]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
14 years agoPrefer memory space register mapping over io space. If memory space
Pyun YongHyeon [Tue, 22 Dec 2009 21:39:34 +0000 (21:39 +0000)]
Prefer memory space register mapping over io space. If memory space
mapping fails fall back to old io space mapping.
While I'm here use PCIR_BAR macro.

14 years agoEnroll these drivers in multipass probing. The motivation behind this
Marius Strobl [Tue, 22 Dec 2009 21:02:46 +0000 (21:02 +0000)]
Enroll these drivers in multipass probing. The motivation behind this
is that the JBus to EBus bridges share the interrupt controller of a
sibling JBus to PCIe bridge (at least as far as the OFW device tree
is concerned, in reality they are part of the same chip) so we have to
probe and attach the latter first. That happens to be also the case
due to the fact that the JBus to PCIe bridges appear first in the OFW
device tree but it doesn't hurt to ensure the right order.

14 years agoInstead of relying on hard resetting of controller to stop
Pyun YongHyeon [Tue, 22 Dec 2009 20:57:30 +0000 (20:57 +0000)]
Instead of relying on hard resetting of controller to stop
receiving incoming traffics, try harder to gracefully stop active
DMA cycles and then stop MACs. This is the way what datasheet
recommends and seems to work reliably. Resetting controller while
active DMAs are in progress is bad thing as we can't predict how
DMAs touche allocated TX/RX buffers. This change ensures controller
stop state before attempting to release allocated TX/RX buffers.
Also update MAC statistics which could have been updated during the
wait time of MAC stop.

While I'm here remove unnecessary controller resets in various
location. ste(4) no longer relies on hard controller reset to stop
controller and resetting controller also clears all configured
settings which makes it hard to implement WOL in near future.
Now resetting a controller is performed in ste_init_locked().

14 years agoDon't build zfsboot, gptzfsboot, and zfsloader if WITHOUT_ZFS is enabled.
John Baldwin [Tue, 22 Dec 2009 20:56:33 +0000 (20:56 +0000)]
Don't build zfsboot, gptzfsboot, and zfsloader if WITHOUT_ZFS is enabled.

MFC after: 1 week

14 years agoUse ALLOW_NEW_SOURCES and BLOCK_OLD_SOURCES to signal a join or leave
Bruce M Simpson [Tue, 22 Dec 2009 20:40:22 +0000 (20:40 +0000)]
Use ALLOW_NEW_SOURCES and BLOCK_OLD_SOURCES to signal a join or leave
with SSM MLDv2 by default.
This is current practice and complies with RFC 4604, as well as being
required by production IPv6 networks in Japan.
The behaviour may be disabled by setting the net.inet6.mld.use_allow
sysctl/tunable to 0.

Requested by: Hideki Yamamoto
MFC after: 1 week

14 years agoReimplement miibus_statchg method. Don't rely on link state change
Pyun YongHyeon [Tue, 22 Dec 2009 20:11:56 +0000 (20:11 +0000)]
Reimplement miibus_statchg method. Don't rely on link state change
interrupt. If we want to use link state change interrupt ste(4)
should also implement auto-negotiation complete handler as well as
various PHY access handling. Now link state change is handled by
mii(4) polling so it will automatically update link state UP/DOWN
events which in turn make ste(4) usable with lagg(4).

r199559 added a private timer to drive watchdog and the timer also
used to drive MAC statistics update. Because the MAC statistics
update is called whenever statistics counter reaches near-full, it
drove watchdog timer too fast such that it caused false watchdog
timeouts under heavy TX traffic conditions.
Fix the regression by separating ste_stats_update() from driving
watchdog timer and introduce a new function ste_tick() that handles
periodic job such as driving watchdog, MAC statistics update and
link state check etc.
While I'm here clear armed watchdog timer in ste_stop().

14 years agoAdd support for Intel SCH PATA controller.
Alexander Motin [Tue, 22 Dec 2009 19:48:06 +0000 (19:48 +0000)]
Add support for Intel SCH PATA controller.

PR: kern/140251

14 years agoIntroduce sc_flags member variable and use it to keep track of
Pyun YongHyeon [Tue, 22 Dec 2009 19:32:16 +0000 (19:32 +0000)]
Introduce sc_flags member variable and use it to keep track of
link state and PHY related information.
Remove ste_link and ste_one_phy variable of softc as it's not used
anymore.
While I'm here add IFF_DRV_RUNNING check in ste_start_locked().

14 years agomerge code from ipfw3-head to reduce contention on the ipfw lock
Luigi Rizzo [Tue, 22 Dec 2009 19:01:47 +0000 (19:01 +0000)]
merge code from ipfw3-head to reduce contention on the ipfw lock
and remove all O(N) sequences from kernel critical sections in ipfw.

In detail:

 1. introduce a IPFW_UH_LOCK to arbitrate requests from
     the upper half of the kernel. Some things, such as 'ipfw show',
     can be done holding this lock in read mode, whereas insert and
     delete require IPFW_UH_WLOCK.

