I have further narrowed my test case for arm alignment issues, I've
commented out the pacemaker service from corosync.conf and am able to
produce my alignment/bus error using the flatiron/test/testcpg  executable.

>From what I've learned of how the dispatch stuff works, and what I saw
with my previous debugging of pacemaker/crmd core files, it seems that
the first ipc message is always aligned ok in
ipc_instance->dispatch_buffer, since the original mmap() returned
aligned memory of dispatch_size(for both chunks in circular_mmap).

But, when there is a second ipc message written after the first one, at
an index of ipc_instance->control_buffer->read bytes into the original
dispatch_buffer mmap'd memory, the index in bytes is basically  the
length of the first message which means the second message is not
aligned properly for access on my arch.

To sum up my findings, it seems that ipc_instance->dispatch_buffer needs
to have the start of each additional message (after the first one)
aligned . Or, am I just not understanding what's going on?

Another question is whether the preference would be to ignore arm arch
alignment issues with flatiron and focus on the new libqb stuff on arm?

Greg
_______________________________________________
Openais mailing list
[email protected]
https://lists.linux-foundation.org/mailman/listinfo/openais

Reply via email to