On Sat, 2005-01-29 at 17:03 -0800, Mark Wong wrote: > On Fri, Jan 28, 2005 at 11:07:04AM -0800, Dave Hansen wrote: > > On Thu, 2005-01-27 at 18:38 -0800, Mark Wong wrote: > > > On Thu, Jan 27, 2005 at 04:35:52PM -0800, Dave Hansen wrote: > > > > Sigh.... Looks like an old CONFIG_NONLINEAR managed to stay around. > > > > Please give the attached patch a shot. Thanks for all of the patience. > > > > There are obviously some growing pains with the new sparsemem port to > > > > the hotplug code (all my fault). > > > > > > Ok, how do these look? > > > > Those look pretty good. The section failure allocations are probably > > due to memory becoming fragmented, and not allowing order-10 allocations > > later in boot. This is quite expected on an i386 machine. But, that's > > just a guess unless I see /proc/buddyinfo. > > Is this snapshot sufficient, or should I redo it after I run the > tests? > > # cat /proc/buddyinfo > Node 0, zone DMA 59 3 0 1 0 0 0 0 0 1 0 > Node 0, zone Normal 512 101 20 0 0 1 1 0 1 1 0 That data isn't completely conclusive, but should be good enough. See the 0's in the last column? That's the pool that the setion_mem_map has to be allocated out of. There appears to be a healthy amount of pages around otherwise, just not in contiguous bits. I'll see what I can come up with to work around this for now. -- Dave
