Merge branch 'softnet-backlog'

qa/1030

@@ -20,6 +20,12 @@ $sudo rm -rf $tmp.* $seq.full
 trap "cd $here; rm -rf $tmp.*; exit \$status" 0 1 2 3 15
 
 # create some data files for testing - from various kernels
+cat <<EOF > $tmp.el9
+001ab57c 00000000 000002d0 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000003 00000000 00000001 00000002
+00041adb 00000000 0000005b 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000003 00000001 00000003 00000000
+00045167 00000000 0000005f 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000002 00000000 00000000
+0003f85b 00000000 0000005b 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000003 00000000 00000000
+EOF
 cat <<EOF > $tmp.el7
 0010c333 00000002 00000003 00000001 00000001 00000001 00000001 00000001 00000004 00000005 00000006
 000ee8f7 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
@@ -48,7 +54,7 @@ pmda=$PCP_PMDAS_DIR/linux/pmda_linux.so,linux_init
 metrics=`pminfo network.softnet | grep -v percpu | LC_COLLATE=POSIX sort`
 cd $here
 
-for file in el7 el6 el5
+for file in el9 el7 el6 el5
 do
     rm -fr $root
     mkdir -p $root/proc/net || _fail "root in use when preparing $file"

qa/1030.out

@@ -1,4 +1,35 @@
 QA output created by 1030
+== Checking metric values - el9
+
+network.softnet.cpu_collision
+    value 0
+
+network.softnet.dropped
+    value 0
+
+network.softnet.flow_limit_count
+    value 0
+
+network.softnet.input_qlen
+    value 4
+
+network.softnet.process_qlen
+    value 2
+
+network.softnet.processed
+    value 2562585
+
+network.softnet.received_rps
+    value 0
+
+network.softnet.time_squeeze
+    value 997
+
+network.softnet.total_backlog
+    value 6
+
+== done
+
 == Checking metric values - el7
 
 network.softnet.cpu_collision
@@ -10,6 +41,12 @@ network.softnet.dropped
 network.softnet.flow_limit_count
     value 6
 
+network.softnet.input_qlen
+Error: Metric not supported by this version of monitored application
+
+network.softnet.process_qlen
+Error: Metric not supported by this version of monitored application
+
 network.softnet.processed
     value 6647645
 
@@ -19,6 +56,9 @@ network.softnet.received_rps
 network.softnet.time_squeeze
     value 3
 
+network.softnet.total_backlog
+Error: Metric not supported by this version of monitored application
+
 == done
 
 == Checking metric values - el6
@@ -32,6 +72,12 @@ network.softnet.dropped
 network.softnet.flow_limit_count
 Error: Metric not supported by this version of monitored application
 
+network.softnet.input_qlen
+Error: Metric not supported by this version of monitored application
+
+network.softnet.process_qlen
+Error: Metric not supported by this version of monitored application
+
 network.softnet.processed
     value 43552
 
@@ -41,6 +87,9 @@ network.softnet.received_rps
 network.softnet.time_squeeze
     value 3
 
+network.softnet.total_backlog
+Error: Metric not supported by this version of monitored application
+
 == done
 
 == Checking metric values - el5
@@ -54,6 +103,12 @@ network.softnet.dropped
 network.softnet.flow_limit_count
 Error: Metric not supported by this version of monitored application
 
+network.softnet.input_qlen
+Error: Metric not supported by this version of monitored application
+
+network.softnet.process_qlen
+Error: Metric not supported by this version of monitored application
+
 network.softnet.processed
     value 9009
 
@@ -63,5 +118,8 @@ Error: Metric not supported by this version of monitored application
 network.softnet.time_squeeze
     value 3
 
+network.softnet.total_backlog
+Error: Metric not supported by this version of monitored application
+
 == done
 

src/pmdas/linux/help

@@ -1581,6 +1581,9 @@ See also network.interface.speed for the Megabytes/second value.
 @ network.softnet.cpu_collision number of times that two cpus collided trying to get the device queue lock
 @ network.softnet.received_rps number of times rps_trigger_softirq has been called
 @ network.softnet.flow_limit_count softnet_data flow limit counter
+@ network.softnet.total_backlog sum of softnet input and process queue lengths
+@ network.softnet.input_qlen softnet input packet queue length
+@ network.softnet.process_qlen softnet process queue length
 @ network.softnet.percpu.processed number of packets (not including netpoll) received by the interrupt handler
 @ network.softnet.percpu.dropped number of packets that were dropped because netdev_max_backlog was exceeded
 @ network.softnet.percpu.time_squeeze number of times ksoftirq ran out of netdev_budget or time slice with work remaining
@@ -1591,6 +1594,9 @@ The network stack has to drop packets when a receive processing a CPUs
 backlog reaches netdev_max_backlog.  The flow_limit_count counter is
 the number of times very active flows have dropped their traffic earlier
 to maintain capacity for other less active flows.
+@ network.softnet.percpu.total_backlog sum of softnet input and process queue lengths
+@ network.softnet.percpu.input_qlen softnet input packet queue length
+@ network.softnet.percpu.process_qlen softnet process queue length
 
