This is just a quick explaination of the problems fixed by the page table races patch. 1. use kmap_atomic in clear_user_*page - fixes potential schedule while holding page table spinlock under heavy load. 2. double spin_unlock fix in memory.c - code could spin_unlock twice under rare alignments while copying page tables 3. make map_user_kiobuf use read lock - prevents stalls during page faults waiting for io to complete 4. move page copy/clear out from page table spinlock - reduce spinlock contention