On 16/08/2018 09:01, Jan Beulich wrote: >>>> On 26.06.18 at 09:18, wrote: >> The STI instances were moved (or added in the INT80 case) to meet TLB >> flush requirements. When XPTI is disabled, they can be put back where >> they were (or omitted in the INT80 case). >> >> Signed-off-by: Jan Beulich <jbeul...@suse.com> >> Tested-by: Juergen Gross <jgr...@suse.com> >> Reviewed-by: Juergen Gross <jgr...@suse.com> >> Reviewed-by: Wei Liu <wei.l...@citrix.com> > I'd not be overly concerned if this was nack-ed for an acceptable > reason, but I'd like to hear so in that case, because the alternative > to this patch is to drop the synthetic NO_XPTI feature flag, which > is not used anywhere else without this change.
There are 2 issues to consider. 1) Juergens performance numbers said this make things worse. (Mean around the same, std deviation shot up iirc) 2) You are vastly complicating the entry paths, because whether interrupts are enabled or not is now variable on runtime configuration. I don't see any benefit from this patch, and I see several negatives, both in terms of performances, and in terms of how easy the code is to reason about. ~Andrew _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
