For your second part (the thread suspend/activate) I agree that at first glance it looks like a bug. Korey has been the most recent person to work with O3 SMT thread handling, and I see his name attached to some of the lines in question, so I'll let him respond...
Steve On Wed, Jul 21, 2010 at 2:46 PM, Ioannis Ilkos <ioannis.ilko...@imperial.ac.uk> wrote: > Hello, > > I have been playing with time in m5 syscall emulation mode and had a couple > of issues: > > First of all I believe there is a bug in the times() syscall implementation > (sim/syscall_emul.hh). The current clocks value passed to the userland is: > int64_t clocks = curTick * OS::M5_SC_CLK_TCK / SimClock::Int::s; > > This evaluates to a wrong expression. Now, according to the man page all the > tms buffer values are supposed to be in clock ticks. Thus a more appropriate > expression for it would be: > int64_t clocks = tc->getCpuPtr()->tickToCycles(curTick); > > Secondly, I have implemented the nanosleep() system call. I believe that the > appropriate way to do this would be suspending (tc->suspend()) the > ThreadContext and re-activating (tc->activate(delay)) after the sleep > duration (please correct me if I am mistaken). This however led to a peculiar > behaviour with O3 in SMT mode: > 1. in FullO3CPU<Impl>::suspendContext (cpu.cc) the CPU _status is set to > Idle, regardless whether there are other running contexts in the CPU. Thus a > thread sleeping leads to all the other SMT threads sleeping. > > 2. in FullO3CPU<Impl>::activateContext (cpu.cc) we schedule a tick event > after the delay parameter passed, effectively overwriting CPU's tickEvent and > moving all threads of the processor tick(delay) ticks forward. Is this the > intended behaviour or is it a bug? Clearly in my case it is not helpful since > the point of getting a thread to sleep relative to the other threads' time. > > Thank you, > Ioannis > > _______________________________________________ > m5-dev mailing list > m5-dev@m5sim.org > http://m5sim.org/mailman/listinfo/m5-dev > _______________________________________________ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
