]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
17 years agoNuke duplicated __FBSDID.
yongari [Tue, 12 Jun 2007 04:33:21 +0000 (04:33 +0000)]
Nuke duplicated __FBSDID.

17 years agoAdd checks for contigmalloc(9) failure.
yongari [Tue, 12 Jun 2007 04:30:30 +0000 (04:30 +0000)]
Add checks for contigmalloc(9) failure.

17 years agoIncrease a maximum segment size of DMA to 4096. Previously it used
yongari [Tue, 12 Jun 2007 02:35:01 +0000 (02:35 +0000)]
Increase a maximum segment size of DMA to 4096. Previously it used
MCLBYTES for the segment size but it used too many Tx descriptors in
TSO case.
While I'm here adjust maximum size of the sum of all segment lengths
in a given DMA mapping to 65535, the maximum size, in bytes, of a IP
packet.

17 years agoAdd nfe(4) to the list of drivers supported by GENERIC kernel.
yongari [Tue, 12 Jun 2007 02:24:30 +0000 (02:24 +0000)]
Add nfe(4) to the list of drivers supported by GENERIC kernel.
While I'm here comment out nve(4) as nfe(4) will take over.

Approved by: re

17 years agoAllow nfe(4) override nve(4).
yongari [Tue, 12 Jun 2007 02:21:02 +0000 (02:21 +0000)]
Allow nfe(4) override nve(4).

17 years agoXref altq(4).
yongari [Tue, 12 Jun 2007 02:19:56 +0000 (02:19 +0000)]
Xref altq(4).

17 years agonfe(4) supports altq(4).
yongari [Tue, 12 Jun 2007 02:19:11 +0000 (02:19 +0000)]
nfe(4) supports altq(4).

17 years agoAdd nfe(4) to the list of device drivers use the miibus interface.
yongari [Tue, 12 Jun 2007 02:18:04 +0000 (02:18 +0000)]
Add nfe(4) to the list of device drivers use the miibus interface.

17 years agoBring overhauled nfe(4) into tree.
yongari [Tue, 12 Jun 2007 02:16:02 +0000 (02:16 +0000)]
Bring overhauled nfe(4) into tree.

o s/printf/device_printf/g
o Nuke OpenBSDism.
o Nuke NetBSD/OpenBSD specific DMA sync operations.(we don't have a way
   to sync a single descriptor within a DMA map.)
o Remove recursive mutex.
o bus_dma(9) clean up.
o 40bit DMA address support.
o Add protection for Rx map load failure.
o Fix a long standing bug for watchdog timeout. [1]
o Add additional protections, missing Tx completion interrupt, losing
   start Tx command, for watchdog timeout.
o Switch to taskqueue(9) API to handle interrupts.
o Use our own timer for watchdog instead of if_watchdog/if_timer
   interface.
o Advertise VLAN header length/capability correctly to upper layer.
o Remove excessive kernel stack consumption in nfe_encap().
o Handle highly fragmented mbuf chains correctly.
o Enable etherenet address reprogramming with ifconfig(8).
o Add ALTQ/TSO, MSI/MSIX support.
o Increased Rx ring to 256 descriptors from 128.
o Align Tx/Rx descriptor ring on sizeof(struct nfe_desc64) boundary.
o Remove alignment restrictions on Tx/Rx buffers.
o Rewritten jumbo frame support code.
o Add support for hardware assistend VLAN tag insertion/stripping.
o Add support for Tx/Rx flow control based on patches from Peer Chen. [2]
o Add a routine that detects whether ethernet address swap routines is
   required. [3]
o Add a workaround that take MAC/PHY out of power down mode.
o Add suspend/resume support.
o style(9) and code clean up.

Special thanks to Shigeaki Tagashira, the original porter of nfe(4),
who submitted lots of patches, performed uncountable number of
regression tests and maintained nfe(4) for a long time. Without his
enthusiastic help and support I could never have completed this
overhauling task.

The only weak point of nfe(4) compared to nve(4) is instability of
manual half-duplex media selection on certain hardwares(auto sensing
media type should work for all cases, though). This was a long
standing bug of nfe(4) and I still have no idea why it doesn't work
on some hardwares.

Obtained from: OpenBSD [1]
Submitted by: Peer Chen < pchen at nvidia dot com > [2], [3]
Reviewed by: Shigeaki Tagashira < shigeaki AT se DOT hiroshima-u DOT ac DOT jp >
Tested by: Shigeaki Tagashira, current
Discussed with: current
Silence from: obrien

17 years agoReset the pointer to the ioctl buffer after it gets changed.
thompsa [Tue, 12 Jun 2007 00:52:40 +0000 (00:52 +0000)]
Reset the pointer to the ioctl buffer after it gets changed.

