PROBLEM: Hangup with Radeon9200 accel. Does not happen when not using agpgart or disabling CPU internal cache.
-Foreword: I am not complaining; just hoping to help in developing a better kernel.
-Symptom: When accelerating graphics, the machine freezes completely (no response to emergency sys req keys or net access).
As a consequence, NOTHING GETS LOGGED (sorry).
-How to reproduce:
1) Use the following hardware:
Radeon 9200 AGP graphics card.
CPU PentiumII Deschutes (Tried the same card on an
Athlon machine -all the rest being equal- and the problem did not happen). baseplate Tyan (This means Intel 440BX chipset)
2) Enable in BIOS the CPU internal cache to
"Write Back" mode.
(Either "Write Thru" or "Disabled" will eliminate the problem, at the expense of an extremely low performance).
3) Use ANY Linux-2.6.0-test kernel, or ANY
Linux-2.4 kernel, (in which case you will need
an updated version for the radeon module).
4) Use XFree86 4.3.0 or over (for radeon acceleration support) with ANY dri version (older or newest).
I have tried the latest DRI CVS without success.
5) In the XF86Config-4 file, specify the chipset of a Radeon 9100 or another radeon for which acceleration is already supported (I know this is not the factor to blame, since -as I said before- it works on an Athlon machine).
6) In the XF86Config-4 file, make sure the option ForcePCIMode is off (when it is enabled, everything works fine and you need no agpgart module, but performance is halved).
_ALL_ the other options are indifferent in order to reproduce the problem, (even the option to use the XFree GPL radeon driver or the propietary FireGL drivers from Ati).
7) Make sure the agpgart module (and the intel_agp module for 2.6 kernels) are loaded before launching XFree.
8) Launch X. Everything seems to work; glxinfo says Yes as to direct rendering. Now launch an OpenGL application, like glxgears:
It runs for some instants, then becomes blocked, as does the rest of the system.
Observation: When a big Graphics Aperture Size (256MB) is specified in BIOS (thus adopted by the agpgart module) and a small one (4MB) in the XF86Config-4 file, then the time until crash is considerably increased. (Does this mean anything to you?)
Please reply with Cc to [email protected] because I am not suscribed to any kernel mailing list.
Thank you.
Quoted 7 lines · Juanjo García Carré
7) Make sure the agpgart module (and the
intel_agpmodule for 2.6 kernels) are loaded before launching XFree.8) Launch X. Everything seems to work; glxinfo says Yes as to direct rendering. Now launch an OpenGL application, like glxgears:
It runs for some instants, then becomes blocked, as does the rest of the system.
I added this to my /etc/X11/XF86Config-4 to prevent my system from locking solid as you describe. This forces XFree86 to drive my Radeon 9200 as if it were a Radeon 9000. Before upgrading X, I had to use 0x4242 to drive my card as a Radeon 8500.
Section "Device"
Identifier "ATI Radeon"
Driver "radeon"
Option "AGPMode" "4"
Option "EnablePageflip" "1"
ChipID 0x4964
#ChipID 0x4242EndSection
-Erik
--
Erik B. Andersen http://codepoet-consulting.com/
--This message was written using 73% post-consumer electrons--
Sorry to tell you that the proposed work-around
ChipID 0x4964
did not work.
Thanks a lot for your time and advice.
A message that is not in the archive (sent off-list or not yet received): [email protected]
A message that is not in the archive (sent off-list or not yet received): [email protected]
I finally managed to find the cause of the problem:
- It was not software.
Sorry for the unnecessary annoyance, people from the kernel and XFree lists! I will try to compensate by telling you what happened, just in case you would face anything similar.
- It was neither the hardware.
- It was BIOS, but...
- It was not a problem with BIOS settings.
- It was not a problem of BIOS version.
- The BIOS in fact can work perfectly.
Intrigued ?
Be prepared for one of the strangest stories of computer malfunction:
I bought a recent graphics card (Radeon 9200) for a not so recent computer (Pentium2 with Intel440BX chipset).
I found Linux freezed completely whenever I tried to use acceleration, which did not happen when using Windows with the same machine, nor the same card on another -dissimilar- machine.
So I begun to search for flaws in the IntelAGP kernel module and/or the Direct Rendering Interface, something that proved unsuccesful. (I deeply thank everyone who tried to help me with this).
Frustrated, I decided to flash the BIOS with the latest version from the manufacturer and... It worked !!!
But... Wait!... Isn't this the same BIOS version that was already installed?
Didn't I tried to load default and fail-safe settings with no positive result?
Mmmmm.... Let's see... Let's try changing some parameters.... It stops working again !!!
Let's return those parameters to their previous values... NO SUCCESS.
This is the bizarre part:
IT COULD NOT BE MADE TO WORK AGAIN, REGARDLESS OF THE PARAMETERS SETTINGS, unless CMOS was cleared using the motherboard jumper.
I checked the fenomenon a dozen times. Moreover, after enough fiddling with the bios parameter settings, even Windows begins to freeze (always regardless of the concrete values introduced)
until CMOS is hard-reset again.
Possible explanations:
1- The system does not truly restart after changing values. This possibility was easy to discard by powering off, waiting 2 minutes and then powering on.
2- Values saved and shown later by the BIOS do not correspond in fact to real ones.
3- The BIOS gets somehow corrupted when changing its parameters. 4- The BIOS is slowly getting corrupted by aging:
We all know today's hardware decomposes with time. I have seen circuit boards that became pastous, sticky, degraded in only a couple of years. (Learn the lesson. Don't throw away your old hardware!) If this is so, in the future I will see how not even the jumper erasing method will work.
For those interested, the "magic BIOS" corresponds to:
BasePlate: Tyan Mod. S1846 Tsunami ATX 2.00.04 031320011030 BIOS: Ami 686 AMIBIOS (c)1999 AE12 4885 BIOS chip: WinBond W29C020-90