Some crashes in the signal handler are the just created "MDEV-30613 output_core_info crashes in my_read()" with a probable cause. Doesn't help the original crash reason however.
A single thread backtrace isn't sufficient on errors like: " InnoDB: innodb_fatal_semaphore_wait_threshold was exceeded for dict_sys.latch" With MTR_PRINT_CORE=detailed in the environment will do multiple threads and be more likely to identify deadlocks between processes (or if the builder is just very slow at this point). Disabling performance schema -DPLUGIN_PERFSCHEMA=NO on s390x might be a way to get some breathing space on this optional feature to resolve later as the unmap crashes seem firmly in this feature's shutdown.
