Arch Linux User Repository

@jhernberg - thanks, that would be helpful * ...and agreed -> people should speak up, as we have no way of how things are working, otherwise. (Morgan Cox should be also testing / updating this package, as soon as we can get some proper feedback too. it's been flagged for quite a while). @JH -> i *can* see how many people have downloaded from SF.net, my source package for 325.15 has been downloaded 14 times, at last check. (yesterday). I've had postive feedback from one person (via email). I've also posted to nvidia devtalk forums (about the changes in 325.xx for -rt / looking for clarification from nvidia-devs); https://devtalk.nvidia.com/default/topic/572468/linux/nvidia-325-15-linux-rt-old-amp-amp-new-nvidia-rt-patch-methods-questions-about-nvidia-installer/ nvidia-rt (325.15); http://sourceforge.net/projects/l-proaudio/files/nvidia-rt-325.15-1.src.tar.gz/download <just to make sure the link is visable / not lost in comments>

@Ninez: I'll try your fix when I have access to that machine again in a few weeks. And I'll second Ninez question, please speak out if you have success. It's really annoying to only get feedback when something is broken, especially as we can't see how many people download a build script and use it...

* Has anyone here beyond myself and jhernberg tested 325.15 with linux-3.10.x-rt? and if so, could they give some feedback? @jhernberg: I think we still have to patch the driver, I don't think disabling it alone is enough; yesterday, while upgrading linux-l-pa to 3.10.6-rt3, i cleaned up my (nvidia) pkgbuild && re-installed nvidia and ran into a problem; Now, before i found out about disabling the PREEMPT_RT conftest (from you/here), i had made a small patch (for nv-linux.h, by comparing 319.32 to 325.15). ~ i never removed this patch (or rather forgot to), until i re-installed nvidia -> it turns out that the non-patched driver experienced lockups on both machines (within 2min on one box, about 25min on the other). I then, re-enabled the patch and voila, nvidia works as expected. anyway, i have posted a source package for nvidia-rt (based on Det's && Arch repo nvidia package) to my SF.net page; http://sourceforge.net/projects/l-proaudio/files/nvidia-rt-325.15-1.src.tar.gz/download 1. patch nv-linux.h to support CONFIG_PREEMPT_RT_FULL explicitly. 2. we use IGNORE_PREEMPT_RT_PRESENCE=1 to override nvidia's PREEMPT_RT conftest (which allows the driver to compile). cheerz

@jhernberg; lol. I didn't get any response from nvidia. ;) -> that is why i sent reports on 3 different occassions. (all "atomic while scheduling bugs") and i take it that since i am seeing no call traces in dmesg anymore about that - these longing standing issues (on my H/W, which were non-critical) appear to have been fixed... you are probably correct, it's likely up to them to fix these issues though. I think nvidia (still) offers both email and their "devtalk" -> linux forum. I would just do both - You could link to the forum post here, maybe someone is having same issues and could pitch/pipe in :)

@Ninez: I will probably try to pursue this with NVIDIA when i get some time. It really is an encouraging sign that they have added the old rt patch, and just disabled it from building on rt by default. Where did you meet with the biggest response from nvidia devs (i suspect they are the only ones that can fix this). I can also point them to a few other things that spike scheduling latencies that they might want to look into..:)

I got tired of not knowing, so even though i ought to have been doing other things, i pulled the gtx 650ti and installed my old 8600gts. I had no problem at all starting Steam or XBMC, so I think my problems are with the 650ti, nvidia and a realtime patched kernel. Seems to indicate that my modified script below most likely works for the people that had working nvidia under linux-rt.

@jhernberg: thx for info. i knew the rt patch wasn't needed (pretty obvious), but disabling the PREEMPT_RT test, wasn't as obvious to me. as far as your H/W - i wonder if it is the cards or possibly mobo? ... I just tested out XBMC here, no lockups. but i'm not using KDE (i use compiz, don't know what impact, if any that would have). You may want to collect info for nvidia, regarding this. Over the last year (or so) i have contacted nvidia on 3 separate occasions, regarding nvidia-rt; in nvforums, then by email then in their devtalk forums. I linked them to RT patch + logging info (traces, etc), each time. ~ you never know, they may actually fix the problem ;)

FWIW, I can boot my q6600 and gtx650ti into kde 4.10.5 and i can even leave chromium playing youtube videos, i can also play skyrim in wine/steam, watch a movie with vlc, etc But any attempt to start something like Steam (native client) or XBMC will hang the screen completely, though ssh login still works. This makes it kind of useless to me, since this is my game machine / media centre and i really don't need realtime on it. I'll keep on trying, maybe someday nvidia will fix it (or i'll figure out that i broke linux-rt). If i find the time someday, i'll pull that 650ti and try an older card.