[LinuxBIOS] LinuxBIOS messing with CMOS on reboot?

[Stefan Reinauer]

Hi,

I'm seeing a weird problem while switching back and forth between 
a legacy bios and LinuxBIOS.

I have a 256 byte cmos image created with cmos_util and a 512k bios
image for each firmware flavour.

Switching from legacy bios to LinuxBIOS works fine using flashrom and
cmos_util.

Switching the other way round always leaves legacy bios with a wrong
checksum on the first boot.

writing the cmos from the file afterwards and rebooting is fine, so it's
not the cmos image that is wrong.

Could LinuxBIOS change the CMOS when doing a reboot?

Or is legacy bios doing some very weird things like safing additional
data in mystic places?

Stefan

[yhlu]

should be second thing.

YH

On 12/3/05, Stefan Reinauer  wrote:
> Hi,
>
> I'm seeing a weird problem while switching back and forth between
> a legacy bios and LinuxBIOS.
>
> I have a 256 byte cmos image created with cmos_util and a 512k bios
> image for each firmware flavour.
>
> Switching from legacy bios to LinuxBIOS works fine using flashrom and
> cmos_util.
>
> Switching the other way round always leaves legacy bios with a wrong
> checksum on the first boot.
>
> writing the cmos from the file afterwards and rebooting is fine, so it's
> not the cmos image that is wrong.
>
> Could LinuxBIOS change the CMOS when doing a reboot?
>
> Or is legacy bios doing some very weird things like safing additional
> data in mystic places?
>
> Stefan

Stefan Reinauer  writes:
> Hi,
>
> I'm seeing a weird problem while switching back and forth between 
> a legacy bios and LinuxBIOS.
>
> I have a 256 byte cmos image created with cmos_util and a 512k bios
> image for each firmware flavour.
>
> Switching from legacy bios to LinuxBIOS works fine using flashrom and
> cmos_util.
>
> Switching the other way round always leaves legacy bios with a wrong
> checksum on the first boot.
>
> writing the cmos from the file afterwards and rebooting is fine, so it's
> not the cmos image that is wrong.
>
> Could LinuxBIOS change the CMOS when doing a reboot?

Possibly.  The only safe way after a bios flash is to toggle the power.
> Or is legacy bios doing some very weird things like safing additional
> data in mystic places?

There are a couple of possibilities.
- cmos_util needs to be explicitly told not to do the linuxbios checksum
   calculation.
- cmos_util has had problems when asked to flash all of the cmos options.
  (Sigsegv ...) 
- The high 128 bytes are only moderately standard so it may be you have
  a board that stores them differently.  We should be ok for intel and
  amd chipsets.

And of course other mystic locations but I would exhaust the other possibilities
first.

Eric

[Stefan Reinauer]

* Eric W. Biederman  [051203 22:28]:
> > writing the cmos from the file afterwards and rebooting is fine, so it's
> > not the cmos image that is wrong.
> >
> > Could LinuxBIOS change the CMOS when doing a reboot?
>
> Possibly.  The only safe way after a bios flash is to toggle the power.

Good point. I will try that to see if it changes the behaviour
> There are a couple of possibilities.
> cmos_util needs to be explicitly told not to do the linuxbios checksum
>  calculation.

i've used file_to_cmos and cmos_to_file to switch. This should not do
any checksumming at all..
> cmos_util has had problems when asked to flash all of the cmos options.
> (Sigsegv ...)

it seems to work fine. 256 bytes (which should be less as an option,
because I don't necessarily want to set back the clock) Also, rereading
the file after amibios corrected the checksum presents an identical file
(only the time changed)
> The high 128 bytes are only moderately standard so it may be you have
> a board that stores them differently.  We should be ok for intel and
> amd chipsets.

It's an AMD k8 chipset. I wondered whether I failed to look at even
higher cmos values via 74/75 or something. but then again, LinuxBIOS
never touches anything except the low 128 bytes.
> And of course other mystic locations but I would exhaust the other possibilities
> first.

Power toggling seems a good start. Closed source proprietary software
sucks for customers. Unfortunately it might rather make them think in terms
of "bios is always complicated"

Stefan