diff options
Diffstat (limited to 'drivers/net/ethernet/intel/ixgbe/ixgbe_ptp.c')
| -rw-r--r-- | drivers/net/ethernet/intel/ixgbe/ixgbe_ptp.c | 718 |
1 files changed, 530 insertions, 188 deletions
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_ptp.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_ptp.c index e5ba04025e2b..ef1504d41890 100644 --- a/drivers/net/ethernet/intel/ixgbe/ixgbe_ptp.c +++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_ptp.c @@ -1,7 +1,7 @@ /******************************************************************************* Intel 10 Gigabit PCI Express Linux driver - Copyright(c) 1999 - 2013 Intel Corporation. + Copyright(c) 1999 - 2015 Intel Corporation. This program is free software; you can redistribute it and/or modify it under the terms and conditions of the GNU General Public License, @@ -27,6 +27,7 @@ *******************************************************************************/ #include "ixgbe.h" #include <linux/ptp_classify.h> +#include <linux/clocksource.h> /* * The 82599 and the X540 do not have true 64bit nanosecond scale @@ -93,7 +94,6 @@ #define IXGBE_INCVAL_SHIFT_82599 7 #define IXGBE_INCPER_SHIFT_82599 24 -#define IXGBE_MAX_TIMEADJ_VALUE 0x7FFFFFFFFFFFFFFFULL #define IXGBE_OVERFLOW_PERIOD (HZ * 30) #define IXGBE_PTP_TX_TIMEOUT (HZ * 15) @@ -104,8 +104,68 @@ */ #define IXGBE_PTP_PPS_HALF_SECOND 500000000ULL +/* In contrast, the X550 controller has two registers, SYSTIMEH and SYSTIMEL + * which contain measurements of seconds and nanoseconds respectively. This + * matches the standard linux representation of time in the kernel. In addition, + * the X550 also has a SYSTIMER register which represents residue, or + * subnanosecond overflow adjustments. To control clock adjustment, the TIMINCA + * register is used, but it is unlike the X540 and 82599 devices. TIMINCA + * represents units of 2^-32 nanoseconds, and uses 31 bits for this, with the + * high bit representing whether the adjustent is positive or negative. Every + * clock cycle, the X550 will add 12.5 ns + TIMINCA which can result in a range + * of 12 to 13 nanoseconds adjustment. Unlike the 82599 and X540 devices, the + * X550's clock for purposes of SYSTIME generation is constant and not dependent + * on the link speed. + * + * SYSTIMEH SYSTIMEL SYSTIMER + * +--------------+ +--------------+ +-------------+ + * X550 | 32 | | 32 | | 32 | + * *--------------+ +--------------+ +-------------+ + * \____seconds___/ \_nanoseconds_/ \__2^-32 ns__/ + * + * This results in a full 96 bits to represent the clock, with 32 bits for + * seconds, 32 bits for nanoseconds (largest value is 0d999999999 or just under + * 1 second) and an additional 32 bits to measure sub nanosecond adjustments for + * underflow of adjustments. + * + * The 32 bits of seconds for the X550 overflows every + * 2^32 / ( 365.25 * 24 * 60 * 60 ) = ~136 years. + * + * In order to adjust the clock frequency for the X550, the TIMINCA register is + * provided. This register represents a + or minus nearly 0.5 ns adjustment to + * the base frequency. It is measured in 2^-32 ns units, with the high bit being + * the sign bit. This register enables software to calculate frequency + * adjustments and apply them directly to the clock rate. + * + * The math for converting ppb into TIMINCA values is fairly straightforward. + * TIMINCA value = ( Base_Frequency * ppb ) / 1000000000ULL + * + * This assumes that ppb is never high enough to create a value bigger than + * TIMINCA's 31 bits can store. This is ensured by the stack. Calculating this + * value is also simple. + * Max ppb = ( Max Adjustment / Base Frequency ) / 1000000000ULL + * + * For the X550, the Max adjustment is +/- 0.5 ns, and the base frequency is + * 12.5 nanoseconds. This means that the Max ppb is 39999999 + * Note: We subtract one in order to ensure no overflow, because the TIMINCA + * register can only hold slightly under 0.5 nanoseconds. + * + * Because TIMINCA is measured in 2^-32 ns units, we have to convert 12.5 ns + * into 2^-32 units, which is + * + * 12.5 * 2^32 = C80000000 + * + * Some revisions of hardware have a faster base frequency than the registers + * were defined for. To fix this, we use a timecounter structure with the + * proper mult and shift to convert the cycles into nanoseconds of time. + */ +#define IXGBE_X550_BASE_PERIOD 0xC80000000ULL +#define INCVALUE_MASK 0x7FFFFFFF +#define ISGN 0x80000000 +#define MAX_TIMADJ 0x7FFFFFFF + /** - * ixgbe_ptp_setup_sdp + * ixgbe_ptp_setup_sdp_x540 * @hw: the hardware private structure * * this function enables or disables the clock out feature on SDP0 for @@ -116,83 +176,116 @@ * aligns the start of the PPS signal to that value. The shift is * necessary because it can change based on the link speed. */ -static void ixgbe_ptp_setup_sdp(struct ixgbe_adapter *adapter) +static void ixgbe_ptp_setup_sdp_x540(struct ixgbe_adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; - int shift = adapter->cc.shift; + int shift = adapter->hw_cc.shift; u32 esdp, tsauxc, clktiml, clktimh, trgttiml, trgttimh, rem; u64 ns = 0, clock_edge = 0; - if ((adapter->flags2 & IXGBE_FLAG2_PTP_PPS_ENABLED) && - (hw->mac.type == ixgbe_mac_X540)) { + /* disable the pin first */ + IXGBE_WRITE_REG(hw, IXGBE_TSAUXC, 0x0); + IXGBE_WRITE_FLUSH(hw); - /* disable the pin first */ - IXGBE_WRITE_REG(hw, IXGBE_TSAUXC, 0x0); - IXGBE_WRITE_FLUSH(hw); + if (!