 @ network.ip6.inreceives count of ip6 inreceives
 @ network.ip6.inhdrerrors count of ip6 inhdrerrors

src/pmdas/linux/linux.h

@@ -305,6 +305,9 @@ typedef struct {
     uint64_t		cpu_collision;
     uint64_t		received_rps;
     uint64_t		flow_limit_count;
+    uint64_t		total_backlog;
+    uint64_t		input_qlen;
+    uint64_t		process_qlen;
 } softnet_t;
 
 typedef struct {

src/pmdas/linux/pmda.c

@@ -7016,6 +7016,30 @@ static pmdaMetric metrictab[] = {
     { NULL, { PMDA_PMID(CLUSTER_NET_SOFTNET,11), PM_TYPE_U64, CPU_INDOM,
       PM_SEM_COUNTER, PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
 
+    /* network.softnet.total_backlog */
+    { NULL, { PMDA_PMID(CLUSTER_NET_SOFTNET,12), PM_TYPE_U64, PM_INDOM_NULL,
+      PM_SEM_INSTANT, PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
+
+    /* network.softnet.percpu.total_backlog */
+    { NULL, { PMDA_PMID(CLUSTER_NET_SOFTNET,13), PM_TYPE_U64, CPU_INDOM,
+      PM_SEM_INSTANT, PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
+
+    /* network.softnet.input_qlen */
+    { NULL, { PMDA_PMID(CLUSTER_NET_SOFTNET,14), PM_TYPE_U64, PM_INDOM_NULL,
+      PM_SEM_INSTANT, PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
+
+    /* network.softnet.percpu.input_qlen */
+    { NULL, { PMDA_PMID(CLUSTER_NET_SOFTNET,15), PM_TYPE_U64, CPU_INDOM,
+      PM_SEM_INSTANT, PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
+
+    /* network.softnet.process_qlen */
+    { NULL, { PMDA_PMID(CLUSTER_NET_SOFTNET,16), PM_TYPE_U64, PM_INDOM_NULL,
+      PM_SEM_INSTANT, PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
+
+    /* network.softnet.percpu.process_qlen */
+    { NULL, { PMDA_PMID(CLUSTER_NET_SOFTNET,17), PM_TYPE_U64, CPU_INDOM,
+      PM_SEM_INSTANT, PMDA_PMUNITS(0,0,1,0,0,PM_COUNT_ONE) }, },
+
 /*
  * tapedev cluster
  */
@@ -9796,6 +9820,42 @@ linux_fetchCallBack(pmdaMetric *mdesc, unsigned int inst, pmAtomValue *atom)
 		return PM_ERR_INST;
 	    atom->ull = cp->softnet->flow_limit_count;
 	    break;
+	case 12: /* network.softnet.total_backlog */
+	    if (!(proc_net_softnet.flags & SN_BACKLOG))
+		return PM_ERR_APPVERSION;
+	    atom->ull = proc_net_softnet.total_backlog;
+	    break;
+	case 13: /* network.softnet.percpu.total_backlog */
+	    if (!(proc_net_softnet.flags & SN_BACKLOG))
+		return PM_ERR_APPVERSION;
+	    if (pmdaCacheLookup(indom, inst, NULL, (void **)&cp) < 0)
+		return PM_ERR_INST;
+	    atom->ull = cp->softnet->total_backlog;
+	    break;
+	case 14: /* network.softnet.input_qlen */
+	    if (!(proc_net_softnet.flags & SN_BACKLOG))
+		return PM_ERR_APPVERSION;
+	    atom->ull = proc_net_softnet.input_qlen;
+	    break;
+	case 15: /* network.softnet.percpu.input_qlen */
+	    if (!(proc_net_softnet.flags & SN_BACKLOG))
+		return PM_ERR_APPVERSION;
+	    if (pmdaCacheLookup(indom, inst, NULL, (void **)&cp) < 0)
+		return PM_ERR_INST;
+	    atom->ull = cp->softnet->input_qlen;
+	    break;
+	case 16: /* network.softnet.process_qlen */
+	    if (!(proc_net_softnet.flags & SN_BACKLOG))
+		return PM_ERR_APPVERSION;
+	    atom->ull = proc_net_softnet.process_qlen;
+	    break;
+	case 17: /* network.softnet.percpu.process_qlen */
+	    if (!(proc_net_softnet.flags & SN_BACKLOG))
+		return PM_ERR_APPVERSION;
+	    if (pmdaCacheLookup(indom, inst, NULL, (void **)&cp) < 0)
+		return PM_ERR_INST;
+	    atom->ull = cp->softnet->process_qlen;
+	    break;
 	default:
 	    return PM_ERR_PMID;
 	}

src/pmdas/linux/proc_net_softnet.c

@@ -23,28 +23,28 @@ refresh_proc_net_softnet(proc_net_softnet_t *all)
     pmInDom	cpus = INDOM(CPU_INDOM);
     percpu_t	*cp;
     softnet_t	*snp;
-    uint64_t	filler;
+    uint64_t	filler, id;
     FILE	*fp;
     static int	logonce;
 
