| Message ID | 20240628133517.8591-7-gakula@marvell.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | Introduce RVU representors | expand |
On Fri, 28 Jun 2024 19:05:13 +0530 Geetha sowjanya wrote: > Adds support to export VF port statistics via representor > netdev. Defines new mbox "NIX_LF_STATS" to fetch VF hw stats. These count all traffic passing to the VF? Both from the representor and directly via forwarding rules?
>-----Original Message----- >From: Jakub Kicinski <kuba@kernel.org> >Sent: Tuesday, July 2, 2024 8:44 AM >To: Geethasowjanya Akula <gakula@marvell.com> >Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org; >davem@davemloft.net; pabeni@redhat.com; edumazet@google.com; Sunil >Kovvuri Goutham <sgoutham@marvell.com>; Subbaraya Sundeep Bhatta ><sbhatta@marvell.com>; Hariprasad Kelam <hkelam@marvell.com> >Subject: [EXTERNAL] Re: [net-next PATCH v7 06/10] octeontx2-pf: Get VF stats >via representor >On Fri, 28 Jun 2024 19:05:13 +0530 Geetha sowjanya wrote: >> Adds support to export VF port statistics via representor netdev. >> Defines new mbox "NIX_LF_STATS" to fetch VF hw stats. > >These count all traffic passing to the VF? Both from the representor and directly >via forwarding rules? Yes, it provide both the stats.
On Tue, 2 Jul 2024 06:41:51 +0000 Geethasowjanya Akula wrote: > >> Adds support to export VF port statistics via representor netdev. > >> Defines new mbox "NIX_LF_STATS" to fetch VF hw stats. > > > >These count all traffic passing to the VF? Both from the representor and directly > >via forwarding rules? > Yes, it provide both the stats. Could you implement IFLA_OFFLOAD_XSTATS_CPU_HIT as well, to indicate how much of the traffic wasn't offloaded?
>-----Original Message----- >From: Jakub Kicinski <kuba@kernel.org> >Sent: Tuesday, July 2, 2024 7:27 PM >To: Geethasowjanya Akula <gakula@marvell.com> >Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org; >davem@davemloft.net; pabeni@redhat.com; edumazet@google.com; Sunil >Kovvuri Goutham <sgoutham@marvell.com>; Subbaraya Sundeep Bhatta ><sbhatta@marvell.com>; Hariprasad Kelam <hkelam@marvell.com> >Subject: Re: [EXTERNAL] Re: [net-next PATCH v7 06/10] octeontx2-pf: Get VF >stats via representor >On Tue, 2 Jul 2024 06:41:51 +0000 Geethasowjanya Akula wrote: >> >> Adds support to export VF port statistics via representor netdev. >> >> Defines new mbox "NIX_LF_STATS" to fetch VF hw stats. >> > >> >These count all traffic passing to the VF? Both from the representor >> >and directly via forwarding rules? >> Yes, it provide both the stats. > >Could you implement IFLA_OFFLOAD_XSTATS_CPU_HIT as well, to indicate >how much of the traffic wasn't offloaded? Will implement while adding TC hw offload support for the representor, which will be submitted as different patch series.
On Wed, 3 Jul 2024 09:08:13 +0000 Geethasowjanya Akula wrote: >> Could you implement IFLA_OFFLOAD_XSTATS_CPU_HIT as well, to indicate >> how much of the traffic wasn't offloaded? > > Will implement while adding TC hw offload support for the representor, > which will be submitted as different patch series. I don't see why we need to wait. Offload is the opposite of the stats we're talking about here.
>-----Original Message----- >From: Jakub Kicinski <kuba@kernel.org> >Sent: Thursday, July 4, 2024 12:03 AM >To: Geethasowjanya Akula <gakula@marvell.com> >Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org; >davem@davemloft.net; pabeni@redhat.com; edumazet@google.com; Sunil >Kovvuri Goutham <sgoutham@marvell.com>; Subbaraya Sundeep Bhatta ><sbhatta@marvell.com>; Hariprasad Kelam <hkelam@marvell.com> >Subject: Re: [EXTERNAL] Re: [net-next PATCH v7 06/10] octeontx2-pf: Get VF >stats via representor > >On Wed, 3 Jul 2024 09: 08: 13 +0000 Geethasowjanya Akula wrote: >> Could >you implement IFLA_OFFLOAD_XSTATS_CPU_HIT as well, to indicate >> how >much of the traffic wasn't offloaded? > > Will implement while adding TC hw >offload >On Wed, 3 Jul 2024 09:08:13 +0000 Geethasowjanya Akula wrote: >>> Could you implement IFLA_OFFLOAD_XSTATS_CPU_HIT as well, to indicate >>> how much of the traffic wasn't offloaded? >> >> Will implement while adding TC hw offload support for the representor, >> which will be submitted as different patch series. > >I don't see why we need to wait. Offload is the opposite of the stats we're >talking about here. Ok, will implement in next version.
