New subject: [PATCH]Fix hopefully last stopper bug for CBFS. Plan to get rid of non-CBFS

30 Sep 2009


      Hi,
attached patch reworks the payload handling with CBFS (ie. selfboot).
We use a bounce buffer (both in CBFS and older) to keep coreboot running
while allowing the payload to use the same memory region, and this is a
tricky part of our code.
Assumptions in the code are:
The bounce buffer is at least 2x coreboot_ram size (lb_end - lb_start),
with the first half being the shadow region that the payload's data
lives in (exactly the bytes that cover the area of coreboot_ram), while
the second half exists to host a copy of coreboot while copying.
The change leads to the following behaviour:
At least 2x coreboot_ram size is looked for.
The area before the bounce buffer is assumed to be reasonably free (but
this isn't validated)
Payload segments that conflict with coreboot_ram are loaded to the
bounce buffer, probably using the memory before it, and filling up more
than coreboot_ram bytes there
Right after copying, the "out of area" bytes are copied to their final
location (around coreboot_ram).
I want to revisit the prefix/suffix calculation, as I'm not 100% sure
that it's doing the right thing (those copy operations should copy
_exactly_ the necessary set of bytes, nothing more), review is welcome
(look for "prefix" and "suffix" in selfboot.c)
The code is runtime tested on qemu-x86 (after cranking up the STACK_SIZE
to >16kb, so lzma could run), with the following scenarios:
RAMBASE = 0x4000, payload starting at 0x00100000
 (no collision between coreboot_ram and payload)
RAMBASE = 0x00100000, payload starting at 0x00100000
 (RAMBASE below or at payload starting address, this worked before)
RAMBASE = 0x00101000, payload starting at 0x00100000
 (RAMBASE above payload starting address, this fails without this patch.
  AMD boards were notorious with that one)
Pending the review of the prefix/suffix calculations (by me and
hopefully others), this is
Signed-off-by: Patrick Georgi patrick.georgi@coresystems.de
Once this is in, there are no CBFS issues that I know of, and hence I
propose to eliminate non-CBFS support in the tree soonish after that
patch is in.
I have some older patch for this, but it will likely not apply anymore,
but I can update it.
Opinions?
Regards,
Patrick