* [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter @ 2024-10-25 19:47 Vadim Fedorenko 2024-10-25 19:47 ` [PATCH net-next v2 2/2] bnxt_en: replace PTP spinlock with seqlock Vadim Fedorenko 2024-10-25 21:31 ` [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter Michael Chan 0 siblings, 2 replies; 8+ messages in thread From: Vadim Fedorenko @ 2024-10-25 19:47 UTC (permalink / raw) To: Vadim Fedorenko, Michael Chan, Pavan Chebbi, Jakub Kicinski Cc: Andrew Lunn, Paolo Abeni, David S. Miller, netdev, Richard Cochran, Vadim Fedorenko This hardware can provide only 48 bits of cycle counter. We can leave only 24 bits in the cache to extend RX timestamps from 32 bits to 48 bits. This make cache writes atomic even on 32 bit platforms and we can simply use READ_ONCE()/WRITE_ONCE() pair and remove spinlock. The configuration structure will be also reduced by 4 bytes. Signed-off-by: Vadim Fedorenko <vadfed@meta.com> --- drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c | 8 ++++---- drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h | 16 +--------------- 2 files changed, 5 insertions(+), 19 deletions(-) diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c index fa514be87650..c7e626b9098a 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c @@ -106,7 +106,7 @@ static void bnxt_ptp_get_current_time(struct bnxt *bp) if (!ptp) return; spin_lock_irqsave(&ptp->ptp_lock, flags); - WRITE_ONCE(ptp->old_time, ptp->current_time); + WRITE_ONCE(ptp->old_time, (u32)(ptp->current_time >> 24)); bnxt_refclk_read(bp, NULL, &ptp->current_time); spin_unlock_irqrestore(&ptp->ptp_lock, flags); } @@ -174,7 +174,7 @@ void bnxt_ptp_update_current_time(struct bnxt *bp) struct bnxt_ptp_cfg *ptp = bp->ptp_cfg; bnxt_refclk_read(ptp->bp, NULL, &ptp->current_time); - WRITE_ONCE(ptp->old_time, ptp->current_time); + WRITE_ONCE(ptp->old_time, (u32)(ptp->current_time >> 24)); } static int bnxt_ptp_adjphc(struct bnxt_ptp_cfg *ptp, s64 delta) @@ -813,7 +813,7 @@ int bnxt_get_rx_ts_p5(struct bnxt *bp, u64 *ts, u32 pkt_ts) if (!ptp) return -ENODEV; - BNXT_READ_TIME64(ptp, time, ptp->old_time); + time = (u64)(READ_ONCE(ptp->old_time) << 24); *ts = (time & BNXT_HI_TIMER_MASK) | pkt_ts; if (pkt_ts < (time & BNXT_LO_TIMER_MASK)) *ts += BNXT_LO_TIMER_MASK + 1; @@ -1079,7 +1079,7 @@ int bnxt_ptp_init(struct bnxt *bp, bool phc_cfg) spin_lock_irqsave(&ptp->ptp_lock, flags); bnxt_refclk_read(bp, NULL, &ptp->current_time); - WRITE_ONCE(ptp->old_time, ptp->current_time); + WRITE_ONCE(ptp->old_time, (u32)(ptp->current_time >> 24)); spin_unlock_irqrestore(&ptp->ptp_lock, flags); ptp_schedule_worker(ptp->ptp_clock, 0); } diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h index f322466ecad3..80046bd314db 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h @@ -106,10 +106,10 @@ struct bnxt_ptp_cfg { /* serialize ts tx request queuing */ spinlock_t ptp_tx_lock; u64 current_time; - u64 old_time; unsigned long next_period; unsigned long next_overflow_check; u32 cmult; + u32 old_time; /* a 23b shift cyclecounter will overflow in ~36 mins. Check overflow every 18 mins. */ #define BNXT_PHC_OVERFLOW_PERIOD (18 * 60 * HZ) @@ -145,20 +145,6 @@ struct bnxt_ptp_cfg { struct bnxt_ptp_stats stats; }; -#if BITS_PER_LONG == 32 -#define BNXT_READ_TIME64(ptp, dst, src) \ -do { \ - unsigned long flags; \ - \ - spin_lock_irqsave(&(ptp)->ptp_lock, flags); \ - (dst) = (src); \ - spin_unlock_irqrestore(&(ptp)->ptp_lock, flags); \ -} while (0) -#else -#define BNXT_READ_TIME64(ptp, dst, src) \ - ((dst) = READ_ONCE(src)) -#endif - #define BNXT_PTP_INC_TX_AVAIL(ptp) \ do { \ spin_lock_bh(&(ptp)->ptp_tx_lock); \ -- 2.43.5 ^ permalink raw reply related [flat|nested] 8+ messages in thread
* [PATCH net-next v2 2/2] bnxt_en: replace PTP spinlock with seqlock 2024-10-25 19:47 [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter Vadim Fedorenko @ 2024-10-25 19:47 ` Vadim Fedorenko 2024-10-26 11:23 ` kernel test robot 2024-10-25 21:31 ` [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter Michael Chan 1 sibling, 1 reply; 8+ messages in thread From: Vadim Fedorenko @ 2024-10-25 19:47 UTC (permalink / raw) To: Vadim Fedorenko, Michael Chan, Pavan Chebbi, Jakub Kicinski Cc: Andrew Lunn, Paolo Abeni, David S. Miller, netdev, Richard Cochran, Vadim Fedorenko We can see high contention on ptp_lock while doing RX timestamping on high packet rates over several queues. Spinlock is not effecient to protect timecounter for RX timestamps when reads are the most usual operations and writes are only occasional. It's better to use seqlock in such cases. Signed-off-by: Vadim Fedorenko <vadfed@meta.com> --- v2: - use read_excl lock to serialize reg access with FW reset --- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 18 ++--- drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c | 73 ++++++------------- drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h | 14 +++- 3 files changed, 46 insertions(+), 59 deletions(-) diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c index 6dd6541d8619..f5ddd01c78eb 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c @@ -2256,9 +2256,7 @@ static int bnxt_rx_pkt(struct bnxt *bp, struct bnxt_cp_ring_info *cpr, struct bnxt_ptp_cfg *ptp = bp->ptp_cfg; unsigned long flags; - spin_lock_irqsave(&ptp->ptp_lock, flags); - ns = timecounter_cyc2time(&ptp->tc, ts); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + ns = bnxt_timecounter_cyc2time(ptp, ts); memset(skb_hwtstamps(skb), 0, sizeof(*skb_hwtstamps(skb))); skb_hwtstamps(skb)->hwtstamp = ns_to_ktime(ns); @@ -2764,12 +2762,12 @@ static int bnxt_async_event_process(struct bnxt *bp, if (!ptp) goto async_event_process_exit; - spin_lock_irqsave(&ptp->ptp_lock, flags); bnxt_ptp_update_current_time(bp); ns = (((u64)BNXT_EVENT_PHC_RTC_UPDATE(data1) << BNXT_PHC_BITS) | ptp->current_time); + write_seqlock_irqsave(&ptp->ptp_lock, flags); bnxt_ptp_rtc_timecounter_init(ptp, ns); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + