On Tue, Sep 01, 2020 at 09:43:34PM -0300, Jason Gunthorpe wrote:
> diff --git a/include/rdma/ib_umem.h b/include/rdma/ib_umem.h
> index b880512ba95f16..ba3b9be0d8c56a 100644
> +++ b/include/rdma/ib_umem.h
> @@ -33,11 +33,17 @@ static inline int ib_umem_offset(struct ib_umem *umem)
> return umem->address & ~PAGE_MASK;
> }
>
> +static inline size_t ib_umem_num_dma_blocks(struct ib_umem *umem,
> + unsigned long pgsz)
> +{
> + return (ALIGN(umem->address + umem->length, pgsz) -
> + ALIGN_DOWN(umem->address, pgsz)) /
> + pgsz;
> +}
Like the other places, this is wrong as well. The IOVA needs to be
used here or it can be +-1 page away from what
rdma_umem_for_each_dma_block() returns.
However, it does work if pgsz is PAGE_SIZE, or for the very common
cases where umem->address == IOVA...
Which, I suppose, is why nobody noticed this until now, as I found
several drivers open coding the above.
Jason
