]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
20 years agoToday, RealTek sent me a driver to test which had been compiled with
Bill Paul [Mon, 2 Aug 2004 18:54:01 +0000 (18:54 +0000)]
Today, RealTek sent me a driver to test which had been compiled with
some debug support turned on. It turns out the sections in this driver
binary had relative virtual addresses (RVAs) that were different
from the raw addresses, and it had a .data section where the virtual size
was much larger than the raw size. (Most production binaries produced
with the Microsoft DDK have RVA == PA.)

There's code in the ndiscvt(8) utility that's supposed to handle
the vsize != rsize case, but it turns out it was slightly broken,
and it failed to handle the RVA != RA case at all. Hopefully, this
commit will fix all that.

20 years agoBack out 1.388.
David E. O'Brien [Mon, 2 Aug 2004 18:48:14 +0000 (18:48 +0000)]
Back out 1.388.
Demanded by: jhb

20 years agoRemove GPT_ENT_TYPE_FREEBSD_UFS2. It was speculatively added before
Marcel Moolenaar [Mon, 2 Aug 2004 18:46:52 +0000 (18:46 +0000)]
Remove GPT_ENT_TYPE_FREEBSD_UFS2. It was speculatively added before
UFS2 was here. It so happened that UFS2 did not need a seperate
partition type. Keep the definition as a comment for documentation
purposes. If there is a benefit for UFS2 file systems to have a
seperate partition type under GPT, then this definition should be
restored as that was the intention of the definition.

20 years agoFix 2 typos in previous commit: both s/strct/struct/
Marcel Moolenaar [Mon, 2 Aug 2004 18:37:55 +0000 (18:37 +0000)]
Fix 2 typos in previous commit: both s/strct/struct/

20 years agoNew release notes (changes in the first half of July 2004):
Hiroki Sato [Mon, 2 Aug 2004 18:24:58 +0000 (18:24 +0000)]
New release notes (changes in the first half of July 2004):
bus_dma(9) alignment and boundary compensation support,
kqueue(2) EVFILT_FS,
KDB framework,
sound and snd_* driver reorganization,
natd(8) globalports option[*],
ppp(8) LQM support[*] and "rad_alive N" option,
GEOM_STRIPE FAST mode support,
MSDOSFS_LARGE[*],
ALTQ framework support added to various network drivers[*],
bsdtar(1) now the default tar(1) utility,
cvs(1) iso8601 option keyword,
multibyte characters support:
fgetwln(3), join(1), nextwctype(3), od(1), regex(3), rev(1),
sed(1) 'y' command, and tr(1),
ftw(3) and nftw(3) implemented,
C99 functions: nearbyint(3),
am-utils 6.0.9->6.0.10p1 import,
GNU grep 2.4d->2.5.1 import, and
tcsh 6.11->6.13.00 import.

MFC:
IPFilter 3.4.31->3.4.35.

[*]Based on work by: josef

20 years agoDon't build and install a mem.ko module.
David E. O'Brien [Mon, 2 Aug 2004 18:14:06 +0000 (18:14 +0000)]
Don't build and install a mem.ko module.
Currently one cannot load the mem.ko module without panicing if mem is
compiled into the kernel and one cannot build a kernel w/o "device mem"
right now either.  Thus it is too dangerous to install mem.ko right now
because if one puts 'mem_load="YES"' in /etc/loader.conf they cannot
boot an "old" kernel (at the time that a kernel doesn't have to be built
with "device mem).

20 years agoAdd the mem and null devices now that they are optional.
Marcel Moolenaar [Mon, 2 Aug 2004 17:53:06 +0000 (17:53 +0000)]
Add the mem and null devices now that they are optional.

20 years agoSort the miscellaneous devices to restore ordering after the insertion
Marcel Moolenaar [Mon, 2 Aug 2004 17:50:39 +0000 (17:50 +0000)]
Sort the miscellaneous devices to restore ordering after the insertion
of the mem and null devices.

20 years agoMake the USB subsystem unloadable and detachable, though currently
Ian Dowse [Mon, 2 Aug 2004 15:37:35 +0000 (15:37 +0000)]
Make the USB subsystem unloadable and detachable, though currently
a significant amount of memory may be leaked each time a host
controller is detached.

