On Fri, 25 Mar 2011, Tejun Heo wrote: > I've looked through the code but can't figure out what the difference > is. The memset code is in mm/percpu-vm.c::pcpu_populate_chunk(). > > for_each_possible_cpu(cpu) > memset((void *)pcpu_chunk_addr(chunk, cpu, 0) + off, 0, size); > > (pcpu_chunk_addr(chunk, cpu, 0) + off) is the same vaddr as will be > obtained by per_cpu_ptr(ptr, cpu), so all allocated memory regions are > accessed before being returned. Dazed and confused (seems like the > theme of today for me). > > Could it be that the vmalloc page is taking more than one faults? The vmalloc page only contains per cpu data from a single cpu right? Could anyone have set write access restrictions that would require a fault to get rid of? Or does an access from a different cpu require a "page table sync"? There is some rather strange looking code in arch/x86/mm/fault.c:vmalloc_fault -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>