[PATCH 1/2] net/cxgbe: fix Rx handling for packed responses

[Abdulrahman Alshawi]

 The Rx path assumes every SGE response starts a new Free List buffer:       BUG_ON(!(len & F_RSPD_NEWBUF));   That is not always true. On T5/T6, small packets can be packed into the same FL buffer. Only the first response for that buffer has NEWBUF set; later responses can refer to the same buffer with a different payload offset.   The current PMD consumes one FL buffer per response. When packed responses are delivered, the FL consumer state goes out of sync with the hardware and the affected ingress queue stops making progress. From user space this shows up as:     - rx_bgN_dropped_packets increasing at line rate   - q_ipackets for the affected queue staying at 0   - imissed increasing   - no recovery until the port is restarted   Fix this by tracking packed-buffer alignment and copying packet payload from the current FL buffer at q->offset, only freeing the FL buffer once it has actually been consumed.   This patch:   - adds sge::fl_align from the ingress pad/pack settings   - introduces cxgbe_copy_rx_pkt() to copy from packed FL buffers   - advances q->offset by len rounded up to fl_align   - frees an FL buffer only after full consumption   - drains the previous buffer when a later response arrives with NEWBUF   Reproduce (T62100-LP-CR, FW 2.1.19.0 / TP 0.1.23.2):     dpdk-testpmd -l 1-9 -a 0000:18:00.4 -- \              --rxq=32 --txq=32 --nb-cores=8 --forward-mode=rxonly -i     testpmd> port stop all   testpmd> flow flush 0   testpmd> flow create 0 ingress pattern eth \            / vlan tci spec 0 tci mask 0x0007 \            / end actions queue index 5 / end   testpmd> port start all   testpmd> start   Without this patch, rx_qN_packets stays at 0 while rx_bgN_dropped_packets rises at line rate. With the patch, rx_qN_packets tracks received traffic and rx_bgN_dropped_packets stays at 0.   Signed-off-by: Abdulrahman Alshawi <ashawi at wirefilter.com> ---  drivers/net/cxgbe/base/adapter.h |   1 +  drivers/net/cxgbe/sge.c          | 122 ++++++++++++++++++++++++-------  2 files changed, 98 insertions(+), 25 deletions(-)   diff --git a/drivers/net/cxgbe/base/adapter.h b/drivers/net/cxgbe/base/adapter.h index 207f3ecb88..e67cf22950 100644 --- a/drivers/net/cxgbe/base/adapter.h +++ b/drivers/net/cxgbe/base/adapter.h @@ -280,6 +280,7 @@ struct sge {     u16 max_ethqsets;           /* # of available Ethernet queue sets */     u32 stat_len;               /* length of status page at ring end */     u32 pktshift;               /* padding between CPL & packet data */ +   u32 fl_align;               /* packed Rx packet alignment */       /* response queue interrupt parameters */     u16 timer_val[SGE_NTIMERS]; diff --git a/drivers/net/cxgbe/sge.c b/drivers/net/cxgbe/sge.c index e9d45f24c4..7cf5c70775 100644 --- a/drivers/net/cxgbe/sge.c +++ b/drivers/net/cxgbe/sge.c @@ -1492,6 +1492,90 @@ static inline void cxgbe_fill_mbuf_info(struct adapter *adap,                    RTE_MBUF_F_RX_L4_CKSUM_BAD);  }   +static int cxgbe_copy_rx_pkt(struct sge_rspq *q, struct sge_eth_rxq *rxq, +               u32 len, struct rte_mbuf **out) +{ +   struct sge *s = &q->adapter->sge; +   struct rte_mbuf *pkt; +   char *dst; +   u32 copied = 0; +   u32 remaining = len; + +   pkt = rte_pktmbuf_alloc(q->mb_pool); +   if (unlikely(!pkt)) { +       rxq->rspq.eth_dev->data->rx_mbuf_alloc_failed++; +       rxq->stats.rx_drops++; +       return -ENOMEM; +   } + +   dst = rte_pktmbuf_append(pkt, len); +   if (unlikely(!dst)) { +       rte_pktmbuf_free(pkt); +       rxq->stats.rx_drops++; +       return -ENOMEM; +   } + +   while (remaining) { +       const struct rx_sw_desc *rsd = &rxq->fl.sdesc[rxq->fl.cidx]; +       struct rte_mbuf *src = rsd->buf; +       u32 bufsz = get_buf_size(q->adapter, rsd); +       u32 copy_len; + +       if (unlikely(!src || q->offset < 0 || +               (u32)q->offset >= bufsz)) { +           rte_pktmbuf_free(pkt); +           rxq->stats.rx_drops++; +           return -EINVAL; +       } + +       copy_len = RTE_MIN(bufsz - (u32)q->offset, remaining); +       rte_memcpy(dst + copied, +             rte_pktmbuf_mtod_offset(src, const void *, q->offset), +             copy_len); + +       copied += copy_len; +       