17 years agoEliminate now-unused SUSER_ALLOWJAIL arguments to priv_check_cred(); in
rwatson [Tue, 12 Jun 2007 00:12:01 +0000 (00:12 +0000)]
Eliminate now-unused SUSER_ALLOWJAIL arguments to priv_check_cred(); in
some cases, move to priv_check() if it was an operation on a thread and
no other flags were present.

Eliminate caller-side jail exception checking (also now-unused); jail
privilege exception code now goes solely in kern_jail.c.

We can't yet eliminate suser() due to some cases in the KAME code where
a privilege check is performed and then used in many different deferred
paths.  Do, however, move those prototypes to priv.h.

Reviewed by: csjp
Obtained from: TrustedBSD Project

17 years ago - Add a missing PROC_SUNLOCK() in tdsignal()
jeff [Mon, 11 Jun 2007 23:27:03 +0000 (23:27 +0000)]
 - Add a missing PROC_SUNLOCK() in tdsignal()

17 years agoClean up, and sometimes remove, a number of audit-related implementation
rwatson [Mon, 11 Jun 2007 22:10:54 +0000 (22:10 +0000)]
Clean up, and sometimes remove, a number of audit-related implementation
comments.

Obtained from: TrutstedBSD Project

17 years ago - Move p_ru to the zero'd section of the proc to keep stats accurate.
jeff [Mon, 11 Jun 2007 21:59:39 +0000 (21:59 +0000)]
 - Move p_ru to the zero'd section of the proc to keep stats accurate.

17 years agoAdd CPU_XSCALE_81342 before I forget again.
cognet [Mon, 11 Jun 2007 21:31:13 +0000 (21:31 +0000)]
Add CPU_XSCALE_81342 before I forget again.

17 years agoIntroduce pmap_kenter_supersection(), which maps 16MB super-sections into
cognet [Mon, 11 Jun 2007 21:29:26 +0000 (21:29 +0000)]
Introduce pmap_kenter_supersection(), which maps 16MB super-sections into
the kernel pmap.
Document a bit more the behavior of the xscale core 3.

17 years agoRe-acquire the PROC_SLOCK before calling calcru(), and release it after,
cognet [Mon, 11 Jun 2007 21:05:41 +0000 (21:05 +0000)]
Re-acquire the PROC_SLOCK before calling calcru(), and release it after,
since calcru() expects it to be locked.

Reviewed by: attilio

17 years ago- Validate incoming addresses and sizes for connectx and bindx.
rrs [Mon, 11 Jun 2007 21:05:29 +0000 (21:05 +0000)]
- Validate incoming addresses and sizes for connectx and bindx.
- For non-sys call version pass the msg_flags.

17 years agoBump __FreeBSD_version for TCP LRO support.
andre [Mon, 11 Jun 2007 20:19:11 +0000 (20:19 +0000)]
Bump __FreeBSD_version for TCP LRO support.

17 years agoAdd reporting and toggling of TCP LRO (large receive offload) support to
andre [Mon, 11 Jun 2007 20:15:20 +0000 (20:15 +0000)]
Add reporting and toggling of TCP LRO (large receive offload) support to
ifconfig(8).

17 years agoCorrect corrupt read when the read starts at a non-aligned offset.
remko [Mon, 11 Jun 2007 20:14:44 +0000 (20:14 +0000)]
Correct corrupt read when the read starts at a non-aligned offset.

PR: kern/77234
MFC After: 1 week
Approved by: imp (mentor)
Requested by: many many people
Submitted by: Andriy Gapon <avg at icyb dot net dot ua>

17 years agoAdd IFCAP_LRO flag for drivers to announce their TCP Large Receive Offload
andre [Mon, 11 Jun 2007 20:08:11 +0000 (20:08 +0000)]
Add IFCAP_LRO flag for drivers to announce their TCP Large Receive Offload
capabilities.

17 years agoExclude inet_addr.c from the build.
cognet [Mon, 11 Jun 2007 19:38:38 +0000 (19:38 +0000)]
Exclude inet_addr.c from the build.
It only provides inet_aton(), which is already provided by the libc. This
causes multiple symbol definitions when linking statically.

