MMm.. I just dug through the 3ware driver source code, and it looks like they F-d up in their interrupt sharing logic. Their interrupt handler (2.6.10) returns "handled==1(TRUE)" regardless of whether the interrupt was for them or not. So a mouse interrupt on the same line (IRQ12) might just get stuck forever there..

Silly 3ware.

The fix is to edit linux-2.6.10/drivers/scsi/3w-9xxx.c and move the "handled = 1" line down a little bit. Source patch attached. This may or may not fix the problem, but it's worth a shot.

Cheers