guest-diag: add --fwsec to the supervised claim harness (stage-3c write run)

1 files changed, 35 insertions, 2 deletions

diff --git a/tools/guest-diag/phys_claim_cycle.sh b/tools/guest-diag/phys_claim_cycle.sh
index 4ecd1fa..898aa15 100755
--- a/tools/guest-diag/phys_claim_cycle.sh
+++ b/tools/guest-diag/phys_claim_cycle.sh
@@ -18,7 +18,7 @@
 # plan ready. The harness itself never retries an aborted cycle.
 #
 # Usage: phys_claim_cycle.sh --yes [--hold SECONDS] [--cycles N] [--layout]
-#                             [--dryrun] [--ko PATH] [--no-restore]
+#                             [--dryrun] [--fwsec] [--ko PATH] [--no-restore]
 #
 #   --yes        required: acknowledge the host-freeze hazard
 #   --hold S     seconds to keep the GPU claimed (default 30)
@@ -29,6 +29,11 @@
 #                READ-ONLY stage-3c validation (VBIOS/BIT/FWSEC parse and the
 #                staging/placement plan, no device writes); expects the
 #                "gsp-phys-fwsec-dryrun:" and "stage 3c dry-run complete" lines
+#   --fwsec      also pass gsp_phys_fwsec=1 (implies --layout): run the REAL
+#                stage-3c write sequence (GSP falcon reset, WPR2 carve, FWSEC);
+#                expects "stage 3c complete" and, on release, the WPR2-down
+#                teardown line; a failed teardown keeps the locks and refuses
+#                the nvidia rebind (the cycle goes red; do not retry)
 #   --ko PATH    module to insmod (default kernel/open-iov-vfio.ko)
 #   --no-restore do not restore power/control=auto or restart persistenced
 #                at the end of each cycle (leaves the D0 hold in place)
@@ -53,6 +58,7 @@ HOLD=30
 CYCLES=1
 LAYOUT=0
 DRYRUN=0
+FWSEC=0
 NO_RESTORE=0
 KO="$REPO_ROOT/kernel/open-iov-vfio.ko"
 
@@ -77,6 +83,7 @@ while [ $# -gt 0 ]; do
         --cycles)     CYCLES="${2:-}"; shift ;;
         --layout)     LAYOUT=1 ;;
         --dryrun)     DRYRUN=1; LAYOUT=1 ;;
+        --fwsec)      FWSEC=1; LAYOUT=1 ;;
         --ko)         KO="${2:-}"; shift ;;
         --no-restore) NO_RESTORE=1 ;;
         -h|--help)    usage; exit 0 ;;
@@ -87,6 +94,10 @@ done
 
 case "$HOLD" in ''|*[!0-9]*) echo "--hold must be a non-negative integer" >&2; exit 1 ;; esac
 case "$CYCLES" in ''|*[!0-9]*|0) echo "--cycles must be a positive integer" >&2; exit 1 ;; esac
+if [ "$DRYRUN" -eq 1 ] && [ "$FWSEC" -eq 1 ]; then
+    echo "--dryrun and --fwsec are mutually exclusive" >&2
+    exit 1
+fi
 if [ "$HOLD" -lt 30 ]; then
     echo "WARN: --hold $HOLD is below the verified 30 s minimum" >&2
 fi
@@ -273,9 +284,11 @@ for n in $(seq 1 "$CYCLES"); do
     [ "$LAYOUT" -eq 1 ] && LAYOUT_ARG="gsp_phys_layout=1"
     DRYRUN_ARG=""
     [ "$DRYRUN" -eq 1 ] && DRYRUN_ARG="gsp_phys_fwsec_dryrun=1"
+    FWSEC_ARG=""
+    [ "$FWSEC" -eq 1 ] && FWSEC_ARG="gsp_phys_fwsec=1"
     # One command: stop persistenced and insmod with the D0 pre-lock in place.
     if ! { sudo systemctl stop nvidia-persistenced && sudo insmod "$KO" \
-            gsp_phys_gpu=1 gsp_phys_hazard_ack=1 $LAYOUT_ARG $DRYRUN_ARG; }; then
+            gsp_phys_gpu=1 gsp_phys_hazard_ack=1 $LAYOUT_ARG $DRYRUN_ARG $FWSEC_ARG; }; then
         save_dmesg "$DLOG"
         echo "ABORT: cycle $n: stop-persistenced && insmod failed" >&2
         exit 2
@@ -313,6 +326,11 @@ for n in $(seq 1 "$CYCLES"); do
             OK_DRYRUN=1
         fi
     fi
+    OK_FWSEC=1
+    if [ "$FWSEC" -eq 1 ]; then
+        OK_FWSEC=0
+        grep -q "stage 3c complete" "$DLOG" && OK_FWSEC=1
+    fi
 
