]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
16 years agoFix logical bug in the bzip2 reading code, which results in bogus EIO
Maxim Sobolev [Tue, 18 Dec 2007 01:50:49 +0000 (01:50 +0000)]
Fix logical bug in the bzip2 reading code, which results in bogus EIO
returned on a perfectly valid bzip2 stream whose decompressed size
is multiple of read-ahead buffer size. Reproduce the problem is easy:
create some power-of-two sized file (truncate -s 1m file will do),
bzip2 it and try to load it as md_image from loader. See how it fails.

The bug doesn't affect gzip code (which most of bzip2-reading code was
copied from) probably due to the fact that libgzip doesn't report
Z_STREAM_END with the last block, but requires extra call to inflate()
to retrieve it and has some extra data in the input stream at that time.
However, apply similar fix to gzipfs.c just in the case the API will
change in the future to do what bzip2 code does.

Add some ifdef'ed code to enable testing bzipfs.c from witin normal
FreeBSD environment as opposed to the restricted loader one, so that
one can use gdb and whatnot.

Sponsored by: Sippy Software, Inc., http://www.sippysoft.com/
MFC in: 7 days

16 years ago- sctp-iterator should run at PI_NET priority ...not 0.
Randall Stewart [Tue, 18 Dec 2007 01:24:15 +0000 (01:24 +0000)]
- sctp-iterator should run at PI_NET priority ...not 0.

MFC after: 1 week

16 years agoCatch up with vfprintf.c,v 1.77.
David Schultz [Tue, 18 Dec 2007 01:20:33 +0000 (01:20 +0000)]
Catch up with vfprintf.c,v 1.77.

16 years agoRemove z_abs(). The z_*() functions were in libf77, and for some reason
David Schultz [Tue, 18 Dec 2007 01:15:20 +0000 (01:15 +0000)]
Remove z_abs(). The z_*() functions were in libf77, and for some reason
someone thought it would be a good idea to copy z_abs() to libm in 1994.
However, it's never been declared or documented anywhere, and I'm
reasonably confident that nobody uses it.

Discussed with: bde, deischen, kan

16 years agoonly include intr_machdep.h when it is needed for intr_bind
Kip Macy [Mon, 17 Dec 2007 23:39:28 +0000 (23:39 +0000)]
only include intr_machdep.h when it is needed for intr_bind
ia64 doesn't have an intr_machdep.h

16 years agoA few more adjustments needed because of changes in xorg. Just installing
Ken Smith [Mon, 17 Dec 2007 23:22:58 +0000 (23:22 +0000)]
A few more adjustments needed because of changes in xorg.  Just installing
xorg-server doesn't include any video drivers so install xorg-drivers as
well.  And if font-alias isn't installed the X server won't start,
complaining it can't find the font "fixed".

Insta-MFC coming, this was tested with a RELENG_6_3 release build and
the necessary packages as part of the first round of testing for 6.3-RC2.

16 years agoRemove file after being repocopied to ../ofw, where it lives on.
Marcel Moolenaar [Mon, 17 Dec 2007 22:50:39 +0000 (22:50 +0000)]
Remove file after being repocopied to ../ofw, where it lives on.

Repocopy by: simon@

16 years agoBuild the OFW loader from its new location after the sources
Marcel Moolenaar [Mon, 17 Dec 2007 22:19:44 +0000 (22:19 +0000)]
Build the OFW loader from its new location after the sources
were repocopied from ./loader to ./ofw.

16 years agoForced commit to record that this file was repocopied from
Marcel Moolenaar [Mon, 17 Dec 2007 22:18:07 +0000 (22:18 +0000)]
Forced commit to record that this file was repocopied from
src/sys/boot/powerpc/loader.

Repocopy by: simon@

16 years agoAdd myself to the copyright.
Andrew Thompson [Mon, 17 Dec 2007 18:49:44 +0000 (18:49 +0000)]
Add myself to the copyright.