  2. introduce a mapping structure to keep rules together. This replaces
     the 'next' chain currently used in ipfw rules. At the moment
     the map is a simple array (sorted by rule number and then rule_id),
     so we can find a rule quickly instead of having to scan the list.
     This reduces many expensive lookups from O(N) to O(log N).

  3. when an expensive operation (such as insert or delete) is done
     by userland, we grab IPFW_UH_WLOCK, create a new copy of the map
     without blocking the bottom half of the kernel, then acquire
     IPFW_WLOCK and quickly update pointers to the map and related info.
     After dropping IPFW_LOCK we can then continue the cleanup protected
     by IPFW_UH_LOCK. So userland still costs O(N) but the kernel side
     is only blocked for O(1).

  4. do not pass pointers to rules through dummynet, netgraph, divert etc,
     but rather pass a <slot, chain_id, rulenum, rule_id> tuple.
     We validate the slot index (in the array of #2) with chain_id,
     and if successful do a O(1) dereference; otherwise, we can find
     the rule in O(log N) through <rulenum, rule_id>

All the above does not change the userland/kernel ABI, though there
are some disgusting casts between pointers and uint32_t

Operation costs now are as follows:

  Function Old Now   Planned
-------------------------------------------------------------------
  + skipto X, non cached O(N) O(log N)
  + skipto X, cached O(1) O(1)
XXX dynamic rule lookup O(1) O(log N)  O(1)
  + skipto tablearg O(N) O(1)
  + reinject, non cached O(N) O(log N)
  + reinject, cached O(1) O(1)
  + kernel blocked during setsockopt() O(N) O(1)
-------------------------------------------------------------------

The only (very small) regression is on dynamic rule lookup and this will
be fixed in a day or two, without changing the userland/kernel ABI

Supported by: Valeria Paoli
MFC after: 1 month

14 years agoAdd minimal dealy while ste(4) is waiting for the end of active DMA
Pyun YongHyeon [Tue, 22 Dec 2009 19:00:18 +0000 (19:00 +0000)]
Add minimal dealy while ste(4) is waiting for the end of active DMA
cycle.

14 years agoAdd bus_dma(9) and endianness support to ste(4).
Pyun YongHyeon [Tue, 22 Dec 2009 18:57:07 +0000 (18:57 +0000)]
Add bus_dma(9) and endianness support to ste(4).
 o Sorted includes and added missing header files.
 o Added basic endianness support. In theory ste(4) should work on
   any architectures.
 o Remove the use of contigmalloc(9), contigfree(9) and vtophys(9).
 o Added 8 byte alignment limitation of TX/RX descriptor.
 o Added 1 byte alignment requirement for TX/RX buffers.
 o ste(4) controllers does not support DAC. Limit DMA address space
   to be within 32bit address.
 o Added spare DMA map to gracefully recover from DMA map failure.
 o Removed dead code for checking STE_RXSTAT_DMADONE bit. The bit
   was already checked in each iteration of loop so it can't be true.
 o Added second argument count to ste_rxeof(). It is used to limit
   number of iterations done in RX handler. ATM polling is the only
   consumer.
 o Removed ste_rxeoc() which was added to address RX stuck issue
   (cvs rev 1.66). Unlike TX descriptors, ST201 supports chaining
   descriptors to form a ring for RX descriptors. If RX descriptor
   chaining is not supported it's possible for controller to stop
   receiving incoming frames once controller pass the end of RX
   descriptor which in turn requires driver post new RX
   descriptors to receive more frames. For TX descriptors which
   does not support chaning, we exactly do manual chaining in
   driver by concatenating new descriptors to the end of previous
   TX chain.
   Maybe the workaround was borrowed from other drivers that does
   not support RX descriptor chaining, which is not valid for ST201
   controllers. I still have no idea how this address RX stuck
   issue and I can't reproduce the RX stuck issue on DFE-550TX
   controller.
 o Removed hw.ste_rxsyncs sysctl as the workaround was removed.
 o TX/RX side bus_dmamap_load_mbuf_sg(9) support.
 o Reimplemented optimized ste_encap().
 o Simplified TX logic of ste_start_locked().
 o Added comments for TFD/RFD requirements.
 o Increased number of RX descriptors to 128 from 64. 128 gave much
   better performance than 64 under high network loads.

14 years agoRemoved duplicate usbd_xfer_state(9) link.
Ruslan Ermilov [Tue, 22 Dec 2009 16:05:28 +0000 (16:05 +0000)]
Removed duplicate usbd_xfer_state(9) link.

14 years agoSort mlinks.
Ruslan Ermilov [Tue, 22 Dec 2009 16:02:08 +0000 (16:02 +0000)]
Sort mlinks.

14 years ago- Rename the __tcpi_(snd|rcv)_mss fields of the tcp_info structure to remove
John Baldwin [Tue, 22 Dec 2009 15:47:40 +0000 (15:47 +0000)]
- Rename the __tcpi_(snd|rcv)_mss fields of the tcp_info structure to remove
  the leading underscores since they are now implemented.
- Implement the tcpi_rto and tcpi_last_data_recv fields in the tcp_info
  structure.