write_sequnlock_irqrestore(&ptp->ptp_lock, flags); } break; } @@ -13496,12 +13494,13 @@ static void bnxt_force_fw_reset(struct bnxt *bp) test_bit(BNXT_STATE_IN_FW_RESET, &bp->state)) return; + /* we have to serialize with bnxt_refclk_read()*/ if (ptp) { unsigned long flags; - spin_lock_irqsave(&ptp->ptp_lock, flags); + write_seqlock_irqsave(&ptp->ptp_lock, flags); set_bit(BNXT_STATE_IN_FW_RESET, &bp->state); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + write_sequnlock_irqrestore(&ptp->ptp_lock, flags); } else { set_bit(BNXT_STATE_IN_FW_RESET, &bp->state); } @@ -13565,12 +13564,13 @@ void bnxt_fw_reset(struct bnxt *bp) struct bnxt_ptp_cfg *ptp = bp->ptp_cfg; int n = 0, tmo; + /* we have to serialize with bnxt_refclk_read()*/ if (ptp) { unsigned long flags; - spin_lock_irqsave(&ptp->ptp_lock, flags); + write_seqlock_irqsave(&ptp->ptp_lock, flags); set_bit(BNXT_STATE_IN_FW_RESET, &bp->state); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + write_sequnlock_irqrestore(&ptp->ptp_lock, flags); } else { set_bit(BNXT_STATE_IN_FW_RESET, &bp->state); } diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c index c7e626b9098a..6c507fd5689c 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c @@ -67,19 +67,21 @@ static int bnxt_ptp_settime(struct ptp_clock_info *ptp_info, if (BNXT_PTP_USE_RTC(ptp->bp)) return bnxt_ptp_cfg_settime(ptp->bp, ns); - spin_lock_irqsave(&ptp->ptp_lock, flags); + write_seqlock_irqsave(&ptp->ptp_lock, flags); timecounter_init(&ptp->tc, &ptp->cc, ns); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + write_sequnlock_irqrestore(&ptp->ptp_lock, flags); return 0; } -/* Caller holds ptp_lock */ static int bnxt_refclk_read(struct bnxt *bp, struct ptp_system_timestamp *sts, u64 *ns) { struct bnxt_ptp_cfg *ptp = bp->ptp_cfg; u32 high_before, high_now, low; + unsigned long flags; + /* We have to serialize reg access and FW reset */ + read_seqlock_excl_irqsave(&ptp->ptp_lock, flags); if (test_bit(BNXT_STATE_IN_FW_RESET, &bp->state)) return -EIO; @@ -93,6 +95,7 @@ static int bnxt_refclk_read(struct bnxt *bp, struct ptp_system_timestamp *sts, low = readl(bp->bar0 + ptp->refclk_mapped_regs[0]); ptp_read_system_postts(sts); } + read_sequnlock_excl_irqrestore(&ptp->ptp_lock, flags); *ns = ((u64)high_now << 32) | low; return 0; @@ -101,14 +104,11 @@ static int bnxt_refclk_read(struct bnxt *bp, struct ptp_system_timestamp *sts, static void bnxt_ptp_get_current_time(struct bnxt *bp) { struct bnxt_ptp_cfg *ptp = bp->ptp_cfg; - unsigned long flags; if (!ptp) return; - spin_lock_irqsave(&ptp->ptp_lock, flags); WRITE_ONCE(ptp->old_time, (u32)(ptp->current_time >> 24)); bnxt_refclk_read(bp, NULL, &ptp->current_time); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); } static int bnxt_hwrm_port_ts_query(struct bnxt *bp, u32 flags, u64 *ts, @@ -151,24 +151,19 @@ static int bnxt_ptp_gettimex(struct ptp_clock_info *ptp_info, { struct bnxt_ptp_cfg *ptp = container_of(ptp_info, struct bnxt_ptp_cfg, ptp_info); - unsigned long flags; u64 ns, cycles; int rc; - spin_lock_irqsave(&ptp->ptp_lock, flags); rc = bnxt_refclk_read(ptp->bp, sts, &cycles); - if (rc) { - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + if (rc) return rc; - } - ns = timecounter_cyc2time(&ptp->tc, cycles); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + + ns = bnxt_timecounter_cyc2time(ptp, cycles); *ts = ns_to_timespec64(ns); return 0; } -/* Caller holds ptp_lock */ void bnxt_ptp_update_current_time(struct bnxt *bp) { struct bnxt_ptp_cfg *ptp = bp->ptp_cfg; @@ -180,7 +175,6 @@ void bnxt_ptp_update_current_time(struct bnxt *bp) static int bnxt_ptp_adjphc(struct bnxt_ptp_cfg *ptp, s64 delta) { struct hwrm_port_mac_cfg_input *req; - unsigned long flags; int rc; rc = hwrm_req_init(ptp->bp, req, HWRM_PORT_MAC_CFG); @@ -194,9 +188,7 @@ static int bnxt_ptp_adjphc(struct bnxt_ptp_cfg *ptp, s64 delta) if (rc) { netdev_err(ptp->bp->dev, "ptp adjphc failed. rc = %x\n", rc); } else { - spin_lock_irqsave(&ptp->ptp_lock, flags); bnxt_ptp_update_current_time(ptp->bp); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); } return rc; @@ -211,9 +203,9 @@ static int bnxt_ptp_adjtime(struct ptp_clock_info *ptp_info, s64 delta) if (BNXT_PTP_USE_RTC(ptp->bp)) return bnxt_ptp_adjphc(ptp, delta); - spin_lock_irqsave(&ptp->ptp_lock, flags); + write_seqlock_irqsave(&ptp->ptp_lock, flags); timecounter_adjtime(&ptp->tc, delta); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + write_sequnlock_irqrestore(&ptp->ptp_lock, flags); return 0; } @@ -246,10 +238,10 @@ static int bnxt_ptp_adjfine(struct ptp_clock_info *ptp_info, long scaled_ppm) if (!BNXT_MH(bp)) return bnxt_ptp_adjfine_rtc(bp, scaled_ppm); - spin_lock_irqsave(&ptp->ptp_lock, flags); + write_seqlock_irqsave(&ptp->ptp_lock, flags); timecounter_read(&ptp->tc); ptp->cc.mult = adjust_by_scaled_ppm(ptp->cmult, scaled_ppm); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + write_sequnlock_irqrestore(&ptp->ptp_lock, flags); return 0; } @@ -257,13 +249,10 @@ void bnxt_ptp_pps_event(struct bnxt *bp, u32 data1, u32 data2) { struct bnxt_ptp_cfg *ptp = bp->ptp_cfg; struct ptp_clock_event event; - unsigned long flags; u64 ns, pps_ts; pps_ts = EVENT_PPS_TS(data2, data1); - spin_lock_irqsave(&ptp->ptp_lock, flags); - ns = timecounter_cyc2time(&ptp->tc, pps_ts); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + ns = bnxt_timecounter_cyc2time(ptp, pps_ts); switch (EVENT_DATA2_PPS_EVENT_TYPE(data2)) { case ASYNC_EVENT_CMPL_PPS_TIMESTAMP_EVENT_DATA2_EVENT_TYPE_INTERNAL: @@ -400,17 +389,13 @@ static int bnxt_get_target_cycles(struct bnxt_ptp_cfg *ptp, u64 target_ns, { u64 cycles_now; u64 nsec_now, nsec_delta; - unsigned long flags; int rc; - spin_lock_irqsave(&ptp->ptp_lock, flags); rc = bnxt_refclk_read(ptp->bp, NULL, &cycles_now); - if (rc) { - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + if (rc) return