From ea558de7238bb12c3435c47f0631e9d17bf4a09f Mon Sep 17 00:00:00 2001 From: Ivan Vecera Date: Sat, 16 Mar 2024 12:38:29 +0100 Subject: i40e: Enforce software interrupt during busy-poll exit As for ice bug fixed by commit b7306b42beaf ("ice: manage interrupts during poll exit") followed by commit 23be7075b318 ("ice: fix software generating extra interrupts") I'm seeing the similar issue also with i40e driver. In certain situation when busy-loop is enabled together with adaptive coalescing, the driver occasionally misses that there are outstanding descriptors to clean when exiting busy poll. Try to catch the remaining work by triggering a software interrupt when exiting busy poll. No extra interrupts will be generated when busy polling is not used. The issue was found when running sockperf ping-pong tcp test with adaptive coalescing and busy poll enabled (50 as value busy_pool and busy_read sysctl knobs) and results in huge latency spikes with more than 100000us. The fix is inspired from the ice driver and do the following: 1) During napi poll exit in case of busy-poll (napo_complete_done() returns false) this is recorded to q_vector that we were in busy loop. 2) Extends i40e_buildreg_itr() to be able to add an enforced software interrupt into built value 2) In i40e_update_enable_itr() enforces a software interrupt trigger if we are exiting busy poll to catch any pending clean-ups 3) Reuses unused 3rd ITR (interrupt throttle) index and set it to 20K interrupts per second to limit the number of these sw interrupts. Test results ============ Prior: [root@dell-per640-07 net]# sockperf ping-pong -i 10.9.9.1 --tcp -m 1000 --mps=max -t 120 sockperf: == version #3.10-no.git == sockperf[CLIENT] send on:sockperf: using recvfrom() to block on socket(s) [ 0] IP = 10.9.9.1 PORT = 11111 # TCP sockperf: Warmup stage (sending a few dummy messages)... sockperf: Starting test... sockperf: Test end (interrupted by timer) sockperf: Test ended sockperf: [Total Run] RunTime=119.999 sec; Warm up time=400 msec; SentMessages=2438563; ReceivedMessages=2438562 sockperf: ========= Printing statistics for Server No: 0 sockperf: [Valid Duration] RunTime=119.549 sec; SentMessages=2429473; ReceivedMessages=2429473 sockperf: ====> avg-latency=24.571 (std-dev=93.297, mean-ad=4.904, median-ad=1.510, siqr=1.063, cv=3.797, std-error=0.060, 99.0% ci=[24.417, 24.725]) sockperf: # dropped messages = 0; # duplicated messages = 0; # out-of-order messages = 0 sockperf: Summary: Latency is 24.571 usec sockperf: Total 2429473 observations; each percentile contains 24294.73 observations sockperf: ---> observation = 103294.331 sockperf: ---> percentile 99.999 = 45.633 sockperf: ---> percentile 99.990 = 37.013 sockperf: ---> percentile 99.900 = 35.910 sockperf: ---> percentile 99.000 = 33.390 sockperf: ---> percentile 90.000 = 28.626 sockperf: ---> percentile 75.000 = 27.741 sockperf: ---> percentile 50.000 = 26.743 sockperf: ---> percentile 25.000 = 25.614 sockperf: ---> observation = 12.220 After: [root@dell-per640-07 net]# sockperf ping-pong -i 10.9.9.1 --tcp -m 1000 --mps=max -t 120 sockperf: == version #3.10-no.git == sockperf[CLIENT] send on:sockperf: using recvfrom() to block on socket(s) [ 0] IP = 10.9.9.1 PORT = 11111 # TCP sockperf: Warmup stage (sending a few dummy messages)... sockperf: Starting