Convert core Sun-3 code to new local_irq*() framework

Mac/m68k early startup fixes (from Ray Knight in 2.4.x)
  - Correct memory mapping for Macintosh.
  - Prevent screen output from appearing twice.

Correct error found in arch/m68k/ whilst auditing the use of misc_register()
in 2.5.44.
misc_register() returns 0 on success, and a negative number on failure.
(Trivial patches misc_register-02[67] from Michael Still <mikal@stillhq.com>)

Convert m68k fbdev drivers to new local_irq*() framework:
  - Atari builtin graphics

Convert core MVME147 code to new local_irq*() framework

Convert m68k char drivers to new local_irq*() framework:
  - Amiga builtin serial
  - Amiga A2232 multiserial
  - MVME167 serial
  - MVME147/162 and BVME6000 SCC

Convert m68k block drivers to new local_irq*() framework:
  - Atari ACSI
  - Amiga floppy
  - Atari floppy
  - Macintosh IIfx/Quadra 900/Quadra 950 floppy

Convert core Mac/m68k code to new local_irq*() framework

Convert core Sun-3x code to new local_irq*() framework

Convert core BVME6000 code to new local_irq*() framework

M68k: Remove bogus prototype for dump_fpu(), which conflicts with the one in
<linux/elfcore.h>, and is wrong anyway.

Convert m68k net drivers to new local_irq*() framework:
  - Amiga A2065 and Ariadne Ethernet
  - Atari Bionet-100 and Pamsnet ACSI Ethernet
  - Atari LANCE Ethernet
  - Mac NC8390, CS89x0, MACE, and Sonic Ethernet
  - Sun-3 i82586 and Sun-3/3x LANCE Ethernet

Convert m68k parport drivers to new local_irq*() framework:
  - Atari parport

Move the definition of struct font_desc back to <video/font.h>, so it can be
used outside of drivers/video/console (needed for Mac/m68k boot progress
messages).

Mac/m68k VIA updates (from Ray Knight in 2.4.x)
  - PowerBooks and early Macs use different bits for nubus.
  - Add loglevel macros to printk.

Remove duplicate Makefile entry for bwtwofb

NCR53C9x ESP SCSI driver updates:
  - Port recent 2.5.x Sun/SPARC ESP SCSI driver changes to the NCR53C9x ESP
    SCSI drivers
  - Move NCR53C9x ESP host templates and host adapter-specific definitions from
    the host adapter-specific header files to the source files
  - Move common NCR53C9x ESP prototypes from the host adapter-specific header
    files to NCR53C9x.h

This affects the following SCSI host adapters:
  - M68k/PPC Amiga Blizzard 1230IV, 1260, and 2060
  - M68k/PPC Amiga BSC Oktagon
  - M68k/PPC Amiga Cyberstorm and Cyberstorm II
  - M68k/PPC Amiga Phase 5 Fastlane
  - MIPS DECstation NCR53C94
  - MIPS JAZZ Fas216
  - M68k Mac 53C9x
  - ia32 MCA 53C9x
  - M68k Sun-3x ESP

M68k: local_irq_count was moved to the hardirq bitmask of preempt_count() (cfr.
asm/hardirq.h).

Amiga floppy: fix incorrect sti()

Ataflop: Add missing =

Scsi_Host.next was removed in 2.5.44, convert to use scsi_host_hn_get()

Atari NCR5380 SCSI: bitops operate on long, not char.
Note: the preprocessor stuff just above checks for MAX_TAGS being a multiple of
32, so dividing by 32 is OK.

M68k: struct fbcon_font_desc was renamed to struct font_desc in 2.5.51

Fix inbalance of curly braces if AUTOSENSE is not defined.

for_each_task() was renamed to for_each_process() in 2.5.35

M68k: add restart_block field to struct thread_info (needed for 2.5.51)

M68k: add *__param to link files (needed for 2.5.52)

M68k: update for changed do_fork() semantics in 2.5.49

M68k module list updates, inspired by ia32 traps.c in 2.5.48

M68k: do_coredump() gained an extra argument in 2.5.52

Add definitions for {,smp_}read_barrier_depends()

When compiling aic7xxx in 2.5.53 with defconfig the kernel always
recompiles because dependency for reg_print.c is not per default in the
aic7xxx Makefile.

Simple correction is to make PRETTY_PRINT dependend on BUILD_FIRMWARE.

If modules don't use module_exit(), they cannot be unloaded.  This
safety mechanism should not apply for modules which don't use
module_init() (implying they have nothing to clean up anyway).

The compat API doesn't cope correctly if struct pci_dev is NULL, as it
can be for the x86 fake pci_dev handling.

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

This was done to make the next reference count patch easier,
and because almost everyone was already calling usb_put_dev() anyway...