There were two changes. First, we need to pack the GPT header to make it
match the spec (duh). Second, there's a subtle bug in how the BIOS recovers
from corrupted headers. The EFI spec says that the primary GPT header must
be at sector 1 (counting from zero) and the secondary GPT header must be at
the last sector on the drive. The BIOS correctly looks in those locations to
find the headers. However, if the secondary GPT header is invalid (as it
usually is due to our build process), the BIOS is supposed to update it from
the primary header. In this case, rather than write to the last sector on
the drive where it just looked, the BIOS trusts the alternate_lba field of
the primary header. That field is supposed to point to the secondary header
location, but the BIOS just blindly uses it no matter where it points. The
cgpt tool wasn't initializing that field, so it pointed to sector 0, which
is the PMBR. The BIOS overwrote that, resulting in an unbootable drive.
Review URL: http://codereview.chromium.org/2844006
This fixes a number of bugs, adds a bunch of commands, and essentially makes
cgpt ready to use as a replacement for gpt. Still to do is to add commands
and options that will let it generated intentionally bad partitions, for use
in testing.
Review URL: http://codereview.chromium.org/2719008
Also includes part of LoadKernel(), which I'll split into a separate
CL. With some hacks, gets into VerifyKernel() before dying because
I'm not passing in the right key blob.
cgptlib is now pretty stable, and worth looking at. LoadKernel() less so.
Thanks,
Randall
Review URL: http://codereview.chromium.org/2438005
