mirror of
https://github.com/Telecominfraproject/OpenCellular.git
synced 2025-12-29 10:00:51 +00:00
e5c69151df
The way Cortex processors handle exceptions allows writing exception
routines directly in C, as return from exception is handled by providing
a special value for the link register.
However, it is not safe to do this when doing context switching. In
particular, C handlers may push some general-purpose registers that
are used by the handler and pop them later, even when context switch
has happened in the meantime. While the processor will restore {r0-r3}
from the stack when returning from an exception, the C handler code
may push, use and pop another register, such as r4.
It turns out that GCC 4.8 would generally only use r3 in svc_handler and
pendsv_handler, but newer versions tend to use r4, thus clobbering r4
that was restored from the context switch and leading up to a fault
when r4 is used by the task code.
An occurrence of this behaviour takes place with GCC > 4.8 in __wait_evt,
where "me" is stored in r4, which gets clobbered after an exception
triggers pendsv_handler. The exception handler uses r4 internally, does
a context switch and then restores the previous value of r4, which is
not restored by the processor's internal, thus clobbering r4.
This ends up with the following assertion failure:
'tskid < TASK_ID_COUNT' in timer_cancel() at common/timer.c:137
For this reason, it is safer to have assembly routines for exception
handlers that do context switching.
BUG=chromium:631514
BRANCH=None
TEST=Build and run speedy EC with a recent GCC version
Change-Id: Ib068bc12ce2204aee3e0f563efcb94f15aa87013
Signed-off-by: Paul Kocialkowski <contact@paulk.fr>
Reviewed-on: https://chromium-review.googlesource.com/362830
Commit-Ready: Shawn N <shawnn@chromium.org>
Tested-by: Shawn N <shawnn@chromium.org>
Reviewed-by: Shawn N <shawnn@chromium.org>