On Sun, May 02, 2004 at 06:11:15PM +0200, Manuel Bouyer wrote:
> It needs N contigous pages. In theory you can have Ram/N allocated and
> have it still fail, if the memory fragmentation is at its maximum.
> I think this can be worked around by bumping a kernel parameter,
> but I don't remember which one :(

Why does it need such a big buffer?  For maximum performance, sure, but..
(Well, I guess using anything more than one page would potentially cause
the same problem..)

(Now I remember what I was going to test on my laptop: I wanted to see if
the CF cards now work on my ThinkPad - I was complaining about data
corruption earlier.  The corruption was caused by writing multiple blocks
to CF using one IDE command - when I hacked the driver to do only one
block at a time, everything worked well (but dead slowly).)

  -jm