Re: [PATCH] mm/slab: align kmalloc to cacheline when DMA API debugging is active

[Mikhail Gavrilov]

On Fri, Mar 27, 2026 at 11:38 AM Harry Yoo (Oracle) <harry@xxxxxxxxxx> wrote:
>
> On Fri, Mar 27, 2026 at 10:58:46AM +0500, Mikhail Gavrilov wrote:
> > When CONFIG_DMA_API_DEBUG is enabled, the DMA debug infrastructure
> > tracks active mappings per cacheline and warns if two different DMA
> > mappings share the same cacheline ("cacheline tracking EEXIST,
> > overlapping mappings aren't supported").
> >
> > On x86_64, ARCH_KMALLOC_MINALIGN defaults to 8, so small kmalloc
> > allocations (e.g. the 8-byte hub->buffer and hub->status in the USB
> > hub driver) frequently land in the same 64-byte cacheline.  When both
> > are DMA-mapped, this triggers a false positive warning.
>
> Is it feasible to suppress the warning if dma_get_cache_alignment() is
> smaller than L1_CACHE_BYTES?


Hi Harry,

Good question. I considered the dma-debug side, but the issue is
that the cacheline overlap check in add_dma_entry() is intentionally
strict -- it catches real bugs on non-coherent architectures where
two DMA buffers sharing a cacheline can corrupt data.

The check already has suppressions for DMA_ATTR_SKIP_CPU_SYNC,
DMA_ATTR_CPU_CACHE_CLEAN, and CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC +
swiotlb.  Adding another special case (e.g. dev_is_dma_coherent())
would weaken the check for all coherent platforms, potentially
hiding real bugs on devices behind non-coherent IOMMUs.

Alan Stern discussed this in the bugzilla [1] and concluded that
the slab alignment approach "seems reasonable" [2], noting that
"turning on debugging should not affect the way the kernel behaves --
otherwise what you're debugging isn't the same as what normally
happens. But given the way the DMA API debugging is set up, I don't
see any alternative."

The memory overhead is only present when CONFIG_DMA_API_DEBUG is
enabled, which is a debug-only option not used in production.

That said, if you'd prefer a dma-debug side fix, I'm happy to
explore that direction instead.

[1] https://bugzilla.kernel.org/show_bug.cgi?id=215740#c31
[2] https://bugzilla.kernel.org/show_bug.cgi?id=215740#c44

-- 
Best Regards,
Mike Gavrilov.