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/mbox.h b/drivers/net/ethernet/marvell/octeontx2/af/mbox.h index a7c32f1cc924..d293a3c35b6b 100644 --- a/drivers/net/ethernet/marvell/octeontx2/af/mbox.h +++ b/drivers/net/ethernet/marvell/octeontx2/af/mbox.h @@ -321,6 +321,7 @@ M(NIX_MCAST_GRP_DESTROY, 0x802c, nix_mcast_grp_destroy, nix_mcast_grp_destroy_re M(NIX_MCAST_GRP_UPDATE, 0x802d, nix_mcast_grp_update, \ nix_mcast_grp_update_req, \ nix_mcast_grp_update_rsp) \ +M(NIX_LF_STATS, 0x802e, nix_lf_stats, nix_stats_req, nix_stats_rsp) \ /* MCS mbox IDs (range 0xA000 - 0xBFFF) */ \ M(MCS_ALLOC_RESOURCES, 0xa000, mcs_alloc_resources, mcs_alloc_rsrc_req, \ mcs_alloc_rsrc_rsp) \ @@ -1366,6 +1367,37 @@ struct nix_bandprof_get_hwinfo_rsp { u32 policer_timeunit; }; +struct nix_stats_req { + struct mbox_msghdr hdr; + u8 reset; + u16 pcifunc; + u64 rsvd; +}; + +struct nix_stats_rsp { + struct mbox_msghdr hdr; + u16 pcifunc; + struct { + u64 octs; + u64 ucast; + u64 bcast; + u64 mcast; + u64 drop; + u64 drop_octs; + u64 drop_mcast; + u64 drop_bcast; + u64 err; + u64 rsvd[5]; + } rx; + struct { + u64 ucast; + u64 bcast; + u64 mcast; + u64 drop; + u64 octs; + } tx; +}; + /* NPC mbox message structs */ #define NPC_MCAM_ENTRY_INVALID 0xFFFF diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu_rep.c b/drivers/net/ethernet/marvell/octeontx2/af/rvu_rep.c index 5f2e2cbd165a..6bc867dea42e 100644 --- a/drivers/net/ethernet/marvell/octeontx2/af/rvu_rep.c +++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu_rep.c @@ -13,6 +13,49 @@ #include "rvu.h" #include "rvu_reg.h" +#define RVU_LF_RX_STATS(reg) \ + rvu_read64(rvu, blkaddr, NIX_AF_LFX_RX_STATX(nixlf, reg)) + +#define RVU_LF_TX_STATS(reg) \ + rvu_read64(rvu, blkaddr, NIX_AF_LFX_TX_STATX(nixlf, reg)) + +int rvu_mbox_handler_nix_lf_stats(struct rvu *rvu, + struct nix_stats_req *req, + struct nix_stats_rsp *rsp) +{ + u16 pcifunc = req->pcifunc; + int nixlf, blkaddr, err; + struct msg_req rst_req; + struct msg_rsp rst_rsp; + + err = nix_get_nixlf(rvu, pcifunc, &nixlf, &blkaddr); + if (err) + return 0; + + if (req->reset) { + rst_req.hdr.pcifunc = pcifunc; + return rvu_mbox_handler_nix_stats_rst(rvu, &rst_req, &rst_rsp); + } + rsp->rx.octs = RVU_LF_RX_STATS(RX_OCTS); + rsp->rx.ucast = RVU_LF_RX_STATS(RX_UCAST); + rsp->rx.bcast = RVU_LF_RX_STATS(RX_BCAST); + rsp->rx.mcast = RVU_LF_RX_STATS(RX_MCAST); + rsp->rx.drop = RVU_LF_RX_STATS(RX_DROP); + rsp->rx.err = RVU_LF_RX_STATS(RX_ERR); + rsp->rx.drop_octs = RVU_LF_RX_STATS(RX_DROP_OCTS); + rsp->rx.drop_mcast = RVU_LF_RX_STATS(RX_DRP_MCAST); + rsp->rx.drop_bcast = RVU_LF_RX_STATS(RX_DRP_BCAST); + + rsp->tx.octs = RVU_LF_TX_STATS(TX_OCTS); + rsp->tx.ucast = RVU_LF_TX_STATS(TX_UCAST); + rsp->tx.bcast = RVU_LF_TX_STATS(TX_BCAST); + rsp->tx.mcast = RVU_LF_TX_STATS(TX_MCAST); + rsp->tx.drop = RVU_LF_TX_STATS(TX_DROP); + + rsp->pcifunc = req->pcifunc; + return 0; +} + static u16 rvu_rep_get_vlan_id(struct rvu *rvu, u16 pcifunc) { int id; diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/rep.c b/drivers/net/ethernet/marvell/octeontx2/nic/rep.c index b993b03622dd..d4e90b13db71 100644 --- a/drivers/net/ethernet/marvell/octeontx2/nic/rep.c +++ b/drivers/net/ethernet/marvell/octeontx2/nic/rep.c @@ -28,6 +28,68 @@ MODULE_DESCRIPTION(DRV_STRING); MODULE_LICENSE("GPL"); MODULE_DEVICE_TABLE(pci, rvu_rep_id_table); +static void rvu_rep_get_stats(struct work_struct *work) +{ + struct delayed_work *del_work = to_delayed_work(work); + struct nix_stats_req *req; + struct nix_stats_rsp *rsp; + struct rep_stats *stats; + struct otx2_nic *priv; + struct rep_dev *rep; + int err; + + rep = container_of(del_work, struct rep_dev, stats_wrk); + priv = rep->mdev; + + mutex_lock(&priv->mbox.lock); + req = otx2_mbox_alloc_msg_nix_lf_stats(&priv->mbox); + if (!req) { + mutex_unlock(&priv->mbox.lock); + return; + } + req->pcifunc = rep->pcifunc; + err = otx2_sync_mbox_msg_busy_poll(&priv->mbox); + if (err) + goto exit; + + rsp = (struct nix_stats_rsp *) + otx2_mbox_get_rsp(&priv->mbox.mbox, 0, &req->hdr); + + if (IS_ERR(rsp)) { + err = PTR_ERR(rsp); + goto exit; + } + + stats = &rep->stats; + stats->rx_bytes = rsp->rx.octs; + stats->rx_frames = rsp->rx.ucast + rsp->rx.bcast + + rsp->rx.mcast; + stats->rx_drops = rsp->rx.drop; + stats->rx_mcast_frames = rsp->rx.mcast; + stats->tx_bytes = rsp->tx.octs; + stats->tx_frames = rsp->tx.ucast + rsp->tx.bcast + rsp->tx.mcast; + stats->tx_drops = rsp->tx.drop; +exit: + mutex_unlock(&priv->mbox.lock); +} + +static void rvu_rep_get_stats64(struct net_device *dev, + struct rtnl_link_stats64 *stats) +{ + struct rep_dev *rep = netdev_priv(dev); + + stats->rx_packets = rep->stats.rx_frames; + stats->rx_bytes = rep->stats.rx_bytes; + stats->rx_dropped = rep->stats.rx_drops; + stats->multicast = rep->stats.rx_mcast_frames; + + stats->tx_packets = rep->stats.tx_frames; + stats->tx_bytes = rep->stats.tx_bytes; + stats->tx_dropped = rep->stats.tx_drops; + + schedule_delayed_work(&rep->stats_wrk, msecs_to_jiffies(100)); +} + static int rvu_eswitch_config(struct otx2_nic *priv, u8 ena) { struct esw_cfg_req *req; @@ -87,6 +149,7 @@ static const struct net_device_ops rvu_rep_netdev_ops = { .ndo_open = rvu_rep_open, .ndo_stop = rvu_rep_stop, .ndo_start_xmit = rvu_rep_xmit, + .ndo_get_stats64 = rvu_rep_get_stats64, }; static int rvu_rep_napi_init(struct otx2_nic *priv, @@ -233,6 +296,8 @@ int rvu_rep_create(struct otx2_nic *priv, struct netlink_ext_ack *extack) free_netdev(ndev); goto exit; } + + INIT_DELAYED_WORK(&rep->stats_wrk, rvu_rep_get_stats); } err = rvu_rep_napi_init(priv, extack); if (err) diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/rep.h b/drivers/net/ethernet/marvell/octeontx2/nic/rep.h index c04874c4d4c6..5d39bf636655 100644 --- a/drivers/net/ethernet/marvell/octeontx2/nic/rep.h +++ b/drivers/net/ethernet/marvell/octeontx2/nic/rep.h @@ -17,9 +17,23 @@ #define PCI_DEVID_RVU_REP 0xA0E0 #define RVU_MAX_REP OTX2_MAX_CQ_CNT + +struct rep_stats { + u64 rx_bytes; + u64 rx_frames; + u64 rx_drops; + u64 rx_mcast_frames; + + u64 tx_bytes; + u64 tx_frames; + u64 tx_drops; +}; + struct rep_dev { struct otx2_nic *mdev; struct net_device *netdev; + struct rep_stats stats; + struct delayed_work stats_wrk; u16 rep_id; u16 pcifunc; };