Quote:
Quote:
I'll late do this. Submitting is useless. This is a very "dirty" patch only suitable to those, who does have their videocards on PCI:0001:xx:xx.x.
That's right, i only saw pegasos 1/2 and some old pseries with a videocards on PCI:0001
It should happen on any system where the PCI controller is split into several domains and the graphics card is not in the 0 domain. On Pegasos, the first PCI bus - domain 0 - is the PCI bus. The second PCI bus - domain 1 - is the "AGP" bus.
This is both a weird concept that was done on the Mai northbridge for Pegasos I and an intentional design decision for the Marvell northbridge (since it could be either on the Marvell..) as it makes using PC southbridges a little less crazy when they assume they are on the only PCI bus.
On the PC the PCI buses are usually split into discrete controllers, and domains are only common in extreme server environments where a system may have 16 or 32 PCI slots or something all with discrete configuration spaces in order to make it fit within the PCI standard (more commonly it'd be something like 4 to 8 on a desktop PC or low end server and there is usually very little problem fitting 8 PCI cards in the 2GB address space).
Quote:
Well, i did a patch against libdrm-2.4.19 with your suggestions and DRI now works on this peg2. Thanks.
What I am disappointed with here is that even with the new libpciaccess library for X.org and all the other technology in the kernel that changed how this works for the better, something somewhere STILL assumes that graphics adapters always live in Domain 0.
After all the fuss about it on Pegasos, the fact it went through every distro bugtracker (Ubuntu, SuSE, reported directly to X.org, Ian Romanick knew about it.. especially since he works at IBM and was responsible for pSeries!) it is surprising that it is still broken.