contrib/ofed/management/infiniband-diags/scripts/ibcheckerrors.in

   1 #!/bin/sh
   2
   3 IBPATH=${IBPATH:-@IBSCRIPTPATH@}
   4
   5 function usage() {
   6         echo Usage: `basename $0` "[-h] [-b] [-v] [-N | -nocolor]"\
   7             "[<topology-file> | -C ca_name -P ca_port -t(imeout) timeout_ms]"
   8         exit -1
   9 }
  10
  11 function user_abort() {
  12         echo "Aborted"
  13         exit 1
  14 }
  15
  16 trap user_abort SIGINT
  17
  18 gflags=""
  19 verbose=""
  20 brief=""
  21 v=0
  22 ntype=""
  23 nodeguid=""
  24 topofile=""
  25 ca_info=""
  26
  27 while [ "$1" ]; do
  28         case $1 in
  29         -h)
  30                 usage
  31                 ;;
  32         -N|-nocolor)
  33                 gflags=-N
  34                 ;;
  35         -v)
  36                 verbose=-v
  37                 brief=""
  38                 v=1
  39                 ;;
  40         -b)
  41                 brief=-b
  42                 verbose=""
  43                 ;;
  44         -P | -C | -t | -timeout)
  45                 case $2 in
  46                 -*)
  47                         usage
  48                         ;;
  49                 esac
  50                 if [ x$2 = x ] ; then
  51                         usage
  52                 fi
  53                 ca_info="$ca_info $1 $2"
  54                 shift
  55                 ;;
  56         -*)
  57                 usage
  58                 ;;
  59         *)
  60                 if [ "$topofile" ]; then
  61                         usage
  62                 fi
  63                 topofile="$1"
  64                 ;;
  65         esac
  66         shift
  67 done
  68
  69 if [ "$topofile" ]; then
  70         netcmd="cat $topofile"
  71 else
  72         netcmd="$IBPATH/ibnetdiscover $ca_info"
  73 fi
  74
  75 text="`eval $netcmd`"
  76 rv=$?
  77 echo "$text" | awk '
  78 BEGIN {
  79         ne=0
  80 }
  81 function check_node(lid, port)
  82 {
  83         if (system("'$IBPATH'/ibchecknode '"$ca_info"' '$gflags' '$verbose' " lid)) {
  84                 ne++
  85                 print "\n# " ntype ": nodeguid 0x" nodeguid " failed"
  86                 return 1;
  87         }
  88         if (system("'$IBPATH'/ibcheckerrs '"$ca_info"' '$gflags' '$verbose' '$brief' " lid " " port))
  89                 return 2;
  90         return 0;
  91 }
  92
  93 /^Ca/ || /^Switch/ || /^Rt/ {
  94                         nnodes++
  95                         ntype=$1; nodeguid=substr($3, 4, 16); ports=$2
  96                         if ('$v')
  97                                 print "\n# Checking " ntype ": nodeguid 0x" nodeguid
  98
  99                         err = 0;
 100                         if (ntype != "Switch")
 101                                 next
 102
 103                         lid = substr($0, index($0, "port 0 lid ") + 11)
 104                         lid = substr(lid, 1, index(lid, " ") - 1)
 105                         err = check_node(lid, 255)
 106                 }
 107 /^\[/   {
 108                 nports++
 109                 port = $1
 110                 sub("\\(.*\\)", "", port)
 111                 gsub("[\\[\\]]", "", port)
 112                 if (ntype != "Switch") {
 113                         lid = substr($0, index($0, " lid ") + 5)
 114                         lid = substr(lid, 1, index(lid, " ") - 1)
 115                         if (check_node(lid, port) == 2)
 116                                 pcnterr++;
 117                 } else if (err &&
 118                            system("'$IBPATH'/ibcheckerrs '"$ca_info"' '$gflags' '$verbose' '$brief' " lid " " port))
 119                         pcnterr++;
 120 }
 121
 122 /^ib/   {print $0; next}
 123 /ibpanic:/      {print $0}
 124 /ibwarn:/       {print $0}
 125 /iberror:/      {print $0}
 126
 127 END {
 128         printf "\n## Summary: %d nodes checked, %d bad nodes found\n", nnodes, ne
 129         printf "##          %d ports checked, %d ports have errors beyond threshold\n", nports, pcnterr
 130         exit (ne + pcnterr)
 131 }
 132 '
 133 exit $rv