Matthew Dillon [Fri, 17 Jul 2009 11:38:35 +0000 (04:38 -0700)]
NFS - Fix unmounting blockages, fix tick interval, hack cpu assignments.
* Umount was having problems due to the way the helper threads were
being shutdown. Should be fixed.
* The NFS timer, which is responsible for handling retransmits and
timeouts, did not need to be set to one tick (10ms). 200ms is
just fine.
* Hack cpu assignments for the tx and rx helper threads.
Matthew Dillon [Fri, 17 Jul 2009 10:02:13 +0000 (03:02 -0700)]
NFS - implement async state machine for nfs_readrpc_bio()
Implement the basic state machine mechanics to allow the two
helper threads to separate transmit and receive work for any
number of NFS RPCs.
Matthew Dillon [Fri, 17 Jul 2009 02:37:21 +0000 (19:37 -0700)]
NFS - refactor nfs_request() and add state machine fields to nfsm_info
* nfs_request() now operates as a state machine and can be told to
return when it reaches any state, allowing RPC sequences to be
split up (to be used by the new iod).
Matthew Dillon [Fri, 17 Jul 2009 01:22:22 +0000 (18:22 -0700)]
NFS - Cleanup pass -factor out strict temporaries from nfsm_info
Matthew Dillon [Fri, 17 Jul 2009 00:03:55 +0000 (17:03 -0700)]
NFS - Fix refactor bugs.
* The new ERROROUT macro unconditionally overwrites error. Don't use
it for cases where we wish to retain a prior error.
Matthew Dillon [Thu, 16 Jul 2009 23:21:36 +0000 (16:21 -0700)]
NFS - Marathon - completely refactor the nfsm_* macros.
Completely refactor the nfsm_*() macros in nfsm_subs.h, converting
all of them to procedures in nfsm_subs.c.
Add a nfsm_info structure to track the persistent state that the code
depends on, and remove all related local variables from the other bits
of the NFS code.
Matthew Dillon [Thu, 16 Jul 2009 15:19:14 +0000 (08:19 -0700)]
NFS - create nfsm_subs.c, clean up externs
* Move nfsm_*() procedures from nfs_subs.c to nfsm_subs.c
* Clean up externs improperly embedded in .c files.
Matthew Dillon [Thu, 16 Jul 2009 06:47:44 +0000 (23:47 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Thu, 16 Jul 2009 06:46:07 +0000 (23:46 -0700)]
NFS - install nfsmount on nfs_mountq, for timeout handling.
* This way the timeout callout can find all the requests.
Matthew Dillon [Thu, 16 Jul 2009 06:21:34 +0000 (23:21 -0700)]
NFS - Remove old nfsiod, start adding new kernel thread infrastructure
* Remove all the nfsiod junk.
* Add two per-mount threads, one for reading from the socket, one for writing
to the socket, in a new file nfs_iod.c
* Implement a quick and dirty synchronous, single threaded nfs_doio()
loop in the writer thread to test basic mechanics.
Matthew Dillon [Thu, 16 Jul 2009 05:37:37 +0000 (22:37 -0700)]
NFS - Minor compile fix
Matthew Dillon [Thu, 16 Jul 2009 05:25:40 +0000 (22:25 -0700)]
tsleep_interlock - Fix bug with TDF_TSLEEPQ
* LWKT previously disallowed the scheduling of a thread flagged
TDF_TSLEEPQ because the tsleep queue was the same as the threadq.
Now that they are different, it is possible for the thread to
still be on a tsleepq when descheduled and later rescheduled.
Remove the check, allowing the thread to be rescheduled by LWKT.
This fixes issues with processes just locking up in "D"isk wait.
Matthew Dillon [Thu, 16 Jul 2009 04:14:04 +0000 (21:14 -0700)]
NFS - move nfs_reqq from global to per-nfsmount.
Matthew Dillon [Thu, 16 Jul 2009 02:48:23 +0000 (19:48 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Thu, 16 Jul 2009 02:41:34 +0000 (19:41 -0700)]
NFS - Use mutex API, begin refactoring the state machine.
