Uwe Hermann wrote:
Hi,
On Tue, May 22, 2007 at 03:14:42PM +0100, Ceri Coburn wrote:
I wanted to limit the number of temp's used as I didn't want ROMCC complaining about no more registers available.
I was seeing the same problem when I tried to integrate your patch -- romcc runs out of registers.
The method with the register_values[] array is used in many other northbridges, I think it should work (in general).
The problem with the 440BX register_values[] is that I messed it up ;-) The bitmasks are set incorrectly (or rather: not at all) currently, that needs fixing. As I understand it, it doesn't matter much if we do 8/16/32 bit reads and writes on PCI. We _do_ read too much stuff if we always do a 32bit read, that true. But if the bitmasks are set correctly, only those bits we _want_ to change will be changed. All other bits (i.e., those we read but actually we didn't want to read) will remain the same.
I _think_ this should not cause any problems, but feel free to correct me.
My bad, I'm still getting to grips with the common parts of the code atm. Well if that way works, you know what they say. If it's not broke, don't fix it.
Anyway, the PAM changes are in the repository now, can you verify that it works for VMWare? It does work on real hardware here...w
I can confirm that the PAM register changes work in VMWare, with the previous values memtest fails within the lower memory region (as expected), with the new values, VMWare passes memtest without any errors.
Ceri