This changes the way futexes are indexed, so that they don't pin pages. 
It also fixes some bugs with private mappings and COW pages.

Currently, all futexes look up the page at the userspace address and pin
it, using the pair (page,offset) as an index into a table of waiting
futexes.  Any page with a futex waiting on it remains pinned in RAM,
which is a problem when many futexes are used, especially with FUTEX_FD.

Another problem is that the page is not always the correct one, if it
can be changed later by a COW (copy on write) operation.  This can
happen when waiting on a futex without writing to it after fork(),
exec() or mmap(), if the page is then written to before attempting to
wake a futex at the same adress. 

There are two symptoms of the COW problem:
 - The wrong process can receive wakeups
 - A process can fail to receive required wakeups. 

This patch fixes both by changing the indexing so that VM_SHARED
mappings use the triple (inode,offset,index), and private mappings use
the pair (mm,virtual_address).

The former correctly handles all shared mappings, including tmpfs and
therefore all kinds of shared memory (IPC shm, /dev/shm and
MAP_ANON|MAP_SHARED).  This works because every mapping which is
VM_SHARED has an associated non-zero vma->vm_file, and hence inode.
(This is ensured in do_mmap_pgoff, where it calls shmem_zero_setup). 

The latter handles all private mappings, both files and anonymous.  It
isn't affected by COW, because it doesn't care about the actual pages,
just the virtual address.

The patch has a few bonuses:

        1. It removes the vcache implementation, as only futexes were
           using it, and they don't any more.

        2. Removing the vcache should make COW page faults a bit faster.

        3. Futex operations no longer take the page table lock, walk
           the page table, fault in pages that aren't mapped in the
           page table, or do a vcache hash lookup - they are mostly a
           simple offset calculation with one hash for the futex
           table.  So they should be noticably faster.

Special thanks to Hugh Dickins, Andrew Morton and Rusty Russell for
insightful feedback.  All suggestions are included.

into home.osdl.org:/home/torvalds/v2.5/linux

From Matias Alejo Garcia <kernel@matiu.com.ar>.

MODULE_ALIAS_LDISC() is not in any way architecture-specific, so don't
put it in architecture header files.

Here's a patch which moves it to a more sensible location.

The "insert_here" list pointer logic was broken, and unnecessary. 

Kill it and its associated logic off completely - just tell the IO
scheduler what kind of insert it is.

This also makes the *_insert_request strategies much easier to follow,
imo.

into nuts.ninka.net:/disk1/davem/BK/net-2.5

Since sdla.c now has locking instead of cli/sti, it shouldn't be
completely busted on SMP.

Also, the frad interface registration needed to be done once
(and check the result), and undone on module unload.

* get rid of leftover sti
* no longer need MOD_INC/DEC stuff
* get rid of dead code related to MOD_INC/DEC
* use module_init/module_exit to cleanly run init code

into nuts.ninka.net:/disk1/davem/BK/net-2.5

try_then_request_module() does what crypto/autoload.c is doing, so
replace it.  Fix try_then_request_module(), too (thanks James).

I should have stayed in bed.

Apparently the new gcc 3.2.2 allows local variable declarations within
the code of a function if it is not used earlier. But older gcc's do
not allow this. 

into nuts.ninka.net:/disk1/davem/BK/net-2.5

Cleanup ikconfig
	- use single_open for built_with file.
	- get rid of unneeded globals
	- use copy_to_user instead of char at a time
	- only need the read routine, proc defaults to correct behaviour
	  for the rest.

When mmap MAP_SHARED is done on a file, it gets marked with VM_MAYSHARE
and, if it's read/write, VM_SHARED.  However, if it is remapped with
mremap(), the MAP_SHARED is only passed into the new mapping based on
VM_SHARED.  This means that remapped read only MAP_SHARED mappings lose
VM_MAYSHARE.  This is causing us a problem on parisc because we have to
align all shared mappings carefully to mitigate cache aliasing problems.

The fix is to key passing the MAP_SHARED flag back into the remapped are
off VM_MAYSHARE not VM_SHARED.

driver->supports_dma was used together with CONFIG_IDEDMA_ONLYDISK to limit
DMA access to disk devices only.  However Alan introduced new scheme in 2.5.63
and this field is not needed any longer because all ide drivers support DMA.

From Benjamin Herrenschmidt <benh@kernel.crashing.org>

into home.osdl.org:/home/torvalds/v2.5/linux

into raven.il.steeleye.com:/home/jejb/BK/scsi-for-linus-2.6

Some references to irq_stat[] snuck into generic code.  It doesn't work on
several architectures.

So revert that little improvement to the original version.

It seems that changing CONFIG_BLK_DEV_INITRD causes the whole kernel to
rebuild due to an inappropriate ifdef in linux/fs.h - we should not
conditionalise prototypes.

In addition, real_root_dev is only used by two files (kernel/sysctl.c
and init/do_mounts_initrd.c) so it makes even less sense that it was in
linux/fs.h

into home.osdl.org:/home/torvalds/v2.5/linux

into kernel.bkbits.net:/home/davem/net-2.5

This patch fixes a logic bug in NAT range calculations, which also
causes a large slowdown when ICMP floods go through NAT.

Author: Karlis Peisenieks

From: Rusty Russell <rusty@rustcorp.com.au>

Rather than hardcoded names in modprobe, modules can offer their own
aliases (which can be overridden by the config file).

Here are the tty-ldisc ones.

From: mike.miller@hp.com

Clean up the error handling in cciss_init_one().

From: Michael Pruznick <michael_pruznick@mvista.com>

build errors:
  hermes.h: In function `hermes_set_irqmask':
  hermes.h:337: parse error before "do"
  hermes.h:337: parse error before ';' token
  hermes.h: In function `hermes_write_words':

In mips, outw_p() is a #define do...while(0) which, in the case of ?:,
results in a statement being used where an expression is required.