On Mon, Aug 20, 2018 at 03:54:02PM +0200, Paolo Bonzini wrote:
>On 19/08/2018 02:17, Wei Yang wrote:
>> This is a trivial optimization of the first loop in update_memslots().
>>
>> Since used_slots records the number of valid slots, it could be used as the
>> boundary of the loop instead of looping the whole array and check npages.
>>
>> Signed-off-by: Wei Yang <richard.weiyang@xxxxxxxxx>
>> ---
>> virt/kvm/kvm_main.c | 6 ++----
>> 1 file changed, 2 insertions(+), 4 deletions(-)
>>
>> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
>> index 5c1911790f25..fac3225eff35 100644
>> --- a/virt/kvm/kvm_main.c
>> +++ b/virt/kvm/kvm_main.c
>> @@ -806,15 +806,13 @@ static void update_memslots(struct kvm_memslots *slots,
>> {
>> int id = new->id;
>> int i = slots->id_to_index[id];
>> + int used_slots = slots->used_slots;
>> struct kvm_memory_slot *mslots = slots->memslots;
>>
>> WARN_ON(mslots[i].id != id);
>> slots->used_slots += (char)change;
>>
>> - while (i < KVM_MEM_SLOTS_NUM - 1 &&
>> - new->base_gfn <= mslots[i + 1].base_gfn) {
>> - if (!mslots[i + 1].npages)
>> - break;
>> + while (i < used_slots - 1 && new->base_gfn <= mslots[i + 1].base_gfn) {
>> mslots[i] = mslots[i + 1];
>> slots->id_to_index[mslots[i].id] = i;
>> i++;
>>
>
>The loop is removing an element from the array, so the correct limit for
>the index would be the *old* length of the array, not the new one.
Yes, this code checks with the *old* length, which is saved at beginning
of the function instead of retrieving it from slots->used_slots.
>Alternatively... just keep the current code. :) It is already loading
>mslots[i+1] to get the base_gfn, so the load is not expensive. This is
>also not really a fast path.
Ah, you are right, mslots[i+1].base_gfn already loaded to memory, so it
won't be expensive to access mslots[i+1].npages.
My intention is to make it *looks* better, by which the loops just use
the number of valid slots as the boundary instead of checking both array
size and npages.
At last, I agree with you that it won't do much to help the performance :)
>
>Paolo
--
Wei Yang
Help you, Help me
