Re: [PATCH v3 2/2] block-dma: properly take MMIO path

[Leon Romanovsky <leon@kernel.org>]

On Wed, Nov 12, 2025 at 09:38:07AM -0500, Keith Busch wrote:
> On Mon, Oct 27, 2025 at 09:30:21AM +0200, Leon Romanovsky wrote:
> > @@ -732,13 +746,20 @@ static void nvme_unmap_metadata(struct request *req)
> >  		return;
> >  	}
> >  
> > -	if (!blk_rq_integrity_dma_unmap(req, dma_dev, &iod->meta_dma_state,
> > -					iod->meta_total_len)) {
> > +	if (iod->flags & IOD_META_P2P)
> > +		map = PCI_P2PDMA_MAP_BUS_ADDR;
> > +	if (iod->flags & IOD_META_MMIO) {
> > +		map = PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
> > +		attrs |= DMA_ATTR_MMIO;
> > +	}
> 
> This should be an 'else if' no?

We can't have both IOD_META_P2P and IOD_META_MMIO. It can be only one or
IOD_META_P2P or IOD_META_MMIO. In this case "else if' is not necessary.

> 

<...>

> > @@ -1036,6 +1067,17 @@ static blk_status_t nvme_map_data(struct request *req)
> >  	if (!blk_rq_dma_map_iter_start(req, dev->dev, &iod->dma_state, &iter))
> >  		return iter.status;
> >  
> > +	switch (iter.p2pdma.map) {
> > +	case PCI_P2PDMA_MAP_BUS_ADDR:
> > +		iod->flags |= IOD_DATA_P2P;
> > +		break;
> > +	case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
> > +		iod->flags |= IOD_DATA_MMIO;
> > +		break;
> > +	default:
> > +		return BLK_STS_RESOURCE;
> > +	}
> 
> Why is it an error if the data isn't a P2PDMA type?

This is probably bug, I was supposed to handle PCI_P2PDMA_MAP_NONE case too.

Thanks