On Fri, Jul 24, 2015 at 04:45:23PM +0200, Vlastimil Babka wrote:
> The function alloc_pages_exact_node() was introduced in 6484eb3e2a81 ("page
> allocator: do not check NUMA node ID when the caller knows the node is valid")
No gold stars for that one.
> as an optimized variant of alloc_pages_node(), that doesn't allow the node id
> to be -1. Unfortunately the name of the function can easily suggest that the
> allocation is restricted to the given node and fails otherwise. In truth, the
> node is only preferred, unless __GFP_THISNODE is passed among the gfp flags.
>
> The misleading name has lead to mistakes in the past, see 5265047ac301 ("mm,
> thp: really limit transparent hugepage allocation to local node") and
> b360edb43f8e ("mm, mempolicy: migrate_to_node should only migrate to node").
>
> To prevent further mistakes and provide a convenience function for allocations
> truly restricted to a node, this patch makes alloc_pages_exact_node() pass
> __GFP_THISNODE to that effect. The previous implementation of
The change of what we have now is a good idea. What you have is a solid
improvement in my view but I see there are a few different suggestions
in the thread. Based on that I think it makes sense to just destroy
alloc_pages_exact_node. In the future "exact" in the allocator API will
mean "exactly this number of pages". Use your __alloc_pages_node helper
and specify __GFP_THISNODE if the caller requires that specific node.
> alloc_pages_exact_node() is copied as __alloc_pages_node() which implies it's
> an optimized variant of __alloc_pages_node() not intended for general usage.
> All three functions are described in the comment.
>
> Existing callers of alloc_pages_exact_node() are adjusted as follows:
> - those that explicitly pass __GFP_THISNODE keep calling
> alloc_pages_exact_node(), but the flag is removed from the call
__alloc_pages_node(__GFP_THISNODE) would be harder to get wrong in the future
> - others are converted to call __alloc_pages_node(). Some may still pass
> __GFP_THISNODE if they serve as wrappers that get gfp_flags from higher
> layers.
>
> There's exception of sba_alloc_coherent() which open-codes the check for
> nid == -1, so it is converted to use alloc_pages_node() instead. This means
> it no longer performs some VM_BUG_ON checks, but otherwise the whole patch
> makes no functional changes.
>
In general, checks for -1 should go away, particularly with new patches.
Use NUMA_NO_NODE.
--
Mel Gorman
SUSE Labs
--
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>