]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
15 years agoBuild fixups for the new USB stack.
Andrew Thompson [Mon, 23 Feb 2009 18:36:54 +0000 (18:36 +0000)]
Build fixups for the new USB stack.

15 years agoChange over the usb kernel options to the new stack (retaining existing
Andrew Thompson [Mon, 23 Feb 2009 18:34:56 +0000 (18:34 +0000)]
Change over the usb kernel options to the new stack (retaining existing
naming). The old usb stack can be compiled in my prefixing the name with 'o'.

15 years agoHook up new USB modules.
Andrew Thompson [Mon, 23 Feb 2009 18:32:59 +0000 (18:32 +0000)]
Hook up new USB modules.

15 years agoMove the new USB stack into its new home.
Andrew Thompson [Mon, 23 Feb 2009 18:31:00 +0000 (18:31 +0000)]
Move the new USB stack into its new home.

15 years agoInitial gdbserver support for PowerPC.
Rafal Jaworowski [Mon, 23 Feb 2009 18:22:49 +0000 (18:22 +0000)]
Initial gdbserver support for PowerPC.

Obtained from: Juniper Networks, Semihalf

15 years agoSVN rev 188743 modified aac_rx_get_fwstatus to use the AAC_RX_OMR0
Ed Maste [Mon, 23 Feb 2009 18:22:06 +0000 (18:22 +0000)]
SVN rev 188743 modified aac_rx_get_fwstatus to use the AAC_RX_OMR0
register instead of AAC_RX_FWSTATUS, as that is the way it's done in
Adaptec's vendor driver and in the Linux drivers.  (The same applies
to aac_rkt_get_fwstatus as well.)

However, a concern has been raised about the compatibility of this
change and old hardware / firmware versions.  In the absense of
specific information, revert to the original behaviour if the firmware
does not support the "New comm." interface.  Users of old cards or
firmware haven't reported the problems that are potentially solved by
switching to OMR0.

15 years agoMove usb to a graveyard location under sys/legacy/dev, it is intended that the
Andrew Thompson [Mon, 23 Feb 2009 18:16:17 +0000 (18:16 +0000)]
Move usb to a graveyard location under sys/legacy/dev, it is intended that the
new USB2 stack will fully replace this for 8.0.

Remove kernel modules, a subsequent commit will update conf/files. Unhook
usbdevs from the build.

15 years agoSome whitespace and style fixes.
John Baldwin [Mon, 23 Feb 2009 15:39:24 +0000 (15:39 +0000)]
Some whitespace and style fixes.

Submitted by: bde (partly)

15 years agoFreeBSD/i386 doesn't include a software FPU emulator anymore, so adjust
John Baldwin [Mon, 23 Feb 2009 15:38:35 +0000 (15:38 +0000)]
FreeBSD/i386 doesn't include a software FPU emulator anymore, so adjust
an iBCS2 syscall to indicate that there is no FPU support at all rather
than emulated support if an FPU is not present.

15 years agoDo not call devices probe/attach if there is nothing new was found.
Alexander Motin [Mon, 23 Feb 2009 08:58:29 +0000 (08:58 +0000)]
Do not call devices probe/attach if there is nothing new was found.

15 years agoFix spelling.
Alexander Motin [Mon, 23 Feb 2009 08:19:30 +0000 (08:19 +0000)]
Fix spelling.

Submitted by: gavin

15 years agoAdd explicit casting in few places.
Pawel Jakub Dawidek [Mon, 23 Feb 2009 07:33:29 +0000 (07:33 +0000)]
Add explicit casting in few places.

It is only really necessary for open(2)'s third argument, which is optional and
obtained through stdarg(3). open(2)'s third argument is 32bit and we pass 64
bits. On little endian it works, because we take lower 32 bits, but on big
endian platforms we take upper 32 bits, so we end up with 0.

Reported by: Milan Čermák <Milan.Cermak@Sun.COM>

15 years agoOptimize free_pv_entry(); specifically, avoid repeated TAILQ_REMOVE()s.
Alan Cox [Mon, 23 Feb 2009 06:00:24 +0000 (06:00 +0000)]
Optimize free_pv_entry(); specifically, avoid repeated TAILQ_REMOVE()s.

MFC after: 1 week

15 years agoUse uiomove_fromphys() instead of the combination of sf_buf and uiomove().
Alan Cox [Sun, 22 Feb 2009 19:50:09 +0000 (19:50 +0000)]
Use uiomove_fromphys() instead of the combination of sf_buf and uiomove().

