guest-diag: single-instance scanner start + full-re-assert backoff

r148 ran THREE guest scanners (pin_gr_stream install_and_start retried
while a partially-failed first attempt had already started one, and the
pid-file kill missed it). Each scanner re-asserted the whole CE pair set
every 2 s; with a fully fragmented pair set (descending PAs, one page per
line) that is ~2 k debugfs writes/s per scanner, and the guest hit load 19
with clients wedged spinning in userspace.

- pin_gr_stream._start_cmd: kill EVERY existing scanner first (pattern
  assembled at runtime with printf so the command's own cmdline cannot
  self-match), then start exactly one.
- guest_pin_stream._emit_ce_mem_delta: the full re-assert keeps its 1.5 s
  cadence for the first STREAM_FULLREM_FAST_S seconds of a mem's life (the
  client-switch flush race), then backs off to STREAM_FULLREM_SLOW_S (10 s)
  as a safety net. Configurable via env.

1 files changed, 14 insertions, 4 deletions

diff --git a/tools/guest-diag/guest_pin_stream.py b/tools/guest-diag/guest_pin_stream.py
index 7d07f55..e80751a 100644
--- a/tools/guest-diag/guest_pin_stream.py
+++ b/tools/guest-diag/guest_pin_stream.py
@@ -458,7 +458,9 @@ def _emit_ce_mem_delta(kc, st, out, fresh_map=False):
     nowm = time.monotonic()
     full = nowm >= st.get("full_next", 0.0)
     if full:
-        st["full_next"] = nowm + KC_FULLREM_S
+        age = nowm - st.get("resolved_at", nowm)
+        st["full_next"] = nowm + (KC_FULLREM_S if age < KC_FULLREM_FAST_S
+                                  else KC_FULLREM_SLOW_S)
     if fresh_map:
         for k in range(npages):
             kc._pa_cache.pop(cpu + (k + off) * 0x1000, None)
@@ -680,8 +682,15 @@ KC_REMAP_S = 30.0		# full re-translation sweep (catches remaps)
 # whole translatable pair set regardless of the delta, healing host-side
 # session-flush wipes (see _emit_ce_mem_delta). Must stay well under the
 # host's gsp_sema_ce_grace_ms (8 s) so a wiped-and-needed page is restored
-# inside the hold window.
-KC_FULLREM_S = 1.5
+# inside the hold window. The fast cadence applies to a mem's first
+# KC_FULLREM_FAST_S seconds (the flush race lives at the client switch);
+# after that the re-assert backs off to KC_FULLREM_SLOW_S - a fully
+# fragmented pair set (descending PAs) emits one page per line, and 4 k
+# lines every 2 s saturated a guest under multi-client load (r148: load 19,
+# three scanners from an install race, clients wedged).
+KC_FULLREM_S = float(os.environ.get("STREAM_FULLREM_S", "1.5"))
+KC_FULLREM_FAST_S = float(os.environ.get("STREAM_FULLREM_FAST_S", "45"))
+KC_FULLREM_SLOW_S = float(os.environ.get("STREAM_FULLREM_SLOW_S", "10.0"))
 
 
 def _chain_mems_at(kc, w):
@@ -870,7 +879,8 @@ def resolve_pb_chain(kc, sem, out):
     gpus, cpu, off, npages = params
     st = {"mem": mem, "gpu": gpu, "cpu": cpu, "gpus": gpus,
           "page_va": cpu, "page_pa": kc.va_to_pa(cpu) or 0,
-          "page_off": off, "npages": npages, "last": {}}
+          "page_off": off, "npages": npages, "last": {},
+          "resolved_at": time.monotonic()}
     # Fresh translation for the first sweep: the Kcore PA cache may hold
     # pairs from before this mem's vmap area was (re)populated.
     kc._pa_cache.clear()