HostDrivers/Support: add and use supdrvOSAreCpusOfflinedOnSuspend(). FreeBSD, Darwin, Solaris need verification.

paranoia.

HostDrivers/Support: Host suspend/resume now recomputes TSC-deltas for all CPUs.

HostDrivers/Support: doxygen nit.

SUPDrvGip.cpp: Increased the TSC frequency refinement period to 12.2 seconds. Decided to remove the wait-for-RTTimeSystemNanoTS-to-change-loop in the TSC frequency refinement callback as it will only get us into trouble where it matters and is pointless on the result of the systems.

SUPDrvGip.cpp: Re-enabled the clock tick sync code in supdrvInitRefineInvariantTscFreqTimer again, but with some precautions preventing us from spinning forever.

SUPDrvGip.cpp: Re-enabled the TSC frequency refinement timer for windows now that RTTimerStop no longer does stupid stuff like trying to flush the DPC queue.

SUPDrvGip.cpp: Forgot something during the Init/Delete removal, would cause NULL pointer crashes in supdrvTscDeltaMethod2ProcessDataOnMaster.

SUPDrvGip.cpp: Temporariliy disabled TSC frequency refinement on windows.

Temporarily disable the time tick loop in supdrvInitRefineInvariantTscFreqTimer.

SUPDrvGip.cpp: Simplify the data structures for the TSC measurements, drop the per-method init/delete stuff.

SUPDrvGip.cpp: Double the timeout to just below 2ms to make it less likely that we time out before we get anything done on slow machines. Windows doesn't seem to care about how long IPI work takes, fortunately. :-) (Usually the job is done in < 50 us, this is just making sure the worst case scenario works.)

SUPDrvGip.cpp: Cleanups.

SUPDrvGip.cpp: Enabled the 2nd algo after some adjusting.

SUPDrvGip.cpp: Sync fixes.

nits

SUPDrvGip.cpp: Sync macro fix. Profiling in progress...

iprt/cdefs.h,VBoxDrvGip.cpp: Using __noinline__ instead of noinline in DECL_NO_INLINE should hopefully fix the issue with linux redefining noinline in a more permanent manner.

Disable the statistics collected in supdrvTscDeltaVerify.

SUPDrv: Register a power manage notification callback so we can cancel the invariant TSC frequency refinement timer.

SUPDrvGip.cpp: Use a interval timer as solaris cannot restart one-shot timers from the callback. This is a better idea anyway as I get ticks I can use to record the refinement history in the u64CpuHz members of CPU entries 1+ :-)

SUPDrvGip.cpp: Use RTMpOnPair.

Eliminated supdrvTscDeltaApply. corrected explanation in supdrvGipSyncAndInvariantTimer.

IPRT, HostDriver, VMMR0: MP notifications fixes for TSC-delta measurements, scheduling of notification callback taken care by the API consumer instead of IPRT.

XXX

SUPDrvGip.cpp: Start the TSC measurement by testing for zero delta. We do alternate TSC reads on the two CPUs and check that the TSC value is ever increasing.

SUPDrvGip.cpp: Fixed warnings and build break.

SUPDrvGip.cpp: todo wrt host suspend during invariant tsc frequency refinement.

done.

SUPDrvGip.cpp: Replaced idTscDeltaInitiator+RTThreadSleep(1) with a mutex semaphore. Allocate the SUPDRVGIPTSCDELTARGS structure (stack might be swappable, also don't waste it). Remove old sync code.

SUPDrvGip.cpp: Enabled new tsc-delta measurement sync code w/ timeout checks.

Must check for timeout in the first loop iteration.

SUPDrvGip.cpp: The TSC-delta thread must clear the work mask even if the CPU got a delta value already or the timer callback will keep it busy for ever (not entirely sure where the darn race is, though). New TSC-delta synchronization code with timeout handling (disable).

SUPDrvGip.cpp: First part of the less noisy tsc-delta-measurment synchronization w/ timeout revamp.

docs.

nit

gee

SUPDrv,TM: Overhauled the CPU/TSC frequency a little, making it possible to run it early and skpping the 200 ms variant for the invariant tsc by executing the refinement timer several times.

Cleaning up the 2nd algo.

SUPDrvGip.cpp: First part of multiple tsc delta algorithm reorg.

SUPDrv,VMM: Added SUPR0TscDeltaMeasureBySetIndex to SUPDrv (bumping version req) so that VMMR0.cpp can, if necessary, trigger a delta measurement prior to doing RC and HM stuff. Also, made the context hook set the return-to-ring3 force flag if we're rescheduled on a CPU with an TSC delta needing measuring. This is probably code that will almost never get called...

SUPDrvGip.cpp: We can skip the determin-async-tsc test on invariant tsc CPUs now.

SUPDrvGip.cpp: Reordered the code, missed a bit.

SUPDrvGip.cpp: Reordered the code.

SUPDrv: Split out the GIP related code into SUPDrvGip.cpp.