]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
20 years agoWhitespace cleanup.
Dag-Erling Smørgrav [Thu, 13 Nov 2003 08:26:29 +0000 (08:26 +0000)]
Whitespace cleanup.

20 years agoRespect RB_KDB flag.
Hidetoshi Shimokawa [Thu, 13 Nov 2003 07:41:55 +0000 (07:41 +0000)]
Respect RB_KDB flag.

20 years agoAdd tunables.
Hidetoshi Shimokawa [Thu, 13 Nov 2003 06:29:40 +0000 (06:29 +0000)]
Add tunables.

20 years agoDon't count PHY errors as input errors. This is important for
Sam Leffler [Thu, 13 Nov 2003 05:35:07 +0000 (05:35 +0000)]
Don't count PHY errors as input errors.  This is important for
5212-based devices because PHY errors are used to collect data
on environmental noise that and doesn't truly reflect the state
of the communications media.  The result is confused users.
Folks that want to watch PHY errors can still get the statistics
through the device ioctl (used by athstats).

20 years agoDon't mmap(2) and munmap(2) zero-length files.
Alan Cox [Thu, 13 Nov 2003 05:26:55 +0000 (05:26 +0000)]
Don't mmap(2) and munmap(2) zero-length files.

Submitted by: Wiktor Niesiobedzki <bsd@w.evip.pl>

20 years agoo insure the current channel is in a good state before starting an AP scan
Sam Leffler [Thu, 13 Nov 2003 05:23:58 +0000 (05:23 +0000)]
o insure the current channel is in a good state before starting an AP scan
o reject scan requests for a device that isn't marked up

This fixes a problem where requesting a scan before marking the device
up would cause a panic because the current channel was set to "any" (0xffff).

20 years agoadd missing inpcb lock before call to tcp_twclose (which reclaims the inpcb)
Sam Leffler [Thu, 13 Nov 2003 05:18:23 +0000 (05:18 +0000)]
add missing inpcb lock before call to tcp_twclose (which reclaims the inpcb)

Supported by: FreeBSD Foundation

20 years agoo reorder some locking asserts to reflect the order of the locks
Sam Leffler [Thu, 13 Nov 2003 05:16:56 +0000 (05:16 +0000)]
o reorder some locking asserts to reflect the order of the locks
o correct a read-lock assert in in_pcblookup_local that should be
  a write-lock assert (since time wait close cleanups may alter state)

Supported by: FreeBSD Foundation

20 years agoSave the device so we can do a device_printf.
Warner Losh [Thu, 13 Nov 2003 04:14:53 +0000 (04:14 +0000)]
Save the device so we can do a device_printf.
Use this in preference to aha_name.
Remove aha_name function and #define it to device_get_unitname()
Minor indentation tweaks resulting therefrom

20 years agoCall free(9) after the vnode interlock is released, avoiding a lock-order
Alan Cox [Thu, 13 Nov 2003 03:56:32 +0000 (03:56 +0000)]
Call free(9) after the vnode interlock is released, avoiding a lock-order
reversal.

20 years agoStop pretending to support kernel profiling. The FAKE_MCOUNT() etc
Peter Wemm [Thu, 13 Nov 2003 02:38:33 +0000 (02:38 +0000)]
Stop pretending to support kernel profiling.  The FAKE_MCOUNT() etc
calls are just gradually getting more and more stale.  At this point it
would be better to start from scratch once prof_machdep.c is adapted.

20 years ago - Add the WITNESS_SKIPSPIN option to the GENERIC kernel so that users who
Jeff Roberson [Thu, 13 Nov 2003 02:04:34 +0000 (02:04 +0000)]
 - Add the WITNESS_SKIPSPIN option to the GENERIC kernel so that users who
   enable WITNESS on alphas will not get the expensive spin lock checking by
   default.  This mirrors the default config settings for i386.

20 years agoMove global variables for icmp_input() to its stack. With SMP or
Andre Oppermann [Thu, 13 Nov 2003 00:32:13 +0000 (00:32 +0000)]
Move global variables for icmp_input() to its stack.  With SMP or
preemption two CPUs can be in the same function at the same time
and clobber each others variables.  Remove register declaration
from local variables.

Reviewed by: sam (mentor)

20 years agoadd in-kernel ttcp performance tool
Sam Leffler [Thu, 13 Nov 2003 00:30:27 +0000 (00:30 +0000)]
add in-kernel ttcp performance tool

20 years agoWarn that you won't be able to build a kernel if you do an installworld
Warner Losh [Thu, 13 Nov 2003 00:10:43 +0000 (00:10 +0000)]
Warn that you won't be able to build a kernel if you do an installworld

20 years ago- Close a race where a thread on another CPU could release a contested lock
John Baldwin [Wed, 12 Nov 2003 23:48:42 +0000 (23:48 +0000)]
- Close a race where a thread on another CPU could release a contested lock
  and empty its turnstile while the blocking threads still pointed to the
  turnstile.  If the thread on the first CPU blocked on a lock owned by
  one of the threads blocked on the turnstile just woken up, then the
  first CPU could try to manipulate a bogus thread queue in the turnstile
  during priority propagation.
- Update locking notes for ts_owner and always clear ts_owner, not just
  under INVARIANTS.

