Distribution: Debian/GNU Unstable Hardware Environment: Amilo A CY26 Software Environment: XFree86 4.4.0 Problem Description: There are probably some IRQ conflicts in Amilo A -laptop. If I have modules snd_ali5451, ohci_hcd, eth0 and yenta loaded (they all use IRQ 11) while trying to start X with DRI support, the computer will crash (black screen, keyboard/network doesn't respond). If I remove the modules, X starts fine and DRI works ok. When the modules are unloaded and X is running with DRI, i tried to load ethernet up and it caused the laptop to crash. The same problem was with 2.4 kernels, but people fixed it by editing "__HAVE_SHARED_IRQ 1" to "__HAVE_SHARED_IRQ 0" in drivers/char/drm/radeon.h. I tried this same fix with 2.6, but all i get are crashing 3D-applications and kernel oopses. X did start DRI-support enabled though. I have tried booting with no acpi support, but the results were the same. This is my first bug-report, so i'm not sure if i have given all the necessary information. I'll post few command outputs & etc. in a separate attachment. (I used "__HAVE_SHARED_IRQ 0" in the logs.) Steps to reproduce: Buy a Amilo A -laptop. Install 2.6.6-mm1 kernel, XFree86 4. 4.0, dri.sourceforge.net patches and try to start X+DRI when ethernet/sound/usb/pcmcia modules are loaded. (Follow the instructions at http: //h00045ad2a5c0.ne.client2.attbi.com/jaydolan/igp.html). Other Amilo A users have reported the same problem.
Created attachment 2837 [details] Few command outputs & logs /proc/interrupts, lspci -vvv, lsmod, dmesg and XFree86 log
Created attachment 2838 [details] Linux 2.6.6-mm1 config
I am also afected by this bug : (Linux nrv 2.6.9-amiloa #8 Tue Nov 9 12:53:25 CET 2004 i686 GNU/Linux) log reported at http://rzr.online.fr/q/drm FYI, This bug is connected to those ones : https://freedesktop.org/show_bug.cgi?id=1678 How to use DRI on AmiloA with 2.6 kernels : http://rzr.online.fr/q/dri base knowlege on amiloA http://rzr.online.fr/q/IRQ By the way what bios are you using ? i am on 1.02
I confirm everything. I have a 2.6.8 kernel with Xorg 6.8.1 Remark : while the radeon module wants to use IRQ #11 (which is very populated already : 5 modules, ohci_hcd, eth0, yenta, yenta, snd_ali5451), Windows XP uses irq # 10 for radeon IGP 320M. Is there a way to have radeon trying another irq ? I've tried lots of irq-related kernel boot options. The only one that had any effect on the result of "cat /proc/interrupts" was "acpi_irq_balance". This way, snd_ali5451 moved to irq #10... But conflicts remains between radeon and the 4 other modules at irq #11.
I've added acpi-bugzilla to the CC list, because I think the bug was wrongly assigned and should show: Category : ACPI Component : config-Interrupts (it is IRQ related) Kernel Version : 2.6.9 I may be wrong, but if so, please assign the bug to the good person.