Reviewed by: rwatson
MFC after: 2 weeks

14 years agoDon't take the driver mutex in mxge_tick(), as it
Andrew Gallatin [Tue, 22 Dec 2009 15:41:58 +0000 (15:41 +0000)]
Don't take the driver mutex in mxge_tick(), as it
is run with the mutex held.

Submitted by: rwatson
MFC after: 3 days

14 years agoDon't print the archive name with -p and -q options.
Jaakko Heinonen [Tue, 22 Dec 2009 15:13:16 +0000 (15:13 +0000)]
Don't print the archive name with -p and -q options.

PR: bin/141280
Approved by: des, trasz (mentor)

14 years agoapm(8) is no longer linked to zzz(8), catch up.
Ruslan Ermilov [Tue, 22 Dec 2009 14:21:08 +0000 (14:21 +0000)]
apm(8) is no longer linked to zzz(8), catch up.
(Fixes "zzz" clash in the whatis(1) database.)

14 years agosome mostly cosmetic changes in preparation for upcoming work:
Luigi Rizzo [Tue, 22 Dec 2009 13:53:34 +0000 (13:53 +0000)]
some mostly cosmetic changes in preparation for upcoming work:

+ in many places, replace &V_layer3_chain with a local
  variable chain;
+ bring the counter of rules and static_len within ip_fw_chain
  replacing static variables;
+ remove some spurious comments and extern declaration;
+ document which lock protects certain data structures

14 years agoMFV of tzcode2009t, r200833
Edwin Groothuis [Tue, 22 Dec 2009 11:21:03 +0000 (11:21 +0000)]
MFV of tzcode2009t, r200833

Comment only, no data changes.

MFC after: 1 week

14 years agoVendor import of tzcode2009t:
Edwin Groothuis [Tue, 22 Dec 2009 11:18:31 +0000 (11:18 +0000)]
Vendor import of tzcode2009t:

Comments only, no data changes.

Obtained from: ftp://elsie.nci.nih.gov/pub/

14 years agoMFV of tzdata2009t, r200831
Edwin Groothuis [Tue, 22 Dec 2009 11:17:10 +0000 (11:17 +0000)]
MFV of tzdata2009t, r200831

zic:
- Fix URL / reference to Calendrical Calculations: Third Edition

libc/stdtime:
- Fix typo in tzfile.5 (no changes in our part)

MFC after: 1 week

14 years agoVendor import of tzcode2009t:
Edwin Groothuis [Tue, 22 Dec 2009 11:08:07 +0000 (11:08 +0000)]
Vendor import of tzcode2009t:

Unused:
- Updates to tz-link.html and tz-art.htm
- Typo in tzfile.5.txt

zic:
- Fix URL / reference to Calendrical Calculations: Third Edition

libc/stdtime:
- Fix typo in tzfile.5

14 years agoCosmetic fixes.
Edward Tomasz Napierala [Tue, 22 Dec 2009 09:03:59 +0000 (09:03 +0000)]
Cosmetic fixes.

14 years agoAdd missed usb product define in r200826.
Andrew Thompson [Tue, 22 Dec 2009 02:15:36 +0000 (02:15 +0000)]
Add missed usb product define in r200826.

14 years agoadd new ID to UFTDI driver.
Andrew Thompson [Tue, 22 Dec 2009 02:13:23 +0000 (02:13 +0000)]
add new ID to UFTDI driver.

Submitted by: YAMAMOTO, Shigeru

14 years ago- add support for more than 2 audio channels. [1]
Andrew Thompson [Tue, 22 Dec 2009 02:11:37 +0000 (02:11 +0000)]
- add support for more than 2 audio channels. [1]
- add support for more sample rates

Submitted by: [1] ariff (earlier version), Hans Petter Selasky

14 years agoSet correct USB device description
Andrew Thompson [Tue, 22 Dec 2009 02:04:16 +0000 (02:04 +0000)]
Set correct USB device description

Submitted by: Paul B Mahol

14 years agoAdd more OHCI pci ids.
Andrew Thompson [Tue, 22 Dec 2009 01:59:17 +0000 (01:59 +0000)]
Add more OHCI pci ids.

Submitted by: Hans Petter Selasky

14 years agoAdd more EHCI pci ids.
Andrew Thompson [Tue, 22 Dec 2009 01:57:34 +0000 (01:57 +0000)]
Add more EHCI pci ids.

Submitted by: Hans Petter Selasky

14 years agoMake graid3 fallback to malloc() when component request size is bigger
Alexander Motin [Mon, 21 Dec 2009 23:31:03 +0000 (23:31 +0000)]
Make graid3 fallback to malloc() when component request size is bigger
then maximal prepared UMA zone size. This fixes crash with MAXPHYS > 128K.

14 years agoSupport the tablet in (at least) the Toshiba Portege M200 Tablet PC.
Gavin Atkinson [Mon, 21 Dec 2009 22:57:40 +0000 (22:57 +0000)]
Support the tablet in (at least) the Toshiba Portege M200 Tablet PC.
This device only appears on the ACPI bus, so isn't caught by the current
entry for it in the uart(4) ISA attachment.