This is not only shorter; it also eliminates unnecessary thread pinning on
architectures that implement a direct map.

MFC after: 3 weeks

15 years agochange tdma slave behaviour: if the channel is locked don't do bmiss handling
Sam Leffler [Sun, 22 Feb 2009 18:48:54 +0000 (18:48 +0000)]
change tdma slave behaviour: if the channel is locked don't do bmiss handling
(so no scanning/roaming)

Reviewed by: Chris Anderson

15 years agodisplay min/max dwell times in msecs in debug msgs
Sam Leffler [Sun, 22 Feb 2009 18:46:36 +0000 (18:46 +0000)]
display min/max dwell times in msecs in debug msgs

15 years agoExplicitly disable generation of Altivec instructions in the kernel on PowerPC,
Nathan Whitehorn [Sun, 22 Feb 2009 18:45:30 +0000 (18:45 +0000)]
Explicitly disable generation of Altivec instructions in the kernel on PowerPC,
and add support to allow users to set their CPUTYPE in make.conf.

15 years agoAdd SATA and USB modes for completeness. USB modes principally can't be set,
Alexander Motin [Sun, 22 Feb 2009 18:22:21 +0000 (18:22 +0000)]
Add SATA and USB modes for completeness. USB modes principally can't be set,
SATA mode setting is not implementes at this time.

Submitted by: Christoph Mallon

15 years agoSimplify the unwiring and activation of pages.
Alan Cox [Sun, 22 Feb 2009 18:15:17 +0000 (18:15 +0000)]
Simplify the unwiring and activation of pages.

MFC after: 1 week

15 years ago- Punctuation fixes.
Pawel Jakub Dawidek [Sun, 22 Feb 2009 15:03:29 +0000 (15:03 +0000)]
- Punctuation fixes.
- New sentence - new line.

Reported by: Ben Kaduk <minimarmot@gmail.com>

- No more than 80 chars per line.

15 years agoTune AHCI verbose messages to make log more readable.
Alexander Motin [Sun, 22 Feb 2009 14:08:20 +0000 (14:08 +0000)]
Tune AHCI verbose messages to make log more readable.

15 years agoMention disk_gone() in disk(9).
Edward Tomasz Napierala [Sun, 22 Feb 2009 13:38:16 +0000 (13:38 +0000)]
Mention disk_gone() in disk(9).

Approved by: rwatson (mentor)

15 years agoFix a typo and add manpage links to geom(4).
Edward Tomasz Napierala [Sun, 22 Feb 2009 13:37:22 +0000 (13:37 +0000)]
Fix a typo and add manpage links to geom(4).

Approved by: rwatson (mentor)

15 years agoFix typo.
Rui Paulo [Sun, 22 Feb 2009 12:40:58 +0000 (12:40 +0000)]
Fix typo.

15 years agoRemove one more place of master/slave terms usage.
Alexander Motin [Sun, 22 Feb 2009 10:26:02 +0000 (10:26 +0000)]
Remove one more place of master/slave terms usage.

15 years agoRemove unnecessary page queues locking around vm_page_wakeup(). (This
Alan Cox [Sun, 22 Feb 2009 02:50:31 +0000 (02:50 +0000)]
Remove unnecessary page queues locking around vm_page_wakeup().  (This
change is applicable to RELENG_7 but not RELENG_6.)

MFC after: 1 week

15 years agoUse root_mount_hold and root_mount_rel to allow the USB bus to be explored
Andrew Thompson [Sun, 22 Feb 2009 01:02:25 +0000 (01:02 +0000)]
Use root_mount_hold and root_mount_rel to allow the USB bus to be explored
before trying to mount root. Unlike USB1 the busses are not explored at attach
but rather after threads are running.

15 years agoUse only higher half of device signature to identify device type. Some
Alexander Motin [Sat, 21 Feb 2009 23:46:34 +0000 (23:46 +0000)]
Use only higher half of device signature to identify device type. Some
devices return incorrect values in lower part confusing detection, while
higher part itself gives enough information for proper detetion.

15 years agoIncrease initial busy waiting time. If drive was hot-plugged, it may need
Alexander Motin [Sat, 21 Feb 2009 23:42:28 +0000 (23:42 +0000)]
Increase initial busy waiting time. If drive was hot-plugged, it may need
much time to spin-up.