16 years agoOops, the previous commit was not needed -- the file was committed but
Bruce Evans [Mon, 17 Dec 2007 18:21:23 +0000 (18:21 +0000)]
Oops, the previous commit was not needed -- the file was committed but
not checked out due to my checkout error.

16 years agoTranslate from the i386 so that this compiles and runs.
Bruce Evans [Mon, 17 Dec 2007 18:12:06 +0000 (18:12 +0000)]
Translate from the i386 so that this compiles and runs.

I hope that this and the i386 version of it will not be needed, but
this is currently about 16 cycles or 36% faster than the C version,
and the i386 version is about 8 cycles or 19% faster than the C
version, due to poor optimization of the C version.

16 years agoFix a MAC label leak for POSIX semaphores, in which per-policy labels
Robert Watson [Mon, 17 Dec 2007 17:26:32 +0000 (17:26 +0000)]
Fix a MAC label leak for POSIX semaphores, in which per-policy labels
would be properly disposed of, but the global label structure for the
semaphore wouldn't be freed.

MFC after: 3 days
Reported by: tanyong <tanyong at ercist dot iscas dot ac dot cn>,
zhouzhouyi

16 years agoFix leaking MAC labels for IPv6 inpcbs by adding missing MAC label
Robert Watson [Mon, 17 Dec 2007 17:20:57 +0000 (17:20 +0000)]
Fix leaking MAC labels for IPv6 inpcbs by adding missing MAC label
destroy call; this transpired because the inpcb alloc path for IPv4/IPv6
is the same code, but IPv6 has a separate free path.  The results was
that as new IPv6 TCP connections were created, kernel memory would
gradually leak.

MFC after: 3 days
Reported by: tanyong <tanyong at ercist dot iscas dot ac dot cn>,
zhouzhouyi

16 years agoMoved logging out of the nss_method_lookup() in order not to
Michael Bushkov [Mon, 17 Dec 2007 16:12:57 +0000 (16:12 +0000)]
Moved logging out of the nss_method_lookup() in order not to
flood logs with failed fallback method lookup attempts.

16 years agoDon't try to build s_nanl.c before it is committed.
Bruce Evans [Mon, 17 Dec 2007 13:20:38 +0000 (13:20 +0000)]
Don't try to build s_nanl.c before it is committed.

16 years agoCalculate p.fs.lookup_step correctly. This should prevent zeroing of
Oleg Bulyzhin [Mon, 17 Dec 2007 10:25:56 +0000 (10:25 +0000)]
Calculate p.fs.lookup_step correctly. This should prevent zeroing of
w_q_lookup table (used in RED algorithm for (1 - w_q)^t computation).

MFC after: 1 months

16 years agodisable update in place on transmit
Kip Macy [Mon, 17 Dec 2007 10:02:29 +0000 (10:02 +0000)]
disable update in place on transmit

16 years agoAdd SB_NOCOALESCE flag to disable socket buffer update in place
Kip Macy [Mon, 17 Dec 2007 10:02:01 +0000 (10:02 +0000)]
Add SB_NOCOALESCE flag to disable socket buffer update in place

16 years agoCalling any function from vfork other than exec* and _exit yields
Diomidis Spinellis [Mon, 17 Dec 2007 09:02:42 +0000 (09:02 +0000)]
Calling any function from vfork other than exec* and _exit yields
undefined behavior.

Noted by: alfred

16 years agoMake TCP offload work on HEAD (modulo negative interaction between sbcompress
Kip Macy [Mon, 17 Dec 2007 08:17:51 +0000 (08:17 +0000)]
Make TCP offload work on HEAD (modulo negative interaction between sbcompress
and t3_push_frames).
 - Import latest changes to cxgb_main.c and cxgb_sge.c from toestack p4 branch
 - make driver local copy of tcp_subr.c and tcp_usrreq.c and override tcp_usrreqs so
   TOE can also functions on versions with unmodified TCP

- add cxgb back to the build

16 years agoCheck NULL pointer.
David Xu [Mon, 17 Dec 2007 08:09:37 +0000 (08:09 +0000)]
Check NULL pointer.