PR: kern/140172
Reviewed by: jhb, marcel
Approved by: ed (mentor)
MFC after: 2 weeks

14 years agoGrammar and minor tweaks to powerd(8) man page.
Gavin Atkinson [Mon, 21 Dec 2009 22:56:06 +0000 (22:56 +0000)]
Grammar and minor tweaks to powerd(8) man page.

PR: docs/133186
Approved by: ed (mentor)
MFC after: 2 weeks

14 years agorc.subr: Use pwait in wait_for_pids.
Jilles Tjoelker [Mon, 21 Dec 2009 22:16:07 +0000 (22:16 +0000)]
rc.subr: Use pwait in wait_for_pids.

This waits for the requested process(es) to terminate, rather than polling
with an interval of 2 seconds.

If pwait is not available, the old method is used.

PR: conf/132766
Reviewed by: dougb

14 years agoSpell AMD properly.
Alexander Motin [Mon, 21 Dec 2009 21:47:33 +0000 (21:47 +0000)]
Spell AMD properly.

14 years agoAdd missing module dependency information.
Marius Strobl [Mon, 21 Dec 2009 21:41:33 +0000 (21:41 +0000)]
Add missing module dependency information.

14 years agoProvide and consume missing module dependency information.
Marius Strobl [Mon, 21 Dec 2009 21:29:16 +0000 (21:29 +0000)]
Provide and consume missing module dependency information.

14 years agoClear all ports interrupt status bits in single write. Clearing one by one
Alexander Motin [Mon, 21 Dec 2009 21:27:56 +0000 (21:27 +0000)]
Clear all ports interrupt status bits in single write. Clearing one by one
causes additional MSIs messages sent if several ports asked for attention
same time. Time window before clearing is not important, as these interrupts
are level triggered by interrupt source.

14 years agoAdd regression test for NFSv4 ACLs on UFS.
Edward Tomasz Napierala [Mon, 21 Dec 2009 20:47:41 +0000 (20:47 +0000)]
Add regression test for NFSv4 ACLs on UFS.

14 years agoSort function prototyes.
Pyun YongHyeon [Mon, 21 Dec 2009 20:42:23 +0000 (20:42 +0000)]
Sort function prototyes.

14 years agoBy popular request, execute rc.d scripts in the same environment
Doug Barton [Mon, 21 Dec 2009 20:32:04 +0000 (20:32 +0000)]
By popular request, execute rc.d scripts in the same environment
as they are at boot time.

Original concept submitted by: ume

14 years agostyle(9)
Pyun YongHyeon [Mon, 21 Dec 2009 20:18:01 +0000 (20:18 +0000)]
style(9)

14 years agoDon't consider non-existence of a PID file an error, we should be able
Xin LI [Mon, 21 Dec 2009 20:12:01 +0000 (20:12 +0000)]
Don't consider non-existence of a PID file an error, we should be able
to proceed anyway as this most likely mean that the process has been
terminated.

PR: bin/140397
Submitted by: Dan Lukes <dan obluda cz>
MFC after: 1 month

14 years agoRemove commented out prototype for ifinit(). This prototype has been
John Baldwin [Mon, 21 Dec 2009 20:09:19 +0000 (20:09 +0000)]
Remove commented out prototype for ifinit().  This prototype has been
commented out since 1.1 and has not been present in <sys/systm.h> since at
least 1.1 of that file.  It is also not needed in FreeBSD due to SYSINIT().

14 years agoRemove trailing white spaces.
Pyun YongHyeon [Mon, 21 Dec 2009 20:02:12 +0000 (20:02 +0000)]
Remove trailing white spaces.

14 years agos/u_intXX_t/uintXX_t/g
Pyun YongHyeon [Mon, 21 Dec 2009 20:00:27 +0000 (20:00 +0000)]
s/u_intXX_t/uintXX_t/g

14 years agoUse vsprintf instead of rolling our own.
Xin LI [Mon, 21 Dec 2009 19:59:38 +0000 (19:59 +0000)]
Use vsprintf instead of rolling our own.

PR: bin/140496
Submitted by: Jeremy Huddleston <jeremyhu apple.com>
MFC after: 1 month

14 years ago o Remove unnecessary return statement.
Pyun YongHyeon [Mon, 21 Dec 2009 19:56:11 +0000 (19:56 +0000)]
 o Remove unnecessary return statement.
 o Remove register keyword.

14 years agoUse vsscanf instead of rolling our own.
Xin LI [Mon, 21 Dec 2009 19:56:03 +0000 (19:56 +0000)]
Use vsscanf instead of rolling our own.

PR: bin/140530
Submitted by: Jeremy Huddleston <jeremyhu apple.com>
MFC after: 1 month

14 years agoK&R -> ANSI prototype.
Xin LI [Mon, 21 Dec 2009 19:55:05 +0000 (19:55 +0000)]
K&R -> ANSI prototype.

