tools: scanerr.py - dump the non-zero RM error record behind a forced 999

The libcuda error scanner (FUN_0054fa00) walks a context list and returns
999 when its force flag is set and any context's RM error record has a
non-zero class. ctx999 only shows the first (zeroed) record; scanerr walks
the whole list and prints every non-zero record so the failing control or
alloc is named. Pushed by push_guest_tools.sh.

2 files changed, 76 insertions, 0 deletions

diff --git a/tools/guest-diag/push_guest_tools.sh b/tools/guest-diag/push_guest_tools.sh
index 42b91ae..b0711d4 100755
--- a/tools/guest-diag/push_guest_tools.sh
+++ b/tools/guest-diag/push_guest_tools.sh
@@ -27,6 +27,8 @@ push_b64 "$REPO/tools/test/cuda_drv_test.c" /tmp/cdt.c
 push_b64 "$HERE/dumpctx.py"     /tmp/dumpctx.py
 push_b64 "$HERE/trkinfo.py"     /tmp/trkinfo.py
 push_b64 "$HERE/bp999.py"       /tmp/bp999.py
+push_b64 "$HERE/ctx999.py"      /tmp/ctx999.py
+push_b64 "$HERE/scanerr.py"     /tmp/scanerr.py
 push_b64 "$HERE/sem_counters.py" /tmp/sem_counters.py
 push_b64 "$HERE/sem_slots.py"   /tmp/sem_slots.py
 push_b64 "$HERE/read_va.py"     /tmp/read_va.py
diff --git a/tools/guest-diag/scanerr.py b/tools/guest-diag/scanerr.py
new file mode 100755
index 0000000..c3af2f5
--- /dev/null
+++ b/tools/guest-diag/scanerr.py
@@ -0,0 +1,74 @@
+# gdb script: at the libcuda error-scanner entry (FUN_0054fa00), walk the
+# context list and dump every non-zero RM error record (magic words,
+# errorClass at +0x8, shorts at +0xc/+0xe). This names the source of a
+# forced 999: the scanner returns 999 when param_2 != 0 and any record's
+# class fields are non-zero.
+# Run: sudo gdb -p <cdt-pid> -batch -x /tmp/scanerr.py
+import gdb
+
+out = gdb.execute("info proc mappings", to_string=True)
+base = None
+for line in out.splitlines():
+    if "libcuda" in line and "r-xp" in line:
+        f = line.split()
+        base = int(f[0], 16) - int(f[3], 16)
+        break
+print("BASE=0x%x" % (base or 0))
+if base is None:
+    raise SystemExit
+
+gdb.execute("set pagination off")
+gdb.execute("set confirm off")
+gdb.execute("break *0x%x" % (base + 0x44fa00))
+try:
+    gdb.execute("continue")
+except gdb.error as e:
+    print("CONTINUE %s" % e)
+print("HIT")
+
+
+def rd64(a):
+    try:
+        return int(gdb.parse_and_eval("*(unsigned long *)0x%x" % a)) & 0xffffffffffffffff
+    except Exception:
+        return None
+
+
+def rd32(a):
+    try:
+        return int(gdb.parse_and_eval("*(unsigned int *)0x%x" % a)) & 0xffffffff
+    except Exception:
+        return None
+
+
+try:
+    p1 = int(gdb.parse_and_eval("$rdi")) & 0xffffffffffffffff
+    p2 = int(gdb.parse_and_eval("$rsi")) & 0xffffffff
+    print("param_1=0x%x param_2=%d" % (p1, p2))
+    pl = rd64(p1)
+    n = 0
+    while pl and n < 64:
+        mgr = rd64(pl + 0x1291 * 8)
+        if mgr:
+            reg = rd64(mgr + 0x10)
+            if reg:
+                rec = rd64(reg + 0x10)
+                if rec:
+                    vals = [rd32(rec + i) or 0 for i in range(0, 0x28, 4)]
+                    if any(vals):
+                        print("ENTRY %d ctx=0x%x mgr=0x%x rec=0x%x rec=%s" %
+                              (n, pl, mgr, rec,
+                               ' '.join('0x%08x' % v for v in vals)))
+        pl = rd64(pl + 0x128e * 8)
+        n += 1
+    print("WALKED %d" % n)
+except Exception as e:
+    print("dump_fail %s" % e)
+try:
+    gdb.execute("bt 8", to_string=True)
+except Exception:
+    pass
+try:
+    gdb.execute("detach")
+except Exception:
+    pass