16 years agoFix an int overflow on very large file systems.
David Schultz [Mon, 17 Dec 2007 08:03:18 +0000 (08:03 +0000)]
Fix an int overflow on very large file systems.

PR: bin/113399
Submitted by: Staffan Ulfberg <staffan@ulfberg.se>

16 years agobuild tcp_offload.c instead of tcp_ofld.c
Kip Macy [Mon, 17 Dec 2007 08:00:08 +0000 (08:00 +0000)]
build tcp_offload.c instead of tcp_ofld.c

16 years agoincorporate feedback since initial commit
Kip Macy [Mon, 17 Dec 2007 07:56:27 +0000 (07:56 +0000)]
incorporate feedback since initial commit
- rename tcp_ofld.[ch] to tcp_offload.[ch]
- document usage and locking conventions of the functions in the
  toe_usrreqs function vector
- document tcpcb, inpcb, and socket fields used by toe
- widen the listen interface into 2 functions
- rename DISABLE_TCP_OFFLOAD to TCP_OFFLOAD_DISABLE
- shrink conditional compilation to reduce the likelihood of bitrot
- replace sc->sc_toepcb checks in tcp_syncache.c with TOEPCB_ISSET

16 years agowiden the routing event interface (arp update, redirect, and eventually pmtu change)
Kip Macy [Mon, 17 Dec 2007 07:40:34 +0000 (07:40 +0000)]
widen the routing event interface (arp update, redirect, and eventually pmtu change)
into separate functions

revert previous commit's changes to arpresolve and add a new interface
arpresolve2 which does arp resolution without an mbuf

16 years agoo Fix a holiday to reflect a reality.
Maxim Konovalov [Mon, 17 Dec 2007 07:36:39 +0000 (07:36 +0000)]
o Fix a holiday to reflect a reality.

Submitted by: Andrey V. Elsukov
MFC after: 1 week

16 years agoAdd missing changes for fixing LOR of umtx lock and thread lock, follow
David Xu [Mon, 17 Dec 2007 05:55:07 +0000 (05:55 +0000)]
Add missing changes for fixing LOR of umtx lock and thread lock, follow
the committing of files:
kern_resource.c revision 1.181
sched_4bsd.c revision 1.111
sched_ule.c revision 1.218

16 years agoUse M_NOWAIT instead of M_WAITOK to cause malloc() to return NULL
Kevin Lo [Mon, 17 Dec 2007 05:08:54 +0000 (05:08 +0000)]
Use M_NOWAIT instead of M_WAITOK to cause malloc() to return NULL

Reviewed by: imp

16 years agoDon't panic in arpresolve if we're given a null mbuf. We could
Kip Macy [Mon, 17 Dec 2007 04:19:25 +0000 (04:19 +0000)]
Don't panic in arpresolve if we're given a null mbuf. We could
insist that the caller just pass in an initialized mbuf even
if didn't have any data - but that seems rather contrived.

16 years agoAdd logbl(3) to libm.
David Schultz [Mon, 17 Dec 2007 03:53:38 +0000 (03:53 +0000)]
Add logbl(3) to libm.

16 years agoAdd function prototypes.
David Xu [Mon, 17 Dec 2007 02:53:11 +0000 (02:53 +0000)]
Add function prototypes.

16 years agoRefactor features a bit in order to make it possible to disable lazy
Jason Evans [Mon, 17 Dec 2007 01:20:04 +0000 (01:20 +0000)]
Refactor features a bit in order to make it possible to disable lazy
deallocation and dynamic load balancing via the MALLOC_LAZY_FREE and
MALLOC_BALANCE knobs.  This is a non-functional change, since these
features are still enabled when possible.

Clean up a few things that more pedantic compiler settings would cause
complaints over.

16 years agoDocument the fact that we have nan(3) now, and make some minor clarifications
David Schultz [Mon, 17 Dec 2007 01:04:43 +0000 (01:04 +0000)]
Document the fact that we have nan(3) now, and make some minor clarifications
in other places.

