[PATCH v3 3/3] dma/ae4dma: add data path operations

[Raghavendra Ningoji]

On Fri, 27 Jun 2026 at 08:23, fengchengwen <fengchengwen at huawei.com> wrote:
>
> > This commit adds:
> >  - copy enqueue (rte_dma_copy): ...
> I don't think it's necessary to write in such detail because the ops
> implemented are defined by the framework. If needed, you can supplement
> by explaining what special features this driver has.

Condensed the commit log in v4. It now just lists the ops and explains
the driver-specific completion handling (the engine signals completion
by advancing read_idx rather than reliably writing a status word back).

> > +    cmd_q->next_write = (uint16_t)((write + 1) % nb);
> the next_write is [0, nb_desc-1], and it will as return value as copy,
> but the dmadev framework expect as [0, 0xFFFF], I doubt your drvier was
> not passed in any DMA test (e.g. dpdk-test, dpdk-dma-perf or examples/dma)
> ...
> > +        *last_idx = (uint16_t)((cmd_q->next_read - 1 + nb) % nb);
> the last_idx should be in range of [0, 0xFFFF]

Rather than having two counters,framework counter and h/w slot count in dricer, we had verified 
internally by wrapping counters of dma-test apps to h/w slots(32).
But again as you mentioned everything has to go with framework.v4 reworks the
indexing so next_write/next_read are free-running 16-bit counters in
[0, 0xFFFF]:
 - rte_dma_copy() returns the pre-increment free-running ring_idx.
 - completed()/completed_status() always set last_idx to the last
   completed ring_idx (next_read - 1), even when no new op is reaped.
 - the HW ring slot is derived as (hw_base + idx) & (nb_desc - 1), and
   in-flight as (next_write - next_read), which is why the depth is kept
   a power of two.
With this plus RTE_DMA_CAPA_OPS_COPY (patch 2/4), the driver follows
the framework ring-index contract.

> > +    case AE4DMA_DMA_ERR_INV_ALIGN:
> > +        /* Name matches DPDK public enum spelling. */
> > +        return RTE_DMA_STATUS_DATA_POISION;
> Suggest add RTE_DMA_STATUS_INVALID_ALIGN enum in rte_dmadev.h

Agreed that reusing DATA_POISION was wrong. For v4 the alignment error
maps to RTE_DMA_STATUS_ERROR_UNKNOWN so we no longer misuse an
unrelated code. Adding a dedicated RTE_DMA_STATUS_INVALID_ALIGNMENT to
the public rte_dmadev.h touches the dmadev API, so I would prefer to do
that as a separate follow-up patch with its own justification - happy
to send it if you agree.

> > +    if (nb < 2 || !rte_is_power_of_2(nb))
> > +        return 0;
> No need to check this

Removed; vchan_setup already guarantees a valid power-of-two depth.
burst_capacity now just computes (nb - 1) - in_flight.

Thanks,
Raghavendra