MFC after: 1 month

14 years agoUse ANSI function definations.
Pyun YongHyeon [Mon, 21 Dec 2009 19:50:29 +0000 (19:50 +0000)]
Use ANSI function definations.

14 years agoUse _once() to initialize the pthread key for thread-local storage to hold
John Baldwin [Mon, 21 Dec 2009 19:43:23 +0000 (19:43 +0000)]
Use _once() to initialize the pthread key for thread-local storage to hold
the results of localtime() instead of using a pthread mutex directly.

MFC after: 1 week

14 years agoImplement NFSv4 ACL support for UFS.
Edward Tomasz Napierala [Mon, 21 Dec 2009 19:39:10 +0000 (19:39 +0000)]
Implement NFSv4 ACL support for UFS.

Reviewed by: rwatson

14 years agoAdd support of using environment variable BURNCD_SPEED to specify
Xin LI [Mon, 21 Dec 2009 19:27:52 +0000 (19:27 +0000)]
Add support of using environment variable BURNCD_SPEED to specify
recodring speed.

PR: bin/140530
Submitted by: Alexander Best <alexbestms wwu.de>
MFC after: 1 month

14 years agoPlug a memory leak.
Xin LI [Mon, 21 Dec 2009 19:18:27 +0000 (19:18 +0000)]
Plug a memory leak.

PR: bin/141835
Submitted by: Henning Petersen <henning.petersen t-online.de>
MFC after: 2 weeks

14 years agoPlug a memory leak.
Xin LI [Mon, 21 Dec 2009 19:15:30 +0000 (19:15 +0000)]
Plug a memory leak.

PR: bin/141836
Submitted by: Henning Petersen <henning.petersen at t-online.de>
MFC after: 2 weeks

14 years agoCast time_t values to intmax_t and use %jd with printf.
Jaakko Heinonen [Mon, 21 Dec 2009 17:41:57 +0000 (17:41 +0000)]
Cast time_t values to intmax_t and use %jd with printf.

OK'ed by: delphij
Approved by: trasz (mentor)

14 years agoRemove non-working special case for pipe(2) from amd64-fbsd32.c and
Jaakko Heinonen [Mon, 21 Dec 2009 17:38:13 +0000 (17:38 +0000)]
Remove non-working special case for pipe(2) from amd64-fbsd32.c and
i386-fbsd.c. Add pipe(2) to syscall table to decode it's pointer
argument properly and re-add special handling for pipe(2) return value
to print_syscall_ret().

PR: bin/120870
Approved by: trasz (mentor)

14 years agoNote that sockstat(1) does not display kernel-owned sockets.
Dag-Erling Smørgrav [Mon, 21 Dec 2009 17:23:04 +0000 (17:23 +0000)]
Note that sockstat(1) does not display kernel-owned sockets.

Submitted by: infofarmer@
MFC after: 2 weeks

14 years agoDon't hide an error if the initial attempt to program a watchdog from
Ruslan Ermilov [Mon, 21 Dec 2009 15:50:37 +0000 (15:50 +0000)]
Don't hide an error if the initial attempt to program a watchdog from
within watchdogd(8) fails.  This is also consistent with watchdog(8).

14 years agoIf none of the watchdogs supports the requested timeout value, the
Ruslan Ermilov [Mon, 21 Dec 2009 15:12:56 +0000 (15:12 +0000)]
If none of the watchdogs supports the requested timeout value, the
code returns EOPNOTSUPP, not EINVAL.

14 years agoMove cursor position after putting a character.
Yoshihiro Takahashi [Mon, 21 Dec 2009 14:59:23 +0000 (14:59 +0000)]
Move cursor position after putting a character.

MFC after: 1 week

14 years agoAdd "FreeBSD-" to the beginning of the filenames for the ISO images.
Ken Smith [Mon, 21 Dec 2009 14:42:35 +0000 (14:42 +0000)]
Add "FreeBSD-" to the beginning of the filenames for the ISO images.
People who collect ISOs from more than just us have been requesting
this for a while.

MFC after: 1 week

14 years ago- Eliminated hard sentence breaks.
Ruslan Ermilov [Mon, 21 Dec 2009 14:39:46 +0000 (14:39 +0000)]
- Eliminated hard sentence breaks.
- Fixed spelling of EINVAL.
- Sorted sections.

14 years agoRemoved one more Alpha leftover.
Ruslan Ermilov [Mon, 21 Dec 2009 12:32:13 +0000 (12:32 +0000)]
Removed one more Alpha leftover.

14 years agoVI_OBJDIRTY vnode flag mirrors the state of OBJ_MIGHTBEDIRTY vm object
Konstantin Belousov [Mon, 21 Dec 2009 12:29:38 +0000 (12:29 +0000)]
VI_OBJDIRTY vnode flag mirrors the state of OBJ_MIGHTBEDIRTY vm object
flag. Besides providing the redundand information, need to update both
vnode and object flags causes more acquisition of vnode interlock.
OBJ_MIGHTBEDIRTY is only checked for vnode-backed vm objects.

Remove VI_OBJDIRTY and make sure that OBJ_MIGHTBEDIRTY is set only for
vnode-backed vm objects.

