latency-test, latency-histogram: warn when rtapi_app lacks RT privileges

When a run-in-place build is used without 'sudo make setuid' (or
'sudo make setcap'), rtapi_app runs unprivileged: realtime threads get
neither SCHED_FIFO scheduling nor locked memory. The resulting latency
readings are wildly inflated and do not reflect the real hardware, which
is easy to mistake for a hardware or kernel problem.

Add a check in both tools that, for a non-root user, locates rtapi_app
and verifies it is either setuid root or carries the cap_sys_nice file
capability. If neither holds, print a warning that the numbers may be
skewed and point at the missing setuid/setcap step. The check stays
silent when run as root or when rtapi_app cannot be located.

Closes #4044

Author: grandixximo

scripts/latency-histogram

@@ -111,6 +111,31 @@ proc which_exe {name} {
   return -code error "$name: executable not found"
 } ;# which_exe
 
+proc check_rt_privileges {} {
+  # Realtime threads need SCHED_FIFO scheduling and locked memory, which
+  # rtapi_app can only obtain when it is setuid root ('sudo make setuid') or
+  # carries file capabilities ('sudo make setcap').  Without them rtapi_app
+  # runs unprivileged and the measured latency is wildly inflated and
+  # meaningless.  See https://github.com/LinuxCNC/linuxcnc/issues/4044
+  if {![catch {exec id -u} uid] && $uid == 0} return ;# root has all privileges
+  if {[catch {which_exe rtapi_app} rtapi]} return ;# cannot locate, stay quiet
+  # setuid-root bit ('make setuid')
+  if {![catch {file attributes $rtapi -permissions} mode] && ($mode & 04000)} return
+  # file capabilities ('make setcap')
+  if {![catch {exec getcap $rtapi} caps] && [string match *cap_sys_nice* $caps]} return
+  set msg "Warning: rtapi_app is not setuid root and has no realtime capabilities."
+  append msg "\nRealtime threads cannot get SCHED_FIFO priority or locked memory,"
+  append msg "\nso the latency shown will be far worse than reality."
+  append msg "\nOn a run-in-place build, run 'sudo make setuid' (or 'sudo make setcap') in src/."
+  append msg "\nSee https://github.com/LinuxCNC/linuxcnc/issues/4044"
+  puts stderr $msg
+  if {$::LH(use_x)} {
+    package require Tk
+    tk_messageBox -parent . -icon warning -type ok \
+        -title "Latency Warning" -message $msg
+  }
+} ;# check_rt_privileges
+
 proc program_check {plist} {
   foreach prog $plist {
     if [catch {
@@ -944,6 +969,7 @@ proc windowToFile { win } {
 if ![info exists ::LH(start)] {
   set_defaults
   config
+  check_rt_privileges
   progress "Loading packages"
   load_packages
   signal trap SIGINT finish

scripts/latency-test

@@ -16,6 +16,31 @@ kversion=$(uname -v)
 calc() { awk "BEGIN { print ($1); }" < /dev/null; }
 icalc() { awk "BEGIN { printf \"%.0f\n\", ($1); }" < /dev/null; }
 
+# Realtime threads need SCHED_FIFO scheduling and locked memory, which
+# rtapi_app can only obtain when it is setuid root ('sudo make setuid') or
+# carries file capabilities ('sudo make setcap').  Without them rtapi_app runs
+# unprivileged and the measured latency is wildly inflated and meaningless.
+# See https://github.com/LinuxCNC/linuxcnc/issues/4044
+check_rt_privileges () {
+    # root already has every privilege rtapi_app needs
+    [ "$(id -u)" = 0 ] && return 0
+    rtapi_app=$(command -v rtapi_app) || return 0
+    [ -n "$rtapi_app" ] || return 0
+    # setuid-root bit ('make setuid')
+    [ -u "$rtapi_app" ] && return 0
+    # file capabilities ('make setcap')
+    if command -v getcap >/dev/null 2>&1; then
+        case $(getcap "$rtapi_app" 2>/dev/null) in
+            *cap_sys_nice*) return 0 ;;
+        esac
+    fi
+    echo "Warning: rtapi_app is not setuid root and has no realtime capabilities." >&2
+    echo "         Realtime threads cannot get SCHED_FIFO priority or locked memory," >&2
+    echo "         so the latency numbers shown below will be far worse than reality." >&2
+    echo "         On a run-in-place build, run 'sudo make setuid' (or 'sudo make setcap')" >&2
+    echo "         in the src/ directory.  See https://github.com/LinuxCNC/linuxcnc/issues/4044" >&2
+}
+
 parse_time () {
     case $1 in
     -)   echo "0" ;;
@@ -170,4 +195,6 @@ L=\$(($SIGNALS
 echo \$L > "$HOME"/.latency
 EOF
 
+check_rt_privileges
+
 halrun lat.hal