Reviewed by:    darrenr

17 years agoFix a spacing nit.
imp [Mon, 11 Jun 2007 19:36:37 +0000 (19:36 +0000)]
Fix a spacing nit.

17 years agoPrefer __packed to __attribute__((__packed__)).
imp [Mon, 11 Jun 2007 19:34:53 +0000 (19:34 +0000)]
Prefer __packed to __attribute__((__packed__)).

OK'd by sam@ months ago...

17 years agoExclude wlan_scan_* from PAE like the rest of wlan.
thompsa [Mon, 11 Jun 2007 19:29:42 +0000 (19:29 +0000)]
Exclude wlan_scan_* from PAE like the rest of wlan.

17 years agoRemove some ioctls that were ill-thought out. There is no user
mjacob [Mon, 11 Jun 2007 19:15:49 +0000 (19:15 +0000)]
Remove some ioctls that were ill-thought out. There is no user
impact as no softwware using these ioctls was ever committed.

Redo locking for ispioctl.

17 years agoMove the oversize ethernet frame size check into DIAGNOSTIC,
gallatin [Mon, 11 Jun 2007 18:45:49 +0000 (18:45 +0000)]
Move the oversize ethernet frame size check into DIAGNOSTIC,
as was proposed when it was originally added.  This allows
LRO to work on non-DIAGNOSTIC kernels without consuming
any mbuf flags.

Discussed with: sam

17 years agoCreate group ftp by default. This is gid 14 as this is the historical
ceri [Mon, 11 Jun 2007 18:36:39 +0000 (18:36 +0000)]
Create group ftp by default.  This is gid 14 as this is the historical
id used by sysinstall when enabling anonymous FTP.

Change the default group used by sysinstall for setting up anonymous FTP
from operator to ftp; there is no reason to use operator and there are
potential security issues when doing so.

PR: 93284
Approved by: ru (mentor)
Reviewed by: simon

17 years agoBack out the previous commit which added an M_LRO mbuf flag
gallatin [Mon, 11 Jun 2007 18:26:25 +0000 (18:26 +0000)]
Back out the previous commit which added an M_LRO mbuf flag
to defeat the mtu check in ether_input.  Mbuf flags are too scarce.
Discussed with: sam

17 years agoCheck against maxsegsz being zero in bus_dma_tag_create and return EINVAL
mjacob [Mon, 11 Jun 2007 17:57:24 +0000 (17:57 +0000)]
Check against maxsegsz being zero in bus_dma_tag_create and return EINVAL
if it is.

Reviewed by: scott long

17 years agoOnly try and set a segment lim size to 1 << 32 iff bus_size_t > 4.
mjacob [Mon, 11 Jun 2007 17:56:17 +0000 (17:56 +0000)]
Only try and set a segment lim size to 1 << 32 iff bus_size_t > 4.

17 years agoSilence a gcc warning in a more canonical way (evl = NULL rather than &evl).
imp [Mon, 11 Jun 2007 15:29:02 +0000 (15:29 +0000)]
Silence a gcc warning in a more canonical way (evl = NULL rather than &evl).
I saw warnings here at one point on the arm build.

17 years agoAllow drivers, such as cxgb and mxge, which support LRO to bypass
gallatin [Mon, 11 Jun 2007 14:59:56 +0000 (14:59 +0000)]
Allow drivers, such as cxgb and mxge, which support LRO to bypass
the MTU check in ether_input() on LRO merged frames.

Discussed with: kmacy

17 years agoSmall LRO related fixes for mxge:
gallatin [Mon, 11 Jun 2007 14:01:10 +0000 (14:01 +0000)]
Small LRO related fixes for mxge:
- Allow LRO to be enabled / disabled at runtime
- Fix a double-free at module unload time.
- Only update timestamp in lro merge when it is present in the frame
Sponsored by: Myricom

17 years agoWhen we return from a "show" function without printing anything except
bde [Mon, 11 Jun 2007 13:02:15 +0000 (13:02 +0000)]
When we return from a "show" function without printing anything except
a warning, return 1 instead of 0 to indicate that we didn't print
anything, so that top-level callers don't print a spurious newline.
This is mainly to fix output formatting when stderr is redirected.  It
also helps in some cases when stderr is interleaved with stdout,
depending on the details of the interleaving (this program has the
usual null explicit support for syncing stderr with stdout).

Return 1 instead of -1 after printing the "malloc failed" warning, since
the return value is boolean.

