -> 8.13
* Palestine will go back to winter time on Sep 4, 2009.
* Samoa will start DST on Oct 4, 2009 and end on Mar 28, 2010.
Alexander Polakov [Tue, 8 Sep 2009 13:11:10 +0000 (17:11 +0400)]
cd9660_util.c: fix a typo.
Alexander Polakov [Tue, 8 Sep 2009 12:20:35 +0000 (16:20 +0400)]
cd9660_util.c: ANSIfy, no functional changes.
Simon Schubert [Tue, 8 Sep 2009 11:58:28 +0000 (13:58 +0200)]
sound: mmap returns page number, not byte offset
Noticed-by: hasso@
Simon Schubert [Tue, 8 Sep 2009 10:29:04 +0000 (12:29 +0200)]
modules: remove KMODDEPS, it is not used anymore
Simon Schubert [Tue, 8 Sep 2009 10:15:13 +0000 (12:15 +0200)]
modules: add/correct module versions and depends
Simon Schubert [Tue, 8 Sep 2009 10:13:49 +0000 (12:13 +0200)]
sys/module.h: include sys/linker_set.h
Simon Schubert [Tue, 8 Sep 2009 09:44:22 +0000 (11:44 +0200)]
ig_hal: add module version
Simon Schubert [Tue, 8 Sep 2009 09:40:17 +0000 (11:40 +0200)]
add tools/chkmoddeps
chkmoddeps is a tool to check for unsatisfied module dependencies.
invoke with pathnames to be searched for kernel and modules.
Simon Schubert [Tue, 8 Sep 2009 06:59:46 +0000 (08:59 +0200)]
Revert "usbd - Do not start moused by default when a usb mouse is connected"
This reverts commit
409ca8073a5e4b42c5f75f3fa5d9be5cdab83c2e, as
discussed on the mailing lists.
Matthew Dillon [Tue, 8 Sep 2009 06:59:18 +0000 (23:59 -0700)]
Revert "USB - Formalize polling mode and fix ohci interrupt storm"
This reverts commit
727afe75ef752829158e906c5b86991478fece53. It broke
too many things, so the OHCI storm will have to be fixed another way.
Matthew Dillon [Tue, 8 Sep 2009 02:23:48 +0000 (19:23 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Tue, 8 Sep 2009 02:21:34 +0000 (19:21 -0700)]
Remove nearly all MP contention from ip_randomid()
* Instead of using two global spinlocks split the id space up into
ncpus_fit independant sections and shuffle within each section.
* Aggregate the random number generation into blocks of 32, on a
per-cpu basis, to reduce cpu contention there.
Using-test-output-by: Aggelos Economopoulos <aoiko@cc.ece.ntua.gr>
Sascha Wildner [Mon, 7 Sep 2009 14:56:21 +0000 (16:56 +0200)]
Add the make_autoclone_dev(9) manual page.
Submitted-by: alexh
Alex Hornung [Mon, 7 Sep 2009 11:34:45 +0000 (12:34 +0100)]
sio(4) - Fix double make_dev if used as console
* sio was calling another make_dev for ttyd0 if it was being used as
console. This is no longer valid with devfs, so we look for the
existing ttyd0, as it must exist already if the device has been
attached.
Dragonfly-bug: http://bugs.dragonflybsd.org/issue1485
Reported-by: Hasso Tepper, Jordan Gordeev
Matthew Dillon [Mon, 7 Sep 2009 06:05:09 +0000 (23:05 -0700)]
USB - Formalize polling mode and fix ohci interrupt storm
* Formalize entering and leaving polling mode via usb_set_polling().
* Call bus->methods->do_poll() once when leaving polling mode to
catch-up any lost interrupts.
* OHCI now acknowledges interrupts from the interrupt handler when
it finds itself in polling mode to prevent the interrupt from
storming. The interrupts are recorded for later use.
Matthew Dillon [Mon, 7 Sep 2009 02:37:55 +0000 (19:37 -0700)]
USB - Preallocate umass buffers to avoid interrupt deadlock
* Try to avoid usb allocations made from the interrupt servicer.
These can deadlock the interrupt handler and any other device
sharing that irq as well as cause USB panics due to the broken
critical section.