Tested by:      sam (1)

20 years agoFix on sparc64.
Maxim Sobolev [Wed, 12 Nov 2003 23:36:17 +0000 (23:36 +0000)]
Fix on sparc64.

Reported by: rwatson/tinderbox
MFC after: 2 weeks

20 years agoDo not fragment a packet with hardware assistance if it has the DF
Andre Oppermann [Wed, 12 Nov 2003 23:35:40 +0000 (23:35 +0000)]
Do not fragment a packet with hardware assistance if it has the DF
bit set.

Reviewed by: sam (mentor)

20 years agoAttempt to save the last dregs of emacs users' sanity by saving the
Ceri Davies [Wed, 12 Nov 2003 23:08:23 +0000 (23:08 +0000)]
Attempt to save the last dregs of emacs users' sanity by saving the
contents of the PR when an interrupt is received during the editor
session.  This stops the use of ^G from deleting a filled PR from
underneath the user.

PR: bin/59201
Submitted by: Heikki Suonsivu <hsu@evoluutio.bbnetworks.net>
MFC After: 2 weeks

20 years agoAdd vendor ID to make Marvell chipset work. E.g. to be found
Wilko Bulte [Wed, 12 Nov 2003 23:01:15 +0000 (23:01 +0000)]
Add vendor ID to make Marvell chipset work. E.g. to be found
on SMC9452TX it seems

Submitted by: Jung-uk Kim <jkim@niksun.com>
Tested by: <Radu Bogdan 'veedee' Rusu> veedee@c7.campus.utcluj.ro
MFC after: 2 weeks

20 years agoAt the request of several developers, restore the DIAGNOSIC code
Kirk McKusick [Wed, 12 Nov 2003 22:28:27 +0000 (22:28 +0000)]
At the request of several developers, restore the DIAGNOSIC code
deleted in 1.81. Increase the initial timeout limit to 2ms to
eliminate spurious messages of excessive timeouts in the NFS
client code.

Requested by: Poul-Henning Kamp <phk@phk.freebsd.dk>
Requested by: Mike Silbersack <silby@silby.com>
Requested by: Sam Leffler <sam@errno.com>

20 years agoMark __mac_get_pid() as MPSAFE in the comment, as it runs without
Robert Watson [Wed, 12 Nov 2003 22:19:15 +0000 (22:19 +0000)]
Mark __mac_get_pid() as MPSAFE in the comment, as it runs without
Giant and is also MPSAFE.

Push Giant further down into __mac_get_fd() and __mac_set_fd(),
grabbing it only for constrained regions dealing with VFS, and
dropping it entirely for operations related to labeling of pipes.

Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

20 years agoBump the major version on libtuil. libutil now relies on the mac_*
Robert Watson [Wed, 12 Nov 2003 21:56:07 +0000 (21:56 +0000)]
Bump the major version on libtuil.  libutil now relies on the mac_*
symbols exported by newer versions of libc, and so we want applications
depending on the newer library code to be required to link against the
newer libc.

Discussed with: scottl, kris, imp

20 years agoFix format strings (intmax_t is %jd, not %qd)
Kris Kennaway [Wed, 12 Nov 2003 21:47:42 +0000 (21:47 +0000)]
Fix format strings (intmax_t is %jd, not %qd)

Reviewed by: tjr

20 years agoreflect ip6_pktopts and ip6_moptions into embeded scope of
Hajimu UMEMOTO [Wed, 12 Nov 2003 21:39:12 +0000 (21:39 +0000)]
reflect ip6_pktopts and ip6_moptions into embeded scope of
destination address.  it makes `ping6 -I <if> <link-local>'
work again.  since we don't merge scope cleanup yet, we need
this for workaround.

20 years agoUpdate from DRI CVS. Includes locking fixes (including PR 59202), changes for
Eric Anholt [Wed, 12 Nov 2003 20:56:30 +0000 (20:56 +0000)]
Update from DRI CVS.  Includes locking fixes (including PR 59202), changes for
Radeon IGP support (still lacking PCI IDs), and DRM interface 1.2 updates which
include finally tying the DRM instances to specific devices rather than relying
on the X Server.

20 years agoAdd a new sysctl knob, net.inet.udp.strict_mcast_mship, to the udp_input path.
Bruce M Simpson [Wed, 12 Nov 2003 20:17:11 +0000 (20:17 +0000)]
Add a new sysctl knob, net.inet.udp.strict_mcast_mship, to the udp_input path.