16 years agoOn platforms where long is 64 bits, make sure gdtoa doesn't use a
David Schultz [Mon, 17 Dec 2007 00:19:49 +0000 (00:19 +0000)]
On platforms where long is 64 bits, make sure gdtoa doesn't use a
long when it really wants an int.

16 years agoFix an amusing typo that has prevented this from compiling since 2004.
David Schultz [Sun, 16 Dec 2007 23:38:55 +0000 (23:38 +0000)]
Fix an amusing typo that has prevented this from compiling since 2004.

16 years agoDon't try the long double tests on i386. Our reduced precision
David Schultz [Sun, 16 Dec 2007 23:36:37 +0000 (23:36 +0000)]
Don't try the long double tests on i386. Our reduced precision
can cause them to fail.

16 years agoFix previous commit: We should only mask the top bit in the STRTOG_NaNBits
David Schultz [Sun, 16 Dec 2007 23:31:55 +0000 (23:31 +0000)]
Fix previous commit: We should only mask the top bit in the STRTOG_NaNBits
case, not the STRTOG_Normal case.

16 years agoRemove hacks to allow libkse to export its symbols in the LIBTHREAD_1_0
Daniel Eischen [Sun, 16 Dec 2007 23:29:57 +0000 (23:29 +0000)]
Remove hacks to allow libkse to export its symbols in the LIBTHREAD_1_0
version namespace which was needed before the library version was
bumped.

16 years agoRemove another Alpha remnant.
David Schultz [Sun, 16 Dec 2007 23:00:18 +0000 (23:00 +0000)]
Remove another Alpha remnant.

16 years agoInclude cdefs.h and param.h for architectures with less header pollution
Kip Macy [Sun, 16 Dec 2007 21:22:24 +0000 (21:22 +0000)]
Include cdefs.h and param.h for architectures with less header pollution

16 years agoRegression tests for nan{,f,l}().
David Schultz [Sun, 16 Dec 2007 21:19:51 +0000 (21:19 +0000)]
Regression tests for nan{,f,l}().

16 years agoImplement and document nan(), nanf(), and nanl(). This commit
David Schultz [Sun, 16 Dec 2007 21:19:28 +0000 (21:19 +0000)]
Implement and document nan(), nanf(), and nanl(). This commit
adds two new directories in msun: ld80 and ld128. These are for
long double functions specific to the 80-bit long double format
used on x86-derived architectures, and the 128-bit format used on
sparc64, respectively.

16 years agoRecognize CFA devices using either identification method.
Poul-Henning Kamp [Sun, 16 Dec 2007 21:19:07 +0000 (21:19 +0000)]
Recognize CFA devices using either identification method.

16 years agoAdd the magic word 0 value for CFA compliant devices.
Poul-Henning Kamp [Sun, 16 Dec 2007 21:18:13 +0000 (21:18 +0000)]
Add the magic word 0 value for CFA compliant devices.

16 years agoExport gdtoa's __ULto{x,Q}_D2A routine in a private namespace so
David Schultz [Sun, 16 Dec 2007 21:15:57 +0000 (21:15 +0000)]
Export gdtoa's __ULto{x,Q}_D2A routine in a private namespace so
libm can use it.

16 years agoArrange so that the NaN returned by strtod("nan", NULL) is the same as
David Schultz [Sun, 16 Dec 2007 21:15:09 +0000 (21:15 +0000)]
Arrange so that the NaN returned by strtod("nan", NULL) is the same as
the NaN returned by strtod("nan()", NULL).

16 years agoFix strto{f,d,ld}() so they don't return signaling NaNs. C99
David Schultz [Sun, 16 Dec 2007 21:14:33 +0000 (21:14 +0000)]
Fix strto{f,d,ld}() so they don't return signaling NaNs. C99
says they are never supposed to, and the fact that they did could
cause apps that run with unmasked FP exceptions to SIGFPE after a
scanf() or strtod(). The vendor stated that he will not be fixing
this, citing portability concerns.