* This is only a temporary fix. Another unaddressed issue is
contigmalloc()'s inability to allocate large physically contiguous
buffers after the machine has been up a while.
Reported-by: Rumko and multiple others over time
Matthew Dillon [Sun, 6 Sep 2009 21:53:51 +0000 (14:53 -0700)]
USB - Avoid unnecessary reprobes.
* Do not unconditionally setdiskinfo when a removable DA* device is
opened. It takes time for the kernel to reprobe the drive and this
seriously messes up mountroot during kernel boot or attempts to
access partitions in slices (instant open + access partition).
If the size hasn't changed, do not call setdiskinfo again.
* This is a bad hack. A better solution (detect actual device
replacement) is needed.
Sascha Wildner [Sun, 6 Sep 2009 18:19:21 +0000 (20:19 +0200)]
LINT: Mention KTR_CTXSW.
Matthew Dillon [Sat, 5 Sep 2009 23:17:59 +0000 (16:17 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Sat, 5 Sep 2009 23:17:10 +0000 (16:17 -0700)]
MFS - Use mmap() instead of malloc().
* mmap does not have the dataspace limitations which malloc does, allowing
us to create a much larger MFS filesystem.
Alexander Polakov [Sat, 5 Sep 2009 21:25:00 +0000 (01:25 +0400)]
smbfs: use libiconv v.2
Sascha Wildner [Sat, 5 Sep 2009 21:18:02 +0000 (23:18 +0200)]
sys/conf/files: Remove duplicates.
Matthew Dillon [Sat, 5 Sep 2009 20:15:53 +0000 (13:15 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Sat, 5 Sep 2009 20:12:49 +0000 (13:12 -0700)]
PIPE - Fix more races in the pipe code
* Numerous additional races can occur with the setting and clearing of
PIPE_SEL.
* EOF must be retested after getting the second token lock because the
acquisition of the second lock can temporarily release the first one.
Reported-by: Tobias Weingartner <weingart@tepid.org>, Rumko
Alexander Polakov [Sat, 5 Sep 2009 19:49:14 +0000 (23:49 +0400)]
netbt: add MODULE_VERSION
Reported-by: Rumko <rumcic@gmail.com>
DragonFly-bug: http://bugs.dragonflybsd.org/issue1487
Alexander Polakov [Sat, 5 Sep 2009 17:55:29 +0000 (21:55 +0400)]
Alexander Polakov [Sat, 5 Sep 2009 16:33:38 +0000 (20:33 +0400)]
mount_msdos: remove conversion tables.
Alexander Polakov [Sat, 5 Sep 2009 10:53:43 +0000 (14:53 +0400)]
msdosfs: add MODULE_VERSION.
Alexander Polakov [Sat, 11 Jul 2009 04:27:42 +0000 (08:27 +0400)]
missing type in iso.h
Alexander Polakov [Fri, 4 Sep 2009 10:39:04 +0000 (14:39 +0400)]
citrus: ooops, actually build utf1632.
Alexander Polakov [Fri, 10 Jul 2009 21:35:56 +0000 (01:35 +0400)]
Move isonum_ declaration to iso.h
Alexander Polakov [Thu, 3 Sep 2009 22:21:41 +0000 (02:21 +0400)]
libiconv: we can now link mount_cd9660 and mount_msdos statically.
Alexander Polakov [Thu, 3 Sep 2009 22:18:30 +0000 (02:18 +0400)]
libkiconv: make libkiconv work when statically linked.
Alexander Polakov [Thu, 3 Sep 2009 22:15:49 +0000 (02:15 +0400)]
citrus: compile in UTF16/32 support for iconv.
Alexander Polakov [Thu, 3 Sep 2009 22:14:15 +0000 (02:14 +0400)]
Connect cd9660_iconv msdosfs_iconv to the build.
Alexander Polakov [Tue, 7 Jul 2009 16:07:50 +0000 (20:07 +0400)]
libiconv: change mount_cd9660 to use libkiconv.
Add "-L locale" option like mount_msdos.
Alexander Polakov [Thu, 3 Sep 2009 21:09:33 +0000 (01:09 +0400)]
libiconv: convert cd9660 to libiconv.
Alexander Polakov [Fri, 5 Jun 2009 10:14:28 +0000 (14:14 +0400)]
libiconv conversion: add libkiconv to the build, fix header path in mount_msdos.
Alexander Polakov [Fri, 5 Jun 2009 08:29:46 +0000 (12:29 +0400)]
Stage 5/5: Change mount_msdos to use libkiconv
mount_msdos should be linked shared now, that's unusual for
/sbin.
Alexander Polakov [Fri, 5 Jun 2009 08:24:06 +0000 (12:24 +0400)]
Stage 4/5: Import libkiconv from FreeBSD.
libkiconv user space library is used by mount_msdos.
Alexander Polakov [Fri, 5 Jun 2009 08:18:02 +0000 (12:18 +0400)]
Stage 3/5: Add msdosfs_iconv.
Contains actual code to do conversion.
It seems like unloading is broken. Should be fixed later.
Alexander Polakov [Fri, 5 Jun 2009 08:10:38 +0000 (12:10 +0400)]
Stage 2/5: Change msdosfs to use libiconv in case.
Code mostly from FreeBSD, ported by me. Still works with old
userspace (if mounted without conversion).
Alexander Polakov [Fri, 5 Jun 2009 08:04:28 +0000 (12:04 +0400)]
libiconv: Uncomment iconv_vfs_refcount.
Alexander Polakov [Fri, 5 Jun 2009 07:59:03 +0000 (11:59 +0400)]
Stage 1/5: Update libiconv to support multibyte encodings.
Code from FreeBSD, ported by Alexander Polakov, fixed by Simon
Schubert.
Alexander Polakov [Thu, 3 Sep 2009 18:51:10 +0000 (22:51 +0400)]
citrus: iconv support for static binaries
Compile iconv into libc.a.
Matthew Dillon [Sat, 5 Sep 2009 04:10:41 +0000 (21:10 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Sepherosa Ziehau [Sat, 5 Sep 2009 03:52:32 +0000 (11:52 +0800)]
soopt_from_kbuf: Use szmin(), since both len and sopt_valsize is size_t
Mainly to avoid integer truncation on 64bit platform.
Reminded-by: dillon@
Sepherosa Ziehau [Sat, 5 Sep 2009 03:27:25 +0000 (11:27 +0800)]
sockopt: Short circuit most of the code in soopt_from_kbuf(), if len == 0
Current kva_p() does not work w/ zero sized kmalloced memory.
Dragonfly-bug: http://bugs.dragonflybsd.org/issue1481
Aggelos Economopoulos [Sat, 5 Sep 2009 00:46:14 +0000 (03:46 +0300)]
add KTR_CTXSW option to log context switches
This logs the current thread and the thread we switch
to on every context switch or thread preemption event
when enabled.
Matthew Dillon [Sat, 5 Sep 2009 01:04:40 +0000 (18:04 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Stathis Kamperis [Fri, 4 Sep 2009 17:22:49 +0000 (20:22 +0300)]
top(1): Silence gcc warnings.
The ORDERKEY_XXX() macros build up a deeply nested set of if-statements.
E.g.,
if (expr)
if (expr)
if (expr)
if (expr)
The last if-statement needs to contain nothing (every expression has
its own side effect, so the whole cascade is actually meaningful). Using
a ; though, makes gcc generate warnings regarding 'empty body in if
statements'. That's why we, instead, use:
ORDERKEY_FOO1
ORDERKEY_FOO2
...
ORDERKEY_FOO3
{}
We could have embedded the {} in the definition of the macro itself, but
that would break if someone rearranged the macros' call order.
Reviewed-by: dillon@
Sascha Wildner [Fri, 4 Sep 2009 20:32:37 +0000 (22:32 +0200)]
nrelease.7: /dev/da8 is the default IMGUSBDEV.
Reported-by: Johannes Hofmann
Matthew Dillon [Fri, 4 Sep 2009 03:56:40 +0000 (20:56 -0700)]
AMD64 - Make machine/endian.h work properly when compiled -m32.
* Fixes issues compiling sys/boot in a 64-bit native environment.
