Hi, By the beginning of February I posted an RFC regarding an approach to alleviate time drift with HPET periodic timers. Ref.: http://article.gmane.org/gmane.comp.emulators.kvm.devel/67346 http://lists.gnu.org/archive/html/qemu-devel/2011-02/msg00327.html During the discussion it became apparent that a generic API that could facilitate (re)injection of lost timer interrupts for any timer device emulation would be preferred over an isolated fix for a specific device. However, at this stage there appears to be no plan to actually implement the generic API. Therefore, we may want to reconsider an improved version of the code that I originally included in the RFC. This improved version compensates the loss of timer interrupts caused by ... - delayed timer callback. - interrupt coalescing (x86 only). Lost timer interrupts are compensated by gradually injecting additional interrupts during the subsequent timer intervals, starting at a rate of one additional interrupt per interval. If further interrupts are lost while compensation is still in progress, the rate is increased. A limit is imposed on the rate and on the 'backlog' of lost interrupts that are to be injected. Compensation of lost timer interrupts can be enabled via a new QEMU command line option [1]. -hpet [device=none|present][,driftfix=none|slew] The code is enclosed in '#ifdef CONFIG_HPET_DRIFTFIX ... #endif' and it is configurable via a new option '--enable-hpet-driftfix' for the 'configure' script. This also makes it easy to identify and to remove if the generic API would be implemented some day. I've tested this code primarily by playing a list of example video files continuously in a Windows Vista(tm) guest. In some cases I observed the guest time fall 15+ seconds behind KVM host time in one minute without the patch, whereas with the patch there was no visible difference between the guest time and KVM host time after several hours of testing [2]. The 'qemu-system-x86_64' test binary image file was built from snapshot ... http://git.kernel.org/?p=virt/kvm/qemu-kvm.git;a=commit;h=e700699e9f6f1e163985a8e4cd76d8568d24c152 ... using only the '--prefix' and '--enable-hpet-driftfix' options of the 'configure' script (other options default). The series of patches is based on the aforementioned snapshot and it consists of three parts. - Part 1 introduces the new QEMU command line option. - Part 2 implements the method to compensate lost timer interrupts. - Part 3 introduces the new option for the 'configure' script. Please review and please comment. Regards, Uli Obergfell [1] Injecting additional timer interrupts to compensate lost interrupts can alleviate long term time drift. However, on a short time scale, this method can have the side effect of making virtual machine time intermittently pass faster than real time (depending on the guest's time keeping algorithm). Compensation is disabled by default and can be enabled for guests where this behaviour is acceptable. [2] Testing required a patch in tight_compress_data(), ui/vnc-enc-tight.c to avoid segmentation fault / core dump of qemu. However, this needs to be discussed in a separate post. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
