Support >32 vCPU's in set_multiqueue. (#375)

2 files changed, 41 insertions, 38 deletions

diff --git a/build_packages.sh b/build_packages.sh
index a2c5cef..6205077 100755
--- a/build_packages.sh
+++ b/build_packages.sh
@@ -42,6 +42,7 @@ function build_distro() {
   fpm \
     -s python \
     -t "${pkg_type}" \
+    --conflicts 'irqbalance' \
     --depends "${depends}" \
     --depends 'python-boto' \
     --depends 'python-setuptools' \
diff --git a/scripts/set_multiqueue b/scripts/set_multiqueue
index 636ee90..9dfe45f 100755
--- a/scripts/set_multiqueue
+++ b/scripts/set_multiqueue
@@ -1,5 +1,5 @@
 #!/bin/bash
-# Copyright 2016 Google Inc. All Rights Reserved.
+# Copyright 2017 Google Inc. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -35,7 +35,7 @@ function set_channels() {
   ethtool -L "${1}" combined "${2}" > /dev/null 2>&1
 }
 
-echo "Running $(basename "$0")."
+echo "Running $(basename $0)."
 NET_DEVS=/sys/bus/virtio/drivers/virtio_net/virtio*
 
 # Loop through all the virtionet devices and enable multi-queue
@@ -68,7 +68,7 @@ do
     irq_dir=/proc/irq/*
     for irq in $irq_dir
     do
-      smp_affinity="${irq}/smp_affinity"
+      smp_affinity="${irq}/smp_affinity_list"
       [ ! -f "${smp_affinity}" ] && continue
       # Classify this IRQ as virtionet intx, virtionet MSI-X, or non-virtionet
       # If the IRQ type is virtionet intx, a subdirectory with the same name as
@@ -78,7 +78,7 @@ do
       # a decimal integer ranging from 0 to K - 1 where K is the number of
       # input (output) queues in the virtionet device.
       virtionet_intx_dir="${irq}/${dev}"
-      virtionet_msix_dir_regex=".*/${dev}-(input|output)\.[0-9]+$"
+      virtionet_msix_dir_regex=".*/${dev}-(input|output)\.([0-9]+)$"
       if [ -d "${virtionet_intx_dir}" ]; then
         # All virtionet intx IRQs are delivered to CPU 0
         echo "Setting ${smp_affinity} to 01 for device ${dev}."
@@ -90,44 +90,46 @@ do
       for entry in ${irq}/${dev}*; do
         if [[ "$entry" =~ ${virtionet_msix_dir_regex} ]]; then
           virtionet_msix_found=1
+          queue_num=${BASH_REMATCH[2]}
         fi
       done
       affinity_hint="${irq}/affinity_hint"
       [ "$virtionet_msix_found" -eq 0 -o ! -f "${affinity_hint}" ] && continue
 
-      # The affinity hint file contains a CPU mask, consisting of
-      # groups of up to 8 hexadecimal digits, separated by a comma. Each bit
-      # position in the CPU mask hex value specifies whether this interrupt
-      # should be delivered to the corresponding CPU. For example, if bits 0
-      # and 3 are set in the affinity hint CPU mask hex value, then the
-      # interrupt should be delivered to CPUs 0 and 3. The virtionet device
-      # driver should set only a single bit in the affinity hint per MSI-X
-      # interrupt, ensuring each TX (RX) queue is used only by a single CPU.
-      # The virtionet driver will only specify an affinity hint if the number of
-      # TX (RX) queues equals the number of online CPUs. If no affinity hint is
-      # specified for an IRQ, the affinity hint file will contain all zeros.
-      affinity_cpumask=$(cat "${affinity_hint}")
-      affinity_hint_enabled=0
-      # Parse the affinity hint, skip if mask is invalid or is empty (all-zeros)
-      OIFS=${IFS}
-      IFS=","
-      for cpu_bitmap in ${affinity_cpumask}; do
-        bitmap_val=$(printf "%d" "0x${cpu_bitmap}" 2>/dev/null)
-        if [ "$?" -ne 0 ]; then
-          echo "Invalid affinity hint ${affinity_hint}: ${affinity_cpumask}."
-          affinity_hint_enabled=0
-          break
-        elif [ "${bitmap_val}" -ne 0 ]; then
-          affinity_hint_enabled=1
-        fi
-      done
-      IFS=${OIFS}
-      if [ "${affinity_hint_enabled}" -eq 0 ]; then
-        echo "Cannot set IRQ affinity ${smp_affinity}, affinity hint disabled."
-      else
-        # Set the IRQ CPU affinity to the virtionet-initialized affinity hint
-        echo "Setting ${smp_affinity} to ${affinity_cpumask} for device ${dev}."
-        echo "${affinity_cpumask}" > "${smp_affinity}"
-      fi
+      # Set the IRQ CPU affinity to the virtionet-initialized affinity hint
+      echo "Setting ${smp_affinity} to ${queue_num} for device ${dev}."
+      echo "${queue_num}" > "${smp_affinity}"
+      real_affinity=`cat ${smp_affinity}`
+      echo "${smp_affinity}: real affinity ${real_affinity}"
     done
 done
+
+XPS=/sys/class/net/e*/queues/tx*/xps_cpus
+num_cpus=$(nproc)
+
+num_queues=0
+for q in $XPS; do
+  num_queues=$((num_queues + 1))
+done
+
+# If we have more CPUs than queues, then stripe CPUs across tx affinity
+# as CPUNumber % queue_count.
+for q in $XPS; do
+  queue_re=".*tx-([0-9]+).*$"
+  if [[ "$q" =~ ${queue_re} ]]; then
+    queue_num=${BASH_REMATCH[1]}
+  fi
+
+  xps=0
+  for cpu in `seq $queue_num $num_queues $((num_cpus - 1))`; do
+    xps=$((xps | (1 << cpu)))
+  done
+
+  # Linux xps_cpus requires a hex number with commas every 32 bits.
+  # It ignores all bits above # cpus, so unconditionally write a
+  # 64 bit hex value, with a comma between dwords.
+  xps_string=`printf "%08x,%08x" $((xps >> 32 & 0xffffffff)) $((xps & 0xffffffff))`
+
+  echo ${xps_string} > $q
+  printf "Queue %d XPS=%s for %s\n" $queue_num `cat $q` $q
+done | sort -n -k2