On Fri, Mar 05, 2021 at 10:58:02PM +1100, Michael Ellerman wrote:

Will Deacon [off-list ref] writes:

quoted

That's very similar to us; we're not relocated, although we are at least
in control of the MMU (which is using a temporary set of page-tables).

prom_init runs as an OF client, with the MMU off (except on some Apple
machines), and we don't own the MMU. So there's really nothing we can do :)

You *could* take over all memory mapping.  This is complex, and I
estimate the change you get this to work correctly on all supported
systems to be between -400% and 0%.

And not very long later Linux jettisons OF completely anyway.

Though now that I look at it, I don't think we should be doing this
level of commandline handling in prom_init. It should just grab the
value from firmware and pass it to the kernel proper, and then all the
prepend/append/force etc. logic should happen there.

That sounds much simpler, yes :-)


Segher