#!/bin/sh
#-*-sh-*-

#
# Copyright © 2009 CNRS
# Copyright © 2009-2022 Inria.  All rights reserved.
# Copyright © 2009-2012 Université Bordeaux
# Copyright © 2014 Cisco Systems, Inc.  All rights reserved.
# See COPYING in top-level directory.
#

HWLOC_VERSION="2.9.3"
prefix="/cvmfs/northgrid.gridpp.ac.uk/simonsobservatory/conda/so-conda-py310-20231220"
exec_prefix="${prefix}"
bindir="${exec_prefix}/bin"
localstatedir="${prefix}/var"
runstatedir="${localstatedir}/run"
# this will be changed into $bindir/lstopo during make install
lstopo="$bindir/lstopo-no-graphics"
hgcpuid="$bindir/hwloc-gather-cpuid"

# make sure we use default numeric formats
LANG=C
LC_ALL=C
export LANG LC_ALL

# don't let ls append special chars after symlinks etc
unalias -a ls

gathercpuid=1
gatherio=0
gatherdmi=0
keep=0

if [ ! -x "$lstopo" ]
then
    error "Could not find lstopo executable in the install or build dir."
    exit 1
fi

error()
{
    echo $@ 2>&1
}

usage()
{
   echo `basename $0`" [options] <savepath>"
   echo "  Saves the Linux topology files (/sys, /proc, ...) under <savepath>.tar.bz2"
   echo "  and the corresponding lstopo verbose output under <savepath>.output"
   echo "Options:"
   echo "  --io        Gather I/O files (takes much longer and generates much larger tarball)"
   echo "  --dmi       Gather SMBIOS files. Works only when run as root. Requires dmi-sysfs kernel module"
   echo "  --no-cpuid  Do not gather x86 CPUID using hwloc-gather-cpuid"
   echo "  --keep      Keep the temporary copy of dumped files"
   echo "  --version   Report version and exit"
   echo "  -h --help   Show this usage"
   echo "Example:"
   echo "  $0 /tmp/\$(uname -n)"
}

while [ x`echo "$1" | cut -c1` = x- ] ; do
  case $1 in
  --io) gatherio=1;;
  --dmi) gatherdmi=1;;
  --no-cpuid) gathercpuid=0;;
  --keep) keep=1;;
  --version) echo `basename $0`" $HWLOC_VERSION"; exit 0;;
  -h|--help) usage; exit 0;;
  *) echo "Unrecognized option: $1"; usage; exit 1;;
  esac
  shift
done

if test $# -lt 1 -o x$1 = x; then
  usage
  exit 1
fi
name="$1"
basename=`basename "$name"`
dirname=`dirname "$name"`

if [ x$gatherio = x0 ]; then
  echo "I/O files won't be saved (--io not given)."
fi
if [ x$gatherdmi = x0 ]; then
  echo "DMI files won't be saved (--dmi not given)."
fi

echo
echo "*** Note that this tool may be slow on large nodes or when I/O is enabled. ***"
echo

if ! mkdir -p "$dirname" ; then
    error "Failed to create directory $dirname."
    exit 1
fi

if [ ! -w  "$dirname" ] ; then
    echo "$dirname is not writable."
    exit 1
fi

destdir=`mktemp -d --tmpdir hwloc-gather-topology.XXXXXXXX`

# Use cat so that we properly get proc/sys files even if their
# file length is wrong
_savefile() {
  local dest="$1"
  local file="$2"
  if test -r "$file"; then
    dir=`dirname "$file"`
    mkdir -p "$dest/$dir" 2>/dev/null
    cat "$file" > "$dest/$file" 2>/dev/null
  fi
}

_savelink() {
  local dest="$1"
  local file="$2"
  dir=`dirname "$file"`
  mkdir -p "$dest/$dir" 2>/dev/null
  cp -P "$file" "$dest/$file"
}

# Get a file
savefile() {
  local dest="$1"
  local file="$2"
  echo " file $file"
  _savefile "$dest" "$file"
}

# Get all files from the given directory
# Ignore errors since some files may be missing, and some may be
# restricted to root (but we don't need them).
savedir() {
  local dest="$1"
  local path="$2"
  echo " directory $path"
  # gather all directories, including empty ones
  find "$path" -type d 2>/dev/null | while read -r dir ; do
    mkdir -p "$dest/$dir" 2>/dev/null
  done
  # gather all files now
  find "$path" -type f 2>/dev/null | while read -r file ; do
    test -e "$dest/$file" || _savefile "$dest" "$file"
  done
  # gather symlinks
  find "$path" -type l 2>/dev/null | while read -r link ; do
    test -e "$dest/$link" || _savelink "$dest" "$link"
  done
}