20 years agoOptimize intr_execute_handlers() by combining the pic_disable_source() and
Scott Long [Mon, 2 Aug 2004 15:31:10 +0000 (15:31 +0000)]
Optimize intr_execute_handlers() by combining the pic_disable_source() and
pic_eoi_source() into one call.  This halves the number of spinlock operations
and indirect function calls in the normal case of handling a normal (ithread)
interrupt.  Optimize the atpic and ioapic drivers to use inlines where
appropriate in supporting the intr_execute_handlers() change.

This knocks 900ns, or roughly 1350 cycles, off of the time spent servicing an
interrupt in the common case on my 1.5GHz P4 uniprocessor system.  SMP systems
likely won't see as much of a gain due to the ioapic being more efficient than
the atpic.  I'll investigate porting this to amd64 soon.

Reviewed by: jhb

20 years agoWhen searching for a suitable block of memory on the free list,
Ian Dowse [Mon, 2 Aug 2004 13:59:02 +0000 (13:59 +0000)]
When searching for a suitable block of memory on the free list,
skip blocks that are too big by a factor of two or greater. This
avoids some cases of extremely inefficient memory use that can occur
when large (e.g. 64k) blocks on the free list get used when allocating
a 4k chunk of 64-byte fragments. Because fragments have their own
free list, the 60k difference got lost forever every time.

20 years agoAttempt to follow the correct procedure for synchronising with the
Ian Dowse [Mon, 2 Aug 2004 12:56:01 +0000 (12:56 +0000)]
Attempt to follow the correct procedure for synchronising with the
system BIOS to disable legacy device emulation as per the "EHCI
Extended Capability: Pre-OS to OS Handoff Synchronisation" section
of the EHCI spec. BIOSes that implement legacy emulation using SMIs
are supposed to disable the emulation when this procedure is performed.

20 years agoRemove la_LN.* from the list of bogus locales. They're incomplete, but
Tim J. Robbins [Mon, 2 Aug 2004 12:50:12 +0000 (12:50 +0000)]
Remove la_LN.* from the list of bogus locales. They're incomplete, but
still potentially useful.

20 years agoExclude bogus la_LN.* and UTF-8 locales from the output of locale -a
Tim J. Robbins [Mon, 2 Aug 2004 12:28:28 +0000 (12:28 +0000)]
Exclude bogus la_LN.* and UTF-8 locales from the output of locale -a
to discourage people from using them.

20 years ago*blush*
Olivier Houchard [Mon, 2 Aug 2004 12:24:18 +0000 (12:24 +0000)]
*blush*
Fix htonl and htons.

20 years agoFix comments.
Olivier Houchard [Mon, 2 Aug 2004 12:23:53 +0000 (12:23 +0000)]
Fix comments.

Spotted out by: mux

20 years agoAdd cross-reference to fmt(1) and a fairly standard ENVIRONMENT section.
Tim J. Robbins [Mon, 2 Aug 2004 11:15:01 +0000 (11:15 +0000)]
Add cross-reference to fmt(1) and a fairly standard ENVIRONMENT section.

20 years agoCross-reference fold(1).
Tim J. Robbins [Mon, 2 Aug 2004 11:12:13 +0000 (11:12 +0000)]
Cross-reference fold(1).

20 years agoAdd support for multibyte characters.
Tim J. Robbins [Mon, 2 Aug 2004 11:10:20 +0000 (11:10 +0000)]
Add support for multibyte characters.

20 years agoRemove 'device mem' from GENERIC, which markm@ mistakingly added.
Suleiman Souhlal [Mon, 2 Aug 2004 11:08:48 +0000 (11:08 +0000)]
Remove 'device mem' from GENERIC, which markm@ mistakingly added.
We don't have mem/kmem yet.

Approved by: grehan (mentor)

20 years agoUnbreak DEVICE_POLLING build / LINT. Sorry!
Max Laier [Mon, 2 Aug 2004 10:08:28 +0000 (10:08 +0000)]
Unbreak DEVICE_POLLING build / LINT. Sorry!

Submitted by: roam

20 years agoDon't use version number in library name. The library version is checked
Pawel Jakub Dawidek [Mon, 2 Aug 2004 09:05:29 +0000 (09:05 +0000)]
Don't use version number in library name. The library version is checked
after dlopen() anyway, so we should be safe.

