Hi, just to let you know, I managed to detect the mode-swithes between privileged and non-privileged mode using one of the simulator API callback functions and it seem to work.
The number of L1 data-cache accesses in non-priviliged mode are exactly the same even though I change the cache-size between experiments, which is what I expect. However, the number of L1 i-cache accesses do differ by a small amount, and I found that the number increases as the execution time increases (due to smaller cache size). I discovered that the number of mode-switches performed in non-privileged mode also increases as the execution time increases, which is logical because the operating system has more opportunities to interrupt the user application. My conclusion is that there are kernel instructions executed in non-priviliged mode and that this "user-mode-OS-overhead" increases as the user-application execution time increases. Does this sound reasonable to you? BTW, apart from mode-switches, are they any other (obvious) code segments in the kernel that can contribute to this overhead? If that is the case I would be very happy to hear about it. Regards, Mladen This message posted from opensolaris.org _______________________________________________ opensolaris-code mailing list opensolaris-code@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
