Re: [PATCH 2/3] KVM: x86/mmu: Drop a redundant remote TLB flush in kvm_zap_gfn_range()

["Maciej S. Szmigiero" <maciej.szmigiero@oracle.com>]

On 25.10.2021 17:39, Sean Christopherson wrote:
> On Fri, Oct 22, 2021, Maciej S. Szmigiero wrote:
>> On 22.10.2021 03:00, Sean Christopherson wrote:
>>> Remove an unnecessary remote TLB flush in kvm_zap_gfn_range() now that
>>> said function holds mmu_lock for write for its entire duration.  The
>>> flush was added by the now-reverted commit to allow TDP MMU to flush while
>>> holding mmu_lock for read, as the transition from write=>read required
>>> dropping the lock and thus a pending flush needed to be serviced.
>>>
>>> Fixes: 5a324c24b638 ("Revert "KVM: x86/mmu: Allow zap gfn range to operate under the mmu read lock"")
>>> Cc: Maxim Levitsky <mlevitsk@redhat.com>
>>> Cc: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
>>> Cc: Ben Gardon <bgardon@google.com>
>>> Signed-off-by: Sean Christopherson <seanjc@google.com>
>>> ---
>>>    arch/x86/kvm/mmu/mmu.c | 3 ---
>>>    1 file changed, 3 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
>>> index f82b192bba0b..e8b8a665e2e9 100644
>>> --- a/arch/x86/kvm/mmu/mmu.c
>>> +++ b/arch/x86/kvm/mmu/mmu.c
>>> @@ -5700,9 +5700,6 @@ void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end)
>>>    						end - 1, true, flush);
>>>    			}
>>>    		}
>>> -		if (flush)
>>> -			kvm_flush_remote_tlbs_with_address(kvm, gfn_start,
>>> -							   gfn_end - gfn_start);
>>>    	}
>>>    	if (is_tdp_mmu_enabled(kvm)) {
>>>
>>
>> Unfortunately, it seems that a pending flush from __kvm_zap_rmaps()
>> can be reset back to false by the following line:
>>> flush = kvm_tdp_mmu_zap_gfn_range(kvm, i, gfn_start, gfn_end, flush);
>>
>> kvm_tdp_mmu_zap_gfn_range() calls __kvm_tdp_mmu_zap_gfn_range with
>> "can_yield" set to true, which passes it to zap_gfn_range, which has
>> this code:
>>> if (can_yield &&
>>>      tdp_mmu_iter_cond_resched(kvm, &iter, flush, shared)) {
>>>        flush = false;
>>>        continue;
>>> }
> 
> That's working by design.  If the MMU (legacy or TDP) yields during zap, it _must_
> flush before dropping mmu_lock so that any SPTE modifications are guaranteed to be
> observed by all vCPUs.  Clearing "flush" is deliberate/correct as another is flush
> is needed if and only if additional SPTE modifications are made.
> 

Got it, thanks.

Maciej