I now have a set of several patches which pass the abi-validation , 1 patch
of checkpatches.sh to prevent future panic instances
and the set of abi-breaking changes which are included the init sequence.
1.
I assume it is best to bundle all to a single patchset. Please correct me
otherwise
2.
Trying to work around the abi in this phase
What is your take regarding adding a state to the init phase:
int rte_get_legacy_panic_state();
void rte_move_to_legacy_panic_state();
This will allow preserving abi for these few functions, replacing for
example the currently abi-breaking
if (eal_thread_init_master())
return -1
with
eal_thread_init_master()
if ( rte_get_legacy_panic_state())
return -1
while calling rte_move_to_legacy_panic_state() from within these void
functions where panic takes place today.
This can also partially serve the cases where panic is called from within
an interrupt handler, and have no applicative context to return to.
thanks
/Arnon
