Up to now, we generated module versions for all objects which
were listed in $(export-objs).

This had one advantage: A changed .config will not affect which
.ver files are built, thus saving recompiles.

However, it is fundamentally broken. To build .ver files, we preprocess
the exporting sources - and the result can obviously depend on
the current .config. Even worse, some files generate errors when
preprocessed with the wrong .config - it doesn't matter a lot that
drivers/sbus/* will generate errors on x86, since it won't be used anyway,
but e.g. kernel/suspend.c cannot be preprocessed unless 
CONFIG_SOFTWARE_SUSPEND is set. - Up to now, we just silently ignore
these errors.

Actually, the whole point behind CONFIG_MODVERSIONS is to make sure
we don't insert modules into a kernel which was configured differently,
and as such the generation of symbols can only work when .config is
known.

So we now only generate symbols for objects which will actually be
compiled - which means less work, and enforce the kernel to be configured
before "make dep".

Keep track of possible targets while we add rules - Later, only
include the command lines / dependencies as needed.

E.g., when defining how to build multi-part objects listed in
$(multi-used-y) and $(multi-used-m), add $(multi-used-y,m) to 
the variable targets. At the end of Rules.make, we then include
the saved command lines for these targets (if they exist), needed
to check whether the command changed and we hence need to rebuild.

Previously, we always had put the .cmd/.d files into the local subdir
- now we put them into the directory where the target lives, like

	some/dir/target.o <-> some/dir/.target.o.cmd

It does not yet make a difference, since we always only generated
targets in the local directory anyway, but that will change with
tracking dependencies for .ver files.

We use A/CFLAGS_KERNEL by default anyway, so no need to set them
explicitly for built-in objects.

This patch doesn't change anything but the order (and some comments) in
Rules.make. The reason for that is that we use Rules.make for three
different purposes, fastdep, modules_install and the actual building.

As an example, we don't need all the dependencies and saved command lines
when installing modules, so it'll speed up this pass to only include the
rules which will actually get used.

So far, we had been lazy and not generated dependencies for 
debugging-only targets, like some/dir/foo.[is]. It actually
cleans up Rules.make to handle them by the same mechanism, though.

Additionally, make use of $(call if_changed) and $(call if_changed_dep)
more consistent, the only difference now is post-processing of
gcc -MD output, which is done for if_changed_dep but not for if_changed.

cpp -MD guesses the target name from the source file name, but not
always correctly, e.g. when compiling fixdep, it generates prequisites
for fixdep.o instead of fixdep.
 
fixdep now fixes up the -MD generated output to contain the right target,
which is the one we pass on the command line.

o When using "make -j<n>", the order in modversions.h could change,
  thus causing unnecessary rebuilds.
o Move a comment in Rules.make which I forgot when I moved the
  associated code.
o When descending into init/ during the final vmlinux link,
  add a '+' do tell make to use its normal "-j" jobserver handling.
o It'd be nice to get the logic between make clean and make mrproper
  clearer. I think make clean should remove all targets used in the
  actual build,
  make mrproper should remove the rest, build-specific files, like
  e.g. the .*.cmd files used to save dependencies and command lines
  and other temporary files.
  A real small step in the direction: Remove .tmpversion/.tmp-export-objs
  at mrproper time.

fs/nfs/dir.c:

   - Call nfs_renew_times() when we deliberately create a negative dentry.
     or rename a file.

fs/nfs/inode.c:
  - Ensure that we initialize NFS_MTIME_UPDATE(inode).

  - If a directory's mtime changed due to some action on our part
    (file creation/deletion/...) don't force a dcache revalidation by
    changing NFS_MTIME_UPDATE(inode).

Fourth and last part of the s/390 update. Docu stuff.

Third patch of the s/390 update. Everything from drivers/s390. This patch
contains the main bunch of changes. Major changes are: 1) the restructured
dasd driver, 2) the new tape driver, 3) the start of the common io layer
rework, 4) the gpl'ed lcs driver and 5) the gpl'ed qdio layer

Second patch of the s/390 update. Contains all the include file changes in
include/asm-{s390,s390x}.

This first patch contains the changes to arch/{s390,s390x}

resume_kernel uses CPU_IDX but never uses GET_CPU_IDX to get the index.

  This is an issue when smp and preemption are both enabled.  I also
removed the unused GET_CURRENT_CPU_IDX.

James introduced bad bug which causes NULL pointer dereference as soon
as you'll try to use secondary head because of screen_base is not
initialized (it is initialized on wrong head).

into home.transmeta.com:/home/torvalds/v2.5/linux

Mike Kravetz pointed out that the set_cpus_allowed optimization he
suggested has a small but possible race condition wherein the system
could still be operating in the context of the task but it is not
running.  On top of this, the runqueue lock can be dropped in
load_balance and thus we can race and set task->cpu at a very unpleasant
time.

My solution is to just remove the optimization.

Apparently my understanding of C brace rules was blurred and there is a
thinko in kernel/capability.c - what we want is what the indention
reflects but what we get is an unconditional return of -EINVAL.

into home.transmeta.com:/home/torvalds/v2.5/linux

Define the for_each_lock macro and start replacing ugly special for loops
with it.  Rejig the interface between sys_flock and flock_lock_file to
always pass a struct file_lock rather than a command.  Eliminate some
gotos by simplifying the logic.  Remove some redundant initialisation.

Fixes for include file cleanups

into home.transmeta.com:/home/torvalds/v2.5/linux

This removes <linux/mm.h> from <linux/vmalloc.h>.

This then goes and fixes all of the files (x86 and PPC) which relied on
implicit includes which don't happen anymore.  This also takes
<linux/kdev_t.h> out of fs/mpage.c and puts it into include/linux/bio.h
where it belongs since <linux/bio.h> references 'kdev_t' directly.

A quick summary of the of the added includes:
arch/i386/kernel/microcode.c: needs extern for num_physpages, in linux/mm.h
include/linux/spinlock.h: local_irq* is defined in <asm/system.h> but
this was never directly included.

This moves the vmalloc wrappers from <linux/vmalloc.h> into mm/vmalloc.c.

Doing this will later allow us to remove <linux/mm.h> from <linux/vmalloc.h>,
along with some other #include fixups.

This patch removes the MSNFS defines.  These locks have a wider purpose
than emulating Microsoft NFS semantics.

This patch removes the annoying and confusing `wait' argument from
many places.  The only change in behaviour is that we now yield once
when unblocking other BSD-style flocks instead of once for each lock.

This slightly improves the semantics for userspace.  Before, when we
had two tasks waiting on a lock, the first one would receive the lock.
Now, the one with the highest priority receives the lock.

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

This adds SA-1111 support for ohci-hcd and fixes usb-ohci too.

thanks to suggestions from David Brownell I modified
pwc driver to compile and work under 2.5.18.
Also it fixes mas ISO transfer sizes which could
potentionaly affect older kernels too.
With these changes it works - I'll test stability and
post results later.

This patch removes the use of urb->next in usbvideo.c and resubmits
the URB in the completion handler.

This patch removes the use of urb->next in konicawc.c and resubmits
the URB in the completion handler.

He has added significant content to these files.