On Wed, 19 Aug 2015, Jiang Liu wrote: > On 2015/8/18 8:31, David Rientjes wrote: > > On Mon, 17 Aug 2015, Jiang Liu wrote: > > > >> Function profile_cpu_callback() allocates memory without specifying > >> __GFP_THISNODE flag, so replace cpu_to_mem() with cpu_to_node() > >> because cpu_to_mem() may cause suboptimal memory allocation if > >> there's no free memory on the node returned by cpu_to_mem(). > >> > > > > Why is cpu_to_node() better with regard to free memory and NUMA locality? > Hi David, > Thanks for review. This is a special case pointed out by Tejun. > For the imagined topology, A<->B<->X<->C<->D, where A, B, C, D has > memory and X is memoryless. > Possible fallback lists are: > B: [ B, A, C, D] > X: [ B, C, A, D] > C: [ C, D, B, A] > > cpu_to_mem(X) will either return B or C. Let's assume it returns B. > Then we will use "B: [ B, A, C, D]" to allocate memory for X, which > is not the optimal fallback list for X. And cpu_to_node(X) returns > X, and "X: [ B, C, A, D]" is the optimal fallback list for X. Ok, that makes sense, but I would prefer that this alloc_pages_exact_node() change to alloc_pages_node() since, as you mention in your commit message, __GFP_THISNODE is not set. In the longterm, if we setup both zonelists correctly (no __GFP_THISNODE and with __GFP_THISNODE), then I'm not sure there's any reason to ever use cpu_to_mem() for alloc_pages(). -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>