]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
17 years agoRemove unecessary locking of process in linux_getpid.
kib [Sat, 18 Nov 2006 10:12:43 +0000 (10:12 +0000)]
Remove unecessary locking of process in linux_getpid.

Suggested by: jhb
Submitted by: rdivacky

17 years ago- Use a standard section 4 SYNOPSIS
brueffer [Sat, 18 Nov 2006 08:07:43 +0000 (08:07 +0000)]
- Use a standard section 4 SYNOPSIS
- Xref uark(4)

MFC after: 3 days

17 years agoChange the internal API for polled commands. Calling mfi_polled_command
scottl [Sat, 18 Nov 2006 07:33:53 +0000 (07:33 +0000)]
Change the internal API for polled commands.  Calling mfi_polled_command
after calling mfi_mapcmd is no longer needed, so long as the MFI_CMD_POLLED
flag is set.  This change eliminates the possibility of a polled command
getting posted twice to the driver.  This is turn fixes panics on shutdown
when INVARIANTS is set.

Sponsored by: Ironport

17 years agoRemove two completely unused files
kmacy [Sat, 18 Nov 2006 07:28:47 +0000 (07:28 +0000)]
Remove two completely unused files

Reviewed by: jb (mentor rwatson)

17 years agoRemove two more duplicated files
kmacy [Sat, 18 Nov 2006 07:24:56 +0000 (07:24 +0000)]
Remove two more duplicated files

Reviewed by: jb (mentor is rwatson)

17 years agoremove 13 (largely) redundant files and switch to the sparc64/sparc64 version
kmacy [Sat, 18 Nov 2006 07:10:52 +0000 (07:10 +0000)]
remove 13 (largely) redundant files and switch to the sparc64/sparc64 version

Reviewed by: jb (mentor rwatson)

17 years ago Make the SAN login/logout stuff more common between different chipsets
mjacob [Sat, 18 Nov 2006 03:53:16 +0000 (03:53 +0000)]
 Make the SAN login/logout stuff more common between different chipsets
and provied an isp_control entry point so that the outer layers can
do PLOGI/LOGO explicitly. Add MS IOCB support. This completes the cycle
for base support for SMI-S.

17 years agoFix msgsnd(3)/msgrcv(3) deadlock under heavy resource pressure by timing out
jkim [Fri, 17 Nov 2006 20:43:01 +0000 (20:43 +0000)]
Fix msgsnd(3)/msgrcv(3) deadlock under heavy resource pressure by timing out
msgsnd and rechecking resources.  This problem was found while I was running
Linux Test Project test suite (test cases: msgctl08, msgctl09).
Change `msgwait' to `msgsnd' and `msgrcv' to distinguish its sleeping
conditions.  Few cosmetic changes to debugging messages.

17 years agoAdd support for 8 byte hardware watches in long mode. Kernel hardware
jhb [Fri, 17 Nov 2006 20:27:01 +0000 (20:27 +0000)]
Add support for 8 byte hardware watches in long mode.  Kernel hardware
watches support 8 byte watches.  For userland, we disallow 8 byte watches
for 32-bit tasks.

17 years agoModified release note: Mention move of GNOME-related ports from
bmah [Fri, 17 Nov 2006 20:15:26 +0000 (20:15 +0000)]
Modified release note:  Mention move of GNOME-related ports from
/usr/X11R6 to /usr/local.

Suggested by: kensmith

17 years agoadd wpa_passphrase utility to the build
sam [Fri, 17 Nov 2006 19:38:51 +0000 (19:38 +0000)]
add wpa_passphrase utility to the build

PR: bin/104457
Submitted by: Henrik Brix Andersen
MFC after: 1 week

17 years ago- Add macro constants for the various fields in %dr7 and use them in place
jhb [Fri, 17 Nov 2006 19:20:32 +0000 (19:20 +0000)]
- Add macro constants for the various fields in %dr7 and use them in place
  of various scattered magic values.
- Pretty print the address of hardware watchpoints in 'show watch' rather
  than just displaying hex.
- Expand address field width on amd64 for 64-bit pointers.

17 years agoDisable code to set max read byte count on the 2400.
mjacob [Fri, 17 Nov 2006 17:32:45 +0000 (17:32 +0000)]
Disable code to set max read byte count on the 2400.

It caused a panic in writing the config register on a system. Turn
it off until we take the time to understand it.