Suggested and reviewed by: alc
Tested by: pho
MFC after: 3 weeks

14 years agoFix a handful of issues with via agp support.
Robert Noland [Mon, 21 Dec 2009 03:28:05 +0000 (03:28 +0000)]
Fix a handful of issues with via agp support.

  * Read the pci capability register to identify AGP 3 support
  * Add missing smaller aperture sizes for AGP3 chips.
  * Fix the aperture size calculation on AGP2 chips.
    All sizes between 32M and 256M reported as 256M.
  * Add \n to error string.

This all seems to get the CLE266 EPIA-M board agp working properly, now
back to work on drm.

MFC after: 2 weeks

14 years agoFix indentation.
Ed Schouten [Sun, 20 Dec 2009 22:55:27 +0000 (22:55 +0000)]
Fix indentation.

14 years agoDisable jumbo frame support for PCIe VT6130/VT6132 controllers.
Pyun YongHyeon [Sun, 20 Dec 2009 19:45:46 +0000 (19:45 +0000)]
Disable jumbo frame support for PCIe VT6130/VT6132 controllers.
Quite contrary to VT6130 datasheet which says it supports up to 8K
jumbo frame, VT6130 does not seem to send jumbo frame that is
larger than 4K in length. Trying to send a frame that is larger
than 4K cause TX MAC hang.
Even though it's possible to allow 4K jumbo frame for VT6130, I
think it's meaningless to allow 4K jumbo frame. I'm not sure VT6132
also has the same limitation but I guess it uses the same MAC of
VT6130.

14 years agoVT6130 datasheet was wrong. If VT6130 receive a jumbo frame the
Pyun YongHyeon [Sun, 20 Dec 2009 19:11:32 +0000 (19:11 +0000)]
VT6130 datasheet was wrong. If VT6130 receive a jumbo frame the
controller will split the jumbo frame into multiple RX buffers.
However it seems the hardware always dma the frame to 8 bytes
boundary for the split frames. Only the first part of the fragment
can have 4 byte alignment and subsequent buffers should be 8 bytes
aligned. Change RX buffer the alignment requirement to 8 bytes from
4 bytes.

14 years agoCorrect fragment bit definition in comments.
Pyun YongHyeon [Sun, 20 Dec 2009 18:53:34 +0000 (18:53 +0000)]
Correct fragment bit definition in comments.

14 years agoAdd VIA CX700/VX800 chipsets SATA/PATA support.
Alexander Motin [Sun, 20 Dec 2009 16:23:11 +0000 (16:23 +0000)]
Add VIA CX700/VX800 chipsets SATA/PATA support.

PR: kern/121521
Tested by: Alex Deiter

14 years agoFairly set master/slave shared PIO/WDMA timings on ITE 821x controllers.
Alexander Motin [Sun, 20 Dec 2009 15:03:57 +0000 (15:03 +0000)]
Fairly set master/slave shared PIO/WDMA timings on ITE 821x controllers.
Previous implementation could only limit mode, but not rise it back.

14 years agoAvoid sharing the file descriptor of the output file with traced
Jaakko Heinonen [Sun, 20 Dec 2009 11:00:53 +0000 (11:00 +0000)]
Avoid sharing the file descriptor of the output file with traced
processes by setting the FD_CLOEXEC flag for the output file.

PR: bin/140493
Submitted by: Erik Lax
OK'ed by: delphij
Approved by: trasz (mentor)

14 years agoAdd fork(2), getegid(2), geteuid(2), getgid(2), getpid(2), getpgid(2),
Jaakko Heinonen [Sun, 20 Dec 2009 10:58:34 +0000 (10:58 +0000)]
Add fork(2), getegid(2), geteuid(2), getgid(2), getpid(2), getpgid(2),
getpgrp(2), getppid(2), getsid(2) and getuid(2) to syscall table to
decode their arguments correctly.

OK'ed by: delphij
Approved by: trasz (mentor)

14 years agoThe service command is an easy interface to the rc.d system.
Doug Barton [Sun, 20 Dec 2009 01:34:12 +0000 (01:34 +0000)]
The service command is an easy interface to the rc.d system.
Its primary purpose is to start and stop services provided by
the rc.d scripts, however it can also be used to list the scripts
using various criteria.

14 years agoSwap VGE_TXQTIMER and VGE_RXQTIMER register definition. Pending
Pyun YongHyeon [Sat, 19 Dec 2009 20:45:23 +0000 (20:45 +0000)]
Swap VGE_TXQTIMER and VGE_RXQTIMER register definition. Pending
timer for Tx queue is at 0x3E.

14 years agoRemove a warning in DELAY about large delays. In kern_shutdown.c
Marcel Moolenaar [Sat, 19 Dec 2009 20:42:56 +0000 (20:42 +0000)]
Remove a warning in DELAY about large delays. In kern_shutdown.c
we use excessive delays quite habitually.

14 years agoLet access overriding to TTYs depend on the cdev_priv, not the vnode.
Ed Schouten [Sat, 19 Dec 2009 18:42:12 +0000 (18:42 +0000)]
Let access overriding to TTYs depend on the cdev_priv, not the vnode.