17 years agoAdd missing \n to printf
phk [Mon, 11 Jun 2007 12:19:34 +0000 (12:19 +0000)]
Add missing \n to printf

17 years agoMFp4 (missed in net80211 megaupdate)
thompsa [Mon, 11 Jun 2007 10:56:06 +0000 (10:56 +0000)]
MFp4 (missed in net80211 megaupdate)
 - Use a seperate taskqueue+thread for reset tasks since iwi_ops will
   block.
 - Return from iwi_ops if the interface has been downed
 - The firmware will fail if we are already associated
 - Add myself to the copyright

17 years agoAdd wlan_scan_ap and wlan_scan_sta to platforms that include wlan.
thompsa [Mon, 11 Jun 2007 08:26:40 +0000 (08:26 +0000)]
Add wlan_scan_ap and wlan_scan_sta to platforms that include wlan.

17 years agoFix what seems to be a copy-paste buglet (`moused_type' is used
keramida [Mon, 11 Jun 2007 08:19:04 +0000 (08:19 +0000)]
Fix what seems to be a copy-paste buglet (`moused_type' is used
in the description of `moused_flags', instead of the later), and
add a description of `moused_XXX_flags' where `XXX' is the port
name of a non-default moused invocation -- including an example
of using "-3" with the default moused(8) instance, but no special
flags for moused(8) invocations handling 3-button USB mice (which
seems a very common scenario these days).

MFC after: 3 days

17 years agoFix an aliasing bug which was finally detected by gcc-4.2. fdlibm has
bde [Mon, 11 Jun 2007 07:48:52 +0000 (07:48 +0000)]
Fix an aliasing bug which was finally detected by gcc-4.2.  fdlibm has
hundreds of similar aliasing bugs, but all except this one seem to have
been fixed by Cygnus and/or NetBSD before the modified version of fdlibm
was imported into FreeBSD in 1994.

PR: standards/113147
Submitted by: Steve Kargl <sgk@troutmask.apl.washington.edu>

17 years agoAdd new files in the net80211 changes.
thompsa [Mon, 11 Jun 2007 07:26:35 +0000 (07:26 +0000)]
Add new files in the net80211 changes.

17 years agoAdd inet_ntoa_r, a reentrant version of inet_ntoa. This is
delphij [Mon, 11 Jun 2007 07:21:21 +0000 (07:21 +0000)]
Add inet_ntoa_r, a reentrant version of inet_ntoa.  This is
available on a lot of platforms, as well as libkern for years.

Submitted by: "MQ"

17 years agoUse tabs after #define.
stefanf [Mon, 11 Jun 2007 06:25:19 +0000 (06:25 +0000)]
Use tabs after #define.

Obtained from: NetBSD
Spotted by: njl

17 years agoExpand DECLARE_USB_DMA_T inline in the one place it is used and eliminate
imp [Mon, 11 Jun 2007 06:21:50 +0000 (06:21 +0000)]
Expand DECLARE_USB_DMA_T inline in the one place it is used and eliminate
from usb_port.h.

17 years agoEliminate USB_ATTACH_SETUP and USB_MATCH_SETUP. They are no longer in
imp [Mon, 11 Jun 2007 06:18:34 +0000 (06:18 +0000)]
Eliminate USB_ATTACH_SETUP and USB_MATCH_SETUP.  They are no longer in
the tree.

17 years agoPrefer device_printf over printf.
imp [Mon, 11 Jun 2007 06:14:42 +0000 (06:14 +0000)]
Prefer device_printf over printf.

17 years agoMinor tweak.
imp [Mon, 11 Jun 2007 06:03:32 +0000 (06:03 +0000)]
Minor tweak.

17 years agoPrefer device_printf to printf
imp [Mon, 11 Jun 2007 06:01:05 +0000 (06:01 +0000)]
Prefer device_printf to printf
Eliminate rue_unit from softc

# saves ~200 bytes

17 years agoPrefer device_printf to printf
imp [Mon, 11 Jun 2007 05:50:47 +0000 (05:50 +0000)]
Prefer device_printf to printf
Remove keu_unit from softc

# this change saves about 180 bytes in the module, all in text

17 years agoGet rid of useless devinfo stuff
imp [Mon, 11 Jun 2007 05:44:28 +0000 (05:44 +0000)]
Get rid of useless devinfo stuff
Kill bogus bzero
prefer device_printf to printf

Reviewed by: alfred@

17 years agoPrefer device_printf to printf where sane.
imp [Mon, 11 Jun 2007 05:42:47 +0000 (05:42 +0000)]
Prefer device_printf to printf where sane.
Elimiante cue_unit from softc.

