itifg

File Release Notes and Changelog

Release Name: itifg-8.2.2-0.1

Notes:
RPM release of itifg-8.2.2-0.

Since the last released rpm, the setup has been reorganised quite a bit.
Notably, only the kernel subpackage will now get a specific target
architecture like i686; the base and development packages will have the
default architecture setting - i386 on a IA32 system. This more closely
reflects the actual compatibility of the software - the base will generally
work unmodified across similar platforms, while the actual module is
highly hardware (and OS release) specific.

Note that the release number of the RPM is set to the number after '-' in
the source code name followed by '.' and a number identifying the rpm
build.

Changes:
This is log of RPM spec file releases since the last official rpm build:

revision 8.14
date: 2005/08/22 12:47:24;  author: toralf;  state: Exp;  lines: +5 -8
itifg setup updated for build with source code release 8.2.2-0. As usual,
the patches had to be adjusted a bit. The Linux 2.6 build param fixes
(check CONFIG_REGPARM etc.) are now in the official source, so an equivalent
of itifg-8.2.1_segfault.patch is not necessary.
----------------------------
revision 8.13
date: 2005/07/11 11:34:48;  author: toralf;  state: Exp;  lines: +2 -2
DEBUG_ITIFG value changed to 7 = 111 binary. Including bit 2 (mask
0x04) generally makes the debug module more useful, as it allows us to
pin down problems more accurately; the bit enables messages indicating
more exactly what the module is doing at a given time (or what it was
doing at the point of failure.)
----------------------------
revision 8.12
date: 2005/07/11 11:04:58;  author: toralf;  state: Exp;  lines: +2 -3
Adjustments to device names + package selection logic - and working
dir on recursive build.
----------------------------
revision 8.11
date: 2005/04/07 10:10:53;  author: toralf;  state: Exp;  lines: +5 -1
New patch: Fixes various bugs in ioctl setup.
----------------------------
revision 8.10
date: 2005/04/06 09:29:08;  author: toralf;  state: Exp;  lines: +19 -12
Patches reorganised - the "segfault" patch now is (must be) applied *before*
the "config" one. That's to get a cleaner variant of the former in order to
try to get it applied in the official sources.
----------------------------
revision 8.9
date: 2005/04/04 10:47:55;  author: toralf;  state: Exp;  lines: +4 -2
Somewhat experimental support for device setup via normal MAKEDEV instead
of MAKEDEV.itifg8 (via confing in /etc/makedev.d)

Using special MAKEDEV is a problem on newer Red Hat releases, as the script
will be removed on reboot.

TODO: Try to avoid hardcoding major device id in config file.
----------------------------
revision 8.8
date: 2005/03/30 13:17:42;  author: toralf;  state: Exp;  lines: +1 -1
Kernel name suffix "hugemem" also supported.
----------------------------
revision 8.7
date: 2005/03/17 10:15:22;  author: toralf;  state: Exp;  lines: +1 -1
/lib/modules/%{kernelRel}/misc listed instead of
/lib/modules/%{kernelRel}/misc/*. It seems like there are no other
drivers in misc, nor a misc directory at all by default, so we should
probably consider ourself as owner of the dir - so that it is properly
removed on upgrade or uninstall.
----------------------------
revision 8.6
date: 2005/03/15 10:28:44;  author: toralf;  state: Exp;  lines: +2 -0
- Module version setup fixed; should actually get version info now.
- Attempts to resolve segfault problems moved to a separate patch.
----------------------------
revision 8.5
date: 2005/03/14 09:36:41;  author: toralf;  state: Exp;  lines: +13 -27
- Kernel id macro setup changed again. The suffix setup in the
  previous versions was probably not correct, as UTS_RELEASE macro
  actually contains the suffix, too. Will now just look for well-known
  suffixes, and also use uname output only as base for version. To get
  CPU, rpm -q --queryformat '%{ARCH}\n' is used, but I still want a
  better way.
- Release policy changed. Will include "source" release no in
  Release:, adding another .<no> to allow multiple RPM releases for
  one source release. Version info will no longer be taken from CVS,
  as syncing the major CVS release no is probably too much hassle with
  this convention.
----------------------------
revision 8.4
date: 2005/03/09 16:04:34;  author: toralf;  state: Exp;  lines: +11 -13
- Updated for source release 8.2.1-17.
- Setup of kernel info macros changed. There is no longer a special
  preprossor input file, as most of the symbols it referenced are no longer
  available with Linux 2.6. Also, setting up version this way seems rather
  pointless since we have already used version specific include. I'm still
  doing it, though - the idea now is to find "suffix" by comparing 'uname -r'
  output with kernel header release def.

TODO: Find out how to distinguish between athlon and i686 where appropriate
      ('uname -m' returns "i686" for both.)
----------------------------
revision 8.3
date: 2005/03/09 12:30:15;  author: toralf;  state: Exp;  lines: +17 -10
Updated for build on Red Hat system with Linux 2.6. Notably, it is essential
that /lib/modules/<version>/build/include is always in the include path on
those setups, as /usr/src/linux is no longer updated. Also, my config.sh
patch removed some bits that were needed on version 2.6 but not 2.4...

+ Cleaner implementation of the update for build root install - all is now
  defined in terms of variables - so the Makefiles and scripts may still be
  used for normal root installation. Note that variable INSTALL_MOD_PATH
  is used for install root for the actual module. The variable name was
  taken from the kernel source makefiles.
----------------------------
revision 8.2
date: 2005/02/22 14:54:29;  author: toralf;  state: Exp;  lines: +79 -22
Will now generate "base" and "kernel" packages directly, via explicit
re-execution of rpmbuild where special target arguments are
passed. Works, but is not very neat...

A somewhat strange side effect: 'rpmbuild --target <target>' with
target set to anything but the default architecture, may be used to
generate kernel module only (with architecture set to the one of the
current kernel, as usual, rather than the <target> value.)

Also, architecture vs. --without setup was cleaned up a bit.
----------------------------
revision 8.1
date: 2005/02/22 11:43:55;  author: toralf;  state: Exp;  lines: +20 -13
- Will now adjust install script so that module will indeed be installed by
  'make install'. Special commands for the installation were removed.
- Another attempt to build base & kernel packages with different
  arch. Currently only amounts to if-tests on architecture - will
  build kernel package if the kernel CPU is specified externally as
  target, base/devel if target is the default one (like i386.) This
  means that default build will usually not add kernel module package;
  --target has to be used to build that package *only*. The --without
  options supported are made somewhat redundant by this.
----------------------------
revision 8.0
date: 2005/01/20 10:23:38;  author: toralf;  state: Exp;  lines: +11 -7
The first attempt to build itifg-8.2.1, with updated patches and all.
Note that the "redhat kernel" one is a lot simpler, as it needs only
to change some version code tests - as the changes that makes it necessary
are Linux 2.6 support backported to 2.4 - and the driver now supports 2.6,
too.
----------------------------
revision 6.8
date: 2004/11/16 12:32:16;  author: toralf;  state: Exp;  lines: +1 -1
Uses pte_offset_map() instead of pte_offset_kernel() for Red Hat kernels.

This is an untested attempt to avoid hang; someone on the itifg
list suggested using pte_offset_map() (not as a way to resolve our problems
with hangs, but in general.)