2.4.10 cdrom readahead botch

[Rico Tudor]

Reading within 160 KB of the end-of-track causes pandemonium.

For those unfamiliar with CDROM media, reading past end-of-track (in TAO
recordings) triggers a hard error from the drive.  For SAO, the error
occurs at the end of the last track of the session.

Through 2.4.9, Linux readahead is about 2 sectors, error recovery takes
a second, and DMA settings are preserved.  Even that readahead glitch
can be avoided by lightly padding the end of a track during a burn.

CDROM support in 2.4.10 has taken a giant step into the toilet, and
effectively breaks existing CDROMs.  Reading within 80 CDROM sectors
causes a boatload of error messages, minutes of delay, ATAPI resets,
loss of DMA settings.  I count 15 ATAPI resets when there should be 0.