15 years ago - Resolve an issue where we may clear an idt while an interrupt on a
Jeff Roberson [Sat, 21 Feb 2009 23:15:34 +0000 (23:15 +0000)]
 - Resolve an issue where we may clear an idt while an interrupt on a
   different cpu is still assigned to that vector by never clearing idt
   entries.  This was only provided as a debugging feature and the bugs
   are caught by other means.
 - Drop the sched lock when rebinding to reassign an interrupt vector
   to a new cpu so that pending interrupts have a chance to be delivered
   before removing the old vector.

Discussed with: tegge, jhb

15 years agoImprove ata_reinit():
Alexander Motin [Sat, 21 Feb 2009 22:57:26 +0000 (22:57 +0000)]
Improve ata_reinit():
 - protect againtst recursions,
 - add new devices detection using ata_identify().

Improve ata_identify():
 - do not add duplicate device if device already exist.

Rework SATA hot-plug events handling. Instead of unsafe duplicate
implementation use common ata_reinit() to handle all state changes.

All together this gives quite stable and robust cold- and hot-plug operation,
invariant to false, lost and duplicate events.

15 years agoReduce the scope of the page queues lock in vm_object_page_remove().
Alan Cox [Sat, 21 Feb 2009 20:57:25 +0000 (20:57 +0000)]
Reduce the scope of the page queues lock in vm_object_page_remove().

MFC after: 1 week

15 years agoo When creating the EBR scheme, set the number of entries
Marcel Moolenaar [Sat, 21 Feb 2009 19:25:13 +0000 (19:25 +0000)]
o  When creating the EBR scheme, set the number of entries
   properly. Otherwise the minimum of 1 is used and you can
   only insert a single partition/slice and only at sector
   0 (index 1).
o  When adding a partition/slice, recalculate the index after
   the start and size of the partition/slice are adjusted to
   make them a multiple of the track size. Since the precheck
   method sets the index based on the start of the partition
   as provided by the user, we know that we're off by at most
   1 and adjusting the index is safe.

15 years agoTeach device drivers' ata_reinit() methods, that there can be more then two
Alexander Motin [Sat, 21 Feb 2009 16:39:26 +0000 (16:39 +0000)]
Teach device drivers' ata_reinit() methods, that there can be more then two
devices per channel.

15 years agoSync with the official Adaptec vendor driver:
Attilio Rao [Sat, 21 Feb 2009 15:40:03 +0000 (15:40 +0000)]
Sync with the official Adaptec vendor driver:

[1] Add the support for the NARK controller which seems a variant of
the i960Rx.
[2] Split up memory regions and other resources in 2 different parts
as long as NARK uses them separately (it is not clear to me
why though as long as there are no more informations available
on this controller). Please note that in all the other cases,
the regions overlaps leaving the default behaviour for all the
other controllers.
[3] Implement a clock daemon responsible for maintain updated the
wall clock time of the controller (run any 30 minutes)*.

Submitted by: Adaptec (driver build 15317 [1, 2] and 15727 [3])
Reviewed by: emaste
Tested by: emaste
Sponsored by: Sandvine Incorporated

* Please note that originally, in the Adaptec driver, the clock daemon
  is not implemented with callouts as in our in-tree driver.

15 years agoFix build when WITH_SSP is set explicitly.
Ruslan Ermilov [Sat, 21 Feb 2009 15:04:31 +0000 (15:04 +0000)]
Fix build when WITH_SSP is set explicitly.

Submitted by: Jeremie Le Hen

15 years agoAdd support for methods to the OSD subsystem. Each object type has a
Jamie Gritton [Sat, 21 Feb 2009 11:15:38 +0000 (11:15 +0000)]
Add support for methods to the OSD subsystem.  Each object type has a
predefined set of methods, which are set in osd_register() and called
via osd_call().  Currently, no methods are defined, though prison
objects will have some in the future.

Expand the locking from a single per-type mutex to three different kinds
of locks (four if you include the requirement that the container
(e.g. prison) be locked when getting/setting data).  This clears up one
existing issue, as well as others added by the method support.

Approved by: bz (mentor)

15 years agoAdd bootcode handling.
Marcel Moolenaar [Sat, 21 Feb 2009 07:01:21 +0000 (07:01 +0000)]
Add bootcode handling.

15 years agoRemove unnecessary opt* definations.
Xin LI [Sat, 21 Feb 2009 03:53:06 +0000 (03:53 +0000)]
Remove unnecessary opt* definations.

