> is a nointr pool safe for this pmap? I could be stupid in this area.
> iirc x86 pmap allocates pv entries from kmem_map to avoid this kind of > recursion. I see, but x86 pmap has been changed to use pool_cache(9) with pool_allocator_meta during vmlocking merge. http://cvsweb.NetBSD.org/bsdweb.cgi/src/sys/arch/x86/x86/pmap.c.diff?r1=1.2&r2=1.3 Is it enough to use pool(9) with pool_allocator_meta (which uses kmem_map), or should I revert that pool change and just make uvm_km_alloc() in old pmap_alloc_pv() use kmem_map instead of kernel_map? --- Izumi Tsutsui