vgsp+guest-diag: kcore-truth pin resolver (pool/pushbuffer/channel ring+userd) + GR ring userd-derived candidates

guest_pin_stream.py now resolves translations entirely from guest truth via
/proc/kcore + a manual page-table walk (no QMP races): the CE semaphore pool
and pushbuffer memory from the UVM rm_mem structures, and each channel's
GPFIFO ring + GPPUT from the tracking semaphore's embedding uvm_channel_t
(channel_info located by its gpFifoGpuVa value). This removes the watcher
pipeline's flakiness on the cudaMalloc stall class (missing pool/pushbuffer
pins).

gsp_sema_service.c: svc_gr_ring_autopin tries the guest-registered USERD
memdesc base as a ring candidate (the fixed FB convention only matches when
the guest's VRAM allocator happens to reproduce the old layout), plus an
optional (gsp_sema_dbg) one-shot FB scan + raw-word diagnostics when every
candidate fails.

tools/cuda-ab/p2: self-checking arbitrary-kernel acceptance suite (10
programs: sequential launches, shapes, params, smem/GEMM, streams, malloc
churn, printf+error, FMA benchmark, mixed, managed probe) with host oracle
expectations and a guest runner.

1 files changed, 161 insertions, 0 deletions

diff --git a/tools/cuda-ab/p2/run_guest.sh b/tools/cuda-ab/p2/run_guest.sh
new file mode 100755
index 0000000..0290df1
--- /dev/null
+++ b/tools/cuda-ab/p2/run_guest.sh
@@ -0,0 +1,161 @@
+#!/bin/bash
+# SPDX-License-Identifier: BSD-3-Clause
+#
+# p2/run_guest.sh - push every P2 program into the running guest, execute it
+# with a timeout, and compare its stable output lines against the recorded
+# host-oracle reference in p2/expected/<name>.txt.
+#
+# Prerequisites:
+#   * p2/expected/<name>.txt exists (run p2/oracle.sh first)
+#   * the guest is running and reachable through scripts/vgpu_vm.py
+#
+# This talks to the live VM; run it only when no campaign cycle is using it.
+#
+# Usage:
+#   ./run_guest.sh                 # all programs
+#   ./run_guest.sh p2_seq5 ...     # selected programs
+#   P2_GUEST_TIMEOUT=600 ./run_guest.sh
+#
+# Exit status: 0 only if every strict program PASSes (the managed-memory
+# probe is informational: it must complete with STATUS OK, its content is
+# recorded but not diffed).
+set -u
+
+HERE="$(cd "$(dirname "$0")" && pwd)"
+REPO="$(cd "$HERE/../.." && pwd)"
+VMEXEC="${P2_VMEXEC:-$REPO/scripts/vgpu_vm.py}"
+FILTER='^(RESULT|CHECK|ERR|STATUS|P2PRINT|PROBE|MAIL)'
+TIMEOUT="${P2_GUEST_TIMEOUT:-300}"
+INFO_PROGS=" p2_managed_probe "
+GUEST_OUT="$HERE/guest"
+
+mkdir -p "$GUEST_OUT"
+
+PROGS=()
+if [ "$#" -gt 0 ]; then
+	PROGS=("$@")
+else
+	for b in "$HERE"/bin/p2_*; do
+		PROGS+=("$(basename "$b")")
+	done
+fi
+
+if [ ! -x "$VMEXEC" ] && [ ! -f "$VMEXEC" ]; then
+	echo "ERROR: VM executor not found: $VMEXEC" >&2
+	exit 2
+fi
+
+# Build any missing binaries first (bin/ may be absent in a fresh checkout).
+need_build=0
+for name in "${PROGS[@]}"; do
+	[ -x "$HERE/bin/$name" ] || need_build=1
+done
+if [ "$need_build" -eq 1 ]; then
+	"$HERE/build.sh" || exit 2
+fi
+
