From: Bram Verweij <bram.verweij@wanadoo.nl>

The problem seems to be that ide-disk.c tries to use PIO mode for
blocks > 137 GB (which is good), and LBA48 + DMA for blocks <= 137GB
(which is known to be a problem, i.e., this is why the no_lba48_dma 
field was introduced in the first place).  Attached is a small patch
that makes ide-disk.c use PIO mode for blocks > 137 GB, and LBA28 DMA
(instead of LBA48 DMA) for blocks <= 137 GB.

bart: argh, I forgot about 'lba48' flag; patch slightly modified by me
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>