(adapter->flags2 & IXGBE_FLAG2_PTP_PPS_ENABLED)) + return; - esdp = IXGBE_READ_REG(hw, IXGBE_ESDP); + esdp = IXGBE_READ_REG(hw, IXGBE_ESDP); - /* - * enable the SDP0 pin as output, and connected to the - * native function for Timesync (ClockOut) - */ - esdp |= (IXGBE_ESDP_SDP0_DIR | - IXGBE_ESDP_SDP0_NATIVE); + /* enable the SDP0 pin as output, and connected to the + * native function for Timesync (ClockOut) + */ + esdp |= IXGBE_ESDP_SDP0_DIR | + IXGBE_ESDP_SDP0_NATIVE; - /* - * enable the Clock Out feature on SDP0, and allow - * interrupts to occur when the pin changes - */ - tsauxc = (IXGBE_TSAUXC_EN_CLK | - IXGBE_TSAUXC_SYNCLK | - IXGBE_TSAUXC_SDP0_INT); + /* enable the Clock Out feature on SDP0, and allow + * interrupts to occur when the pin changes + */ + tsauxc = IXGBE_TSAUXC_EN_CLK | + IXGBE_TSAUXC_SYNCLK | + IXGBE_TSAUXC_SDP0_INT; - /* clock period (or pulse length) */ - clktiml = (u32)(IXGBE_PTP_PPS_HALF_SECOND << shift); - clktimh = (u32)((IXGBE_PTP_PPS_HALF_SECOND << shift) >> 32); + /* clock period (or pulse length) */ + clktiml = (u32)(IXGBE_PTP_PPS_HALF_SECOND << shift); + clktimh = (u32)((IXGBE_PTP_PPS_HALF_SECOND << shift) >> 32); - /* - * Account for the cyclecounter wrap-around value by - * using the converted ns value of the current time to - * check for when the next aligned second would occur. - */ - clock_edge |= (u64)IXGBE_READ_REG(hw, IXGBE_SYSTIML); - clock_edge |= (u64)IXGBE_READ_REG(hw, IXGBE_SYSTIMH) << 32; - ns = timecounter_cyc2time(&adapter->tc, clock_edge); + /* Account for the cyclecounter wrap-around value by + * using the converted ns value of the current time to + * check for when the next aligned second would occur. + */ + clock_edge |= (u64)IXGBE_READ_REG(hw, IXGBE_SYSTIML); + clock_edge |= (u64)IXGBE_READ_REG(hw, IXGBE_SYSTIMH) << 32; + ns = timecounter_cyc2time(&adapter->hw_tc, clock_edge); - div_u64_rem(ns, IXGBE_PTP_PPS_HALF_SECOND, &rem); - clock_edge += ((IXGBE_PTP_PPS_HALF_SECOND - (u64)rem) << shift); + div_u64_rem(ns, IXGBE_PTP_PPS_HALF_SECOND, &rem); + clock_edge += ((IXGBE_PTP_PPS_HALF_SECOND - (u64)rem) << shift); - /* specify the initial clock start time */ - trgttiml = (u32)clock_edge; - trgttimh = (u32)(clock_edge >> 32); + /* specify the initial clock start time */ + trgttiml = (u32)clock_edge; + trgttimh = (u32)(clock_edge >> 32); - IXGBE_WRITE_REG(hw, IXGBE_CLKTIML, clktiml); - IXGBE_WRITE_REG(hw, IXGBE_CLKTIMH, clktimh); - IXGBE_WRITE_REG(hw, IXGBE_TRGTTIML0, trgttiml); - IXGBE_WRITE_REG(hw, IXGBE_TRGTTIMH0, trgttimh); + IXGBE_WRITE_REG(hw, IXGBE_CLKTIML, clktiml); + IXGBE_WRITE_REG(hw, IXGBE_CLKTIMH, clktimh); + IXGBE_WRITE_REG(hw, IXGBE_TRGTTIML0, trgttiml); + IXGBE_WRITE_REG(hw, IXGBE_TRGTTIMH0, trgttimh); - IXGBE_WRITE_REG(hw, IXGBE_ESDP, esdp); - IXGBE_WRITE_REG(hw, IXGBE_TSAUXC, tsauxc); - } else { - IXGBE_WRITE_REG(hw, IXGBE_TSAUXC, 0x0); - } + IXGBE_WRITE_REG(hw, IXGBE_ESDP, esdp); + IXGBE_WRITE_REG(hw, IXGBE_TSAUXC, tsauxc); IXGBE_WRITE_FLUSH(hw); } /** - * ixgbe_ptp_read - read raw cycle counter (to be used by time counter) + * ixgbe_ptp_read_X550 - read cycle counter value + * @hw_cc: cyclecounter structure + * + * This function reads SYSTIME registers. It is called by the cyclecounter + * structure to convert from internal representation into nanoseconds. We need + * this for X550 since some skews do not have expected clock frequency and + * result of SYSTIME is 32bits of "billions of cycles" and 32 bits of + * "cycles", rather than seconds and nanoseconds. + */ +static cycle_t ixgbe_ptp_read_X550(const struct cyclecounter *hw_cc) +{ + struct ixgbe_adapter *adapter = + container_of(hw_cc, struct ixgbe_adapter, hw_cc); + struct ixgbe_hw *hw = &adapter->hw; + struct timespec64 ts; + + /* storage is 32 bits of 'billions of cycles' and 32 bits of 'cycles'. + * Some revisions of hardware run at a higher frequency and so the + * cycles are not guaranteed to be nanoseconds. The timespec64 created + * here is used for its math/conversions but does not necessarily + * represent nominal time. + * + * It should be noted that this cyclecounter will overflow