On 06/19/2016 01:35 AM, Benjamin Herrenschmidt wrote: > On Fri, 2016-06-17 at 16:32 +0200, Cédric Le Goater wrote: >> The instruction set PPC_POWER_BR contains nearly all the deleted >> instructions from isa2. rfi is not part of it and should. Also, only >> the cpus "PowerPC 601*" make a use of it in their insns_flags. > > Are you sure those arent the old POWER instructions as in pre-powerPC > architecture that 601 (and only 601) supports ?
OK. I get it now. All the deleted instructions from POWER are/should be under the set PPC_POWER. All the deleted instructions from POWER2 are under the set PPC_POWER2. None of these sets are in use. For the "PowerPC 601*" cpus, we moved a couple from set PPC_POWER to subset PPC_POWER_BR. rfi is special. it is under PPC_FLOW and all CPUs can use it >> So, we would want this set to be in all the "PowerPC {6,7}*" cpus. >> Are there more ? > > All 32-bit hash based CPUs are arch 1.x and support rfi > > All 64-bit hash based CPUs we support (ie, POWER4 and later) are > architecture 2.x and later. > > So my test is correct in the context of what we emulate today. OK. so this is an openbios issue when run under a ppc64. shouldn't we be using an openbios-ppc64 in that case ? We could be more strict with the rfi instruction. C.