This switch toggles between strict multicast delivery, and traditional
multicast delivery.

The traditional (default) behaviour is to deliver multicast datagrams to all
sockets which are members of that group, regardless of the network interface
where the datagrams were received.

The strict behaviour is to deliver multicast datagrams received on a
particular interface only to sockets whose membership is bound to that
interface.

Note that as a matter of course, multicast consumers specifying INADDR_ANY
for their interface get joined on the interface where the default route
happens to be bound. This switch has no effect if the interface which the
consumer specifies for IP_ADD_MEMBERSHIP is not UP and RUNNING.

The original patch has been cleaned up somewhat from that submitted. It has
been tested on a multihomed machine with multiple QuickTime RTP streams
running over the local switch, which doesn't do IGMP snooping.

PR: kern/58359
Submitted by: William A. Carrel
Reviewed by: rwatson
MFC after: 1 week

20 years agoAdd a card carryring member of the Pointless Comparison Collection.
Poul-Henning Kamp [Wed, 12 Nov 2003 20:08:47 +0000 (20:08 +0000)]
Add a card carryring member of the Pointless Comparison Collection.

20 years agoFix some typos.
John Baldwin [Wed, 12 Nov 2003 19:46:17 +0000 (19:46 +0000)]
Fix some typos.

20 years agodropwithreset is not needed in this case as tcp_drop() is already notifying
Andre Oppermann [Wed, 12 Nov 2003 19:38:01 +0000 (19:38 +0000)]
dropwithreset is not needed in this case as tcp_drop() is already notifying
the other side. Before we were sending two RST packets.

20 years agoAdd an entry about the changes in the statfs structure that require
Kirk McKusick [Wed, 12 Nov 2003 19:13:02 +0000 (19:13 +0000)]
Add an entry about the changes in the statfs structure that require
kernel and user land applications doing statfs to be in sync.

Requested by: Johan Karlsson <johan@FreeBSD.org>

20 years ago- setsockopt/sysctl takes int, not u_long.
Hajimu UMEMOTO [Wed, 12 Nov 2003 18:56:59 +0000 (18:56 +0000)]
- setsockopt/sysctl takes int, not u_long.
- be more picky about argument parsing - like ERANGE.
- use u_long for args, not to lose accuracy/prevent overflow.
- socklen_t audit.
- Add -I (use icmp) option.
- warn if multiple addresses are present for dest.
- no need to pass tz.
- type pedant.  check -p range.
- grab hlim from sysctl.
- typo in port number setting.

Obtained from: KAME

20 years agoGarbage collect unused values.
John Baldwin [Wed, 12 Nov 2003 18:14:34 +0000 (18:14 +0000)]
Garbage collect unused values.

20 years ago- Move manipulation of td_intr_nesting_level out of assembly interrupt
John Baldwin [Wed, 12 Nov 2003 18:13:57 +0000 (18:13 +0000)]
- Move manipulation of td_intr_nesting_level out of assembly interrupt
  vector stubs and into the C functions they call.
- Move disabling and EOIing of interrupt sources out of PIC driver entry
  points and into intr_execute_handlers().  Intr_execute_handlers() only
  disables a source for an interrupt if it is a stray interrupt or has
  threaded handlers.  Sources with fast handlers no longer disable (mask)
  the source while executing the handlers.
- Move the setting of clkintr_pending into intr_execute_handlers() and set
  the variable for any interrupt source with a vector of 0.  (Should only
  be true for IRQ 0.)  This fixes clkintr_pending in the NO_MIXED_MODE
  case.
- Implement lapic_eoi() and use it to implement ioapic_eoi_source().
- Rename atpic_sched_ithd() to atpic_handle_intr() since it is used to
  handle all atpic interrupts and not just threaded ones.

Inspired by: peter's changes to amd64 in p4 (1)
Requested by: bde (2)

20 years agoAdd a new configuration variable - nas_ipaddr, which if set allows to
Maxim Sobolev [Wed, 12 Nov 2003 17:47:23 +0000 (17:47 +0000)]
Add a new configuration variable - nas_ipaddr, which if set allows to
set NAS-IP-Address attribute in requests generated by the pam_radius
module. This attribute is mandatory for some Radius servers out there.

Reviewed by: des
MFC after: 2 weeks

20 years agoThe partition naming on ia64 (e.g. da0p1) cannot be selected based
Marcel Moolenaar [Wed, 12 Nov 2003 17:44:37 +0000 (17:44 +0000)]
The partition naming on ia64 (e.g. da0p1) cannot be selected based
on whether the parent chunk is of type whole. This also applies to
MBR slices for non-GPT disks. Since most of the GPT handling is
conditionally compiled, do the same with the partition naming.

This fixes a braino that caused slices to be named as GPT partitions
and generally messing up an install.