at a + * non-bitmask field since we have to convert our billions of cycles + * into an actual cycles count. This results in some possible weird + * situations at high cycle counter stamps. However given that 32 bits + * of "seconds" is ~138 years this isn't a problem. Even at the + * increased frequency of some revisions, this is still ~103 years. + * Since the SYSTIME values start at 0 and we never write them, it is + * highly unlikely for the cyclecounter to overflow in practice. + */ + IXGBE_READ_REG(hw, IXGBE_SYSTIMR); + ts.tv_nsec = IXGBE_READ_REG(hw, IXGBE_SYSTIML); + ts.tv_sec = IXGBE_READ_REG(hw, IXGBE_SYSTIMH); + + return (u64)timespec64_to_ns(&ts); +} + +/** + * ixgbe_ptp_read_82599 - read raw cycle counter (to be used by time counter) * @cc: the cyclecounter structure * * this function reads the cyclecounter registers and is called by the * cyclecounter structure used to construct a ns counter from the * arbitrary fixed point registers */ -static cycle_t ixgbe_ptp_read(const struct cyclecounter *cc) +static cycle_t ixgbe_ptp_read_82599(const struct cyclecounter *cc) { struct ixgbe_adapter *adapter = - container_of(cc, struct ixgbe_adapter, cc); + container_of(cc, struct ixgbe_adapter, hw_cc); struct ixgbe_hw *hw = &adapter->hw; u64 stamp = 0; @@ -203,20 +296,79 @@ static cycle_t ixgbe_ptp_read(const struct cyclecounter *cc) } /** - * ixgbe_ptp_adjfreq + * ixgbe_ptp_convert_to_hwtstamp - convert register value to hw timestamp + * @adapter: private adapter structure + * @hwtstamp: stack timestamp structure + * @systim: unsigned 64bit system time value + * + * We need to convert the adapter's RX/TXSTMP registers into a hwtstamp value + * which can be used by the stack's ptp functions. + * + * The lock is used to protect consistency of the cyclecounter and the SYSTIME + * registers. However, it does not need to protect against the Rx or Tx + * timestamp registers, as there can't be a new timestamp until the old one is + * unlatched by reading. + * + * In addition to the timestamp in hardware, some controllers need a software + * overflow cyclecounter, and this function takes this into account as well. + **/ +static void ixgbe_ptp_convert_to_hwtstamp(struct ixgbe_adapter *adapter, + struct skb_shared_hwtstamps *hwtstamp, + u64 timestamp) +{ + unsigned long flags; + struct timespec64 systime; + u64 ns; + + memset(hwtstamp, 0, sizeof(*hwtstamp)); + + switch (adapter->hw.mac.type) { + /* X550 and later hardware supposedly represent time using a seconds + * and nanoseconds counter, instead of raw 64bits nanoseconds. We need + * to convert the timestamp into cycles before it can be fed to the + * cyclecounter. We need an actual cyclecounter because some revisions + * of hardware run at a higher frequency and thus the counter does + * not represent seconds/nanoseconds. Instead it can be thought of as + * cycles and billions of cycles. + */ + case ixgbe_mac_X550: + case ixgbe_mac_X550EM_x: + /* Upper 32 bits represent billions of cycles, lower 32 bits + * represent cycles. However, we use timespec64_to_ns for the + * correct math even though the units haven't been corrected + * yet. + */ + systime.tv_sec = timestamp >> 32; + systime.tv_nsec = timestamp & 0xFFFFFFFF; + + timestamp = timespec64_to_ns(&systime); + break; + default: + break; + } + + spin_lock_irqsave(&adapter->tmreg_lock, flags); + ns = timecounter_cyc2time(&adapter->hw_tc, timestamp); + spin_unlock_irqrestore(&adapter->tmreg_lock, flags); + + hwtstamp->hwtstamp = ns_to_ktime(ns); +} + +/** + * ixgbe_ptp_adjfreq_82599 * @ptp: the ptp clock structure * @ppb: parts per billion adjustment from base * * adjust the frequency of the ptp cycle counter by the * indicated ppb from the base frequency. */ -static int ixgbe_ptp_adjfreq(struct ptp_clock_info *ptp, s32 ppb) +static int ixgbe_ptp_adjfreq_82599(struct ptp_clock_info *ptp, s32 ppb) { struct ixgbe_adapter *adapter = container_of(ptp, struct ixgbe_adapter, ptp_caps); struct ixgbe_hw *hw = &adapter->hw; - u64 freq; - u32 diff, incval; + u64 freq, incval; + u32 diff; int neg_adj = 0; if (ppb < 0) { @@ -235,12 +387,16 @@ static int ixgbe_ptp_adjfreq(struct ptp_clock_info *ptp, s32 ppb) switch (hw->mac.type) { case ixgbe_mac_X540: - IXGBE_WRITE_REG(hw, IXGBE_TIMINCA, incval); + if (incval > 0xFFFFFFFFULL) + e_dev_warn("PTP ppb adjusted SYSTIME rate overflowed!\n"); + IXGBE_WRITE_REG(hw, IXGBE_TIMINCA, (u32)incval); break; case ixgbe_mac_82599EB: + if (incval > 0x00FFFFFFULL) + e_dev_warn("PTP ppb adjusted SYSTIME rate overflowed!