# Save a class directory, including symlink targets (except in /devices/pci) and their device symlinks if any
saveclassdir() {
  local dest="$1"
  local classname="$2"

  test -d /sys/class/$classname/ || return

  savedir "$dest" /sys/class/$classname/

  for dev in $(ls /sys/class/$classname/); do
    target=$(readlink /sys/class/$classname/$dev 2>/dev/null | grep -v /devices/pci)
    if test "x$target" != x; then
       savedir "$dest" "/sys/class/$classname/$target"
    fi
    device=$(readlink /sys/class/$classname/$target/device 2>/dev/null | grep -v /devices/pci)
    if test "x$device" != x; then
       savedir "$dest" "/sys/class/$classname/$target/$device"
    fi
  done
}

# Save a bus devices directory, include symlinks target (except in /devices/pci) with a possible relative path, and their device symlink if any
savebusdir() {
  local dest="$1"
  local busname="$2"
  local relpath="$3"
  local withdriver="$4"

  test -d /sys/bus/$busname/devices/ || return

  savedir "$dest" /sys/bus/$busname/devices/

  if test "x$withdriver" = x1; then
    savedir "$dest" /sys/bus/$busname/drivers/
  fi

  for dev in $(ls /sys/bus/$busname/devices/); do
    target=$(readlink /sys/bus/$busname/devices/$dev 2>/dev/null | grep -v /devices/pci)
    if test "x$target" != x; then
       savedir "$dest" "/sys/bus/$busname/devices/$target/$relpath"
    fi
    device=$(readlink /sys/bus/$busname/devices/$target/$relpath/device 2>/dev/null | grep -v /devices/pci)
    if test "x$device" != x; then
       savedir "$dest" "/sys/bus/$busname/devices/$target/$relpath/$device"
    fi
  done
}

# savedir "$destdir/$basename" /sys/bus/dax/devices/
# readlink /sys/bus/dax/devices/* 2>/dev/null | grep -v /devices/pci | while read -r path ; do savedir "$destdir/$basename" "/sys/bus/dax/devices/$path/.." ; done

# Get an entire mount point, after decoding its path
# we don't support path with \n since it would break in 'find ... | while read ..." above
savemntpnt() {
  local encodedpath=$1
  if echo "$1" | grep "\\012" ; then
    echo "Ignoring mount point whose filename contains an end of line."
    return
  fi
  local path=$(echo "$1" | sed -e 's@\\134@\\@g' -e 's@\\040@ @g' -e 's@\\011@	@g')
  savedir "$destdir/$basename" "${path}/"
}

#
# Main stuff
#
echo "Gathering main files and directories..."

# Gather the following list of files
savefile "$destdir/$basename" /proc/cmdline
savefile "$destdir/$basename" /proc/cpuinfo
savefile "$destdir/$basename" /proc/meminfo
savefile "$destdir/$basename" /proc/mounts
savefile "$destdir/$basename" /proc/stat
savefile "$destdir/$basename" /proc/version
savefile "$destdir/$basename" /proc/self/cpuset
savefile "$destdir/$basename" /proc/self/cgroup
savefile "$destdir/$basename" /proc/driver/nvidia

# Gather cpu and node information
# (no need for savebusdir since we already gather the target /sys/devices/system/{cpu,node} explicitly)
savedir "$destdir/$basename" /sys/devices/system/cpu/
savedir "$destdir/$basename" /sys/bus/cpu/devices/
savedir "$destdir/$basename" /sys/devices/system/node/
savedir "$destdir/$basename" /sys/bus/node/devices/

# Gather DMI IDs
# (no need for aveclassdir since we only want "id" and it usually points to /sys/devices/virtual/dmi/id/)
savedir "$destdir/$basename" /sys/class/dmi/id/
savedir "$destdir/$basename" /sys/devices/virtual/dmi/id/

# Gather hugepage information
savedir "$destdir/$basename" /sys/kernel/mm/hugepages/

# Gather the default /var/run/hwloc (in case it was created by a system-wide hwloc-dump-hwdata)
if test -d /var/run/hwloc; then
  savedir "$destdir/$basename" /var/run/hwloc
fi
# Then, gather what the current hwloc installation could have created in a different $runstatedir
if test "$runstatedir" != "/var/run" -a -d "$runstatedir/hwloc"; then
  savedir "$destdir/$basename" "$runstatedir/hwloc/"
  mkdir -p "$destdir/$basename/var/run"
  ln -sr "$destdir/$basename/$runstatedir/hwloc" "$destdir/$basename/var/run/hwloc.runstatedir"
fi
# And gather what custom $HWLOC_DUMPED_HWDATA_DIR
if test "x$HWLOC_DUMPED_HWDATA_DIR" != x -a -d "$HWLOC_DUMPED_HWDATA_DIR"; then
  savedir "$destdir/$basename" "$HWLOC_DUMPED_HWDATA_DIR"
  mkdir -p "$destdir/$basename/var/run"
  ln -sr "$destdir/$basename/$HWLOC_DUMPED_HWDATA_DIR" "$destdir/$basename/var/run/hwloc.HWLOC_DUMPED_HWDATA_DIR"
fi
# Now link to /var/run/hwloc (used by default in lstopo -i) to something sane if needed
if ! test -d "$destdir/$basename/var/run/hwloc"; then
  if test -e "$destdir/$basename/var/run/hwloc.HWLOC_DUMPED_HWDATA_DIR"; then
    ln -sr "$destdir/$basename/var/run/hwloc.HWLOC_DUMPED_HWDATA_DIR" "$destdir/$basename/var/run/hwloc"
  else if test -e "$destdir/$basename/$runstatedir/hwloc"; then
    ln -sr "$destdir/$basename/var/run/hwloc.runstatedir" "$destdir/$basename/var/run/hwloc"
  fi fi
fi