16 years agoSome changes the vendor didn't want:
David Schultz [Sun, 16 Dec 2007 21:13:54 +0000 (21:13 +0000)]
Some changes the vendor didn't want:

- Accept the '0x' prefix so strtod("nan(0x...)", NULL) returns the same
  thing as gcc's builtin nan("0x...") for such strings.
- Don't return uninitialized memory.
- Finish processing the string up to the closing ')' (provided it's
  lexically valid) for compatibility with C99 and *scanf().

16 years agoFix previous commit. The code ended up in the wrong function.
Rui Paulo [Sun, 16 Dec 2007 20:37:27 +0000 (20:37 +0000)]
Fix previous commit. The code ended up in the wrong function.

Approved by:      njl (mentor)

16 years agoReport erase interval (correctly) in sectors.
Poul-Henning Kamp [Sun, 16 Dec 2007 20:19:55 +0000 (20:19 +0000)]
Report erase interval (correctly) in sectors.

16 years agoRename the undocumented -E option to -X.
Poul-Henning Kamp [Sun, 16 Dec 2007 19:41:31 +0000 (19:41 +0000)]
Rename the undocumented -E option to -X.

Implement -E option which will erase the filesystem sectors before
making the new filesystem.  Reserved space in front of the superblock
(bootcode) is not erased.

NB: Erasing can take as long time as writing every sector sequentially.

This is relevant for all flash based disks which use wearlevelling.

16 years agoChop DIOCGDELETE from userland up in 1024 sector chunks to give geom_disk
Poul-Henning Kamp [Sun, 16 Dec 2007 19:38:26 +0000 (19:38 +0000)]
Chop DIOCGDELETE from userland up in 1024 sector chunks to give geom_disk
or any other bio chopping geom a reasonable size of work.

Check for delivered signals between chunks, because the request size
and service time is unbounded.

16 years agoAdd back in cxgb now that the includes are believed to be fixed
Kip Macy [Sun, 16 Dec 2007 18:45:23 +0000 (18:45 +0000)]
Add back in cxgb now that the includes are believed to be fixed

16 years agoUse the vm include convention of busdma
Kip Macy [Sun, 16 Dec 2007 18:43:57 +0000 (18:43 +0000)]
Use the vm include convention of busdma

16 years agoneed M_IOVEC define
Kip Macy [Sun, 16 Dec 2007 18:36:57 +0000 (18:36 +0000)]
need M_IOVEC define

16 years agoDon't globally include mvec.h its only needed by cxgb_sge.c
Kip Macy [Sun, 16 Dec 2007 18:26:04 +0000 (18:26 +0000)]
Don't globally include mvec.h its only needed by cxgb_sge.c

16 years agoDon't limit BIO_DELETE requests to MAXPHYS, they perform no data
Poul-Henning Kamp [Sun, 16 Dec 2007 18:03:31 +0000 (18:03 +0000)]
Don't limit BIO_DELETE requests to MAXPHYS, they perform no data
transfers, so they are not subject to the VM system limitation.

16 years agoAdd a berase() function which uses ioctl(DIOCGDELETE) to erase a slab
Poul-Henning Kamp [Sun, 16 Dec 2007 18:02:37 +0000 (18:02 +0000)]
Add a berase() function which uses ioctl(DIOCGDELETE) to erase a slab
of the disk.

16 years agoEliminate gcc "variable clobbered" warnings by declaring the variables
Diomidis Spinellis [Sun, 16 Dec 2007 17:47:34 +0000 (17:47 +0000)]
Eliminate gcc "variable clobbered" warnings by declaring the variables
living across the vfork as volatile.

