RFC: Changing CBMEM console to run as a persistent ring-buffer

[Julius Werner]

Hi folks,

I'm planning to make some changes to coreboot's CBMEM console with
https://review.coreboot.org/#/c/18301 that should make it easier to debug
problems spanning multiple reboots. I'm trying my best to avoid breaking
previous behavior and keeping it compatible with older code, but there'll
be some edge cases so Aaron suggested I announce it here.

The basic idea is to turn the console into a persistent ring buffer. If
coreboot finds an already initialized console in memory during boot, it
will just continue appending to it rather than begin writing from the start
of the buffer again. This means that if you run cbmem -c, the most recent
boot will always be at the bottom of the output, and you can scroll further
up to see logs from the boot before that (assuming you didn't lose DRAM
power or something). It may also give some runtime firmware components
(like we have on ARM64) a good place to log stuff and persist crash dumps
in the future.

The console buffer will fill up eventually with enough reboots, so when
that happens we will start discarding the oldest lines from the log. This
means the behavior if the log buffer is too small for even a single full
boot will become slightly different: we'll only keep the latest lines,
whereas previously we would only keep the earliest lines and a counter of
dropped characters. I don't think this should be an issue though, since we
normally should have enough buffer to fit a whole boot anyway (the default
is currently 128KB which ought to be enough for any board).

The change may also cause some hiccups if you're using a newer version of
coreboot with an older version of cbmem (or SeaBIOS or whatever else reads
the console): it will not crash and it will still print the whole log, but
if the log has rolled over (into "ring buffer mode") it will print lines
out of order. This is unfortunately the best I can do with the way current
readers are implemented. I'm of course also updating the code for cbmem so
as soon as you deploy the new version it will be able to display buffers
from both old and new coreboot versions correctly. (I'll send patches to
align SeaBIOS and the Linux memconsole driver in the same manner as soon as
the coreboot patch is approved.)

Let me know if you have any concerns about this.