Pointy hat: marcel

20 years agoRemove extraneous & to fix compile.
John Baldwin [Wed, 12 Nov 2003 17:21:57 +0000 (17:21 +0000)]
Remove extraneous & to fix compile.

20 years agoThe snmp_netgraph module depends on libnetgraph. So add a dependency and
Hartmut Brandt [Wed, 12 Nov 2003 17:09:13 +0000 (17:09 +0000)]
The snmp_netgraph module depends on libnetgraph. So add a dependency and
add libnetgraph to the list of prebuilt libraries in the main Makefile.

Reviewed by: ru

20 years agoMNAMELEN is back to an int again after Kirk's statfs commit
Peter Wemm [Wed, 12 Nov 2003 17:09:12 +0000 (17:09 +0000)]
MNAMELEN is back to an int again after Kirk's statfs commit

kern/vfs_mount.c:1305: warning: signed size_t format, different type arg (arg 4)
*** Error code 1

20 years agoBump the netgraph header version to 6 for the change of the name
Hartmut Brandt [Wed, 12 Nov 2003 17:03:40 +0000 (17:03 +0000)]
Bump the netgraph header version to 6 for the change of the name
length definitions.

Reminded by: jdp

20 years agoReindent to non-style(9) compliant 4 space indent to match rest of file.
John Baldwin [Wed, 12 Nov 2003 16:24:16 +0000 (16:24 +0000)]
Reindent to non-style(9) compliant 4 space indent to match rest of file.

20 years agoAdd an ACPI_FUNCTION_TRACE() invocation to acpi_OverrideInterruptLevel()
John Baldwin [Wed, 12 Nov 2003 16:23:21 +0000 (16:23 +0000)]
Add an ACPI_FUNCTION_TRACE() invocation to acpi_OverrideInterruptLevel()
to fix compile with ACPI_DEBUG.

Reported by: tinderbox

20 years agoRemove extraneous fullpath variable, which broke lint build. This
Robert Watson [Wed, 12 Nov 2003 15:09:39 +0000 (15:09 +0000)]
Remove extraneous fullpath variable, which broke lint build.  This
extra argument to the devfs MAC policy entry points was accidentally
merged from the MAC branch during my earlier commit to these policies,
and is not scheduled to be merged just yet.

20 years agoFixed some style bugs (insertion sort error and extra blank line).
Bruce Evans [Wed, 12 Nov 2003 15:07:18 +0000 (15:07 +0000)]
Fixed some style bugs (insertion sort error and extra blank line).

20 years agoFix a typo in a comment.
John Baldwin [Wed, 12 Nov 2003 14:55:45 +0000 (14:55 +0000)]
Fix a typo in a comment.

Submitted by: das

20 years agoSync the list of supported devices in the manual page with the hardware
Simon L. B. Nielsen [Wed, 12 Nov 2003 14:13:36 +0000 (14:13 +0000)]
Sync the list of supported devices in the manual page with the hardware
notes and the driver source.

20 years agoDocument that the length of a command string is now 32. Also change all
Hartmut Brandt [Wed, 12 Nov 2003 13:24:03 +0000 (13:24 +0000)]
Document that the length of a command string is now 32. Also change all
occurences of the old *LEN definitions to the new *SIZ definitions.

Correct the example how to compare a nodename.
strncmp(name, "fred", NG_NODELEN) may step behind "fred" if the node
name is really fred.

Reminded by: ru

20 years agoGet rid of the old constants which did not include the trailing NUL
Hartmut Brandt [Wed, 12 Nov 2003 13:04:44 +0000 (13:04 +0000)]
Get rid of the old constants which did not include the trailing NUL
and use the new constants which do.

20 years ago- vlan_start(): Increment the correct interface statistics member.
Ruslan Ermilov [Wed, 12 Nov 2003 12:58:19 +0000 (12:58 +0000)]
- vlan_start(): Increment the correct interface statistics member.

Reviewed by: mdodd

- vlan_input(): Macroize the VLAN tag extraction from mbuf.

20 years agoBetter English usage.
Tony Finch [Wed, 12 Nov 2003 10:53:52 +0000 (10:53 +0000)]
Better English usage.

Submitted by: wollman

20 years agoReplace B_PHYS conditional assignment to bio_offset with KASSERT check
Poul-Henning Kamp [Wed, 12 Nov 2003 10:27:06 +0000 (10:27 +0000)]
Replace B_PHYS conditional assignment to bio_offset with KASSERT check
to see that the originating code already did it right.

20 years agoDon't mess around with spare fields of public structures.
Poul-Henning Kamp [Wed, 12 Nov 2003 09:54:07 +0000 (09:54 +0000)]
Don't mess around with spare fields of public structures.

20 years agoDon't mess about with spare fields in public structures.
Poul-Henning Kamp [Wed, 12 Nov 2003 09:52:10 +0000 (09:52 +0000)]
Don't mess about with spare fields in public structures.