Reported by and Testing by: Anton

17 years agoMissed two nits in rev 1.7.
obrien [Fri, 17 Nov 2006 16:49:40 +0000 (16:49 +0000)]
Missed two nits in rev 1.7.

17 years agoFix the output format of MI ddb watches in 'show watch' on 64-bit
jhb [Fri, 17 Nov 2006 16:41:56 +0000 (16:41 +0000)]
Fix the output format of MI ddb watches in 'show watch' on 64-bit
platforms.

17 years agoTrim some noise from bootverbose:
jhb [Fri, 17 Nov 2006 16:41:03 +0000 (16:41 +0000)]
Trim some noise from bootverbose:
- Drop the printf in intr_machdep.c when we assign an interrupt souce to
  a CPU.  Each source already has a more detailed printf.
- Don't output a line for each ioapic pin showing its initial state, this
  has outlived its usefulness.
- When an APIC enumerator sets the bus, polarity, or trigger mode of an
  ioapic pin, just return success without printing anything if the new
  value matches the current one.

MFC after: 2 weeks

17 years agoA few more style fixes.
jhb [Fri, 17 Nov 2006 16:37:35 +0000 (16:37 +0000)]
A few more style fixes.

17 years agoWake up PIOCWAIT handler on the process exit in addition to the stop
kib [Fri, 17 Nov 2006 14:52:38 +0000 (14:52 +0000)]
Wake up PIOCWAIT handler on the process exit in addition to the stop
events. &p->p_stype is explicitely woken up on process exit for us.

Now, truss /nonexistent exits with error instead of waiting until killed
by signal.

Reported by: Nikos Vassiliadis nvass at teledomenet gr
Reviewed by: jhb
MFC after: 1 week

17 years agoErm we really want to mask all interrupts in the range, just not the first
cognet [Fri, 17 Nov 2006 11:56:56 +0000 (11:56 +0000)]
Erm we really want to mask all interrupts in the range, just not the first
one.

Submitted by: ru

17 years agoDeal more generically with the byteswap needed on !littleendian HW.
sos [Fri, 17 Nov 2006 11:13:47 +0000 (11:13 +0000)]
Deal more generically with the byteswap needed on !littleendian HW.
Account for the odd layout on ARM bigendian HW.

17 years agoShift to vim-lite package because vim package doesn't build at the moment.
kensmith [Fri, 17 Nov 2006 05:10:41 +0000 (05:10 +0000)]
Shift to vim-lite package because vim package doesn't build at the moment.

MFC after: 3 days

17 years agoAlpha no longer supported.
kensmith [Fri, 17 Nov 2006 04:24:33 +0000 (04:24 +0000)]
Alpha no longer supported.

Noticed by: ru@

17 years agoCompile -- remove an unused global variable avail_end.
kevlo [Fri, 17 Nov 2006 00:53:39 +0000 (00:53 +0000)]
Compile -- remove an unused global variable avail_end.

Approved by: cognet

17 years ago*smack* - forgot to do i386 compile, so last
mjacob [Fri, 17 Nov 2006 00:19:55 +0000 (00:19 +0000)]
*smack* - forgot to do i386 compile, so last
commit broke things.

17 years agoFinally fix local command responses to set residual correctly.
mjacob [Thu, 16 Nov 2006 23:47:16 +0000 (23:47 +0000)]
Finally fix local command responses to set residual correctly.
This allows us to play nicely on SANs when we have target mode
enabled in f/w but have neither the scsi_targbh enabled or
scsi_targ with a target enabled.

17 years agoMove the documentation to its own separate disc to make room for more
kensmith [Thu, 16 Nov 2006 23:09:35 +0000 (23:09 +0000)]
Move the documentation to its own separate disc to make room for more
packages on disc2.  This will also let users decide if they want to
have a CD of the docs at all - unless they're disconnected from the
net they will probably find the Web site more useful.

Reviewed by: ru
MFC after: 3 days

17 years agovfs_hash_insert() vputs() the losing vnode before returning, in the event of
mohans [Thu, 16 Nov 2006 23:03:46 +0000 (23:03 +0000)]
vfs_hash_insert() vputs() the losing vnode before returning, in the event of
a race where a duplicate vnode is entered into the vfs hash. nfs_nget() shouldn't
be releasing the vnode in that case.

17 years agoFix to readdir+ reply handling. When inserting an entry into the namecache,
mohans [Thu, 16 Nov 2006 23:02:37 +0000 (23:02 +0000)]
Fix to readdir+ reply handling. When inserting an entry into the namecache,
initialize the nfsnode's ctime. Otherwise a subsequent lookup purges the
just entered namecache entry.

