[coreboot] cbfs for safe flashing
Myles Watson
mylesgw at gmail.com
Wed Jul 14 14:47:33 CEST 2010
> -----Original Message-----
> From: coreboot-bounces at coreboot.org [mailto:coreboot-bounces at coreboot.org]
> On Behalf Of Ed Swierk
> Sent: Wednesday, July 14, 2010 12:45 AM
> To: Coreboot
> Subject: [coreboot] cbfs for safe flashing
>
> Hi folks,
>
> I'm using Coreboot to implement an old-school Linux-as-bootloader for
> a prototype board, which has an 8-MByte SPI flash attached to an AMD
> SB800 southbridge. I'd like to take advantage of that nice roomy
> flash, as well as the normal/fallback capabilities of Coreboot and the
> layout and partial-rewrite features of flashrom, to provide a safe
> firmware upgrade path for end users.
>
> Ideally I'd divide up the flash so that the normal Coreboot+payload
> are separate from the fallback Coreboot+payload and bootblock. The end
> user would only rewrite the normal Coreboot+payload. If the rewrite
> fails leaving the normal area of the flash completely scrambled, the
> board would still boot using the fallback Coreboot+payload. And for
> extra credit, I'd try to put the critical fallback bits in an area of
> the flash that can be turned read-only once it's programmed at the
> factory.
>
> Following the flashrom layout format, here's how I imagine
> partitioning the flash:
>
> 00000000:003fffff normal+payload
> 00400000:ffffffff fallback+payload+bootblock
Wouldn't you want three areas? I'd think you'd want another for just the
bootblock.
>
> Following Patrick's helpful recipe
> (http://www.coreboot.org/pipermail/coreboot/2010-February/055944.html)
> I figured out how to build a rom image with a cbfs containing normal
> and fallback files. Unfortunately the build system insists on placing
> each new file at the beginning of the free space; I didn't see any
> obvious way to convince cbfstool to leave a gap after normal+payload
> and place the fallback files at 00400000. I hacked around this by
> creating a dummy pad file and adding it before the fallback files. Is
> there a better way to do this?
It seems like there used to be a way to specify an address, which would
force CBFS to add the empty space for you.
> I'm thinking of extending the cbfs format to allow more than one
> top-level header in the bootblock. In my scheme, there would be two:
> one pointing to the first normal file, and another pointing to the
> first fallback file. Searching for a file would involve scanning each
> series of files linked from the headers in the bootblock, and would
> simply terminate if no matching filename were found in any of them.
>
> This implies changing cbfstool and any code in Coreboot that touches
> cbfs. Before I jump off the deep end, is there an easier or better way
> to achieve this?
Instead of erasing one image to test it, you could add blank images with the
same name. That would allow you to test the case of a bad flash. The case
where CBFS is corrupted shouldn't happen too often, and when it does it is
only inconvenient to have it take a long time to boot. The fact that it
recovers will make up for that :)
Thanks,
Myles
More information about the coreboot
mailing list