[Arm-netbook] It arrived in the post just now!

[Henrik Nordström]

tor 2013-08-01 klockan 02:01 +0100 skrev luke.leighton:
>  ok.  right.  there's i think _one_ possible last spare pin on the
> 68-pin header (i've been thinking of making UART_TX and UART_RX part
> of the EOMA-68 spec) - which could potentially be used here to get a
> CPU card to go into "bootloader" mode.

Are you sure it should go into EOMA68 standard? It is not strictly
needed for EOMA68-A20 if the SD port in the EOMA68 connector works.

But yes probably something is needed to avoid requiring a recovery
button on each card. Most CPUs do require pin strapping to select boot
source, usually not as friendly as A10/A13/A10s/A20 where the CPU tries
the different sources in order. Even the A31 does require pin strapping
to select boot source, fortunately just a single pin for FEL.

>  however... that's typically going to be CPU-card-specific.  but... if
> you need to boot and install a CPU Card from scratch you *will* know
> which one you're dealing with, therefore you'd put the right OS on,
> use the right tools etc. etc. and would use the appropriate boot
> pinout setup etc. etc.

Yes, what "bootloader" mode means will differ between cards. For some it
will simply mean "boot from external storage instead of internal flash",
for others "enter USB diagnostics/install" such as the FEL mode on
Allwinner CPUs, for others something else, what external connectors a
EOMA68 card have is card specific. For some maybe nothing (not required,
i.e. when the boot media itself can be changed).

For those not capable of booting from any of the available storage types
or connectors other solutions need to be found such as switching
external USB port from USB to something else. I.e. turn it into a JTAG
port or similar. All card specific outside EOMA68.

If end users are to use this to recover "bricked" cards then we can't
rely on any other EOMA68 pins such as the UART pins there.

If it's not for endusers then the pin can mean that the card is
connected to an engineering/service station and EOMA68 pins can get
other defined meaning, but that route easily leads to CPU specific
requirements.

With fully assembled cards built for mechanical durability we can not
count on users being able to at all disasemble the PCMCIA casing to
enter service mode as doing so will likely destroy the casing. It's easy
to open the current EOMA68-A20 only because it's not fully assembled
missing external faceplate, and also having some margin inside allowing
the casing to flex and not providing any tension to the metal shield to
grip into the plastic.

For the production run you want to adjust things so that the card i
flush against the plastic of the casing so the metal shields get a good
grip, add some fillers between PCB and shield to provide additinal
support for the metal shields, where one of these is thermally
conducting connecting the CPU to the metal shield for cooling
(preferably both CPU and PCB sides), add a suitable external faceplate
and a bit more glue on that side as well to hold it together. Opening
that will be a nightmare.

Regards
Henrik