Created attachment 4253 [details] Comparison of IRQs between Windows XP and Linux It's a table of comparison of IRQ assignment is : - Windows XP - Linux 2.6.8 - Linux 2.6.8 pci=noacpi - Linux 2.6.8 acpi_irq_balance
Is this still a problem in 2.6.13? (it was possibly fixed by the implicit return workaround) If so, please attach the output from acpidump, available in the latest pmtools here: http://ftp.kernel.org/pub/linux/kernel/people/lenb/acpi/utils along with lspci -vv, dmesg -s64000 and /proc/interrupts
I did not try 2.6.13 yet, but this problem still exists in 2.6.12-cko3 (this patchset includes ACPI-version 20050708). But the reason of this is clear now: This crash occurs when the PCMCIA0-device (the upper slot) and the RADEON-device (which always shares the IRQ with Firewire) do not! have the same IRQ. I know this is somewhat weird but all tests verify this, if PCMCIA0 and Radeon have the same IRQ (no matter which one of the IRQs that are for PCI-Devices=3,5,9,10,11) 3D works perfectly without crashing. It is also independent from the fact whether there is a card in the PCMCIA-slot or not. The latest bios version for the Amilo A 7600 (1.10) from summer 2004 does not fix this and it is very unlikely that there will be a 1.11 from FSC. Some of the Amilo A 7600-owners (me too) boot with a fixed DSDT to solve this (and a problem with two of the three USB-slots not working with newer kernels, but this is another topic). This fixed DSDT (tested by some people, also with PCMCIA-devices) is available on http://acpi.sourceforge.net/dsdt/tables/Fujitsu-Siemens/Amilo_A_7600/Fujitsu-Siemens-Amilo_A_7600-1.0.9-custom.asl.gz The changed lines concerning this problem are line 378 (section LNKA=PCMCIA0) and 466 (section LNKC=Radeon and Firewire). They get the same IRQ number, IRQ 3 can be used as the Amilo A 7600 has only one COM-Port. In the original, unfixed DSDT LNKA has IRQ 11 and LNKC has IRQ 10 (this is the same as in Windows XP, it works there, but with Linux not). A detailed descripton is also there (also the one of the USB-problem): http://groups.yahoo.com/group/amilo/message/539 (OK the pronounced HowTo on the end is still not ready I had much to do with my diploma thesis). Maybe if the kernel gets fixed this will not be necessary anymore. It would be desirable if this problem would be fixed in future kernels as X.org 6.8.1 and better support 3D with the Radeon IGP 320M and some distributions (Fedora3...) enable 3D by default. Then people complain "cant use distribution XYZ because my computer freezes when starting X". This can discourage some beginners as loading a fixed DSDT is not so easy. If you wish to solve this problem in the vanilla-kernel and I can help you with logs etc. just ask. But unfortunately I am not a programmer. Sorry if my English is not so good. Thomas
Created attachment 5682 [details] output of ./acpidump > acpidump.out, with fixed DSDT
Created attachment 5684 [details] output of dmesg -s64000 > dmesg.txt, with fixed DSDT Cant get to loginscreen without loading custom DSDT because of segmentation faults and many services do not start (SuSE 9.3 with 2.6.12-cko3)
Created attachment 5685 [details] output of cat /proc/interrupts > interrupts.txt, with fixed DSDT
Created attachment 5686 [details] output of lspci -vv > lspci.txt, with fixed DSDT
Created attachment 5687 [details] original DSDT, without fixing This is the output of "cat /proc/acpi/dsdt > dsdt.dat followed by ./iasl -d dsdt.dat" As I cant get to a login screen with Kernel 2.6.12-cko3 without fixed DSDT this the output from December 2004 with SuSE 9.0 and Kernel 2.4.28-lck1. But there should be no differences to 2.6.x - kernels (2.6.11-cko5 booted fine in runlevel 3 without custom DSDT).
Created attachment 5748 [details] original DSDT, without fixing
Created attachment 5749 [details] fixed DSDT
Created attachment 5750 [details] output of lspci without a fixed DSDT
Created attachment 5751 [details] Interrupts with fixed DSDT
Created attachment 5752 [details] Interrupts without fixed DSDT
Created attachment 5753 [details] dmesg with DSDT
Created attachment 5754 [details] dmesg without fixed DSDT
I have now kernel 2.6.13-rc6-git13 installed. I can boot my SuSE 9.3 with this kernel-version also without the custom DSDT (in contrast to 2.6.12). So I can confirm that the problem that the computer freezes if X is started with Direct Rendering enabled also exists with 2.6.13-rc6-git13 and provide current logfiles. (it seems that for Vesa and 2D-Mode no IRQ for the graphics card is used). Thomas
I am also affected by this bug DRI works on : 2.6.12.4 not 2.6.13 I will probally drop additional notes at http://rzr.online.fr/q/Raptor
is this still a problem with linux-2.6.22.stable?
please re-open if this is still an issue with linux-2.6.22.stable or later
I have this problem on a newly installed and updted ubuntu 7.10, with kernel 2.6.22-14-generic
Yes, this is still an issue with latest kernel from cooker. Please reopen (I can't).