20 years agoMake sure to return errors if we have any.
Poul-Henning Kamp [Wed, 12 Nov 2003 09:46:54 +0000 (09:46 +0000)]
Make sure to return errors if we have any.

Submitted by:    Pawel Jakub Dawidek <nick@garage.freebsd.pl>

20 years agoAdd an entry about the changes in netgraph that require kernel and
Hartmut Brandt [Wed, 12 Nov 2003 09:11:18 +0000 (09:11 +0000)]
Add an entry about the changes in netgraph that require kernel and
user land netgraph stuff to be in sync.

Reviewed by: imp

20 years agoDouble length of node names, hook names, command strings and types. Add
Hartmut Brandt [Wed, 12 Nov 2003 09:10:11 +0000 (09:10 +0000)]
Double length of node names, hook names, command strings and types. Add
defines for these constants that include the trailing NUL byte. These
new constants have SIZ in their name instead of LEN. As soon as all
consumers in the tree are converted to use the new defines the old
defines will be put under BURN_BRIDGES.

Reviewed by: archie, julian, ru
Approved by: re (in principle)

20 years agoUse __sfvwrite() instead of __sputc() via __fputwc() to write to fake
Tim J. Robbins [Wed, 12 Nov 2003 08:49:12 +0000 (08:49 +0000)]
Use __sfvwrite() instead of __sputc() via __fputwc() to write to fake
string files (__SSTR flag set). This is necessary because __sputc()
does not respect the __SALC flag, and crashes trying to flush the buffer
instead of resizing it.

PR: 59167

20 years agoRemove leftovers from old color determination scheme.
Dag-Erling Smørgrav [Wed, 12 Nov 2003 08:37:10 +0000 (08:37 +0000)]
Remove leftovers from old color determination scheme.
Tune the speed at which colors change.

20 years agoFBSDID style nit.
David E. O'Brien [Wed, 12 Nov 2003 08:33:18 +0000 (08:33 +0000)]
FBSDID style nit.

20 years agobraino in age computation
Dag-Erling Smørgrav [Wed, 12 Nov 2003 08:32:34 +0000 (08:32 +0000)]
braino in age computation

20 years agoUse graded colors to more clearly indicate relative age.
Dag-Erling Smørgrav [Wed, 12 Nov 2003 08:30:51 +0000 (08:30 +0000)]
Use graded colors to more clearly indicate relative age.

20 years agoCorrect the on-disc path to the porters-handbook
Christian Brueffer [Wed, 12 Nov 2003 08:26:08 +0000 (08:26 +0000)]
Correct the on-disc path to the porters-handbook

20 years agoTurn the table around: platforms across, branches down.
Dag-Erling Smørgrav [Wed, 12 Nov 2003 08:16:16 +0000 (08:16 +0000)]
Turn the table around: platforms across, branches down.
Also fix some bogus tabification in here documents.

20 years agoUpdate the five files derived from /sys/kern/syscalls.master
Kirk McKusick [Wed, 12 Nov 2003 08:09:19 +0000 (08:09 +0000)]
Update the five files derived from /sys/kern/syscalls.master
after the additions made for the new statfs structure (version
1.157). These must be updated in a separate checkin after
syscalls.master has been checked in so that they reflect its
new CVS identity. As these are purely derived files, it is not
clear to me why they are under CVS at all. I presume that it has
something to do with having `make world' operate properly.

20 years agoMove cd9660 module from 3rd floppy to 2nd to unbreak release.
Jun Kuriyama [Wed, 12 Nov 2003 08:08:16 +0000 (08:08 +0000)]
Move cd9660 module from 3rd floppy to 2nd to unbreak release.

20 years agoDocument the 'fetch-recursive-list' and 'deinstall-all' targets.
Christian Brueffer [Wed, 12 Nov 2003 08:06:35 +0000 (08:06 +0000)]
Document the 'fetch-recursive-list' and 'deinstall-all' targets.

No MFC reminder, I'll sync this page with the one in -stable in
a couple of days.

PR: docs/46181
Patch submitted by: Sergey Matveychuk <sem@ciam.ru>

20 years agoUpdate the statfs structure with 64-bit fields to allow
Kirk McKusick [Wed, 12 Nov 2003 08:01:40 +0000 (08:01 +0000)]
Update the statfs structure with 64-bit fields to allow
accurate reporting of multi-terabyte filesystem sizes.

You should build and boot a new kernel BEFORE doing a `make world'
as the new kernel will know about binaries using the old statfs
structure, but an old kernel will not know about the new system
calls that support the new statfs structure. Running an old kernel
after a `make world' will cause programs such as `df' that do a
statfs system call to fail with a bad system call.

Reviewed by: Bruce Evans <bde@zeta.org.au>
Reviewed by: Tim Robbins <tjr@freebsd.org>
Reviewed by: Julian Elischer <julian@elischer.org>
Reviewed by: the hoards of <arch@freebsd.org>
Sponsored by:   DARPA & NAI Labs.