Suggested by: ru

20 years agoRemove an implicit int parameter by using prototypes.
Stefan Farfeleder [Mon, 2 Aug 2004 08:46:23 +0000 (08:46 +0000)]
Remove an implicit int parameter by using prototypes.

20 years agoDon't pass function pointers via a void * parameter.
Stefan Farfeleder [Mon, 2 Aug 2004 08:18:43 +0000 (08:18 +0000)]
Don't pass function pointers via a void * parameter.

20 years ago- Signal handlers must have an int argument.
Stefan Farfeleder [Mon, 2 Aug 2004 08:10:28 +0000 (08:10 +0000)]
- Signal handlers must have an int argument.
- Use prototypes.
- Add a cast for a signed vs unsigned comparison.
- Mark as WARNS?=3 clean.

20 years agoImprove the wording of the last commit, and update the document date. (1)
Murray Stokely [Mon, 2 Aug 2004 07:28:02 +0000 (07:28 +0000)]
Improve the wording of the last commit, and update the document date. (1)
While here, update an example.

(1) Submitted by: ru

20 years agoConsistently list _CPUCFLAGS.
David E. O'Brien [Mon, 2 Aug 2004 04:19:22 +0000 (04:19 +0000)]
Consistently list _CPUCFLAGS.

20 years agoEliminate the acquisition and release of Giant around the call to
Alan Cox [Mon, 2 Aug 2004 03:31:05 +0000 (03:31 +0000)]
Eliminate the acquisition and release of Giant around the call to
pmap_mincore() in mincore(2).  Either pmap locking exists (alpha, amd64,
i386, ia64) or pmap_mincore() is unimplemented (arm, powerpc, sparc64).

20 years agoDocument incorrect handling of multibyte characters with -I and -J options.
Tim J. Robbins [Mon, 2 Aug 2004 03:07:42 +0000 (03:07 +0000)]
Document incorrect handling of multibyte characters with -I and -J options.

20 years agoAdd PowerPC bridge instruction flag to the assembler to help out
Peter Grehan [Mon, 2 Aug 2004 03:06:21 +0000 (03:06 +0000)]
Add PowerPC bridge instruction flag to the assembler to help out
with test code written in the loader.

20 years agoG5 support: handle the case where the OpenFirmware memory array uses
Peter Grehan [Mon, 2 Aug 2004 03:05:09 +0000 (03:05 +0000)]
G5 support: handle the case where the OpenFirmware memory array uses
64 bits for the phys address, but only 32 for the virtual address.

20 years agoKernel traps were not being passed to trap_fatal in some
Peter Grehan [Mon, 2 Aug 2004 02:37:29 +0000 (02:37 +0000)]
Kernel traps were not being passed to trap_fatal in some
circumstances.

Spotted by:  gallatin

20 years agoDocument machdep.enable_panic_key.
Scott Long [Mon, 2 Aug 2004 02:07:56 +0000 (02:07 +0000)]
Document machdep.enable_panic_key.

Submitted by: Craig Rodrigues

20 years agoFix the build by providing 'PHYS_TO_DMAP' and 'M_MEMDESC'.
David E. O'Brien [Mon, 2 Aug 2004 02:07:20 +0000 (02:07 +0000)]
Fix the build by providing 'PHYS_TO_DMAP' and 'M_MEMDESC'.

20 years agoBump __FreeBSD_version to designate uma_zone functions changing type.
Brian Feldman [Mon, 2 Aug 2004 01:49:22 +0000 (01:49 +0000)]
Bump __FreeBSD_version to designate uma_zone functions changing type.

20 years agoAdd what appears to be a missing '*/' at the end of a comment.
Robert Watson [Mon, 2 Aug 2004 01:38:27 +0000 (01:38 +0000)]
Add what appears to be a missing '*/' at the end of a comment.

20 years ago- Fix unloading by the same way it is done in my other classes:
Pawel Jakub Dawidek [Mon, 2 Aug 2004 00:37:40 +0000 (00:37 +0000)]
- Fix unloading by the same way it is done in my other classes:
  set gp->softc to NULL and return ENXIO when it is NULL, so GEOM
  will not panic or hang, but unload one device on every 'unload'.
  This make 'unload' command usable, but it have to be executed
  <number of devices> + 1 times.
- Made use of 'pp' variable.

