On Mon, 29 Nov 2021, Rustam Kovhaev wrote:

quoted

I'm not sure that helps.
The header can't share a cache line with the previous item (because it
might be mapped for DMA) so will always take a full cache line.

I thought that DMA API allocates buffers that are larger than page size.
DMA pool seems to be able to give out smaller buffers, but underneath it
seems to be calling page allocator.
The SLOB objects that have this header are all less than page size, and
they cannot end up in DMA code paths, or can they?

kmalloc slab allocations must return dmaable memory. If the underlying
hardware can only dma to a cache line border then all objects must be
aligned that way.