guest-diag: per-page pushbuffer translation, per-client re-pin

1 files changed, 40 insertions, 30 deletions

diff --git a/tools/guest-diag/semfix_watch.py b/tools/guest-diag/semfix_watch.py
index 8fd534d..c0ce072 100644
--- a/tools/guest-diag/semfix_watch.py
+++ b/tools/guest-diag/semfix_watch.py
@@ -197,16 +197,22 @@ def resolve_pushbuffer(q, sem):
 
 
 def pin_pushbuffer(q, gpu_vas, cpu, gpa):
+    """Pin the whole pushbuffer with *per-page* translations (the layout is
+    not linear across the allocation: page 32 jumped on a real client)."""
     pinned = 0
     for gv in gpu_vas:
+        miss = 0
         for i in range(PUSHBUF_MAX_PAGES):
-            if i in (0, 1, 2, 3, 8, 32, 128, 512):
-                g = q.gva2gpa(cpu + i * 0x1000)
-                if g != gpa - i * 0x1000:
-                    print("semfix: pushbuf va=0x%x: page %d not linear (gpa=%s), stop at %d"
-                          % (gv, i, hex(g) if g else None, pinned), flush=True)
+            g = q.gva2gpa(cpu + i * 0x1000)
+            if not g:
+                miss += 1
+                if miss >= 32:
+                    print("semfix: pushbuf va=0x%x: 32 unmapped pages at %d, stop (%d pinned)"
+                          % (gv, i, pinned), flush=True)
                     return pinned
-            if not pin(gv + i * 0x1000, gpa - i * 0x1000):
+                continue
+            miss = 0
+            if not pin(gv + i * 0x1000, g):
                 print("semfix: pushbuf va=0x%x: pin failed at page %d" % (gv, i), flush=True)
                 return pinned
             pinned += 1
@@ -220,22 +226,28 @@ def main():
     q = Qmp(QMP_SOCK)
     seen = set()
     seen_pb = set()
-    pb_done = False
+    pinned_ranges = set()
+
+    def pin_range_once(gpu_vas, cpu, gpa):
+        key = tuple(gpu_vas), cpu
+        if key in pinned_ranges:
+            return
+        pinned_ranges.add(key)
+        pin_pushbuffer(q, gpu_vas, cpu, gpa)
+
     print("semfix: watching (interval %.1fs)" % interval, flush=True)
     while True:
         sems, pbs = guest_trace()
-        if not pb_done:
-            for pb in pbs - seen_pb:
-                seen_pb.add(pb)
-                try:
-                    gpu_vas, cpu, gpa = resolve_pb_ptr(q, pb)
-                except Exception as e:
-                    print("semfix: pb=0x%x resolve failed: %s" % (pb, e), flush=True)
-                    continue
-                if gpu_vas:
-                    pin_pushbuffer(q, gpu_vas, cpu, gpa)
-                    pb_done = True
-                    break
+        for pb in pbs - seen_pb:
+            seen_pb.add(pb)
+            try:
+                gpu_vas, cpu, gpa = resolve_pb_ptr(q, pb)
+            except Exception as e:
+                print("semfix: pb=0x%x resolve failed: %s" % (pb, e), flush=True)
+                continue
+            if gpu_vas:
+                pin_range_once(gpu_vas, cpu, gpa)
+            else:
                 print("semfix: pb=0x%x no pushbuffer rm_mem (retry)" % pb, flush=True)
         for sem in sems - seen:
             try:
@@ -251,17 +263,15 @@ def main():
                 ok = pin(gv, gpa)
                 print("semfix: sem=0x%x va=0x%x -> gpa=0x%x pin=%s"
                       % (sem, gv, gpa, "ok" if ok else "FAILED"), flush=True)
-            if not pb_done:
-                try:
-                    gpu_vas, cpu, gpa = resolve_pushbuffer(q, sem)
-                except Exception as e:
-                    print("semfix: pushbuf resolve failed: %s" % e, flush=True)
-                    gpu_vas = []
-                if gpu_vas:
-                    pin_pushbuffer(q, gpu_vas, cpu, gpa)
-                    pb_done = True
-                else:
-                    print("semfix: pushbuf not found near sem=0x%x (retry)" % sem, flush=True)
+            try:
+                gpu_vas, cpu, gpa = resolve_pushbuffer(q, sem)
+            except Exception as e:
+                print("semfix: pushbuf resolve failed: %s" % e, flush=True)
+                gpu_vas = []
+            if gpu_vas:
+                pin_range_once(gpu_vas, cpu, gpa)
+            else:
+                print("semfix: pushbuf not found near sem=0x%x (retry)" % sem, flush=True)
         time.sleep(interval)