+for name in "${PROGS[@]}"; do
+	if [ ! -x "$HERE/bin/$name" ]; then
+		echo "ERROR: missing binary bin/$name - run p2/build.sh" >&2
+		exit 2
+	fi
+	if [ ! -f "$HERE/expected/$name.txt" ]; then
+		echo "ERROR: missing expected/$name.txt - run p2/oracle.sh" >&2
+		exit 2
+	fi
+done
+
+ping="$("$VMEXEC" exec "echo VMPING" 2>&1 || true)"
+if [[ "$ping" != *VMPING* ]]; then
+	echo "ERROR: guest not reachable via $VMEXEC (is the VM running?)" >&2
+	exit 2
+fi
+
+push_bin() {
+	local name="$1" b64 total off size=70000 chunk out
+
+	b64="$(base64 -w0 "$HERE/bin/$name")"
+	total=${#b64}
+	off=0
+	"$VMEXEC" exec "rm -f /tmp/$name.b64" >/dev/null 2>&1
+	while [ "$off" -lt "$total" ]; do
+		chunk="${b64:$off:$size}"
+		"$VMEXEC" exec "echo -n '$chunk' >> /tmp/$name.b64" >/dev/null 2>&1
+		off=$((off + size))
+	done
+	out="$("$VMEXEC" exec "base64 -d /tmp/$name.b64 > /tmp/$name && chmod +x /tmp/$name && rm -f /tmp/$name.b64 && echo PUSH_OK" 2>&1)"
+	[[ "$out" == *PUSH_OK* ]]
+}
+
+declare -a R_NAME R_STATUS R_NOTE
+
+echo "== P2 guest acceptance: $(date -u +%Y-%m-%dT%H:%M:%SZ) =="
+echo "== timeout=${TIMEOUT}s, expected dir: $HERE/expected =="
+
+for name in "${PROGS[@]}"; do
+	gs="$GUEST_OUT/$name.txt"
+	fail=0
+	note=""
+
+	if ! push_bin "$name"; then
+		R_NAME+=("$name"); R_STATUS+=("FAIL"); R_NOTE+=("push failed")
+		continue
+	fi
+
+	"$VMEXEC" exec --timeout "$TIMEOUT" "/tmp/$name" > "$gs" 2>&1
+	rc=$?
+
+	if [[ " $INFO_PROGS " == *" $name "* ]]; then
+		if [ "$rc" -eq 0 ] && grep -q '^STATUS OK$' "$gs" && grep -q '^PROBE ' "$gs"; then
+			note="$(grep -m1 '^PROBE ' "$gs" | cut -c1-70)"
+			R_NAME+=("$name"); R_STATUS+=("INFO"); R_NOTE+=("$note")
+		else
+			R_NAME+=("$name"); R_STATUS+=("FAIL")
+			R_NOTE+=("expected-failure probe did not complete cleanly (rc=$rc)")
+		fi
+		continue
+	fi
+
+	if [ "$rc" -ne 0 ]; then
+		fail=1
+		note="guest exit=$rc"
+	fi
+	if ! grep -q '^STATUS OK$' "$gs"; then
+		fail=1
+		note="${note:+$note; }no STATUS OK"
+	fi
+
+	grep -E "$FILTER" "$gs" | LC_ALL=C sort > "$gs.filtered"
+	grep -E "$FILTER" "$HERE/expected/$name.txt" | LC_ALL=C sort > "$gs.expected"
+
+	if ! cmp -s "$gs.filtered" "$gs.expected"; then
+		fail=1
+		note="${note:+$note; }stable-line mismatch"
+		if [ ! -s "$gs.filtered" ]; then
+			note="${note:+$note; }no stable lines (timeout/hang?)"
+		fi
+	fi
+
+	if [ "$fail" -eq 0 ]; then
+		R_NAME+=("$name"); R_STATUS+=("PASS"); R_NOTE+=("")
+	else
+		R_NAME+=("$name"); R_STATUS+=("FAIL"); R_NOTE+=("$note")
+	fi
+done
+
+echo
+printf '%-22s %-6s %s\n' "PROGRAM" "RESULT" "NOTES"
+printf '%-22s %-6s %s\n' "----------------------" "------" "-----"
+any_fail=0
+for i in "${!R_NAME[@]}"; do
+	printf '%-22s %-6s %s\n' "${R_NAME[$i]}" "${R_STATUS[$i]}" "${R_NOTE[$i]}"
+	[ "${R_STATUS[$i]}" = "FAIL" ] && any_fail=1
+done
+echo
+
+if [ "$any_fail" -ne 0 ]; then
+	echo "P2: FAIL (guest output in $GUEST_OUT, diffs: *.filtered vs *.expected)"
+	exit 1
+fi
+echo "P2: PASS ($((${#R_NAME[@]})) programs; guest output in $GUEST_OUT)"
+exit 0