rc; - } - nsec_now = timecounter_cyc2time(&ptp->tc, cycles_now); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + + nsec_now = bnxt_timecounter_cyc2time(ptp, cycles_now); nsec_delta = target_ns - nsec_now; *cycles_delta = div64_u64(nsec_delta << ptp->cc.shift, ptp->cc.mult); @@ -697,7 +682,6 @@ static int bnxt_stamp_tx_skb(struct bnxt *bp, int slot) struct skb_shared_hwtstamps timestamp; struct bnxt_ptp_tx_req *txts_req; unsigned long now = jiffies; - unsigned long flags; u64 ts = 0, ns = 0; u32 tmo = 0; int rc; @@ -711,9 +695,7 @@ static int bnxt_stamp_tx_skb(struct bnxt *bp, int slot) tmo, slot); if (!rc) { memset(×tamp, 0, sizeof(timestamp)); - spin_lock_irqsave(&ptp->ptp_lock, flags); - ns = timecounter_cyc2time(&ptp->tc, ts); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + ns = bnxt_timecounter_cyc2time(ptp, ts); timestamp.hwtstamp = ns_to_ktime(ns); skb_tstamp_tx(txts_req->tx_skb, ×tamp); ptp->stats.ts_pkts++; @@ -767,9 +749,9 @@ static long bnxt_ptp_ts_aux_work(struct ptp_clock_info *ptp_info) bnxt_ptp_get_current_time(bp); ptp->next_period = now + HZ; if (time_after_eq(now, ptp->next_overflow_check)) { - spin_lock_irqsave(&ptp->ptp_lock, flags); + write_seqlock_irqsave(&ptp->ptp_lock, flags); timecounter_read(&ptp->tc); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + write_sequnlock_irqrestore(&ptp->ptp_lock, flags); ptp->next_overflow_check = now + BNXT_PHC_OVERFLOW_PERIOD; } if (rc == -EAGAIN) @@ -829,7 +811,6 @@ void bnxt_tx_ts_cmp(struct bnxt *bp, struct bnxt_napi *bnapi, u32 opaque = tscmp->tx_ts_cmp_opaque; struct bnxt_tx_ring_info *txr; struct bnxt_sw_tx_bd *tx_buf; - unsigned long flags; u64 ts, ns; u16 cons; @@ -844,9 +825,7 @@ void bnxt_tx_ts_cmp(struct bnxt *bp, struct bnxt_napi *bnapi, le32_to_cpu(tscmp->tx_ts_cmp_flags_type), le32_to_cpu(tscmp->tx_ts_cmp_errors_v)); } else { - spin_lock_irqsave(&ptp->ptp_lock, flags); - ns = timecounter_cyc2time(&ptp->tc, ts); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); + ns = bnxt_timecounter_cyc2time(ptp, ts); timestamp.hwtstamp = ns_to_ktime(ns); skb_tstamp_tx(tx_buf->skb, ×tamp); } @@ -1005,9 +984,9 @@ int bnxt_ptp_init_rtc(struct bnxt *bp, bool phc_cfg) if (rc) return rc; } - spin_lock_irqsave(&bp->ptp_cfg->ptp_lock, flags); + write_seqlock_irqsave(&bp->ptp_cfg->ptp_lock, flags); bnxt_ptp_rtc_timecounter_init(bp->ptp_cfg, ns); - spin_unlock_irqrestore(&bp->ptp_cfg->ptp_lock, flags); + write_sequnlock_irqrestore(&bp->ptp_cfg->ptp_lock, flags); return 0; } @@ -1042,7 +1021,7 @@ int bnxt_ptp_init(struct bnxt *bp, bool phc_cfg) bnxt_ptp_free(bp); WRITE_ONCE(ptp->tx_avail, BNXT_MAX_TX_TS); - spin_lock_init(&ptp->ptp_lock); + seqlock_init(&ptp->ptp_lock); spin_lock_init(&ptp->ptp_tx_lock); if (BNXT_PTP_USE_RTC(bp)) { @@ -1075,12 +1054,8 @@ int bnxt_ptp_init(struct bnxt *bp, bool phc_cfg) atomic64_set(&ptp->stats.ts_err, 0); if (bp->flags & BNXT_FLAG_CHIP_P5_PLUS) { - unsigned long flags; - - spin_lock_irqsave(&ptp->ptp_lock, flags); bnxt_refclk_read(bp, NULL, &ptp->current_time); WRITE_ONCE(ptp->old_time, (u32)(ptp->current_time >> 24)); - spin_unlock_irqrestore(&ptp->ptp_lock, flags); ptp_schedule_worker(ptp->ptp_clock, 0); } ptp->txts_tmo = BNXT_PTP_DFLT_TX_TMO; diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h index 80046bd314db..e2b401251399 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h @@ -102,7 +102,7 @@ struct bnxt_ptp_cfg { struct timecounter tc; struct bnxt_pps pps_info; /* serialize timecounter access */ - spinlock_t ptp_lock; + seqlock_t ptp_lock; /* serialize ts tx request queuing */ spinlock_t ptp_tx_lock; u64 current_time; @@ -168,4 +168,16 @@ void bnxt_ptp_rtc_timecounter_init(struct bnxt_ptp_cfg *ptp, u64 ns); int bnxt_ptp_init_rtc(struct bnxt *bp, bool phc_cfg); int bnxt_ptp_init(struct bnxt *bp, bool phc_cfg); void bnxt_ptp_clear(struct bnxt *bp); +static inline u64 bnxt_timecounter_cyc2time(struct bnxt_ptp_cfg *ptp, u64 ts) +{ + unsigned int seq; + u64 ns; + + do { + seq = read_seqbegin(&ptp->ptp_lock); + ns = timecounter_cyc2time(&ptp->tc, ts); + } while (read_seqretry(&ptp->ptp_lock, seq)); + + return ns; +} #endif -- 2.43.5 ^ permalink raw reply related [flat|nested] 8+ messages in thread
* Re: [PATCH net-next v2 2/2] bnxt_en: replace PTP spinlock with seqlock 2024-10-25 19:47 ` [PATCH net-next v2 2/2] bnxt_en: replace PTP spinlock with seqlock Vadim Fedorenko @ 2024-10-26 11:23 ` kernel test robot 0 siblings, 0 replies; 8+ messages in thread From: kernel test robot @ 2024-10-26 11:23 UTC (permalink / raw) To: Vadim Fedorenko, Vadim Fedorenko, Michael Chan, Pavan Chebbi, Jakub Kicinski Cc: llvm, oe-kbuild-all, Andrew Lunn, Paolo Abeni, netdev, Richard Cochran Hi Vadim, kernel test robot noticed the following build warnings: [auto build test WARNING on net-next/main] url: https://github.com/intel-lab-lkp/linux/commits/Vadim-Fedorenko/bnxt_en-replace-PTP-spinlock-with-seqlock/20241026-035014 base: net-next/main patch link: https://lore.kernel.org/r/20241025194753.3070604-2-vadfed%40meta.com patch subject: [PATCH net-next v2 2/2] bnxt_en: replace PTP spinlock with seqlock config: i386-buildonly-randconfig-004-20241026 (https://download.01.org/0day-ci/archive/20241026/202410261913.KgwVFAOV-lkp@intel.com/config) compiler: clang version 19.1.2 (https://github.com/llvm/llvm-project 7ba7d8e2f7b6445b60679da826210cdde29eaf8b) reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20241026/202410261913.KgwVFAOV-lkp@intel.com/reproduce) If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <lkp@intel.com> | Closes: https://lore.kernel.org/oe-kbuild-all/202410261913.KgwVFAOV-lkp@intel.com/ All warnings (new ones prefixed by >>): In file included from drivers/net/ethernet/broadcom/bnxt/bnxt.c:21: In file included from include/linux/pci.h:1650: In file included from include/linux/dmapool.h:14: In file included from include/linux/scatterlist.h:8: In file included from include/linux/mm.h:2213: