Hello! > In helpr.c:mpidr_read we calculate aff0 & aff1 from cs->cpu_index in the > same way but AFAIK > the input parameter i.e. CPUARMState doesn't has the #cores-in-soc nor any > other useful > information. We can add this field and initialize it in virt.c but then we > need to touch all other > machines. Or we can say that if this number is zero than there is no aff1 and > aff0 is cs- > >cpu_index. Please note that this is not according to spec but this is the > >current implementation > (before the GICv3 patch) so according to the purists the current > implementation with GICv2 > should be limited to 4 cortex-a57 cores! > In psci.c we need to break the aff0 & aff1 that the kernel is giving us back > to linear number for > this we again need the #cores-is-soc, but again all we have is ARMCPU and > CPUARMState.
Sorry for perhaps being too insisting, but have you seen my implementation? Nobody seems to comment on it... It is very easy to convert into what you are suggesting. It already has CPU_PER_CORES parameter which is currently a #define. If you turn it into a global variable, this will be the only change you need. It would default to 8 so that old machines don't need to be changed. But you could set it from inside virt before instantiating CPUs if you want to change it. When CPUs are instantiated their IDs will be auto-generated from index (raw order number) according to this parameter. Kind regards, Pavel Fedin Expert Engineer Samsung Electronics Research center Russia
