Re: [PATCH] KVM: PPC: e500mc: Add support for single threaded vcpus on e6500 core

[Scott Wood <scottwood@freescale.com>]

On Tue, 2014-08-12 at 01:53 +0200, Alexander Graf wrote:
> 
> > Am 12.08.2014 um 01:36 schrieb Scott Wood <scottwood@freescale.com>:
> > 
> >> On Wed, 2014-08-06 at 19:33 +0300, Mihai Caraman wrote:
> >> @@ -390,19 +400,30 @@ static void kvmppc_core_vcpu_free_e500mc(struct kvm_vcpu *vcpu)
> >> 
> >> static int kvmppc_core_init_vm_e500mc(struct kvm *kvm)
> >> {
> >> -    int lpid;
> >> +    int i, lpid;
> >> 
> >> -    lpid = kvmppc_alloc_lpid();
> >> -    if (lpid < 0)
> >> -        return lpid;
> >> +    /* The lpid pool supports only 2 entries now */
> >> +    if (threads_per_core > 2)
> >> +        return -ENOMEM;
> >> +
> >> +    /* Each VM allocates one LPID per HW thread index */
> >> +    for (i = 0; i < threads_per_core; i++) {
> >> +        lpid = kvmppc_alloc_lpid();
> >> +        if (lpid < 0)
> >> +            return lpid;
> >> +
> >> +        kvm->arch.lpid_pool[i] = lpid;
> >> +    }
> > 
> > Wouldn't it be simpler to halve the size of the lpid pool that the
> > allocator sees, and just OR in the high bit based on the low bit of the
> > cpu number?
> 
> Heh, I wrote the same and then removed the section from my reply again. It wouldn't really make that much of a difference if you think it through completely.
> 
> But yes, it certainly would be quite a bit more natural. I'm ok either way.

It's not a huge difference, but it would at least get rid of some of the
ifdeffing in the headers.  It'd also be nicer when debugging to have the
LPIDs correlated.

-Scott