include/linux/vmstat.h:518:36: warning: arithmetic between different enumeration types ('enum node_stat_item' and 'enum lru_list') [-Wenum-enum-conversion] 518 | return node_stat_name(NR_LRU_BASE + lru) + 3; // skip "nr_" | ~~~~~~~~~~~ ^ ~~~ >> drivers/net/ethernet/broadcom/bnxt/bnxt.c:2257:19: warning: unused variable 'flags' [-Wunused-variable] 2257 | unsigned long flags; | ^~~~~ 2 warnings generated. vim +/flags +2257 drivers/net/ethernet/broadcom/bnxt/bnxt.c a7445d69809fe3 Michael Chan 2023-12-01 2010 c0c050c58d8409 Michael Chan 2015-10-22 2011 /* returns the following: c0c050c58d8409 Michael Chan 2015-10-22 2012 * 1 - 1 packet successfully received c0c050c58d8409 Michael Chan 2015-10-22 2013 * 0 - successful TPA_START, packet not completed yet c0c050c58d8409 Michael Chan 2015-10-22 2014 * -EBUSY - completion ring does not have all the agg buffers yet c0c050c58d8409 Michael Chan 2015-10-22 2015 * -ENOMEM - packet aborted due to out of memory c0c050c58d8409 Michael Chan 2015-10-22 2016 * -EIO - packet aborted due to hw error indicated in BD c0c050c58d8409 Michael Chan 2015-10-22 2017 */ e44758b78ae814 Michael Chan 2018-10-14 2018 static int bnxt_rx_pkt(struct bnxt *bp, struct bnxt_cp_ring_info *cpr, e44758b78ae814 Michael Chan 2018-10-14 2019 u32 *raw_cons, u8 *event) c0c050c58d8409 Michael Chan 2015-10-22 2020 { e44758b78ae814 Michael Chan 2018-10-14 2021 struct bnxt_napi *bnapi = cpr->bnapi; b6ab4b01f53b5f Michael Chan 2016-01-02 2022 struct bnxt_rx_ring_info *rxr = bnapi->rx_ring; c0c050c58d8409 Michael Chan 2015-10-22 2023 struct net_device *dev = bp->dev; c0c050c58d8409 Michael Chan 2015-10-22 2024 struct rx_cmp *rxcmp; c0c050c58d8409 Michael Chan 2015-10-22 2025 struct rx_cmp_ext *rxcmp1; c0c050c58d8409 Michael Chan 2015-10-22 2026 u32 tmp_raw_cons = *raw_cons; a7445d69809fe3 Michael Chan 2023-12-01 2027 u16 cons, prod, cp_cons = RING_CMP(tmp_raw_cons); c0c050c58d8409 Michael Chan 2015-10-22 2028 struct bnxt_sw_rx_bd *rx_buf; c0c050c58d8409 Michael Chan 2015-10-22 2029 unsigned int len; 6bb19474391d17 Michael Chan 2017-02-06 2030 u8 *data_ptr, agg_bufs, cmp_type; ee536dcbdce496 Andy Gospodarek 2022-04-08 2031 bool xdp_active = false; c0c050c58d8409 Michael Chan 2015-10-22 2032 dma_addr_t dma_addr; c0c050c58d8409 Michael Chan 2015-10-22 2033 struct sk_buff *skb; b231c3f3414cfc Andy Gospodarek 2022-04-08 2034 struct xdp_buff xdp; 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2035 u32 flags, misc; c13e268c076865 Michael Chan 2023-12-07 2036 u32 cmpl_ts; 6bb19474391d17 Michael Chan 2017-02-06 2037 void *data; c0c050c58d8409 Michael Chan 2015-10-22 2038 int rc = 0; c0c050c58d8409 Michael Chan 2015-10-22 2039 c0c050c58d8409 Michael Chan 2015-10-22 2040 rxcmp = (struct rx_cmp *) c0c050c58d8409 Michael Chan 2015-10-22 2041 &cpr->cp_desc_ring[CP_RING(cp_cons)][CP_IDX(cp_cons)]; c0c050c58d8409 Michael Chan 2015-10-22 2042 8fe88ce7ab3181 Michael Chan 2019-07-29 2043 cmp_type = RX_CMP_TYPE(rxcmp); 8fe88ce7ab3181 Michael Chan 2019-07-29 2044 8fe88ce7ab3181 Michael Chan 2019-07-29 2045 if (cmp_type == CMP_TYPE_RX_TPA_AGG_CMP) { 8fe88ce7ab3181 Michael Chan 2019-07-29 2046 bnxt_tpa_agg(bp, rxr, (struct rx_agg_cmp *)rxcmp); 8fe88ce7ab3181 Michael Chan 2019-07-29 2047 goto next_rx_no_prod_no_len; 8fe88ce7ab3181 Michael Chan 2019-07-29 2048 } 8fe88ce7ab3181 Michael Chan 2019-07-29 2049 c0c050c58d8409 Michael Chan 2015-10-22 2050 tmp_raw_cons = NEXT_RAW_CMP(tmp_raw_cons); c0c050c58d8409 Michael Chan 2015-10-22 2051 cp_cons = RING_CMP(tmp_raw_cons); c0c050c58d8409 Michael Chan 2015-10-22 2052 rxcmp1 = (struct rx_cmp_ext *) c0c050c58d8409 Michael Chan 2015-10-22 2053 &cpr->cp_desc_ring[CP_RING(cp_cons)][CP_IDX(cp_cons)]; c0c050c58d8409 Michael Chan 2015-10-22 2054 c0c050c58d8409 Michael Chan 2015-10-22 2055 if (!RX_CMP_VALID(rxcmp1, tmp_raw_cons)) c0c050c58d8409 Michael Chan 2015-10-22 2056 return -EBUSY; c0c050c58d8409 Michael Chan 2015-10-22 2057 828affc27ed434 Michael Chan 2021-08-15 2058 /* The valid test of the entry must be done first before 828affc27ed434 Michael Chan 2021-08-15 2059 * reading any further. 828affc27ed434 Michael Chan 2021-08-15 2060 */ 828affc27ed434 Michael Chan 2021-08-15 2061 dma_rmb(); c0c050c58d8409 Michael Chan 2015-10-22 2062 prod = rxr->rx_prod; c0c050c58d8409 Michael Chan 2015-10-22 2063 a7445d69809fe3 Michael Chan 2023-12-01 2064 if (cmp_type == CMP_TYPE_RX_L2_TPA_START_CMP || a7445d69809fe3 Michael Chan 2023-12-01 2065 cmp_type == CMP_TYPE_RX_L2_TPA_START_V3_CMP) { 39b2e62be3704d Michael Chan 2023-12-01 2066 bnxt_tpa_start(bp, rxr, cmp_type, 39b2e62be3704d Michael Chan 2023-12-01 2067 (struct rx_tpa_start_cmp *)rxcmp, c0c050c58d8409 Michael Chan 2015-10-22 2068 (struct rx_tpa_start_cmp_ext *)rxcmp1); c0c050c58d8409 Michael Chan 2015-10-22 2069 4e5dbbda4c40a2 Michael Chan 2017-02-06 2070 *event |= BNXT_RX_EVENT; e7e70fa6784b48 Colin Ian King 2018-01-16 2071 goto next_rx_no_prod_no_len; c0c050c58d8409 Michael Chan 2015-10-22 2072 c0c050c58d8409 Michael Chan 2015-10-22 2073 } else if (cmp_type == CMP_TYPE_RX_L2_TPA_END_CMP) { e44758b78ae814 Michael Chan 2018-10-14 2074 skb = bnxt_tpa_end(bp, cpr, &tmp_raw_cons, c0c050c58d8409 Michael Chan 2015-10-22 2075 (struct rx_tpa_end_cmp *)rxcmp, 4e5dbbda4c40a2 Michael Chan 2017-02-06 2076 (struct rx_tpa_end_cmp_ext *)rxcmp1, event); c0c050c58d8409 Michael Chan 2015-10-22 2077 1fac4b2fdbccab Tobias Klauser 2017-09-26 2078 if (IS_ERR(skb)) c0c050c58d8409 Michael Chan 2015-10-22 2079 return -EBUSY; c0c050c58d8409 Michael Chan 2015-10-22 2080 c0c050c58d8409 Michael Chan 2015-10-22 2081 rc = -ENOMEM; c0c050c58d8409 Michael Chan 2015-10-22 2082 if (likely(skb)) { ee5c7fb3404724 Sathya Perla 2017-07-24 2083 bnxt_deliver_skb(bp, bnapi, skb); c0c050c58d8409 Michael Chan 2015-10-22 2084 rc = 1; c0c050c58d8409 Michael Chan 2015-10-22 2085 } 4e5dbbda4c40a2 Michael Chan 2017-02-06 2086 *event |= BNXT_RX_EVENT; e7e70fa6784b48 Colin Ian King 