17 years agoNote uhidev.4 removal.
brueffer [Thu, 16 Nov 2006 22:34:17 +0000 (22:34 +0000)]
Note uhidev.4 removal.

Approved by: rwatson (mentor)
MFC after: 1 week

17 years agoRemove uhidev.4 and references to it, the corresponding code was never
brueffer [Thu, 16 Nov 2006 22:32:27 +0000 (22:32 +0000)]
Remove uhidev.4 and references to it, the corresponding code was never
imported from NetBSD.

Submitted by: markus
MFC after: 1 week

17 years agoFix message size tests for msgsnd(3) and msgrcv(3). msgsz is the length of
jkim [Thu, 16 Nov 2006 19:51:10 +0000 (19:51 +0000)]
Fix message size tests for msgsnd(3) and msgrcv(3).  msgsz is the length of
actual message and it should not include size of message type.

17 years agoSwitch to linux_base-fc4 for the Linux emulation package.
kensmith [Thu, 16 Nov 2006 19:09:41 +0000 (19:09 +0000)]
Switch to linux_base-fc4 for the Linux emulation package.

MFC after: 3 days

17 years agoSwitch to emulators/linux_base-fc4 since that's the new default
kensmith [Thu, 16 Nov 2006 19:08:27 +0000 (19:08 +0000)]
Switch to emulators/linux_base-fc4 since that's the new default
Linux emulator.

MFC after: 3 days

17 years agoimplemented more validation checks about incoming responses per RFC2080, and one...
suz [Thu, 16 Nov 2006 19:03:03 +0000 (19:03 +0000)]
implemented more validation checks about incoming responses per RFC2080, and one additional heuristic check for safer operation.

Obtained from: KAME
MFC after: 1 week

17 years agoLook for capabilities in PCI-PCI bridges using the same CAP PTR register
jhb [Thu, 16 Nov 2006 17:31:33 +0000 (17:31 +0000)]
Look for capabilities in PCI-PCI bridges using the same CAP PTR register
as for type 0 devices.

Submitted by: grehan
MFC after: 1 week

17 years agoDocument the new -P option of sockstat(1) in the release notes.
keramida [Thu, 16 Nov 2006 16:07:15 +0000 (16:07 +0000)]
Document the new -P option of sockstat(1) in the release notes.

Reviewed by: bmah

17 years agoGrammar fix from Ceri.
ru [Thu, 16 Nov 2006 13:43:05 +0000 (13:43 +0000)]
Grammar fix from Ceri.

17 years agoRemove an unused variable.
ru [Thu, 16 Nov 2006 13:32:30 +0000 (13:32 +0000)]
Remove an unused variable.

17 years agoDocument that visual bell is the global setting, while
ru [Thu, 16 Nov 2006 12:37:35 +0000 (12:37 +0000)]
Document that visual bell is the global setting, while
other bell types can be set individually for each vty.

17 years agoReplace magic numbers for console bell types with defines.
ru [Thu, 16 Nov 2006 12:27:51 +0000 (12:27 +0000)]
Replace magic numbers for console bell types with defines.

17 years agoo Make pv_maxchunks no less than maxproc. This helps to survive a
maxim [Thu, 16 Nov 2006 11:46:24 +0000 (11:46 +0000)]
o Make pv_maxchunks no less than maxproc.  This helps to survive a
forkbomb explosion.

Reviewed by: alc
Security: local DoS
X-MFC atfer: RELENG_6 is not affected due to a different pv_entry
allocation code.

17 years agoAdd the Transcend 2GB card specs.
benno [Thu, 16 Nov 2006 08:04:29 +0000 (08:04 +0000)]
Add the Transcend 2GB card specs.

17 years agoResize the hash table upwards if the number of collision entries is greater than...
kmacy [Thu, 16 Nov 2006 07:50:33 +0000 (07:50 +0000)]
Resize the hash table upwards if the number of collision entries is greater than 1/4 of the
total
it is possible that the resize threshold should be revised upwards

Approved by: scottl (standing in for mentor rwatson)

17 years agoDue to an incorrect macro, it appears that this driver has always been
scottl [Thu, 16 Nov 2006 06:28:54 +0000 (06:28 +0000)]
Due to an incorrect macro, it appears that this driver has always been
accidentally truncating off the VLAN tag field in the TX descriptor.  Fix
this by splitting up the vlan_tag and flags fields into separate fields,
and handling them appropriately.

