Re: [PATCH 1/2] KVM: PPC: e6500: Handle LRAT error exception

[Scott Wood <scottwood@freescale.com>]

On Wed, 2015-09-30 at 14:27 +0300, Laurentiu Tudor wrote:
> On 09/30/2015 01:32 PM, Laurentiu Tudor wrote:
> > On 09/25/2015 03:10 AM, Scott Wood wrote:
> > > On Thu, 2015-09-24 at 16:11 +0300, Laurentiu Tudor wrote:
> 
> [snip]
> 
> > > > b/arch/powerpc/kvm/e500_mmu_host.c
> > > > index 12d5c67..99ad88a 100644
> > > > --- a/arch/powerpc/kvm/e500_mmu_host.c
> > > > +++ b/arch/powerpc/kvm/e500_mmu_host.c
> > > > @@ -96,6 +96,112 @@ static inline void __write_host_tlbe(struct 
> > > > kvm_book3e_206_tlb_entry *stlbe,
> > > >                                     stlbe->mas2, stlbe->mas7_3);
> > > >  }
> > > >  
> > > > +#if defined(CONFIG_64BIT) && defined(CONFIG_KVM_BOOKE_HV)
> > > > +static int lrat_next(void)
> > > > +{
> > > 
> > > Will anything break by removing the CONFIG_64BIT condition, even if we 
> > > don't 
> > > have a 32-bit target that uses this?
> > 
> > Not completly certain but i remember getting compile or link errors
> > on 32-bit e500mc or e500v2. I can recheck if you want.
> > 
> > > > +void kvmppc_lrat_map(struct kvm_vcpu *vcpu, gfn_t gfn)
> > > > +{
> > > > +     struct kvm_memory_slot *slot;
> > > > +     unsigned long pfn;
> > > > +     unsigned long hva;
> > > > +     struct vm_area_struct *vma;
> > > > +     unsigned long psize;
> > > > +     int tsize;
> > > > +     unsigned long tsize_pages;
> > > > +
> > > > +     slot = gfn_to_memslot(vcpu->kvm, gfn);
> > > > +     if (!slot) {
> > > > +             pr_err_ratelimited("%s: couldn't find memslot for gfn 
> > > > %lx!\n",
> > > > +                                __func__, (long)gfn);
> > > > +             return;
> > > > +     }
> > > > +
> > > > +     hva = slot->userspace_addr;
> > > 
> > > What if the faulting address is somewhere in the middle of the slot?  
> > > Shouldn't you use gfn_to_hva_memslot() like kvmppc_e500_shadow_map()?  
> > > In 
> > > fact there's probably a lot of logic that should be shared between 
> > > these two 
> > > functions.
> > 
> > So if my understanding is correct most of the gfn -> pfn translation
> > stuff done in kvmppc_e500_shadow_map() should also be present in here.
> > If that's the case maybe i should first extract this code (which includes
> > VM_PFNMAP handling) in a separate function and call it from both 
> > kvmppc_lrat_map()
> > and kvmppc_e500_shadow_map(). 
> > 
> 
> Off-topic, but just noticed that kvmppc_e500_shadow_map() is marked as 
> inline.
> Was that on purpose? Is inlining such a large function worth anything?

I don't remember the intent.  It was probably a lot smaller back then.  That 
said, it's only used two places, with probably pretty good temporal 
separation between performance-intensive uses of one versus the other (so not 
a huge icache concern), and a pretty good portion of the function will be 
optimized out in the caller with tlbsel = 0.

-Scott