Hi,
I have pushed a patch [1] to use a barrier to synchronize pending EA at the entry and exit of exception handlers in BL31(EL3S). There is an interesting discussion as seen in the gerrit review comments of the patch. It looks like this patch is just enabling early panic/crash of the system in the event of SErrors but it does come with performance penalty (due to DSBs) for common exceptions like SMC calls. We believe this needs broader discussion to understand the practical approach to handle SErrors, especially in SoCs without RAS support. Please let us know your thoughts on this patch.
[1] https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/3440
Thanks, Madhukar
tf-a@lists.trustedfirmware.org