On 18.01.2013, at 01:11, Scott Wood wrote:
> On 01/17/2013 04:50:39 PM, Alexander Graf wrote:
>> @@ -1024,9 +1001,11 @@ void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
>> {
>> struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
>> struct tlbe_priv *priv;
>> - struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
>> + struct kvm_book3e_206_tlb_entry *gtlbe, stlbe = {};
>
> Is there a code path in which stlbe gets used but not fully filled in
> without this?
I am hoping not, but when I wrote this patch gcc suddenly jumped at me claiming that the whole struct can get used uninitialized:
arch/powerpc/kvm/e500_mmu_host.c: In function ‘kvmppc_mmu_map’:
arch/powerpc/kvm/e500_mmu_host.c:533: error: ‘stlbe.mas1’ may be used uninitialized in this function
arch/powerpc/kvm/e500_mmu_host.c:533: error: ‘stlbe.mas2’ may be used uninitialized in this function
arch/powerpc/kvm/e500_mmu_host.c:533: error: ‘stlbe.mas7_3’ may be used uninitialized in this function
arch/powerpc/kvm/e500_mmu_host.c:533: error: ‘stlbe.mas8’ may be used uninitialized in this function
If you have any idea where this could come from, please let me know :).
>
>> int tlbsel = tlbsel_of(index);
>> int esel = esel_of(index);
>> + /* Needed for initial map, where we can't use the cached value */
>> + int force_map = index & KVM_E500_INDEX_FORCE_MAP;
>> int stlbsel, sesel;
>> gtlbe = get_entry(vcpu_e500, tlbsel, esel);
>> @@ -1038,7 +1017,7 @@ void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
>> priv = &vcpu_e500->gtlb_priv[tlbsel][esel];
>> /* Only triggers after clear_tlb_refs */
>> - if (unlikely(!(priv->ref.flags & E500_TLB_VALID)))
>> + if (force_map || unlikely(!(priv->ref.flags & E500_TLB_VALID)))
>> kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe);
>> else
>> kvmppc_e500_setup_stlbe(vcpu, gtlbe, BOOK3E_PAGESZ_4K,
>
> It seems a bit odd to overload index rather than adding a flags
> parameter...
Yeah, I mostly wanted to refrain from touching 440 code. But if you prefer that, I can certainly do so :).
> It also seems like it would be cleaner to just invalidate the old entry
> in tlbwe, and then this function doesn't need to change at all. I am a
> bit confused by how invalidation is currently operating -- why is
> E500_TLB_VALID not cleared on invalidations (except for MMU API stuff and
> MMU notifiers)?
Consider me as confused as you are.
Alex
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
