* [dpdk-stable] [PATCH 01/12] net/bnxt: fix burst mode get for Arm [not found] <20200909155302.28656-1-lance.richardson@broadcom.com> @ 2020-09-09 15:52 ` Lance Richardson 2020-09-09 15:52 ` [dpdk-stable] [PATCH 02/12] net/bnxt: fix rxq/txq get information Lance Richardson 1 sibling, 0 replies; 5+ messages in thread From: Lance Richardson @ 2020-09-09 15:52 UTC (permalink / raw) To: Ajit Khaparde, Somnath Kotur, Ruifeng Wang; +Cc: dev, stable Transmit and receive burst mode get operations incorrectly return "Vector SSE" on ARM64 platforms, change to return "Vector Neon" instead. Fixes: 3983583414 ("net/bnxt: support NEON") Reviewed-by: Ajit Kumar Khaparde <ajit.khaparde@broadcom.com> Signed-off-by: Lance Richardson <lance.richardson@broadcom.com> Cc: stable@dpdk.org --- drivers/net/bnxt/bnxt_ethdev.c | 60 +++++++++++++++++++++------------- 1 file changed, 38 insertions(+), 22 deletions(-) diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c index 75d055be00..7a77922c0c 100644 --- a/drivers/net/bnxt/bnxt_ethdev.c +++ b/drivers/net/bnxt/bnxt_ethdev.c @@ -2615,46 +2615,62 @@ bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id, qinfo->conf.tx_deferred_start = txq->tx_deferred_start; } +static const struct { + eth_rx_burst_t pkt_burst; + const char *info; +} bnxt_rx_burst_info[] = { + {bnxt_recv_pkts, "Scalar"}, +#if defined(RTE_ARCH_X86) + {bnxt_recv_pkts_vec, "Vector SSE"}, +#elif defined(RTE_ARCH_ARM64) + {bnxt_recv_pkts_vec, "Vector Neon"}, +#endif +}; + static int bnxt_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id, struct rte_eth_burst_mode *mode) { eth_rx_burst_t pkt_burst = dev->rx_pkt_burst; + size_t i; - if (pkt_burst == bnxt_recv_pkts) { - snprintf(mode->info, sizeof(mode->info), "%s", - "Scalar"); - return 0; - } -#if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64) - if (pkt_burst == bnxt_recv_pkts_vec) { - snprintf(mode->info, sizeof(mode->info), "%s", - "Vector SSE"); - return 0; + for (i = 0; i < RTE_DIM(bnxt_rx_burst_info); i++) { + if (pkt_burst == bnxt_rx_burst_info[i].pkt_burst) { + snprintf(mode->info, sizeof(mode->info), "%s", + bnxt_rx_burst_info[i].info); + return 0; + } } -#endif return -EINVAL; } +static const struct { + eth_tx_burst_t pkt_burst; + const char *info; +} bnxt_tx_burst_info[] = { + {bnxt_xmit_pkts, "Scalar"}, +#if defined(RTE_ARCH_X86) + {bnxt_xmit_pkts_vec, "Vector SSE"}, +#elif defined(RTE_ARCH_ARM64) + {bnxt_xmit_pkts_vec, "Vector Neon"}, +#endif +}; + static int bnxt_tx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id, struct rte_eth_burst_mode *mode) { eth_tx_burst_t pkt_burst = dev->tx_pkt_burst; + size_t i; - if (pkt_burst == bnxt_xmit_pkts) { - snprintf(mode->info, sizeof(mode->info), "%s", - "Scalar"); - return 0; - } -#if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64) - if (pkt_burst == bnxt_xmit_pkts_vec) { - snprintf(mode->info, sizeof(mode->info), "%s", - "Vector SSE"); - return 0; + for (i = 0; i < RTE_DIM(bnxt_tx_burst_info); i++) { + if (pkt_burst == bnxt_tx_burst_info[i].pkt_burst) { + snprintf(mode->info, sizeof(mode->info), "%s", + bnxt_tx_burst_info[i].info); + return 0; + } } -#endif return -EINVAL; } -- 2.25.1 ^ permalink raw reply [flat|nested] 5+ messages in thread
* [dpdk-stable] [PATCH 02/12] net/bnxt: fix rxq/txq get information [not found] <20200909155302.28656-1-lance.richardson@broadcom.com> 2020-09-09 15:52 ` [dpdk-stable] [PATCH 01/12] net/bnxt: fix burst mode get for Arm Lance Richardson @ 2020-09-09 15:52 ` Lance Richardson 2020-09-11 14:41 ` [dpdk-stable] [dpdk-dev] " Ferruh Yigit 1 sibling, 1 reply; 5+ messages in thread From: Lance Richardson @ 2020-09-09 15:52 UTC (permalink / raw) To: Ajit Khaparde, Somnath Kotur; +Cc: dev, stable Return correct values for Rx/Tx offloads and for rx_drop_en. Fixes: 2fc201884be8 ("net/bnxt: support rxq/txq get information") Reviewed-by: Ajit Kumar Khaparde <ajit.khaparde@broadcom.com> Signed-off-by: Lance Richardson <lance.richardson@broadcom.com> Cc: stable@dpdk.org --- drivers/net/bnxt/bnxt_ethdev.