Basically this commit changes two things, which improves access to TTYs
in exceptional conditions. Basically the problem was that when you ran
jexec(8) to attach to a jail, you couldn't use /dev/tty (well, also the
node of the actual TTY, e.g. /dev/pts/X). This is very inconvenient if
you want to attach to screens quickly, use ssh(1), etc.

The fixes:

- Cache the cdev_priv of the controlling TTY in struct session. Change
  devfs_access() to compare against the cdev_priv instead of the vnode.
  This allows you to bypass UNIX permissions, even across different
  mounts of devfs.

- Extend devfs_prison_check() to unconditionally expose the device node
  of the controlling TTY, even if normal prison nesting rules normally
  don't allow this. This actually allows you to interact with this
  device node.

To be honest, I'm not really happy with this solution. We now have to
store three pointers to a controlling TTY (s_ttyp, s_ttyvp, s_ttydp).
In an ideal world, we should just get rid of the latter two and only use
s_ttyp, but this makes certian pieces of code very impractical (e.g.
devfs, kern_exit.c).

Reported by: Many people

14 years agoReference the correct man page for firmware(9).
Benedict Reuschling [Sat, 19 Dec 2009 16:36:57 +0000 (16:36 +0000)]
Reference the correct man page for firmware(9).

PR:             docs/140986
Submitted by:   Glen Barber (glen dot j dot barber at gmail dot com)
MFC after:      4 days
Reviewed by:    sam
Approved by:    sam, jkois (mentor)

14 years agoApply fix for Solaris bug 6462803: zfs snapshot -r failed because
Xin LI [Sat, 19 Dec 2009 11:49:20 +0000 (11:49 +0000)]
Apply fix for Solaris bug 6462803: zfs snapshot -r failed because
filesystem was busy
(onnv revision 8989)

Submitted by: mm
Approved by: pjd
Obtained from: OpenSolaris
MFC after: 2 weeks

14 years agoApply fix for Solaris bug 6801979: zfs recv can fail with E2BIG
Xin LI [Sat, 19 Dec 2009 11:47:22 +0000 (11:47 +0000)]
Apply fix for Solaris bug 6801979: zfs recv can fail with E2BIG
(onnv revision 8986)

Requested by: mm
Submitted by: pjd
Obtained from: OpenSolaris
MFC after: 2 weeks

14 years agoApply fix Solaris bug 6462803 zfs snapshot -r failed because
Xin LI [Sat, 19 Dec 2009 11:43:39 +0000 (11:43 +0000)]
Apply fix Solaris bug 6462803 zfs snapshot -r failed because
filesystem was busy.

Submitted by: mm
Approved by: pjd
MFC after: 2 weeks

14 years agoInterpret VAPPEND correctly in vaccess_acl_nfs4(9).
Edward Tomasz Napierala [Sat, 19 Dec 2009 11:41:52 +0000 (11:41 +0000)]
Interpret VAPPEND correctly in vaccess_acl_nfs4(9).

14 years agoFix a problem with how mergemaster handles the hard links for /.cshrc
Doug Barton [Sat, 19 Dec 2009 05:20:26 +0000 (05:20 +0000)]
Fix a problem with how mergemaster handles the hard links for /.cshrc
and /.profile. The problem is that install(1) will unlink the old file
before it installs the new one, which means that in the best case we
have to compare the changes for the old file twice.

So, change the logic to first test to see if the link exists, then
install the file. Then if the link was there and we're using -i, just
create the link in /root and be done with it. Otherwise display the
message to the user and give them the option.

Because we are now sorting things before doing the comparison we can
know conclusively that the files in / should be the sources, and the
files in /root will be the targets, so adjust the paths accordingly.

While I'm here, split a too-long error message into two lines and
just return at the end of handling these files instead of setting
the variable that says "do nothing" and then returning at the end
of the function anyway.

14 years agoFix an indentation issue, no functional changes
Doug Barton [Sat, 19 Dec 2009 01:32:34 +0000 (01:32 +0000)]
Fix an indentation issue, no functional changes

14 years agoIn the places where find is used that the user may see the results,
Doug Barton [Sat, 19 Dec 2009 01:30:45 +0000 (01:30 +0000)]
In the places where find is used that the user may see the results,
first pipe it to sort so that order of processing will be deterministic
and like things will be grouped together.

14 years agoAdd rudimentary WOL support. While I'm here remove enabling
Pyun YongHyeon [Fri, 18 Dec 2009 22:14:28 +0000 (22:14 +0000)]
Add rudimentary WOL support. While I'm here remove enabling
busmastering/memory address in resume path. Bus driver will handle
that.

14 years agoMake sure to enable Next Page bit for IP1001. Otherwise the PHY
Pyun YongHyeon [Fri, 18 Dec 2009 21:57:51 +0000 (21:57 +0000)]
Make sure to enable Next Page bit for IP1001. Otherwise the PHY
fails to re-establishe 1000baseT link after downgrading to
10/100Mbps link.

