1 files changed, 92 insertions, 0 deletions
diff --git a/tools/guest-diag/bpz.py b/tools/guest-diag/bpz.py
new file mode 100644
index 0000000..53d46a6
--- /dev/null
+++ b/tools/guest-diag/bpz.py
@@ -0,0 +1,92 @@
+# gdb script: name the silent-rollback 999 site in the cuCtxCreate path.
+#
+# Sites (file vaddr of libcuda.so.570.144; runtime = base + off):
+#   0x222e99  FUN_00322190: cmp $0x88,%eax (work-submission offset delta
+#             computed from two RM helper calls); next insn jumps to the
+#             999 store when delta > 0x88.
+#   0x223402  FUN_00322190: rc=999 (delta > 0x88)
+#   0x223452  FUN_00322190: rc=999 (neither GR nor CE engine pointer found)
+#   0x2231ab  FUN_00322190: store of FUN_0031a3b0 (engine event pump) return
+#   0x277840  FUN_00377680: rc=999 (context-register helper failed)
+#   0x27640b  FUN_00374240: rc=999 (malloc(0x40) failed)
+#   0x44fb80  FUN_0054fa00: scanner forced 999
+#   0x44fb6c  FUN_0054fa00: scanner return 0x2cf
+#
+# Run: sudo gdb -p <cdt-pid> -batch -x /tmp/bpz.py
+import gdb
+
+out = gdb.execute("info proc mappings", to_string=True)
+base = None
+for line in out.splitlines():
+    if "libcuda" in line and "r-xp" in line:
+        f = line.split()
+        base = int(f[0], 16) - int(f[3], 16)
+        break
+print("BASE=0x%x" % (base or 0))
+if base is None:
+    print("NO_LIBCUDA")
+    raise SystemExit
+
+gdb.execute("set pagination off")
+gdb.execute("set confirm off")
+
+SITES = [
+    (0x222e99, "SIZEDELTA"),
+    (0x223402, "SIZE999"),
+    (0x223452, "ENGINE999"),
+    (0x2231ab, "AFTER_PUMP"),
+    (0x277840, "CTXREF999"),
+    (0x27640b, "MALLOC999"),
+    (0x44fb80, "SCAN999"),
+    (0x44fb6c, "SCAN2cf"),
+]
+for off, name in SITES:
+    gdb.execute("break *0x%x" % (base + off))
+
+for hit in range(80):
+    try:
+        gdb.execute("continue")
+    except gdb.error as e:
+        print("CONTINUE_END %s" % e)
+        break
+    try:
+        rip = int(gdb.parse_and_eval("$rip")) & 0xffffffffffffffff
+        rax = int(gdb.parse_and_eval("$rax")) & 0xffffffffffffffff
+        r14 = int(gdb.parse_and_eval("$r14")) & 0xffffffffffffffff
+        r12 = int(gdb.parse_and_eval("$r12")) & 0xffffffffffffffff
+    except Exception as e:
+        print("REGS_GONE %s" % e)
+        break
+    off = rip - base
+    name = dict(SITES).get(off, "?")
+    print("=== HIT %d rip=0x%x off=0x%x %s rax=0x%x r14=0x%x r12=0x%x" %
+          (hit, rip, off, name, rax, r14, r12))
+    try:
+        if off in (0x222e99, 0x223402):
+            print("  delta(eax)=0x%x u32(r14)=0x%x u32(r14+0x14)=0x%x" %
+                  (rax & 0xffffffff,
+                   int(gdb.parse_and_eval("*(unsigned int *)0x%x" % r14)),
+                   int(gdb.parse_and_eval("*(unsigned int *)0x%x" % (r14 + 0x14)))))
+        if off == 0x223452:
+            print("  ctx+0x9830=0x%x ctx+0x9838=0x%x" %
+                  (int(gdb.parse_and_eval("*(unsigned long *)0x%x" % (r12 + 0x9830))),
+                   int(gdb.parse_and_eval("*(unsigned long *)0x%x" % (r12 + 0x9838)))))
+    except Exception as e:
+        print("  deref_fail %s" % e)
+    try:
+        print(gdb.execute("bt 10", to_string=True))
+    except Exception as e:
+        print("bt_fail %s" % e)
+
+for t in gdb.selected_inferior().threads():
+    t.switch()
+    print("== tid %s" % t.num)
+    try:
+        print(gdb.execute("bt 6", to_string=True).rstrip())
+    except Exception as e:
+        print("bt_fail %s" % e)
+
+try:
+    gdb.execute("detach")
+except Exception:
+    pass