Hi all,

I was doing an experiment with the Berkeley DB benchmark ex_tpc. What I
notice is that increasing the disk latency from 1us to 10us causes the
following: panic in a vanilla kernel:

panic: Inconsistent DMA transfer state: dmaState = 2 devState = 0

In addition, I notice that if I bind network and disk interrupts to a core,
this error occurs in twice as many cases. The tested simulation featured two
m5 default setting O3 cores (I have also seen the same DMA problem with
simpler O3 models and the TimingSimpleCPU), the vanilla 2.6.27 m5 kernel
and a 3 level cache (using the MESI express snoop protocol). I am currently
running gem5-changeset:8339.

I have also found this thread which seems to have a similar problem, but
considers the x86 ISA and mentions that the APIC is the heart of the
problem.

http://www.mail-archive.com/[email protected]/msg03192.html

If anyone happens to know how to fix the problem or efficient documentation
on how Ide works, your thoughts and words would be greatly appreciated.

Thanks,
-Rick
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to