Re: Savage IX/Thinkpad T20: Lockup with DMA
Am Donnerstag, den 26.01.2006, 13:38 -0500 schrieb Alex Deucher: On 1/26/06, Michael Karcher [EMAIL PROTECTED] wrote: Hello DRI developers, on my ThinkPad T20, I experience Lockups with either type of DMA, be it Vertex DMA or AGP texture access. With Option DmaMode None and Option AGPMode 2 glxgears runs perfectly, as does glxinfo, whereas ppracer locks up the system at start (supposedly because of using AGP texturing). If I add Option BusType PCI, everything works as expected. It's a known problem with savage based thinkpads and AGP. I was never able to track it down. you might try messing with the shadowstatus option. I want to say that may be related, but it's been a while since I looked into it. I already use ShadowStatus on, as the machine locked up on around 70% of X server start attempts (with DRI on) if I didn't (The lockup was even before the video memory is cleared). I didn't explicitly mention it, because ShadowStatus on is the default option if DRI is enabled for quite some time now. However, thanks for that hint. Regards, Michael Karcher --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid=103432bid=230486dat=121642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA / Driver broken if bpp!=16
Am Freitag, den 27.01.2006, 11:33 +0100 schrieb Michael Karcher: Am Donnerstag, den 26.01.2006, 22:31 +0100 schrieb Michael Karcher: Am Donnerstag, den 26.01.2006, 14:05 -0700 schrieb Brian Paul: I don't have a Savage card to test anything myself. Looks like the driver's ColorMask code depends on the card model. Which card are you using? As the subject says, it is a Savage IX built into a Thinkpad T20. I probably should add that I am using 16bpp, which makes bitmask errors possible. I now tried other depths, which just showed another bunch of problems. If I switch to 24bpp, I get really funny colors in 2D mode, they change with the xgamma setting. I guess the gamma correction tables are messed up. 3D acceleration produces very strange results (picture width horizontally half of what it should be, distorted geometry and colors) which give the impression that some part of hard- or software is assuming 2 bytes per pixel instead of four. Software fallback (as Felix mentioned) in the red/blue stereo case works OK (except for the funny colors already seen in 2D), and looks the same as with LIBGL_ALWAYS_INDIRECT, but of course it is slow. If the color depth makes a difference then it could be related to the z-buffer depth or maybe color dithering. There is nothing you can do about the z-buffer depth (on Savage4 you could experiment with a floating-point z-buffer). But try disabling color dithering in DRIconf. Regards, Felix After that experience I tried 15bpp. In this case 2D looks quite OK. The gradient in the window title of my sawfish theme looks a bit odd with too much red mixed in some steps, but that might be an application bug. 3D acceleration does not work correctly, as warning messages of libgl already say driver claims to not support visual 0x... The output of 3D acceleration looks like it is meant for 16bpp. Software fallback in the red/blue-stereo case does *not* work correctly: The image width doubles, and at some time the application crashes. It looks like rendering for 32bpp, which might be OK. A final note to the 16bpp case: The performance of xmakemol and the User-to-system CPU usage ratio seem to indicate the software fallback, so the question remains: Why is the output wrong? Michael Karcher --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid=103432bid=230486dat=121642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel -- | Felix Kühling [EMAIL PROTECTED] http://fxk.de.vu | | PGP Fingerprint: 6A3C 9566 5B30 DDED 73C3 B152 151C 5CC1 D888 E595 | --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid3432bid#0486dat1642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA
On 1/27/06, Michael Karcher [EMAIL PROTECTED] wrote: Am Donnerstag, den 26.01.2006, 13:38 -0500 schrieb Alex Deucher: On 1/26/06, Michael Karcher [EMAIL PROTECTED] wrote: Hello DRI developers, on my ThinkPad T20, I experience Lockups with either type of DMA, be it Vertex DMA or AGP texture access. With Option DmaMode None and Option AGPMode 2 glxgears runs perfectly, as does glxinfo, whereas ppracer locks up the system at start (supposedly because of using AGP texturing). If I add Option BusType PCI, everything works as expected. It's a known problem with savage based thinkpads and AGP. I was never able to track it down. you might try messing with the shadowstatus option. I want to say that may be related, but it's been a while since I looked into it. I already use ShadowStatus on, as the machine locked up on around 70% of X server start attempts (with DRI on) if I didn't (The lockup was even before the video memory is cleared). I didn't explicitly mention it, because ShadowStatus on is the default option if DRI is enabled for quite some time now. However, thanks for that hint. I was going to say try forcing it off (I think the windows driver does this on thinkpads), but if you get lockups, I guess that's not an option. Alex Regards, Michael Karcher --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid3432bid#0486dat1642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA / Driver broken if bpp!=16