* Use the mtx_*() API instead of roll-your-own locks for the send and
receive locks.
* Refactor nfs_request(). Break the procedure up into multiple pieces
for upcoming nfsiod/nfsd work (as in: getting rid of them).
The main thing here is to split off the 'setup', 'send', 'receive',
and 'process reply' parts. This will make it easier to construct
a kernel thread to i.e. just do the 'setup/send' part, and another
to do the 'receive/reply' part.
Matthew Dillon [Thu, 16 Jul 2009 02:38:39 +0000 (19:38 -0700)]
MPSAFE - mutex - better exclusive lock sequencer, bug fixes, abort
* Redo the exclusive lock chaining algorithm. Use an explicit
link structure and directly pass ownership to the next thread
waiting on an exclusive lock.
* Exclusive locks can be aborted via mtx_lock_ex_link() and
mtx_abort_ex_link().
* Lots of misc bug fixes.
Sascha Wildner [Wed, 15 Jul 2009 20:23:43 +0000 (22:23 +0200)]
Fix LINT build.
Matthew Dillon [Wed, 15 Jul 2009 19:13:13 +0000 (12:13 -0700)]
MPSAFE - mutexes
* Add additional inline functions to test the lock state.
Matthew Dillon [Wed, 15 Jul 2009 19:04:28 +0000 (12:04 -0700)]
MPSAFE - mutex enhancements.
* Add timeout to mtx_lock_{sh,ex}
* Add quick version, mtx_lock_{sh,ex}_quick() which does not take a
flags or timeout parameter.
Matthew Dillon [Wed, 15 Jul 2009 18:38:50 +0000 (11:38 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Wed, 15 Jul 2009 18:37:45 +0000 (11:37 -0700)]
MPSAFE - Add a set of general blocking/spinnable mutex functions.
* (serial wakeup code wasn't included in last commit)
Matthew Dillon [Wed, 15 Jul 2009 18:29:43 +0000 (11:29 -0700)]
MPSAFE - Add a set of general blocking/spinnable mutex functions.
These locks are intended to eventually replace lockmgr locks for most
use cases.
* Optimized based on in-line atomic_cmpset_int() calls with fallback call.
* Recursive shared and exclusive locks. Downgrading, and non-blocking
upgrading.
* Interlocked wakeup flags.
* Serial wakeup for exclusive waiters (i.e. optimal in the face of a large
number of waiting threads). Mass-wakeup for shared waiters.
* Additional entry points for spinning.
* Ref-count support, separate from lock count.
Matthew Dillon [Wed, 15 Jul 2009 16:44:22 +0000 (09:44 -0700)]
Adjust atomic_cmpset_int/long - Faster version, fix amd64 issue.
* Instead of using the condition code just compare %eax (or %rax)
against the old value. This is considerably faster then using
sete/movzbl and GCC will also optimize the caller's test for
both 0 or non-zero.
* AMD64. long is 64 bits, use cmpxchgq (it was previously using cmpxchgl).
Michael Neumann [Wed, 15 Jul 2009 16:34:58 +0000 (18:34 +0200)]
hammer expand: Also increase number of total bigblocks
Michael Neumann [Wed, 15 Jul 2009 15:37:44 +0000 (17:37 +0200)]
hammer expand: Layer 1 formatting (step 2/2)
With this commit online HAMMER filesystem expansion is functional, but
highly experimental and might destroy your filesystem.
Michael Neumann [Wed, 15 Jul 2009 12:36:58 +0000 (14:36 +0200)]
Minor adjustment which does not change behaviour.
This has no effect because of the bzero(). Nevertheless it's
more logical to generate the crc after setting blocks_free to 0.
Michael Neumann [Wed, 15 Jul 2009 11:58:39 +0000 (13:58 +0200)]
hammer expand: Layer 2 formatting (step 1.5/2)
Sascha Wildner [Wed, 15 Jul 2009 10:56:15 +0000 (12:56 +0200)]
isdnd(8): Fix some NULL dereferences.
Taken-from: FreeBSD
Matthew Dillon [Wed, 15 Jul 2009 06:23:55 +0000 (23:23 -0700)]
pipes - set kern.pipe.mpsafe default to 1.
Enable MPSAFE operation on pipes by default.
Matthew Dillon [Wed, 15 Jul 2009 06:09:25 +0000 (23:09 -0700)]
Whitespace cleanup
Matthew Dillon [Wed, 15 Jul 2009 06:00:52 +0000 (23:00 -0700)]
NFS - Make the NFS read-ahead defaults & limits more generous.
Matthew Dillon [Wed, 15 Jul 2009 03:19:13 +0000 (20:19 -0700)]
undo - Add <sys/param.h> to get MAXPATHLEN
Reported-by: corecode
Matthew Dillon [Wed, 15 Jul 2009 03:09:41 +0000 (20:09 -0700)]
LWKT - Make sure an exiting thread is not sitting on a tsleep queue.
Matthew Dillon [Wed, 15 Jul 2009 03:05:51 +0000 (20:05 -0700)]
LWKT - Rename lwkt_yield_quick() to a more appropriate splz_check().
Matthew Dillon [Wed, 15 Jul 2009 02:31:18 +0000 (19:31 -0700)]
MPSAFE - tsleep_interlock, BUF/BIO, cluster, swap_pager.
* tsleep_interlock()/tsleep() could miss wakeups during periods of
heavy cpu activity. What would happen is code inbetween the two
calls would try to send an IPI (say, issue a wakeup()), but while
sending the IPI the kernel would be forced to process incoming IPIs
synchronous to avoid a deadlock.
The new tsleep_interlock()/tsleep() code adds another TAILQ_ENTRY to
the thread structure allowing tsleep_interlock() to formally place
the thread on the appropriate sleep queue without having to deschedule
the thread. Any wakeup which occurs between the interlock and the
real tsleep() call will remove the thread from the queue and the
later tsleep() call will recognize this and simply return without sleeping.
The new tsleep() call requires PINTERLOCKED to be passed to tsleep
so tsleep() knows that the thread has already been placed on a sleep
queue.
* Continue making BUF/BIO MPSAFE. Remove B_ASYNC and B_WANT from buf->b_flag
and add a new bio->bio_flags field to the bio. Add BIO_SYNC, BIO_WANT,
and BIO_DONE. Use atomic_cmpset_int() (aka cmpxchg) to interlock
biodone() against biowait().
vn_strategy() and dev_dstrategy() call semantics now require that
synchronous BIO's install a bio_done function and set BIO_SYNC in
the bio.
* Clean up the cluster code a bit.
* Redo the swap_pager code. Instead of issuing I/O during the collection,
which depended on critical sections to avoid races in the cluster append,
we now build the entire collection first and then dispatch the I/O.
This allows us to use only async completion for the BIOs, instead of
a hybrid sync-or-async completion.
Matthew Dillon [Wed, 15 Jul 2009 02:24:43 +0000 (19:24 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Wed, 15 Jul 2009 02:23:34 +0000 (19:23 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Wed, 15 Jul 2009 02:23:01 +0000 (19:23 -0700)]
Minor adjustments of recent commits to fix UP kernel build.
Sascha Wildner [Wed, 15 Jul 2009 02:09:19 +0000 (04:09 +0200)]
Fix some typos in manpages.
Michael Neumann [Tue, 14 Jul 2009 17:30:15 +0000 (19:30 +0200)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Michael Neumann [Tue, 14 Jul 2009 17:29:13 +0000 (19:29 +0200)]
hammer expand: Do not assume sequential vol_no's
Hasso Tepper [Tue, 14 Jul 2009 16:14:15 +0000 (19:14 +0300)]
Fix randomly seen problems with linking C++ programs.