15 years agoDefine extern **environ in global scope instead of in function.
Xin LI [Sat, 21 Feb 2009 03:50:25 +0000 (03:50 +0000)]
Define extern **environ in global scope instead of in function.

15 years agoUtilize calloc() instead of bzero'ing manually.
Xin LI [Sat, 21 Feb 2009 03:43:20 +0000 (03:43 +0000)]
Utilize calloc() instead of bzero'ing manually.

15 years agoInitialize 'sz' as 0.
Xin LI [Sat, 21 Feb 2009 03:42:31 +0000 (03:42 +0000)]
Initialize 'sz' as 0.

15 years agoUtilize strlcpy().
Xin LI [Sat, 21 Feb 2009 03:38:46 +0000 (03:38 +0000)]
Utilize strlcpy().

15 years agoRevert r188878 for the moment, this breaks kdump.
Andrew Thompson [Sat, 21 Feb 2009 03:06:29 +0000 (03:06 +0000)]
Revert r188878 for the moment, this breaks kdump.

15 years agoInclude altivec.c, missed on previous commit r188860. This should unbreak the
Nathan Whitehorn [Sat, 21 Feb 2009 02:15:08 +0000 (02:15 +0000)]
Include altivec.c, missed on previous commit r188860. This should unbreak the
build.

15 years agoProvide compat typedefs for usb1.
Andrew Thompson [Sat, 21 Feb 2009 01:14:49 +0000 (01:14 +0000)]
Provide compat typedefs for usb1.

15 years agoDisable port hardware on detach. First switch it to slumber mode to
Alexander Motin [Fri, 20 Feb 2009 23:17:52 +0000 (23:17 +0000)]
Disable port hardware on detach. First switch it to slumber mode to
power-down peer transmitter, then disable completely.
Side effect of this is saving about 0.5W of power per detached device.

15 years agomark nodes created by way of neighbor discovery need to be marked as ERP
Sam Leffler [Fri, 20 Feb 2009 23:05:05 +0000 (23:05 +0000)]
mark nodes created by way of neighbor discovery need to be marked as ERP
so they get the correct tx parameters setup by node_settxparms

15 years agoDon't read the TOC for DIOCGPROVIDERALIAS. This causes the
Marcel Moolenaar [Fri, 20 Feb 2009 23:01:27 +0000 (23:01 +0000)]
Don't read the TOC for DIOCGPROVIDERALIAS. This causes the
acd code to create new provider (?), but not from an expected
context, which results in a panic (GEOM topology not held).

15 years agoFix typo: s/SLUMPER/SLUMBER/
Alexander Motin [Fri, 20 Feb 2009 22:46:24 +0000 (22:46 +0000)]
Fix typo: s/SLUMPER/SLUMBER/

15 years agocorrect SIFS setting; there is a 2usec adjustment between the calculated
Sam Leffler [Fri, 20 Feb 2009 22:06:58 +0000 (22:06 +0000)]
correct SIFS setting; there is a 2usec adjustment between the calculated
value and what the hardware requires (based on inspection of INI values)

Submitted by: Jiri Fojtasek <jiri.fojtasek@hlohovec.net>

15 years agodon't adjust core clk conversions for 1/2 and 1/4 rate channels; the
Sam Leffler [Fri, 20 Feb 2009 22:05:08 +0000 (22:05 +0000)]
don't adjust core clk conversions for 1/2 and 1/4 rate channels; the
mac runs at full speed so doing this breaks conversion for ifs parameters

Submitted by: Felix Fietkau <nbd@openwrt.org>

15 years agoo reset aggressive mode flag; it was being left set after marking an
Sam Leffler [Fri, 20 Feb 2009 21:57:05 +0000 (21:57 +0000)]
o reset aggressive mode flag; it was being left set after marking an
  interface down
o only allow the first vap to initialize shared wme parameters

15 years agomisc cleanup of wme parameter setting
Sam Leffler [Fri, 20 Feb 2009 21:55:23 +0000 (21:55 +0000)]
misc cleanup of wme parameter setting

15 years agoReduce soft-reset assertion time from 5ms to 50us.
Alexander Motin [Fri, 20 Feb 2009 20:29:32 +0000 (20:29 +0000)]
Reduce soft-reset assertion time from 5ms to 50us.
ATA specification declares minimal reset time of 5us. SATA keeps it, but
requires devices to handle commands transmitted even one by one without
any gap.

