Fix non-x86 __PRE_RAM__ assertions and add FATAL_ASSERTS Kconfig option

This patch fixes a bug that caused non-x86 boards to use the poor man's
assert() version with a lot more instructions per invocation and
hexadecimal line numbers in __PRE_RAM__ environments. This was really
just an oversight in the ARM port... even x86 uses a proper printk() in
most cases (those with CAR) and there's no reason not to do so on the
generally even more flexible SRAM-based architectures.

Additionally, it adds a new Kconfig option to make failed assertions and
BUG() calls halt again. This seems to have been the original intention,
but was commented out once out of fear that this might prevent
production systems from booting. It is still a useful debugging feature
though (since otherwise assertions can easily just scroll past and get
overlooked), so the user should be able to decide the this based on his

(Also changed error messages for both to include the word "ERROR", since
grepping for that is the most sophisticated way we currently have to
detect firmware problems. Some automated Chromium OS suspend tests check
for that.)

TEST=Booted Jerry. Compared binary sizes before and after, new version's
bootblock is some ~600 bytes smaller.

Change-Id: I894da18d77e12bf104e443322e2d58e60564e4b7
Signed-off-by: Patrick Georgi <>
Original-Commit-Id: 6a5343124719c18a1c969477e3d18bda13c0bf26
Original-Change-Id: I0268cfd67d8c894406b18bb3759a577944bcffb1
Original-Signed-off-by: Julius Werner <>
Original-Reviewed-by: Vadim Bendebury <>
Original-Reviewed-by: Aaron Durbin <>
Tested-by: build bot (Jenkins)
Reviewed-by: Edward O'Callaghan <>
diff --git a/src/Kconfig b/src/Kconfig
index 6f7f459..a6032c7 100644
--- a/src/Kconfig
+++ b/src/Kconfig
@@ -808,6 +808,12 @@
 	  If enabled, coreboot will wait for a GDB connection.
+	bool "Halt when hitting a BUG() or assertion error"
+	default n
+	help
+	  If enabled, coreboot will call hlt() on a BUG() or failed ASSERT().
 config DEBUG_CBFS
 	bool "Output verbose CBFS debug messages"
 	default n