On 10/7/24 16:18, Rafael J. Wysocki wrote: > On Mon, Oct 7, 2024 at 4:44 PM kernel test robot <oliver.sang@xxxxxxxxx> wrote: >> >> >> >> Hello, >> >> kernel test robot noticed a 51.9% regression of fsmark.app_overhead on: > > What exactly is fsmark.app_overhead? What does it measure? "App overhead is time in microseconds spent in the test not doing file writing related system calls." So the loop is: /* * MAIN FILE WRITE LOOP: * This loop measures the specific steps in creating files: * Step 1: Make up a file name * Step 2: Creat(file_name); * Step 3: write file data * Step 4: fsync() file data (optional) * Step 5: close() file descriptor */ And it gets the timestamps before and after each syscall. It then subtracts all those times (spent in syscalls) from the total time. I'll go take a look tomorrow, what's interesting is that the exit-latency (which is the thing supposed to cause regressions here) should be time spent blocking on syscalls. If that is all correct that's the part that shouldn't regress on cpuidle.