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c index 7a77922c0c..5585f872d0 100644 --- a/drivers/net/bnxt/bnxt_ethdev.c +++ b/drivers/net/bnxt/bnxt_ethdev.c @@ -2588,8 +2588,9 @@ bnxt_rxq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id, qinfo->nb_desc = rxq->nb_rx_desc; qinfo->conf.rx_free_thresh = rxq->rx_free_thresh; - qinfo->conf.rx_drop_en = 0; + qinfo->conf.rx_drop_en = 1; qinfo->conf.rx_deferred_start = rxq->rx_deferred_start; + qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads; } static void @@ -2613,6 +2614,7 @@ bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id, qinfo->conf.tx_free_thresh = txq->tx_free_thresh; qinfo->conf.tx_rs_thresh = 0; qinfo->conf.tx_deferred_start = txq->tx_deferred_start; + qinfo->conf.offloads = dev->data->dev_conf.txmode.offloads; } static const struct { -- 2.25.1 ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [dpdk-stable] [dpdk-dev] [PATCH 02/12] net/bnxt: fix rxq/txq get information 2020-09-09 15:52 ` [dpdk-stable] [PATCH 02/12] net/bnxt: fix rxq/txq get information Lance Richardson @ 2020-09-11 14:41 ` Ferruh Yigit 2020-09-18 18:41 ` Lance Richardson 0 siblings, 1 reply; 5+ messages in thread From: Ferruh Yigit @ 2020-09-11 14:41 UTC (permalink / raw) To: Lance Richardson, Ajit Khaparde, Somnath Kotur; +Cc: dev, stable On 9/9/2020 4:52 PM, Lance Richardson wrote: > Return correct values for Rx/Tx offloads and for rx_drop_en. > > Fixes: 2fc201884be8 ("net/bnxt: support rxq/txq get information") > Reviewed-by: Ajit Kumar Khaparde <ajit.khaparde@broadcom.com> > Signed-off-by: Lance Richardson <lance.richardson@broadcom.com> > Cc: stable@dpdk.org > --- > drivers/net/bnxt/bnxt_ethdev.c | 4 +++- > 1 file changed, 3 insertions(+), 1 deletion(-) > > diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c > index 7a77922c0c..5585f872d0 100644 > --- a/drivers/net/bnxt/bnxt_ethdev.c > +++ b/drivers/net/bnxt/bnxt_ethdev.c > @@ -2588,8 +2588,9 @@ bnxt_rxq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id, > qinfo->nb_desc = rxq->nb_rx_desc; > > qinfo->conf.rx_free_thresh = rxq->rx_free_thresh; > - qinfo->conf.rx_drop_en = 0; > + qinfo->conf.rx_drop_en = 1; Why 0 is wrong but 1 is correct? Technically 'rx_drop_en' is a user configuration, which is set via 'rte_eth_rx_queue_setup()' API. bnxt seems not honoring this config option at all. Based on HW capability, I think two things can be done, 1) Configure the HW based on config request, and return configured value in 'bnxt_rxq_info_get_op()'. see 'ixgbe'. 2) If HW is not configurable, check the value in 'rte_eth_rx_queue_setup()' a) return error if unsupported value requested. see 'sfc'. b) log a warning and overwrite the requested config with whatever supported. And for both a & b, return current config in the 'bnxt_rxq_info_get_op()' > qinfo->conf.rx_deferred_start = rxq->rx_deferred_start; > + qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads; This is for queue specific offloads, you are returning port offloads. As far as I can see bnxt doesn't have any queue specific offload, so this can be dropped. ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [dpdk-stable] [dpdk-dev] [PATCH 02/12] net/bnxt: fix rxq/txq get information 2020-09-11 14:41 ` [dpdk-stable] [dpdk-dev] " Ferruh Yigit @ 2020-09-18 18:41 ` Lance Richardson 2020-09-21 11:05 ` Ferruh Yigit 0 siblings, 1 reply; 5+ messages in thread From: Lance Richardson @ 2020-09-18 18:41 UTC (permalink / raw) To: Ferruh Yigit; +Cc: Ajit Khaparde, Somnath Kotur, dev, stable On Fri, Sep 11, 2020 at 10:41 AM Ferruh Yigit <ferruh.yigit@intel.com> wrote: > > > - qinfo->conf.rx_drop_en = 0; > > + qinfo->conf.rx_drop_en = 1; > > Why 0 is wrong but 1 is correct? > > Technically 'rx_drop_en' is a user configuration, which is set via > 'rte_eth_rx_queue_setup()' API. > > bnxt seems not honoring this config option at all. > > Based on HW capability, I think two things can be done, > 1) Configure the HW based on config request, and return configured value in > 'bnxt_rxq_info_get_op()'. see 'ixgbe'. > > 2) If HW is not configurable, check the value in 'rte_eth_rx_queue_setup()' > a) return error if unsupported value requested. see 'sfc'. > b) log a warning and overwrite the requested config with whatever supported. > And for both a & b, return current config in the 'bnxt_rxq_info_get_op()' > > > qinfo->conf.rx_deferred_start = rxq->rx_deferred_start; > > + qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads; > Hi Ferruh, Apologies, this somehow didn't make it to my inbox. I believe case (2) applies here, rx_drop_en is not currently configurable in hw, so this change was intended to accurately report the effective value. I'm not sure whether (2a) or (2b) would be better, but (2b) seems less likely to cause issues for