With -flto enabled, the compiler can optimize code across multiple files, it may be able to keep more variables in registers, so some code might not pick up changes to some variables which are modified in an interrupt handler, and not declared as volatile.
Without -flto, every function call to another source file acts as a memory barrier, the compiler must save all global values back to memory, and reload them afterwards, because it has no way to know whether the called function changes them as a side effect.
The -flto option is deliberately hidden in STM32CubeMX, because HAL functions use volatile at the wrong places, as if the developers who have never understood its purpose had to fill some quota. Frequent calls to functions in another source file significantly increases the chance that this mess works, as long as these calls are all implicit memory barriers.