2018-01-16 2087 goto next_rx_no_prod_no_len; c0c050c58d8409 Michael Chan 2015-10-22 2088 } c0c050c58d8409 Michael Chan 2015-10-22 2089 c0c050c58d8409 Michael Chan 2015-10-22 2090 cons = rxcmp->rx_cmp_opaque; fa7e28127a5ad9 Michael Chan 2016-05-10 2091 if (unlikely(cons != rxr->rx_next_cons)) { bbd6f0a9481399 Michael Chan 2021-04-23 2092 int rc1 = bnxt_discard_rx(bp, cpr, &tmp_raw_cons, rxcmp); fa7e28127a5ad9 Michael Chan 2016-05-10 2093 1b5c8b63d6a4a2 Michael Chan 2020-10-04 2094 /* 0xffff is forced error, don't print it */ 1b5c8b63d6a4a2 Michael Chan 2020-10-04 2095 if (rxr->rx_next_cons != 0xffff) a1b0e4e684e9c3 Michael Chan 2019-04-08 2096 netdev_warn(bp->dev, "RX cons %x != expected cons %x\n", a1b0e4e684e9c3 Michael Chan 2019-04-08 2097 cons, rxr->rx_next_cons); fea2993aecd74d Jakub Kicinski 2023-07-19 2098 bnxt_sched_reset_rxr(bp, rxr); bbd6f0a9481399 Michael Chan 2021-04-23 2099 if (rc1) fa7e28127a5ad9 Michael Chan 2016-05-10 2100 return rc1; bbd6f0a9481399 Michael Chan 2021-04-23 2101 goto next_rx_no_prod_no_len; fa7e28127a5ad9 Michael Chan 2016-05-10 2102 } a1b0e4e684e9c3 Michael Chan 2019-04-08 2103 rx_buf = &rxr->rx_buf_ring[cons]; a1b0e4e684e9c3 Michael Chan 2019-04-08 2104 data = rx_buf->data; a1b0e4e684e9c3 Michael Chan 2019-04-08 2105 data_ptr = rx_buf->data_ptr; 6bb19474391d17 Michael Chan 2017-02-06 2106 prefetch(data_ptr); c0c050c58d8409 Michael Chan 2015-10-22 2107 c61fb99cae5195 Michael Chan 2017-02-06 2108 misc = le32_to_cpu(rxcmp->rx_cmp_misc_v1); c61fb99cae5195 Michael Chan 2017-02-06 2109 agg_bufs = (misc & RX_CMP_AGG_BUFS) >> RX_CMP_AGG_BUFS_SHIFT; c0c050c58d8409 Michael Chan 2015-10-22 2110 c0c050c58d8409 Michael Chan 2015-10-22 2111 if (agg_bufs) { c0c050c58d8409 Michael Chan 2015-10-22 2112 if (!bnxt_agg_bufs_valid(bp, cpr, agg_bufs, &tmp_raw_cons)) c0c050c58d8409 Michael Chan 2015-10-22 2113 return -EBUSY; c0c050c58d8409 Michael Chan 2015-10-22 2114 c0c050c58d8409 Michael Chan 2015-10-22 2115 cp_cons = NEXT_CMP(cp_cons); 4e5dbbda4c40a2 Michael Chan 2017-02-06 2116 *event |= BNXT_AGG_EVENT; c0c050c58d8409 Michael Chan 2015-10-22 2117 } 4e5dbbda4c40a2 Michael Chan 2017-02-06 2118 *event |= BNXT_RX_EVENT; c0c050c58d8409 Michael Chan 2015-10-22 2119 c0c050c58d8409 Michael Chan 2015-10-22 2120 rx_buf->data = NULL; c0c050c58d8409 Michael Chan 2015-10-22 2121 if (rxcmp1->rx_cmp_cfa_code_errors_v2 & RX_CMP_L2_ERRORS) { 8e44e96c6c8e8f Michael Chan 2019-04-08 2122 u32 rx_err = le32_to_cpu(rxcmp1->rx_cmp_cfa_code_errors_v2); 8e44e96c6c8e8f Michael Chan 2019-04-08 2123 c0c050c58d8409 Michael Chan 2015-10-22 2124 bnxt_reuse_rx_data(rxr, cons, data); c0c050c58d8409 Michael Chan 2015-10-22 2125 if (agg_bufs) 4a228a3a5e58e5 Michael Chan 2019-07-29 2126 bnxt_reuse_rx_agg_bufs(cpr, cp_cons, 0, agg_bufs, 4a228a3a5e58e5 Michael Chan 2019-07-29 2127 false); c0c050c58d8409 Michael Chan 2015-10-22 2128 c0c050c58d8409 Michael Chan 2015-10-22 2129 rc = -EIO; 8e44e96c6c8e8f Michael Chan 2019-04-08 2130 if (rx_err & RX_CMPL_ERRORS_BUFFER_ERROR_MASK) { a75fbb3aa47a62 Edwin Peer 2024-04-30 2131 bnapi->cp_ring.sw_stats->rx.rx_buf_errors++; 1c7fd6ee2fe4ec Randy Schacher 2023-11-20 2132 if (!(bp->flags & BNXT_FLAG_CHIP_P5_PLUS) && 8d4bd96b54dcb5 Michael Chan 2020-10-04 2133 !(bp->fw_cap & BNXT_FW_CAP_RING_MONITOR)) { 8fbf58e17dce8f Michael Chan 2020-10-04 2134 netdev_warn_once(bp->dev, "RX buffer error %x\n", 19b3751ffa713d Michael Chan 2019-11-18 2135 rx_err); fea2993aecd74d Jakub Kicinski 2023-07-19 2136 bnxt_sched_reset_rxr(bp, rxr); 8e44e96c6c8e8f Michael Chan 2019-04-08 2137 } 19b3751ffa713d Michael Chan 2019-11-18 2138 } 0b397b17a4120c Michael Chan 2019-04-25 2139 goto next_rx_no_len; c0c050c58d8409 Michael Chan 2015-10-22 2140 } c0c050c58d8409 Michael Chan 2015-10-22 2141 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2142 flags = le32_to_cpu(rxcmp->rx_cmp_len_flags_type); 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2143 len = flags >> RX_CMP_LEN_SHIFT; 11cd119d31a71b Michael Chan 2017-02-06 2144 dma_addr = rx_buf->mapping; c0c050c58d8409 Michael Chan 2015-10-22 2145 b231c3f3414cfc Andy Gospodarek 2022-04-08 2146 if (bnxt_xdp_attached(bp, rxr)) { bbfc17e50ba2ed Michael Chan 2022-12-26 2147 bnxt_xdp_buff_init(bp, rxr, cons, data_ptr, len, &xdp); 4c6c123c9af9c9 Andy Gospodarek 2022-04-08 2148 if (agg_bufs) { 4c6c123c9af9c9 Andy Gospodarek 2022-04-08 2149 u32 frag_len = bnxt_rx_agg_pages_xdp(bp, cpr, &xdp, 4c6c123c9af9c9 Andy Gospodarek 2022-04-08 2150 cp_cons, agg_bufs, 4c6c123c9af9c9 Andy Gospodarek 2022-04-08 2151 false); 73011773070999 Jakub Kicinski 2024-04-23 2152 if (!frag_len) 73011773070999 Jakub Kicinski 2024-04-23 2153 goto oom_next_rx; 4c6c123c9af9c9 Andy Gospodarek 2022-04-08 2154 } ee536dcbdce496 Andy Gospodarek 2022-04-08 2155 xdp_active = true; ee536dcbdce496 Andy Gospodarek 2022-04-08 2156 } ee536dcbdce496 Andy Gospodarek 2022-04-08 2157 9f4b28301ce6a5 Andy Gospodarek 2022-04-08 2158 if (xdp_active) { 1614f06e09ad6b Somnath Kotur 2024-04-02 2159 if (bnxt_rx_xdp(bp, rxr, cons, &xdp, data, &data_ptr, &len, event)) { c6d30e8391b85e Michael Chan 2017-02-06 2160 rc = 1; c6d30e8391b85e Michael Chan 2017-02-06 2161 goto next_rx; c6d30e8391b85e Michael Chan 2017-02-06 2162 } b231c3f3414cfc Andy Gospodarek 2022-04-08 2163 } ee536dcbdce496 Andy Gospodarek 2022-04-08 2164 c0c050c58d8409 Michael Chan 2015-10-22 2165 if (len <= bp->rx_copy_thresh) { 0ae1fafc8be6d4 Somnath Kotur 2024-04-02 2166 if (!xdp_active) 6bb19474391d17 Michael Chan 2017-02-06 2167 skb = bnxt_copy_skb(bnapi, data_ptr, len, dma_addr); 0ae1fafc8be6d4 Somnath Kotur 2024-04-02 2168 else 0ae1fafc8be6d4 Somnath Kotur 2024-04-02 2169 skb = bnxt_copy_xdp(bnapi, &xdp, len, dma_addr); c0c050c58d8409 Michael Chan 2015-10-22 2170 bnxt_reuse_rx_data(rxr, cons, data); c0c050c58d8409 Michael Chan 2015-10-22 2171 if (!skb) { a7559bc8c17c3f Andy Gospodarek 2022-04-08 2172 if (agg_bufs) { a7559bc8c17c3f Andy Gospodarek 