20 years agoBelatedly include RELENG_4_9 in the build, and rotate RELENG_4_7 out.
Dag-Erling Smørgrav [Wed, 12 Nov 2003 07:50:36 +0000 (07:50 +0000)]
Belatedly include RELENG_4_9 in the build, and rotate RELENG_4_7 out.

20 years agoMinor diff reduction with p4
Warner Losh [Wed, 12 Nov 2003 05:44:44 +0000 (05:44 +0000)]
Minor diff reduction with p4

20 years agoFix a harmless typo (use of res instead of rle: if rle is NULL, we'd
Warner Losh [Wed, 12 Nov 2003 05:21:06 +0000 (05:21 +0000)]
Fix a harmless typo (use of res instead of rle: if rle is NULL, we'd
still get a panic, just not a nice message) and update to new
__FBSDID.

Submitted by: charnier@

20 years agoavoid module name conflict with opencrypto/rijndael.c.
Hajimu UMEMOTO [Wed, 12 Nov 2003 04:22:37 +0000 (04:22 +0000)]
avoid module name conflict with opencrypto/rijndael.c.

Reported by: tinderbox

20 years agoImprove debug message.
Hidetoshi Shimokawa [Wed, 12 Nov 2003 04:06:21 +0000 (04:06 +0000)]
Improve debug message.

20 years ago- Reserve a ocb for management ORB.
Hidetoshi Shimokawa [Wed, 12 Nov 2003 03:45:10 +0000 (03:45 +0000)]
- Reserve a ocb for management ORB.
- Requeue XPT_SCSI_IO if ocb is short.

20 years agoGC prototype for mac_destroy_vnode_label(), missed in last commit.
Robert Watson [Wed, 12 Nov 2003 03:33:43 +0000 (03:33 +0000)]
GC prototype for mac_destroy_vnode_label(), missed in last commit.

20 years agoOops, fix typo in my name.
Hidetoshi Shimokawa [Wed, 12 Nov 2003 03:29:57 +0000 (03:29 +0000)]
Oops, fix typo in my name.

20 years agoRemove ia64_highfp_load() now that it's unused.
Marcel Moolenaar [Wed, 12 Nov 2003 03:24:34 +0000 (03:24 +0000)]
Remove ia64_highfp_load() now that it's unused.

20 years agoModify the MAC Framework so that instead of embedding a (struct label)
Robert Watson [Wed, 12 Nov 2003 03:14:31 +0000 (03:14 +0000)]
Modify the MAC Framework so that instead of embedding a (struct label)
in various kernel objects to represent security data, we embed a
(struct label *) pointer, which now references labels allocated using
a UMA zone (mac_label.c).  This allows the size and shape of struct
label to be varied without changing the size and shape of these kernel
objects, which become part of the frozen ABI with 5-STABLE.  This opens
the door for boot-time selection of the number of label slots, and hence
changes to the bound on the number of simultaneous labeled policies
at boot-time instead of compile-time.  This also makes it easier to
embed label references in new objects as required for locking/caching
with fine-grained network stack locking, such as inpcb structures.

This change also moves us further in the direction of hiding the
structure of kernel objects from MAC policy modules, not to mention
dramatically reducing the number of '&' symbols appearing in both the
MAC Framework and MAC policy modules, and improving readability.

While this results in minimal performance change with MAC enabled, it
will observably shrink the size of a number of critical kernel data
structures for the !MAC case, and should have a small (but measurable)
performance benefit (i.e., struct vnode, struct socket) do to memory
conservation and reduced cost of zeroing memory.

NOTE: Users of MAC must recompile their kernel and all MAC modules as a
result of this change.  Because this is an API change, third party
MAC modules will also need to be updated to make less use of the '&'
symbol.

Suggestions from: bmilekic
Obtained from: TrustedBSD Project
Sponsored by: DARPA, Network Associates Laboratories

20 years ago1. Consolidate mount struct allocation/destruction into a common code in
Alexander Kabaev [Wed, 12 Nov 2003 02:54:47 +0000 (02:54 +0000)]
1. Consolidate mount struct allocation/destruction into a common code in
vfs_mount_alloc/vfs_mount_destroy functions and take care to completely
destroy the mount point along with its locks. Mount struct has grown in
coplexity recently and depending on each failure path to destroy it
completely isn't working anymore.

2. Eliminate largely identical vfs_mount and vfs_unmount question by
moving the code to handle both cases into a newly introduced vfs_domount
function.

3. Simplify nfs_mount_diskless to always expect an allocated mount
struct and never attempt an allocation/destruction itself. The
vfs_allocroot allocation was there to support 'magic' swap space
configuration for diskless clients that was already removed by PHK some
time ago.