Reported-by: corecode
Matthew Dillon [Thu, 3 Sep 2009 23:53:50 +0000 (16:53 -0700)]
modules: Kill debugging.
Requested-by: corecode
Simon Schubert [Thu, 3 Sep 2009 22:47:11 +0000 (00:47 +0200)]
modules: correct dependencies / version declarations
Simon Schubert [Thu, 3 Sep 2009 20:42:52 +0000 (22:42 +0200)]
test/amd64: build and install modules
also don't try to install gcc44 since we don't build it.
Simon Schubert [Thu, 3 Sep 2009 21:59:53 +0000 (23:59 +0200)]
loader: update elf module code from FreeBSD
Simon Schubert [Thu, 3 Sep 2009 21:58:17 +0000 (23:58 +0200)]
modules: add elf_obj linker for amd64
Obtained-from: FreeBSD
Simon Schubert [Thu, 3 Sep 2009 23:17:18 +0000 (01:17 +0200)]
modules: pull in most of FreeBSD's module linker changes
The big user-visible difference is that the module dependency
system changed. Module dependencies are now the same as for
the loader, and the hack linker (and Makefile) magic is not
needed anymore.
Obtained-from: FreeBSD
Simon Schubert [Thu, 3 Sep 2009 20:45:30 +0000 (22:45 +0200)]
modules: disable dummy module declarations
dummy modules serve a questionable goal and conflict with the
module linker update.
Simon Schubert [Thu, 3 Sep 2009 20:29:02 +0000 (22:29 +0200)]
modules: drop a.out module support
Simon Schubert [Thu, 3 Sep 2009 20:41:01 +0000 (22:41 +0200)]
amd64: declare initializecpu outside of SMP
initializecpu() is not a SMP-only function; we have to
always declare it. smp.h is the wrong file, but I didn't
find any better location.
Simon Schubert [Thu, 3 Sep 2009 20:34:03 +0000 (22:34 +0200)]
acpi: restructure genwakecode.sh
genwakecode.sh used some obscure hexdump/sed magic to embed a binary
into source. The specific usage of sed breaks in at least the UTF-8
locale. Rework the script to use hexdump and sed in a less obscure
way.
Matthew Dillon [Thu, 3 Sep 2009 20:37:20 +0000 (13:37 -0700)]
HAMMER Util - hammer iostats now reports undo space used.
Matthew Dillon [Thu, 3 Sep 2009 20:36:32 +0000 (13:36 -0700)]
HAMMER - Add vfs.hammer.stats_undo
* Statistics on number of bytes of undo space written.
Matthew Dillon [Thu, 3 Sep 2009 15:26:26 +0000 (08:26 -0700)]
CAM - Check DA media on open, basic suppor for removable DA media (vs CD).
* Install setdiskinfo on probe for DA if no media is present, similar to how
the CD works. This allows open() to function.
* Recheck media presence on open().
Reported-by: "Simon 'corecode' Schubert" <corecode@fs.ei.tum.de>
Matthew Dillon [Thu, 3 Sep 2009 15:25:26 +0000 (08:25 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Thu, 3 Sep 2009 15:18:43 +0000 (08:18 -0700)]
HAMMER - Fix lost inode issue (primarily with nohistory mounts)
* When a HAMMER cursor is unlocked it becomes tracked and unrelated
B-Tree operations will cause the tracked cursor's nodes and indices
to be updated. The cursor structure also has a leaf element pointer
which was not being properly updated. This could lead to panics and
lost inodes.
Properly adjust the leaf element pointer in tracked cursors.
* The bug primarily occurs with nohistory mounts or nohistory sub-trees
due to the larger number of physical deletions made to the B-Tree, but
could also occur (rarely) with normal mounts.
* Add additional assertions to catch any further occurrences (though I
think all the cases have been covered now).
* Add a new sysctl vfs.hammer.error_panic which can be set to e.g. 9 to
cause critical errors to panic immediately instead of returning
through the call stack, making debugging possible.
Reported-by: Numerous people
Alex Hornung [Thu, 3 Sep 2009 13:18:25 +0000 (14:18 +0100)]
subr_disk - add debug info
* Add tunable and sysctl int kern.disk_debug which enables debug info
in critical parts of subr_disk (probing, creating, unprobing,
destroying).
