@@ -730,6 +730,15 @@ static u8 iavf_select_rx_desc_format(struct iavf_adapter *adapter)
if (!RXDID_ALLOWED(adapter))
return VIRTCHNL_RXDID_1_32B_BASE;
+ /* Rx timestamping requires the use of flexible NIC descriptors */
+ if (iavf_ptp_cap_supported(adapter, VIRTCHNL_1588_PTP_CAP_RX_TSTAMP)) {
+ if (supported_rxdids & BIT(VIRTCHNL_RXDID_2_FLEX_SQ_NIC))
+ return VIRTCHNL_RXDID_2_FLEX_SQ_NIC;
+
+ dev_dbg(&adapter->pdev->dev,
+ "Unable to negotiate flexible descriptor format.\n");
+ }
+
/* Warn if the PF does not list support for the default legacy
* descriptor format. This shouldn't happen, as this is the format
* used if VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC is not supported. It is
@@ -445,6 +445,9 @@ void iavf_ptp_release(struct iavf_adapter *adapter)
}
adapter->aq_required &= ~IAVF_FLAG_AQ_SEND_PTP_CMD;
spin_unlock(&adapter->ptp.aq_cmd_lock);
+
+ adapter->ptp.hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
+ iavf_ptp_disable_rx_tstamp(adapter);
}
/**
@@ -477,3 +480,69 @@ void iavf_ptp_process_caps(struct iavf_adapter *adapter)
iavf_ptp_disable_rx_tstamp(adapter);
}
}
+
+/**
+ * iavf_ptp_extend_32b_timestamp - Convert a 32b nanoseconds timestamp to 64b
+ * nanoseconds
+ * @cached_phc_time: recently cached copy of PHC time
+ * @in_tstamp: Ingress/egress 32b nanoseconds timestamp value
+ *
+ * Hardware captures timestamps which contain only 32 bits of nominal
+ * nanoseconds, as opposed to the 64bit timestamps that the stack expects.
+ *
+ * Extend the 32bit nanosecond timestamp using the following algorithm and
+ * assumptions:
+ *
+ * 1) have a recently cached copy of the PHC time
+ * 2) assume that the in_tstamp was captured 2^31 nanoseconds (~2.1
+ * seconds) before or after the PHC time was captured.
+ * 3) calculate the delta between the cached time and the timestamp
+ * 4) if the delta is smaller than 2^31 nanoseconds, then the timestamp was
+ * captured after the PHC time. In this case, the full timestamp is just
+ * the cached PHC time plus the delta.
+ * 5) otherwise, if the delta is larger than 2^31 nanoseconds, then the
+ * timestamp was captured *before* the PHC time, i.e. because the PHC
+ * cache was updated after the timestamp was captured by hardware. In this
+ * case, the full timestamp is the cached time minus the inverse delta.
+ *
+ * This algorithm works even if the PHC time was updated after a Tx timestamp
+ * was requested, but before the Tx timestamp event was reported from
+ * hardware.
+ *
+ * This calculation primarily relies on keeping the cached PHC time up to
+ * date. If the timestamp was captured more than 2^31 nanoseconds after the
+ * PHC time, it is possible that the lower 32bits of PHC time have
+ * overflowed more than once, and we might generate an incorrect timestamp.
+ *
+ * This is prevented by (a) periodically updating the cached PHC time once
+ * a second, and (b) discarding any Tx timestamp packet if it has waited for
+ * a timestamp for more than one second.
+ *
+ * Return: extended timestamp (to 64b)
+ */
+u64 iavf_ptp_extend_32b_timestamp(u64 cached_phc_time, u32 in_tstamp)
+{
+ const u64 mask = GENMASK_ULL(31, 0);
+ u32 delta;
+ u64 ns;
+
+ /* Calculate the delta between the lower 32bits of the cached PHC
+ * time and the in_tstamp value
+ */
+ delta = (in_tstamp - (u32)(cached_phc_time & mask));
+
+ /* Do not assume that the in_tstamp is always more recent than the
+ * cached PHC time. If the delta is large, it indicates that the
+ * in_tstamp was taken in the past, and should be converted
+ * forward.
