Re: Interest in nvidia-driver debugging on 5-*
On 13 Jun 2003, Samy Al Bahra wrote: Hey, I believe the fix is as simple as saving and restoring %gs in the nVidia driver, and/or not letting the nVidia use %gs at all (segment registers generally belong to the OS, in any case, and stealing a CPU register of any kind for exclusive use of a driver is a pig trick). Please take a look at http://www.minion.de/ Christian Zander has written a patch to deal with the gs restoration problem and other issues. It is well tested and adopted. Maxime Henrion has also written several patches which may be found in the ports system (x11/nvidia-driver). I think Terry is jumping to conclusions on what Evan's problem could be. I have this seen this cause application to crash at exit (if using OpenGL) but never a crash in kernel-space from the nVidia drivers. No, I don't think Terry is jumping to conclusion. Read the archives of this problem. The Nvidia drivers and OpenGL for FreeBSD won't work with -current and any threading library other than libc_r. -- Dan Eischen ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
Samy Al Bahra wrote: I think Terry is jumping to conclusions on what Evan's problem could be. I have this seen this cause application to crash at exit (if using OpenGL) but never a crash in kernel-space from the nVidia drivers. 1) It works with libc_r, but not with libthr and not with libkse. 2) The %gs register is not used by libc_r, but is used for threads context for the other two. 3) Most of the other code is shared between the libraries. Draw your own conclusions... 8-). Also read the recent post about nVidia to the threads list about %gs. -- Terry ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
Daniel Eischen Wrote: No, I don't think Terry is jumping to conclusion. Read the archives of this problem. The Nvidia drivers and OpenGL for FreeBSD won't work with -current and any threading library other than libc_r. I didn't deal with any users having kernel crashes from this %gs restoration bug only OpenGL-linked applications crashing at exit (such as QT applications, do look at minion.de for a better description). You're incorrect in many premises here (as Terry kindly pointed out). Terry Lambert Wrote: Draw your own conclusions... 8-). Also read the recent post about nVidia to the threads list about %gs. It's kind of hard with no debugging information :) Will look at that thread though. -- +---+ | Samy Al Bahra | [EMAIL PROTECTED] | +---+ Arabeyes.org Kerneled.com FreeBSD.org ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
Samy Al Bahra wrote: Daniel Eischen Wrote: No, I don't think Terry is jumping to conclusion. Read the archives of this problem. The Nvidia drivers and OpenGL for FreeBSD won't work with -current and any threading library other than libc_r. I didn't deal with any users having kernel crashes from this %gs restoration bug only OpenGL-linked applications crashing at exit (such as QT applications, do look at minion.de for a better description). You're incorrect in many premises here (as Terry kindly pointed out). Uh, Daniel and I are in total agreement on this one; I don't know what I'm supposed to have pointed out, but I was basically going off a conversation between Daniel and an nVidia user, and my understanding of the threads library code in the three libraries. BTW: a corrupted %gs can crash a user space program that's threaded and expects a different value, without actually crashing the kernel: it all depends on who trashes who's %gs first, since if you crash in user space from a %gs set by the kernel to an unexpected value, then you aren't going to get the opportunity to set it to a value that will crash the kernel becase, well, having crashed, you aren't runnung any more to execute the code to do it. 8-). -- Terry ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
hello, Any chance someone can support nforce2's chipset agp? Not all of us own nvidia graphics cards :-( I don't know how hard it is to get this working on freebsd. here's some code: http://groups.google.com/groups?hl=enlr=ie=UTF-8oe=utf-8threadm=20030524205432.GA17179%40suse.de.lucky.linux.kernelrnum=18prev=/groups%3Fq%3Dnvidia%2Bagp%2Bnforce2%2Blinux%26num%3D100%26hl%3Den%26lr%3D%26ie%3DUTF-8%26oe%3Dutf-8%26sa%3DN%26scoring%3Dd -- Bruno Miguel Afonso Biological Eng. student. D.E.Q. @ I.S.T. - Portugal ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
On 2003-06-12 11:51 +, Evan Dower wrote: I am highly interested in getting nvidia-driver to work well for FBSD, and in particular my own computer. I do not have the expertise to hack on either it or the kernel (yet), but I am eager to provide debugging information. nvidia-driver frequently crashes my machine, so I expect that my input could be valuable. Is anybody interested in my assistance? Is anybody actively working on/with/around the nvidia-driver, or does the license make that infeasible? Thanks a bundle, Evan Dower Undergraduate, Computer Science University of Washington NVIDIA is handling everything to do with the driver. Expect a new release shortly *cough*. Until then, perhaps some of the information in the FAQ will help you: http://www.soulwax.net/nvidia/faq.shtml Good luck. -- Munish Chopra ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