4. Include a vfs_buildopts cleanups by Peter Edwards to validate the
sanity of nmount parameters passed from userland.

Submitted by:  (4) Peter Edwards <peter.edwards@openet-telecom.com>
Reviewed by:    rwatson

20 years agoMdoc Janitor:
Hiten Pandya [Wed, 12 Nov 2003 02:35:20 +0000 (02:35 +0000)]
Mdoc Janitor:

* cleanup hard sentence breaks.

* sprinle some .Dq macros.

20 years agoMdoc Janitor:
Hiten Pandya [Wed, 12 Nov 2003 02:26:47 +0000 (02:26 +0000)]
Mdoc Janitor:

* Add missing `utility' word to sentence describing
  ips(4) card configuration.

* Remove extraneous use of .Pp, and describe the tunable
  hw.ips.0.disable in a better way.

* Replace wrongly used .Op mdoc macros with the .Bq macro.
  The .Op macro should only be used when describing a
  ``usage'' line of a utility/command.

* Add .Er, for marking errno defines (ENOMEM etc etc)

20 years agoCosmetic sync with i386
Peter Wemm [Wed, 12 Nov 2003 01:49:49 +0000 (01:49 +0000)]
Cosmetic sync with i386

20 years agoRemove a gremlin so that this code compiles under -stable without
Josef Karthauser [Wed, 12 Nov 2003 01:40:11 +0000 (01:40 +0000)]
Remove a gremlin so that this code compiles under -stable without
a "syntax error before `struct'" error.

20 years agoFurther work-out the handling of the high FP registers. The most
Marcel Moolenaar [Wed, 12 Nov 2003 01:26:02 +0000 (01:26 +0000)]
Further work-out the handling of the high FP registers. The most
important change is in cpu_switch() where we disable the high FP
registers for the thread that we switch-out if the CPU currently
has its high FP registers. This avoids that the high FP registers
remain enabled for the thread even when the CPU has unloaded them
or the thread migrated to another processor.
Likewise, when we switch-in a thread of that has its high FP
registers on the CPU, we enable them. This avoids an otherwise
harmless, but unnecessary trap to have them enabled.

The code that handles the disabled high FP trap (in trap()) has
been turned into a critical section for the most part to avoid
being preempted. If there's a race, we bail out and have the
processor trap again if necessary.

Avoid using the generic ia64_highfp_save() function when the
context is predictable. The function adds unnecessary overhead.
Don't use ia64_highfp_load() for the same reason. The function
is now unused and can be removed.

These changes make the lazy context switching of the high FP
registers in an UP kernel functional.

20 years agoOh dear, forgot this file in the turnstile commit. This header defines
John Baldwin [Tue, 11 Nov 2003 23:08:26 +0000 (23:08 +0000)]
Oh dear, forgot this file in the turnstile commit.  This header defines
the turnstile API and includes several comments.

Reminded by: peter

20 years agoAdd ID for ALC658 CODEC.
Jun Kuriyama [Tue, 11 Nov 2003 22:15:17 +0000 (22:15 +0000)]
Add ID for ALC658 CODEC.

Tested on: GIGABYTE GA-8S655FX-L
Reviewed by: orion

20 years agoAdd an implementation of turnstiles and change the sleep mutex code to use
John Baldwin [Tue, 11 Nov 2003 22:07:29 +0000 (22:07 +0000)]
Add an implementation of turnstiles and change the sleep mutex code to use
turnstiles to implement blocking isntead of implementing a thread queue
directly.  These turnstiles are somewhat similar to those used in Solaris 7
as described in Solaris Internals but are also different.

Turnstiles do not come out of a fixed-sized pool.  Rather, each thread is
assigned a turnstile when it is created that it frees when it is destroyed.
When a thread blocks on a lock, it donates its turnstile to that lock to
serve as queue of blocked threads.  The queue associated with a given lock
is found by a lookup in a simple hash table.  The turnstile itself is
protected by a lock associated with its entry in the hash table.  This
means that sched_lock is no longer needed to contest on a mutex.  Instead,
sched_lock is only used when manipulating run queues or thread priorities.
Turnstiles also implement priority propagation inherently.

Currently turnstiles only support mutexes.  Eventually, however, turnstiles
may grow two queue's to support a non-sleepable reader/writer lock
implementation.  For more details, see the comments in sys/turnstile.h and
kern/subr_turnstile.c.

The two primary advantages from the turnstile code include: 1) the size
of struct mutex shrinks by four pointers as it no longer stores the
thread queue linkages directly, and 2) less contention on sched_lock in
SMP systems including the ability for multiple CPUs to contend on different
locks simultaneously (not that this last detail is necessarily that much of
a big win).  Note that 1) means that this commit is a kernel ABI breaker,
so don't mix old modules with a new kernel and vice versa.

