[J-core] Instruction Request: Dedicated CPUID Function

Cedric BAIL cedric.bail at free.fr
Tue Oct 24 21:44:14 EDT 2017

On Mon, Oct 23, 2017 at 6:03 PM, Rob Landley <rob at landley.net> wrote:
> On 10/23/2017 06:11 PM, D. Jeff Dionne wrote:
>> My point is there is n bytes of storage practically available on chip
>> for that, where 128 <= n <= 2k.  At 2k, it's already 1/4 of a nice,
>> practical on chip 1st stage bootloader...
> The kernel already knows how to gunzip itself during early boot, can we
> teach it to decompress the device tree? If so that gives us 2-4x the
> space...

Why not go one step further. The problem of storing the DT out of the
SoC is a synchronisation issue. Not having it in the hardware put less
pressure on getting it right also at manufacturing time. So why not
just put a public elliptic curve signature of the DT in the chip ?
This would cost 160bits. I am assuming here that the early stage boot
loader has enough logic to actually get to the next step in the boot
without reading the DT. Maybe I am wrong on that assumption.
Cedric BAIL

More information about the J-core mailing list