Binutils 2.17 supports COMDAT groups and .nsubspa comdat option. It should
fix randomly seen linking errors like:
`.gnu.linkonce.t.something' referenced in section `.gnu.linkonce.r.something'
of some_object: defined in discarded section `.gnu.linkonce.t.something'
of some_object.
Digged-out-by: corecode@
Michael Neumann [Tue, 14 Jul 2009 14:01:40 +0000 (16:01 +0200)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Michael Neumann [Tue, 14 Jul 2009 13:55:02 +0000 (15:55 +0200)]
Expand HAMMER filesystem step 1/2
Format the volume and increase other volumes' vol_count field.
The space of the new volume is not yet given to the filesystem (that's step 2).
Example:
newfs_hammer -L TEST /dev/da0
mount_hammer /dev/da0 /hammer
hammer expand /hammer /dev/da1
umount /hammer
mount_hammer /dev/da0:/dev/da1 /hammer
# the following fails (wrong number of volumes)
mount_hammer /dev/da0 /hammer
Sascha Wildner [Tue, 14 Jul 2009 12:54:36 +0000 (14:54 +0200)]
Fix typos in mse(4) and ppp(8) manpages: compliment -> complement.
Sepherosa Ziehau [Mon, 13 Jul 2009 13:44:11 +0000 (21:44 +0800)]
Shuffle IPI handlers to the last two blocks of the IDT.
Relative position of the IPI handlers is kept. This gives us
more contiguous free IDT vectors.
Sepherosa Ziehau [Mon, 13 Jul 2009 06:10:30 +0000 (14:10 +0800)]
Add age(4) to GENERIC and LINT
Sepherosa Ziehau [Mon, 13 Jul 2009 06:06:44 +0000 (14:06 +0800)]
Add ae(4) manpage
Submitted-by: Alexander Polakov <polachok@gmail.com>
Obtained-from: FreeBSD
Sepherosa Ziehau [Mon, 13 Jul 2009 06:03:24 +0000 (14:03 +0800)]
Add ae(4) to LINT and GENERIC
Sepherosa Ziehau [Mon, 13 Jul 2009 05:51:23 +0000 (13:51 +0800)]
Add ae(4) into module building
Sepherosa Ziehau [Mon, 13 Jul 2009 05:45:25 +0000 (13:45 +0800)]
Add driver for Attansic/Atheros L2 FastEthernet
Submitted-by: Alexander Polakov <polachok@gmail.com> w/ modification
Obtained-from: FreeBSD
Sepherosa Ziehau [Sun, 12 Jul 2009 08:05:49 +0000 (16:05 +0800)]
lapic_init() -> lapic_map()
Sascha Wildner [Tue, 14 Jul 2009 12:01:34 +0000 (14:01 +0200)]
Remove the distribute, redistribute and distributeworld targets.
We don't split things into different distributions and none of this
was ever used.
Pointed-out-by: pavalos
Matthew Dillon [Tue, 14 Jul 2009 10:36:06 +0000 (03:36 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Tue, 14 Jul 2009 10:34:56 +0000 (03:34 -0700)]
tsleep() - Use PINTERLOCKED in msleep() and serialized_sleep().
Stathis Kamperis [Wed, 1 Jul 2009 20:49:43 +0000 (23:49 +0300)]
termios.h: Expose tcgetsid(3) function.
Dragonfly-bug: <http://bugs.dragonflybsd.org/issue1419>
Reviewed-by: hasso@, corecode@
Matthew Dillon [Tue, 14 Jul 2009 10:14:33 +0000 (03:14 -0700)]
tsleep() - Add PINTERLOCKED flag to catch edge case.
When the tsleep_interlock() + UNLOCK + tsleep() combination is used it is
possible for an incoming wakeup IPI to be processed even if the combination
is used within a critical section, because operations inbetween the two
may send an IPI. Under heavy loads sending an IPI can force incoming IPIs
to be processed synchronously to avoid deadlocks.
It is also possible for tsleep itself to create this condition when it
releases the user process schedule prior to descheduling itself.
