When zeroing a range of normal memory, it is much more efficient to use the DC VZA instruction (if the implementation supports it) than store instructions . Trusted Firmware currently doesn't make use of this instruction. The DCZID_EL0 register should be read to determine if DC VZA is supported and the block size to use.

In particular:

• The memset() implementation in lib/stdlib/mem.c is often called with zero as the val argument, which results in a loop of STRB instructions.
• The zeromem16 implementation in lib/aarch64/misc_helpers.S uses STP and STRB instructions. This is used for zeroing out the BSS and coherent RAM sections.

There may be other places.