I am comparing the latencies between vmalloc and vzalloc. According to the definitions of these functions, vzalloc should take more time because of filling zero, and I got a proof on Kernel 4.15, vmalloc took significatly fewer time to allocat memory. However, the latency of vmalloc significantly increase in kernel 6.2, thus the latency of vmalloc is almost identical to vzalloc. Dose anybody notice the behavior and can explain the root cause?
Attach my experiment result. In the experiment, the buffer size is 50MB. The systems are clean Ubuntu LTS 16.04(kernel 4.15) and 22.04(kernel 6.2) Each value is an average latency, and the latency was measured by ktime_get_ns(). The action for fill is to access a byte to simulate real access.