PINTERLOCKED causes tsleep to check whether the bit set by tsleep_interlock()
is still set. If it is not set we simply return without sleeping.
Matthew Dillon [Tue, 14 Jul 2009 07:14:12 +0000 (00:14 -0700)]
HAMMER - new hammer_lock code, fix mplock bug in last commit, mpsafe getattr.
* Fix a mplock bug in the last commit that can cause a panic in hammer_read.
The mplock check code wasn't setting up the hammer transaction properly.
* VMP_GETATTR is now set on all HAMMER vnodes. If sysctl vfs.getattr_mpsafe=1
all fstat() operations will run without the MP lock.
* Rewrite HAMMER's lock code, making it MPSAFE. The new lock code uses
atomic_cmpset_int().
Matthew Dillon [Tue, 14 Jul 2009 07:08:11 +0000 (00:08 -0700)]
MPSAFE - Regenerate system calls for MPSAFE changes.
Matthew Dillon [Tue, 14 Jul 2009 06:02:21 +0000 (23:02 -0700)]
MPSAFE - Implement sysctl vfs.getattr_mpsafe, make fstat() MPSAFE
Make fstat() MPSAFE. Adjust fileops->fo_stat functions to be MPSAFE as
necessary (uipc socket stat mainly).
Matthew Dillon [Tue, 14 Jul 2009 05:58:59 +0000 (22:58 -0700)]
MPSAFE - Document a few procedures as MPSAFE (non-operational)
Matthew Dillon [Tue, 14 Jul 2009 05:26:53 +0000 (22:26 -0700)]
MPSAFE - system calls set_tls_area() and get_tls_area() are MPSAFE
Matthew Dillon [Tue, 14 Jul 2009 03:49:10 +0000 (20:49 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Tue, 14 Jul 2009 03:46:45 +0000 (20:46 -0700)]
HAMMER - Add MPSAFE suppor for hammer_vop_read().
This is still a bit of a hack. If vfs.read_mpsafe is set HAMMER will
run MPSAFE for reads from regular files which are cached in the buffer
cache. Data which is not cached will fall-back to getting the MP lock.
At the moment enabling this operation will cause atime to not get updated
on fully cached files, because the rest of HAMMER is not MPSAFE yet so I
cannot modify the internal hammer_inode state.
Matthew Dillon [Tue, 14 Jul 2009 03:41:30 +0000 (20:41 -0700)]
MPSAFE - Add read() and write() path MPSAFE support, and more.
* Add sysctl vfs.read_mpsafe, vfs.write_mpsafe, and vfs.getattr_mpsafe.
Only vfs.read_mpsafe is currently implemented, and only for HAMMER
filesystems.
* read() and write() will now call VOP_READ()/VOP_WRITE() without the
MP lock if the vnode->v_flag says it is ok to do so.
* read() and write() now serialize reads and writes which depend on the
file offset stored in the fp, when the concurrent operations share the
same fp.
* Add vnode flags VMP_READ, VMP_WRITE, VMP_GETATTR.
* Add vnode flag VNOTSEEKABLE indicating that fp->f_offset will never be
used and no serialization is needed (which e.g. can mess up read()'s
blocking devices such as ttys).
* Correct bug in FIONREAD related to the data field being too small.
Sascha Wildner [Tue, 14 Jul 2009 02:04:49 +0000 (04:04 +0200)]
mxge: Clean up Makefile a little more.
Sascha Wildner [Tue, 14 Jul 2009 01:42:10 +0000 (03:42 +0200)]
mxge: Add opt_inet.h to the Makefile.
Matthew Dillon [Tue, 14 Jul 2009 00:19:09 +0000 (17:19 -0700)]
CCMS - Correct bug, add #include <sys/spinlock2.h> to kern_ccms.c
Matthew Dillon [Tue, 14 Jul 2009 00:17:21 +0000 (17:17 -0700)]
CCMS - bug in last commit, remove #include <sys/spinlock2.h> from ccms.h
Matthew Dillon [Mon, 13 Jul 2009 23:48:45 +0000 (16:48 -0700)]
Normalize the defines for the vnode flags (non-operational change).
