Fast booting: coreboot versus optimized UEFI

Are there some hard numbers for coreboot initialization and bootloader
payload time (from power-up/reset to the start of the OS kernel load)?
Non-optimized UEFI starts up very slowly, in some cases slower than a
legacy BIOS, but some optimized UEFI frameworks start up quite quickly
and can begin booting within a few seconds of power on/reset.  However,
I suspect that even optimized UEFI is 2-3 slower than coreboot and a
bootloader payload.  Are there boards with both coreboot and UEFI
support where the respective boot times are compared and published on
the Internet?

Sincerely,

Ken Fuchs

[Peter Stuge]

[e-mail] wrote:
> Are there some hard numbers for coreboot initialization and
> bootloader payload time (from power-up/reset to the start of the OS
> kernel load)?

About a year and a half ago I did some numbers on ALIX.1C which is a
Geode LX800 board.

I was using Linux 2.6 as payload.

coreboot v3 ran for some 600ms until Linux start, but there was some
discussion that there was a bit of "invisible" code running before I
started measuring, which could run up to a few 100ms, because I timed
the serial output rather than the system reset signal.

//Peter

[Patrick Georgi]

Am Freitag, den 30.10.2009, 12:35 -0500 schrieb [e-mail]:
> Are there some hard numbers for coreboot initialization and bootloader
> payload time (from power-up/reset to the start of the OS kernel load)?
> Non-optimized UEFI starts up very slowly, in some cases slower than a
> legacy BIOS, but some optimized UEFI frameworks start up quite quickly
> and can begin booting within a few seconds of power on/reset.  However,
> I suspect that even optimized UEFI is 2-3 slower than coreboot and a
> bootloader payload.  Are there boards with both coreboot and UEFI
> support where the respective boot times are compared and published on
> the Internet?

It really depends on the chipset. I ran a couple of tests a while ago,
and can give you three data points (from memory):
- Via vt8454c (via cx700 chipset) boots to payload in 0.9 seconds
- Kontron 986lcd-m (intel i945) boots to payload in 2 seconds
- An AMD Fam10h board boots to payload in ~3.3 seconds

All of these were cold boots and included a full hardware detection and
init cycle (unlike the recent hack published by Phoenix, which is more
of a wakeup from suspend than a real boot).

Test setup was:
- System started from cold (power supply was plugged in just before the
test)
- no logging in coreboot
- a single character printed on serial when in the payload loader
- a tool to add timing information to serial output (eg. the one in the
SeaBIOS tree)

The serial port on the host (which ran the timing utility) emits some
junk byte when the lines are high, so I could figure out the delta
between the moment the serial port was powered on, and the moment where
the byte was printed in the payload loader routine.

That "junk byte" seemed to happen in the same instant where I started
the system (we initialize the serial port quite early).

Not the most scientific setup, but good enough for me :-)

After that measured time comes the payload, but both SeaBIOS and FILO
can be configured to load the OS very quickly. HD spinup might be a
limiting factor there.

Patrick