2022-04-08 2173 if (!xdp_active) 4a228a3a5e58e5 Michael Chan 2019-07-29 2174 bnxt_reuse_rx_agg_bufs(cpr, cp_cons, 0, 4a228a3a5e58e5 Michael Chan 2019-07-29 2175 agg_bufs, false); a7559bc8c17c3f Andy Gospodarek 2022-04-08 2176 else a7559bc8c17c3f Andy Gospodarek 2022-04-08 2177 bnxt_xdp_buff_frags_free(rxr, &xdp); a7559bc8c17c3f Andy Gospodarek 2022-04-08 2178 } 73011773070999 Jakub Kicinski 2024-04-23 2179 goto oom_next_rx; c0c050c58d8409 Michael Chan 2015-10-22 2180 } c0c050c58d8409 Michael Chan 2015-10-22 2181 } else { c61fb99cae5195 Michael Chan 2017-02-06 2182 u32 payload; c61fb99cae5195 Michael Chan 2017-02-06 2183 c6d30e8391b85e Michael Chan 2017-02-06 2184 if (rx_buf->data_ptr == data_ptr) c61fb99cae5195 Michael Chan 2017-02-06 2185 payload = misc & RX_CMP_PAYLOAD_OFFSET; c6d30e8391b85e Michael Chan 2017-02-06 2186 else c6d30e8391b85e Michael Chan 2017-02-06 2187 payload = 0; 6bb19474391d17 Michael Chan 2017-02-06 2188 skb = bp->rx_skb_func(bp, rxr, cons, data, data_ptr, dma_addr, c61fb99cae5195 Michael Chan 2017-02-06 2189 payload | len); 73011773070999 Jakub Kicinski 2024-04-23 2190 if (!skb) 73011773070999 Jakub Kicinski 2024-04-23 2191 goto oom_next_rx; c0c050c58d8409 Michael Chan 2015-10-22 2192 } c0c050c58d8409 Michael Chan 2015-10-22 2193 c0c050c58d8409 Michael Chan 2015-10-22 2194 if (agg_bufs) { 32861236190bf1 Andy Gospodarek 2022-04-08 2195 if (!xdp_active) { 23e4c0469ad03f Andy Gospodarek 2022-04-08 2196 skb = bnxt_rx_agg_pages_skb(bp, cpr, skb, cp_cons, agg_bufs, false); 73011773070999 Jakub Kicinski 2024-04-23 2197 if (!skb) 73011773070999 Jakub Kicinski 2024-04-23 2198 goto oom_next_rx; 1dc4c557bfedfc Andy Gospodarek 2022-04-08 2199 } else { 1dc4c557bfedfc Andy Gospodarek 2022-04-08 2200 skb = bnxt_xdp_build_skb(bp, skb, agg_bufs, rxr->page_pool, &xdp, rxcmp1); 1dc4c557bfedfc Andy Gospodarek 2022-04-08 2201 if (!skb) { 1dc4c557bfedfc Andy Gospodarek 2022-04-08 2202 /* we should be able to free the old skb here */ a7559bc8c17c3f Andy Gospodarek 2022-04-08 2203 bnxt_xdp_buff_frags_free(rxr, &xdp); 73011773070999 Jakub Kicinski 2024-04-23 2204 goto oom_next_rx; 1dc4c557bfedfc Andy Gospodarek 2022-04-08 2205 } c0c050c58d8409 Michael Chan 2015-10-22 2206 } 32861236190bf1 Andy Gospodarek 2022-04-08 2207 } c0c050c58d8409 Michael Chan 2015-10-22 2208 c0c050c58d8409 Michael Chan 2015-10-22 2209 if (RX_CMP_HASH_VALID(rxcmp)) { a7445d69809fe3 Michael Chan 2023-12-01 2210 enum pkt_hash_types type; a7445d69809fe3 Michael Chan 2023-12-01 2211 a7445d69809fe3 Michael Chan 2023-12-01 2212 if (cmp_type == CMP_TYPE_RX_L2_V3_CMP) { a7445d69809fe3 Michael Chan 2023-12-01 2213 type = bnxt_rss_ext_op(bp, rxcmp); a7445d69809fe3 Michael Chan 2023-12-01 2214 } else { c0c050c58d8409 Michael Chan 2015-10-22 2215 u32 hash_type = RX_CMP_HASH_TYPE(rxcmp); c0c050c58d8409 Michael Chan 2015-10-22 2216 a7445d69809fe3 Michael Chan 2023-12-01 2217 /* RSS profiles 1 and 3 with extract code 0 for inner a7445d69809fe3 Michael Chan 2023-12-01 2218 * 4-tuple a7445d69809fe3 Michael Chan 2023-12-01 2219 */ c0c050c58d8409 Michael Chan 2015-10-22 2220 if (hash_type != 1 && hash_type != 3) c0c050c58d8409 Michael Chan 2015-10-22 2221 type = PKT_HASH_TYPE_L3; a7445d69809fe3 Michael Chan 2023-12-01 2222 else a7445d69809fe3 Michael Chan 2023-12-01 2223 type = PKT_HASH_TYPE_L4; a7445d69809fe3 Michael Chan 2023-12-01 2224 } c0c050c58d8409 Michael Chan 2015-10-22 2225 skb_set_hash(skb, le32_to_cpu(rxcmp->rx_cmp_rss_hash), type); c0c050c58d8409 Michael Chan 2015-10-22 2226 } c0c050c58d8409 Michael Chan 2015-10-22 2227 a7445d69809fe3 Michael Chan 2023-12-01 2228 if (cmp_type == CMP_TYPE_RX_L2_CMP) a7445d69809fe3 Michael Chan 2023-12-01 2229 dev = bnxt_get_pkt_dev(bp, RX_CMP_CFA_CODE(rxcmp1)); a7445d69809fe3 Michael Chan 2023-12-01 2230 skb->protocol = eth_type_trans(skb, dev); c0c050c58d8409 Michael Chan 2015-10-22 2231 c2f8063309da9b Michael Chan 2023-12-01 2232 if (skb->dev->features & BNXT_HW_FEATURE_VLAN_ALL_RX) { c2f8063309da9b Michael Chan 2023-12-01 2233 skb = bnxt_rx_vlan(skb, cmp_type, rxcmp, rxcmp1); c2f8063309da9b Michael Chan 2023-12-01 2234 if (!skb) 96bdd4b9ea7ef9 Michael Chan 2021-07-18 2235 goto next_rx; 96bdd4b9ea7ef9 Michael Chan 2021-07-18 2236 } c0c050c58d8409 Michael Chan 2015-10-22 2237 c0c050c58d8409 Michael Chan 2015-10-22 2238 skb_checksum_none_assert(skb); c0c050c58d8409 Michael Chan 2015-10-22 2239 if (RX_CMP_L4_CS_OK(rxcmp1)) { c0c050c58d8409 Michael Chan 2015-10-22 2240 if (dev->features & NETIF_F_RXCSUM) { c0c050c58d8409 Michael Chan 2015-10-22 2241 skb->ip_summed = CHECKSUM_UNNECESSARY; c0c050c58d8409 Michael Chan 2015-10-22 2242 skb->csum_level = RX_CMP_ENCAP(rxcmp1); c0c050c58d8409 Michael Chan 2015-10-22 2243 } c0c050c58d8409 Michael Chan 2015-10-22 2244 } else { 665e350ddbfde8 Satish Baddipadige 2015-12-27 2245 if (rxcmp1->rx_cmp_cfa_code_errors_v2 & RX_CMP_L4_CS_ERR_BITS) { 665e350ddbfde8 Satish Baddipadige 2015-12-27 2246 if (dev->features & NETIF_F_RXCSUM) a75fbb3aa47a62 Edwin Peer 2024-04-30 2247 bnapi->cp_ring.sw_stats->rx.rx_l4_csum_errors++; c0c050c58d8409 Michael Chan 2015-10-22 2248 } 665e350ddbfde8 Satish Baddipadige 2015-12-27 2249 } c0c050c58d8409 Michael Chan 2015-10-22 2250 c13e268c076865 Michael Chan 2023-12-07 2251 if (bnxt_rx_ts_valid(bp, flags, rxcmp1, &cmpl_ts)) { 1c7fd6ee2fe4ec Randy Schacher 2023-11-20 2252 if (bp->flags & BNXT_FLAG_CHIP_P5_PLUS) { 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2253 u64 ns, ts; 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2254 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2255 if (!bnxt_get_rx_ts_p5(bp, &ts, cmpl_ts)) { 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2256 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg; 4ab3e4983bcc9d Vadim Fedorenko 2024-10-16 @2257 unsigned long flags; 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2258 e3c23abd87fef3 Vadim Fedorenko 2024-10-25 2259 ns = bnxt_timecounter_cyc2time(ptp, ts); 