14 years agoFix ub_env_enum(): syscall() returns 0 when properly invoked.
Marcel Moolenaar [Fri, 18 Dec 2009 21:12:37 +0000 (21:12 +0000)]
Fix ub_env_enum(): syscall() returns 0 when properly invoked.

14 years agoMake the wchan names of pts(4) fit in top(1).
Ed Schouten [Fri, 18 Dec 2009 20:11:29 +0000 (20:11 +0000)]
Make the wchan names of pts(4) fit in top(1).

Just like a similar change we made to the TTY code about half a year
ago, make these strings look similar.

Suggested by: Jille Timmermans <jille@quis.cx>

14 years agoDocument more VIA Velocity family controllers to vge(4). Previously
Pyun YongHyeon [Fri, 18 Dec 2009 18:45:36 +0000 (18:45 +0000)]
Document more VIA Velocity family controllers to vge(4). Previously
it mentioned only VT6122. While I'm here remove the mention of
VT3119 which seems to be VIA's internal model name and VT3119
wouldn't be available to end users.

Reviewed by: brueffer

14 years agoAdd new "hw.psm.tap_enabled" tunable and sysctl.
Jean-Sébastien Pédron [Fri, 18 Dec 2009 17:46:57 +0000 (17:46 +0000)]
Add new "hw.psm.tap_enabled" tunable and sysctl.

This tunable allows one to enable (1) or disable (0) gestures like tap
and tap-hold on Synaptics TouchPad when the Extended mode isn't enabled
(ie. "hw.psm.synaptics_support" not set).

By default, the value is -1 in order to keep the current behaviour of
not enabling/disabling gestures explicitly.

PR: kern/139272
Submitted by: David Horn <dhorn2000 AT gmail DOT com>
Reviewed by: David Horn <dhorn2000 AT gmail DOT com>

14 years agoAdded proper attribution.
Ruslan Ermilov [Fri, 18 Dec 2009 17:22:21 +0000 (17:22 +0000)]
Added proper attribution.

Requested by: luigi

14 years agoMention the unification of rc.firewall and rc.firewall6.
Hajimu UMEMOTO [Fri, 18 Dec 2009 16:35:28 +0000 (16:35 +0000)]
Mention the unification of rc.firewall and rc.firewall6.

Suggested by: David Horn <dhorn2000__at__gmail.com>

14 years ago- Add missing newlines to some error messages.
John Baldwin [Fri, 18 Dec 2009 16:14:31 +0000 (16:14 +0000)]
- Add missing newlines to some error messages.
- Add interrupt descriptions when using mulitple MSI-X interrupts.

Reviewed by: jfv

14 years ago- Create a separate section in in the MI NOTES file for PCI wireless NIC
John Baldwin [Fri, 18 Dec 2009 16:13:21 +0000 (16:13 +0000)]
- Create a separate section in in the MI NOTES file for PCI wireless NIC
  drivers and move bwi(4) there from the PCI Ethernet NIC section.
- Move ath(4) and ral(4) to the MI NOTES file.

Reviewed by: rpaulo

14 years agoRecognize Intel CPUs with Family 0x6, Models 0x1E and 0x1F.
Joseph Koshy [Fri, 18 Dec 2009 15:01:46 +0000 (15:01 +0000)]
Recognize Intel CPUs with Family 0x6, Models 0x1E and 0x1F.

Submitted by: Marc Unangst <mju at panasas dot com>

14 years agoRemove duplicate devstat_start_transaction_bio() call. It is already called
Alexander Motin [Fri, 18 Dec 2009 14:41:30 +0000 (14:41 +0000)]
Remove duplicate devstat_start_transaction_bio() call. It is already called
from geom_disk. Dulicate call causes wrong queue depth and busy accounting.

14 years agoSignal 0 is used to check the permission for current process to signal
Konstantin Belousov [Fri, 18 Dec 2009 14:27:18 +0000 (14:27 +0000)]
Signal 0 is used to check the permission for current process to signal
target one. Since r184058, linux_do_tkill() calls tdsignal() instead of
kill(), without checking for validity of supplied signal number. Prevent
panic when supplied signal is 0 by finishing work after checks.

Found and tested by: scf
MFC after: 3 days

14 years ago- Fixed incorrect watchdog timeout setting: MSB of a 2-byte
Ruslan Ermilov [Fri, 18 Dec 2009 12:10:42 +0000 (12:10 +0000)]
- Fixed incorrect watchdog timeout setting: MSB of a 2-byte
  value is obtained by dividing it by 256, not by 2550; also,
  one second is 10^9 nanoseconds, not 1800000000 nanoseconds.

- Due to rounding error, setting watchdog to a really small
  timeout (<1 sec) was turning the watchdog off.  It should
  set the watchdog to a small timeout instead.

- Implemented error checking in ipmi_wd_event(), as required
  by watchdog(9).

PR: kern/130512
Submitted by: Dmitrij Tejblum

- Additionally, check that the timeout value is within the
  supported range, and if it's too large, act as required by
  watchdog(9).

MFC after: 3 days