test... sockperf: Test end (interrupted by timer) sockperf: Test ended sockperf: [Total Run] RunTime=119.999 sec; Warm up time=400 msec; SentMessages=2400055; ReceivedMessages=2400054 sockperf: ========= Printing statistics for Server No: 0 sockperf: [Valid Duration] RunTime=119.549 sec; SentMessages=2391186; ReceivedMessages=2391186 sockperf: ====> avg-latency=24.965 (std-dev=5.934, mean-ad=4.642, median-ad=1.485, siqr=1.067, cv=0.238, std-error=0.004, 99.0% ci=[24.955, 24.975]) sockperf: # dropped messages = 0; # duplicated messages = 0; # out-of-order messages = 0 sockperf: Summary: Latency is 24.965 usec sockperf: Total 2391186 observations; each percentile contains 23911.86 observations sockperf: ---> observation = 195.841 sockperf: ---> percentile 99.999 = 45.026 sockperf: ---> percentile 99.990 = 39.009 sockperf: ---> percentile 99.900 = 35.922 sockperf: ---> percentile 99.000 = 33.482 sockperf: ---> percentile 90.000 = 28.902 sockperf: ---> percentile 75.000 = 27.821 sockperf: ---> percentile 50.000 = 26.860 sockperf: ---> percentile 25.000 = 25.685 sockperf: ---> observation = 12.277 Fixes: 0bcd952feec7 ("ethernet/intel: consolidate NAPI and NAPI exit") Reported-by: Hugo Ferreira Reviewed-by: Michal Schmidt Signed-off-by: Ivan Vecera Reviewed-by: Jesse Brandeburg Tested-by: Pucha Himasekhar Reddy (A Contingent worker at Intel) Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/i40e/i40e.h | 1 + drivers/net/ethernet/intel/i40e/i40e_main.c | 6 ++ drivers/net/ethernet/intel/i40e/i40e_register.h | 3 + drivers/net/ethernet/intel/i40e/i40e_txrx.c | 82 ++++++++++++++++++------- drivers/net/ethernet/intel/i40e/i40e_txrx.h | 1 + 5 files changed, 72 insertions(+), 21 deletions(-) diff --git a/drivers/net/ethernet/intel/i40e/i40e.h b/drivers/net/ethernet/intel/i40e/i40e.h index ba24f3fa92c3..2fbabcdb5bb5 100644 --- a/drivers/net/ethernet/intel/i40e/i40e.h +++ b/drivers/net/ethernet/intel/i40e/i40e.h @@ -955,6 +955,7 @@ struct i40e_q_vector { struct rcu_head rcu; /* to avoid race with update stats on free */ char name[I40E_INT_NAME_STR_LEN]; bool arm_wb_state; + bool in_busy_poll; int irq_num; /* IRQ assigned to this q_vector */ } ____cacheline_internodealigned_in_smp; diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c index f86578857e8a..6576a0081093 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_main.c +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c @@ -3911,6 +3911,12 @@ static void i40e_vsi_configure_msix(struct i40e_vsi *vsi) q_vector->tx.target_itr >> 1); q_vector->tx.current_itr = q_vector->tx.target_itr; + /* Set ITR for software interrupts triggered after exiting + * busy-loop polling. + */ + wr32(hw, I40E_PFINT_ITRN(I40E_SW_ITR, vector - 1), + I40E_ITR_20K); + wr32(hw, I40E_PFINT_RATEN(vector - 1), i40e_intrl_usec_to_reg(vsi->int_rate_limit)); diff --git a/drivers/net/ethernet/intel/i40e/i40e_register.h b/drivers/net/ethernet/intel/i40e/i40e_register.h index 14ab642cafdb..432afbb64201 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_register.h +++ b/drivers/net/ethernet/intel/i40e/i40e_register.h @@ -333,8 +333,11 @@ #define I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT 3 #define I40E_PFINT_DYN_CTLN_ITR_INDX_MASK I40E_MASK(0x3, I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) #define I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT 5 +#define I40E_PFINT_DYN_CTLN_INTERVAL_MASK I40E_MASK(0xFFF, I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT) #define I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_SHIFT 24 #define I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK I40E_MASK(0x1, I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_SHIFT) +#define I40E_PFINT_DYN_CTLN_SW_ITR_INDX_SHIFT 25 +#define I40E_PFINT_DYN_CTLN_SW_ITR_INDX_MASK I40E_MASK(0x3, I40E_PFINT_DYN_CTLN_SW_ITR_INDX_SHIFT) #define I40E_PFINT_ICR0 0x00038780 /* Reset: CORER */ #define I40E_PFINT_ICR0_INTEVENT_SHIFT 0 #define I40E_PFINT_ICR0_INTEVENT_MASK I40E_MASK(0x1, I40E_PFINT_ICR0_INTEVENT_SHIFT) diff --git a/drivers/net/ethernet/intel/i40e/i40e_txrx.c b/drivers/net/ethernet/intel/i40e/i40e_txrx.c index 0d7177083708..1a12b732818e 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_txrx.c +++ b/drivers/net/ethernet/intel/i40e/i40e_txrx.c @@ -2630,7 +2630,22 @@ process_next: return failure ? budget : (int)total_rx_packets; } -static inline u32 i40e_buildreg_itr(const int type, u16 itr) +/** + * i40e_buildreg_itr - build a value for writing to I40E_PFINT_DYN_CTLN register + * @itr_idx: interrupt throttling index + * @interval: interrupt throttling interval value in usecs + * @force_swint: force software interrupt + * + * The function builds a value for I40E_PFINT_DYN_CTLN register that + * is used to update interrupt throttling interval for specified ITR index + * and optionally enforces a software interrupt. If the @itr_idx is equal + * to I40E_ITR_NONE then no interval change is applied and only @force_swint + * parameter is taken into account. If the interval change and enforced + * software interrupt are not requested then the built value just enables + * appropriate vector interrupt. + **/ +static u32 i40e_buildreg_itr(enum i40e_dyn_idx itr_idx, u16 interval, + bool force_swint) { u32 val; @@ -2644,23 +2659,33 @@ static inline u32 i40e_buildreg_itr(const int type, u16 itr) * an event in the PBA anyway so we need to rely on the automask * to hold pending events for us until the interrupt is re-enabled * - * The itr value is reported in microseconds, and the register - * value is recorded in 2 microsecond units. For this reason we - * only need to shift by the interval shift - 1 instead of the - * full value. + * We have to shift the given value as it is reported in microseconds + * and the register value is recorded in 2 microsecond units. */ - itr &= I40E_ITR_MASK; + interval >>= 1; + /* 1. Enable vector interrupt + * 2. Update the interval for the specified ITR index + * (I40E_ITR_NONE in the register is used to indicate that + * no interval update is requested) + */ val = I40E_PFINT_DYN_CTLN_INTENA_MASK | - (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) | - (itr << (I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT - 1)); + FIELD_PREP(I40E_PFINT_DYN_CTLN_ITR_INDX_MASK, itr_idx) | + FIELD_PREP(I40E_PFINT_DYN_CTLN_INTERVAL_MASK, interval); + + /* 3. Enforce software interrupt trigger if requested + * (These software interrupts rate is limited by ITR2 that is + * set to 20K interrupts per second) + */ + if (force_swint) + val |= I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK | + I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK | + FIELD_PREP(I40E_PFINT_DYN_CTLN_SW_ITR_INDX_MASK, + I40E_SW_ITR); return val; } -/* a small macro to shorten up some long lines */ -#define INTREG I40E_PFINT_DYN_CTLN - /* The act of updating the ITR will cause it to immediately trigger. In order * to prevent this from throwing off adaptive update statistics we defer the * update so that it can only happen so often. So after either Tx or Rx are @@ -2679,8 +2704,10 @@ static inline u32 i40e_buildreg_itr(const int type, u16 itr) static inline void i40e_update_enable_itr(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector) { + enum i40e_dyn_idx itr_idx = I40E_ITR_NONE; struct i40e_hw *hw = &vsi->back->hw; - u32 intval; + u16 interval = 0; + u32 itr_val; /* If we don't have MSIX, then we only need to re-enable icr0 */ if (!test_bit(I40E_FLAG_MSIX_ENA, vsi->back->flags)) { @@ -2702,8 +2729,8 @@ static inline void i40e_update_enable_itr(struct i40e_vsi *vsi, */ if (q_vector->rx.target_itr < q_vector->rx.current_itr) { /* Rx ITR needs to be reduced, this is highest priority */ - intval = i40e_buildreg_itr(I40E_RX_ITR, - q_vector->rx.target_itr); + itr_idx = I40E_RX_ITR; + interval = q_vector->rx.target_itr; q_vector->rx.current_itr = q_vector->rx.target_itr; q_vector->itr_countdown = ITR_COUNTDOWN_START; } else if ((q_vector->tx.target_itr < q_vector->tx.current_itr) || @@ -2712,25 +2739,36 @@ static inline void i40e_update_enable_itr(struct i40e_vsi *vsi, /* Tx ITR needs to be reduced, this is second priority * Tx ITR needs to be increased more than Rx, fourth priority */ - intval = i40e_buildreg_itr(I40E_TX_ITR, - q_vector->tx.target_itr); + itr_idx = I40E_TX_ITR; + interval = q_vector->tx.target_itr; q_vector->tx.current_itr = q_vector->tx.target_itr; q_vector->itr_countdown = ITR_COUNTDOWN_START; } else if (q_vector->rx.current_itr != q_vector->rx.target_itr) { /* Rx ITR needs to be increased, third priority */ - intval = i40e_buildreg_itr(I40E_RX_ITR, - q_vector->rx.target_itr); + itr_idx = I40E_RX_ITR; + interval = q_vector->rx.target_itr; q_vector->rx.current_itr = q_vector->rx.target_itr; q_vector->itr_countdown = ITR_COUNTDOWN_START; } else { /* No ITR update, lowest priority */ - intval = i40e_buildreg_itr(I40E_ITR_NONE, 0); if (q_vector->itr_countdown) q_vector->itr_countdown--; } - if (!test_bit(__I40E_VSI_DOWN, vsi->state)) - wr32(hw, INTREG(q_vector->reg_idx), intval); + /* Do not update interrupt control register if VSI is down */ + if (test_bit(__I40E_VSI_DOWN, vsi->state)) + return; + + /* Update ITR interval if necessary and enforce software interrupt + * if we are exiting busy poll. + */ + if (q_vector->in_busy_poll) { + itr_val = i40e_buildreg_itr(itr_idx, interval, true); + q_vector->in_busy_poll = false; + } else { + itr_val = i40e_buildreg_itr(itr_idx, interval, false); + } + wr32(hw, I40E_PFINT_DYN_CTLN(q_vector->reg_idx), itr_val); } /** @@ -2845,6 +2883,8 @@ tx_only: */ if (likely(napi_complete_done(napi, work_done))) i40e_update_enable_itr(vsi, q_vector); + else + q_vector->in_busy_poll = true; return min(work_done, budget - 1); } diff --git a/drivers/net/ethernet/intel/i40e/i40e_txrx.h b/drivers/net/ethernet/intel/i40e/i40e_txrx.h index abf15067eb5d..2cdc7de6301c 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_txrx.h +++ b/drivers/net/ethernet/intel/i40e/i40e_txrx.h @@ -68,6 +68,7 @@ enum