Version 0.1.0
  Re-ported from the FreeBSD driver version 1.0.11 (from the end of
March). Made IRQ stuff sharable. Changed the setting of dma_end[eo]
registers; while this is a variance from the original code FreeBSD
code, I am setting them to stop at the last long-word in the buffer,
rather than at the address of the buffer plus the buffer size (this
possibly might have been an off-by-one sort of error, and may have
been causing my problems with the bigphysarea code). Seems to work
this way, anyway.
 I did more examination of range-checking and determined that the
FreeBSD driver was in all probability never using it. I took all the
range_enable stuff out and turned it unconditionally on (I think one
would need to be dealing with something like 16MB to start to possibly
have trouble with it). After doing this, I started getting occasional
dma odd address errors. My SHOW statements look like all the addresses
should be valid, but something is triggering a write that probably
would be trashing random memory if allowed. I'm happy with range
checking enabled.
  With the dma bug fixed, bigphysaddr works, so I've finally ditched
the MeteorBuffer hack and gone over to this.
  I've had to pull some bizarre hacks to be able to use range checking,
but it seems to work.

Version 0.2.0:
 Put code back in to set PCI_LATENCY. Changed memory allocation
in various ways, most importantly by moving it from init code to open code
so the frame buffers are allocated at open() time and closed at close() time.
Got the signal-on-frame-capture mechanism working, rewrote view to use it.
Seems to be working quite well.

Version 0.3.0: 

 In meteor.c:
 Changed memory management: Made initial frame buffer allocation, done in open(),
conditional upon METEOR_ALLOC being defined. #ifdeffed METEOR_ALLOC out. It seems
doubtful that users will not set their own geometry and format.
 Improved range-check-checking code in alloc_frame_buffer().
 Put dynamic frame buffer allocation back into the SetGeometry code in the ioctl().
 Improved signal handling.
 Added a bunch of debugging chatter, conditional on DEBUG_METEOR being defined.

 In Makefile:
   Replaced rmmod with modprobe -r. No confusing spurious error message.
   Added SMP and SMP_PROF to CFLAGS. ADJUST TO SUIT YOUR SYSTEM.
  
 Improved signal handling in view.cc. Changed printf()s to more informative
 perror()s in view.cc and single.c

Version 0.4.0:

  Put a break where it was needed at the end of METEORCAPTUR ioctl() handling.
  Changed the range-checking code in alloc_frame_buffer() to dynamically determine
 the size of the saa7116 dma_end registers and thus the maxLimit and boundary.

Version 0.9: (close to 1.0)

 Put a break where it was needed at the end of METEORCAPTUR ioctl() handling.
 Changed the range-checking code in alloc_frame_buffer() to dynamically determine
  the size of the saa7116 dma_end registers and thus the maxRange.
  Moved this code into inline probe_range().
 Changed a wrong-looking <= to a >= in the TimeStamp code in meteor_intr().
 Changed alloc_meteor_buffer to PAGE_ALIGN wantSize, keeps meteor_mmap() happy.
 Moved the pci_latency code to inline set_latency().
 More comments and PRINTK(())s.

Version 0.9.1:

 Changed uses of schedule() as delay to wait_jiffies() (schedule() might
  be too quick on a lightly-loaded SMP system).
 Changed meteor_unwind() to do a reset_saa7116() for each meteor.
 Enabled field_toggle as per FreeBSD driver.
 Renamed view.cc to vw.cc to avoid name conflict.
 Updated bigphysarea-patch to linux-1.3.95.
 Took the sleep() out of single; not needed now.

Version 0.9.2:

 Made the delay time for dealing with hardware registers WAIT_JIFFS (defined
  in meteor.h).
 Made checkPCI() to check and clear PCI status. Called from
 start_capture() and from meteor_intr().

