Re: [PATCH rdma-next v11 4/6] RDMA/bnxt_re: Refactor bnxt_re_create_cq()

[Jason Gunthorpe <jgg@ziepe.ca>]

On Tue, Feb 10, 2026 at 10:29:37PM +0530, Sriharsha Basavapatna wrote:
> +static int bnxt_re_setup_sginfo(struct bnxt_re_dev *rdev,
> +				struct ib_umem *umem,
> +				struct bnxt_qplib_sg_info *sginfo)
> +{
> +	unsigned long page_size;
> +
> +	if (!umem)
> +		return -EINVAL;
> +
> +	page_size = ib_umem_find_best_pgsz(umem, SZ_4K, 0);
> +	if (!page_size || page_size != SZ_4K)
> +		return -EINVAL;
> +
> +	sginfo->umem = umem;
> +	sginfo->npages = ib_umem_num_dma_blocks(umem, page_size);

This ends up doing ib_umem_num_dma_blocks() twice:

bnxt_qplib_alloc_init_hwq()
 bnxt_qplib_create_cq()
  bnxt_re_create_cq()

And then again for a third time:

static int __alloc_pbl(struct bnxt_qplib_res *res,
		       struct bnxt_qplib_pbl *pbl,
		       struct bnxt_qplib_sg_info *sginfo)

	if (sginfo->umem)
		pages = ib_umem_num_dma_blocks(sginfo->umem, sginfo->pgsize);
	else
		pages = sginfo->npages;

etc. :\

It looks to me like npages is only expected to be used in kernel mode
where there is no umem.

So maybe don't add another num_dma_blocks here and make a note to go
and clean up this code properly as a followup? Properly means the
function that allocates the memory for the sginfo fills it in
completely instead of sprinkling it all over.

Jason