Yes this is big but 2.4 has been indented, updated and there wasnt a
sane way to deal with it

There are some bits left to do but this folds in all the main stuff from
2.4 and 2.5 including Al's recent change

This doesnt fix all the isdn code but it sorts out the tqueue stuff so we
are no worse than before

Linus, this has been tested by some people in the field to not break
things, and it's the start of some other changes I'm making, so please put
this in your tree so I'm not merging huge patches but instead am merging a
little as I go.

This enables tagged command queueing support in the ide layer again and
marks it as an experimental feature.

This changes the way we do pool lookups when freeing a bio.  Right now
we use bi_max as a handle into bvec_array[], to find the pool where it
came from.  This used to be just fine, because if you had a private bio,
you could specify your own destructor.  But now we have bio_add_page()
which also needs to know where the bio came from, or more precisely, it
needs to know how many entries the bio can hold.

So I've changed bi_max to bi_max_vecs, it now contains the number of vec
entries in the bio. Privately allocated bio's (or on stack) can now just
set bio->bi_max_vecs to reflect the max size. The pool index for the
default destructor is stored in the top bits of bi_flags.

Make things static, and remove hardcoded value

As per comment..

Hide ISA only drivers during configuration..

From Pavel Machek. Without these two lines,
the throttling limits are never reached, and a machine
will happily cook itself.

Sammy changed address, and lets BUG() instead of
locking up the machine.

Daniel Phillips' indexed directory support.  Ported from ext2 by
Christopher Li.  Contributions from Andreas Dilger, Stephen Tweedie,
lots from Ted.

It requires e2fsprogs-1.29; I've updated the Changes file to reflect
that.

From Bill Irwin

The simple_link(), simple_unlink(), simple_rename(),
simple_sync_file(), simple_rmdir(), and simple_empty() functions are
easy to duplicate.  Basically, I duplicated them in an fs patch of
mine, and Linus told me to put them in libfs.c

Pat Mochel has acked the changes for driverfs (in that he'll convert
the stuff over when they're available from libfs), and my hugetlbfs
implementation was the thing that spurred the whole incident.

From Christoph Hellwig.

If we always pass &contig_page_data into free_area_init_node for the
non-distcontig case we can merge both versions of that function into
one.  Move that one to page_alloc.c and thus kill numa.c which was
totally misnamed, btw.

Patch from Christoph Hellwig

It turns alloc_pages_node into a static inline, in the same fashion as
alloc_pages.  There is no need for #ifdef CONFIG_NUMA in the patch, as
the numa node identification functions are sensible enough to do the
right thing for non-NUMA systems.  Moves alloc_pages_node from numa.c
to gfp.h, and removes the EXPORT_SYMBOL which is no longer needed as
this is now an inline.

This used to be a macro wrapper around get_zeroed_page().

It was a dreadful choice of name, because it hid the fact that this
function memsets the page.  Probably, lots of the callers didn't
actually want to do that but this patch does not make such
optimisations.

At least now, people may spot a `get_zeroed_page' in there and realise
that the page doesn't actually need to be initialised.

The patch also changes get_zeroed_page() to go BUG if called with
__GFP_HIGHMEM.  It returns the address of the page's memory and hence
can only be used for direct-mapped pages (wli).

From Martin Bligh.

This patch remaps the lmem_map (struct page) arrays for each node onto
their own nodes.  This is non-trivial, since all of ZONE_NORMAL, and
hence permanently mapped KVA resides on node 0.

Very early in the boot sequence, it calculates the size of the lmem_map
arrays (rounding up to the nearest large page size), and reserves a
suitable amount of permanent KVA by shifting down max_low_pfn to create
a gap between max_low_pfn and highstart_pfn (both of which are normally
about 896Mb).

It then uses the new set_pmd_pfn function to set up the pmds correctly
so that the large pages point at the physical addresses reserved from
the remote nodes.

Tested on NUMA-Q and some ratty old i386 PC kicking around under my
desk (on 2.5.36-mm1).  Was good for a 20% improvement in system time on
kernel compile when I initially benchmarked it against 2.5.32 or
something - due to a reduction in inter-node traffic, better
interconnect cache usage and locality.  Should have no effect on any
system other than i386 NUMA systems.

From Martin Bligh

This patch fixes a nasty bug that wli found in buffer.c which
cause an oops - we were using contig_page_data on a discontigmem
machine. It's a slightly modified version of the fix wli suggested,
tested on NUMA-Q.

I've also added code to not define contig_page_data for discontigmem
systems, to stop this from happening again. I wrapped a couple of
bootmem functions that were using it in #ifndef CONFIG_DISCONTIGMEM.
I suppose it's possible (though unlikely) that some other discontig
arch might need to wrap a couple of functions in their tree similarly,
but any borkage will just give a simple clear compiler error telling
them exactly where the problem is.

A discontigmem compilation fix From Bill Irwin

- Don't convert READA into READ in ll_rw_block().

This patch basically just renames usages of CONFIG_MULTIQUAD to
CONFIG_X86_NUMAQ. The original multiquad option covered a variety
of sins, and just made a mess (my fault). After extensive community
discussion, this is the naming scheme everyone agreed to - the
CONFIG_X86_NUMAQ option already exists, we're just renaming usages
of MULTIQUAD to it, and removing the MULTIQUAD option.

Stuff that's really to do with clustered APIC mode is put under
CONFIG_CLUSTERED_APIC instead, as it's not specific to the NUMA-Q.

Note that the definitions of clustered_apic_mode are still duplicated,
so the changes appear twice.  I'm not making it worse, just haven't
fixed this one yet, will do that next.

Patch was written by Matt Dobson.

Or in this case two.  This makes the migration and ksoftirq threads
match the naming convention of the workqueue threads.

Apply it and marvel at how neater a ps looks.

I tried building the aacraid driver in 2.5.41 and got an error
saying that O_TARGET usage was obsolete.  The following change
to the Makefile allows me to build.