On 04.03.15 15:14, Andreas Färber wrote: > Alex, > > Am 04.03.2015 um 13:28 schrieb Alexander Graf: >> On 04.03.15 02:31, Alexey Kardashevskiy wrote: >>> 5b79b1c "target-ppc: Create versionless CPU class per family if KVM" added >>> a dynamic CPU class registration with the name of the CPU family which >>> QEMU is running on. For example, this allowed specifying "-cpu POWER7" >>> on every version of POWER7 machine, not just the one which POWER7 was >>> an alias of. I.e. before 5b79b1c, "-cpu POWER7" would not work on real >>> POWER7 2.1 and would work on POWER7 2.3 only. The same story for POWER8. >>> >>> However that patch broke POWER5+ support as POWER5+ CPU uses the same >>> name as the CPU class so dynamic registering of the POWER5+ class failed. >>> >>> This redefines POWER5+ server CPUs by adding a version to them and adding >>> an alias for TCG case. KVM will use dynamically registered CPUs. >>> >>> While we are here, do the same for 970 CPU. >>> >>> Signed-off-by: Alexey Kardashevskiy <a...@ozlabs.ru> >> >> Thanks, applied to ppc-next. > > As Alexey predicted, I object. > > The 970 part looks good and could be applied immediately if it were a > separate patch. > > But the POWER5+ part I have my doubts about: Was there really a v0.0???
IIUC most IBM POWER PVRs are regular in that their lower 16 bits are major/minor. The PVR we have as "POWER5+" has those bits as 0, thus v0.0 is the correct translation of that. Whether that CPU ever existed is a different question and arguably out of scope for the patch. > Others start with v1.0 and I have: > > revision : 2.1 (pvr 003b 0201) I wouldn't object to removing the v0.0 version altogether and just make POWER5+ be an alias to v2.1 instead. But that's something for a follow-up patch ;). Alex
