gr: r405b late-write probes - retire LATE-PROBE (+500 ms re-scan) and a tunable settle window (gsp_phys_gr_settle_ms)

The r405 streams2 forensics show a new shape of the no-write class: the
in-place scale launch wrote 1 of its 32 expected pages and the
partial-changed count was content-stable for 300 ms, so the retire wait
declared 'settled' and the copy-out shipped one page. Two runtime knobs
discriminate and test the fix in one cycle: gsp_phys_gr_late_probe=1
re-scans +500 ms after the retire wait and logs whether the engine was
still writing; gsp_phys_gr_settle_ms raises the content-stable window
(default 300 ms).

1 files changed, 56 insertions, 1 deletions

diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c
index 01e0d12..b5898b6 100644
--- a/kernel/gsp_phys_booter.c
+++ b/kernel/gsp_phys_booter.c
@@ -141,6 +141,44 @@ static inline bool gsp_phys_gr_copywit_enabled(void)
 }
 
 /*
+ * r405 late-write probes (the r405 "partial write" class: the scale launch
+ * of streams2 wrote 1 of its 32 expected in-place pages and every other
+ * launch settled with far fewer changed pages than the kernel should
+ * produce). Two knobs, both runtime-settable:
+ *
+ *  - gsp_phys_gr_late_probe (default 0): after the retire wait concludes,
+ *    sleep +500 ms and re-scan the tracked pages. Run-once diagnostic:
+ *    "ENGINE STILL WRITING" proves the copy-out shipped before the stores
+ *    landed (the fix is a longer settle window); "stable" proves the engine
+ *    genuinely stopped, which needs a different treatment.
+ *  - gsp_phys_gr_settle_ms (default 300): the content-stable window the
+ *    partial-changed rule requires before it declares the stores settled.
+ *    Raise it to give a still-writing engine its time (A/B in one cycle).
+ */
+static bool gsp_phys_gr_late_probe_param;
+module_param_named(gsp_phys_gr_late_probe, gsp_phys_gr_late_probe_param,
+		   bool, 0644);
+MODULE_PARM_DESC(gsp_phys_gr_late_probe,
+		 "Diagnostic: after the GR retire wait, re-scan the tracked pages +500 ms later and log whether the engine is still writing (default 0)");
+
+static int gsp_phys_gr_settle_ms_param = 300;
+module_param_named(gsp_phys_gr_settle_ms, gsp_phys_gr_settle_ms_param,
+		   int, 0644);
+MODULE_PARM_DESC(gsp_phys_gr_settle_ms,
+		 "GR retire wait: content-stable window (ms) required before a partial changed-page count counts as settled (default 300)");
+
+static inline u32 gsp_phys_gr_settle_iters(void)
+{
+	int ms = READ_ONCE(gsp_phys_gr_settle_ms_param);
+
+	if (ms < 60)
+		ms = 60;
+	if (ms > 1400)
+		ms = 1400;
+	return (u32)(ms / 5);
+}
+
+/*
  * COPYWIT fail4 r224 follow-up: the r224 OUT-PAGE-WIT dumps proved the
  * engine's outputs are structurally corrupted (row bands >= 16 replicate a
  * single tile per buffer) while the copy-out write/read pairs agree
@@ -17725,7 +17763,8 @@ gsp_phys_gr_output_retire_wait(const struct gsp_phys_gr_submit *req,
 		 * kernel costs at most 300 ms extra; a ramping engine gets
 		 * the time its stores need.
 		 */
-		if (changed > 0 && hash == prev_hash && iter >= 60u) {
+		if (changed > 0 && hash == prev_hash &&
+		    iter >= gsp_phys_gr_settle_iters()) {
 			why = " (settled)";
 			break;
 		}
@@ -17738,6 +17777,22 @@ gsp_phys_gr_output_retire_wait(const struct gsp_phys_gr_submit *req,
 	}
 	pr_info("vgsp-gr: output retirement wait: %u/%u pages changed after %u ms%s\n",
 		changed, n_all, jiffies_to_msecs(jiffies - wait_start), why);
+	/*
+	 * r405: run-once late-write probe (see the param comment above).
+	 */
+	if (READ_ONCE(gsp_phys_gr_late_probe_param) && changed < n_all) {
+		u32 late_n = 0, late;
+		unsigned long lp_start = jiffies;
+
+		msleep(500);
+		late = gsp_phys_gr_scan_pages(req, csum_in, &late_n, NULL);
+		pr_info("vgsp-gr: retire LATE-PROBE %u -> %u of %u after +%u ms%s\n",
+			changed, late, late_n,
+			jiffies_to_msecs(jiffies - lp_start),
+			late != changed ?
+				" (ENGINE STILL WRITING - stores landed late)" :
+				" (stable - writes are final)");
+	}
 	return changed;
 }