15 years agoDisable port interrupts before doing PHY reset.
Alexander Motin [Fri, 20 Feb 2009 20:20:00 +0000 (20:20 +0000)]
Disable port interrupts before doing PHY reset.
PHY reset causes drive connect/disconnect events, unwanted at this moment.

15 years agoAdd Altivec support for supported CPUs. This is derived from the FPU support
Nathan Whitehorn [Fri, 20 Feb 2009 17:48:40 +0000 (17:48 +0000)]
Add Altivec support for supported CPUs. This is derived from the FPU support
code, and also reducing the size of trapcode to fit inside a 32 byte handler
slot.

Reviewed by: grehan
MFC after: 2 weeks

15 years agoEliminate stale comments.
Alan Cox [Fri, 20 Feb 2009 16:19:34 +0000 (16:19 +0000)]
Eliminate stale comments.

15 years agoAdd the add-stream capability. Still needs more
Randall Stewart [Fri, 20 Feb 2009 15:03:54 +0000 (15:03 +0000)]
Add the add-stream capability. Still needs more
testing..

MFC after: 1 month

15 years agoDelete files which serve no purpose in this environment.
Michael Reifenberger [Fri, 20 Feb 2009 14:53:49 +0000 (14:53 +0000)]
Delete files which serve no purpose in this environment.

Inspired by: simon@

15 years agoFix a bug. The sending was being restricted improperly by
Randall Stewart [Fri, 20 Feb 2009 14:33:45 +0000 (14:33 +0000)]
Fix a bug. The sending was being restricted improperly by
the max_burst. It should only be gated by cwnd in the
lower level send.

Obtained from: Michael Tuexen
MFC after: 1 week.

15 years agowhen initializing, also build _includes and _libraries
Luigi Rizzo [Fri, 20 Feb 2009 14:28:56 +0000 (14:28 +0000)]
when initializing, also build _includes and _libraries

On passing, fix a wrong comment

MFC after: 3 days

15 years agoFix pc98 build error due to missing variable.
Alexander Motin [Fri, 20 Feb 2009 14:10:23 +0000 (14:10 +0000)]
Fix pc98 build error due to missing variable.

Submitted by: avg

15 years agoDon't make Linux stat() open character devices to resolve its name.
Ed Schouten [Fri, 20 Feb 2009 13:05:29 +0000 (13:05 +0000)]
Don't make Linux stat() open character devices to resolve its name.

The existing code calls kern_open() to resolve the vnode of a pathname
right after a stat(). This is not correct, because it causes random
character devices to be opened in /dev. This means ls'ing a tape
streamer will cause it to rewind, for example. Changes I have made:

- Add kern_statat_vnhook() to allow binary emulators to `post-process'
  struct stat, using the proper vnode.

- Remove unneeded printf's from stat() and statfs().

- Make the Linuxolator use kern_statat_vnhook(), replacing
  translate_path_major_minor_at().

- Let translate_fd_major_minor() use vp->v_rdev instead of
  vp->v_un.vu_cdev.

Result:

crw-rw-rw- 1 root root   0, 14 Feb 20 13:54 /dev/ptmx
crw--w---- 1 root adm  136,  0 Feb 20 14:03 /dev/pts/0
crw--w---- 1 root adm  136,  1 Feb 20 14:02 /dev/pts/1
crw--w---- 1 ed   tty  136,  2 Feb 20 14:03 /dev/pts/2

Before this commit, ptmx also had a major number of 136, because it
silently allocated and deallocated a pseudo-terminal. Device nodes that
cannot be opened now have proper major/minor-numbers.

Reviewed by: kib, netchild, rdivacky (thanks!)

15 years agoDoh! r188813 modified an auto-generated file. Fix it for real:
Mike Makonnen [Fri, 20 Feb 2009 11:09:55 +0000 (11:09 +0000)]
Doh! r188813 modified an auto-generated file. Fix it for real:
  o grammar fix
  o remove mention of libkse

Submitted by: Niclas Zeising <niclas.zeising@gmail.com>

15 years agoHandle nForce MCP67 and MCP73 SATA controllers as AHCI. They report itself
Alexander Motin [Fri, 20 Feb 2009 08:49:56 +0000 (08:49 +0000)]
Handle nForce MCP67 and MCP73 SATA controllers as AHCI. They report itself
as ATA RAID, but generic ATAPCI driver unable to detect drives there. AHCI
driver reported to handle them fine. Linux does the same.

