On Fri, May 11, 2018 at 10:37:42AM -0500, Rob Herring wrote:
> On Fri, May 11, 2018 at 10:02 AM, Rich Felker <dal...@libc.org> wrote:
> > On Fri, May 11, 2018 at 08:45:59AM -0500, Rob Herring wrote:
> >> Commit 0fa1c579349f ("of/fdt: use memblock_virt_alloc for early alloc")
> >> inadvertently switched the DT unflattening allocations from memblock to
> >> bootmem which doesn't work because the unflattening happens before
> >> bootmem is initialized. Swapping the order of bootmem init and
> >> unflattening could also fix this, but removing bootmem is desired. So
> >> enable NO_BOOTMEM on SH like other architectures have done.
> >>
> >> Fixes: 0fa1c579349f ("of/fdt: use memblock_virt_alloc for early alloc")
> >> Reported-by: Rich Felker <dal...@libc.org>
> >> Cc: Yoshinori Sato <ys...@users.sourceforge.jp>
> >> Signed-off-by: Rob Herring <r...@kernel.org>
> >> ---
> >> This is compile tested only, but similar to microblaze and h8300
> >> conversions. The additional complexity for SH is NUMA support (which to
> >> me looks like an abuse of NUMA support to map a small amount of
> >> on-chip? RAM to NUMA nodes).
> >
> > Thanks! I was just reading the corresponding microblaze commit, and
> > think this approach makes sense. I'll test it now with both DT and
> > non-DT sh systems and let you know if it works. If it's good would you
> > like me to send it upstream via arch/sh? I already have 2
> > regression-fix patches to submit in a pull request asap.
>
> Yes, please do.
OK. So far looks good. J2 works again (DT-based) and I'm now doing a
new build to test under qemu (r2d board) for non-DT sh.
Rich