i40e_dyn_idx { /* these are indexes into ITRN registers */ #define I40E_RX_ITR I40E_IDX_ITR0 #define I40E_TX_ITR I40E_IDX_ITR1 +#define I40E_SW_ITR I40E_IDX_ITR2 /* Supported RSS offloads */ #define I40E_DEFAULT_RSS_HENA ( \ -- cgit v1.2.3 From eb58c598ce45b7e787568fe27016260417c3d807 Mon Sep 17 00:00:00 2001 From: Aleksandr Loktionov Date: Wed, 13 Mar 2024 10:44:00 +0100 Subject: i40e: fix i40e_count_filters() to count only active/new filters The bug usually affects untrusted VFs, because they are limited to 18 MACs, it affects them badly, not letting to create MAC all filters. Not stable to reproduce, it happens when VF user creates MAC filters when other MACVLAN operations are happened in parallel. But consequence is that VF can't receive desired traffic. Fix counter to be bumped only for new or active filters. Fixes: 621650cabee5 ("i40e: Refactoring VF MAC filters counting to make more reliable") Signed-off-by: Aleksandr Loktionov Reviewed-by: Arkadiusz Kubalewski Reviewed-by: Paul Menzel Tested-by: Rafal Romanowski Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/i40e/i40e_main.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c index 6576a0081093..48b9ddb2b1b3 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_main.c +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c @@ -1253,8 +1253,11 @@ int i40e_count_filters(struct i40e_vsi *vsi) int bkt; int cnt = 0; - hash_for_each_safe(vsi->mac_filter_hash, bkt, h, f, hlist) - ++cnt; + hash_for_each_safe(vsi->mac_filter_hash, bkt, h, f, hlist) { + if (f->state == I40E_FILTER_NEW || + f->state == I40E_FILTER_ACTIVE) + ++cnt; + } return cnt; } -- cgit v1.2.3 From f37c4eac99c258111d414d31b740437e1925b8e8 Mon Sep 17 00:00:00 2001 From: Aleksandr Loktionov Date: Wed, 13 Mar 2024 10:56:39 +0100 Subject: i40e: fix vf may be used uninitialized in this function warning To fix the regression introduced by commit 52424f974bc5, which causes servers hang in very hard to reproduce conditions with resets races. Using two sources for the information is the root cause. In this function before the fix bumping v didn't mean bumping vf pointer. But the code used this variables interchangeably, so stale vf could point to different/not intended vf. Remove redundant "v" variable and iterate via single VF pointer across whole function instead to guarantee VF pointer validity. Fixes: 52424f974bc5 ("i40e: Fix VF hang when reset is triggered on another VF") Signed-off-by: Aleksandr Loktionov Reviewed-by: Arkadiusz Kubalewski Reviewed-by: Przemek Kitszel Reviewed-by: Paul Menzel Tested-by: Rafal Romanowski Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c | 34 ++++++++++------------ 1 file changed, 16 insertions(+), 18 deletions(-) diff --git a/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c b/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c index 83a34e98bdc7..4c13f78af675 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +++ b/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c @@ -1624,8 +1624,8 @@ bool i40e_reset_all_vfs(struct i40e_pf *pf, bool flr) { struct i40e_hw *hw = &pf->hw; struct i40e_vf *vf; - int i, v; u32 reg; + int i; /* If we don't have any VFs, then there is nothing to reset */ if (!pf->num_alloc_vfs) @@ -1636,11 +1636,10 @@ bool i40e_reset_all_vfs(struct i40e_pf *pf, bool flr) return