Sponsored by: Ironport
MFC After: 3 days

17 years agoIn bce_start_locked, check the used_tx_bd count rather than the
jdp [Thu, 16 Nov 2006 04:04:07 +0000 (04:04 +0000)]
In bce_start_locked, check the used_tx_bd count rather than the
descriptor's mbuf pointer to see if the transmit ring is full.  The
mbuf pointer is set only in the last descriptor of a
multi-descriptor packet.  By relying on the mbuf pointers of the
earlier descriptors, the driver would sometimes overwrite a
descriptor belonging to a packet that wasn't completed yet.  Also,
tx_chain_prod wasn't updated inside the loop, causing the wrong
descriptor to be checked after the first iteration.  The upshot of
all this was the loss of some transmitted packets at medium to high
packet rates.

In bce_tx_encap, remove a couple of old statements that shuffled
around the tx_mbuf_map pointers.  These now correspond 1-to-1 with
the transmit descriptors, and they are not supposed to be changed.

Correct a couple of inaccurate comments.

MFC after: 1 month

17 years agoAfter tests on 2 different AMD platforms with several
mjacob [Thu, 16 Nov 2006 02:40:18 +0000 (02:40 +0000)]
After tests on 2 different AMD platforms with several
different cards (SAS, 4Gb FC), MSI seems to work with
the cards.

This was of some concern because some  PCI cards
claim to work with MSI but don't.

17 years agoChange sleepq_add(9) argument from 'struct mtx *' to 'struct lock_object *',
pjd [Thu, 16 Nov 2006 01:02:00 +0000 (01:02 +0000)]
Change sleepq_add(9) argument from 'struct mtx *' to 'struct lock_object *',
which allows to use it with different kinds of locks. For example it allows
to implement Solaris conditions variables which will be used in ZFS port on
top of sx(9) locks.

Reviewed by: jhb

17 years agoMfp4: Remove mci_device.c. It should have been removed when it was
imp [Thu, 16 Nov 2006 00:55:24 +0000 (00:55 +0000)]
Mfp4: Remove mci_device.c.  It should have been removed when it was
      merged ito sd-card.c, but this is an imperfect world.

17 years agoMove to using a common arm_init.S. These things are more similar than
imp [Thu, 16 Nov 2006 00:53:28 +0000 (00:53 +0000)]
Move to using a common arm_init.S.  These things are more similar than
different at this point.

17 years agoMFp4: Don't fix the size at 8k, and some minor cleanups. andre@ contributed
imp [Thu, 16 Nov 2006 00:49:50 +0000 (00:49 +0000)]
MFp4: Don't fix the size at 8k, and some minor cleanups.  andre@ contributed
      to fixing this problem.

17 years agoMFp4: Improvements, including the ability to download to an arbitrary
imp [Thu, 16 Nov 2006 00:48:53 +0000 (00:48 +0000)]
MFp4: Improvements, including the ability to download to an arbitrary
      part of the spi flash.

17 years agoTweaks for better boot flavor support.
imp [Thu, 16 Nov 2006 00:48:04 +0000 (00:48 +0000)]
Tweaks for better boot flavor support.

17 years agoMfp4: We no longer need ee.h included here.
imp [Thu, 16 Nov 2006 00:47:31 +0000 (00:47 +0000)]
Mfp4: We no longer need ee.h included here.

17 years agoIncrease the timeout for some SAN commands.
mjacob [Thu, 16 Nov 2006 00:39:56 +0000 (00:39 +0000)]
Increase the timeout for some SAN commands.

Only complain about FC Reponse errors if they're nonzero.

Shorten some PortID printouts for local loop.

Add an internal isp_xcmd_t data structure which we'll use for some
CT-Passthru support as part of adding SMI-S.

17 years agominor change to reduce some diff noise
mjacob [Thu, 16 Nov 2006 00:31:46 +0000 (00:31 +0000)]
minor change to reduce some diff noise

17 years agoAdd big endian support.
jb [Wed, 15 Nov 2006 21:46:36 +0000 (21:46 +0000)]
Add big endian support.

Submitted by: scottl
Reviewed by: mjacob

17 years agoGet the parent dma tag if one exists. This is required on sun4v. Other
jb [Wed, 15 Nov 2006 21:41:59 +0000 (21:41 +0000)]
Get the parent dma tag if one exists. This is required on sun4v. Other
arches will default to NULL if they have no parent.

