From 244ae992e3e80e5c9c272c77324c831148457f95 Mon Sep 17 00:00:00 2001 From: Vinicius Costa Gomes Date: Tue, 20 Feb 2024 15:57:10 -0800 Subject: [PATCH 1/3] igc: Fix missing time sync events Fix "double" clearing of interrupts, which can cause external events or timestamps to be missed. The IGC_TSIRC Time Sync Interrupt Cause register can be cleared in two ways, by either reading it or by writing '1' into the specific cause bit. This is documented in section 8.16.1. The following flow was used: 1. read IGC_TSIRC into 'tsicr'; 2. handle the interrupts present in 'tsirc' and mark them in 'ack'; 3. write 'ack' into IGC_TSICR; As both (1) and (3) will clear the interrupt cause, if the same interrupt happens again between (1) and (3) it will be ignored, causing events to be missed. Remove the extra clear in (3). Fixes: 2c344ae24501 ("igc: Add support for TX timestamping") Reviewed-by: Kurt Kanzenbach Tested-by: Kurt Kanzenbach # Intel i225 Signed-off-by: Vinicius Costa Gomes Tested-by: Naama Meir Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/igc/igc_main.c | 12 +----------- 1 file changed, 1 insertion(+), 11 deletions(-) diff --git a/drivers/net/ethernet/intel/igc/igc_main.c b/drivers/net/ethernet/intel/igc/igc_main.c index 81c21a893ede9..e447ba0370568 100644 --- a/drivers/net/ethernet/intel/igc/igc_main.c +++ b/drivers/net/ethernet/intel/igc/igc_main.c @@ -5302,25 +5302,22 @@ igc_features_check(struct sk_buff *skb, struct net_device *dev, static void igc_tsync_interrupt(struct igc_adapter *adapter) { - u32 ack, tsauxc, sec, nsec, tsicr; struct igc_hw *hw = &adapter->hw; + u32 tsauxc, sec, nsec, tsicr; struct ptp_clock_event event; struct timespec64 ts; tsicr = rd32(IGC_TSICR); - ack = 0; if (tsicr & IGC_TSICR_SYS_WRAP) { event.type = PTP_CLOCK_PPS; if (adapter->ptp_caps.pps) ptp_clock_event(adapter->ptp_clock, &event); - ack |= IGC_TSICR_SYS_WRAP; } if (tsicr & IGC_TSICR_TXTS) { /* retrieve hardware timestamp */ igc_ptp_tx_tstamp_event(adapter); - ack |= IGC_TSICR_TXTS; } if (tsicr & IGC_TSICR_TT0) { @@ -5334,7 +5331,6 @@ static void igc_tsync_interrupt(struct igc_adapter *adapter) wr32(IGC_TSAUXC, tsauxc); adapter->perout[0].start = ts; spin_unlock(&adapter->tmreg_lock); - ack |= IGC_TSICR_TT0; } if (tsicr & IGC_TSICR_TT1) { @@ -5348,7 +5344,6 @@ static void igc_tsync_interrupt(struct igc_adapter *adapter) wr32(IGC_TSAUXC, tsauxc); adapter->perout[1].start = ts; spin_unlock(&adapter->tmreg_lock); - ack |= IGC_TSICR_TT1; } if (tsicr & IGC_TSICR_AUTT0) { @@ -5358,7 +5353,6 @@ static void igc_tsync_interrupt(struct igc_adapter *adapter) event.index = 0; event.timestamp = sec * NSEC_PER_SEC + nsec; ptp_clock_event(adapter->ptp_clock, &event); - ack |= IGC_TSICR_AUTT0; } if (tsicr & IGC_TSICR_AUTT1) { @@ -5368,11 +5362,7 @@ static void igc_tsync_interrupt(struct igc_adapter *adapter) event.index = 1; event.timestamp = sec * NSEC_PER_SEC + nsec; ptp_clock_event(adapter->ptp_clock, &event); - ack |= IGC_TSICR_AUTT1; } - - /* acknowledge the interrupts */ - wr32(IGC_TSICR, ack); } /** From ee14cc9ea19ba9678177e2224a9c58cce5937c73 Mon Sep 17 00:00:00 2001 From: Vinicius Costa Gomes Date: Tue, 20 Feb 2024 15:57:11 -0800 Subject: [PATCH 2/3] igb: Fix missing time sync events Fix "double" clearing of interrupts, which can cause external events or timestamps to be missed. The E1000_TSIRC Time Sync Interrupt Cause register can be cleared in two ways, by either reading it or by writing '1' into the specific cause bit. This is documented in section 8.16.1. The following flow was used: 1. read E1000_TSIRC into 'tsicr'; 2. handle the interrupts present into 'tsirc' and mark them in 'ack'; 3. write 'ack' into E1000_TSICR; As both (1) and (3) will clear the interrupt cause, if the same interrupt happens again between (1) and (3) it will be ignored, causing events to be missed. Remove the extra clear in (3). Fixes: 00c65578b47b ("igb: enable internal PPS for the i210") Acked-by: Richard Cochran Signed-off-by: Vinicius Costa Gomes Tested-by: Pucha Himasekhar Reddy (A Contingent worker at Intel) Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/igb/igb_main.c | 23 +++++------------------ 1 file changed, 5 insertions(+), 18 deletions(-) diff --git a/drivers/net/ethernet/intel/igb/igb_main.c b/drivers/net/ethernet/intel/igb/igb_main.c index cebb44f51d5f5..7662c42e35c11 100644 --- a/drivers/net/ethernet/intel/igb/igb_main.c +++ b/drivers/net/ethernet/intel/igb/igb_main.c @@ -6985,44 +6985,31 @@ static void igb_extts(struct igb_adapter *adapter, int tsintr_tt) static void igb_tsync_interrupt(struct igb_adapter *adapter) { struct e1000_hw *hw = &adapter->hw; - u32 ack = 0, tsicr = rd32(E1000_TSICR); + u32 tsicr = rd32(E1000_TSICR); struct ptp_clock_event event; if (tsicr & TSINTR_SYS_WRAP) { event.type = PTP_CLOCK_PPS; if (adapter->ptp_caps.pps) ptp_clock_event(adapter->ptp_clock, &event); - ack |= TSINTR_SYS_WRAP; } if (tsicr & E1000_TSICR_TXTS) { /* retrieve hardware timestamp */ schedule_work(&adapter->ptp_tx_work); - ack |= E1000_TSICR_TXTS; } - if (tsicr & TSINTR_TT0) { + if (tsicr & TSINTR_TT0) igb_perout(adapter, 0); - ack |= TSINTR_TT0; - } - if (tsicr & TSINTR_TT1) { + if (tsicr & TSINTR_TT1) igb_perout(adapter, 1); - ack |= TSINTR_TT1; - } - if (tsicr & TSINTR_AUTT0) { + if (tsicr & TSINTR_AUTT0) igb_extts(adapter, 0); - ack |= TSINTR_AUTT0; - } - if (tsicr & TSINTR_AUTT1) { + if (tsicr & TSINTR_AUTT1) igb_extts(adapter, 1); - ack |= TSINTR_AUTT1; - } - - /* acknowledge the interrupts */ - wr32(E1000_TSICR, ack); } static irqreturn_t igb_msix_other(int irq, void *data) From 257310e998700e60382fbd3f4fd275fdbd9b2aaf Mon Sep 17 00:00:00 2001 From: Przemek Kitszel Date: Tue, 27 Feb 2024 15:31:06 +0100 Subject: [PATCH 3/3] ice: fix stats being updated by way too large values Simplify stats accumulation logic to fix the case where we don't take previous stat value into account, we should always respect it. Main netdev stats of our PF (Tx/Rx packets/bytes) were reported orders of magnitude too big during OpenStack reconfiguration events, possibly other reconfiguration cases too. The regression was reported to be between 6.1 and 