On 07.05.2004 16:56, Udo Wolter wrote:
There might be a race condition somewhere wrt SMP -- as Johannes already pointed out, he'll have a look at it.
Are there any kernel changes which could make my situation better ? The preemptive kernel parameter or the MSI vector based IRQs ? Or even vice versa, is it better to switch something off ? Or any other kernel parameter ?
Not really. Call it a race condition or something else -- it's a logical programming error that needs to be fixed when it comes to SMP or hyperthreading.
The two-staged notification and transfer mechanism of the av7110 makes it a little difficult to handle, because both CPUs can execute the tasklets scheduled by the irq handlers independently.
I'd stand a lot if it just costs only 5% performance but I want to stay in hyperthreading. Switching off hyperthreading costs 30-40% performance. :(
In the meantime, you need to turn off hyperthreading and kernel preemption unless Johannes, me or someone else has fixed this problem thorougly.
Mermgfurt,
Udo
CU Michael.
-- Info: To unsubscribe send a mail to [EMAIL PROTECTED] with "unsubscribe linux-dvb" as subject.