Tested on: i386 SMP, sparc64 SMP, alpha SMP

20 years ago- Disable AGP on ALI chipsets if aperture size is 0.
Eric Anholt [Tue, 11 Nov 2003 21:49:18 +0000 (21:49 +0000)]
- Disable AGP on ALI chipsets if aperture size is 0.
- Fail in agp_alloc_gatt if the aperture size is 0 instead of panicing in
  contigmalloc.

Reported by: Bjoern Fischer <bfischer@Techfak.Uni-Bielefeld.DE>
Reviewed by: jhb
MFC after: 1 week

20 years agoFollowing the repo-copy from src/sys/modules/mac_none/Makefile,
Robert Watson [Tue, 11 Nov 2003 21:23:54 +0000 (21:23 +0000)]
Following the repo-copy from src/sys/modules/mac_none/Makefile,
I neglected to update the filenames/etc in mac_stub.  Do so now.

20 years agoDon't probe busses in the MP Table for the MP Table PCI bridge drivers
John Baldwin [Tue, 11 Nov 2003 21:19:43 +0000 (21:19 +0000)]
Don't probe busses in the MP Table for the MP Table PCI bridge drivers
if the bus number doesn't correspond to a PCI bus in the MP Table.

Reported by: jhay

20 years agoHelp bmah out and add the ips(4) entity.
Tom Rhodes [Tue, 11 Nov 2003 19:20:13 +0000 (19:20 +0000)]
Help bmah out and add the ips(4) entity.

20 years agocleanup rijndael API.
Hajimu UMEMOTO [Tue, 11 Nov 2003 18:58:54 +0000 (18:58 +0000)]
cleanup rijndael API.
since there are naming conflicts with opencrypto, #define was
added to rename functions intend to avoid conflicts.

Obtained from: KAME

20 years agoAttach ips.4 to the build.
Tom Rhodes [Tue, 11 Nov 2003 18:48:02 +0000 (18:48 +0000)]
Attach ips.4 to the build.

20 years agoAdd a manual page for the ips(4) driver.
Tom Rhodes [Tue, 11 Nov 2003 18:47:01 +0000 (18:47 +0000)]
Add a manual page for the ips(4) driver.

Requested by: obrien
Reviewed by: scottl (older version)

20 years ago - Add some information about how init, securelevel, and jails
Ken Smith [Tue, 11 Nov 2003 18:37:50 +0000 (18:37 +0000)]
- Add some information about how init, securelevel, and jails
  interact with each other.
- Minor markup fix (.Dq -> .Va for a variable)

Reviewed by: rwatson
Approved by: blackend (mentor)

20 years ago - Add a note that there are two MIB variables that have per-jail
Ken Smith [Tue, 11 Nov 2003 18:34:29 +0000 (18:34 +0000)]
- Add a note that there are two MIB variables that have per-jail
  settings.

Reviewed by: rwatson
Approved by: blackend (mentor)

20 years ago - Markup fix-ups (add .Dq, and some hard line breaks at the end
Ken Smith [Tue, 11 Nov 2003 18:31:36 +0000 (18:31 +0000)]
- Markup fix-ups (add .Dq, and some hard line breaks at the end
  of sentences).

Approved by: blackend (mentor)

20 years ago - Add a note about how jail(2) effects the securelevel.
Ken Smith [Tue, 11 Nov 2003 18:21:20 +0000 (18:21 +0000)]
- Add a note about how jail(2) effects the securelevel.

Reviewed by: rwatson
Approved by: blackend (mentor)

20 years agoSome motherboards like to remap the SCI (normally IRQ 9) up to a PCI
John Baldwin [Tue, 11 Nov 2003 18:20:10 +0000 (18:20 +0000)]
Some motherboards like to remap the SCI (normally IRQ 9) up to a PCI
interrupt such as IRQ 22 or 19.  However, the ACPI BIOS still routes
interrupts from some PCI devices to the same intpin calling the pin
IRQ 22.  Thus, ACPI expects to address a single interrupt source via two
different names.  To work around this, if the SCI is remapped to a non-ISA
interrupt (i.e., greater than 15), then we use
acpi_OverrideInterruptLevel() function to tell ACPI to use IRQ 22 or 19
rather than IRQ 9 for the SCI.

Previously we would change IRQ 22 or 19's name to IRQ 9 when we encountered
such an Interrupt Source Override entry in the MADT which routed the SCI
properly but left PCI devices mapped to IRQ 22 or 19 w/o a routable
interrupt.

Tested by: sos

20 years agocorrect typos
Sam Leffler [Tue, 11 Nov 2003 18:16:54 +0000 (18:16 +0000)]
correct typos

Pointed out by: Mike Silbersack

20 years agoAdd an acpi_OverrideInterruptLevel() method that OSPM can use to override
John Baldwin [Tue, 11 Nov 2003 18:12:12 +0000 (18:12 +0000)]
Add an acpi_OverrideInterruptLevel() method that OSPM can use to override
the InterruptLevel used for the SCI.