20 years ago* Add a "how" argument to uma_zone constructors and initialization functions
Brian Feldman [Mon, 2 Aug 2004 00:18:36 +0000 (00:18 +0000)]
* Add a "how" argument to uma_zone constructors and initialization functions
  so that they know whether the allocation is supposed to be able to sleep
  or not.
* Allow uma_zone constructors and initialation functions to return either
  success or error.  Almost all of the ones in the tree currently return
  success unconditionally, but mbuf is a notable exception: the packet
  zone constructor wants to be able to fail if it cannot suballocate an
  mbuf cluster, and the mbuf allocators want to be able to fail in general
  in a MAC kernel if the MAC mbuf initializer fails.  This fixes the
  panics people are seeing when they run out of memory for mbuf clusters.
* Allow debug.nosleepwithlocks on WITNESS to be disabled, without changing
  the default.

Both bmilekic and jeff have reviewed the changes made to make failable
zone allocations work.

20 years agoSecond part of ALTQ driver modifications, covering:
Max Laier [Sun, 1 Aug 2004 23:58:04 +0000 (23:58 +0000)]
Second part of ALTQ driver modifications, covering:
an(4), ath(4), hme(4), ndis(4), vr(4) and wi(4)

Please help testing: http://people.freebsd.org/~mlaier/ALTQ_driver/

Tested by: Vaidas Damosevicius (an, ath, wi)
Roman Divacky (vr)
Submitted by: yongari (hme)

20 years agoComment kse_create() and make a few minor code cleanups
Julian Elischer [Sun, 1 Aug 2004 23:02:00 +0000 (23:02 +0000)]
Comment kse_create() and make a few minor code cleanups

Reviewed by: davidxu

20 years agoRemove unnecessary use of the __ia64__ conditional. This slightly improves
Marcel Moolenaar [Sun, 1 Aug 2004 22:44:40 +0000 (22:44 +0000)]
Remove unnecessary use of the __ia64__ conditional. This slightly improves
maintainability and generally avoids confusion.

20 years agoThe watchdog callout executes with the (non-sleepable) ifnet lock held
Bill Paul [Sun, 1 Aug 2004 22:25:12 +0000 (22:25 +0000)]
The watchdog callout executes with the (non-sleepable) ifnet lock held
now, but it's possible for ndis_reset_nic() to sleep (sometimes the
MiniportReset() method returns NDIS_STATUS_PENDING and we have
to wait for completion). To get around this, execute the ndis_reset_nic()
routine in the NDIS_TASKQUEUE thread.

20 years agoAfter changing LIBDIR to SHLIBDIR, because of dependencies problems,
Pawel Jakub Dawidek [Sun, 1 Aug 2004 22:24:07 +0000 (22:24 +0000)]
After changing LIBDIR to SHLIBDIR, because of dependencies problems,
new problem shows up: symblic links (<libname>.so) are created under
/usr/lib/ now, instead of under /lib/geom/ where geom(8) looks for them.
Introduce a workaround to fix this by teaching geom(8) to open libraries
via /lib/geom/<libname>.so.<major_number> instead of /lib/geom/<libname>.so.

20 years agoUUCP's uucico(8) has not been in the base system for some time now,
Mark Murray [Sun, 1 Aug 2004 21:33:47 +0000 (21:33 +0000)]
UUCP's uucico(8) has not been in the base system for some time now,
so reflect this in the default. The uucp uid is a bit funny, and
is used by mtree in /var/spool for locks, so we can't remove it
without thinking about it a bit harder.

20 years agoIn ndis_alloc_bufpool() and ndis_alloc_packetpool(), the test to see if
Bill Paul [Sun, 1 Aug 2004 21:15:29 +0000 (21:15 +0000)]
In ndis_alloc_bufpool() and ndis_alloc_packetpool(), the test to see if
allocating pool memory succeeded was checking the wrong pointer (should
have been looking at *pool, not pool). Corrected this.

20 years agoThis commit was generated by cvs2svn to compensate for changes in r132977,
Garance A Drosehn [Sun, 1 Aug 2004 20:45:54 +0000 (20:45 +0000)]
This commit was generated by cvs2svn to compensate for changes in r132977,
which included commits to RCS files with non-trunk default branches.