remaining -= copy_len; +       q->offset += copy_len; + +       if (remaining) { +           free_rx_bufs(&rxq->fl, 1); +           q->offset = 0; +           continue; +       } + +       q->offset = RTE_ALIGN_CEIL(q->offset, s->fl_align); +       if ((u32)q->offset >= bufsz) { +           free_rx_bufs(&rxq->fl, 1); +           q->offset = 0; +       } +   } + +   *out = pkt; +   return 0; +} + +static unsigned int cxgbe_fl_pkt_align(struct adapter *adap) +{ +   u32 sge_control = t4_read_reg(adap, A_SGE_CONTROL); +   unsigned int ingpad_shift, ingpad, fl_align; + +   ingpad_shift = CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5 ? +             X_INGPADBOUNDARY_SHIFT : X_T6_INGPADBOUNDARY_SHIFT; +   ingpad = 1U << (G_INGPADBOUNDARY(sge_control) + ingpad_shift); +   fl_align = ingpad; + +   if (!is_t4(adap->params.chip)) { +       u32 sge_control2 = t4_read_reg(adap, A_SGE_CONTROL2); +       unsigned int ingpack = G_INGPACKBOUNDARY(sge_control2); + +       ingpack = ingpack == X_INGPACKBOUNDARY_16B ? +            16 : 1U << (ingpack + X_INGPACKBOUNDARY_SHIFT); +       fl_align = RTE_MAX(ingpad, ingpack); +   } + +   return fl_align ? fl_align : RTE_CACHE_LINE_SIZE; +} +  /**   * process_responses - process responses from an SGE response queue   * @q: the ingress queue to process @@ -1535,14 +1619,12 @@ static int process_responses(struct sge_rspq *q, int budget,             stat_pidx = ntohs(q->stat->pidx);             stat_pidx_diff = P_IDXDIFF(q, stat_pidx);             while (stat_pidx_diff && budget_left) { -               const struct rx_sw_desc *rsd = -                   &rxq->fl.sdesc[rxq->fl.cidx];                 const struct rss_header *rss_hdr =                     (const void *)q->cur_desc;                 const struct cpl_rx_pkt *cpl =                     (const void *)&q->cur_desc[1]; -               struct rte_mbuf *pkt, *npkt; -               u32 len, bufsz; +               struct rte_mbuf *pkt; +               u32 len;                   rc = (const struct rsp_ctrl *)                     ((const char *)q->cur_desc + @@ -1553,28 +1635,16 @@ static int process_responses(struct sge_rspq *q, int budget,                     break;                   len = ntohl(rc->pldbuflen_qid); -               BUG_ON(!(len & F_RSPD_NEWBUF)); -               pkt = rsd->buf; -               npkt = pkt; -               len = G_RSPD_LEN(len); -               pkt->pkt_len = len; - -               /* Chain mbufs into len if necessary */ -               while (len) { -                   struct rte_mbuf *new_pkt = rsd->buf; - -                   bufsz = min(get_buf_size(q->adapter, -                               rsd), len); -                   new_pkt->data_len = bufsz; -                   unmap_rx_buf(&rxq->fl); -                   len -= bufsz; -                   npkt->next = new_pkt; -                   npkt = new_pkt; -                   pkt->nb_segs++; -                   rsd = &rxq->fl.sdesc[rxq->fl.cidx]; +               if (len & F_RSPD_NEWBUF) { +                   if (q->offset > 0) { +                       free_rx_bufs(&rxq->fl, 1); +                       q->offset = 0; +                   }                 } -               npkt->next = NULL; -               pkt->nb_segs--; +               len = G_RSPD_LEN(len); +               ret = cxgbe_copy_rx_pkt(q, rxq, len, &pkt); +               if (unlikely(ret)) +                   break;                   cxgbe_fill_mbuf_info(q->adapter, cpl, pkt);   @@ -2379,6 +2449,7 @@ int t4_sge_init(struct adapter *adap)     sge_control = t4_read_reg(adap, A_SGE_CONTROL);     s->pktshift = G_PKTSHIFT(sge_control);     s->stat_len = (sge_control & F_EGRSTATUSPAGESIZE) ? 128 : 64; +   s->fl_align = cxgbe_fl_pkt_align(adap);     ret = t4_sge_init_soft(adap);     if (ret < 0) {         dev_err(adap, "%s: t4_sge_init_soft failed, error %d\n", @@ -2516,6 +2587,7 @@ int t4vf_sge_init(struct adapter *adap)     s->stat_len = ((sge_control & F_EGRSTATUSPAGESIZE)             ? 128 : 64);     s->pktshift = G_PKTSHIFT(sge_control); +   s->fl_align = RTE_CACHE_LINE_SIZE;       /*     * A FL with <= fl_starve_thres buffers is starving and a periodic --  2.39.5 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/stable/attachments/20260427/546c2605/attachment.htm>