regression with SG DMA buf allocations and IOMMU in low-mem

[Kai Vehmanen]

Hi Takashi and list,

I've been debugging a SOF DSP load fail on VT-D/IOMMU systems [1] and 
that brought me to these two commits from you:

"ALSA: memalloc: Add fallback SG-buffer allocations for x86"
and

"ALSA: memalloc: Revive x86-specific WC page allocations again"

We have an allocation with:
snd_dma_alloc_pages(SNDRV_DMA_TYPE_DEV_SG, &pci->dev, size, dmab);

.. but in rare low-memory cases, the dma_alloc_noncontiguous()
call will fail in core/memalloc.c and we go to the fallback path.

So we go to snd_dma_sg_fallback_alloc(), but here it seems something is 
missing. The pages are not allocated with DMA mapping API anymore, so 
IOMMU won't know about the memory and in our case the DSP load will fail 
to a IOMMU fault. 

Looking at 5.15 code, the fallback looks very different, but 
in fallback we still use the DMA mapping API via snd_dma_dev_alloc()
so even if we go to fallback path, mapping is still ok.

Any ideas which way this should be fixed? Given the many changes
I thought it's better to ask early on the list about this.

[1] https://github.com/thesofproject/linux/issues/3915

Br, Kai