qualcommax: pin EDMA processing to dedicated CPUs

IPQ60xx and IPQ807x use threaded NAPI for EDMA. The NAPI threads
inherit CPU0 affinity, and running both there can eventually starve the
RX fill ring under sustained traffic.

Add an IPQ60xx smp_affinity service and extend the existing IPQ807x
service to discover the EDMA IRQs and NAPI IDs at runtime. Place RX on
CPU1 and TX on CPU2. Retry EDMA setup asynchronously because its NAPI
threads may appear after the init service runs.

Provide the same UCI enable and logging controls on both subtargets.
IPQ50xx uses a different Ethernet driver and is left unchanged.

Link: https://github.com/openwrt/openwrt/pull/24203
Signed-off-by: Robert Marko <robimarko@gmail.com>

target/linux/qualcommax/ipq60xx/base-files/etc/init.d/smp_affinity

@@ -0,0 +1,105 @@
+#!/bin/sh /etc/rc.common
+
+START=94
+
+PROG=smp_affinity
+
+log_msg() {
+	local type="$1" id="$2" name="$3" affinity="$4"
+
+	logger -t "$PROG" "Pinning $type($id) $name to CPU $affinity"
+}
+
+get_irq() {
+	awk -v name="$1" '$NF == name {
+		sub(/:$/, "", $1)
+		print $1
+		exit
+	}' /proc/interrupts
+}
+
+get_napi_threads() {
+	local comm comm_path id pid
+
+	TX_NAPI_ID=
+	TX_PID=
+	RX_NAPI_ID=
+	RX_PID=
+
+	for comm_path in /proc/[0-9]*/comm; do
+		read -r comm < "$comm_path"
+		case "$comm" in
+		napi/eth0-[0-9]*)
+			id="${comm##*-}"
+			pid="${comm_path#/proc/}"
+			pid="${pid%/comm}"
+
+			if [ -z "$TX_NAPI_ID" ] || [ "$id" -lt "$TX_NAPI_ID" ]; then
+				TX_NAPI_ID="$id"
+				TX_PID="$pid"
+			fi
+
+			if [ -z "$RX_NAPI_ID" ] || [ "$id" -gt "$RX_NAPI_ID" ]; then
+				RX_NAPI_ID="$id"
+				RX_PID="$pid"
+			fi
+			;;
+		esac
+	done
+}
+
+pin_irq() {
+	local irq="$1" mask="$2" affinity="$3" name="$4"
+
+	[ -n "$irq" ] || return
+	$enable_log && log_msg IRQ "$irq" "$name" "$affinity"
+	echo "$mask" > "/proc/irq/$irq/smp_affinity"
+}
+
+pin_thread() {
+	local pid="$1" mask="$2" affinity="$3" name="$4"
+
+	[ -n "$pid" ] || return
+	$enable_log && log_msg thread "$pid" "$name" "$affinity"
+	taskset -p "$mask" "$pid" >/dev/null
+}
+
+enable_affinity_edma() {
+	local seconds=0 rx_irq tx_irq
+
+	while [ "$seconds" -lt 60 ]; do
+		rx_irq="$(get_irq edma_rxdesc)"
+		tx_irq="$(get_irq edma_txcmpl)"
+		get_napi_threads
+
+		[ -n "$rx_irq" ] && [ -n "$tx_irq" ] &&
+			[ -n "$RX_PID" ] && [ -n "$TX_PID" ] &&
+			[ "$RX_PID" != "$TX_PID" ] && break
+
+		seconds=$((seconds + 1))
+		sleep 1
+	done
+
+	[ -n "$rx_irq" ] && [ -n "$tx_irq" ] &&
+		[ -n "$RX_PID" ] && [ -n "$TX_PID" ] &&
+		[ "$RX_PID" != "$TX_PID" ] || return
+
+	pin_irq "$rx_irq" 2 1 edma_rxdesc
+	pin_thread "$RX_PID" 2 1 "EDMA RX NAPI"
+	pin_irq "$tx_irq" 4 2 edma_txcmpl
+	pin_thread "$TX_PID" 4 2 "EDMA TX NAPI"
+}
+
+boot() {
+	local enable
+
+	config_load smp_affinity
+
+	config_get_bool enable "general" enable 1
+	config_get_bool enable_log "general" enable_log 1
+
+	[ "$enable" -eq 1 ] && enable=true || enable=false
+	[ "$enable_log" -eq 1 ] && enable_log=true || enable_log=false
+
+	$enable && enable_affinity_edma &
+}