17 years agoDon't lose leading '/' for pathnames exactly 101 bytes long.
kientzle [Mon, 11 Jun 2007 05:17:30 +0000 (05:17 +0000)]
Don't lose leading '/' for pathnames exactly 101 bytes long.
Also, update the test harness to exercise this case.

17 years ago- Don't force to be the GDB port since dcons(4) is in GENERIC now.
simokawa [Mon, 11 Jun 2007 04:08:50 +0000 (04:08 +0000)]
- Don't force to be the GDB port since dcons(4) is in GENERIC now.

To enable the GDB port of dcons(4), you need to put
  dcons_gdb=1
in /boot/loader.conf.

17 years agonote 802.11 changes
sam [Mon, 11 Jun 2007 04:06:50 +0000 (04:06 +0000)]
note 802.11 changes

17 years agoadd 11n stats
sam [Mon, 11 Jun 2007 04:05:49 +0000 (04:05 +0000)]
add 11n stats

17 years agoo add 11n knob
sam [Mon, 11 Jun 2007 04:05:15 +0000 (04:05 +0000)]
o add 11n knob
o gcc42 stuff

17 years agoo add minimal radiotap support for 11n
sam [Mon, 11 Jun 2007 04:04:30 +0000 (04:04 +0000)]
o add minimal radiotap support for 11n

17 years agotrack net80211 changes to get scan results ioctl
sam [Mon, 11 Jun 2007 03:57:46 +0000 (03:57 +0000)]
track net80211 changes to get scan results ioctl

17 years agoUpdate for revised 802.11 support:
sam [Mon, 11 Jun 2007 03:56:33 +0000 (03:56 +0000)]
Update for revised 802.11 support:
o revised channel handling support; ifconfig now queries the kernel to
  find the list of available channels and handles channel promotion;
  channel attributes can be specified as part of the channel; e.g. 36:a
  for channel 36 in 11a (as opposed to turbo A or HT A)
o use channel list to map between freq and IEEE channel #; this eliminates
  all knowledge of how the mapping is done and fixes handling of cases
  where channels overlap in the IEEE channel # space but are distinct in
  the frequency+attributes space (e.g. PSB)
o add new knobs: bgscan, ff (Atheors fast frames), dturbo (Atheros
  Dynamic Turbo mode), bgscanidle, bgscanintvl, scanvalid, roam:rssi11a,
  roam:rssi11b, roam:rssi11g, roam:rate11a, roam:rate11b, roam:rate11g
  (roaming parameters), burst, doth (forthcoming 11h support)
o print contents of WME, ATH, WPA, RSN, information elements with -v option
o print signal strength in dBm
o print noise floor in dBm
o add list txpow to print tx power caps/channel
o change default channel display in status to be more informative

17 years agoUpdate 802.11 wireless support:
sam [Mon, 11 Jun 2007 03:36:55 +0000 (03:36 +0000)]
Update 802.11 wireless support:
o major overhaul of the way channels are handled: channels are now
  fully enumerated and uniquely identify the operating characteristics;
  these changes are visible to user applications which require changes
o make scanning support independent of the state machine to enable
  background scanning and roaming
o move scanning support into loadable modules based on the operating
  mode to enable different policies and reduce the memory footprint
  on systems w/ constrained resources
o add background scanning in station mode (no support for adhoc/ibss
  mode yet)
o significantly speedup sta mode scanning with a variety of techniques
o add roaming support when background scanning is supported; for now
  we use a simple algorithm to trigger a roam: we threshold the rssi
  and tx rate, if either drops too low we try to roam to a new ap
o add tx fragmentation support
o add first cut at 802.11n support: this code works with forthcoming
  drivers but is incomplete; it's included now to establish a baseline
  for other drivers to be developed and for user applications
o adjust max_linkhdr et. al. to reflect 802.11 requirements; this eliminates
  prepending mbufs for traffic generated locally
o add support for Atheros protocol extensions; mainly the fast frames
  encapsulation (note this can be used with any card that can tx+rx
  large frames correctly)
o add sta support for ap's that beacon both WPA1+2 support
o change all data types from bsd-style to posix-style
o propagate noise floor data from drivers to net80211 and on to user apps
o correct various issues in the sta mode state machine related to handling
  authentication and association failures
o enable the addition of sta mode power save support for drivers that need
  net80211 support (not in this commit)
o remove old WI compatibility ioctls (wicontrol is officially dead)
o change the data structures returned for get sta info and get scan
  results so future additions will not break user apps
o fixed tx rate is now maintained internally as an ieee rate and not an
  index into the rate set; this needs to be extended to deal with
  multi-mode operation
o add extended channel specifications to radiotap to enable 11n sniffing

