On Fri, 20 Apr 2012 21:55:55 -0300 Marcelo Tosatti <mtosatti@xxxxxxxxxx> wrote: > More importantly than the particular flush TLB case, the point is > every piece of code that reads and writes sptes must now be aware that > mmu_lock alone does not guarantee stability. Everything must be audited. In addition, please give me some stress-test cases to verify these in the real environments. Live migration with KSM, with notifier call, etc? Although the current logic is verified by dirty-log api test, the new logic may need another api test program. Note: the problem is that live migration can fail silently. We cannot know the data loss is from guest side problem or get_dirty side. > Where the bulk of the improvement comes from again? If there is little > or no mmu_lock contention (which we have no consistent data to be honest > in your testcase) is the bouncing off mmu_lock's cacheline that hurts? This week, I was doing some simplified "worst-latency-tests" for my work. It was difficult than I thought. But Xiao's "lock-less" should see the reduction of mmu_lock contention more easily, if there is really some. To make things simple, e.g., we can do the same kind of write-loop as XBZRLE people are doing in the guest - with more VCPUs if possible. Thanks, Takuya -- 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
