Today, I played my new version Linux kernel 5.5.9 on WSL2. Quickly, I hit a very interesting bug. Please see the following code

1
2
3
4
5
6
7
8
9
#include <thread>
#include <chrono>
using namespace std::literals::chrono_literals;

int main() {
    auto start = std::chrono::high_resolution_clock::now();
    while (std::chrono::high_resolution_clock::now() - start < 1s);
    return 0;
}

If you run the above code, you would get the following result:

xyb@ubuntu:~$ time gcc a.c
xyb@ubuntu:~$ time ./a.out

real    0m1.011s
user    0m0.010s <-- This is wrong!
sys     0m0.000s

The user time is just 0.010s. It's a wrong value obviously. After a lot of investigation, I found it might be a bug in the latest Linux kernel. In linux, the unit of shced_clock() should be nano seconds, but in hyperv_timer.c, the unit, I believe, is 100 nano seconds.

I made a fix here: https://github.com/xieyubo/WSL2-Linux-Kernel/commit/85fc62846acb12a06b18f6e60bba849315f51d86. After this fix, re-run the above application:

xyb@ubuntu:~$ time ./a.out

real    0m1.003s
user    0m0.999s
sys     0m0.004s

The value looks right now.