Matthew Dillon [Mon, 13 Jul 2009 23:44:51 +0000 (16:44 -0700)]
CCMS - Make the cache coherency locks MPSAFE.
Matthew Dillon [Mon, 13 Jul 2009 22:57:53 +0000 (15:57 -0700)]
BUF/BIO - MPSAFE work on core buffer cache routines.
Do initial MPSAFE work on getblk(), bread(), brelse(), bqrelse(),
biowait(), and friends. Use get_mplock()/rel_mplock() in areas not
yet converted.
Add a flags argument to findblk() and have it return a locked buffer
cache buffer by default. Callers may request that the buffer be
locked non-blocking, or that it not be locked at all.
Adjust all callers of findblk() for those cases where an unlocked
buffer is desired to lock and re-check the returned buffer's parameters,
since otherwise it can change out from under the caller whether a critical
section is used or not (for SMP).
Sascha Wildner [Mon, 13 Jul 2009 22:03:33 +0000 (00:03 +0200)]
telnet: Fix a bug (really use the return value of tty_istrapsig()).
Michael Neumann [Mon, 13 Jul 2009 21:40:38 +0000 (23:40 +0200)]
Expand HAMMER filesystem stubs
Matthew Dillon [Mon, 13 Jul 2009 21:11:11 +0000 (14:11 -0700)]
pipe - add required critical section around tsleep_interlock() call.
Matthew Dillon [Mon, 13 Jul 2009 20:41:12 +0000 (13:41 -0700)]
buffer cache - protect bufqueues[] with the bufspin spinlock.
Sascha Wildner [Mon, 13 Jul 2009 19:16:16 +0000 (21:16 +0200)]
mxge.4: Some small changes.
* We use local time line, not FreeBSD's.
* Kick unneeded .Pp.
* First appeared in 2.3.
Aggelos Economopoulos [Mon, 13 Jul 2009 18:47:37 +0000 (21:47 +0300)]
Merge branch 'mxge'
Aggelos Economopoulos [Mon, 13 Jul 2009 18:46:35 +0000 (21:46 +0300)]
mxge.4: sync with the status of the driver in DragonFly
Aggelos Economopoulos [Mon, 13 Jul 2009 18:45:09 +0000 (21:45 +0300)]
mxge: import mxge.4 from FreeBSD
Matthew Dillon [Mon, 13 Jul 2009 18:39:03 +0000 (11:39 -0700)]
Document pmap_qremove() as being MPSAFE.
Matthew Dillon [Mon, 13 Jul 2009 18:38:09 +0000 (11:38 -0700)]
Make getpbuf(), trypbuf(), and relpbuf() MPSAFE.
Matthew Dillon [Mon, 13 Jul 2009 18:37:35 +0000 (11:37 -0700)]
Make biowait() MPSAFE.
Sascha Wildner [Mon, 13 Jul 2009 18:07:54 +0000 (20:07 +0200)]
telnet.1: Remove unnecessary .Pp
Sascha Wildner [Mon, 13 Jul 2009 17:58:58 +0000 (19:58 +0200)]
priv.9: Update for recent changes.
Approved-by: mneumann
Matthew Dillon [Mon, 13 Jul 2009 17:55:24 +0000 (10:55 -0700)]
buffer cache - Control all access to the buf red-black trees with vp->v_token
Access to the buffer cache's RB trees is now controlled via vp->v_token
instead of a critical section. We still hold the BGL but this is not
quite as simple as it seems because an interrupt calling biodone() on a
B_INVAL buffer may now potentially block, where as it would not have before.
The buffer is locked.
Matthew Dillon [Mon, 13 Jul 2009 17:53:03 +0000 (10:53 -0700)]
lwkt_token - Add ASSERT_LWKT_TOKEN_HELD(token) macro.
