Michael Hohmuth <hohmuth@innocent.com> writes:
can't meet real-time requirements; the task_delete operation seems to be especially bad, sometimes disabling interrupts for more than one second.
Yes, the measurements showed interrupt a latency of sometimes 1.5 seconds (even if the interrupt thread had the highest priority) when in the L4-Linux System had to serve a lot of calls to fork/exec. This was tested with the lat_proc benchmark from the lmbench-suite that tries to measure the latency for creating simple processes (that only call exit) Still there is only one explanation for this: disabled interrupts while flushing the complete address space of a task when doing task_delete. ciao, Torsten. -- Torsten Paul Es ist leichter, einen Atomkern zu paul@os.inf.tu-dresden.de // spalten als ein Vorurteil. __________________________ooO_(+ +)_Ooo________Albert Einstein_(1879-1955) U