existing applications. > > This is for queue specific offloads, you are returning port offloads. > As far as I can see bnxt doesn't have any queue specific offload, so this can be > dropped. It wasn't clear to me whether this was intended to report the difference between the offload configuration for the queue and the offload configuration for the port or the effective offload configuration for the queue. I noticed that several other PMDs (e.g. mlx5, netvsc, sfc) report the offload configuration for the port in rx/tx_queue_info_get(). The sfc PMD reports the offload configuration for the port combined with queue-specific offloads, based on those examples this seemed to be correct. I guess you're saying those are also incorrect? Thanks, Lance ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [dpdk-stable] [dpdk-dev] [PATCH 02/12] net/bnxt: fix rxq/txq get information 2020-09-18 18:41 ` Lance Richardson @ 2020-09-21 11:05 ` Ferruh Yigit 0 siblings, 0 replies; 5+ messages in thread From: Ferruh Yigit @ 2020-09-21 11:05 UTC (permalink / raw) To: Lance Richardson; +Cc: Ajit Khaparde, Somnath Kotur, dev, stable On 9/18/2020 7:41 PM, Lance Richardson wrote: > On Fri, Sep 11, 2020 at 10:41 AM Ferruh Yigit <ferruh.yigit@intel.com> wrote: >> >>> - qinfo->conf.rx_drop_en = 0; >>> + qinfo->conf.rx_drop_en = 1; >> >> Why 0 is wrong but 1 is correct? >> >> Technically 'rx_drop_en' is a user configuration, which is set via >> 'rte_eth_rx_queue_setup()' API. >> >> bnxt seems not honoring this config option at all. >> >> Based on HW capability, I think two things can be done, >> 1) Configure the HW based on config request, and return configured value in >> 'bnxt_rxq_info_get_op()'. see 'ixgbe'. >> >> 2) If HW is not configurable, check the value in 'rte_eth_rx_queue_setup()' >> a) return error if unsupported value requested. see 'sfc'. >> b) log a warning and overwrite the requested config with whatever supported. >> And for both a & b, return current config in the 'bnxt_rxq_info_get_op()' >> >>> qinfo->conf.rx_deferred_start = rxq->rx_deferred_start; >>> + qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads; >> > Hi Ferruh, > > Apologies, this somehow didn't make it to my inbox. > > I believe case (2) applies here, rx_drop_en is not currently configurable in hw, > so this change was intended to accurately report the effective value. I'm not > sure whether (2a) or (2b) would be better, but (2b) seems less likely to cause > issues for existing applications. > I guess (2a) may cause more issues with existing applications, applications previously running without problem may now start failing if PMD starts returning error. In (2b) execution will be same, and if PMD already doesn't support NOT dropping, functionality will be same, only configuration will reflect what is actually happening instead of what user thinks happening. > >> >> This is for queue specific offloads, you are returning port offloads. >> As far as I can see bnxt doesn't have any queue specific offload, so this can be >> dropped. > > It wasn't clear to me whether this was intended to report the difference between > the offload configuration for the queue and the offload configuration > for the port > or the effective offload configuration for the queue. I noticed that > several other > PMDs (e.g. mlx5, netvsc, sfc) report the offload configuration for the port in > rx/tx_queue_info_get(). The sfc PMD reports the offload configuration for the > port combined with queue-specific offloads, based on those examples this > seemed to be correct. I guess you're saying those are also incorrect? > May bad, they look OK, "the effective offload configuration for the queue" makes more sense here. I guess I confused same fields should hold the queue specific offloads when used for 'rte_eth_rx_queue_setup()' ^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2020-09-21 11:05 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
[not found] <20200909155302.28656-1-lance.richardson@broadcom.com>
2020-09-09 15:52 ` [dpdk-stable] [PATCH 01/12] net/bnxt: fix burst mode get for Arm Lance Richardson
2020-09-09 15:52 ` [dpdk-stable] [PATCH 02/12] net/bnxt: fix rxq/txq get information Lance Richardson
2020-09-11 14:41 ` [dpdk-stable] [dpdk-dev] " Ferruh Yigit
2020-09-18 18:41 ` Lance Richardson
2020-09-21 11:05 ` Ferruh Yigit
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).