Version 0.9.3:

 Fixed what was definitely at least partially a strange hardware problem
  (probably a Neptune PCI weirdness) by judicious use of delays.
 Rewrote the wait_jiffies() code, various other bits.
 No known or suspected problems remaining, even on Neptune chipsets. Lots of
  little tweaks.

Version pre-1.0:

 Changed wait_jiffies() to be uninterruptible. Turned off all debugging chatter.

Version 1.0: 5/1/96
  Made checkPCI() inline, #ifdeffed it on DEBUG_METEOR so it should
disappear unless debugging is set.
 Changed Makefile to compile -Wall -Wstrict-prototypes. This revealed
a typo in ioctl METEORSBT254. Fixed. Added parens, prototypes etc
until no warnings.

Version 1.1: 5/6/96
 Implemented METEORSTATUS ioctl fix posted by Jim Lowe.

Version 1.2 14/May/96 Ian Reid (ian@robots.ox.ac.uk)

 Included Jim Bray's patch 1.1.1 tidying meteor_init
 Fixed frames per second so that it actually implements
  the call in the saa7116.
 Fixed bug in SVIDEO input which ignored chrominance
 Fixed #define METEOR_INPUT_DEV_RCA (previously defined to METEOR_GEO_DEV0)
 Fixed #define METEOR_DEV_MASK (previously 0x2000f000)
 Fixed spelling of PLANAR (previously PLANER)
 Added a selectable system default (SYSTEM=PAL/SECAM/NTSC)

Version 1.3 22/May/96 Ian Reid (ian@robots.ox.ac.uk)

 Fixed problem with PAL RGB mode (from jonas@mcs.com)
 Fixed problem with frames per second code
 Fixed problem with PCF8574_xxxx_WRITE
 Added METEOR_GEO_ODD_EVEN to enable separate capture of odd/even
  fields
 Added METEOR_SIG_[FRAME|FIELD] to toggle between interrupts after
  each field or frame (frame still the default)
 Added METEOR_FIELD_MODE to capture fields as if they were frames
  *** this has not been thoroughly tested on all image types and
      on synchronous capture ***

Version 1.4 17/July/96

 Fixed prob inadvertently introduced to set_fps in the last fix
  (thanks Pedro Felzenszwalb pff@cs.cornell.edu)
 Tested and fixed FIELD_MODE stuff
 Added support for YUV in synchronous mode
 Allow lowat==hiwat==0 in synchronous mode to mean capture
  in round-robin fashion without user intervention.  Also
  added cur_frame to struct meteor_mem so that this info is
  available in shared memory.
 It appears that not only is there a problem with the number of bits
  in the dma_end registers (only 22), but that the dma regs themselves don't
  increment over 4Mb boundaries.  Instead they wrap around to the start of the 
  4Mb block.  Furthermore, since they only increment in 4 byte words, if the
  image width is not divisible by 8 then YUV images can overflow (since the U
  and V parts are saved separately using dma regs 1 and 2).
  The changes to fix these problems are due to anuj@fwi.uva.nl and include a
  complete rewrite of alloc_frame_buffer which may not be completely backwardly
  compatibile, but is much cleaner and more general.  You can now grab a 
  sequence of images once (using a new capture flag METEOR_CAP_CONT_ONCE added
  by anuj@fwi.uva.nl) and use buffers of >4Mb.  In order to allow this, a new
  ioctl (METEORGFROFF) returns the offsets into the frame buffer of each frame
  and the meteor_mem structure (see ioctl_meteor.h for the meteor_frame_offset
  structure).

Version 1.4a 1/9/96
  Very minor changes to tidy v1.4 and update some info in README
  Bug fixes from Mark Wolski (mwolski@rvssun3.gmr.com)
  Fix to FIELD_MODE so it resets properly

Version 1.4b 5/11/96
  Minor bug fixes
  Added equivalence between YUV_422 and YUV_PLANAR
 
Version 1.4c 20/2/97
  Bug fixes (from burschka@mail.lpr.e-technik.tu-muenchen.de)
