[PATCH 0/5] arm/tms570: include hardware initialization and selftests to run RTEMS directly from Flash without loader on TMS570LS3137 HDK.

[Pavel Pisa]

Hello Gedare,

thanks for the review. It is huge piece and not in a state
as nice as I would like it.

On Tuesday 13 of September 2016 21:42:48 Gedare Bloom wrote:
> I'm slowly reading through these. Is there a doc or is it worth it to
> generate one that maps the HalcoGen functions to their RTEMS renamed
> versions? I'm not sure if it is of much value.

I am not sure at the end. I have tried to change names
to RTEMS, Linux, POSIX style. The rest of BSP (which
is written mostly from scratch) already follows that.
Other option is to use Ti CammelCase and add tms570_ prefix.
Genrally I am not sure. I have not renamed functions
in the ARM core support tms570_sys_core.S .
But if we agree that all should follow single naming
convention I take a while for that.
But there could be value in preserving Ti names either.

> I'm slowly going through this big one. One quick note I wanted to make
> is to ask you to make comments where you have used #if 0 ... #endif to
> comment out code.

Yes, I should comment that more.
Some small parts are only comments by code sections.
But more text would help there.

But the biggest commented out blocks are in bspstarthooks-hwinit.c.
I probably try to move this code to subroutines.
The two bigger parts are internal main SRAM selftest
and test of reaction to SRAM content protection against bitflip
errors.

I plan to try to reenable at least selftest for cases where
it does not lead to crash. But these testes has to be skipped
for case when RTEMS application is linked and run from SRAM
and even for case when SRAM is used as overlay for FLASH
when RTEMS application is not starting from address zero.
Or at least I have problems with these cases.
When we have already finally link mode where RTEMS
can be the first (and only) Flash content starting
directly then for this mode code should work.
But requires more experiments.

Masks for tms570_pbist_run should be converted to
some defines as well -> to my TODO.

Complete RAM ECC reporting checking then triggers
real uncorrectable error fault which leads to ARM level
exception which is catch and only if that all happens
then selftest continues, else fault is reported.
All that requires that code runs from Flash and does
not touch RAM except for precise triggering events.
So again quite huge task and requiring modification
of RTEMS ARM exception table.

> Is "partest" halcogen term, or you chose it? I might prefer
> "parity_check" instead (and remove any redundant "parity_check_check".

I looked for some shorthand to have sane length names.
But I agree that they could be misleading. May it be parck_
to be short? On the other hand parity_check_ is more clear.
But in the fact it is "selftest of parity check reporting
mechanism health".

Best wishes,

               Pavel