Erik> David MT, should there be a new syscall for io stuff, so the
Erik> Jensen differences can be hidden. Is there any way to hide
Erik> those differences in PAL code instead? I assume that would be
Erik> faster... The way things are now, there is no way Red Hat can
Erik> fully support Jensen's.
No, it can be handled transparently in user mode. As mentioned in an
earlier mail to linux-alpha (which you may not have received yet), the
general idea is to use the (cached) "system type" in /proc/cpuinfo to
dynamically adjust to the I/O system. The easiest solution would be
to use a table of function pointers to dispatch to the correct
routine.
--david