\n"); IXGBE_WRITE_REG(hw, IXGBE_TIMINCA, (1 << IXGBE_INCPER_SHIFT_82599) | - incval); + ((u32)incval & 0x00FFFFFFUL)); break; default: break; @@ -250,6 +406,43 @@ static int ixgbe_ptp_adjfreq(struct ptp_clock_info *ptp, s32 ppb) } /** + * ixgbe_ptp_adjfreq_X550 + * @ptp: the ptp clock structure + * @ppb: parts per billion adjustment from base + * + * adjust the frequency of the SYSTIME registers by the indicated ppb from base + * frequency + */ +static int ixgbe_ptp_adjfreq_X550(struct ptp_clock_info *ptp, s32 ppb) +{ + struct ixgbe_adapter *adapter = + container_of(ptp, struct ixgbe_adapter, ptp_caps); + struct ixgbe_hw *hw = &adapter->hw; + int neg_adj = 0; + u64 rate = IXGBE_X550_BASE_PERIOD; + u32 inca; + + if (ppb < 0) { + neg_adj = 1; + ppb = -ppb; + } + rate *= ppb; + rate = div_u64(rate, 1000000000ULL); + + /* warn if rate is too large */ + if (rate >= INCVALUE_MASK) + e_dev_warn("PTP ppb adjusted SYSTIME rate overflowed!\n"); + + inca = rate & INCVALUE_MASK; + if (neg_adj) + inca |= ISGN; + + IXGBE_WRITE_REG(hw, IXGBE_TIMINCA, inca); + + return 0; +} + +/** * ixgbe_ptp_adjtime * @ptp: the ptp clock structure * @delta: offset to adjust the cycle counter by @@ -263,10 +456,11 @@ static int ixgbe_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta) unsigned long flags; spin_lock_irqsave(&adapter->tmreg_lock, flags); - timecounter_adjtime(&adapter->tc, delta); + timecounter_adjtime(&adapter->hw_tc, delta); spin_unlock_irqrestore(&adapter->tmreg_lock, flags); - ixgbe_ptp_setup_sdp(adapter); + if (adapter->ptp_setup_sdp) + adapter->ptp_setup_sdp(adapter); return 0; } @@ -283,11 +477,11 @@ static int ixgbe_ptp_gettime(struct ptp_clock_info *ptp, struct timespec64 *ts) { struct ixgbe_adapter *adapter = container_of(ptp, struct ixgbe_adapter, ptp_caps); - u64 ns; unsigned long flags; + u64 ns; spin_lock_irqsave(&adapter->tmreg_lock, flags); - ns = timecounter_read(&adapter->tc); + ns = timecounter_read(&adapter->hw_tc); spin_unlock_irqrestore(&adapter->tmreg_lock, flags); *ts = ns_to_timespec64(ns); @@ -308,17 +502,16 @@ static int ixgbe_ptp_settime(struct ptp_clock_info *ptp, { struct ixgbe_adapter *adapter = container_of(ptp, struct ixgbe_adapter, ptp_caps); - u64 ns; unsigned long flags; - - ns = timespec64_to_ns(ts); + u64 ns = timespec64_to_ns(ts); /* reset the timecounter */ spin_lock_irqsave(&adapter->tmreg_lock, flags); - timecounter_init(&adapter->tc, &adapter->cc, ns); + timecounter_init(&adapter->hw_tc, &adapter->hw_cc, ns); spin_unlock_irqrestore(&adapter->tmreg_lock, flags); - ixgbe_ptp_setup_sdp(adapter); + if (adapter->ptp_setup_sdp) + adapter->ptp_setup_sdp(adapter); return 0; } @@ -343,33 +536,26 @@ static int ixgbe_ptp_feature_enable(struct ptp_clock_info *ptp, * event when the clock SDP triggers. Clear mask when PPS is * disabled */ - if (rq->type == PTP_CLK_REQ_PPS) { - switch (adapter->hw.mac.type) { - case ixgbe_mac_X540: - if (on) - adapter->flags2 |= IXGBE_FLAG2_PTP_PPS_ENABLED; - else - adapter->flags2 &= ~IXGBE_FLAG2_PTP_PPS_ENABLED; + if (rq->type != PTP_CLK_REQ_PPS || !adapter->ptp_setup_sdp) + return -ENOTSUPP; - ixgbe_ptp_setup_sdp(adapter); - return 0; - default: - break; - } - } + if (on) + adapter->flags2 |= IXGBE_FLAG2_PTP_PPS_ENABLED; + else + adapter->flags2 &= ~IXGBE_FLAG2_PTP_PPS_ENABLED; - return -ENOTSUPP; + adapter->ptp_setup_sdp(adapter); + return 0; } /** * ixgbe_ptp_check_pps_event * @adapter: the private adapter structure - * @eicr: the interrupt cause register value * * This function is called by the interrupt routine when checking for * interrupts. It will check and handle a pps event. */ -void ixgbe_ptp_check_pps_event(struct ixgbe_adapter *adapter, u32 eicr) +void ixgbe_ptp_check_pps_event(struct ixgbe_adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; struct ptp_clock_event event; @@ -425,7 +611,9 @@ void ixgbe_ptp_rx_hang(struct ixgbe_adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; u32 tsyncrxctl = IXGBE_READ_REG(hw, IXGBE_TSYNCRXCTL); + struct ixgbe_ring *rx_ring; unsigned long rx_event; + int n; /* if we don't have a valid timestamp in the registers, just update the * timeout counter and exit @@ -437,19 +625,43 @@ void ixgbe_ptp_rx_hang(struct ixgbe_adapter *adapter) /* determine the most recent watchdog or rx_timestamp event */ rx_event = adapter->last_rx_ptp_check; - if (time_after(adapter->last_rx_timestamp, rx_event)) - rx_event = adapter->last_rx_timestamp; + for (n = 0; n < adapter->num_rx_queues; n++) { + rx_ring = adapter->rx_ring[n]; + if (time_after(rx_ring->last_rx_timestamp, rx_event)) + rx_event = rx_ring->last_rx_timestamp; + } /* only need to read the high RXSTMP register to clear the lock */ - if (time_is_before_jiffies(rx_event + 5*HZ)) { + if (time_is_before_jiffies(rx_event + 5 * HZ)) { IXGBE_READ_REG(hw, IXGBE_RXSTMPH); adapter->last_rx_ptp_check = jiffies; + adapter->rx_hwtstamp_cleared++; e_warn(drv, "clearing RX Timestamp hang\n"); } } /** + * ixgbe_ptp_clear_tx_timestamp - utility function to clear Tx timestamp state + * @adapter: the private adapter structure + * + * This function should be called whenever the state related to a Tx timestamp + * needs to be cleared. This helps ensure that all related bits are reset for + * the next Tx timestamp event. + */ +static void ixgbe_ptp_clear_tx_timestamp(struct ixgbe_adapter *adapter) +{ + struct ixgbe_hw *hw = &adapter->hw; + + IXGBE_READ_REG(hw, IXGBE_TXSTMPH); + if (adapter->ptp_tx_skb) { + dev_kfree_skb_any(adapter->ptp_tx_skb); + adapter->ptp_tx_skb = NULL; + } + clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state); +} + +/** * ixgbe_ptp_tx_hwtstamp - utility function which checks for TX time stamp * @adapter: the private adapter struct * @@ -461,23 +673,15 @@ static void ixgbe_ptp_tx_hwtstamp(struct ixgbe_adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; struct skb_shared_hwtstamps shhwtstamps; - u64 regval = 0, ns; - unsigned long flags; + u64 regval = 0; regval |= (u64)IXGBE_READ_REG(hw, IXGBE_TXSTMPL); regval |= (u64)IXGBE_READ_REG(hw, IXGBE_TXSTMPH) << 32; - spin_lock_irqsave(&adapter->tmreg_lock, flags); - ns = timecounter_cyc2time(&adapter->tc, regval); - spin_unlock_irqrestore(&adapter->tmreg_lock, flags); - - memset(&shhwtstamps, 0, sizeof(shhwtstamps)); - shhwtstamps.hwtstamp = ns_to_ktime(ns); + ixgbe_ptp_convert_to_hwtstamp(adapter, &shhwtstamps, regval); skb_tstamp_tx(adapter->ptp_tx_skb, &shhwtstamps); - dev_kfree_skb_any(adapter->ptp_tx_skb); - adapter->ptp_tx_skb = NULL; - clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state); + ixgbe_ptp_clear_tx_timestamp(adapter); } /** @@ -497,38 +701,85 @@ static void ixgbe_ptp_tx_hwtstamp_work(struct work_struct *work) IXGBE_PTP_TX_TIMEOUT); u32 tsynctxctl; - if (timeout) { - dev_kfree_skb_any(adapter->ptp_tx_skb); - adapter->ptp_tx_skb = NULL; - clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state); - e_warn(drv, "clearing Tx Timestamp hang\n"); + /* we have to have a valid skb to poll for a timestamp */ + if (!adapter->ptp_tx_skb) { + ixgbe_ptp_clear_tx_timestamp(adapter); return; } + /* stop polling once we have a valid timestamp */ tsynctxctl = IXGBE_READ_REG(hw, IXGBE_TSYNCTXCTL); - if (tsynctxctl & IXGBE_TSYNCTXCTL_VALID) + if (tsynctxctl & IXGBE_TSYNCTXCTL_VALID) { ixgbe_ptp_tx_hwtstamp(adapter); - else + return; + } + + if (timeout) { + ixgbe_ptp_clear_tx_timestamp(adapter); + adapter->tx_hwtstamp_timeouts++; + e_warn(drv, "clearing Tx Timestamp hang\n"); + } else { /* reschedule to keep checking if it's not available yet */ schedule_work(&adapter->ptp_tx_work); + } } /** - * ixgbe_ptp_rx_hwtstamp - utility function which checks for RX time stamp - * @adapter: pointer to adapter struct + * ixgbe_ptp_rx_pktstamp - utility function to get RX time stamp from buffer + * @q_vector: structure containing interrupt and ring information + * @skb: the packet + * + * This function will be called by the Rx routine of the timestamp for this + * packet is stored in the buffer. The value is stored in little endian format + * starting at the end of the packet data. + */ +void ixgbe_ptp_rx_pktstamp(struct ixgbe_q_vector *q_vector, + struct sk_buff *skb) +{ + __le64 regval; + + /* copy the bits out of the skb, and then trim the skb length */ + skb_copy_bits(skb, skb->len - IXGBE_TS_HDR_LEN, ®val, + IXGBE_TS_HDR_LEN); + __pskb_trim(skb, skb->len - IXGBE_TS_HDR_LEN); + + /* The timestamp is recorded in little endian format, and is stored at + * the end of the packet. + * + * DWORD: N N + 1 N + 2 + * Field: End of Packet SYSTIMH SYSTIML + */ + ixgbe_ptp_convert_to_hwtstamp(q_vector->adapter, skb_hwtstamps(skb), + le64_to_cpu(regval)); +} + +/** + * ixgbe_ptp_rx_rgtstamp - utility function which checks for RX time stamp + * @q_vector: structure containing interrupt and ring information * @skb: particular skb to send timestamp with * * if the timestamp is valid, we convert it into the timecounter ns * value, then store that result into the shhwtstamps structure which * is passed up the network stack */ -void ixgbe_ptp_rx_hwtstamp(struct ixgbe_adapter *adapter, struct sk_buff *skb) +void ixgbe_ptp_rx_rgtstamp(struct ixgbe_q_vector *q_vector, + struct sk_buff *skb) { - struct ixgbe_hw *hw = &adapter->hw; - struct skb_shared_hwtstamps *shhwtstamps; - u64 regval = 0, ns; + struct ixgbe_adapter *adapter; + struct ixgbe_hw *hw; + u64 regval = 0; u32 tsyncrxctl; - unsigned long flags; + + /* we cannot process timestamps on a ring without a q_vector */ + if (!q_vector || !q_vector->adapter) + return; + + adapter = q_vector->adapter; + hw = &adapter->hw; + + /* Read the tsyncrxctl register afterwards in order to prevent taking an + * I/O hit on every packet. + */ tsyncrxctl = IXGBE_READ_REG(hw, IXGBE_TSYNCRXCTL); if (!