28.05.2021 14:01, Paolo Bonzini wrote:
On 28/05/21 12:24, Paolo Bonzini wrote:

It's still more complicated, because you need to add some kind of

     method = s->method;

This would even have to be a separate, one-line critical section...


Hm, so, we should set .use_copy_range in task, when it is initialized.


     ret = block_copy_do_copy(..., method);
     if (ret < 0 && method <= COPY_RANGE_SMALL) {
         method = COPY_RANGE_READ_WRITE;
         ret = block_copy_do_copy(..., method);
         }
     lock();
         if (method == s->method) {
             /* compute new method */
         }

which makes it more complicated than this patch IMO.  But yeah at least it's a 
viable alternative to the atomics.



OK, I'm OK with patch as is. Finally I can refactor it later on top if needed.. 
I'll try now do some refactoring, you'll probably want to base on it, or 
vise-versa, I'll rebase it later on top of these patches.

--
Best regards,
Vladimir

Reply via email to