On Wed, Jul 22, 2020 at 9:52 PM Palmer Dabbelt <palmer@xxxxxxxxxxx> wrote:
> On Wed, 22 Jul 2020 02:43:50 PDT (-0700), Arnd Bergmann wrote:
> > On Tue, Jul 21, 2020 at 9:06 PM Palmer Dabbelt <palmer@xxxxxxxxxxx> wrote:
> > The eventual goal is to have a split of 3840MB for either user or linear map
> > plus and 256MB for vmalloc, including the kernel. Switching between linear
> > and user has a noticeable runtime overhead, but it relaxes both the limits
> > for user memory and lowmem, and it provides a somewhat stronger
> > address space isolation.
>
> Ya, I think we decided not to do that, at least for now. I guess the right
> answer there will depend on what 32-bit systems look like, and since we don't
> have any I'm inclined to just stick to the fast option.
Makes sense. Actually on 32-bit Arm we see fewer large-memory
configurations in new machines than we had in the past before 64-bit
machines were widely available at low cost, so I expect not to see a
lot new hardware with more than 1GB of DDR3 (two 256Mbit x16 chips)
for cost reasons, and rv32 is likely going to be similar, so you may never
really see a need for highmem or the above hack to increase the
size of the linear mapping.
I just noticed that rv32 allows 2GB of lowmem rather than just the usual
768MB or 1GB, at the expense of addressable user memory. This seems
like an unusual choice, but I also don't see any reason to change this
or make it more flexible unless actual users appear.
Arnd
