> Alan loves to complain about lack of MMIO flush, but in practice this is
> rarely the source of problems such as the one you describe.
> But if its broken its broken, and we need to revert. Any luck getting
> benh access to the device?
Especially since reverting it will break a whole bunch of cell blades,
which wouldn't be nice (those cannot do PIO and have that controller).
In the meantime, we might "workaround" with a hack to only enable MMIO
on those cell
blades, something like:
mmio = 1;
That might get us out of the regression until we find the proper