20 years agoImport of a BSD-licensed version of `patch', which will eventually
Garance A Drosehn [Sun, 1 Aug 2004 20:45:54 +0000 (20:45 +0000)]
Import of a BSD-licensed version of `patch', which will  eventually
replace the version we currently have in src/gnu/usr.bin/patch/.
Among other things, this version includes a --posix option for strict
POSIX conformance.

This version is the current source from OpenBSD as of today.  It is
their 3.5-release, plus a few updates to patch.c and pch.c that they
made about three weeks ago.

20 years agoTypo.
Pawel Jakub Dawidek [Sun, 1 Aug 2004 20:41:58 +0000 (20:41 +0000)]
Typo.

20 years agoAnother stupid error from my side. PPPOE_NONSTANDARD was first defined
Gleb Smirnoff [Sun, 1 Aug 2004 20:39:33 +0000 (20:39 +0000)]
Another stupid error from my side. PPPOE_NONSTANDARD was first defined
in enum {}, and then redefined with #define.
No warnings from compiler, though.

Submitted by: bz
Pointy hat to: glebius

20 years agoCorrect the explanation of the -X option.
Tim Kientzle [Sun, 1 Aug 2004 20:09:08 +0000 (20:09 +0000)]
Correct the explanation of the -X option.
Thanks to: Pav Lucistnik

20 years agoBig mess 'o changes:
Bill Paul [Sun, 1 Aug 2004 20:04:31 +0000 (20:04 +0000)]
Big mess 'o changes:

- Give ndiscvt(8) the ability to process a .SYS file directly into
  a .o file so that we don't have to emit big messy char arrays into
  the ndis_driver_data.h file. This behavior is currently optional, but
  may become the default some day.

- Give ndiscvt(8) the ability to turn arbitrary files into .ko files
  so that they can be pre-loaded or kldloaded. (Both this and the
  previous change involve using objcopy(1)).

- Give NdisOpenFile() the ability to 'read' files out of kernel memory
  that have been kldloaded or pre-loaded, and disallow the use of
  the normal vn_open() file opening method during bootstrap (when no
  filesystems have been mounted yet). Some people have reported that
  kldloading if_ndis.ko works fine when the system is running multiuser
  but causes a panic when the modile is pre-loaded by /boot/loader. This
  happens with drivers that need to use NdisOpenFile() to access
  external files (i.e. firmware images). NdisOpenFile() won't work
  during kernel bootstrapping because no filesystems have been mounted.
  To get around this, you can now do the following:

        o Say you have a firmware file called firmware.img
        o Do: ndiscvt -f firmware.img -- this creates firmware.img.ko
        o Put the firmware.img.ko in /boot/kernel
        o add firmware.img_load="YES" in /boot/loader.conf
        o add if_ndis_load="YES" and ndis_load="YES" as well

  Now the loader will suck the additional file into memory as a .ko. The
  phony .ko has two symbols in it: filename_start and filename_end, which
  are generated by objcopy(1). ndis_open_file() will traverse each module
  in the module list looking for these symbols and, if it finds them, it'll
  use them to generate the file mapping address and length values that
  the caller of NdisOpenFile() wants.

  As a bonus, this will even work if the file has been statically linked
  into the kernel itself, since the "kernel" module is searched too.
  (ndiscvt(8) will generate both filename.o and filename.ko for you).

- Modify the mechanism used to provide make-pretend FASTCALL support.
  Rather than using inline assembly to yank the first two arguments
  out of %ecx and %edx, we now use the __regparm__(3) attribute (and
  the __stdcall__ attribute) and use some macro magic to re-order
  the arguments and provide dummy arguments as needed so that the
  arguments passed in registers end up in the right place. Change
  taken from DragonflyBSD version of the NDISulator.

20 years agoAdd the I/O device for those architectures that have it.
Mark Murray [Sun, 1 Aug 2004 19:37:34 +0000 (19:37 +0000)]
Add the I/O device for those architectures that have it.

20 years agoFor the "portable" distribution, the configure script will overwrite
Tim Kientzle [Sun, 1 Aug 2004 19:30:56 +0000 (19:30 +0000)]
For the "portable" distribution, the configure script will overwrite
"Makefile," so I'm moving all the FreeBSD build machinery to
"Makefile.freebsd", with the default "Makefile" containing a single
include.

20 years agoAdd a HARDWARE section which lists supported devices.
Simon L. B. Nielsen [Sun, 1 Aug 2004 19:29:21 +0000 (19:29 +0000)]
Add a HARDWARE section which lists supported devices.

