Is there a reason why for 1G and 2M requests setup_guest() does pass XENMEMF_populate_on_demand, yet in the event of falling back to 4k allocations it doesn''t? A POD guest that can''t get large pages allocated seems penalized enough already, so I don''t see why its startup gets even further slowed down by requiring the pages to be reclaimed (if that''s possible at all, considering that the pages don''t get cleared explicitly) rather than added to the POD cache from the beginning. Thanks, Jan