[Intel-gfx] G45 BIOS mis-initializes stolen GTT PTEs

[Eric Anholt]

On Mon, 2008-10-13 at 23:50 -0700, Keith Packard wrote:
> Ok, so I've managed to get my G45 running without DRI.
> 
> What I discovered is that the very last PTE for stolen memory (pages
> allocated by the BIOS for graphics use) was not initialized by the BIOS
> correctly. When the graphics hardware tried to access that page, it
> would get a segfault and lock up until it was reset.
> 
> Here's a patch which finds that PTE and fixes it if it is broken. If
> your G45 is crashing randomly after startup, please give this patch a
> try and see if it works.
> 
> As you can see, it's an ugly hack that doesn't check to make sure you're
> running on a G45, so it's not exactly ready for upstream, but it should
> detect and repair this specific bug.

I also don't think it's safe -- if the bios is reserving that page for
something, we probably don't want to go stepping on it.  See the pair of
patches I posted for an (I believe) improved fix.

I've been debating whether we want the "is there something we're calling
mapped to stolen that isn't mapped" part of your patch should go in agp.
At this point, I'm thinking that it's better to just have the userland
tool for gtt inspection and only make people that are writing agp
drivers run gtt inspection.

-- 
Eric Anholt
eric at anholt.net                         eric.anholt at intel.com


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
URL: <http://lists.freedesktop.org/archives/intel-gfx/attachments/20081014/de39ee57/attachment.sig>