6.2, so I was almost certain that on of the two "preserve stats over reset" commits were the culprit. While reading the code, it was found that in some cases we will increase the stats by arbitrarily large number (thanks to ignoring "-prev" part of condition, after zeroing it). Note that this fixes also the case where we were around limits of u64, but that was not the regression reported. Full disclosure: I remember suggesting this particular piece of code to Ben a few years ago, so blame on me. Fixes: 2fd5e433cd26 ("ice: Accumulate HW and Netdev statistics over reset") Reported-by: Nebojsa Stevanovic Link: https://lore.kernel.org/intel-wired-lan/VI1PR02MB439744DEDAA7B59B9A2833FE912EA@VI1PR02MB4397.eurprd02.prod.outlook.com Reported-by: Christian Rohmann Link: https://lore.kernel.org/intel-wired-lan/f38a6ca4-af05-48b1-a3e6-17ef2054e525@inovex.de Reviewed-by: Jacob Keller Signed-off-by: Przemek Kitszel Reviewed-by: Simon Horman Tested-by: Pucha Himasekhar Reddy (A Contingent worker at Intel) Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/ice/ice_main.c | 24 +++++++++++------------ 1 file changed, 11 insertions(+), 13 deletions(-) diff --git a/drivers/net/ethernet/intel/ice/ice_main.c b/drivers/net/ethernet/intel/ice/ice_main.c index df6a68ab747ee..6d256dbcb77d0 100644 --- a/drivers/net/ethernet/intel/ice/ice_main.c +++ b/drivers/net/ethernet/intel/ice/ice_main.c @@ -6737,6 +6737,7 @@ static void ice_update_vsi_ring_stats(struct ice_vsi *vsi) { struct rtnl_link_stats64 *net_stats, *stats_prev; struct rtnl_link_stats64 *vsi_stats; + struct ice_pf *pf = vsi->back; u64 pkts, bytes; int i; @@ -6782,21 +6783,18 @@ static void ice_update_vsi_ring_stats(struct ice_vsi *vsi) net_stats = &vsi->net_stats; stats_prev = &vsi->net_stats_prev; - /* clear prev counters after reset */ - if (vsi_stats->tx_packets < stats_prev->tx_packets || - vsi_stats->rx_packets < stats_prev->rx_packets) { - stats_prev->tx_packets = 0; - stats_prev->tx_bytes = 0; - stats_prev->rx_packets = 0; - stats_prev->rx_bytes = 0; + /* Update netdev counters, but keep in mind that values could start at + * random value after PF reset. And as we increase the reported stat by + * diff of Prev-Cur, we need to be sure that Prev is valid. If it's not, + * let's skip this round. + */ + if (likely(pf->stat_prev_loaded)) { + net_stats->tx_packets += vsi_stats->tx_packets - stats_prev->tx_packets; + net_stats->tx_bytes += vsi_stats->tx_bytes - stats_prev->tx_bytes; + net_stats->rx_packets += vsi_stats->rx_packets - stats_prev->rx_packets; + net_stats->rx_bytes += vsi_stats->rx_bytes - stats_prev->rx_bytes; } - /* update netdev counters */ - net_stats->tx_packets += vsi_stats->tx_packets - stats_prev->tx_packets; - net_stats->tx_bytes += vsi_stats->tx_bytes - stats_prev->tx_bytes; - net_stats->rx_packets += vsi_stats->rx_packets - stats_prev->rx_packets; - net_stats->rx_bytes += vsi_stats->rx_bytes - stats_prev->rx_bytes; - stats_prev->tx_packets = vsi_stats->tx_packets; stats_prev->tx_bytes = vsi_stats->tx_bytes; stats_prev->rx_packets = vsi_stats->rx_packets;