An Updated ATI Kernel Mode-Setting Driver

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

  • agd5f
    replied
    The closed source memory manager is designed around a completely different driver stack, all parts of the stack (2D, 3D, etc.) have have to be substantially rewritten to use it.

    Leave a comment:


  • Ant P.
    replied
    On the other hand, maybe VIA would... their drivers on any OS make fglrx look like NASA software by comparison.

    Leave a comment:


  • nanonyme
    replied
    Hasn't nVidia had a kernel memory manager longest of all the drivers? You'd think it's so well optimized they wouldn't care what their competitors do.

    Leave a comment:


  • Ant P.
    replied
    Originally posted by Xavier View Post
    So, why hasn't that wonderful memory manager been used in the OSS radeon driver, instead of TTM ?
    Because it'd probably have to be 99% rewritten to fit in with the OSS API. Also because they're paranoid that nVidia will just rip it off.

    Leave a comment:


  • nanonyme
    replied
    Originally posted by Xavier View Post
    So, why hasn't that wonderful memory manager been used in the OSS radeon driver, instead of TTM ?
    Well, they didn't at any point say they want to opensource fglrx, did they? Besides, who are they to deny radeon developers the pleasure of yanking their hair out in frustration while trying to figure out how a memory manager should work.

    Leave a comment:


  • Xavier
    replied
    Hi,

    Originally posted by bridgman View Post
    There is very little compatibility between the open and Catalyst driver internals. The Catalyst drivers have had a high performance kernel memory manager for a number of years, with similar capabilities to GEM/TTM plus a few years of performance tuning, and bringing that memory manager into fglrx was one part of the new 3D stack we released back in September 07.

    The memory manager is tightly integrated with both the 3D drivers and the low-level GPU management across all OSes, so switching to GEM/TTM is not a practical option.
    So, why hasn't that wonderful memory manager been used in the OSS radeon driver, instead of TTM ?

    Leave a comment:


  • highlandsun
    replied
    Originally posted by nanonyme View Post
    Not all patches are meant for end-users, some are actually just parts of the development process.
    Understood. But usually, developers share patches with each other by directly pulling from each other's git repos. Since this was made available as a standalone patch, it seemed to be aimed at a slightly wider audience.

    Anyway, after running it for a short period of time my system started to slow down. dmesg showed that it was being flooded with messages:

    Aug 13 21:23:30 violino kernel: [ 740.664131] [drm] Emit age: 26591 (Current: 26586)
    Aug 13 21:23:30 violino kernel: [ 740.664131] [drm] Emit age: 26592 (Current: 26586)
    Aug 13 21:23:30 violino kernel: [ 740.664131] [drm] Emit age: 26593 (Current: 26586)

    it was pegging klogd... So I've gone back to no KMS, which is better anyway since it has acceleration...

    Leave a comment:


  • agd5f
    replied
    Originally posted by highlandsun View Post
    And a warning - it appears that after you've loaded the radeon kernel module once, it's not safe to load it again with modesetting enabled. I was able to unload it and reload it with modeset=0, which I needed to do a few times because I didn't have all the KMS pieces in place. But the next time I tried to unload and reload it (with modesetting active) I just got a blank screen and everything was hung, no SysRq or anything...
    You need to unbind the console from the fb device, e.g.,
    echo 0 > /sys/class/vtconsole/vtcon1/bind
    You will lose your console however, so best to do this from a second PC over ssh.

    Leave a comment:


  • nanonyme
    replied
    Originally posted by highlandsun View Post
    That says it provides no hardware acceleration. Aside from testing purposes or curiosity, why would someone else want to use this patch?
    Not all patches are meant for end-users, some are actually just parts of the development process.

    Leave a comment:


  • highlandsun
    replied
    It applied OK on top of my 2.6.31-rc6 tree. There was one already-applied patch (adding the RS880 pciid) which I ignored. Now to see if it actually does anything useful...

    edit: after I rebuilt libdrm with --enable-radeon-experimental-api and rebuilt xf86-video-ati, X started working. But, like the patch says, no acceleration of any kind. No Xv, no EXA.

    And a warning - it appears that after you've loaded the radeon kernel module once, it's not safe to load it again with modesetting enabled. I was able to unload it and reload it with modeset=0, which I needed to do a few times because I didn't have all the KMS pieces in place. But the next time I tried to unload and reload it (with modesetting active) I just got a blank screen and everything was hung, no SysRq or anything...
    Last edited by highlandsun; 14 August 2009, 12:59 AM.

    Leave a comment:

Working...
X