Re: [PATCH v2] dma-debug: suppress cacheline overlap warning when arch has no DMA alignment requirement
From: Robin Murphy
Date: Wed Apr 01 2026 - 11:36:15 EST
On 2026-04-01 2:25 pm, Andy Shevchenko wrote:
On Wed, Apr 1, 2026 at 3:11 PM Robin Murphy <robin.murphy@xxxxxxx> wrote:
On 2026-03-30 8:44 am, Marek Szyprowski wrote:
On 27.03.2026 13:41, Mikhail Gavrilov wrote:
...
TBH I'd be inclined to have CONFIG_DMA_DEBUG raise ARCH_DMA_MINALIGN as
appropriate such that genuine false-positives can't happen, rather than
effectively defeat the whole check,
I dunno if you read v1 thread, where I proposed to unroll the check
and use pr_debug_once() for the cases which we expect not to panic,
but would be good to have a track of.
I had not seen v1, as I took the last 3 days off and hadn't got that far up my inbox yet - I guess it's at least reassuring to have reached similar conclusions independently :)
The fundamental issue here is that dma-debug doesn't realistically have a way to know whether the thing being mapped is intentionally a whole dedicated kmalloc allocation - where we can trust SLUB (and DMA_BOUNCE_UNALIGNED_KMALLOC if appropriate) to do the right thing across different systems - or just something which might happen to line up by coincidence on someone's development machine, but for portability they definitely do still need to take explicit care about (e.g. struct devres::data).
Certainly some milder and/or opt-in reporting of potential-false-positive overlaps wouldn't be a terrible idea, if someone wants to have a crack at it.
Thanks,
Robin.