Reviewed by: mjacob

17 years agoAdjust assertions to allow for magical properties of the 'lbolt' wait
jhb [Wed, 15 Nov 2006 20:44:07 +0000 (20:44 +0000)]
Adjust assertions to allow for magical properties of the 'lbolt' wait
channel for tsleep():
- Allow tsleep() on &lbolt without Giant with a timeout 0 since &lbolt has
  an implied timeout.
- If &lbolt is used with msleep() pass NULL to sleepq_add() for the lock
  object.  Unlike other sleepq channels, &lbolt doesn't have an associated
  owning lock.

17 years agoTurn off MSI until some testing is done.
mjacob [Wed, 15 Nov 2006 20:18:09 +0000 (20:18 +0000)]
Turn off MSI until some testing is done.

17 years agoAdd MSI support to em(4), bce(4), and mpt(4). For now, we only support
jhb [Wed, 15 Nov 2006 20:04:57 +0000 (20:04 +0000)]
Add MSI support to em(4), bce(4), and mpt(4).  For now, we only support
devices that support a maximum of 1 message, and we use that 1 message
instead of the INTx rid 0 IRQ with the same interrupt handler, etc.

17 years agoAdd a note about the MSI support with details on how to turn MSI off if
jhb [Wed, 15 Nov 2006 20:02:20 +0000 (20:02 +0000)]
Add a note about the MSI support with details on how to turn MSI off if
things break.

17 years agoVarious whitespace and style fixes.
jhb [Wed, 15 Nov 2006 19:53:48 +0000 (19:53 +0000)]
Various whitespace and style fixes.

17 years agoAdd a BIND-4 related obsolete files
matteo [Wed, 15 Nov 2006 19:05:51 +0000 (19:05 +0000)]
Add a BIND-4 related obsolete files

Submitted by: Riccardo Torrini vic@gufi.org
MFC after: 1 week

17 years agoFix a typo that broke MSI (MSI-X worked fine) in the later revisions of
jhb [Wed, 15 Nov 2006 18:40:00 +0000 (18:40 +0000)]
Fix a typo that broke MSI (MSI-X worked fine) in the later revisions of
the MSI patches.

17 years ago- Use our standard section 4 SYNOPSIS
brueffer [Wed, 15 Nov 2006 16:59:15 +0000 (16:59 +0000)]
- Use our standard section 4 SYNOPSIS
- Add a HARDWARE section
- Mention the first release to include this driver

17 years agoBuilding faithd(8) has no sense if WITHOUT_INET6 is set.
yar [Wed, 15 Nov 2006 14:57:05 +0000 (14:57 +0000)]
Building faithd(8) has no sense if WITHOUT_INET6 is set.

PR: bin/100805 (audit-trail)
Noticed by: Alex Kozlov

17 years agoAdd a short regression test to try to exercise races in the non-atomic
rwatson [Wed, 15 Nov 2006 12:43:45 +0000 (12:43 +0000)]
Add a short regression test to try to exercise races in the non-atomic
nature of implied connect via sendto().  Oddly, uipc_usrreq.c implements
this for stream sockets, but doesn't set the flag in its protocol
definition so that it can actually be used.  As such, the stream test is
implemented but doesn't run for now.

17 years agoGroup pid and parent are shared in a case of CLONE_THREAD not CLONE_VM.
kib [Wed, 15 Nov 2006 11:04:37 +0000 (11:04 +0000)]
Group pid and parent are shared in a case of CLONE_THREAD not CLONE_VM.
This fix lets clone02 LTP test pass with 2.6 emulation. In reality 99%
of the cases are that CLONE_VM and CLONE_THREAD are both set so it
seemed to work.

Submitted by: rdivacky

17 years agoIn rev 1.188 of linux_misc.c the added check for valid options ommited
kib [Wed, 15 Nov 2006 10:01:06 +0000 (10:01 +0000)]
In rev 1.188 of linux_misc.c the added check for valid options ommited
__WCLONE. This fixes it thus fixing skype/teamspeak to not keep zombies
after exit.

Submitted by: rdivacky
Reported by: Bakul Shah (bakul at bitblocks com)

17 years agoo Add uark(4), a driver for Arkmicro Technologies ARK3116 based serial
maxim [Wed, 15 Nov 2006 09:13:25 +0000 (09:13 +0000)]
o Add uark(4), a driver for Arkmicro Technologies ARK3116 based serial
adapters.

