Firmware Execution Model

In this page, we will explain how we design and implement firmware execution model (embedded CPU).

This model stores the number of instructions per function, which is calculated by aarch64 compiler. When the function called, corresponding instructions are added to statistics.

You can configure total number of cores dedicated to each firmware component (HIL, ICL and FTL).

• HILCoreCount: # cores dedicated to HIL functions.

• ICLCoreCount: # cores dedicated to ICL functions.

• FTLCoreCount: # cores dedicated to FTL functions.

If you set core count as zero, firmware execution latency will not added.

CPU and Core

The core is defined as SimpleSSD::CPU::CPU::Core in cpu/cpu.hh. Instruction accumulation will be performed in SimpleSSD::CPU::CPU.

CPU class will select Core which has lease utilization at time point. CPU::execute and CPU::applyLatency functions will add up instructions to Core statistics.

All functions related to I/O handling has instruction count. This value is hard-coded into the source-code (See CPU::CPU constructor). The value is auto-generated by python script, it generates total number of instructions from specified file and function. You can find script at cpu/generator/generate.py. To use script, you need aarch64-linux-gnu-g++ and aarch64-linux-gnu-objdump. For installing these program (ARMv8 toolchain), see Download toolchain.