Noted by: kan

16 years agoRemove cxgb module from build until I can compensate for the lack of header
Kip Macy [Sun, 16 Dec 2007 17:43:40 +0000 (17:43 +0000)]
Remove cxgb module from build until I can compensate for the lack of header
pollution on architectures other than amd64

16 years agoReport CFA extension, CFA -> (Compact) Flash Association
Poul-Henning Kamp [Sun, 16 Dec 2007 15:11:40 +0000 (15:11 +0000)]
Report CFA extension, CFA -> (Compact) Flash Association

16 years agoWhen moving a directory across devices to a place where a directory
Diomidis Spinellis [Sun, 16 Dec 2007 14:14:31 +0000 (14:14 +0000)]
When moving a directory across devices to a place where a directory
with the same name exists, delete that directory first, before performing
the copy.  This ensures that mv(1) across devices follows the semantics
of rename(2), as required by POSIX.

This change could introduce the potential of data loss, even if the
copy fails, violating the atomicity properties of rename(2).  This is
(mostly) mitigated by first renaming the destination and obliterating
it only after a succesfull copy.

The above logic also led to the introduction of code that will cleanup
the results of a partial copy, if a cross-device copy fails.

PR: bin/118367
MFC after: 1 month

16 years agoAutogenerate hardware notes for hptrr(4).
Bruce A. Mah [Sun, 16 Dec 2007 13:14:47 +0000 (13:14 +0000)]
Autogenerate hardware notes for hptrr(4).

16 years ago- Don't return 0xffff if PHY id isn't equal 0. This allows PHYs with
Stanislav Sedov [Sun, 16 Dec 2007 12:57:12 +0000 (12:57 +0000)]
- Don't return 0xffff if PHY id isn't equal 0. This allows PHYs with
  non-zero addresses to be used.

Approved by: cognet
MFC after: 2 weeks

16 years ago- Add quirk for the TrendNet TU-S9 adapter, which uses new version of PL2303
Stanislav Sedov [Sun, 16 Dec 2007 12:39:50 +0000 (12:39 +0000)]
- Add quirk for the TrendNet TU-S9 adapter, which uses new version of PL2303
  chip (revision 0x400). This allows it to be correctly identified as PL2303X.

Approved by: maxim
MFC after: 1 week

16 years agoVerify that the moved source is no longer there.
Diomidis Spinellis [Sun, 16 Dec 2007 08:25:34 +0000 (08:25 +0000)]
Verify that the moved source is no longer there.

16 years agoRegression tests for upcoming makeup of mv.
Diomidis Spinellis [Sun, 16 Dec 2007 08:11:38 +0000 (08:11 +0000)]
Regression tests for upcoming makeup of mv.
Case 20 corresponds to PR bin/118367.

16 years agoMake the reported number of tests match their actual number.
Diomidis Spinellis [Sun, 16 Dec 2007 08:02:09 +0000 (08:02 +0000)]
Make the reported number of tests match their actual number.
This fixes the reporting under prove(1)

16 years agocd to the correct directory so that the tests can be run from prove(1)
Diomidis Spinellis [Sun, 16 Dec 2007 07:55:33 +0000 (07:55 +0000)]
cd to the correct directory so that the tests can be run from prove(1)

16 years agoAdd a missing mutex unlock.
Scott Long [Sun, 16 Dec 2007 07:49:44 +0000 (07:49 +0000)]
Add a missing mutex unlock.

Reported by: Michael Jung

16 years agoturn off building of cxgb properly ... sigh
Kip Macy [Sun, 16 Dec 2007 07:44:08 +0000 (07:44 +0000)]
turn off building of cxgb properly ... sigh

16 years agoDon't use old-style mbuf iovecs
Kip Macy [Sun, 16 Dec 2007 07:41:57 +0000 (07:41 +0000)]
Don't use old-style mbuf iovecs

16 years agoUpdate tod_connect call to reflect updated interface
Kip Macy [Sun, 16 Dec 2007 07:37:48 +0000 (07:37 +0000)]
Update tod_connect call to reflect updated interface

16 years agodisable cxgb build to prevent tinderbox whining
Kip Macy [Sun, 16 Dec 2007 07:36:35 +0000 (07:36 +0000)]
disable cxgb build to prevent tinderbox whining

16 years agoo Get missed "%" in output back.
Maxim Konovalov [Sun, 16 Dec 2007 07:18:59 +0000 (07:18 +0000)]
o Get missed "%" in output back.