20 years agoMake the HARDWARE section better suited to the upcoming auto generated
Simon L. B. Nielsen [Sun, 1 Aug 2004 19:26:42 +0000 (19:26 +0000)]
Make the HARDWARE section better suited to the upcoming auto generated
Hardware Notes:

- Only include text related the device listings and hardware support
  in the HARDWARE section.
- Make the HARDWARE section preamble text have a call to the Nm macro,
  so the driver name will appear in the Hardware Notes.
- Add the manufacturer name to each item in the device list, where
  appropriate.
- Clean trailing punctuation characters from the lists.

20 years ago- Add a HARDWARE section which lists supported devices.
Simon L. B. Nielsen [Sun, 1 Aug 2004 19:08:04 +0000 (19:08 +0000)]
- Add a HARDWARE section which lists supported devices.
- Minor cleanup of the device lists when I'm here anyway.

20 years agoC standard does not permit empty initializer list.
Tim Kientzle [Sun, 1 Aug 2004 19:02:49 +0000 (19:02 +0000)]
C standard does not permit empty initializer list.
Thanks to: Stefan Farfeleder

20 years agoInclude <stdlib.h> for exit(), use prototypes and bump WARNS to 6.
Stefan Farfeleder [Sun, 1 Aug 2004 18:52:40 +0000 (18:52 +0000)]
Include <stdlib.h> for exit(), use prototypes and bump WARNS to 6.

20 years agoRemove extraneous ';'.
Mark Murray [Sun, 1 Aug 2004 18:51:44 +0000 (18:51 +0000)]
Remove extraneous ';'.

20 years agoImplement basic support for EHCI interrupt pipes. This is unlikely
Ian Dowse [Sun, 1 Aug 2004 18:47:42 +0000 (18:47 +0000)]
Implement basic support for EHCI interrupt pipes. This is unlikely
to be particularly correct or optimal, but it seems to be enough
to allow the attachment of USB2 hubs and USB2 devices connected via
USB2 hubs. None of the split transaction support is implemented in
our USB stack, so USB1 peripherals will definitely not work when
connected via USB2 hubs.

20 years agoYA oops. Remove code that was being tested locally.
Mark Murray [Sun, 1 Aug 2004 18:22:44 +0000 (18:22 +0000)]
YA oops. Remove code that was being tested locally.

20 years agoRemove local hack that was not supposed to be committed.
Mark Murray [Sun, 1 Aug 2004 18:12:25 +0000 (18:12 +0000)]
Remove local hack that was not supposed to be committed.

Spotted by: Antoine Brodin - antoine dot brodin at laposte dot net

20 years agoTurn off PREEMPTION by default while it gets debugged. It's been causing
Scott Long [Sun, 1 Aug 2004 14:31:45 +0000 (14:31 +0000)]
Turn off PREEMPTION by default while it gets debugged.  It's been causing
4 weeks of problems including deadlocks and instant panics.  Note that the
real bugs are likely in the scheduler.

20 years agoMFi386: revision 1.1167
Yoshihiro Takahashi [Sun, 1 Aug 2004 13:01:36 +0000 (13:01 +0000)]
MFi386: revision 1.1167

20 years agoMFi386: revision 1.501.
Yoshihiro Takahashi [Sun, 1 Aug 2004 13:00:04 +0000 (13:00 +0000)]
MFi386: revision 1.501.

20 years agoChange the default to switch on DMA on ATAPI devices if they can
Søren Schmidt [Sun, 1 Aug 2004 12:31:38 +0000 (12:31 +0000)]
Change the default to switch on DMA on ATAPI devices if they can
do UDMA2 (ATA33) mode and beyond.

20 years agoAnnounce the memory device module update.
Mark Murray [Sun, 1 Aug 2004 11:46:00 +0000 (11:46 +0000)]
Announce the memory device module update.

20 years agoBreak out the MI part of the /dev/[k]mem and /dev/io drivers into
Mark Murray [Sun, 1 Aug 2004 11:40:54 +0000 (11:40 +0000)]
Break out the MI part of the /dev/[k]mem and /dev/io drivers into
their own directory and module, leaving the MD parts in the MD
area (the MD parts _are_ part of the modules). /dev/mem and /dev/io
are now loadable modules, thus taking us one step further towards
a kernel created entirely out of modules. Of course, there is nothing
preventing the kernel from having these statically compiled.