Submitted by: Alex Rodin
Obtained from: OpenBSD
Reviewed by: -usb
MFC after: 6 weeks

17 years agoHeavily re-factor tte_hash to remove redundant code
kmacy [Wed, 15 Nov 2006 06:29:52 +0000 (06:29 +0000)]
Heavily re-factor tte_hash to remove redundant code
Add hash resizing support - doesn't quite work yet

17 years agoAdd a copy of the sparc64 boot crunch configuration as
jb [Wed, 15 Nov 2006 06:12:26 +0000 (06:12 +0000)]
Add a copy of the sparc64 boot crunch configuration as
a starting point for sun4v. With this commit, the 'make release'
works on sun4v and boots.

I still have to check that it actually installs to disk.

17 years agoAdd archive_write_open_filename()/archive_read_open_filename() as
kientzle [Wed, 15 Nov 2006 05:33:38 +0000 (05:33 +0000)]
Add archive_write_open_filename()/archive_read_open_filename() as
synonyms for archive_write_open_file()/archive_read_open_file().
The new names are much clearer.

17 years agoStyle: Use the correct type for a sizeof() variable.
kientzle [Wed, 15 Nov 2006 05:20:14 +0000 (05:20 +0000)]
Style: Use the correct type for a sizeof() variable.

17 years agoChange the internal API for writing data to an entry; make the
kientzle [Wed, 15 Nov 2006 05:14:20 +0000 (05:14 +0000)]
Change the internal API for writing data to an entry; make the
internal format-specific functions return the same as the public
function, so that the public API layer doesn't have to guess the
correct return value.  This addresses an obscure problem that occurs
when someone tries to write more data than the size of the entry (as
indicated in the entry header).  In this case, the return value from
archive_write_data() was incorrect, reflecting the requested write
rather than the amount actually written.

MFC after: 15 days

17 years agoadd trap trace to tl1 trap
kmacy [Wed, 15 Nov 2006 03:53:27 +0000 (03:53 +0000)]
add trap trace to tl1 trap

17 years agoadd trap tracing to dev_mondo
kmacy [Wed, 15 Nov 2006 03:20:12 +0000 (03:20 +0000)]
add trap tracing to dev_mondo

17 years agoadd trap tracing to cpu mondo handler and tsb miss handler
kmacy [Wed, 15 Nov 2006 03:16:30 +0000 (03:16 +0000)]
add trap tracing to cpu mondo handler and tsb miss handler

17 years agoAdded yet another extra fxp(4) PCI ID.
rink [Tue, 14 Nov 2006 18:54:31 +0000 (18:54 +0000)]
Added yet another extra fxp(4) PCI ID.

PR: kern/104896
Submitted by: Yoshihiko Sarumaru <mistral@imasy.or.jp>
Reviewed by: imp (mentor), jfv
Approved by: imp (mentor)
MFC after: 3 days

17 years agoFix a couple of comment typos.
jhb [Tue, 14 Nov 2006 17:54:55 +0000 (17:54 +0000)]
Fix a couple of comment typos.

Reported by: ru

17 years ago- Add in FreeBSD native ioctl that models the Linux version.
ambrisko [Tue, 14 Nov 2006 16:48:00 +0000 (16:48 +0000)]
- Add in FreeBSD native ioctl that models the Linux version.
- Add a translation so the Linux ioctl's don't conflict with
  the FreeBSD definition.
- Assume Linux 32bit emulation on amd64.
This was tested on i386 and amd64 with the 32bit Linux MegaCli.
Eventually we should do a 32bit native FreeBSD translation app.

17 years agoUnbreak pc98.
ru [Tue, 14 Nov 2006 14:28:09 +0000 (14:28 +0000)]
Unbreak pc98.

17 years agoAdd `PN_XNUM', for supporting ELF objects with a large number of program
jkoshy [Tue, 14 Nov 2006 12:10:39 +0000 (12:10 +0000)]
Add `PN_XNUM', for supporting ELF objects with a large number of program
header table entries.

17 years agoUse tabs instead of spaces.
pjd [Tue, 14 Nov 2006 11:40:45 +0000 (11:40 +0000)]
Use tabs instead of spaces.

17 years agoCorrect a typo.
bz [Tue, 14 Nov 2006 11:16:34 +0000 (11:16 +0000)]
Correct a typo.

