Re: [PATCH v6-S1 0/6] dpaa2: bus, DMA and mempool base fixes

[Stephen Hemminger <stephen@networkplumber.org>]

On Tue,  6 Oct 2026 20:37:43 +0530
Prashant Gupta <prashant.gupta_3@nxp.com> wrote:

> This series is the first of four that upstream the missing NXP dpaa2
> driver changes. It collects the foundational bus/fslmc, dma/dpaa2 and
> mempool/dpaa2 fixes that the later series build on:
> 
> - defer fslmc bus initialization to probe; reduce probe logging
>   and skip ignored devices during DPRC population,
> - use memcpy to fill the dpaa2 QDMA completion index ring,
> - release SG FLE on completion ring overflow in the dpaa2 QDMA
>   dequeue path,
> - validate the FLE pool IOVA mapping at vchan setup and free the pool
>   on failure,
> - resolve the mempool ops index locally in secondary processes by
>   scanning rte_mempool_ops_table, removing the need for IPC.

When running AI locally which also has context from previous reviews,
it found a couple of things. I goaded it to be more concise and
provide more concrete suggestions.

Review: [PATCH v6-S1 0/6] DPAA2 fixes

Applies to main, every commit builds with -Dwerror=true.
Patch 5 needs rework, patch 4 a small follow-up. 1-3 look good.


[PATCH 5/6] dma/dpaa2: validate FLE pool IOVA mapping at vchan setup

Error: vchan setup fails on 2M hugepages.

DPAA2_VADDR_TO_IOVA_AND_CHECK() on a whole chunk only succeeds if
the chunk fits in one fslmc dmaseg, and fslmc creates one dmaseg
per memseg (hugepage). The FLE pool is 8192 x 2312 bytes, about
19 MB, so on 2M pages the check always fails.

Also the reference offset comes from fle_pool->mz (the mempool
header), not from the object memory.

Take the offset from the first chunk and compare the rest:

struct dpaa2_qdma_fle_pool_check {
	uint64_t iova2va_offset;
	bool bad;
};

static void
dpaa2_qdma_fle_pool_iova_check(struct rte_mempool *mp __rte_unused,
	void *opaque, struct rte_mempool_memhdr *memhdr,
	unsigned int mem_idx)
{
	struct dpaa2_qdma_fle_pool_check *check = opaque;
	uint64_t offset;

	if (memhdr->iova == RTE_BAD_IOVA) {
		check->bad = true;
		return;
	}

	offset = (uint64_t)memhdr->addr - memhdr->iova;
	if (mem_idx == 0)
		check->iova2va_offset = offset;
	else if (offset != check->iova2va_offset)
		check->bad = true;
}

and in dpaa2_qdma_vchan_setup() drop the mz based iova/va:

	rte_mempool_mem_iter(qdma_dev->vqs[vchan].fle_pool,
		dpaa2_qdma_fle_pool_iova_check, &fle_check);
	if (fle_check.bad) {
		DPAA2_QDMA_ERR("%s spans inconsistent IOVA offsets",
			       pool_name);
		ret = -EINVAL;
		goto err_pool;
	}
	qdma_dev->vqs[vchan].fle_iova2va_offset =
		fle_check.iova2va_offset;

Warning: the later error paths (both rte_mempool_get_bulk() calls,
ring_cntx_idx alloc) still return with fle_pool allocated, so the
retry case in the commit message is still broken. Send all of them
to one label:

	return 0;

err_pool:
	rte_mempool_free(qdma_dev->vqs[vchan].fle_pool);
	qdma_dev->vqs[vchan].fle_pool = NULL;
	return ret;
}


[PATCH 4/6] dma/dpaa2: release SG FLE on completion ring overflow

Warning: the LONG branch still records the FLE before the ring
insert. Make it match the SG branch:

		fle_sdd = (void *)(uintptr_t)DPAA2_GET_FD_FLC(fd);
		ret = qdma_cntx_idx_ring_eq(qdma_vq->ring_cntx_idx,
				&idx, 1, free_space);
		if (unlikely(ret != 1)) {
			rte_mempool_put(qdma_vq->fle_pool, fle_sdd);
			return -ENOSPC;
		}
		qdma_vq->fle_elem[*fle_elem_nb] = fle_sdd;
		(*fle_elem_nb)++;

Warning: drop "This matches how the caller treats a failed dq_fd()
as having consumed nothing." The FD is already pulled from QBMAN,
so on -ENOSPC its completions are lost, not retried.


[PATCH 6/6] mempool/dpaa2: look up ops index locally in secondary

Info: drop the unrelated blank line removal before
bp_list->next; this goes to stable.