Re: [PATCH] kvm: don't lose the higher 32 bits of tlbs_dirty

[Paolo Bonzini <pbonzini@redhat.com>]

On 14/12/20 18:20, Sean Christopherson wrote:
> On Sun, Dec 13, 2020, Lai Jiangshan wrote:
>> From: Lai Jiangshan <laijs@linux.alibaba.com>
>>
>> In kvm_mmu_notifier_invalidate_range_start(), tlbs_dirty is used as:
>> 	need_tlb_flush |= kvm->tlbs_dirty;
>> with need_tlb_flush's type being int and tlbs_dirty's type being long.
>>
>> It means that tlbs_dirty is always used as int and the higher 32 bits
>> is useless.
> 
> It's probably worth noting in the changelog that it's _extremely_ unlikely this
> bug can cause problems in practice.  It would require encountering tlbs_dirty
> on a 4 billion count boundary, and KVM would need to be using shadow paging or
> be running a nested guest.
> 
>> We can just change need_tlb_flush's type to long to
>> make full use of tlbs_dirty.
> 
> Hrm, this does solve the problem, but I'm not a fan of continuing to use an
> integer variable as a boolean.  Rather than propagate tlbs_dirty to
> need_tlb_flush, what if this bug fix patch checks tlbs_dirty directly, and then
> a follow up patch converts need_tlb_flush to a bool and removes the unnecessary
> initialization (see below).

Indeed, the compiler should be able to convert || to | if useful and 
valid (it may or may not do it depending on the sizes of types involved, 
but that's Someone Else's Problem and this is not really a path where 
every instruction matter).

Paolo

> E.g. the net result of both patches would be:
> 
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 3abcb2ce5b7d..93b6986d3dfc 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -473,7 +473,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
>                                          const struct mmu_notifier_range *range)
>   {
>          struct kvm *kvm = mmu_notifier_to_kvm(mn);
> -       int need_tlb_flush = 0, idx;
> +       bool need_tlb_flush;
> +       int idx;
> 
>          idx = srcu_read_lock(&kvm->srcu);
>          spin_lock(&kvm->mmu_lock);
> @@ -483,11 +484,10 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
>           * count is also read inside the mmu_lock critical section.
>           */
>          kvm->mmu_notifier_count++;
> -       need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
> -                                            range->flags);
> -       need_tlb_flush |= kvm->tlbs_dirty;
> +       need_tlb_flush = !!kvm_unmap_hva_range(kvm, range->start, range->end,
> +                                              range->flags);
>          /* we've to flush the tlb before the pages can be freed */
> -       if (need_tlb_flush)
> +       if (need_tlb_flush || kvm->tlbs_dirty)
>                  kvm_flush_remote_tlbs(kvm);
> 
>          spin_unlock(&kvm->mmu_lock);
> 
> Cc: stable@vger.kernel.org
> Fixes: a4ee1ca4a36e ("KVM: MMU: delay flush all tlbs on sync_page path")
> 
>> Signed-off-by: Lai Jiangshan <laijs@linux.alibaba.com>
>> ---
>>   virt/kvm/kvm_main.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
>> index 2541a17ff1c4..4e519a517e9f 100644
>> --- a/virt/kvm/kvm_main.c
>> +++ b/virt/kvm/kvm_main.c
>> @@ -470,7 +470,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
>>   					const struct mmu_notifier_range *range)
>>   {
>>   	struct kvm *kvm = mmu_notifier_to_kvm(mn);
>> -	int need_tlb_flush = 0, idx;
>> +	long need_tlb_flush = 0;
> 
> need_tlb_flush doesn't need to be initialized here, it's explicitly set via the
> call to kvm_unmap_hva_range().
> 
>> +	int idx;
>>   
>>   	idx = srcu_read_lock(&kvm->srcu);
>>   	spin_lock(&kvm->mmu_lock);
>> -- 
>> 2.19.1.6.gb485710b
>>
>