On 2/19/24 12:36, Rodrigo Queiro wrote: > What do you think about removing the checks on the model number, and > assuming that all processors with family == 6 support the Nehalem > MSRs, including the SMI counter? Any user running a recent cyclictest > on pre-Nehalem hardware could still run without --smi to prevent > seeing bogus SMI counts. When I added that option the smi was a relatively new thing... nowadays I think it is pretty standard... so simplifying it is a good thing. Nevertheless, the SMI counter is just part of the story. Most of the hardware related latency problems goes outside of SMI "domain" so... the best thing to do is to run rtla hwnoise (or hwlat) on the hardware to detect hardware latencies. -- Daniel
