guest-diag: single-read counter sampler (kills phantom 4-of-20 view); vmrestart reaps stale watchers

2 files changed, 23 insertions, 18 deletions

diff --git a/tools/guest-diag/track_waits.sh b/tools/guest-diag/track_waits.sh
index 1e3cc91..54c4905 100644
--- a/tools/guest-diag/track_waits.sh
+++ b/tools/guest-diag/track_waits.sh
@@ -28,31 +28,31 @@ pid = int(sys.argv[1]); page = int(sys.argv[2], 16)
 count = int(sys.argv[3]); stride = int(sys.argv[4], 16); off = int(sys.argv[5], 16)
 pm = os.open('/proc/%d/pagemap' % pid, os.O_RDONLY)
 mem = os.open('/proc/%d/mem' % pid, os.O_RDONLY)
+# Window dump start; slots live at [off, off+count*stride) inside the page.
+WIN = 0x200
+WIN_BASE = page + 0xe00
+SLOT0 = off - 0xe00
 while True:
     ts = time.strftime('%T')
-    vals = []
-    for i in range(count):
-        va = page + off + i * stride
-        try:
-            os.lseek(pm, (va // 4096) * 8, 0)
-            e = int.from_bytes(os.read(pm, 8), 'little')
-            present = (e >> 63) & 1
-            if present:
-                v = int.from_bytes(os.pread(mem, 4, va), 'little')
-                vals.append('%d:0x%x' % (i, v))
-            else:
-                vals.append('%d:-' % i)
-        except OSError:
-            vals.append('%d:ERR' % i)
-    print('%s %s' % (ts, ' '.join(vals)), flush=True)
-    # Raw hex dump of the counter window: shows the true slot layout and any
-    # value a stride-16 sample would misread.
     try:
         os.lseek(pm, (page // 4096) * 8, 0)
         e = int.from_bytes(os.read(pm, 8), 'little')
         if (e >> 63) & 1:
-            raw = os.pread(mem, 0x200, page + 0xe00)
+            # One read for both views: per-slot values and the raw dump come
+            # from the same buffer, so they can never disagree. (The old
+            # 20-single-word reader plus a later bulk read produced a
+            # phantom "4 of 20 slots" pattern from a torn/remapped read.)
+            raw = os.pread(mem, WIN, WIN_BASE)
+            vals = []
+            for i in range(count):
+                o = SLOT0 + i * stride
+                v = int.from_bytes(raw[o:o + 4], 'little')
+                vals.append('%d:0x%x' % (i, v))
+            print('%s %s' % (ts, ' '.join(vals)), flush=True)
             print('%s RAW %s' % (ts, raw.hex()), flush=True)
+        else:
+            print('%s %s' % (ts, ' '.join('%d:-' % i for i in range(count))),
+                  flush=True)
     except OSError:
         pass
     time.sleep(0.5)
diff --git a/tools/guest-diag/vmrestart.sh b/tools/guest-diag/vmrestart.sh
index dbe60ec..5d15852 100755
--- a/tools/guest-diag/vmrestart.sh
+++ b/tools/guest-diag/vmrestart.sh
@@ -6,6 +6,11 @@ set -u
 cd "$(dirname "$(readlink -f "$0")")/../.."
 
 sudo pkill -f qemu-system-x86_64 2>/dev/null
+# Reap host-side watchers too: a pin_gr_stream/semfix pair left over from the
+# previous boot holds the vsock port and the single QMP socket slot and will
+# silently corrupt the next run (two QMP clients).
+pkill -f "guest-diag/pin_gr_stream.py" 2>/dev/null
+pkill -f "guest-diag/semfix_uvmt.py" 2>/dev/null
 sleep 4
 sudo rm -f /tmp/qemu-guest-agent.sock /tmp/qemu-qmp.sock