Am Donnerstag, 10. Oktober 2002 18:25 schrieb Fabrice Bellet: > On Thu, Oct 10, 2002 at 04:55:34PM +0200, Dieter Nützel wrote: > > Am Donnerstag, 10. Oktober 2002 01:42 schrieb Fabrice Bellet: > > > I updated my dri HEAD cvs tree two hours ago, and I rebuilt > > > the drm module too. > > > > Which kernel version? > > 2.4.20-pre9. > > > > Hmm, all the interrupts are routed to the same CPU ? the SCSI disk > > > is on the first scsi0 controller (irq 5 I assume). > > > > Are you running a "current" kernel with IRQ balancing? > > Have you tried with APIC enabled? > > ACPI is useful, too (IRQ routing). > > Exact. I reenabled ACPI in bios, added support in kernel, and > removed the noapic option at boot time, and IRQ are now correctly > balanced on both CPUs.
Good to hear ;-) > But the same behaviour still occurs : > . X server freeze, taking 100% CPU time, > . boths client applications are blocked in r200GetLock() > . X is doing ioctl()s somewhere in miSpriteComposite() > > In some cases, I can recover from this situation, by killing > the clients and the X process. Removing the drm driver generates > error messages : > > [drm:radeon_ioremapfree:mappings] *ERROR* Attempt to free NULL pointer > [drm:radeon_ioremapfree:mappings] *ERROR* Excess frees: 5 frees, 4 allocs > [drm] Module unloaded > > I other cases, the machine definitly hangs when killing X, or > hands when restarting X. Yes exactly, one of the worst observation I ever made. I would have posted about it much earlier... Anyone _EVER_ tried to stop X, "rmmod DRM module", and restart X, again? X server looks up solid with an old framebuffer content except of some few upper lines. => REBOOT needed. So there must be a module (re)initialization bug. > > > > Does this work with R200_NO_IRQS? > > > > > > Setting R200_NO_IRQS=1 didn't change the behaviour. The machine crashed > > > after both clients existed with r200WaitForFrameCompletion: > > > drmRadeonIrqWait: -16 > > > > Like with multiple context here. > > > > I'll try with R200_NO_IRQS and/or R200_DEBUG, now. > > After doing some tests again, I noticed that R200_NO_IRQS doesn't > basically resolve the locking problem. But, when enabled, > I no longer have the drmRadeonIrqWait: -16 in the client's logs. > The clients apps no longer exit on this error, but stay blocked > somewhere in r200GetLock(). Thanks, as like as here. SMP only? -Dieter ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