Submitted by: sem
MFC after: 1 week

16 years agoMove arp update upcall to always be called for ARP replies - previous invocation
Kip Macy [Sun, 16 Dec 2007 06:42:33 +0000 (06:42 +0000)]
Move arp update upcall to always be called for ARP replies - previous invocation
would not always get called at the appropriate times

16 years agoRefactor select to reduce contention and hide internal implementation
Jeff Roberson [Sun, 16 Dec 2007 06:21:20 +0000 (06:21 +0000)]
Refactor select to reduce contention and hide internal implementation
details from consumers.

 - Track individual selecters on a per-descriptor basis such that there
   are no longer collisions and after sleeping for events only those
   descriptors which triggered events must be rescaned.
 - Protect the selinfo (per descriptor) structure with a mtx pool mutex.
   mtx pool mutexes were chosen to preserve api compatibility with
   existing code which does nothing but bzero() to setup selinfo
   structures.
 - Use a per-thread wait channel rather than a global wait channel.
 - Hide select implementation details in a seltd structure which is
   opaque to the rest of the kernel.
 - Provide a 'selsocket' interface for those kernel consumers who wish to
   select on a socket when they have no fd so they no longer have to
   be aware of select implementation details.

Tested by: kris
Reviewed on: arch

16 years ago - Don't depend on header pollution to declare struct thread.
Jeff Roberson [Sun, 16 Dec 2007 06:12:53 +0000 (06:12 +0000)]
 - Don't depend on header pollution to declare struct thread.

16 years agoFix tinderbox on sun4v
Kip Macy [Sun, 16 Dec 2007 06:07:34 +0000 (06:07 +0000)]
Fix tinderbox on sun4v
include cpufunc.h so that nanoseconds(void) is defined

16 years agoMakefile for the TCP offload module
Kip Macy [Sun, 16 Dec 2007 05:40:18 +0000 (05:40 +0000)]
Makefile for the TCP offload module

16 years agoDon't use old style mbuf iovec interface
Kip Macy [Sun, 16 Dec 2007 05:34:18 +0000 (05:34 +0000)]
Don't use old style mbuf iovec interface

16 years agoUpdate the toedev's connect interface to reflect the fact that the inpcb
Kip Macy [Sun, 16 Dec 2007 05:30:21 +0000 (05:30 +0000)]
Update the toedev's connect interface to reflect the fact that the inpcb
doesn't cache the rtentry in HEAD.

16 years agoAdd driver for TCP offload
Kip Macy [Sun, 16 Dec 2007 05:27:26 +0000 (05:27 +0000)]
Add driver for TCP offload

Sponsored by: Chelsio Inc.

16 years agoUpdate the buffer management support code needed by the tcp offload module
Kip Macy [Sun, 16 Dec 2007 05:19:48 +0000 (05:19 +0000)]
Update the buffer management support code needed by the tcp offload module

16 years agoSanitize of a routine that is going away
Kip Macy [Sun, 16 Dec 2007 05:14:25 +0000 (05:14 +0000)]
Sanitize of a routine that is going away

16 years agooverlead mbuf fields for use by toe
Kip Macy [Sun, 16 Dec 2007 05:11:42 +0000 (05:11 +0000)]
overlead mbuf fields for use by toe

16 years agoAdd system includes for mvec.h
Kip Macy [Sun, 16 Dec 2007 05:11:10 +0000 (05:11 +0000)]
Add system includes for mvec.h

16 years agoAdd socket option for setting and retrieving the congestion control algorithm.
Kip Macy [Sun, 16 Dec 2007 03:30:07 +0000 (03:30 +0000)]
Add socket option for setting and retrieving the congestion control algorithm.
The name used is to allow compatibility with Linux.

