Re: [PATCH 00/16] remove eight obsolete architectures

[Palmer Dabbelt]

On Thu, 15 Mar 2018 03:42:25 PDT (-0700), Arnd Bergmann wrote:
> On Thu, Mar 15, 2018 at 10:59 AM, Hannes Reinecke <hare@xxxxxxx> wrote:
> > On 03/15/2018 10:42 AM, David Howells wrote:
> > > Do we have anything left that still implements NOMMU?
> > RISC-V ?
> > (evil grin :-)
>
> Is anyone producing a chip that includes enough of the Privileged ISA spec
> to have things like system calls, but not the MMU parts?
>
> I thought at least initially the kernel only supports hardware that has a rather
> complete feature set.

We currently do not have a NOMMU port.  As far as I know, everyone who's
currently producing RISC-V hardware with enough memory to run Linux has S mode
with paging support.  The ISA allows for S mode without paging but there's no
hardware for that -- if you're going to put a DRAM controller on there then
paging seems pretty cheap.  You could run a NOMMU port on a system with S-mode
and paging, but With all the superpage stuff I don't think you'll get an
appreciable performance win for any workload running without an MMU so there's
nothing to justify the work (and incompatibility) of a NOMMU port there.

While I think you could implement a NOMMU port on a machine with only M and U
modes (and therefor no address translation at all), I don't know of any MU-only
machines that have enough memory to run Linux (ours have less than 32KiB).  A
SBI-free Linux would be a prerequisite for this, but there's some interest in
that outside of a NOMMU port so it might materialize anyway.

Of course, QEMU could probably be tricked into emulating one of these machines
with little to no effort :)...  That said, I doubt we'll see a NOMMU port
materialize without some real hardware as it's a lot of work for a QEMU-only
target.