On Wed, Sep 04, 2019 at 12:47:13PM -0700, Bart Van Assche wrote: > On 9/4/19 11:02 AM, Peter Zijlstra wrote: > > On Wed, Sep 04, 2019 at 10:38:59AM -0700, Bart Van Assche wrote: > > > I think it is widely known that rdtsc is a relatively slow x86 instruction. > > > So I expect that using that instruction will cause a measurable overhead if > > > it is called frequently enough. I'm not aware of any publicly available > > > measurement data however. > > > > https://www.agner.org/optimize/instruction_tables.pdf > > > > RDTSC, Ryzen: ~36 > > RDTSC, Skylake: ~20 > > > > Sadly those same tables don't list the cost of actual exceptions or even > > IRET :/ > > Thanks Peter for having looked up these numbers. These numbers are much > better than last time I checked. Ming, would CONFIG_IRQ_TIME_ACCOUNTING help > your workload? In my fio test on azure L80sv2, IRQ_TIME_ACCOUNTING isn't enabled. However the irq flood detection introduces two RDTSC for each do_IRQ(), not see obvious IOPS difference. Thanks, Ming