20 years agoComment some of the 'io' functions.
Alfred Perlstein [Sun, 1 Aug 2004 09:19:41 +0000 (09:19 +0000)]
Comment some of the 'io' functions.

20 years ago- Launch main provider when there are no more disks in NEW state.
Pawel Jakub Dawidek [Sun, 1 Aug 2004 09:01:50 +0000 (09:01 +0000)]
- Launch main provider when there are no more disks in NEW state.
- Log syncid bump at debug level 1.

20 years agoAdd some minor changes related to PCMCIA attribute memory mapping
Bill Paul [Sun, 1 Aug 2004 06:42:44 +0000 (06:42 +0000)]
Add some minor changes related to PCMCIA attribute memory mapping
(which I apparently forgot to commit earlier).

Acquire NDIS_LOCK() in ndis_linksts_done().

20 years agoDocument EXTSRCDIR.
Murray Stokely [Sun, 1 Aug 2004 06:36:03 +0000 (06:36 +0000)]
Document EXTSRCDIR.

Submitted by: Pawel Worach <pawel.worach@telia.com>

20 years agoAdd missing thread suspension/resumption code, fix a bug in pt_thr_sstep,
David Xu [Sun, 1 Aug 2004 04:57:04 +0000 (04:57 +0000)]
Add missing thread suspension/resumption code, fix a bug in pt_thr_sstep,
don't overwrite other debug flags.

20 years agoAdd code to implement register structure converter.
David Xu [Sun, 1 Aug 2004 02:08:39 +0000 (02:08 +0000)]
Add code to implement register structure converter.

20 years agoUpdate release notes to reflect sendmail 8.13.1 import
Gregory Neil Shapiro [Sun, 1 Aug 2004 01:20:44 +0000 (01:20 +0000)]
Update release notes to reflect sendmail 8.13.1 import

20 years agoUpdate notes after sendmail 8.13.1 import
Gregory Neil Shapiro [Sun, 1 Aug 2004 01:18:40 +0000 (01:18 +0000)]
Update notes after sendmail 8.13.1 import

20 years agoReflect changes in sendmail 8.13 source tree
Gregory Neil Shapiro [Sun, 1 Aug 2004 01:16:59 +0000 (01:16 +0000)]
Reflect changes in sendmail 8.13 source tree

20 years agoResolve conflicts from sendmail 8.13.1 import
Gregory Neil Shapiro [Sun, 1 Aug 2004 01:16:16 +0000 (01:16 +0000)]
Resolve conflicts from sendmail 8.13.1 import

20 years agoThis commit was generated by cvs2svn to compensate for changes in r132943,
Gregory Neil Shapiro [Sun, 1 Aug 2004 01:04:57 +0000 (01:04 +0000)]
This commit was generated by cvs2svn to compensate for changes in r132943,
which included commits to RCS files with non-trunk default branches.

20 years agoImport sendmail 8.13.1
Gregory Neil Shapiro [Sun, 1 Aug 2004 01:04:57 +0000 (01:04 +0000)]
Import sendmail 8.13.1

20 years agoSpecify the locking for some proc fields
Julian Elischer [Sat, 31 Jul 2004 23:51:04 +0000 (23:51 +0000)]
Specify the locking for some proc fields

20 years agoIf there are no valid components after the timeout, just destroy device.
Pawel Jakub Dawidek [Sat, 31 Jul 2004 22:10:51 +0000 (22:10 +0000)]
If there are no valid components after the timeout, just destroy device.
There is probably nothing to wait for.

20 years agoPropagate size changes upwards.
Lukas Ertl [Sat, 31 Jul 2004 21:34:21 +0000 (21:34 +0000)]
Propagate size changes upwards.

20 years agoFix a stupid error in my previous commit, which broke operation
Gleb Smirnoff [Sat, 31 Jul 2004 21:32:55 +0000 (21:32 +0000)]
Fix a stupid error in my previous commit, which broke operation
of many nodes.

