On 11/06/12 18:46, Alexander Graf wrote: > On 06/11/2012 04:06 PM, Cornelia Huck wrote: >> From: Christian Borntraeger<borntraeger@xxxxxxxxxx> >> >> The initial cpu reset sets the cpu in the stopped state. >> Several places check for the cpu state (e.g. sigp set prefix) and >> not setting the STOPPED state triggered errors with newer guest >> kernels after reboot. >> >> Signed-off-by: Christian Borntraeger<borntraeger@xxxxxxxxxx> >> Signed-off-by: Cornelia Huck<cornelia.huck@xxxxxxxxxx> > > Is this only broken as of now or has this been broken before? So do we need to send this to -stable for anyone? And what happens on RESET? Looks like this problem appeared as a combination of newer guest kernels (something like >=3.2 or so) with a reworked SMP bringup behaviour and a host fix (151104a7b3a82f9c56d636595ae58084049d2559 KVM: s390: make sigp restart return busy when stop pending) which both are correct in itself but uncovered this third problem. With this patch I have a qemu/kernel combination which works reliably with different guest kernels for guest initiated resets as well as qemu monitor system_reset as well as system_shutdown+system_reset+cont. It will probably not be the last fix (e.g. initial reset should also take care of pending CPU local interrupts) but we are making progress. Christian PS: The working combination has two patches pending for qemu which need some cleanup, will send soon -- 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