17 years agoPush things closer to path failover by implementing loop down and
mjacob [Tue, 14 Nov 2006 08:45:48 +0000 (08:45 +0000)]
Push things closer to path failover by implementing loop down and
gone device timers and zombie state entries. There are tunables
that can be used to select a number of parameters.

loop_down_limit - how long to wait for loop to come back up before
declaring
all devices dead (default 300 seconds)

gone_device_time- how long to wait for a device that has appeared
to leave the loop or fabric to reappear (default 30 seconds)

Internal tunables include (which should be externalized):

quick_boot_time- how long to wait when booting for loop to come up

change_is_bad- whether or not to accept devices with the same
WWNN/WWPN that reappear at a different PortID as being the 'same'
device.

Keen students of some of the subtle issues here will ask how
one can keep devices from being re-accepted at all (the answer
is to set a gone_device_time to zero- that effectively would
be the same thing).

17 years agoAdd Kris Kennaway to reflect his performance improvements to
kmacy [Tue, 14 Nov 2006 07:57:19 +0000 (07:57 +0000)]
Add Kris Kennaway to reflect his performance improvements to
the lock profiling code

17 years agoFix a copy-paste bug in NON-KSE case.
davidxu [Tue, 14 Nov 2006 05:48:27 +0000 (05:48 +0000)]
Fix a copy-paste bug in NON-KSE case.

17 years agoFix debugging output of '-d', to more accurately reflect if
rodrigc [Tue, 14 Nov 2006 01:07:42 +0000 (01:07 +0000)]
Fix debugging output of '-d', to more accurately reflect if
we exec an external mount program, or just call nmount()
to mount a filesystem.

Noticed by: kris

17 years agoMD support for PCI Message Signalled Interrupts on amd64 and i386:
jhb [Mon, 13 Nov 2006 22:23:34 +0000 (22:23 +0000)]
MD support for PCI Message Signalled Interrupts on amd64 and i386:
- Add a new apic_alloc_vectors() method to the local APIC support code
  to allocate N contiguous IDT vectors (aligned on a M >= N boundary).
  This function is used to allocate IDT vectors for a group of MSI
  messages.
- Add MSI and MSI-X PICs.  The PIC code here provides methods to manage
  edge-triggered MSI messages as x86 interrupt sources.  In addition to
  the PIC methods, msi.c also includes methods to allocate and release
  MSI and MSI-X messages.  For x86, we allow for up to 128 different
  MSI IRQs starting at IRQ 256 (IRQs 0-15 are reserved for ISA IRQs,
  16-254 for APIC PCI IRQs, and IRQ 255 is reserved).
- Add pcib_(alloc|release)_msi[x]() methods to the MD x86 PCI bridge
  drivers to bubble the request up to the nexus driver.
- Add pcib_(alloc|release)_msi[x]() methods to the x86 nexus drivers that
  ask the MSI PIC code to allocate resources and IDT vectors.

MFC after: 2 months

17 years agoFirst cut at MI support for PCI Message Signalled Interrupts (MSI):
jhb [Mon, 13 Nov 2006 21:47:30 +0000 (21:47 +0000)]
First cut at MI support for PCI Message Signalled Interrupts (MSI):
- Add 3 new functions to the pci_if interface along with suitable wrappers
  to provide the device driver visible API:
  - pci_alloc_msi(dev, int *count) backed by PCI_ALLOC_MSI().  '*count'
    here is an in and out parameter.  The driver stores the desired number
    of messages in '*count' before calling the function.  On success,
    '*count' holds the number of messages allocated to the device.  Also on
    success, the driver can access the messages as SYS_RES_IRQ resources
    starting at rid 1.  Note that the legacy INTx interrupt resource will
    not be available when using MSI.  Note that this function will allocate
    either MSI or MSI-X messages depending on the devices capabilities and
    the 'hw.pci.enable_msix' and 'hw.pci.enable_msi' tunables.  Also note
    that the driver should activate the memory resource that holds the
    MSI-X table and pending bit array (PBA) before calling this function
    if the device supports MSI-X.
  - pci_release_msi(dev) backed by PCI_RELEASE_MSI().  This function
    releases the messages allocated for this device.  All of the
    SYS_RES_IRQ resources need to be released for this function to succeed.
  - pci_msi_count(dev) backed by PCI_MSI_COUNT().  This function returns
    the maximum number of MSI or MSI-X messages supported by this device.
    MSI-X is preferred if present, but this function will honor the
    'hw.pci.enable_msix' and 'hw.pci.enable_msi' tunables.  This function
    should return the largest value that pci_alloc_msi() can return
    (assuming the MD code is able to allocate sufficient backing resources
    for all of the messages).
- Add default implementations for these 3 methods to the pci_driver generic
  PCI bus driver.  (The various other PCI bus drivers such as for ACPI and
  OFW will inherit these default implementations.)  This default
  implementation depends on 4 new pcib_if methods that bubble up through
  the PCI bridges to the MD code to allocate IRQ values and perform any
  needed MD setup code needed:
  - PCIB_ALLOC_MSI() attempts to allocate a group of MSI messages.
  - PCIB_RELEASE_MSI() releases a group of MSI messages.
  - PCIB_ALLOC_MSIX() attempts to allocate a single MSI-X message.
  - PCIB_RELEASE_MSIX() releases a single MSI-X message.
- Add default implementations for these 4 methods that just pass the
  request up to the parent bus's parent bridge driver and use the
  default implementation in the various MI PCI bridge drivers.
- Add MI functions for use by MD code when managing MSI and MSI-X
  interrupts:
  - pci_enable_msi(dev, address, data) programs the MSI capability address
    and data registers for a group of MSI messages
  - pci_enable_msix(dev, index, address, data) initializes a single MSI-X
    message in the MSI-X table
  - pci_mask_msix(dev, index) masks a single MSI-X message
  - pci_unmask_msix(dev, index) unmasks a single MSI-X message
  - pci_pending_msix(dev, index) returns true if the specified MSI-X
    message is currently pending
- Save the MSI capability address and data registers in the pci_cfgreg
  block in a PCI devices ivars and restore the values when a device is
  resumed.  Note that the MSI-X table is not currently restored during
  resume.
- Add constants for MSI-X register offsets and fields.
- Record interesting data about any MSI-X capability blocks we come
  across in the pci_cfgreg block in the ivars for PCI devices.