16 years agoMake files under src/sys/powerpc/aim, as well as Open Firmware related
Marcel Moolenaar [Sun, 16 Dec 2007 01:02:47 +0000 (01:02 +0000)]
Make files under src/sys/powerpc/aim, as well as Open Firmware related
files dependent upon option/cpu AIM. This is in preparation of adding
support for Book-E (e500) support.

Obtained from: Juniper, Semihalf

16 years ago - Define lpohead even if userspace includes this file.
Jeff Roberson [Sun, 16 Dec 2007 00:52:32 +0000 (00:52 +0000)]
 - Define lpohead even if userspace includes this file.

16 years agoRename OEA to AIM. The former means nothing as it applies to all
Marcel Moolenaar [Sun, 16 Dec 2007 00:45:56 +0000 (00:45 +0000)]
Rename OEA to AIM. The former means nothing as it applies to all
processors (it's the PowerPC Operating Environment Architecture).
AIM designates the processors made by the Apple-IBM-Motorola
alliance and those we typically support.

While here, remove the NetBSD option IPKDB. It's not an option
used by us. Also, PPC_HAVE_FPU is not used by us either. Remove
that too.

Obtained from: Juniper, Semihalf

16 years ago- fix tab to space issue, hmm maybe I should use vi.
Randall Stewart [Sat, 15 Dec 2007 23:14:53 +0000 (23:14 +0000)]
- fix tab to space issue, hmm maybe I should use vi.

16 years ago - Re-implement lock profiling in such a way that it no longer breaks
Jeff Roberson [Sat, 15 Dec 2007 23:13:31 +0000 (23:13 +0000)]
 - Re-implement lock profiling in such a way that it no longer breaks
   the ABI when enabled.  There is no longer an embedded lock_profile_object
   in each lock.  Instead a list of lock_profile_objects is kept per-thread
   for each lock it may own.  The cnt_hold statistic is now always 0 to
   facilitate this.
 - Support shared locking by tracking individual lock instances and
   statistics in the per-thread per-instance lock_profile_object.
 - Make the lock profiling hash table a per-cpu singly linked list with a
   per-cpu static lock_prof allocator.  This removes the need for an array
   of spinlocks and reduces cache contention between cores.
 - Use a seperate hash for spinlocks and other locks so that only a
   critical_enter() is required and not a spinlock_enter() to modify the
   per-cpu tables.
 - Count time spent spinning in the lock statistics.
 - Remove the LOCK_PROFILE_SHARED option as it is always supported now.
 - Specifically drop and release the scheduler locks in both schedulers
   since we track owners now.

In collaboration with: Kip Macy
Sponsored by: Nokia

16 years agofix bonehead cut and paste error in last commit
Kip Macy [Sat, 15 Dec 2007 22:06:23 +0000 (22:06 +0000)]
fix bonehead cut and paste error in last commit

16 years agoMove NIC driver (cxgb) into its own directory. Add toecore to support
Kip Macy [Sat, 15 Dec 2007 22:03:19 +0000 (22:03 +0000)]
Move NIC driver (cxgb) into its own directory. Add toecore to support
registering different TOMs.

16 years agoImport updated support code for the TOM (tcp offload module).
Kip Macy [Sat, 15 Dec 2007 21:54:59 +0000 (21:54 +0000)]
Import updated support code for the TOM (tcp offload module).

16 years agoCreate separate capability flags for TCP over IPv4 and TCP over IPv6
Kip Macy [Sat, 15 Dec 2007 21:01:48 +0000 (21:01 +0000)]
Create separate capability flags for TCP over IPv4 and TCP over IPv6

16 years agoadd interface capability for TOE
Kip Macy [Sat, 15 Dec 2007 20:22:09 +0000 (20:22 +0000)]
add interface capability for TOE

16 years agomake naming prefixes consistent across tom_info
Kip Macy [Sat, 15 Dec 2007 20:20:08 +0000 (20:20 +0000)]
make naming prefixes consistent across tom_info