Re: Whats the purpose of get_cycles_sync()

On Tue, Oct 30, 2007 at 09:21:02PM +0100, Andi Kleen wrote:
"Joerg Roedel" <joerg.roedel@xxxxxxx> writes:

I would like to answer what the special purpose of the get_cycles_sync()
function is in the x86 architecture. In special I ask myself why
this function has to be *sync*?

Vojtech had one test that tested time monotonicity over CPUs
and it constantly failed until we added the CPUID on K8 C stepping.
He can give details on the test.

I suspect the reason was because the CPU reordered the RDTSCs so that
a later RDTSC could return a value before an earlier one. This can
happen because gettimeofday() is so fast that a tight loop calling it can
fit more than one iteration into the CPU's reordering window.

The K8's still guarantee that subsequent RDTSCs return increasing
values, even if the processor reorders them.

What could have been happening then was that the RDTSC instruction might
have been reordered by the CPU out of the seqlock, causing trouble in
the calculation.

Anyway, adding the CPUID didn't solve all the problems we've seen back
then, and so far none of the approaches for using TSC without acquiring
a spinlock on multi-socket AMD boxes worked 100% correctly.

That is why newer kernels use RDTSCP if available which doesn't need
to be intercepted and is synchronous. And since all AMD SVM systems
have RDTSCP they are fine.

On Intel Core2 without RDTSCP the CPUID can be still intercepted right
now, but the real fix there is to readd FEATURE_SYNC_TSC for Core2 --
the RDTSC there is always monotonic per CPU and the patch that changed
that (f3d73707a1e84f0687a05144b70b660441e999c7) was bogus and must be
reverted. I didn't catch that in time unfortunately.

--
Vojtech Pavlik
Director SuSE Labs
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Relevant Pages

  • Re: what exactly does CPU_TIME measure?
    ... clock cycles then rdtsc makes more sense. ... speed CPU) but clock cycles are. ... so you can get accurate timing even for small ...
    (comp.lang.fortran)
  • Re: Is timeGetTime monotonic and uniform?
    ... that returns the CPU ticks since boot time in EDX:EAX. ... Some time ago I read that when you get the RDTSC overhead (ie the time ... of checking whether the above recommendation is really necessary, ... forcing all other instructions out of the internal CPU ...
    (microsoft.public.win32.programmer.mmedia)
  • Re: assembly in future C standard
    ... should the compiler author implement method 1 ... Some compilers will give an option to which set of cpu instruction ... complience with standard XX run on a cpu predated XX. ... Again ill use an example of RDTSC: ...
    (comp.lang.c)
  • Re: rdtsc and the sleep fcn (Was problem with rdtsc instruction on Pentium M)
    ... > function goes into the idle loop, causing the tsc to stop counting. ... the CPU is slowed down which stuffs ... The solution is not to use RDTSC. ...
    (comp.lang.asm.x86)
  • Re: Whats the purpose of get_cycles_sync()
    ... Vojtech had one test that tested time monotonicity over CPUs ... RDTSC instruction out-of-order, thats clear. ... that this function executes CPUID if RDTSC itself ... by hypervisors (KVM intercepts it for example). ...
    (Linux-Kernel)