ixgbe: Check PTP Rx timestamps via BPF filter

This patch fixes a potential Rx timestamp deadlock that causes the Rx
timestamping to stall indefinitely. The issue could occur when a PTP packet is
timestamped by hardware but never reaches the Rx queue. In order to prevent a
permanent loss of timestamping, the RXSTMP(L/H) registers have to be read to
unlock them. (This used to only occur when a packet that was timestamped
reached the software.) However the registers can't be read early otherwise
there is no way to correlate them to the packet.

This patch introduces a filter function which can be used to determine if a
packet should have been timestamped. Supplied with the filter setup by the
hwtstamp ioctl, check to make sure the PTP protocol and message type match the
expected values. If so, then read the timestamp registers (to free them.) At
this point check the descriptor bit, if the bit is set then we know this
packet correlates to the timestamp stored in the RXTSTAMP registers.
Otherwise, assume that packet was dropped by the hardware, and ignore this
timestamp value. However, we have at least unlocked the rxtstamp registers for
future timestamping.

Due to the way the driver handles skb data, it cannot be directly accessed. In
order to work around this, a copy of the skb data into a linear buffer is
made. From this buffer it becomes possible to read the data correctly

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Richard Cochran <richardcochran@gmail.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

drivers/net/ethernet/intel/ixgbe/ixgbe.h

@@ -561,6 +561,7 @@ struct ixgbe_adapter {
 	spinlock_t tmreg_lock;
 	struct cyclecounter cc;
 	struct timecounter tc;
+	int rx_hwtstamp_filter;
 	u32 base_incval;
 	u32 cycle_speed;
 #endif /* CONFIG_IXGBE_PTP */
@@ -718,6 +719,7 @@ extern void ixgbe_ptp_overflow_check(struct ixgbe_adapter *adapter);
 extern void ixgbe_ptp_tx_hwtstamp(struct ixgbe_q_vector *q_vector,
 				  struct sk_buff *skb);
 extern void ixgbe_ptp_rx_hwtstamp(struct ixgbe_q_vector *q_vector,
+				  union ixgbe_adv_rx_desc *rx_desc,
 				  struct sk_buff *skb);
 extern int ixgbe_ptp_hwtstamp_ioctl(struct ixgbe_adapter *adapter,
 				    struct ifreq *ifr, int cmd);

drivers/net/ethernet/intel/ixgbe/ixgbe_main.c

@@ -1397,8 +1397,7 @@ static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
 
 #ifdef CONFIG_IXGBE_PTP
-	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_TS))
-		ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector, skb);
+	ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
 #endif
 
 	if ((dev->features & NETIF_F_HW_VLAN_RX) &&

drivers/net/ethernet/intel/ixgbe/ixgbe_ptp.c

@@ -26,6 +26,7 @@
 *******************************************************************************/
 #include "ixgbe.h"
 #include <linux/export.h>
+#include <linux/ptp_classify.h>
 
 /*
  * The 82599 and the X540 do not have true 64bit nanosecond scale
@@ -100,6 +101,10 @@
 #define NSECS_PER_SEC 1000000000ULL
 #endif
 
+static struct sock_filter ptp_filter[] = {
+	PTP_FILTER
+};
+
 /**
  * ixgbe_ptp_read - read raw cycle counter (to be used by time counter)
  * @cc - the cyclecounter structure
@@ -425,6 +430,68 @@ void ixgbe_ptp_overflow_check(struct ixgbe_adapter *adapter)
 	}
 }
 
+/**
+ * ixgbe_ptp_match - determine if this skb matches a ptp packet
+ * @skb: pointer to the skb
+ * @hwtstamp: pointer to the hwtstamp_config to check
+ *
+ * Determine whether the skb should have been timestamped, assuming the
+ * hwtstamp was set via the hwtstamp ioctl. Returns non-zero when the packet
+ * should have a timestamp waiting in the registers, and 0 otherwise.
+ *
+ * V1 packets have to check the version type to determine whether they are
+ * correct. However, we can't directly access the data because it might be
+ * fragmented in the SKB, in paged memory. In order to work around this, we
+ * use skb_copy_bits which will properly copy the data whether it is in the
+ * paged memory fragments or not. We have to copy the IP header as well as the
+ * message type.
+ */
+static int ixgbe_ptp_match(struct sk_buff *skb, int rx_filter)
+{
+	struct iphdr iph;
+	u8 msgtype;
+	unsigned int type, offset;
+
+	if (rx_filter == HWTSTAMP_FILTER_NONE)
+		return 0;
+
+	type = sk_run_filter(skb, ptp_filter);
+
+	if (likely(rx_filter == HWTSTAMP_FILTER_PTP_V2_EVENT))
+		return type & PTP_CLASS_V2;
+
+	/* For the remaining cases actually check message type */
+	switch (type) {
+	case PTP_CLASS_V1_IPV4:
+		skb_copy_bits(skb, OFF_IHL, &iph, sizeof(iph));
+		offset = ETH_HLEN + (iph.ihl << 2) + UDP_HLEN + OFF_PTP_CONTROL;
+		break;
+	case PTP_CLASS_V1_IPV6:
+		offset = OFF_PTP6 + OFF_PTP_CONTROL;
+		break;
+	default:
+		/* other cases invalid or handled above */
+		return 0;
+	}
+
+	/* Make sure our buffer is long enough */
+	if (skb->len < offset)
+		return 0;
+
+	skb_copy_bits(skb, offset, &msgtype, sizeof(msgtype));
+
+	switch (rx_filter) {
+	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
+		return (msgtype == IXGBE_RXMTRL_V1_SYNC_MSG);
+		break;
+	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
+		return (msgtype == IXGBE_RXMTRL_V1_DELAY_REQ_MSG);
+		break;
+	default:
+		return 0;
+	}
+}
+
 /**
  * ixgbe_ptp_tx_hwtstamp - utility function which checks for TX time stamp
  * @q_vector: structure containing interrupt and ring information
@@ -474,6 +541,7 @@ void ixgbe_ptp_tx_hwtstamp(struct ixgbe_q_vector *q_vector,
 /**
  * ixgbe_ptp_rx_hwtstamp - utility function which checks for RX time stamp
  * @q_vector: structure containing interrupt and ring information
+ * @rx_desc: the rx descriptor
  * @skb: particular skb to send timestamp with
  *
  * if the timestamp is valid, we convert it into the timecounter ns
@@ -481,6 +549,7 @@ void ixgbe_ptp_tx_hwtstamp(struct ixgbe_q_vector *q_vector,
  * is passed up the network stack
  */
 void ixgbe_ptp_rx_hwtstamp(struct ixgbe_q_vector *q_vector,
+			   union ixgbe_adv_rx_desc *rx_desc,
 			   struct sk_buff *skb)
 {
 	struct ixgbe_adapter *adapter;
@@ -498,21 +567,33 @@ void ixgbe_ptp_rx_hwtstamp(struct ixgbe_q_vector *q_vector,
 	hw = &adapter->hw;
 