false; /* Begin reset on all VFs at once */ - for (v = 0; v < pf->num_alloc_vfs; v++) { - vf = &pf->vf[v]; + for (vf = &pf->vf[0]; vf < &pf->vf[pf->num_alloc_vfs]; ++vf) { /* If VF is being reset no need to trigger reset again */ if (!test_bit(I40E_VF_STATE_RESETTING, &vf->vf_states)) - i40e_trigger_vf_reset(&pf->vf[v], flr); + i40e_trigger_vf_reset(vf, flr); } /* HW requires some time to make sure it can flush the FIFO for a VF @@ -1649,14 +1648,13 @@ bool i40e_reset_all_vfs(struct i40e_pf *pf, bool flr) * the VFs using a simple iterator that increments once that VF has * finished resetting. */ - for (i = 0, v = 0; i < 10 && v < pf->num_alloc_vfs; i++) { + for (i = 0, vf = &pf->vf[0]; i < 10 && vf < &pf->vf[pf->num_alloc_vfs]; ++i) { usleep_range(10000, 20000); /* Check each VF in sequence, beginning with the VF to fail * the previous check. */ - while (v < pf->num_alloc_vfs) { - vf = &pf->vf[v]; + while (vf < &pf->vf[pf->num_alloc_vfs]) { if (!test_bit(I40E_VF_STATE_RESETTING, &vf->vf_states)) { reg = rd32(hw, I40E_VPGEN_VFRSTAT(vf->vf_id)); if (!(reg & I40E_VPGEN_VFRSTAT_VFRD_MASK)) @@ -1666,7 +1664,7 @@ bool i40e_reset_all_vfs(struct i40e_pf *pf, bool flr) /* If the current VF has finished resetting, move on * to the next VF in sequence. */ - v++; + ++vf; } } @@ -1676,39 +1674,39 @@ bool i40e_reset_all_vfs(struct i40e_pf *pf, bool flr) /* Display a warning if at least one VF didn't manage to reset in * time, but continue on with the operation. */ - if (v < pf->num_alloc_vfs) + if (vf < &pf->vf[pf->num_alloc_vfs]) dev_err(&pf->pdev->dev, "VF reset check timeout on VF %d\n", - pf->vf[v].vf_id); + vf->vf_id); usleep_range(10000, 20000); /* Begin disabling all the rings associated with VFs, but do not wait * between each VF. */ - for (v = 0; v < pf->num_alloc_vfs; v++) { + for (vf = &pf->vf[0]; vf < &pf->vf[pf->num_alloc_vfs]; ++vf) { /* On initial reset, we don't have any queues to disable */ - if (pf->vf[v].lan_vsi_idx == 0) + if (vf->lan_vsi_idx == 0) continue; /* If VF is reset in another thread just continue */ if (test_bit(I40E_VF_STATE_RESETTING, &vf->vf_states)) continue; - i40e_vsi_stop_rings_no_wait(pf->vsi[pf->vf[v].lan_vsi_idx]); + i40e_vsi_stop_rings_no_wait(pf->vsi[vf->lan_vsi_idx]); } /* Now that we've notified HW to disable all of the VF rings, wait * until they finish. */ - for (v = 0; v < pf->num_alloc_vfs; v++) { + for (vf = &pf->vf[0]; vf < &pf->vf[pf->num_alloc_vfs]; ++vf) { /* On initial reset, we don't have any queues to disable */ - if (pf->vf[v].lan_vsi_idx == 0) + if (vf->lan_vsi_idx == 0) continue; /* If VF is reset in another thread just continue */ if (test_bit(I40E_VF_STATE_RESETTING, &vf->vf_states)) continue; - i40e_vsi_wait_queues_disabled(pf->vsi[pf->vf[v].lan_vsi_idx]); + i40e_vsi_wait_queues_disabled(pf->vsi[vf->lan_vsi_idx]); } /* Hw may need up to 50ms to finish disabling the RX queues. We @@ -1717,12 +1715,12 @@ bool i40e_reset_all_vfs(struct i40e_pf *pf, bool flr) mdelay(50); /* Finish the reset on each VF */ - for (v = 0; v < pf->num_alloc_vfs; v++) { + for (vf = &pf->vf[0]; vf < &pf->vf[pf->num_alloc_vfs]; ++vf) { /* If VF is reset in another thread just continue */ if (test_bit(I40E_VF_STATE_RESETTING, &vf->vf_states)) continue; - i40e_cleanup_reset_vf(&pf->vf[v]); + i40e_cleanup_reset_vf(vf); } i40e_flush(hw); -- cgit v1.2.3