Drivers:
o ath: add support for bg scanning, tx fragmentation, fast frames,
       dynamic turbo (lightly tested), 11n (sniffing only and needs
       new hal)
o awi: compile tested only
o ndis: lightly tested
o ipw: lightly tested
o iwi: add support for bg scanning (well tested but may have some
       rough edges)
o ral, ural, rum: add suppoort for bg scanning, calibrate rssi data
o wi: lightly tested

This work is based on contributions by Atheros, kmacy, sephe, thompsa,
mlaier, kevlo, and others.  Much of the scanning work was supported by
Atheros.  The 11n work was supported by Marvell.

17 years agoConditionally acquire Giant in vm_contig_launder_page().
alc [Mon, 11 Jun 2007 03:20:16 +0000 (03:20 +0000)]
Conditionally acquire Giant in vm_contig_launder_page().

17 years agoDiff reduction against other *BSDs: ANSIfy function
delphij [Mon, 11 Jun 2007 03:05:54 +0000 (03:05 +0000)]
Diff reduction against other *BSDs: ANSIfy function
prototypes.  No function changes.

17 years agoAdd _PATH_FWMEM for libkvm.
simokawa [Mon, 11 Jun 2007 02:21:18 +0000 (02:21 +0000)]
Add _PATH_FWMEM for libkvm.

17 years agoConnect icsphy(4) to the build.
yongari [Mon, 11 Jun 2007 02:07:08 +0000 (02:07 +0000)]
Connect icsphy(4) to the build.

17 years agoAdd icsphy(4), Integrated Circuit Systems PHY driver, ported from
yongari [Mon, 11 Jun 2007 02:04:50 +0000 (02:04 +0000)]
Add icsphy(4), Integrated Circuit Systems PHY driver, ported from
NetBSD. ATM the only consumer of the PHY is XBox with nfe(4) driver.

Submitted by: Ed Schouten <ed@fxq.nl>
Tested by: Ed Schouten <ed@fxq.nl>

17 years agoAdd ICS1889/ICS1892/ICS1893 PHY.
yongari [Mon, 11 Jun 2007 02:02:20 +0000 (02:02 +0000)]
Add ICS1889/ICS1892/ICS1893 PHY.

Submitted by: Ed Schouten <ed@fxq.nl>

17 years agoIncrease a maximum segment size of DMA to 4096. Previously it used
yongari [Mon, 11 Jun 2007 02:00:50 +0000 (02:00 +0000)]
Increase a maximum segment size of DMA to 4096. Previously it used
MCLBYTES for the segment size but it used too many Tx descriptors in
TSO case.
While I'm here adjust maximum size of the sum of all segment lengths
in a given DMA mapping to 65535, the maximum size, in bytes, of a IP
packet.

17 years agoDisable automatic IP ID increment. Due to a hardware bug the automatic
yongari [Mon, 11 Jun 2007 01:55:09 +0000 (01:55 +0000)]
Disable automatic IP ID increment. Due to a hardware bug the automatic
IP ID increment in TSO case generated corrupted IP packets.
This change brings back TSO capability.

17 years agoFilter/compress the amount of channel trigger. This should reduce
ariff [Mon, 11 Jun 2007 00:49:46 +0000 (00:49 +0000)]
Filter/compress the amount of channel trigger. This should reduce
much of lock/unlock contentions within the interrupt handler. Most
of these drivers only need PCMTRIG_START or STOP (ABORT).

Discussed with: scottl

17 years agoUse default options for default partitioning schemes, rather than
marcel [Mon, 11 Jun 2007 00:38:06 +0000 (00:38 +0000)]
Use default options for default partitioning schemes, rather than
making the relevant files standard. This avoids duplication and
makes it easier to override/disable unwanted schemes. Since ARM
doesn't have a DEFAULTS configuration file, leave the source
files for the BSD and MBR partitioning schemes in files.arm for
now.

17 years agoAdd the machine-specific definitions for configuring the new physical
alc [Sun, 10 Jun 2007 23:39:07 +0000 (23:39 +0000)]
Add the machine-specific definitions for configuring the new physical
memory allocator.

Set the size of phys_avail[] using one of these definitions.