I was afraid that might be the case :-( From: Munish Chopra [EMAIL PROTECTED] On 2003-06-12 11:51 +, Evan Dower wrote: I am highly interested in getting nvidia-driver to work well for FBSD, and in particular my own computer. I do not have the expertise to hack on either it or the kernel (yet), but I am eager to provide debugging information. nvidia-driver frequently crashes my machine, so I expect that my input could be valuable. Is anybody interested in my assistance? Is anybody actively working on/with/around the nvidia-driver, or does the license make that infeasible? Thanks a bundle, Evan Dower Undergraduate, Computer Science University of Washington NVIDIA is handling everything to do with the driver. Expect a new release shortly *cough*. Until then, perhaps some of the information in the FAQ will help you: http://www.soulwax.net/nvidia/faq.shtml Good luck. -- Munish Chopra ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED] _ MSN 8 helps eliminate e-mail viruses. Get 2 months FREE*. http://join.msn.com/?page=features/virus ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
Evan Dower wrote: I am highly interested in getting nvidia-driver to work well for FBSD, and in particular my own computer. I do not have the expertise to hack on either it or the kernel (yet), but I am eager to provide debugging information. nvidia-driver frequently crashes my machine, so I expect that my input could be valuable. Is anybody interested in my assistance? Is anybody actively working on/with/around the nvidia-driver, or does the license make that infeasible? As was pointed out the other day, nVidia use the %gs register without saving and restoring it, in order to maintain their internal thread state. Now that FreeBSD also uses this register, that causes crashes. I believe the fix is as simple as saving and restoring %gs in the nVidia driver, and/or not letting the nVidia use %gs at all (segment registers generally belong to the OS, in any case, and stealing a CPU register of any kind for exclusive use of a driver is a pig trick). -- Terry ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
Evan Dower wrote: NVIDIA is handling everything to do with the driver. Expect a new release shortly *cough*. I was afraid that might be the case :-( The only thing you care about is entry and exit through driver entry points... and maybe interrupts. It would be easy enough to wrapper all entry points to restore %gs on entry and save it on exit. For the interrupt handler, you would have to mess with the registration mechanism to wrap that. You might be able to get by with poking wrapper routines into the device declaration entry points on module registration with an if nVidia hack. This type of hack would actually be best, from the perspective of the performance of everything else, since it would be a one time setup, instead of something that has to be checked on each entry to see if it's happened yet. -- Terry ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
Hey, I believe the fix is as simple as saving and restoring %gs in the nVidia driver, and/or not letting the nVidia use %gs at all (segment registers generally belong to the OS, in any case, and stealing a CPU register of any kind for exclusive use of a driver is a pig trick). Please take a look at http://www.minion.de/ Christian Zander has written a patch to deal with the gs restoration problem and other issues. It is well tested and adopted. Maxime Henrion has also written several patches which may be found in the ports system (x11/nvidia-driver). I think Terry is jumping to conclusions on what Evan's problem could be. I have this seen this cause application to crash at exit (if using OpenGL) but never a crash in kernel-space from the nVidia drivers. -- +---+ | Samy Al Bahra | [EMAIL PROTECTED] | +---+ Arabeyes.org Kerneled.com FreeBSD.org ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Interest in nvidia-driver debugging on 5-*
On 13 Jun 2003 05:11:34 +, Samy Al Bahra [EMAIL PROTECTED] wrote: Hey, I believe the fix is as simple as saving and restoring %gs in the nVidia driver, and/or not letting the nVidia use %gs at all (segment registers generally belong to the OS, in any case, and stealing a CPU register of any kind for exclusive use of a driver is a pig trick). Please take a look at http://www.minion.de/ Christian Zander has written a patch to deal with the gs restoration problem and other issues. It is well tested and adopted. Maxime Henrion has also written several patches which may be found in the ports system (x11/nvidia-driver). The %gs register issue patch for linux_sysvec.c has been committed in the source tree like a month ago or so, btw. :-) Cheers, Mezz I think Terry is jumping to conclusions on what Evan's problem could be. I have this seen this cause application to crash at exit (if using OpenGL) but never a crash in kernel-space from the nVidia drivers. -- bsdforums.org 's moderator, mezz. ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to [EMAIL PROTECTED]
