On Mon, Apr 28, 2025 at 08:22:25PM +0300, Leon Romanovsky wrote:
> On Mon, Apr 28, 2025 at 10:46:19AM -0600, Keith Busch wrote:
> > On Mon, Apr 28, 2025 at 12:22:29PM +0300, Leon Romanovsky wrote:
> > > + do {
> > > + if (WARN_ON_ONCE(mapped == entries)) {
> > > + iter.status = BLK_STS_IOERR;
> > > + break;
> > > + }
> > > + nvme_pci_sgl_set_data(&sg_list[mapped++], &iter);
> >
> > I think this should say "++mapped" so that the data blocks start at
> > index 1 (continued below...)
> >
> > > + iod->total_len += iter.len;
> > > + } while (blk_rq_dma_map_iter_next(req, dev->dev, &iod->dma_meta_state,
> > > + &iter));
> > >
> > > -out_unmap_sg:
> > > - dma_unmap_sgtable(dev->dev, &iod->meta_sgt, rq_dma_dir(req), 0);
> > > -out_free_sg:
> > > - mempool_free(iod->meta_sgt.sgl, dev->iod_meta_mempool);
> > > - return BLK_STS_RESOURCE;
> > > + nvme_pci_sgl_set_seg(sg_list, sgl_dma, mapped);
> >
> > because this here is setting sg_list index 0 to be the segment
> > descriptor.
> >
> > And you also need to increment sgl_dma to point to the element after
> > sg_list, otherwise it's pointing right back to itself, creating a looped
> > list.
>
> Thanks for pointing to the difference between data_map and metadata_map,
Yeah, They're different because the SQE has 16 bytes available for the
data map, so the command can fit an SGL descriptor directly in its
submission entry. The metadata field only has 8 bytes in the SQE, so we
have to set it to the address of an external SGL segment descriptor.
