Re: radeon DRM PCI/AGP detection

[Ian R. <id...@us...>]

Jon Smirl wrote:

> drmAgpAcquire() will always fail on a PCI card, right? It would be
> much better to use a DRM call to tell the difference.

For a "true" PCI card, yes.  This is one of the changes I made to fail
more gracefully on PCI G450 cards.

> This will still allow an AGP card to run in PCI mode if you really
> want to. drmAgpAcquire() - find out if you have AGP - then ignore the
> result.

The way I'm doing it for MGA is a bit different.  I've added a "DMA
bootstrap" ioctl.  User-mode tells the kernel what DMA configuration
(e.g., AGP mode, size of buffers, etc.) it wants.  The kernel then does
what it can and tells user-mode what happend.  In this case, if you
request a non-zero AGP mode, but PCI DMA is used, the returned AGP mode
will be zero.

I should have this patch out tomorrow or Friday.