Tested on: em (i386, MSI), bce (amd64/i386, MSI), mpt (amd64, MSI-X)
Reviewed by: scottl, grehan, jfv
MFC after: 2 months

17 years agoVarious fixes:
jhb [Mon, 13 Nov 2006 21:14:54 +0000 (21:14 +0000)]
Various fixes:
- Remove an extra entry from the array for 0x0f prefixed instruction groups.
  This fixes decoding of instructions where the second opcode >= 0x80.
- Add support for the 64-bit immediate mov instructions.
- When short_addr is enabled, don't parse the modr/m byte for a 16-bit
  address, but as a 32-bit address.
- Support %rip relative addressing.
- Don't print a displacement of 0 if there is a base or index register.

MFC after: 3 days

17 years agoFix NKPT comments to match reality. Note that the current value
ru [Mon, 13 Nov 2006 20:33:54 +0000 (20:33 +0000)]
Fix NKPT comments to match reality.  Note that the current value
of NKPT is no longer enough to run amd64 with 16G of RAM, as it
doesn't have space for mapping a kernel (16M kernel would require
additionally 8 page tables).

17 years agoNew release note: traceroute(8) -D (+MFC).
bmah [Mon, 13 Nov 2006 19:13:39 +0000 (19:13 +0000)]
New release note:  traceroute(8) -D (+MFC).

17 years agoAdd SCTP as a known upper layer protocol over v6.
bz [Mon, 13 Nov 2006 19:07:32 +0000 (19:07 +0000)]
Add SCTP as a known upper layer protocol over v6.
We are not yet aware of the protocol internals but this way
SCTP traffic over v6 will not be discarded.

Reported by: Peter Lei via rrs
Tested by:   Peter Lei <peterlei cisco.com>

17 years agoNo change in functionality, but fill in a missing error
kientzle [Mon, 13 Nov 2006 16:50:18 +0000 (16:50 +0000)]
No change in functionality, but fill in a missing error
message when reading a truncated tar archive.

17 years agoFix minor formatting issues:
ru [Mon, 13 Nov 2006 09:46:16 +0000 (09:46 +0000)]
Fix minor formatting issues:

- make document title match filename;
- remove hard sentence breaks, whitespace at EOL, and double whitespace;
- sort SEE ALSO xrefs, adding missing section numbers;
- fix a misspelled macro name.

17 years agoThe utility accepts multiple arguments.
ru [Mon, 13 Nov 2006 09:12:25 +0000 (09:12 +0000)]
The utility accepts multiple arguments.

17 years agoSort xrefs.
ru [Mon, 13 Nov 2006 08:56:42 +0000 (08:56 +0000)]
Sort xrefs.

17 years agoFix a comment.
ru [Mon, 13 Nov 2006 06:26:57 +0000 (06:26 +0000)]
Fix a comment.