On 7/10/2026 3:22 PM, Raghavendra Ningoji wrote: > On Mon, 7 Jul 2026 at 06:49, fengchengwen <[email protected]> wrote: >> >>> + if (last_idx != NULL) >>> + *last_idx = (uint16_t)(cmd_q->next_read - 1); >> >> 1\ the last_idx always non-NULL for driver, so no need for >> 'if (last_idx != NULL)' >> 2\ last_idx should be the last success completed index > > Both fixed in v5. The NULL check is dropped, and last_idx now reports > the ring_idx of the last *successfully* completed op. > >> Consider four reqests, the hardware mark each request as following: >> req1 success >> req2 err >> req3 success >> req4 err >> the cpl_count will be 4, and err_count will be 2 >> It will return 2 in current impl, and last_idx will be the req4's >> But in this function, it should return 1, and last_idx should be req1's > > Agreed - thanks for the clear example. rte_dma_completed() is reworked > in v5 to stop at the first failed descriptor: it returns only the run of > successful ops preceding the error (1 in your example), sets *has_error, > and leaves the failed op in place with last_idx pointing at the last > success (req1). The failed op and everything after it are then drained > and reported by rte_dma_completed_status(). > > On a related note, while implementing the above I found the last_idx > documentation for rte_dma_completed() and rte_dma_completed_status() > to be identical: > > @param[out] last_idx > The last completed operation's ring_idx. > > Since rte_dma_completed() only reports operations that completed > *successfully* (it stops at the first error), its last_idx is really the > last successfully completed operation's ring_idx, whereas for > rte_dma_completed_status() it is the last completed operation regardless > of status. Would a small doc clarification along these lines be welcome > (as a separate patch)?
I think it is necessary to update. Please also update rte_dma_completed_status, both in a separate patch Thanks > > rte_dma_completed(): > @param[out] last_idx > The last successfully completed operation's ring_idx. > > I'm happy to send that as a separate dmadev doc patch if you agree. > > Thanks, > Raghavendra
