Hello Helge, Thank you for picking up patches #1 and #2. Sorry for the very slow reply; I only recently caught up with this thread.
The zicntr=false gating patch (commit ebfde6dafff2) is in master and I have not seen any follow-up problems with it. The reservation invalidation patch (commit 5089493e45d3) is a different story. I have followed the performance report behind the revert proposal: the patch sends every store through EXCP_ATOMIC and walks all CPUs on each store, which turns an xz run on 2 MB into minutes instead of seconds. I agree that cost is not acceptable, so please let it be reverted. I will not ask for it to be kept in its current form. If I can work out a version that avoids the global serialization, I will send it as a separate follow-up with the multi-threaded numbers in the cover letter. Until then the reverted state is the right one. No action needed from you on either patch. Thanks again for taking care of them. Regards, Wangyang
