ice: add trace events for tx timestamps
authorJacob Keller <jacob.e.keller@intel.com>
Tue, 1 Mar 2022 23:02:28 +0000 (15:02 -0800)
committerTony Nguyen <anthony.l.nguyen@intel.com>
Wed, 16 Mar 2022 17:38:15 +0000 (10:38 -0700)
We've previously run into many issues related to the latency of a Tx
timestamp completion with the ice hardware. It can be difficult to
determine the root cause of a slow Tx timestamp. To aid in this,
introduce new trace events which capture timing data about when the
driver reaches certain points while processing a transmit timestamp

 * ice_tx_tstamp_request: Trace when the stack initiates a new timestamp
   request.

 * ice_tx_tstamp_fw_req: Trace when the driver begins a read of the
   timestamp register in the work thread.

 * ice_tx_tstamp_fw_done: Trace when the driver finishes reading a
   timestamp register in the work thread.

 * ice_tx_tstamp_complete: Trace when the driver submits the skb back to
   the stack with a completed Tx timestamp.

These trace events can be enabled using the standard trace event
subsystem exposed by the ice driver. If they are disabled, they become
no-ops with no run time cost.

The following is a simple GNU AWK script which can highlight one
potential way to use the trace events to capture latency data from the
trace buffer about how long the driver takes to process a timestamp:

-----
  BEGIN {
      PREC=256
  }

  # Detect requests
  /tx_tstamp_request/ {
      time=strtonum($4)
      skb=$7

      # Store the time of request for this skb
      requests[skb] = time
      printf("skb %s: idx %d at %.6f\n", skb, idx, time)
  }

  # Detect completions
  /tx_tstamp_complete/ {
      time=strtonum($4)
      skb=$7
      idx=$9

      if (skb in requests) {
          latency = (time - requests[skb]) * 1000
          printf("skb %s: %.3f to complete\n", skb, latency)
          if (latency > 4) {
              printf(">>> HIGH LATENCY <<<\n")
          }
          printf("\n")
      } else {
          printf("!!! skb %s (idx %d) at %.6f\n", skb, idx, time)
      }
  }
-----

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Gurucharan <gurucharanx.g@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
drivers/net/ethernet/intel/ice/ice_ptp.c
drivers/net/ethernet/intel/ice/ice_trace.h

index 000c39d..a1cd332 100644 (file)
@@ -3,6 +3,7 @@
 
 #include "ice.h"
 #include "ice_lib.h"
+#include "ice_trace.h"
 
 #define E810_OUT_PROP_DELAY_NS 1
 
@@ -2063,11 +2064,15 @@ static void ice_ptp_tx_tstamp_work(struct kthread_work *work)
                struct sk_buff *skb;
                int err;
 
+               ice_trace(tx_tstamp_fw_req, tx->tstamps[idx].skb, idx);
+
                err = ice_read_phy_tstamp(hw, tx->quad, phy_idx,
                                          &raw_tstamp);
                if (err)
                        continue;
 
+               ice_trace(tx_tstamp_fw_done, tx->tstamps[idx].skb, idx);
+
                /* Check if the timestamp is invalid or stale */
                if (!(raw_tstamp & ICE_PTP_TS_VALID) ||
                    raw_tstamp == tx->tstamps[idx].cached_tstamp)
@@ -2093,6 +2098,8 @@ static void ice_ptp_tx_tstamp_work(struct kthread_work *work)
                tstamp = ice_ptp_extend_40b_ts(pf, raw_tstamp);
                shhwtstamps.hwtstamp = ns_to_ktime(tstamp);
 
+               ice_trace(tx_tstamp_complete, skb, idx);
+
                skb_tstamp_tx(skb, &shhwtstamps);
                dev_kfree_skb_any(skb);
        }
@@ -2131,6 +2138,7 @@ s8 ice_ptp_request_ts(struct ice_ptp_tx *tx, struct sk_buff *skb)
                tx->tstamps[idx].start = jiffies;
                tx->tstamps[idx].skb = skb_get(skb);
                skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
+               ice_trace(tx_tstamp_request, skb, idx);
        }
 
        spin_unlock(&tx->lock);
index cf68524..ae98d5a 100644 (file)
@@ -216,6 +216,30 @@ DEFINE_EVENT(ice_xmit_template, name, \
 DEFINE_XMIT_TEMPLATE_OP_EVENT(ice_xmit_frame_ring);
 DEFINE_XMIT_TEMPLATE_OP_EVENT(ice_xmit_frame_ring_drop);
 
+DECLARE_EVENT_CLASS(ice_tx_tstamp_template,
+                   TP_PROTO(struct sk_buff *skb, int idx),
+
+                   TP_ARGS(skb, idx),
+
+                   TP_STRUCT__entry(__field(void *, skb)
+                                    __field(int, idx)),
+
+                   TP_fast_assign(__entry->skb = skb;
+                                  __entry->idx = idx;),
+
+                   TP_printk("skb %pK idx %d",
+                             __entry->skb, __entry->idx)
+);
+#define DEFINE_TX_TSTAMP_OP_EVENT(name) \
+DEFINE_EVENT(ice_tx_tstamp_template, name, \
+            TP_PROTO(struct sk_buff *skb, int idx), \
+            TP_ARGS(skb, idx))
+
+DEFINE_TX_TSTAMP_OP_EVENT(ice_tx_tstamp_request);
+DEFINE_TX_TSTAMP_OP_EVENT(ice_tx_tstamp_fw_req);
+DEFINE_TX_TSTAMP_OP_EVENT(ice_tx_tstamp_fw_done);
+DEFINE_TX_TSTAMP_OP_EVENT(ice_tx_tstamp_complete);
+
 /* End tracepoints */
 
 #endif /* _ICE_TRACE_H_ */