Re: [RFC 07/11] eth: ixgbe: use dma_alloc_coherent_node()

[Jakub Kicinski <kuba@kernel.org>]

On Tue, 27 Oct 2026 17:06:44 +0100 Björn Töpel wrote:
> Jakub Kicinski <kuba@kernel.org> writes:
> > On Tue, 27 Oct 2026 10:06:03 +0100 Björn Töpel wrote:  
> >> Not really this series, but the fallback here is odd, no? I thought the
> >> allocator already did cross-zone fallback?  
> >
> > Fair, I was too lazy to check. I'll remove the fallbacks in the same
> > patch. Seems obvious enough. LMK if you prefer separate.  
> 
> Hmm. Seems like dma_alloc_contiguous() doesn't spill over to other
> nodes. It only tries the per-NUMA CMA... so keeping it is bit more
> robust. (Then again, I dunno...)
> 
> I'd say keep it.

Or we fold the retry into the helper? I tried to write the kdoc:

/**
 * dma_alloc_attrs_node - allocate coherent memory on a given NUMA node
 * @dev: device to allocate the memory for
 * @size: size of the allocation
 * @dma_handle: returns the DMA address of the allocation
 * @flag: GFP flags for the allocation
 * @attrs: DMA_ATTR_* flags for the allocation
 * @node: NUMA node to allocate the memory on, NUMA_NO_NODE for no preference
 *
 * Same as dma_alloc_attrs() but places the memory on @node rather than on the
 * node of @dev. Like with kmalloc_node() the node is only a preference,
 * the page allocator may fall back to other nodes if @node is out of memory.
 * The contiguous memory allocator (CMA) does not: only the CMA area of @node
 * and the global CMA area are tried. The allocation may fail even though the
 * CMA area of another node, e.g. that of @dev, still has free space.
 *
 * Allocations served from per-device coherent pools or CMA areas, restricted
 * or atomic pools, or by dma_map_ops implementations do not take @node into
 * account.
 *
 * Return: the CPU address of the allocation, or NULL on failure.
 */

This feels like too much explaining of the internals for a commonly
used APIs. If we retry in the helper all we have to say is basically
"acts like kmalloc_node()".

WDYT?