Matthew Dillon [Mon, 13 Jul 2009 17:42:34 +0000 (10:42 -0700)]
bio_track - fix bug in last commit (the negative count assertion was wrong)
Matthew Dillon [Mon, 13 Jul 2009 16:30:22 +0000 (09:30 -0700)]
Make operations on struct bio_track MPSAFE.
Use atomic ops to make reference counting and waiting on bio_track structures
MPSAFE.
Aggelos Economopoulos [Mon, 13 Jul 2009 15:12:38 +0000 (18:12 +0300)]
mxge: use ether_input_chain()
Aggelos Economopoulos [Mon, 13 Jul 2009 15:11:10 +0000 (18:11 +0300)]
mxge: call bus_setup_intr() after the if serializer has been set
Aggelos Economopoulos [Mon, 13 Jul 2009 15:10:01 +0000 (18:10 +0300)]
mxge: properly initialize m_pkthdr.len as well
Aggelos Economopoulos [Mon, 13 Jul 2009 15:05:50 +0000 (18:05 +0300)]
mxge: use normal ethernet mtu by default
We don't properly support jumbo frames yet, so
change the default mtu
Aggelos Economopoulos [Mon, 13 Jul 2009 15:04:19 +0000 (18:04 +0300)]
mxge: properly remove the sysctls
Matthew Dillon [Mon, 13 Jul 2009 14:57:26 +0000 (07:57 -0700)]
vnode - move v_pollinfo.vpi_token to v_token
* Rename the lwkt_token used for the vnode's v_pollinfo for general use by
the vnode. This token will soon be used to lock the buffer cache.
Matthew Dillon [Mon, 13 Jul 2009 14:40:48 +0000 (07:40 -0700)]
pipe - fix UP build error.
Reported-by: Hasso Tepper <hasso@estpak.ee>
Matthew Dillon [Mon, 13 Jul 2009 05:44:08 +0000 (22:44 -0700)]
SYSPERF - update pipe1, pipe2
* pipe1 now does a 1-second timing run.
* pipe2 now does a 2-second timing run, just so it runs just a tad longer.
Matthew Dillon [Mon, 13 Jul 2009 05:24:54 +0000 (22:24 -0700)]
Merge branch 'master' of ssh://crater.dragonflybsd.org/repository/git/dragonfly
Matthew Dillon [Mon, 13 Jul 2009 05:14:43 +0000 (22:14 -0700)]
pipe - replace use of DELAY() with a better solution, add features.
* Use the new tsc functions to poll for new read data in a pipe for a short
period of time on a SMP box. This greatly increases the odds of a pipe
writer on one cpu being able to pipeline data to a reader on another cpu
without having to use an IPI or tsleep/wakeup.
For the pipe1 test this brings the synchronous communications path
over a pipe (Awrite, Bread, Bwrite, Aread) down from 7uS to around 2uS.
For the pipe2 test this value greatly reduces the number of IPIs
and improves bandwidth by a few hundred megabytes/sec (the old DELAY
did the same thing so there is no change for the pipe2 test).
* Add sysctl kern.pipe.delay which defaults to 5000 nanoseconds (5uS).
This is the maximum a pipe reader will wait for additional data before
falling back to tsleep/wakeup (and related ipis). pipe_delay may be
set to 0 to disable the function. I value of at least 3000 is
recomended. Pipelining large buffers efficiently requires a higher value,
say up to 8000 or so.
* Allow kern.pipe.mpsafe to be set to 2 which adds a predictive wakeup
when a writer is found to be stalled. This currently has no significant
effect on operations due to token collisions.
* Add statistics: kern.pipe.wblocked and kern.pipe.rblocked, counting
the number of times a pipe blocks in "pipewr" or "piperd".
* Fix MP races in pipe_ioctl().
Matthew Dillon [Mon, 13 Jul 2009 05:12:50 +0000 (22:12 -0700)]
clocks - Add fine-grained spin delay function via TSC
Add int64_t target = tsc_get_target(ns) and tsc_test_target(target).
See routines for details. These functions are available when the system
supports an extremely fine-grained counter such as a TSC and may be used
to generate finely tuned delays.