# Gather cgroup/cpuset mntpnts
cat /proc/mounts | while read -r dummy1 mntpath mnttype mntopts dummy2 ; do
  [ x$mnttype = xcpuset ] && savemntpnt "$mntpath"
  [ x$mnttype = xcgroup ] && echo $mntopts | grep -w cpuset >/dev/null && savemntpnt "$mntpath"
  [ x$mnttype = xcgroup2 ] && savemntpnt "$mntpath"
done

#
# Optionally gather I/O directories too
#
if [ x$gatherio = x1 ]; then
  echo "Gathering I/O files..."
  # gather all PCI stuff, lots of links (/sys/bus/pci/devices/* and /sys/class/foo/*) point there
  savedir "$destdir/$basename" /sys/bus/pci/devices/
  savedir "$destdir/$basename" /sys/bus/pci/slots/
  ls -d /sys/devices/pci* 2>/dev/null | while read -r path ; do savedir "$destdir/$basename" "$path" ; done
  # gather class and bus links, we'll parse that the target path for PCI busids
  savebusdir "$destdir/$basename" cxl . 1
  saveclassdir "$destdir/$basename" block
  saveclassdir "$destdir/$basename" dax
  savebusdir "$destdir/$basename" dax .. 1
  saveclassdir "$destdir/$basename" dma
  saveclassdir "$destdir/$basename" drm
  saveclassdir "$destdir/$basename" infiniband
  saveclassdir "$destdir/$basename" net
  saveclassdir "$destdir/$basename" ve
  saveclassdir "$destdir/$basename" bxi
  saveclassdir "$destdir/$basename" mic
  # udev block data
  ls -d /run/udev/data/b* 2>/dev/null | while read -r path ; do savefile "$destdir/$basename" "$path" ; done
fi

#
# Optionally gather DMI directories too
#
if [ x$gatherdmi = x1 ]; then
  echo "Gathering DMI files..."
  savedir "$destdir/$basename" /sys/firmware/dmi/
fi

#
# Export /proc/hwloc-nofile-info during lstopo (needs HWLOC_DUMP_NOFILE_INFO with HWLOC_THISSYSTEM=1)
#
echo "Exporting /proc/hwloc-nofile-info"
export HWLOC_DUMP_NOFILE_INFO="$destdir/$basename/proc/hwloc-nofile-info"
"$lstopo" - >/dev/null
# disable HWLOC_DUMP_NOFILE_INFO for next lstopo invocation
export HWLOC_DUMP_NOFILE_INFO=

#
# Export cpuid if available
#
if [ x$gathercpuid = x1 -a -e "$hgcpuid" ]; then
  echo
  echo "Exporting x86 CPUID using hwloc-gather-cpuid"
  $hgcpuid --silent "$destdir/$basename/cpuid"
fi

# Create the archive and optionally keep the tree in /tmp for testing
echo
( cd "$destdir/" && tar cfj "$basename.tar.bz2" "$basename" )
mv "$destdir/$basename.tar.bz2" "$dirname/$basename.tar.bz2"
if test x$keep = x1; then
  echo "Topology files gathered in $dirname/$basename.tar.bz2 and kept in $destdir/$basename/"
else
  echo "Topology files gathered in $dirname/$basename.tar.bz2"
  rm -rf "$destdir"
fi

# Generate the output as well
# we need "Topology not from this system" in the output so as to make test-topology.sh happy
export HWLOC_THISSYSTEM=0
"$lstopo" - -v > "$dirname/$basename.output"
echo "Expected topology output stored in $dirname/$basename.output"
"$lstopo" -.xml --whole-io --disallowed > "$dirname/$basename.xml"
echo "XML topology stored in $dirname/$basename.xml"

echo
echo "WARNING: Do not post these files on a public list or website unless you"
echo "WARNING: are sure that no information about this platform is sensitive."

exit 0