On 1/27/06, Michael Karcher [EMAIL PROTECTED] wrote: Am Donnerstag, den 26.01.2006, 22:31 +0100 schrieb Michael Karcher: Am Donnerstag, den 26.01.2006, 14:05 -0700 schrieb Brian Paul: I don't have a Savage card to test anything myself. Looks like the driver's ColorMask code depends on the card model. Which card are you using? As the subject says, it is a Savage IX built into a Thinkpad T20. I probably should add that I am using 16bpp, which makes bitmask errors possible. I now tried other depths, which just showed another bunch of problems. If I switch to 24bpp, I get really funny colors in 2D mode, they change with the xgamma setting. I guess the gamma correction tables are messed up. 3D acceleration produces very strange results (picture width horizontally half of what it should be, distorted geometry and colors) which give the impression that some part of hard- or software is assuming 2 bytes per pixel instead of four. Software fallback (as Felix mentioned) in the red/blue stereo case works OK (except for the funny colors already seen in 2D), and looks the same as with LIBGL_ALWAYS_INDIRECT, but of course it is slow. After that experience I tried 15bpp. In this case 2D looks quite OK. The gradient in the window title of my sawfish theme looks a bit odd with too much red mixed in some steps, but that might be an application bug. 3D acceleration does not work correctly, as warning messages of libgl already say driver claims to not support visual 0x... The output of 3D acceleration looks like it is meant for 16bpp. Software fallback in the red/blue-stereo case does *not* work correctly: The image width doubles, and at some time the application crashes. It looks like rendering for 32bpp, which might be OK. Only depth 16 and 24 are accelerated. Alex A final note to the 16bpp case: The performance of xmakemol and the User-to-system CPU usage ratio seem to indicate the software fallback, so the question remains: Why is the output wrong? Michael Karcher --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid3432bid#0486dat1642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Savage IX/Thinkpad T20: Lockup with DMA
Hello DRI developers, on my ThinkPad T20, I experience Lockups with either type of DMA, be it Vertex DMA or AGP texture access. With Option DmaMode None and Option AGPMode 2 glxgears runs perfectly, as does glxinfo, whereas ppracer locks up the system at start (supposedly because of using AGP texturing). If I add Option BusType PCI, everything works as expected. I am using: X Window System Version 6.9.0 (Debian 6.9.0.dfsg.1-2 20060106075605 David Nusinow [EMAIL PROTECTED]) common-20060124-linux.i386.tar.bz2 savage-20060124-linux.i386.tar.bz2 kernel 2.6.15-rc4 (is this a problem?) Thanks for your work, Michael Karcher --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid=103432bid=230486dat=121642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA
Am Donnerstag, den 26.01.2006, 13:46 +0100 schrieb Michael Karcher: Hello DRI developers, on my ThinkPad T20, I experience Lockups with either type of DMA, be it Vertex DMA or AGP texture access. With Option DmaMode None and Option AGPMode 2 glxgears runs perfectly, as does glxinfo, whereas ppracer locks up the system at start (supposedly because of using AGP texturing). If I add Option BusType PCI, everything works as expected. Did you suspend/resume the notebook? There was a problem with AGP after resume that can be worked around by using BusType PCI. I suspected that the bug was actually in the VIA AGP driver, not in the Savage driver. I never got around to looking into it in detail though. Am Donnerstag, den 26.01.2006, 14:46 +0100 schrieb Michael Karcher: Hello developers, The red/blue stereo rendering of xmakemol doesn't work with savage hardware acceleration, but looks fine without (LIBGL_ALWAYS_INDIRECT=1). This is true for the current DRI snapshot as well as for an older snapshot from about August of last year. Do you need further information for debugging this problem? Screenshots are on http://www.physik.fu-berlin.de/~karcher/good.png (software rendered) and http://www.physik.fu-berlin.de/~karcher/bad.png (hardware rendered). No one has been working on the Savage 3D driver in a while. Your odds of getting someone to look into this problem are fairly low ATM. Sorry. Feel free to open a bug in the freedesktop.org bugzilla. Regards, Felix -- | Felix Kühling [EMAIL PROTECTED] http://fxk.de.vu | | PGP Fingerprint: 6A3C 9566 5B30 DDED 73C3 B152 151C 5CC1 D888 E595 | --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid3432bid#0486dat1642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA
On 1/26/06, Michael Karcher [EMAIL PROTECTED] wrote: Hello DRI developers, on my ThinkPad T20, I experience Lockups with either type of DMA, be it Vertex DMA or AGP texture access. With Option DmaMode None and Option AGPMode 2 glxgears runs perfectly, as does glxinfo, whereas ppracer locks up the system at start (supposedly because of using AGP texturing). If I add Option BusType PCI, everything works as expected. I am using: X Window System Version 6.9.0 (Debian 6.9.0.dfsg.1-2 20060106075605 David Nusinow [EMAIL PROTECTED]) common-20060124-linux.i386.tar.bz2 savage-20060124-linux.i386.tar.bz2 kernel 2.6.15-rc4 (is this a problem?) Thanks for your work, Michael Karcher It's a known problem with savage based thinkpads and AGP. I was never able to track it down. you might try messing with the shadowstatus option. I want to say that may be related, but it's been a while since I looked into it. Alex --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid3432bid#0486dat1642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA
Am Donnerstag, den 26.01.2006, 12:06 -0500 schrieb Felix Kühling: Am Donnerstag, den 26.01.2006, 13:46 +0100 schrieb Michael Karcher: Hello DRI developers, on my ThinkPad T20, I experience Lockups with either type of DMA, be it Vertex DMA or AGP texture access. With Option DmaMode None and Option AGPMode 2 glxgears runs perfectly, as does glxinfo, whereas ppracer locks up the system at start (supposedly because of using AGP texturing). If I add Option BusType PCI, everything works as expected. Did you suspend/resume the notebook? There was a problem with AGP after resume that can be worked around by using BusType PCI. I suspected that the bug was actually in the VIA AGP driver, not in the Savage driver. I never got around to looking into it in detail though. Am Donnerstag, den 26.01.2006, 14:46 +0100 schrieb Michael Karcher: Hello developers, The red/blue stereo rendering of xmakemol doesn't work with savage hardware acceleration, but looks fine without (LIBGL_ALWAYS_INDIRECT=1). This is true for the current DRI snapshot as well as for an older snapshot from about August of last year. Do you need further information for debugging this problem? Screenshots are on http://www.physik.fu-berlin.de/~karcher/good.png (software rendered) and http://www.physik.fu-berlin.de/~karcher/bad.png (hardware rendered). No one has been working on the Savage 3D driver in a while. Your odds of getting someone to look into this problem are fairly low ATM. Sorry. Feel free to open a bug in the freedesktop.org bugzilla. Correcting myself. There have been updates to the Savage driver after I stopped working on it by Ian Romanick, Brian Paul and Eric Anholt and maybe others. Sorry folks, I appreciate your work! Is anyone going to look into this xmakemol issue? Regards, Felix Regards, Felix -- | Felix Kühling [EMAIL PROTECTED] http://fxk.de.vu | | PGP Fingerprint: 6A3C 9566 5B30 DDED 73C3 B152 151C 5CC1 D888 E595 | --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid3432bid#0486dat1642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA
Felix Kühling wrote: Am Donnerstag, den 26.01.2006, 12:06 -0500 schrieb Felix Kühling: Am Donnerstag, den 26.01.2006, 13:46 +0100 schrieb Michael Karcher: Hello DRI developers, on my ThinkPad T20, I experience Lockups with either type of DMA, be it Vertex DMA or AGP texture access. With Option DmaMode None and Option AGPMode 2 glxgears runs perfectly, as does glxinfo, whereas ppracer locks up the system at start (supposedly because of using AGP texturing). If I add Option BusType PCI, everything works as expected. Did you suspend/resume the notebook? There was a problem with AGP after resume that can be worked around by using BusType PCI. I suspected that the bug was actually in the VIA AGP driver, not in the Savage driver. I never got around to looking into it in detail though. Am Donnerstag, den 26.01.2006, 14:46 +0100 schrieb Michael Karcher: Hello developers, The red/blue stereo rendering of xmakemol doesn't work with savage hardware acceleration, but looks fine without (LIBGL_ALWAYS_INDIRECT=1). This is true for the current DRI snapshot as well as for an older snapshot from about August of last year. Do you need further information for debugging this problem? Screenshots are on http://www.physik.fu-berlin.de/~karcher/good.png (software rendered) and http://www.physik.fu-berlin.de/~karcher/bad.png (hardware rendered). No one has been working on the Savage 3D driver in a while. Your odds of getting someone to look into this problem are fairly low ATM. Sorry. Feel free to open a bug in the freedesktop.org bugzilla. Correcting myself. There have been updates to the Savage driver after I stopped working on it by Ian Romanick, Brian Paul and Eric Anholt and maybe others. Sorry folks, I appreciate your work! Is anyone going to look into this xmakemol issue? I took a quick look at xmakemol. The first thing I'd check is if the driver code for glColorMask is operating properly. For red/blue stereo, the scene is rendered twice, first with glColorMask(T,F,F,T), then glColorMask(F,F,T,T). I don't have a Savage card to test anything myself. Looks like the driver's ColorMask code depends on the card model. Which card are you using? BTW, the author of xmakemol needs to be taught how to use function prototypes. Put them in a header file; don't repeat the prototype inside each and every function it's used in! -Brian --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid3432bid#0486dat1642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA
Am Donnerstag, den 26.01.2006, 12:06 -0500 schrieb Felix Kühling: on my ThinkPad T20, I experience Lockups with either type of DMA, be it Vertex DMA or AGP texture access. With Option DmaMode None and Option AGPMode 2 glxgears runs perfectly, as does glxinfo, whereas ppracer locks up the system at start (supposedly because of using AGP texturing). If I add Option BusType PCI, everything works as expected. Did you suspend/resume the notebook? There was a problem with AGP after resume that can be worked around by using BusType PCI. I suspected that the bug was actually in the VIA AGP driver, not in the Savage driver. I never got around to looking into it in detail though. I know about these problems. The driver from mid of last year had these problems: With AGPMode 1, I had serious problems, with AGPMode 2, everything worked fine until the first suspend, with BusType PCI, everything worked fine. The current snapshot crashes on BusType PCI without DmaMode None. It surely is not a VIA AGP problem, my chipset is a i440BX. But the agpgart backend drivers are all quite like to each other, so i would expect common bugs. Michael Karcher --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid3432bid#0486dat1642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA
Am Donnerstag, den 26.01.2006, 14:05 -0700 schrieb Brian Paul: I don't have a Savage card to test anything myself. Looks like the driver's ColorMask code depends on the card model. Which card are you using? As the subject says, it is a Savage IX built into a Thinkpad T20. I probably should add that I am using 16bpp, which makes bitmask errors possible. Michael Karcher PS: Brian, sorry for the private copy. I didn't pay enough attention. --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid=103432bid=230486dat=121642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
Re: Savage IX/Thinkpad T20: Lockup with DMA
Am Donnerstag, den 26.01.2006, 14:05 -0700 schrieb Brian Paul: Felix Kühling wrote: Am Donnerstag, den 26.01.2006, 12:06 -0500 schrieb Felix Kühling: Am Donnerstag, den 26.01.2006, 13:46 +0100 schrieb Michael Karcher: Hello DRI developers, on my ThinkPad T20, I experience Lockups with either type of DMA, be it Vertex DMA or AGP texture access. With Option DmaMode None and Option AGPMode 2 glxgears runs perfectly, as does glxinfo, whereas ppracer locks up the system at start (supposedly because of using AGP texturing). If I add Option BusType PCI, everything works as expected. Did you suspend/resume the notebook? There was a problem with AGP after resume that can be worked around by using BusType PCI. I suspected that the bug was actually in the VIA AGP driver, not in the Savage driver. I never got around to looking into it in detail though. Am Donnerstag, den 26.01.2006, 14:46 +0100 schrieb Michael Karcher: Hello developers, The red/blue stereo rendering of xmakemol doesn't work with savage hardware acceleration, but looks fine without (LIBGL_ALWAYS_INDIRECT=1). This is true for the current DRI snapshot as well as for an older snapshot from about August of last year. Do you need further information for debugging this problem? Screenshots are on http://www.physik.fu-berlin.de/~karcher/good.png (software rendered) and http://www.physik.fu-berlin.de/~karcher/bad.png (hardware rendered). No one has been working on the Savage 3D driver in a while. Your odds of getting someone to look into this problem are fairly low ATM. Sorry. Feel free to open a bug in the freedesktop.org bugzilla. Correcting myself. There have been updates to the Savage driver after I stopped working on it by Ian Romanick, Brian Paul and Eric Anholt and maybe others. Sorry folks, I appreciate your work! Is anyone going to look into this xmakemol issue? I took a quick look at xmakemol. The first thing I'd check is if the driver code for glColorMask is operating properly. For red/blue stereo, the scene is rendered twice, first with glColorMask(T,F,F,T), then glColorMask(F,F,T,T). I took a brief look to refresh my memory. The difference between Savage3D-based and Savage4-based cards is that Savage4 can disable all drawing in the case glColorMask(F,F,F,F). Maybe useful if you want to render only to the Z-buffer. But I'm not even sure that would work correctly. In all other cases that are not glColorMask(T,T,T,T) the driver will fallback to software. The code looks correct to me but I don't think I ever tested it thoroughly. I don't have a Savage card to test anything myself. Me neither. Looks like the driver's ColorMask code depends on the card model. Which card are you using? Micheal is using a Savage3D-based card. Regards, Felix BTW, the author of xmakemol needs to be taught how to use function prototypes. Put them in a header file; don't repeat the prototype inside each and every function it's used in! -Brian -- | Felix Kühling [EMAIL PROTECTED] http://fxk.de.vu | | PGP Fingerprint: 6A3C 9566 5B30 DDED 73C3 B152 151C 5CC1 D888 E595 | --- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnkkid3432bid#0486dat1642 -- ___ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel