Problem: PnP BIOS driver reports outdated information

Problem: PnP BIOS driver reports outdated information

[Thomas Hood]

The PnP BIOS interface driver builds a device list, including
resource information for each device, at init time.  Other
drivers search this list using pnpbios_find_device().

The problem is that the information in the device list may
be out of date if setpnp has been used.

Perhaps it will be said that setpnp is a "debugging tool" and
ordinarily one shouldn't change resource assignments, but
setpnp is useful for more than debugging.  E.g., with current
kernels it is necessary for me to run setpnp during the boot
sequence in order to set up certain devices properly.  
Sometimes, also, there aren't enough IRQs to go around, and
one may use setpnp to disable and enable different devices.

Thus I think that the PnP BIOS driver needs to be fixed so
that it will use and report the correct (current) resource
values.  Looking at the code, I see no reason why there
needs to be a device list at all.  Functions such as
pnpbios_find_device() could just as well scan through 
device info returned (fresh) by the PnP BIOS.  I don't 
think speed is a big issue here.  Are there other reasons
for maintaining a device list in the driver?

--
Thomas Hood
jdthood_AT_yahoo.co.uk

Re: Problem: PnP BIOS driver reports outdated information

[Alan Cox]

> think speed is a big issue here.  Are there other reasons
> for maintaining a device list in the driver?

If you query the current device status on a Vaio and some other boxes
using the 32bit API your computer dies horribly shortly afterwards.

So yes - we should be handling setpnp in the kernel, but no we can't query
the bios for the data

Alan

Re: Problem: PnP BIOS driver reports outdated information

[Thomas Hood]

Alan Cox wrote:
> If you query the current device status on a Vaio and some other boxes
> using the 32bit API your computer dies horribly shortly afterwards.
> 
> So yes - we should be handling setpnp in the kernel, but no we can't query
> the bios for the data

How about this: pnpbios functions that scan the device list
optionally (depending on a flag of some sort, set according to
whether or not one has an evil BIOS) cause the list to be rebuilt
first.  This would require only slight changes to the code.

I would offer a patch, but I don't know how the aforementioned
flag should be implemented.

Thomas Hood
jdthood_AT_yahoo.co.uk

Re: Problem: PnP BIOS driver reports outdated information

[Alan Cox]

> How about this: pnpbios functions that scan the device list
> optionally (depending on a flag of some sort, set according to
> whether or not one has an evil BIOS) cause the list to be rebuilt

We can use DMI to figure out if the BIOS is likely to be problematic
A DMI match on Intel or Vaio should cut out most BIOSes that either explode
on boot or have the weird corrupting case.

> I would offer a patch, but I don't know how the aforementioned
> flag should be implemented.

Use the existing vaio flag that is used by the vaio drivers currently -
"is_sony_vaio_laptop"