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2260 memset(skb_hwtstamps(skb), 0, 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2261 sizeof(*skb_hwtstamps(skb))); 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2262 skb_hwtstamps(skb)->hwtstamp = ns_to_ktime(ns); 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2263 } 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2264 } 7f5515d19cd7aa Pavan Chebbi 2021-06-27 2265 } ee5c7fb3404724 Sathya Perla 2017-07-24 2266 bnxt_deliver_skb(bp, bnapi, skb); c0c050c58d8409 Michael Chan 2015-10-22 2267 rc = 1; c0c050c58d8409 Michael Chan 2015-10-22 2268 c0c050c58d8409 Michael Chan 2015-10-22 2269 next_rx: 6a8788f25625ea Andy Gospodarek 2018-01-09 2270 cpr->rx_packets += 1; 6a8788f25625ea Andy Gospodarek 2018-01-09 2271 cpr->rx_bytes += len; e7e70fa6784b48 Colin Ian King 2018-01-16 2272 0b397b17a4120c Michael Chan 2019-04-25 2273 next_rx_no_len: 0b397b17a4120c Michael Chan 2019-04-25 2274 rxr->rx_prod = NEXT_RX(prod); c09d22674b9420 Michael Chan 2023-11-20 2275 rxr->rx_next_cons = RING_RX(bp, NEXT_RX(cons)); 0b397b17a4120c Michael Chan 2019-04-25 2276 e7e70fa6784b48 Colin Ian King 2018-01-16 2277 next_rx_no_prod_no_len: c0c050c58d8409 Michael Chan 2015-10-22 2278 *raw_cons = tmp_raw_cons; c0c050c58d8409 Michael Chan 2015-10-22 2279 c0c050c58d8409 Michael Chan 2015-10-22 2280 return rc; 73011773070999 Jakub Kicinski 2024-04-23 2281 73011773070999 Jakub Kicinski 2024-04-23 2282 oom_next_rx: a75fbb3aa47a62 Edwin Peer 2024-04-30 2283 cpr->sw_stats->rx.rx_oom_discards += 1; 73011773070999 Jakub Kicinski 2024-04-23 2284 rc = -ENOMEM; 73011773070999 Jakub Kicinski 2024-04-23 2285 goto next_rx; c0c050c58d8409 Michael Chan 2015-10-22 2286 } c0c050c58d8409 Michael Chan 2015-10-22 2287 -- 0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter 2024-10-25 19:47 [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter Vadim Fedorenko 2024-10-25 19:47 ` [PATCH net-next v2 2/2] bnxt_en: replace PTP spinlock with seqlock Vadim Fedorenko @ 2024-10-25 21:31 ` Michael Chan 2024-10-25 21:53 ` Vadim Fedorenko 1 sibling, 1 reply; 8+ messages in thread From: Michael Chan @ 2024-10-25 21:31 UTC (permalink / raw) To: Vadim Fedorenko Cc: Vadim Fedorenko, Pavan Chebbi, Jakub Kicinski, Andrew Lunn, Paolo Abeni, David S. Miller, netdev, Richard Cochran [-- Attachment #1: Type: text/plain, Size: 796 bytes --] On Fri, Oct 25, 2024 at 12:48 PM Vadim Fedorenko <vadfed@meta.com> wrote: > > This hardware can provide only 48 bits of cycle counter. We can leave > only 24 bits in the cache to extend RX timestamps from 32 bits to 48 > bits. This make cache writes atomic even on 32 bit platforms and we can > simply use READ_ONCE()/WRITE_ONCE() pair and remove spinlock. The > configuration structure will be also reduced by 4 bytes. ptp->old_time serves 2 purposes: to cache the upper 16 bits of the HW counter and for rollover check. With this patch reducing ptp->old_time to 24 bits, we now use the upper 16 bits for the cache and the next 8 bits for the rollover check. I think this will work. But since the field is now 32-bit, why not use the full 32 bits instead of 24 bits? Thanks. [-- Attachment #2: S/MIME Cryptographic Signature --] [-- Type: application/pkcs7-signature, Size: 4209 bytes --] ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter 2024-10-25 21:31 ` [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter Michael Chan @ 2024-10-25 21:53 ` Vadim Fedorenko 2024-10-25 22:13 ` Michael Chan 0 siblings, 1 reply; 8+ messages in thread From: Vadim Fedorenko @ 2024-10-25 21:53 UTC (permalink / raw) To: Michael Chan Cc: Pavan Chebbi, Jakub Kicinski, Andrew Lunn, Paolo Abeni, David S. Miller, netdev, Richard Cochran On 25/10/2024 22:31, Michael Chan wrote: > On Fri, Oct 25, 2024 at 12:48 PM Vadim Fedorenko <vadfed@meta.com> wrote: >> >> This hardware can provide only 48 bits of cycle counter. We can leave >> only 24 bits in the cache to extend RX timestamps from 32 bits to 48 >> bits. This make cache writes atomic even on 32 bit platforms and we can >> simply use READ_ONCE()/WRITE_ONCE() pair and remove spinlock. The >> configuration structure will be also reduced by 4 bytes. > > ptp->old_time serves 2 purposes: to cache the upper 16 bits of the HW > counter and for rollover check. With this patch reducing > ptp->old_time to 24 bits, we now use the upper 16 bits for the cache > and the next 8 bits for the rollover check. I think this will work. > But since the field is now 32-bit, why not use the full 32 bits > instead of 24 bits? Thanks. As you confirmed that the HW has 48 bits of cycle counter, we have to cache 16 bits only. The other 8 bits are used for the rollover check. We can even use less bits for it. What is the use for another 8 bits? With this patch the rollover will happen every ~16 ms, but will be caught even till ~4s. If we will cache upper 32 bits, the rollover will happen every 64us and we will have to update cache value more often. But at the same time it will not bring any value, because the upper limit will still be ~4s. That's why I don't see any benefits of using all 32 bits. ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter 2024-10-25 21:53 ` Vadim Fedorenko @ 2024-10-25 22:13 ` Michael Chan 2024-10-26 22:20 ` Vadim Fedorenko 0 siblings, 1 reply; 8+ messages in thread From: Michael Chan @ 2024-10-25 22:13 UTC (permalink / raw) To: Vadim Fedorenko Cc: Pavan Chebbi, Jakub Kicinski, Andrew Lunn, Paolo Abeni, David S. Miller, netdev, Richard Cochran [-- Attachment #1: Type: text/plain, Size: 1801 bytes --] On Fri, Oct 25, 2024 at 2:53 PM Vadim Fedorenko <vadim.fedorenko@linux.dev> wrote: > > On 25/10/2024 22:31, Michael Chan wrote: > > On Fri, Oct 25, 2024 at 12:48 PM Vadim Fedorenko <vadfed@meta.com> wrote: > >> > >> This hardware can provide only 48 bits of cycle counter. We can leave > >> only 24 bits in the cache to extend RX timestamps from 32 bits to 48 > >> bits. This make cache writes atomic even on 32 bit platforms and we can > >> simply use READ_ONCE()/WRITE_ONCE() pair and remove spinlock. The > >> configuration structure will be also reduced by 4 bytes. > > > > ptp->old_time serves 2 purposes: to cache the upper 16 bits of the HW > > counter and for rollover check. With this patch reducing > > ptp->old_time to 24 bits, we now use the upper 16 bits for the cache > > and the next 8 bits for the rollover check. I think this will work. > > But since the field is now 32-bit, why not use the full 32 bits > > instead of 24 bits? Thanks. > > As you confirmed that the HW has 48 bits of cycle counter, we have to > cache 16 bits only. The other 8 bits are used for the rollover check. We > can even use less bits for it. What is the use for another 8 bits? With > this patch the rollover will happen every ~16 ms, but will be caught > even till ~4s. If we will cache upper 32 bits, the rollover will happen > every 64us and we will have to update cache value more often. But at the > same time it will not bring any value, because the upper limit will > still be ~4s. That's why I don't see any benefits of using all 32 bits. I agree the extra bits have no value other than to fill the 32-bit field. But it should not affect the frequency of caching ptp->old_time. It should be updated in bnxt_ptp_ts_aux_work() at a fixed interval (1 second). [-- Attachment #2: S/MIME Cryptographic Signature --] [-- Type: application/pkcs7-signature, Size: 4209 bytes --] ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter 2024-10-25 22:13 ` Michael Chan @ 2024-10-26 22:20 ` Vadim Fedorenko 2024-10-28 16:18 ` Michael Chan 0 siblings, 1 reply; 8+ messages in thread From: Vadim Fedorenko @ 2024-10-26 22:20 UTC (permalink / raw) To: Michael Chan Cc: Pavan Chebbi, Jakub Kicinski, Andrew Lunn, Paolo Abeni, David S. Miller, netdev, Richard Cochran On 25/10/2024 23:13, Michael Chan wrote: > On Fri, Oct 25, 2024 at 2:53 PM Vadim Fedorenko > <vadim.fedorenko@linux.dev> wrote: >> >> On 25/10/2024 22:31, Michael Chan wrote: >>> On Fri, Oct 25, 2024 at 12:48 PM Vadim Fedorenko <vadfed@meta.com> wrote: >>>> >>>> This hardware can provide only 48 bits of cycle counter. We can leave >>>> only 24 bits in the cache to extend RX timestamps from 32 bits to 48 >>>> bits. This make cache writes atomic even on 32 bit platforms and we can >>>> simply use READ_ONCE()/WRITE_ONCE() pair and remove spinlock. The >>>> configuration structure will be also reduced by 4 bytes. >>> >>> ptp->old_time serves 2 purposes: to cache the upper 16 bits of the HW >>> counter and for rollover check. With this patch reducing >>> ptp->old_time to 24 bits, we now use the upper 16 bits for the cache >>> and the next 8 bits for the rollover check. I think this will work. >>> But since the field is now 32-bit, why not use the full 32 bits >>> instead of 24 bits? Thanks. >> >> As you confirmed that the HW has 48 bits of cycle counter, we have to >> cache 16 bits only. The other 8 bits are used for the rollover check. We >> can even use less bits for it. What is the use for another 8 bits? With >> this patch the rollover will happen every ~16 ms, but will be caught >> even till ~4s. If we will cache upper 32 bits, the rollover will happen >> every 64us and we will have to update cache value more often. But at the >> same time it will not bring any value, because the upper limit will >> still be ~4s. That's why I don't see any benefits of using all 32 bits. > > I agree the extra bits have no value other than to fill the 32-bit > field. But it should not affect the frequency of caching > ptp->old_time. It should be updated in bnxt_ptp_ts_aux_work() at a > fixed interval (1 second). Ok, BNXT_LO_TIMER_MASK/BNXT_HI_TIMER_MASK use 24 bits only. I don't see any reason to keep more bits out of 48 bit of counter and I tried to be consistent across the code. It doesn't matter in terms of performance should we shift 16 or 24 bits. But because there will be another version (I forgot to remove one variable), I can change the code to fill 32-bit field if you insist. ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter 2024-10-26 22:20 ` Vadim Fedorenko @ 2024-10-28 16:18 ` Michael Chan 0 siblings, 0 replies; 8+ messages in thread From: Michael Chan @ 2024-10-28 16:18 UTC (permalink / raw) To: Vadim Fedorenko Cc: Pavan Chebbi, Jakub Kicinski, Andrew Lunn, Paolo Abeni, David S. Miller, netdev, Richard Cochran [-- Attachment #1: Type: text/plain, Size: 588 bytes --] On Sat, Oct 26, 2024 at 3:20 PM Vadim Fedorenko <vadim.fedorenko@linux.dev> wrote: > Ok, BNXT_LO_TIMER_MASK/BNXT_HI_TIMER_MASK use 24 bits only. I don't see > any reason to keep more bits out of 48 bit of counter and I tried to be > consistent across the code. It doesn't matter in terms of performance > should we shift 16 or 24 bits. But because there will be another version > (I forgot to remove one variable), I can change the code to fill 32-bit > field if you insist. 24-bit will be fine. but please document that 8 bits will be used for roll-over check. Thanks. [-- Attachment #2: S/MIME Cryptographic Signature --] [-- Type: application/pkcs7-signature, Size: 4209 bytes --] ^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2024-10-28 16:18 UTC | newest] Thread overview: 8+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2024-10-25 19:47 [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter Vadim Fedorenko 2024-10-25 19:47 ` [PATCH net-next v2 2/2] bnxt_en: replace PTP spinlock with seqlock Vadim Fedorenko 2024-10-26 11:23 ` kernel test robot 2024-10-25 21:31 ` [PATCH net-next v2 1/2] bnxt_en: cache only 24 bits of hw counter Michael Chan 2024-10-25 21:53 ` Vadim Fedorenko 2024-10-25 22:13 ` Michael Chan 2024-10-26 22:20 ` Vadim Fedorenko 2024-10-28 16:18 ` Michael Chan
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.