Submitted by: Andrey V. Elsukov on stable@

15 years agoTell CAM that CISS logical devices support tagged queueing. This fixes
Scott Long [Fri, 20 Feb 2009 07:43:43 +0000 (07:43 +0000)]
Tell CAM that CISS logical devices support tagged queueing.  This fixes
the low "max device openings" count that has led to poor performance in
FreeBSD 7.0 and 7.1

15 years agoTell CAM that SPI disconnect works, which in turn will let it use tags.
Scott Long [Fri, 20 Feb 2009 07:40:54 +0000 (07:40 +0000)]
Tell CAM that SPI disconnect works, which in turn will let it use tags.
This fixes the low "max device openings" count that has lead to poor
performance in FreeBSD 7.0 and 7.1.

Extra thanks goes to Mike Tancsa at Sentex for providing a debug system for
this.

15 years agoAdd basic support for DDF, often found on Adaptec HostRAID controllers.
Scott Long [Fri, 20 Feb 2009 06:10:12 +0000 (06:10 +0000)]
Add basic support for DDF, often found on Adaptec HostRAID controllers.
Spares and rebuilds are not supported, so this code should be considered
for entertainment purposes only.

15 years agoProvide compatibility symlink for logical partitions:
Marcel Moolenaar [Fri, 20 Feb 2009 04:48:40 +0000 (04:48 +0000)]
Provide compatibility symlink for logical partitions:
1.  Extend geom_dev by having it create the symlink (i.e. call
    make_dev_alias) based on the DIOCGPROVIDERALIAS ioctl.
    In this way the functionaility is generic and thus usable
    by any geom/provider.
2.  Have g_part handle said ioctl through the devalias method,
    so that it's under control of the scheme itself. By design
    the alias will not be created for newly added partitions.

15 years agoFix an infinite loop created when the last logical partition is
Marcel Moolenaar [Fri, 20 Feb 2009 04:10:31 +0000 (04:10 +0000)]
Fix an infinite loop created when the last logical partition is
removed.

15 years ago- Add glarkin and mva to mentor section
Beech Rintoul [Fri, 20 Feb 2009 02:48:58 +0000 (02:48 +0000)]
- Add glarkin and mva to mentor section

15 years agouse /boot/loader by default, because the boot code seems to
Luigi Rizzo [Fri, 20 Feb 2009 00:05:33 +0000 (00:05 +0000)]
use /boot/loader by default, because the boot code seems to
have problems with kernels larger than 4MB.
Add a flag to avoid the /boot/loader and use the old method.

Add support for an additional makefile to perform custom manipulation
(this is not documented yet).

Add support for building an ISO image (not complete)

15 years agoupdate 'sio' with 'uart' and apply related changes.
Luigi Rizzo [Thu, 19 Feb 2009 23:29:05 +0000 (23:29 +0000)]
update 'sio' with 'uart' and apply related changes.

15 years agoEnable caching of negative pathname lookups in the NFS client. To avoid
John Baldwin [Thu, 19 Feb 2009 22:28:48 +0000 (22:28 +0000)]
Enable caching of negative pathname lookups in the NFS client.  To avoid
stale entries, we save a copy of the directory's modification time when
the first negative cache entry was added in the directory's NFS node.
When a negative cache entry is hit during a pathname lookup, the parent
directory's modification time is checked.  If it has changed, all of the
negative cache entries for that parent are purged and the lookup falls
back to using the RPC.  This required adding a new cache_purge_negative()
method to the name cache to purge only negative cache entries for a given
directory.

Submitted by: mohans, Rick Macklem, Ricardo Labiaga @ NetApp
Reviewed by: mohans

15 years agoWhen fetching attributes for a file for NFSv3 mounts, do not perform an
John Baldwin [Thu, 19 Feb 2009 22:18:00 +0000 (22:18 +0000)]
When fetching attributes for a file for NFSv3 mounts, do not perform an
opportunistic ACCESS RPC to populate both the access and attribute caches
of the file and instead always use a GETATTR RPC.  On many modern NFS
servers, an ACCESS RPC is much more expensive to service than a GETATTR
RPC.

Submitted by: mohans

15 years agoDon't clear the attribute cache of a file when it is closed. A subsequent
John Baldwin [Thu, 19 Feb 2009 22:10:39 +0000 (22:10 +0000)]
Don't clear the attribute cache of a file when it is closed.  A subsequent
open() of the same file will load fresh attributes, so they do not need to
be explicitly flushed in close() to guarantee close to open consistency.
However, other file desciptors may still reference this file and clearing
the attributes in close() forces those other file descriptors to fetch
fresh attributes the next time they need them.

