Regression in the siliconmotion driver

[Paulo Cesar Pereira de Andrade]

currojerez at gmail.com wrote:
> Hi
> The current siliconmotion HEAD has some problems with non-sm501 chipsets.
> It seems this patch fixes the issues: 
> http://delcorp.org/~curro/smi/0001-Fix-XAA-ShadowFB-and-VT-switching-for-non-sm501-chi.patch

  Thanks. I should have better checked when creating the CHECK_SECONDARY
macro. And thanks for spotting the inversed logic in SMI_LeaveVT(), I
am still using a "patched" X Server that mmaps the VGA region, so it
can read/write there without crashing...

> CHECK_SECONDARY() breaks XAA acceleration when using ShadowFB, so i've
> modified it to do nothing for non-sm501 cards.

  The sources I have don't work with ShadowFB, and that macro is "expanded
in place" everywhere (I made it macro instead of several cut&paste...).

> BTW, I tried to understand better how the second texture engine works
> by looking at the Windows Direct3d implementation provided by SMI: It
> never accesses the second texture registers... It seems it doesn't
> even advertise multitexturing capabilities.
>
> I suspect the 3d engine (or the databook) is buggy and the second
> texture engine doesn't work at all...
>
> I also tried to get DMA working without success (just lockups). The windows driver didn't gave me any insight on this, as it does everything through PIO.

  Can't help much here (I only have access to two hardware
implementation versions of the smi501), but hopefully someone from
smi is reading, and can give some
advice...

Paulo