Re: [PATCH] IDE 58

[Benjamin Herrenschmidt]

>Alan, we talked about this and the driver/hardware has a flaw.
>If you count the total number of single IO operations to check
>status/error et al., it is out right fugly.  Preprocessing will kill us
>like today.

So we can still end up having both

 - The single ops indirected like I first proposed (with maybe the
addition of slow versions but that could be a parameter, and 32 bits
versions)
 - A couple of "apply this taskfile" versions with well known semantics
used only in normal cases for perfs.

Ben.