OK. After some digging in core dumps and sources seems this is upstream regression.
In recent versions they've added some platform-depended definitions to mysys/my_rdtsc.c
$this->bbcode_second_pass_code('', '
385,386d364
< #elif defined(__GNUC__) && defined (__aarch64__)
< mti->cycles.routine= MY_TIMER_ROUTINE_AARCH64;
')
It's defined as #29 somewhere else, so basically it means "somewhere in 500 Mb of code there is a function with number 29 contains illegal aarch64 instruction". Yet have no time to find it
which I suppose causes crash immediately at start:
$this->bbcode_second_pass_code('', '
Reading symbols from /mnt/sda1/src/mariadb/dist.atomic/11/usr/bin/mariadbd...
(No debugging symbols found in /mnt/sda1/src/mariadb/dist.atomic/11/usr/bin/mariadbd)
[New LWP 132777]
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/usr/lib/libthread_db.so.1".
Core was generated by `/usr/bin/mariadbd'.
Program terminated with signal SIGILL, Illegal instruction.
#0 0x0000005583be6524 in my_timer_init ()
(gdb) bt
#0 0x0000005583be6524 in my_timer_init ()
#1 0x000000558335ebe0 in mysqld_main(int, char**) ()
#2 0x0000007f911e2d90 in __libc_start_main () from /usr/lib/libc.so.6
#3 0x00000055833527d0 in _start ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)
')