(tsyncrxctl & IXGBE_TSYNCRXCTL_VALID)) @@ -537,17 +788,7 @@ void ixgbe_ptp_rx_hwtstamp(struct ixgbe_adapter *adapter, struct sk_buff *skb) regval |= (u64)IXGBE_READ_REG(hw, IXGBE_RXSTMPL); regval |= (u64)IXGBE_READ_REG(hw, IXGBE_RXSTMPH) << 32; - spin_lock_irqsave(&adapter->tmreg_lock, flags); - ns = timecounter_cyc2time(&adapter->tc, regval); - spin_unlock_irqrestore(&adapter->tmreg_lock, flags); - - shhwtstamps = skb_hwtstamps(skb); - shhwtstamps->hwtstamp = ns_to_ktime(ns); - - /* Update the last_rx_timestamp timer in order to enable watchdog check - * for error case of latched timestamp on a dropped packet. - */ - adapter->last_rx_timestamp = jiffies; + ixgbe_ptp_convert_to_hwtstamp(adapter, skb_hwtstamps(skb), regval); } int ixgbe_ptp_get_ts_config(struct ixgbe_adapter *adapter, struct ifreq *ifr) @@ -610,14 +851,20 @@ static int ixgbe_ptp_set_timestamp_mode(struct ixgbe_adapter *adapter, case HWTSTAMP_FILTER_NONE: tsync_rx_ctl = 0; tsync_rx_mtrl = 0; + adapter->flags &= ~(IXGBE_FLAG_RX_HWTSTAMP_ENABLED | + IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER); break; case HWTSTAMP_FILTER_PTP_V1_L4_SYNC: tsync_rx_ctl |= IXGBE_TSYNCRXCTL_TYPE_L4_V1; tsync_rx_mtrl |= IXGBE_RXMTRL_V1_SYNC_MSG; + adapter->flags &= ~(IXGBE_FLAG_RX_HWTSTAMP_ENABLED | + IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER); break; case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ: tsync_rx_ctl |= IXGBE_TSYNCRXCTL_TYPE_L4_V1; tsync_rx_mtrl |= IXGBE_RXMTRL_V1_DELAY_REQ_MSG; + adapter->flags &= ~(IXGBE_FLAG_RX_HWTSTAMP_ENABLED | + IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER); break; case HWTSTAMP_FILTER_PTP_V2_EVENT: case HWTSTAMP_FILTER_PTP_V2_L2_EVENT: @@ -631,9 +878,21 @@ static int ixgbe_ptp_set_timestamp_mode(struct ixgbe_adapter *adapter, tsync_rx_ctl |= IXGBE_TSYNCRXCTL_TYPE_EVENT_V2; is_l2 = true; config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT; + adapter->flags &= ~(IXGBE_FLAG_RX_HWTSTAMP_ENABLED | + IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER); break; case HWTSTAMP_FILTER_PTP_V1_L4_EVENT: case HWTSTAMP_FILTER_ALL: + /* The X550 controller is capable of timestamping all packets, + * which allows it to accept any filter. + */ + if (hw->mac.type >= ixgbe_mac_X550) { + tsync_rx_ctl |= IXGBE_TSYNCRXCTL_TYPE_ALL; + config->rx_filter = HWTSTAMP_FILTER_ALL; + adapter->flags |= IXGBE_FLAG_RX_HWTSTAMP_ENABLED; + break; + } + /* fall through */ default: /* * register RXMTRL must be set in order to do V1 packets, @@ -641,16 +900,46 @@ static int ixgbe_ptp_set_timestamp_mode(struct ixgbe_adapter *adapter, * Delay_Req messages and hardware does not support * timestamping all packets => return error */ + adapter->flags &= ~(IXGBE_FLAG_RX_HWTSTAMP_ENABLED | + IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER); config->rx_filter = HWTSTAMP_FILTER_NONE; return -ERANGE; } if (hw->mac.type == ixgbe_mac_82598EB) { + adapter->flags &= ~(IXGBE_FLAG_RX_HWTSTAMP_ENABLED | + IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER); if (tsync_rx_ctl | tsync_tx_ctl) return -ERANGE; return 0; } + /* Per-packet timestamping only works if the filter is set to all + * packets. Since this is desired, always timestamp all packets as long + * as any Rx filter was configured. + */ + switch (hw->mac.type) { + case ixgbe_mac_X550: + case ixgbe_mac_X550EM_x: + /* enable timestamping all packets only if at least some + * packets were requested. Otherwise, play nice and disable + * timestamping + */ + if (config->rx_filter == HWTSTAMP_FILTER_NONE) + break; + + tsync_rx_ctl = IXGBE_TSYNCRXCTL_ENABLED | + IXGBE_TSYNCRXCTL_TYPE_ALL | + IXGBE_TSYNCRXCTL_TSIP_UT_EN; + config->rx_filter = HWTSTAMP_FILTER_ALL; + adapter->flags |= IXGBE_FLAG_RX_HWTSTAMP_ENABLED; + adapter->flags &= ~IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER; + is_l2 = true; + break; + default: + break; + } + /* define ethertype filter for timestamping L2 packets */ if (is_l2) IXGBE_WRITE_REG(hw, IXGBE_ETQF(IXGBE_ETQF_FILTER_1588), @@ -678,8 +967,8 @@ static int ixgbe_ptp_set_timestamp_mode(struct ixgbe_adapter *adapter, IXGBE_WRITE_FLUSH(hw); /* clear TX/RX time stamp registers, just to be sure */ - regval = IXGBE_READ_REG(hw, IXGBE_TXSTMPH); - regval = IXGBE_READ_REG(hw, IXGBE_RXSTMPH); + ixgbe_ptp_clear_tx_timestamp(adapter); + IXGBE_READ_REG(hw, IXGBE_RXSTMPH); return 0; } @@ -712,23 +1001,9 @@ int ixgbe_ptp_set_ts_config(struct ixgbe_adapter *adapter, struct ifreq *ifr) -EFAULT : 0; } -/** - * ixgbe_ptp_start_cyclecounter - create the cycle counter from hw - * @adapter: pointer to the adapter structure - * - * This function should be called to set the proper values for the TIMINCA - * register and tell the cyclecounter structure what the tick rate of SYSTIME - * is. It does not directly modify SYSTIME registers or the timecounter - * structure. It should be called whenever a new TIMINCA value is necessary, - * such as during initialization or when the link speed changes. - */ -void ixgbe_ptp_start_cyclecounter(struct ixgbe_adapter *adapter) +static void ixgbe_ptp_link_speed_adjust(struct ixgbe_adapter *adapter, + u32 *shift, u32 *incval) { - struct ixgbe_hw *hw = &adapter->hw; - u32 incval = 0; - u32 shift = 0; - unsigned long flags; - /** * Scale the NIC cycle counter by a large factor so that * relatively small corrections to the frequency can be added @@ -745,36 +1020,98 @@ void ixgbe_ptp_start_cyclecounter(struct ixgbe_adapter *adapter) */ switch (adapter->link_speed) { case IXGBE_LINK_SPEED_100_FULL: - incval = IXGBE_INCVAL_100; - shift = IXGBE_INCVAL_SHIFT_100; + *shift = IXGBE_INCVAL_SHIFT_100; + *incval = IXGBE_INCVAL_100; break; case IXGBE_LINK_SPEED_1GB_FULL: - incval = IXGBE_INCVAL_1GB; - shift = IXGBE_INCVAL_SHIFT_1GB; + *shift = IXGBE_INCVAL_SHIFT_1GB; + *incval = IXGBE_INCVAL_1GB; break; case IXGBE_LINK_SPEED_10GB_FULL: default: - incval = IXGBE_INCVAL_10GB; - shift = IXGBE_INCVAL_SHIFT_10GB; + *shift = IXGBE_INCVAL_SHIFT_10GB; + *incval = IXGBE_INCVAL_10GB; break; } +} - /** - * Modify the calculated values to fit within the correct - * number of bits specified by the hardware. The 82599 doesn't - * have the same space as the X540, so bitshift the calculated - * values to fit. +/** + * ixgbe_ptp_start_cyclecounter - create the cycle counter from hw + * @adapter: pointer to the adapter structure + * + * This function should be called to set the proper values for the TIMINCA + * register and tell the cyclecounter structure what the tick rate of SYSTIME + * is. It does not directly modify SYSTIME registers or the timecounter + * structure. It should be called whenever a new TIMINCA value is necessary, + * such as during initialization or when the link speed changes. + */ +void ixgbe_ptp_start_cyclecounter(struct ixgbe_adapter *adapter) +{ + struct ixgbe_hw *hw = &adapter->hw; + struct cyclecounter cc; + unsigned long flags; + u32 incval = 0; + u32 tsauxc = 0; + u32 fuse0 = 0; + + /* For some of the boards below this mask is technically incorrect. + * The timestamp mask overflows at approximately 61bits. However the + * particular hardware does not overflow on an even bitmask value. + * Instead, it overflows due to conversion of upper 32bits billions of + * cycles. Timecounters are not really intended for this purpose so + * they do not properly function if the overflow point isn't 2^N-1. + * However, the actual SYSTIME values in question take ~138 years to + * overflow. In practice this means they won't actually overflow. A + * proper fix to this problem would require modification of the + * timecounter delta calculations. */ + cc.mask = CLOCKSOURCE_MASK(64); + cc.mult = 1; + cc.shift = 0; + switch (hw->mac.type) { + case ixgbe_mac_X550EM_x: + /* SYSTIME assumes X550EM_x board frequency is 300Mhz, and is + * designed to represent seconds and nanoseconds when this is + * the case. However, some revisions of hardware have a 400Mhz + * clock and we have to compensate for this frequency + * variation using corrected mult and shift values. + */ + fuse0 = IXGBE_READ_REG(hw, IXGBE_FUSES0_GROUP(0)); + if (!(fuse0 & IXGBE_FUSES0_300MHZ)) { + cc.mult = 3; + cc.shift = 2; + } + /* fallthrough */ + case ixgbe_mac_X550: + cc.read = ixgbe_ptp_read_X550; + + /* enable SYSTIME counter */ + IXGBE_WRITE_REG(hw, IXGBE_SYSTIMR, 0); + IXGBE_WRITE_REG(hw, IXGBE_SYSTIML, 0); + IXGBE_WRITE_REG(hw, IXGBE_SYSTIMH, 0); + tsauxc = IXGBE_READ_REG(hw, IXGBE_TSAUXC); + IXGBE_WRITE_REG(hw, IXGBE_TSAUXC, + tsauxc & ~IXGBE_TSAUXC_DISABLE_SYSTIME); + IXGBE_WRITE_REG(hw, IXGBE_TSIM, IXGBE_TSIM_TXTS); + IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_TIMESYNC); + + IXGBE_WRITE_FLUSH(hw); + break; case ixgbe_mac_X540: + cc.read = ixgbe_ptp_read_82599; + + ixgbe_ptp_link_speed_adjust(adapter, &cc.shift, &incval); IXGBE_WRITE_REG(hw, IXGBE_TIMINCA, incval); break; case ixgbe_mac_82599EB: + cc.read = ixgbe_ptp_read_82599; + + ixgbe_ptp_link_speed_adjust(adapter, &cc.shift, &incval); incval >>= IXGBE_INCVAL_SHIFT_82599; - shift -= IXGBE_INCVAL_SHIFT_82599; + cc.shift -= IXGBE_INCVAL_SHIFT_82599; IXGBE_WRITE_REG(hw, IXGBE_TIMINCA, - (1 << IXGBE_INCPER_SHIFT_82599) | - incval); + (1 << IXGBE_INCPER_SHIFT_82599) | incval); break; default: /* other devices aren't supported */ @@ -787,13 +1124,7 @@ void ixgbe_ptp_start_cyclecounter(struct ixgbe_adapter *adapter) /* need lock to prevent incorrect read while modifying cyclecounter */ spin_lock_irqsave(&adapter->tmreg_lock, flags); - - memset(&adapter->cc, 0, sizeof(adapter->cc)); - adapter->cc.read = ixgbe_ptp_read; - adapter->cc.mask = CYCLECOUNTER_MASK(64); - adapter->cc.shift = shift; - adapter->cc.mult = 1; - + memcpy(&adapter->hw_cc, &cc, sizeof(adapter->hw_cc)); spin_unlock_irqrestore(&adapter->tmreg_lock, flags); } @@ -814,29 +1145,27 @@ void ixgbe_ptp_reset(struct ixgbe_adapter *adapter) struct ixgbe_hw *hw = &adapter->hw; unsigned long flags; - /* set SYSTIME registers to 0 just in case */ - IXGBE_WRITE_REG(hw, IXGBE_SYSTIML, 0x00000000); - IXGBE_WRITE_REG(hw, IXGBE_SYSTIMH, 0x00000000); - IXGBE_WRITE_FLUSH(hw); - /* reset the hardware timestamping mode */ ixgbe_ptp_set_timestamp_mode(adapter, &adapter->tstamp_config); + /* 82598 does not support PTP */ + if (hw->mac.type == ixgbe_mac_82598EB) + return; + ixgbe_ptp_start_cyclecounter(adapter); spin_lock_irqsave(&adapter->tmreg_lock, flags); - - /* reset the ns time counter */ - timecounter_init(&adapter->tc, &adapter->cc, + timecounter_init(&adapter->hw_tc, &adapter->hw_cc, ktime_to_ns(ktime_get_real())); - spin_unlock_irqrestore(&adapter->tmreg_lock, flags); - /* - * Now that the shift has been calculated and the systime + adapter->last_overflow_check = jiffies; + + /* Now that the shift has been calculated and the systime * registers reset, (re-)enable the Clock out feature */ - ixgbe_ptp_setup_sdp(adapter); + if (adapter->ptp_setup_sdp) + adapter->ptp_setup_sdp(adapter); } /** @@ -845,11 +1174,11 @@ void ixgbe_ptp_reset(struct ixgbe_adapter *adapter) * * This function performs setup of the user entry point function table and * initializes the PTP clock device, which is used to access the clock-like - * features of the PTP core. It will be called by ixgbe_ptp_init, only if - * there isn't already a clock device (such as after a suspend/resume cycle, - * where the clock device wasn't destroyed). + * features of the PTP core. It will be called by ixgbe_ptp_init, and may + * reuse a previously initialized clock (such as during a suspend/resume + * cycle). */ -static int ixgbe_ptp_create_clock(struct ixgbe_adapter *adapter) +static long ixgbe_ptp_create_clock(struct ixgbe_adapter *adapter) { struct net_device *netdev = adapter->netdev; long err; @@ -869,11 +1198,12 @@ static int ixgbe_ptp_create_clock(struct ixgbe_adapter *adapter) adapter->ptp_caps.n_ext_ts = 0; adapter->ptp_caps.n_per_out = 0; adapter->ptp_caps.pps = 1; - adapter->ptp_caps.adjfreq = ixgbe_ptp_adjfreq; + adapter->ptp_caps.adjfreq = ixgbe_ptp_adjfreq_82599; adapter->ptp_caps.adjtime = ixgbe_ptp_adjtime; adapter->ptp_caps.gettime64 = ixgbe_ptp_gettime; adapter->ptp_caps.settime64 = ixgbe_ptp_settime; adapter->ptp_caps.enable = ixgbe_ptp_feature_enable; + adapter->ptp_setup_sdp = ixgbe_ptp_setup_sdp_x540; break; case ixgbe_mac_82599EB: snprintf(adapter->ptp_caps.name, @@ -885,14 +1215,31 @@ static int ixgbe_ptp_create_clock(struct ixgbe_adapter *adapter) adapter->ptp_caps.n_ext_ts = 0; adapter->ptp_caps.n_per_out = 0; adapter->ptp_caps.pps = 0; - adapter->ptp_caps.adjfreq = ixgbe_ptp_adjfreq; + adapter->ptp_caps.adjfreq = ixgbe_ptp_adjfreq_82599; + adapter->ptp_caps.adjtime = ixgbe_ptp_adjtime; + adapter->ptp_caps.gettime64 = ixgbe_ptp_gettime; + adapter->ptp_caps.settime64 = ixgbe_ptp_settime; + adapter->ptp_caps.enable = ixgbe_ptp_feature_enable; + break; + case ixgbe_mac_X550: + case ixgbe_mac_X550EM_x: + snprintf(adapter->ptp_caps.name, 16, "%s", netdev->name); + adapter->ptp_caps.owner = THIS_MODULE; + adapter->ptp_caps.max_adj = 30000000; + adapter->ptp_caps.n_alarm = 0; + adapter->ptp_caps.n_ext_ts = 0; + adapter->ptp_caps.n_per_out = 0; + adapter->ptp_caps.pps = 0; + adapter->ptp_caps.adjfreq = ixgbe_ptp_adjfreq_X550; adapter->ptp_caps.adjtime = ixgbe_ptp_adjtime; adapter->ptp_caps.gettime64 = ixgbe_ptp_gettime; adapter->ptp_caps.settime64 = ixgbe_ptp_settime; adapter->ptp_caps.enable = ixgbe_ptp_feature_enable; + adapter->ptp_setup_sdp = NULL; break; default: adapter->ptp_clock = NULL; + adapter->ptp_setup_sdp = NULL; return -EOPNOTSUPP; } @@ -961,18 +1308,13 @@ void ixgbe_ptp_suspend(struct ixgbe_adapter *adapter) if (!test_and_clear_bit(__IXGBE_PTP_RUNNING, &adapter->state)) return; - /* since this might be called in suspend, we don't clear the state, - * but simply reset the auxiliary PPS signal control register - */ - IXGBE_WRITE_REG(&adapter->hw, IXGBE_TSAUXC, 0x0); + adapter->flags2 &= ~IXGBE_FLAG2_PTP_PPS_ENABLED; + if (adapter->ptp_setup_sdp) + adapter->ptp_setup_sdp(adapter); /* ensure that we cancel any pending PTP Tx work item in progress */ cancel_work_sync(&adapter->ptp_tx_work); - if (adapter->ptp_tx_skb) { - dev_kfree_skb_any(adapter->ptp_tx_skb); - adapter->ptp_tx_skb = NULL; - clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state); - } + ixgbe_ptp_clear_tx_timestamp(adapter); } /** |