1 files changed, 80 insertions, 0 deletions
diff --git a/tools/guest-diag/bpz2.py b/tools/guest-diag/bpz2.py
new file mode 100644
index 0000000..2d399ff
--- /dev/null
+++ b/tools/guest-diag/bpz2.py
@@ -0,0 +1,80 @@
+# gdb script: capture the post-pump failure call in FUN_00374240 (ctx init).
+#
+# From the previous run: the engine event pump FUN_0031a3b0 returned 0
+# (0x2231ab) and the four known 999 stores did NOT fire, so the rc=999 comes
+# from a call AFTER the pump on the setup path. Every one of those calls in
+# FUN_00374240 stores its status into r14d and jumps to 0x274680 on failure,
+# and the function returns r14d at 0x27428a. Break at each store site and at
+# the final return; print the status and a backtrace when it is non-zero.
+#
+# Run: sudo gdb -p <cdt-pid> -batch -x /tmp/bpz2.py
+import gdb
+import time
+
+out = gdb.execute("info proc mappings", to_string=True)
+base = None
+for line in out.splitlines():
+    if "libcuda" in line and "r-xp" in line:
+        f = line.split()
+        base = int(f[0], 16) - int(f[3], 16)
+        break
+print("BASE=0x%x" % (base or 0))
+if base is None:
+    print("NO_LIBCUDA")
+    raise SystemExit
+
+gdb.execute("set pagination off")
+gdb.execute("set confirm off")
+
+SITES = [
+    (0x2231ab, "AFTER_PUMP"),
+    (0x2233f7, "TAIL_SCAN"),
+    (0x27537d, "RET_CTXENG"),
+    (0x27539d, "RET_A4090"),
+    (0x2753da, "RET_35fb00"),
+    (0x275444, "RET_a5c50"),
+    (0x275474, "RET_ENG60"),
+    (0x27549d, "RET_ENG68"),
+    (0x2754c6, "RET_ENG70"),
+    (0x2754f6, "RET_ENG78"),
+    (0x275541, "RET_ENGLOOP"),
+    (0x27558c, "RET_ENG2f8"),
+    (0x2755bf, "RET_ENG300"),
+    (0x27428a, "CTX_RET"),
+    (0x277840, "CTXREF999"),
+    (0x44fb80, "SCAN999"),
+    (0x44fb6c, "SCAN2cf"),
+]
+for off, name in SITES:
+    gdb.execute("break *0x%x" % (base + off))
+
+t0 = time.time()
+for hit in range(200):
+    try:
+        gdb.execute("continue")
+    except gdb.error as e:
+        print("CONTINUE_END %s" % e)
+        break
+    try:
+        rip = int(gdb.parse_and_eval("$rip")) & 0xffffffffffffffff
+        rax = int(gdb.parse_and_eval("$rax")) & 0xffffffffffffffff
+        r14 = int(gdb.parse_and_eval("$r14")) & 0xffffffffffffffff
+    except Exception as e:
+        print("REGS_GONE %s" % e)
+        break
+    off = rip - base
+    name = dict(SITES).get(off, "?")
+    st = rax & 0xffffffff
+    print("HIT %d t=%.3f off=0x%x %s rax=0x%x r14d=0x%x" %
+          (hit, time.time() - t0, off, name, rax, r14 & 0xffffffff))
+    if st != 0 or off in (0x27428a, 0x277840, 0x44fb80, 0x44fb6c):
+        print("  >>> NONZERO/TERMINAL %s status=0x%x" % (name, st))
+        try:
+            print(gdb.execute("bt 12", to_string=True))
+        except Exception as e:
+            print("bt_fail %s" % e)
+
+try:
+    gdb.execute("detach")
+except Exception:
+    pass