Reviewed by: mohans
MFC after: 1 week

15 years agoDisplay an error message when the requested mode is not known. So the user can
Roman Divacky [Thu, 19 Feb 2009 20:45:37 +0000 (20:45 +0000)]
Display an error message when the requested mode is not known. So the user can
distinguish between a typo in the mode name and that the device does not
support a certain mode (till now both causes show the same result, i.e. the old
mode is displayed).

Submitted by: Christoph Mallon <christoph.mallon gmx.de>
Approved by: kib (mentor)

15 years agoInclude proper header files for system functions that are used and
Warner Losh [Thu, 19 Feb 2009 20:07:59 +0000 (20:07 +0000)]
Include proper header files for system functions that are used and
that gcc complains about this code changing from the built-in
versions.

15 years agoThese warnings are only relevant on NetBSD it seems. They don't seem
Warner Losh [Thu, 19 Feb 2009 19:53:58 +0000 (19:53 +0000)]
These warnings are only relevant on NetBSD it seems.  They don't seem
to be relevant to FreeBSD at all.

15 years agoInclude string.h for strncpy prototype
Warner Losh [Thu, 19 Feb 2009 19:45:49 +0000 (19:45 +0000)]
Include string.h for strncpy prototype
Include strings.h for bcopy and bzero prototype

15 years agoEnable building of ficl on MIPS. It compiles, but needs more testing.
Warner Losh [Thu, 19 Feb 2009 18:53:08 +0000 (18:53 +0000)]
Enable building of ficl on MIPS.  It compiles, but needs more testing.

15 years agoAdd basic mips support. This has been compile tested only.
Warner Losh [Thu, 19 Feb 2009 18:52:31 +0000 (18:52 +0000)]
Add basic mips support.  This has been compile tested only.

15 years agoThese symbols don't belong here. Remove them. Since mips hasn't had
Warner Losh [Thu, 19 Feb 2009 18:03:41 +0000 (18:03 +0000)]
These symbols don't belong here.  Remove them.  Since mips hasn't had
a release, I think there's no impact here...

Reviewed by: cognet@

15 years agoSquash some small bugs in pts(4).
Ed Schouten [Thu, 19 Feb 2009 17:54:42 +0000 (17:54 +0000)]
Squash some small bugs in pts(4).

- Don't return a negative errno when using an unknown ioctl() on a
  pseudo-terminal master device. Be sure to convert ENOIOCTL to ENOTTY,
  just like the TTY layer does.

- Even though we should return st_rdev of the master device node when
  emulating pty(4) devices, FIODGNAME should still return the name of
  the slave device. Otherwise ptsname(3) and ttyname(3) return an
  invalid device name.

15 years agouse 1/2 and 1/4 rate phy types to simplify txtime calculation
Sam Leffler [Thu, 19 Feb 2009 17:44:23 +0000 (17:44 +0000)]
use 1/2 and 1/4 rate phy types to simplify txtime calculation

15 years agoProperly convert bit value to a bit field. Before we were storing
Warner Losh [Thu, 19 Feb 2009 16:30:11 +0000 (16:30 +0000)]
Properly convert bit value to a bit field.  Before we were storing
values like 0x80 or 0x40 into a uint8_t foo:1 bitfield.  This would
result in the bit always being 0.  One of these caused a warning for
overflow (one that was 0x80), but the other didn't.  They were both
wrong.

This is why I hate code that mixes c struct bitfields and #defines.
The rest of the fields accessed by the program should be audited.

15 years agoAdd myself.
Max Brazhnikov [Thu, 19 Feb 2009 16:16:44 +0000 (16:16 +0000)]
Add myself.

15 years agoAdd myself.
Max Brazhnikov [Thu, 19 Feb 2009 16:15:30 +0000 (16:15 +0000)]
Add myself.

15 years agostyle nit in r188815
Andriy Gapon [Thu, 19 Feb 2009 15:37:43 +0000 (15:37 +0000)]
style nit in r188815

Pointed out by: jhb, rpaulo
Approved by: jhb (mentor)

15 years agofs/udf: fix incorrect error return (-1) when reading a large dir
Andriy Gapon [Thu, 19 Feb 2009 15:05:30 +0000 (15:05 +0000)]
fs/udf: fix incorrect error return (-1) when reading a large dir

Not enough space in user-land buffer is not an error, userland
will read further until eof is reached. So instead of propagating
-1 to caller we convert it to zero/success.

cd9660 code works exactly the same way.