     if [ "$NR_LOCKS" -eq 0 ] && [ "$OK_CLAIM" -eq 0 ]; then
         echo "ABORT: cycle $n: no locks and no claim took effect" >&2
@@ -336,6 +354,9 @@ for n in $(seq 1 "$CYCLES"); do
     if [ "$DRYRUN" -eq 1 ] && [ "$OK_DRYRUN" -ne 1 ]; then
         warn "cycle $n: gsp_phys_fwsec_dryrun=1 but the read-only dry-run did not complete (missing 'gsp-phys-fwsec-dryrun:'/'stage 3c dry-run complete')"
     fi
+    if [ "$FWSEC" -eq 1 ] && [ "$OK_FWSEC" -ne 1 ]; then
+        warn "cycle $n: gsp_phys_fwsec=1 but 'stage 3c complete' is missing (FWSEC did not report success)"
+    fi
 
     IN_RAIL_BEFORE="$(in_window_rail_count "$DLOG")"
     OUT_RAIL_BEFORE="$(( $(rail_count "$DLOG") - IN_RAIL_BEFORE ))"
@@ -376,6 +397,11 @@ for n in $(seq 1 "$CYCLES"); do
 
     OK_REBIND=0;  grep -q "rebound to nvidia (rc=1)" "$DLOG" && OK_REBIND=1
     OK_BEFORE=0;  grep -q "board check before rebind.*id=10de:1ed3" "$DLOG" && OK_BEFORE=1
+    OK_TEARDOWN=1
+    if [ "$FWSEC" -eq 1 ]; then
+        OK_TEARDOWN=0
+        grep -q "teardown complete: WPR2 down" "$DLOG" && OK_TEARDOWN=1
+    fi
     NR_RELEASED="$(grep -c "power domain released" "$DLOG" 2>/dev/null || true)"
     if [ "$OK_REBIND" -ne 1 ] || [ "$NR_RELEASED" -ne 5 ]; then
         ALL_GREEN=0
@@ -387,6 +413,10 @@ for n in $(seq 1 "$CYCLES"); do
         warn "cycle $n: missing 'board check before rebind id=10de:1ed3' line"
         ALL_GREEN=0
     fi
+    if [ "$FWSEC" -eq 1 ] && [ "$OK_TEARDOWN" -ne 1 ]; then
+        warn "cycle $n: gsp_phys_fwsec=1 but no 'teardown complete: WPR2 down' on release (WPR2 may still be up; do not retry)"
+        ALL_GREEN=0
+    fi
 
     if [ "$NO_RESTORE" -eq 0 ]; then
         set_power_control "$GPU_BDF" auto || warn "cycle $n: failed to restore GPU power/control=auto"
@@ -426,6 +456,9 @@ for n in $(seq 1 "$CYCLES"); do
     if [ "$OK_DRYRUN" -ne 1 ]; then
         ALL_GREEN=0
     fi
+    if [ "$OK_FWSEC" -ne 1 ]; then
+        ALL_GREEN=0
+    fi
 
     if [ "$ALL_GREEN" -eq 1 ] && [ "$IN_RAIL" -eq 0 ] && [ "$HEALTH_RED" -eq 0 ]; then
         log "cycle $n: GREEN (locks=$NR_LOCKS, released=$NR_RELEASED, in-window rail=0, outside rail=$OUT_RAIL, hold=${HOLD}s)"