Re: [PATCH iwl-next 9/9] igc: Add support to get frame preemption statistics via ethtool

From: Abdul Rahim, Faizal
Date: Mon Dec 23 2024 - 04:53:02 EST




On 17/12/2024 12:05 am, Vladimir Oltean wrote:
On Mon, Dec 16, 2024 at 01:47:20AM -0500, Faizal Rahim wrote:
Implemented "ethtool --include-statistics --show-mm" callback for IGC.

Tested preemption scenario to check preemption statistics:
1) Trigger verification handshake on both boards:
$ sudo ethtool --set-mm enp1s0 pmac-enabled on
$ sudo ethtool --set-mm enp1s0 tx-enabled on
$ sudo ethtool --set-mm enp1s0 verify-enabled on
2) Set preemptible or express queue in taprio for tx board:
$ sudo tc qdisc replace dev enp1s0 parent root handle 100 taprio \
num_tc 4 map 0 1 2 3 0 0 0 0 0 0 0 0 0 0 0 0 \
queues 1@0 1@1 1@2 1@3 base-time 0 sched-entry S F 100000 \
fp E E P P

Hmm, the prio_tc_map pattern changed since the last time I looked at igc
examples? It was in decreasing order before? How do you handle backwards
compatibility with the Tx ring strict priority default configuration?
I haven't downloaded the entire set locally, will do so later.


I tested like this for i226:
CMD=(
"tc qdisc replace dev $IFACE parent root handle 100 taprio "
"num_tc 4 "
"map 3 2 1 0 3 3 3 3 3 3 3 3 3 3 3 3 "
"queues 1@0 1@1 1@2 1@3 "
"base-time $BASE "
"${SCHED_ENTRY[*]} "
"flags 0x1 "
"txtime-delay $TXTIME_DELAY "
"clockid CLOCK_TAI "

But I mistakenly copied the mapping from a different scenario where socket priority -> tc -> hw_queue mapping is not important to my test objective in that scenario.

I'll update the description to use decreasing order then.

+
+ return ooo_smdc + ooo_frame_cnt + ooo_frag_cnt + miss_frame_frag_cnt;
+}
+
+static void igc_ethtool_get_mm_stats(struct net_device *dev,
+ struct ethtool_mm_stats *stats)
+{
+ struct igc_adapter *adapter = netdev_priv(dev);
+ struct igc_hw *hw = &adapter->hw;
+
+ stats->MACMergeFrameAssErrorCount = igc_ethtool_get_frame_ass_error(dev);
+ stats->MACMergeFrameSmdErrorCount = 0; /* Not available in IGC */
+ stats->MACMergeFrameAssOkCount = rd32(IGC_PRMPTDRCNT);
+ stats->MACMergeFragCountRx = rd32(IGC_PRMEVNTRCNT);
+ stats->MACMergeFragCountTx = rd32(IGC_PRMEVNTTCNT);
+ stats->MACMergeHoldCount = 0; /* Not available in IGC */

Don't report counters as zero when in reality you don't know.

Just don't assign values to these. mm_prepare_data() -> ethtool_stats_init()
presets them to 0xffffffffffffffff (ETHTOOL_STAT_NOT_SET), and
mm_put_stats() -> mm_put_stat() detects whether they are still equal to
this value, and if they are, does not report netlink attributes for them.


Got it.


diff --git a/drivers/net/ethernet/intel/igc/igc_regs.h b/drivers/net/ethernet/intel/igc/igc_regs.h
index 12ddc5793651..f40946cce35a 100644
--- a/drivers/net/ethernet/intel/igc/igc_regs.h
+++ b/drivers/net/ethernet/intel/igc/igc_regs.h
@@ -222,6 +222,25 @@
#define IGC_FTQF(_n) (0x059E0 + (4 * (_n))) /* 5-tuple Queue Fltr */
+/* Time sync registers - preemption statistics */
+#define IGC_PRMEVNTTCNT 0x04298 /* TX Preemption event counter */
+#define IGC_PRMEVNTRCNT 0x0429C /* RX Preemption event counter */
+#define IGC_PRMPTDRCNT 0x04284 /* Good RX Preempted Packets */
+
+ /* Preemption Exception Counter */
+#define IGC_PRMEXPRCNT 0x042A0
+/* Received out of order packets with SMD-C and NOT ReumeRx */
+#define IGC_PRMEXPRCNT_OOO_SMDC 0x000000FF
+/* Received out of order packets with SMD-C and wrong Frame CNT */
+#define IGC_PRMEXPRCNT_OOO_FRAME_CNT 0x0000FF00
+#define IGC_PRMEXPRCNT_OOO_FRAME_CNT_SHIFT 8
+/* Received out of order packets with SMD-C and wrong Frag CNT */
+#define IGC_PRMEXPRCNT_OOO_FRAG_CNT 0x00FF0000
+#define IGC_PRMEXPRCNT_OOO_FRAG_CNT_SHIFT 16
+/* Received packets with SMD-S and ReumeRx */

What is ReumeRx?

Resume receive. It’s a typo in the i226 documentation that I (shamelessly) copied into the code without checking properly. It's meant to indicate that an RX flag in the i226 firmware is set. I’ll remove the 'ResumeRX' part since it adds confusion.