Approved by: re

17 years ago- Add codec id for ALC660 [1]
ariff [Sun, 10 Jun 2007 23:01:40 +0000 (23:01 +0000)]
- Add codec id for ALC660 [1]
- Add codec id for AD1988B, along with fixing its line-in and other
  issues (with proper quirks). [2]

Submitted by: [1] barbara.xxx1975@libero.it
              [2] Oliver Brandmueller ob@e-Gitt.NET
MFC after: 3 days

17 years agoOptimize vmmeter locking.
attilio [Sun, 10 Jun 2007 21:59:14 +0000 (21:59 +0000)]
Optimize vmmeter locking.
In particular:
- Add an explicative table for locking of struct vmmeter members
- Apply new rules for some of those members
- Remove some unuseful comments

Heavily reviewed by: alc, bde, jeff
Approved by: jeff (mentor)

17 years agoFix a case in tcp_do_segment() where tcp_update_sack_list() would
andre [Sun, 10 Jun 2007 21:07:21 +0000 (21:07 +0000)]
Fix a case in tcp_do_segment() where tcp_update_sack_list() would
be called with an incorrect segment end value.  tcp_reass() may
trim segments when they overlap with already existing ones in the
reassembly queue.  Instead of saving the segment end value before
the call to tcp_reass() compute it on the fly based on the effective
segment length afterwards.

This bug was not really problematic as no information got lost and
the eventual SACK information computation was correct nontheless.

MFC after: 1 week

17 years agoFix style for comments, be more verbose and add some more.
andre [Sun, 10 Jun 2007 20:59:22 +0000 (20:59 +0000)]
Fix style for comments, be more verbose and add some more.

17 years agoSome improvements to the int-type printing code based on suggestions by bde.
dwmalone [Sun, 10 Jun 2007 20:11:52 +0000 (20:11 +0000)]
Some improvements to the int-type printing code based on suggestions by bde.

17 years agoFix a number of WARNS, including printf, constness and unsigned comparison
dwmalone [Sun, 10 Jun 2007 19:32:20 +0000 (19:32 +0000)]
Fix a number of WARNS, including printf, constness and unsigned comparison
warnings.

17 years agoSome style improvements suggested by bde, including removing an
dwmalone [Sun, 10 Jun 2007 19:13:40 +0000 (19:13 +0000)]
Some style improvements suggested by bde, including removing an
unused include, adding parens for return and sizeof and renaming,
adding some missing whitespace and sorting some variables.

17 years agoMerge NetBSD changes, among them:
stefanf [Sun, 10 Jun 2007 19:06:09 +0000 (19:06 +0000)]
Merge NetBSD changes, among them:

el.c 1.44, el.h 1.17, editline.3 1.53, histedit.h 1.31:
# add EL_GETFP, and EL_SETFP.

el.c 1.42, term.c 1.46, term.h 1.18, editline.3 1.52, histedit.h 1.29:
# - Add more readline functions, enough for gdb-6.5
# - Make el_get varyadic, and implement EL_GETTC.
# - XXX: the EL_SETTC api will change in the future.

Note: The latter change breaks the ABI of the el_get() function.

Approved by: re (kensmith)

17 years agoNow pam_nologin(8) will provide an account management function
yar [Sun, 10 Jun 2007 18:57:20 +0000 (18:57 +0000)]
Now pam_nologin(8) will provide an account management function
instead of an authentication function.  There are a design reason
and a practical reason for that.  First, the module belongs in
account management because it checks availability of the account
and does no authentication.  Second, there are existing and potential
PAM consumers that skip PAM authentication for good or for bad.
E.g., sshd(8) just prefers internal routines for public key auth;
OTOH, cron(8) and atrun(8) do implicit authentication when running
a job on behalf of its owner, so their inability to use PAM auth
is fundamental, but they can benefit from PAM account management.

Document this change in the manpage.

Modify /etc/pam.d files accordingly, so that pam_nologin.so is listed
under the "account" function class.

Bump __FreeBSD_version (mostly for ports, as this change should be
invisible to C code outside pam_nologin.)

PR: bin/112574
Approved by: des, re

17 years agoWork around a firmware bug in the HP rx2660, where in ACPI an I/O port
marcel [Sun, 10 Jun 2007 16:53:01 +0000 (16:53 +0000)]
Work around a firmware bug in the HP rx2660, where in ACPI an I/O port
is really a memory mapped I/O address. The bug is in the GAS that
describes the address and in particular the SpaceId field. The field
should not say the address is an I/O port when it clearly is not.