-    /* size > (11*7)+1 bytes, where 7 == strlen("%08llx "), and +1 for '\0' */
+    /* size > (15*7)+1 bytes, where 7 == strlen("%08llx "), and +1 for '\0' */
     static char fmt[128] = { '\0' };
 
     if (fmt[0] == '\0') {
 	/*
 	 * one trip initialization to decide the correct sscanf format
-	 * for a uint64_t data type .. needs to be
-	 * "%08lx %08lx %08lx %08lx %08lx %08lx %08lx %08lx %08lx %08lx %08lx"
+	 * for a uint64_t data type .. needs to be either
+	 * "%08lx %08lx ... %08lx %08lx"
 	 * or
-	 * "%08llx %08llx %08llx %08llx %08llx %08llx %08llx %08llx %08llx %08llx %08llx"
+	 * "%08llx %08llx ... %08llx %08llx"
 	 */
 	fmt[0] = '\0';
 	if (strcmp(FMT_INT64, "lld") == 0) {
-	    for (i = 0; i < 11; i++)
+	    for (i = 0; i < 15; i++)
 		strcat(fmt, "%08llx ");
 	}
 	else {
-	    for (i = 0; i < 11; i++)
+	    for (i = 0; i < 15; i++)
 		strcat(fmt, "%08lx ");
 	}
 	/* chop off last ' ' */
@@ -74,22 +74,32 @@ refresh_proc_net_softnet(proc_net_softnet_t *all)
 	memset(snp, 0, sizeof(*snp));
 	i = sscanf(buf, fmt, &snp->processed, &snp->dropped, &snp->time_squeeze,
 		&filler, &filler, &filler, &filler, &filler,
-		&snp->cpu_collision, &snp->received_rps, &snp->flow_limit_count);
+		&snp->cpu_collision, &snp->received_rps, &snp->flow_limit_count,
+		&snp->total_backlog, &id, &snp->input_qlen, &snp->process_qlen);
+	if (i >= 13 && (int)id != cpu) {
+	    fprintf(stderr, "refresh_proc_net_softnet: warning: inconsistent cpu %d not %d\n", cpu, (int)id);
+	    logonce = 1;
+	}
 
 	/* update aggregate CPU stats as well */
 	all->processed += snp->processed;
 	all->dropped += snp->dropped;
 	all->time_squeeze += snp->time_squeeze;
-	all->cpu_collision += snp->cpu_collision;
+	all->cpu_collision += snp->cpu_collision; /* no longer used in kernel */
 	all->received_rps += snp->received_rps;
 	all->flow_limit_count += snp->flow_limit_count;
+	all->total_backlog += snp->total_backlog;
+	all->input_qlen += snp->input_qlen;
+	all->process_qlen += snp->process_qlen;
 
 	if (i >= 9)
 	    snp->flags |= SN_PROCESSED | SN_DROPPED | SN_TIME_SQUEEZE | SN_CPU_COLLISION;
 	if (i >= 10)
 	    snp->flags |= SN_RECEIVED_RPS;
 	if (i >= 11)
 	    snp->flags |= SN_FLOW_LIMIT_COUNT;
+	if (i >= 15)
+	    snp->flags |= SN_BACKLOG;
 
 	if (cpu > 0 && snp->flags != all->flags && logonce <= 1) {
 	    fprintf(stderr, "refresh_proc_net_softnet: warning: inconsistent flags, cpu %d\n", cpu);

src/pmdas/linux/proc_net_softnet.h

@@ -1,11 +1,11 @@
 /*
- * Copyright (c) 2015-2017 Red Hat.
- * 
+ * Copyright (c) 2015-2017,2024 Red Hat.
+ *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU General Public License as published by the
  * Free Software Foundation; either version 2 of the License, or (at your
  * option) any later version.
- * 
+ *
  * This program is distributed in the hope that it will be useful, but
  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
@@ -19,6 +19,7 @@ enum {
     SN_CPU_COLLISION	= 1<<3,
     SN_RECEIVED_RPS	= 1<<4,
     SN_FLOW_LIMIT_COUNT	= 1<<5,
+    SN_BACKLOG		= 1<<6,
 };
 
 typedef softnet_t proc_net_softnet_t;

src/pmdas/linux/root_linux

@@ -1363,6 +1363,9 @@ network.softnet {
     cpu_collision	60:57:3
     received_rps	60:57:4
     flow_limit_count	60:57:5
+    total_backlog	60:57:12
+    input_qlen		60:57:14
+    process_qlen	60:57:16
     percpu
 }
 
@@ -1373,6 +1376,9 @@ network.softnet.percpu {
     cpu_collision	60:57:9
     received_rps	60:57:10
     flow_limit_count	60:57:11
+    total_backlog	60:57:13
+    input_qlen		60:57:15
+    process_qlen	60:57:17
 }
 
 network.ip6 {