On Wed, 17 Nov 2010, Andreas Dilger wrote: > >> - avoid doing anything other than GFP_KERNEL allocations for __vmalloc(): > >> the only current users are gfs2, ntfs, and ceph (the page allocator > >> __vmalloc() can be discounted since it's done at boot and GFP_ATOMIC > >> here has almost no chance of failing since the size is determined based > >> on what is available). > > > > ^^ this > > > > Using vmalloc anywhere is lame. > > I agree. What we really want is 1MB kmalloc() to work... :-/ > Order-8 allocations are already have a higher liklihood of succeeding because of memory compaction, which was explicitly targeted to aid in order-9 hugepage allocations. The problem is that it's useless for GFP_NOFS. I think removing gfp_t arguments from all of the public vmalloc interface will inevitably be where we go with this and everything will assume GFP_KERNEL | __GFP_HIGHMEM. If you _really_ need 1MB of physically contiguous memory, then you'll need to find a way to do it in a reclaimable context. If we actually can remove the dependency that gfs2, ntfs, and ceph have in the kernel.org kernel, then this support may be pulled out from under you; the worst-case scenario for Lustre is that you'll have to modify the callchains like I suggested in my original email to pass the gfp mask all the way down to the pte allocators if you can't find a way to do it under GFP_KERNEL. -- 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 policy in Canada: sign http://dissolvethecrtc.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>