Looks like boot-timestamps.c first tries to read the ACPI FPDT table (it seems acpi-fpdt.c uses the "OS Loader StartImage Start" and "ExitBootServices Exit" fields), but if that's unavailable, then it uses timestamps stored by systemd-boot in EFI variables (/sys/firmware/efi/efivars/LoaderTime*). The latter seems to be estimating it from RDTSC (src/boot/efi/util.c).
On Wed, Oct 13, 2021 at 3:05 PM jiansong Xu <jiansong...@oracle.com> wrote: > *This is bios from oracle, we noticed that the firmware times reported > were incorrect.* > > *Startup finished in** 59min 7.944s** (firmware) + 33.051s (loader) + > 4.428s* > *(kernel) + 1min 18.870s (initrd) + 44.494s (userspace) = 1h 1min 48.789s* > > > > *May I ask what is the source of the **firmware** ? is it read from acpi > table? * > > > -- Mantas Mikulėnas