With an additional check for the IA64_BUS_SPACE_IO case in the bus
access functions, and the fact that I/O ports pretty much not used
in general on ia64, make the calculation of the I/O port address a
function. This avoids inlining the work-around into every driver,
and also helps reduce overall code bloat.

17 years agoFix bogon in previous commit: <machine/cpu.h> is still needed.
marcel [Sun, 10 Jun 2007 16:32:08 +0000 (16:32 +0000)]
Fix bogon in previous commit: <machine/cpu.h> is still needed.

17 years agoCast len to be a uintmax_t and make format in KASSERT match so as
mjacob [Sun, 10 Jun 2007 15:46:34 +0000 (15:46 +0000)]
Cast len to be a uintmax_t and make format in KASSERT match so as
to avoid different sizes on different platforms types of complaints.

Reviewed by: Ariff

17 years agoRemove const type qualifier from a function- gcc4.2 doesn't accept it.
mjacob [Sun, 10 Jun 2007 15:45:29 +0000 (15:45 +0000)]
Remove const type qualifier from a function- gcc4.2 doesn't accept it.

Reviewed by: Ariff

17 years agoInitialize the dma tag's bounce_zone to NULL if we didn't allocate it.
cognet [Sun, 10 Jun 2007 12:33:01 +0000 (12:33 +0000)]
Initialize the dma tag's bounce_zone to NULL if we didn't allocate it.

17 years agoRemove /usr/X11R6 and subdirs.
brueffer [Sun, 10 Jun 2007 08:51:03 +0000 (08:51 +0000)]
Remove /usr/X11R6 and subdirs.

MFC after: 3 days

17 years agoPrefer device_printf over printf
imp [Sun, 10 Jun 2007 07:33:48 +0000 (07:33 +0000)]
Prefer device_printf over printf
Eliminate cdce_unit from softc.

17 years agoPrefer device_printf over printf. Eliminate axe_unit as it is no
imp [Sun, 10 Jun 2007 07:24:32 +0000 (07:24 +0000)]
Prefer device_printf over printf.  Eliminate axe_unit as it is no
longer required.

17 years agoBackout mess mistakenly committed with manpage update.
ceri [Sun, 10 Jun 2007 06:18:04 +0000 (06:18 +0000)]
Backout mess mistakenly committed with manpage update.

17 years agoDocument SCTP support.
ceri [Sun, 10 Jun 2007 06:11:03 +0000 (06:11 +0000)]
Document SCTP support.

17 years agoRemove 'inline' qualifiers from functions which are not, in fact, inlines.
mjacob [Sun, 10 Jun 2007 04:54:42 +0000 (04:54 +0000)]
Remove 'inline' qualifiers from functions which are not, in fact, inlines.

17 years agoCast the ioctl define to the type of the variable being switched on.
mjacob [Sun, 10 Jun 2007 04:53:13 +0000 (04:53 +0000)]
Cast the ioctl define to the type of the variable being switched on.

17 years agoRemove unused variable.
mjacob [Sun, 10 Jun 2007 04:51:03 +0000 (04:51 +0000)]
Remove unused variable.

17 years agoInit timespec to zero fo quiesce warnings.
mjacob [Sun, 10 Jun 2007 04:42:20 +0000 (04:42 +0000)]
Init timespec to zero fo quiesce warnings.

17 years agoQuiesce warnings by initializing irql values to zero.
mjacob [Sun, 10 Jun 2007 04:40:13 +0000 (04:40 +0000)]
Quiesce warnings by initializing irql values to zero.

17 years agoEnsure that newpath is always initialized, even for the error case.
mjacob [Sun, 10 Jun 2007 04:37:22 +0000 (04:37 +0000)]
Ensure that newpath is always initialized, even for the error case.

17 years agoCommit for Nate his "guidelines for submitting quirks".
mjacob [Sun, 10 Jun 2007 04:31:55 +0000 (04:31 +0000)]
Commit for Nate his "guidelines for submitting quirks".

17 years agoThe definition of CACHELINESIZE moved from <machine/cpu.h> to
marcel [Sun, 10 Jun 2007 03:13:56 +0000 (03:13 +0000)]
The definition of CACHELINESIZE moved from <machine/cpu.h> to
<machine/md_var.h>.

17 years agoRemove unused variable.
mjacob [Sun, 10 Jun 2007 01:50:05 +0000 (01:50 +0000)]
Remove unused variable.