Pointy hat to: glebius

20 years agoHandle spoil event in dedicated function: g_mirror_spoiled().
Pawel Jakub Dawidek [Sat, 31 Jul 2004 21:08:17 +0000 (21:08 +0000)]
Handle spoil event in dedicated function: g_mirror_spoiled().
The different between the new function and g_mirror_orphan() (which was
used previously) is that syncid is bumped immediately, instead of on
first write, because when consumer was spoiled, it means, that its
provider was opened for writing, so we can't trust that its data
will be valid when it will be connected again.

20 years agoAdd copyright notices.
Max Khon [Sat, 31 Jul 2004 20:47:57 +0000 (20:47 +0000)]
Add copyright notices.

Prodded by: imp

20 years agoExpand the license referenced indirectly inline.
Warner Losh [Sat, 31 Jul 2004 18:49:53 +0000 (18:49 +0000)]
Expand the license referenced indirectly inline.

20 years agoHelp simon out and add a HARDWARE section.
Tom Rhodes [Sat, 31 Jul 2004 15:14:28 +0000 (15:14 +0000)]
Help simon out and add a HARDWARE section.

20 years agoAllow for capital letters as size suffixes.
Pawel Jakub Dawidek [Sat, 31 Jul 2004 15:13:08 +0000 (15:13 +0000)]
Allow for capital letters as size suffixes.

Inspired by: le
Approved by: green (maintainer)

20 years agoftpd(8) seems to be WARNS=2 clean now.
Yaroslav Tykhiy [Sat, 31 Jul 2004 15:07:33 +0000 (15:07 +0000)]
ftpd(8) seems to be WARNS=2 clean now.

Tested on: i386, ia64, amd64, sparc64, alpha

20 years agoChange ``(foo *)0'' to ``NULL'' where it's possible
Yaroslav Tykhiy [Sat, 31 Jul 2004 15:03:17 +0000 (15:03 +0000)]
Change ``(foo *)0'' to ``NULL'' where it's possible
(and it appears possible throughout ftpd(8) source.)

It is not a mere issue of style: Null pointers in C
seem to have been mistaken one way or another quite often.

20 years agoKill a small herd of casts to off_t where they were not needed.
Yaroslav Tykhiy [Sat, 31 Jul 2004 14:46:41 +0000 (14:46 +0000)]
Kill a small herd of casts to off_t where they were not needed.
Thank Fortune, the C compiler can figure out by itself the proper
conversion for assignments, comparisons, and prototyped function
arguments.

20 years agoPrintf(3) off_t values through conversion to intmax_t since
Yaroslav Tykhiy [Sat, 31 Jul 2004 14:22:02 +0000 (14:22 +0000)]
Printf(3) off_t values through conversion to intmax_t since
we've got <stdint.h> et al now.  (This makes ftpd(8) WARNS=2 clean.)

20 years agoSave context in kernel fashion, so it can be restored by
David Xu [Sat, 31 Jul 2004 14:18:26 +0000 (14:18 +0000)]
Save context in kernel fashion, so it can be restored by
kse_switchin syscall.

20 years agoRemove unused field.
David Xu [Sat, 31 Jul 2004 14:14:55 +0000 (14:14 +0000)]
Remove unused field.

20 years agoKill an unused variable (heading to WARNS=2.)
Yaroslav Tykhiy [Sat, 31 Jul 2004 14:03:59 +0000 (14:03 +0000)]
Kill an unused variable (heading to WARNS=2.)

20 years agoConvert a couple of bogus null statements to the right form.
Yaroslav Tykhiy [Sat, 31 Jul 2004 14:03:14 +0000 (14:03 +0000)]
Convert a couple of bogus null statements to the right form.
(Heading to WARNS=2.)

20 years agoTurn on PCB_FULLCTX for set_mcontext, functions like kse_switchin
David Xu [Sat, 31 Jul 2004 14:02:29 +0000 (14:02 +0000)]
Turn on PCB_FULLCTX for set_mcontext, functions like kse_switchin
needs to fully restore asynchronous context which did not come
from fast syscall.

20 years agoRemove unused field.
Pawel Jakub Dawidek [Sat, 31 Jul 2004 13:03:38 +0000 (13:03 +0000)]
Remove unused field.

20 years agoDestroy synchronization geom immediately. This should fix unloading without
Pawel Jakub Dawidek [Sat, 31 Jul 2004 11:22:03 +0000 (11:22 +0000)]
Destroy synchronization geom immediately. This should fix unloading without
stopping all mirrors.