PR: kern/78987
Reviewed by: jhb (mentor)
Approved by: jhb (mentor)

15 years agoacpi_cpu: fixup for PIIX4E PCI config related to C2
Andriy Gapon [Thu, 19 Feb 2009 14:39:52 +0000 (14:39 +0000)]
acpi_cpu: fixup for PIIX4E PCI config related to C2

This is triggered only if BIOS configures ACPI_BITREG_BUS_MASTER_RLD
aka BRLD_EN_BM to 1.
Rationale:
1. we do not support C3 on PIIX4E
2. bus master activity need not break out of C2 state
3. because of CPU_QUIRK_NO_BM_CTRL quirk we may reset bus master
   status which would result in immediate break out from C2

So if you have seen
cpu0: too many short sleeps, backing off to C1
with this chipset before you may want to try cx_lowest of C2 again.

Reviewed by: rpaulo (mentor), njl
Approved by: rpaulo (mentor)

15 years agoo grammar fix
Mike Makonnen [Thu, 19 Feb 2009 13:45:25 +0000 (13:45 +0000)]
o grammar fix
o remove mention of libkse

15 years agoUse channel driver's attach/detach routines instead of ata_attach()/
Alexander Motin [Thu, 19 Feb 2009 12:47:24 +0000 (12:47 +0000)]
Use channel driver's attach/detach routines instead of ata_attach()/
ata_detach() to implement IOCATAATTACH/IOCATADETACH ioctls.
This will permit channel drivers to properly shutdown port hardware on channel
detach and init it on attach.

15 years agorevamp after 1/2 and 1/4 rate changes
Sam Leffler [Thu, 19 Feb 2009 05:36:07 +0000 (05:36 +0000)]
revamp after 1/2 and 1/4 rate changes

15 years agoFixup handling of roaming and xmit parameters to support 1/2 and 1/4-width
Sam Leffler [Thu, 19 Feb 2009 05:28:03 +0000 (05:28 +0000)]
Fixup handling of roaming and xmit parameters to support 1/2 and 1/4-width
channel modes:
o usurp 'h' mode flag for half-width channels
o add 'q' mode flag for quarter-width channels
o rewrite rate parameter parsing to handle fractional values
o merge mode loops to eliminate ordering assumptions
o replace 0x80 with IEEE80211_RATE_MCS

15 years agoremove private support for IEEE80211_MODE_HALF and IEEE80211_MODE_QUARTER
Sam Leffler [Thu, 19 Feb 2009 05:22:40 +0000 (05:22 +0000)]
remove private support for IEEE80211_MODE_HALF and IEEE80211_MODE_QUARTER
now that net80211 has them

15 years agoAdd modes for 1/2 and 1/4-width channels so we have separate roaming
Sam Leffler [Thu, 19 Feb 2009 05:21:54 +0000 (05:21 +0000)]
Add modes for 1/2 and 1/4-width channels so we have separate roaming
and xmit parameters.  This makes it possible to use tdma on fractional
channels.
o add IEEE80211_MODE_HALF and IEEE80211_MODE_QUARTER; note these are
  band-agnostic (may need revisiting)
o setup all default rates in ic_sup_rates instead of doing it only
  for active modes; we need these to calculate the default tx parameters
  which are not recalculated after a regulatory update (can't just
  recalculate after installing a new channel list because we might
  clobber user settings)
o remove special case code in ieee80211_get_suprates; this is now
  a candidate for an inline or removal
o add various entries for new modes (roaming+tx params, wme, rate
  mapping, scan set setup, country ie construction, tdma, basic rates)

Note these modes are intentionally not visible through if_media.

15 years agoo looseen abi dependencies on IEEE80211_MODE_MAX by accepting smaller
Sam Leffler [Thu, 19 Feb 2009 05:03:10 +0000 (05:03 +0000)]
o looseen abi dependencies on IEEE80211_MODE_MAX by accepting smaller
  parameters for IEEE80211_IOC_ROAM and IEEE80211_IOC_TXPARAMS; this
  lets us add more modes and still have old apps work
o consolidate loops to remote assumptions about mode ordering

15 years agouse c99 initializers
Sam Leffler [Thu, 19 Feb 2009 04:52:03 +0000 (04:52 +0000)]
use c99 initializers