[coreboot] libpci accesses in CAR, I need a suggestion

[Pete Batard]

On 2011.08.10 23:42, Stefan Reinauer wrote:
> I think CAR only works well for data, not for instructions. That said,
> you can cache the ROM, but probably not copy it to cache and jmp there.

I've had success doing just that (copying/uploading instructions to 
cache and then jumping to it) in UBRX [1]. This is basically the 
approach I'd envision to solve the flashrom in panic-room issues on my end.

Of course this requires initializing and using the L2-Unified cache for 
the program, rather than L1, since L1-Data and L1-Instructions are 
pretty much always separated, which in turn requires a different 
approach to the panic room, but at least on x86, this shouldn't be much 
of an issue.

For this to work, one only needs a panic room that includes a basic 
scripting console (such as the one UBRX provides as per see [2]), to 
allow the *user* to set MTRRs and L1+L2 CAR according to their CPU, as 
well as the upload and flush of a binary executable into L2-Unified.

While this may sound like a lot of extra work for the user, eventually 
this should boil down to a copy/paste of a pre-existing CPU script, 
which we'd provide, and some Y-modem transfer of an executable payload. 
In my view, this is the panic-room approach that offers the greatest 
flexibility as, for a flashrom payload, the immediate advantages would be:
- flashrom executable from user upload rather than flash, meaning both 
freed up flash space and bugfixes/improvement to flashrom being able to 
be applied immediately rather than requiring a bootblock reflash
- size of the flashrom executable only limited by the size of L2, which 
is a lot larger than L1
- whilst still bare-metal, flashrom can be compiled with gcc rather than 
ROM_CC, as stack (from the L1-Data CAR init) is available
- the whole BIOS is flashable at once, since flashrom is executed from 
cache and not XIP.

Of course you guys may think this is a shameless plug for UBRX (and 
you're probably right) but I have reason to think that, at least for 
x86, this approach could prove very effective. Oh and of course, what 
works for flashrom, would also work for SerialICE and other goodies...

Regards,

/Pete

[1] 
http://code.google.com/p/akeo/source/browse/ubrx/scripts/intel_hello_world_pIIIs1.txt
[2] http://code.google.com/p/akeo/source/browse/ubrx/USAGE