+ */
+ if (delta > (mask / 2)) {
+ /* reverse the delta calculation here */
+ delta = ((u32)(cached_phc_time & mask) - in_tstamp);
+ ns = cached_phc_time - delta;
+ } else {
+ ns = cached_phc_time + delta;
+ }
+
+ return ns;
+}
@@ -6,6 +6,9 @@
#include <linux/ptp_clock_kernel.h>
+/* bit indicating whether a 40bit timestamp is valid */
+#define IAVF_PTP_40B_TSTAMP_VALID BIT(0)
+
/* structure used to queue PTP commands for processing */
struct iavf_ptp_aq_cmd {
struct list_head list;
@@ -38,5 +41,6 @@ void iavf_virtchnl_send_ptp_cmd(struct iavf_adapter *adapter);
long iavf_ptp_do_aux_work(struct ptp_clock_info *ptp);
int iavf_ptp_get_ts_config(struct iavf_adapter *adapter, struct ifreq *ifr);
int iavf_ptp_set_ts_config(struct iavf_adapter *adapter, struct ifreq *ifr);
+u64 iavf_ptp_extend_32b_timestamp(u64 cached_phc_time, u32 in_tstamp);
#endif /* _IAVF_PTP_H_ */
@@ -1131,6 +1131,48 @@ static void iavf_flex_rx_hash(struct iavf_ring *ring,
}
}
+/**
+ * iavf_flex_rx_tstamp - Capture Rx timestamp from the descriptor
+ * @rx_ring: descriptor ring
+ * @rx_desc: specific descriptor
+ * @skb: skb currently being received
+ *
+ * Read the Rx timestamp value from the descriptor and pass it to the stack.
+ *
+ * This function only operates on the VIRTCHNL_RXDID_2_FLEX_SQ_NIC flexible
+ * descriptor writeback format.
+ */
+static void iavf_flex_rx_tstamp(struct iavf_ring *rx_ring,
+ union iavf_rx_desc *rx_desc,
+ struct sk_buff *skb)
+{
+ struct skb_shared_hwtstamps *skb_tstamps;
+ struct iavf_adapter *adapter;
+ u32 tstamp;
+ u64 ns;
+
+ /* Skip processing if timestamps aren't enabled */
+ if (!(rx_ring->flags & IAVF_TXRX_FLAGS_HW_TSTAMP))
+ return;
+
+ /* Check if this Rx descriptor has a valid timestamp */
+ if (!(rx_desc->flex_wb.ts_low & IAVF_PTP_40B_TSTAMP_VALID))
+ return;
+
+ adapter = netdev_priv(rx_ring->netdev);
+
+ /* the ts_low field only contains the valid bit and sub-nanosecond
+ * precision, so we don't need to extract it.
+ */
+ tstamp = le32_to_cpu(rx_desc->flex_wb.flex_ts.ts_high);
+ ns = iavf_ptp_extend_32b_timestamp(adapter->ptp.cached_phc_time,
+ tstamp);
+
+ skb_tstamps = skb_hwtstamps(skb);
+ memset(skb_tstamps, 0, sizeof(*skb_tstamps));
+ skb_tstamps->hwtstamp = ns_to_ktime(ns);
+}
+
/**
* iavf_process_skb_fields - Populate skb header fields from Rx descriptor
* @rx_ring: rx descriptor ring packet is being transacted on
@@ -1152,6 +1194,7 @@ static void iavf_process_skb_fields(struct iavf_ring *rx_ring,
} else {
iavf_flex_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
iavf_flex_rx_csum(rx_ring->vsi, skb, rx_desc);
+ iavf_flex_rx_tstamp(rx_ring, rx_desc, skb);
}
skb_record_rx_queue(skb, rx_ring->queue_index);