 	tsyncrxctl = IXGBE_READ_REG(hw, IXGBE_TSYNCRXCTL);
+
+	/* Check if we have a valid timestamp and make sure the skb should
+	 * have been timestamped */
+	if (likely(!(tsyncrxctl & IXGBE_TSYNCRXCTL_VALID) ||
+		   !ixgbe_ptp_match(skb, adapter->rx_hwtstamp_filter)))
+		return;
+
+	/*
+	 * Always read the registers, in order to clear a possible fault
+	 * because of stagnant RX timestamp values for a packet that never
+	 * reached the queue.
+	 */
 	regval |= (u64)IXGBE_READ_REG(hw, IXGBE_RXSTMPL);
 	regval |= (u64)IXGBE_READ_REG(hw, IXGBE_RXSTMPH) << 32;
 
 	/*
-	 * If this bit is set, then the RX registers contain the time stamp. No
-	 * other packet will be time stamped until we read these registers, so
-	 * read the registers to make them available again. Because only one
-	 * packet can be time stamped at a time, we know that the register
-	 * values must belong to this one here and therefore we don't need to
-	 * compare any of the additional attributes stored for it.
+	 * If the timestamp bit is set in the packet's descriptor, we know the
+	 * timestamp belongs to this packet. No other packet can be
+	 * timestamped until the registers for timestamping have been read.
+	 * Therefor only one packet with this bit can be in the queue at a
+	 * time, and the rx timestamp values that were in the registers belong
+	 * to this packet.
 	 *
 	 * If nothing went wrong, then it should have a skb_shared_tx that we
 	 * can turn into a skb_shared_hwtstamps.
 	 */
-	if (!(tsyncrxctl & IXGBE_TSYNCRXCTL_VALID))
+	if (unlikely(!ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_TS)))
 		return;
 
 	spin_lock_irqsave(&adapter->tmreg_lock, flags);
@@ -598,19 +679,20 @@ int ixgbe_ptp_hwtstamp_ioctl(struct ixgbe_adapter *adapter,
 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
 		tsync_rx_ctl |= IXGBE_TSYNCRXCTL_TYPE_EVENT_V2;
-		config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
 		is_l2 = true;
 		is_l4 = true;
+		config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
 		break;
 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
 	case HWTSTAMP_FILTER_ALL:
 	default:
 		/*
-		 * register RXMTRL must be set, therefore it is not
-		 * possible to time stamp both V1 Sync and Delay_Req messages
-		 * and hardware does not support timestamping all packets
-		 * => return error
+		 * register RXMTRL must be set in order to do V1 packets,
+		 * therefore it is not possible to time stamp both V1 Sync and
+		 * Delay_Req messages and hardware does not support
+		 * timestamping all packets => return error
 		 */
+		config.rx_filter = HWTSTAMP_FILTER_NONE;
 		return -ERANGE;
 	}
 
@@ -620,6 +702,9 @@ int ixgbe_ptp_hwtstamp_ioctl(struct ixgbe_adapter *adapter,
 		return 0;
 	}
 
+	/* Store filter value for later use */
+	adapter->rx_hwtstamp_filter = config.rx_filter;
+
 	/* define ethertype filter for timestamped packets */
 	if (is_l2)
 		IXGBE_WRITE_REG(hw, IXGBE_ETQF(3),
@@ -855,6 +940,10 @@ void ixgbe_ptp_init(struct ixgbe_adapter *adapter)
 		return;
 	}
 
+	/* initialize the ptp filter */
+	if (ptp_filter_init(ptp_filter, ARRAY_SIZE(ptp_filter)))
+		e_dev_warn("ptp_filter_init failed\n");
+
 	spin_lock_init(&adapter->tmreg_lock);
 
 	ixgbe_ptp_start_cyclecounter(adapter);