Announcement

Collapse
No announcement yet.

status of r600/r700 AGP support?

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • status of r600/r700 AGP support?

    I've tried KMS-support for r600/r700 cards with the 2.6.32 release candidate and can say, it looks nice. 2D and 3D acceleratioand even UTS/DFS seemed to work with my AGP card. But after a while I got again screen corruptions like before KMS and UTS/DFS enabled. Additionally I found now some entries from the edac driver in the log:

    Oct 13 09:40:50 datengrab kernel: Northbridge Error, node 0, core: -1
    Oct 13 09:40:50 datengrab kernel: K8 ECC error.

    So the radeon driver is still not usable over a longer period.
    Does anybody know how far the investigation of the AGP problems are?

  • #2
    Does it work if you disable UTS/DFS ?

    Comment


    • #3
      Originally posted by bridgman View Post
      Does it work if you disable UTS/DFS ?
      I assume still through the xorg.conf as without KMS? I'll test, but it will take some time as I don't know how to reproduce the errors reliable.

      Comment


      • #4
        Hmmm... good question. I'm not sure, hopefully agd5f or someone else will jump in.

        The Northbridge error does not appear to be graphics related - I don't think we do anything with ECC in the graphics drivers.

        Comment


        • #5
          Originally posted by bridgman View Post
          Hmmm... good question. I'm not sure, hopefully agd5f or someone else will jump in.

          The Northbridge error does not appear to be graphics related - I don't think we do anything with ECC in the graphics drivers.
          The ECC error messages only came only with 2.6.32-rc4 and KMS. So either the edac driver catches more MCEs through the updates the driver got, or it's really triggerd by KMS. I can't really test this, since it needs 2.6.32 for KMS and the fglrx doesn't work with this version.

          I have a Opteron 144 on aTyan Tiger K8W S2875 with enabled ECC for RAM and Caches working. So it's not neccessary that KMS is doing anything with ECC. It "only" corrupts the data which is recognized by the hardware und reported by the edac driver (just a thought).

          Comment


          • #6
            If you disable KMS at boot does the error still occur ?

            Comment


            • #7
              Originally posted by bridgman View Post
              If you disable KMS at boot does the error still occur ?
              Will give it a try later. For now I've disabled UTS/DFS (options in xorg.conf work) and try if the error still occur.

              PS: Disabling UTS/DFS didn't help:

              Northbridge Error, node 0, core: -1
              K8 ECC error.
              Northbridge Error, node 0, core: -1
              K8 ECC error.

              and again screen corruptions: http://www.stud.tu-ilmenau.de/~johi-...chirmfoto1.png

              PPS: With KMS enabled and UTS/DFS disabled (at least the log says so), the system seems to be more responsible and fluid than with UTS/DFS enabled. With KMS disabled, UTS/DFS are disabled by default for AGP cards. The system is then really really slow and nearly unusable. X eats more than 80% CPU. That's the reason why I was still using fglrx. No more ECC errors since I've disabled KMS for now.
              Last edited by PuckPoltergeist; 10-13-2009, 06:58 PM.

              Comment


              • #8
                KMS + RV670 AGP is going well for me so far. I haven't seen any corruption yet, and I would if dfs is used with ums.

                I also get a game perf boost, in enemy territory as much as X3 on a complex scene. This is compared with ums + agp as well as kms pcie - quite suprising really, though I think there must be some issue with ums + agp that is slowing things (maybe having to use no_wb=1 doesn't help - but then kms doesn't use wb).

                There was some uts corruption recently in drm-next that is fixed now - I didn't see it on fonts, but you could possibly be running a kernel that still has that.

                Disabling uts did fix it though - did your xorg.log confirm uts/dfs were off when you tested?

                Comment


                • #9
                  Originally posted by legume View Post
                  There was some uts corruption recently in drm-next that is fixed now - I didn't see it on fonts, but you could possibly be running a kernel that still has that.
                  I'm using linux-2.6 master. There where some checkins since I've built the kernel but nothing radeon related.

                  Disabling uts did fix it though - did your xorg.log confirm uts/dfs were off when you tested?
                  Yes, it did so.

                  Comment


                  • #10
                    Originally posted by PuckPoltergeist View Post
                    I'm using linux-2.6 master. There where some checkins since I've built the kernel but nothing radeon related.


                    Yes, it did so.
                    OK I guess it's a separate issue, though if I am looking at the right tree it looks like linus hasn't pulled the fix yet.

                    It may not show unless you have a recent git ddx as well and it only seemed to affect some image types for me.

                    Comment


                    • #11
                      little update:

                      I'm now on linux-2.6.32. It seems the screen corruptions are gone. But I still get the ECC error messages after a while

                      Northbridge Error, node 0, core: -1
                      K8 ECC error.
                      Northbridge Error, node 0, core: -1
                      K8 ECC error.
                      Northbridge Error, node 0, core: -1
                      K8 ECC error.
                      Northbridge Error, node 0, core: -1
                      K8 ECC error.

                      Comment


                      • #12
                        Originally posted by PuckPoltergeist View Post
                        little update:

                        I'm now on linux-2.6.32. It seems the screen corruptions are gone.

                        Cheered too soon. The graphic errors are back.

                        Comment


                        • #13
                          Pardon the thread necromancy.

                          Just this morning I got a "Northbridge error", accompanied by font corruption very similar to the screenshot posted by the thread starter, only on an rv280. Software is 64-bit Fedora 12 with latest updates. Hardware is an Athlon64 X2 on VIA K8T800 Pro chipset.

                          So it seems this (AGP-related?) problem isn't limited to R600.

                          Comment

                          Working...
                          X