Alex Hornung [Thu, 3 Sep 2009 07:57:34 +0000 (08:57 +0100)]
acpi_wakeup - Mark used vars with attribute used
* Mark obscurely used variables (such as variables only used from inside
inline assembly) with __attribute__((used)).
This also solves an issue with llvm/clang.
Alex Hornung [Thu, 3 Sep 2009 07:54:41 +0000 (08:54 +0100)]
mxge - Avoid using #cpu
* Don't use #cpu preprocessor macro, as it isn't needed. CPU is detected
properly by defines as CPU __CPU__ __CPU.
This also fixes compilation problems with llvm/clang.
Alex Hornung [Thu, 3 Sep 2009 09:33:22 +0000 (10:33 +0100)]
linuxemu - fix typo
* Commit
e54488bbec5c9f80e95cedd395b0e3d31fde253d introduced a typo
in linux_file breaking anything using linux getdents stuff.
This also fixes issue 1471 and brings linuxulator back into working
condition.
Dragonfly-bug: http://bugs.dragonflybsd.org/issue1471
Reported-By: ftigeot
Matthew Dillon [Thu, 3 Sep 2009 05:15:12 +0000 (22:15 -0700)]
TCP - Reduce context switching when handling large send*() or write()'s
* Large sends or writes would cause sosend() to sleep and wakeup for every
tcp segment acknowledged, resulting in massive inefficiencies. To deal
with this, if no LOWAT is set we automatically set the LOWAT to 1/2 the
send buffer size and adjust the wakeup code.
Matthew Dillon [Thu, 3 Sep 2009 05:12:06 +0000 (22:12 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Thu, 3 Sep 2009 05:06:22 +0000 (22:06 -0700)]
NFE - Fix bug with imtimer transitions and improve performance
* When the imtimer transitions to a new state via a sysctl the
code called ifp->if_init() which more often then not caused
the device to stop taking all interrupts.
* Change the way dynamic interrupt moderation works. Timer moderation is
not turned on until the discrete interrupt rate exceeds the threshold.
e.g. by default 500uS is 2000 interrupts/second so timer moderation
is not turned on until the discrete interrupt rate exceeds 2000 ips.
This allows the device to respond interactively as long as traffic levels
are reasonable, before converting into moderation-timer-based batching.
Timer moderation is turned off and we go back to discrete interrupts
if the average rate over ~4 seconds falls below the threshold.
* Change the interrupt moderation default from 125uS to 500uS. 125uS can
saturate the cpu if the interrupt line is shared with other devices.
A larger moderation timer is also more reasonable now that discrete
interrupts are left intact until the rate exceeds the threshold.
The moderation timer should be roughly designed to deal with the txring
and rxring size.
Alexander Polakov [Wed, 2 Sep 2009 15:25:08 +0000 (19:25 +0400)]
test commit
Alexander Polakov [Wed, 2 Sep 2009 14:54:57 +0000 (18:54 +0400)]
citrus: add a list of supported locales and build UTF8 by default.
Alexander Polakov [Wed, 2 Sep 2009 13:38:51 +0000 (17:38 +0400)]
citrus: fix libc_rtld build with STATIC_LOCALES set.
Fix locales build for libc_rtld
Disabled locales for libc_rtld.
Alexander Polakov [Wed, 2 Sep 2009 12:54:19 +0000 (16:54 +0400)]
make.conf(5): mention STATIC_LOCALES
Sascha Wildner [Wed, 2 Sep 2009 13:17:55 +0000 (15:17 +0200)]
libobjc (gcc44): Don't build when NO_OBJC is set.
Reported-by: Venkatesh Srinivas
Alex Hornung [Wed, 2 Sep 2009 08:12:58 +0000 (09:12 +0100)]
ttyslot(3) - Update for unix98 ptys
* Update ttyslot(3) to be compatible with unix98 ptys. This fixes
problems with logins on unix98 ptys not appearing in utmp.
* ttyslot was calling rindex, which is invalid for unix98 pty files,
so now we just strip "/dev/".
Obtained-from: FreeBSD