mirror of
https://github.com/Telecominfraproject/OpenCellular.git
synced 2026-01-08 16:41:55 +00:00
72481572aa
Vboot hash calculation takes ~350 ms during EC boot. Since the hash
task is higher priority than the hook task, this starves all the hooks
during boot.
We could, in theory, fix that simply by swapping the priority of the
hook and hash tasks. But then watchdog detection (in the hook task)
wouldn't detect hangs in the hash task.
A better fix (implemented here) is to convert the hashing operation to
a series of deferred function calls. This gets rid of the hash task
entirely, and allows all pending hooks and other deferred function
calls to take place between each chunk of hashing.
On STM32-based boards, we need to bump up the hook task stack size,
since hashing is called from several layers deep in the hook task
instead of at the top of its own task, but this is still a net win of
several hundred bytes of SRAM.
BUG=chrome-os-partner:24892
BRANCH=rambi
TEST=Boot EC; look for "hash start" and "hash done" debug output.
'taskinfo' shows at least 32 bytes of unused stack for HOOKS task.
'hash ro' runs properly from EC console.
Change-Id: I9e580dc10fc0bc8e44896d84451218ef67578bbe
Signed-off-by: Randall Spangler <rspangler@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/181954
In the most general case, the flash layout looks something like this: +---------------------+ | Reserved for EC use | +---------------------+ +---------------------+ | Vblock B | +---------------------+ | RW firmware B | +---------------------+ +---------------------+ | Vblock A | +---------------------+ | RW firmware A | +---------------------+ +---------------------+ | FMAP | +---------------------+ | Public root key | +---------------------+ | Read-only firmware | +---------------------+ BIOS firmware (and kernel) put the vblock info at the start of each image where it's easy to find. The Blizzard EC expects the firmware vector table to come first, so we have to put the vblock at the end. This means we have to know where to look for it, but that's built into the FMAP and the RO firmware anyway, so that's not an issue. The RO firmware doesn't need a vblock of course, but it does need some reserved space for vboot-related things. Using SHA256/RSA4096, the vblock is 2468 bytes (0x9a4), while the public root key is 1064 bytes (0x428) and the current FMAP is 644 bytes (0x284). If we reserve 4K at the top of each FW image, that should give us plenty of room for vboot-related stuff.