Re: MTRR patch causes system lockups

Emil Briggs (briggs@bucky.physics.ncsu.edu)
Sat, 17 Jan 1998 23:34:59 -0500 (EST)

>> The MTRR setting patch can cause a system lockup for certain
>> motherboards/bioses using the Natoma chipset. I've tested this
>> under two P6 motherboards so far and obtained the same results.
>> Heres the conditions under which the lockup occurs.
>>
>>
>> 1. write combining to the linear frame buffer is enabled
>> 2. passive release is not enabled in the PIIX3 (bit 1 offset 82h)
>> 3. access to the floppy drive at the same time as large transfers
>> to the LFB are occuring.
>>
>
>Here's a kernel patch that should fix the problem (for 2.1.75 onwards)
>
>Putting PCI-related code in with MTRR code is a little ugly, so I've
>put the fix in drivers/pci/quirks.c instead (it is a PCI quirk, after
>all).
>

You're right -- it belongs in quirks. It might be a good idea to
add a line to the configure help for CONFIG_MTRR -- let people
know that if they have a 440FX chipset and they set CONFIG_MTTR
to yes that they also need to make sure passive release is enabled.

>You'll need to say yes to CONFIG_PCI_OPTIMIZE.
>
>(CONFIG_PCI_OPTIMIZE is really a misnomer in 2.1.75 onwards. What do
>people think of CONFIG_PCI_QUIRKS, or having a CONFIG option for each
>quirk?).
>

Definitely if there's a lot of quirks to take care of.
So far it looks like this is the first one besides the bridge
optimizations.

>Please let me know whether it does the trick.
>

Works fine on 2.1.78. Not sure how many bios's have this problem.
I checked two SuperMicro motherboards (a P6SNE and a P6DNE with
AMI bios) and both of them failed to set the passive release
bit correctly.

Emil