target/linux/qualcommax/ipq60xx/base-files/etc/uci-defaults/15_smp_affinity

@@ -0,0 +1,11 @@
+#!/bin/sh
+
+uci -q get smp_affinity && exit 0
+touch /etc/config/smp_affinity
+
+uci -q batch << EOF
+  set smp_affinity.general=smp_affinity
+  set smp_affinity.general.enable='1'
+  set smp_affinity.general.enable_log='1'
+  commit smp_affinity
+EOF

target/linux/qualcommax/ipq807x/base-files/etc/init.d/smp_affinity

@@ -81,6 +81,68 @@ set_affinity() {
     done
 }
 
+get_napi_threads() {
+
+  local comm comm_path id pid
+
+  TX_NAPI_ID=
+  TX_PID=
+  RX_NAPI_ID=
+  RX_PID=
+
+  for comm_path in /proc/[0-9]*/comm; do
+    read -r comm < "$comm_path"
+    case "$comm" in
+      napi/eth0-[0-9]*)
+        id="${comm##*-}"
+        pid="${comm_path#/proc/}"
+        pid="${pid%/comm}"
+
+        if [ -z "$TX_NAPI_ID" ] || [ "$id" -lt "$TX_NAPI_ID" ]; then
+          TX_NAPI_ID="$id"
+          TX_PID="$pid"
+        fi
+
+        if [ -z "$RX_NAPI_ID" ] || [ "$id" -gt "$RX_NAPI_ID" ]; then
+          RX_NAPI_ID="$id"
+          RX_PID="$pid"
+        fi
+        ;;
+    esac
+  done
+}
+
+set_thread_affinity() {
+
+  local pid="$1" affinity="$2" bitmask
+
+  [ -n "$pid" ] || return
+  bitmask=$(cpus_to_bitmask "$affinity") &&
+    taskset -p "$bitmask" "$pid" >/dev/null
+}
+
+enable_affinity_edma() {
+
+  local seconds=0
+
+  while [ "$seconds" -lt 60 ]; do
+    get_napi_threads
+    [ -n "$RX_PID" ] && [ -n "$TX_PID" ] &&
+      [ "$RX_PID" != "$TX_PID" ] && break
+
+    seconds=$((seconds + 1))
+    sleep 1
+  done
+
+  [ -n "$RX_PID" ] && [ -n "$TX_PID" ] &&
+    [ "$RX_PID" != "$TX_PID" ] || return
+
+  set_affinity 'edma_rxdesc' 1
+  set_thread_affinity "$RX_PID" 1
+  set_affinity 'edma_txcmpl' 2
+  set_thread_affinity "$TX_PID" 2
+}
+
 enable_affinity_ipq807x() {
 
   # assign 4 rx interrupts to each core
@@ -99,11 +161,6 @@ enable_affinity_ipq807x() {
   set_affinity 'ppdu-end-interrupts-mac2'      2
   set_affinity 'ppdu-end-interrupts-mac3'      3
 
-  # assign 4 lan/wan to core 4
-  set_affinity 'edma_txcmpl'                   3
-  set_affinity 'edma_rxfill'                   3
-  set_affinity 'edma_rxdesc'                   3
-  set_affinity 'edma_misc'                     3
 }
 
 boot() {
@@ -118,5 +175,8 @@ boot() {
   [ "$enable"     -eq 1 ] && enable=true     || enable=false
   [ "$enable_log" -eq 1 ] && enable_log=true || enable_log=false
 
-  $enable && enable_affinity_ipq807x
+  $enable && {
+    enable_affinity_ipq807x
+    enable_affinity_edma &
+  }
 }