vgsp: m178 GR launch reliability + diagnostics; tools: pin latching robustness

Kernel: authorship instrumentation + swrel gate for in-flight GR release VAs;
first-poll shortening + fast extended re-arm (params gr_retry_fast=1,
gr_first_poll_ms=2500); per-attempt win0 bank recompute + shadow recopy in the
outcome-retry loop; poll-break/deposit-readback diagnostics; C1-C10 sema
diagnostics (crel-entry, relx, opdef-clear, walk, pub-adv, hold-fallback,
busy-defer, CAND-OVERWRITE, inline recs); busy-defer hold instead of
software-release consume; counter fast path; QMD word 48/49 skip.
Tools: heartbeat thread, scanner-alive probe on reinstall, full periodic
re-assert, #nopid handling, per-client reset fix (debounce), dead-pid drop,
supervisor grace+threshold.

1 files changed, 47 insertions, 0 deletions

diff --git a/kernel/gsp_phys_gpu.c b/kernel/gsp_phys_gpu.c
index b1945ee..95bd627 100644
--- a/kernel/gsp_phys_gpu.c
+++ b/kernel/gsp_phys_gpu.c
@@ -974,6 +974,53 @@ int gsp_phys_rmcall_gr_nowrite_retry_value(void)
 EXPORT_SYMBOL_GPL(gsp_phys_rmcall_gr_nowrite_retry_value);
 
 /*
+ * F2 fast re-arm (gsp_phys_gr_retry_fast; default 1). r198: the first
+ * forwarded-launch timeout costs 2x5 s before the fresh-entry ladder runs,
+ * yet the m155 extended re-arm (lane-enable replay + GPFIFO/TSG schedule +
+ * doorbell) plus one fresh ring entry recovers most of these launches. With
+ * this on, the first timeout immediately runs that re-arm + a fresh entry
+ * and re-polls with a short window; only if that still fails does the
+ * historical timeout ladder continue. Set to 0 for today's exact behavior.
+ */
+static int gsp_phys_gr_retry_fast_param = 1;
+module_param_named(gsp_phys_gr_retry_fast,
+		   gsp_phys_gr_retry_fast_param, int, 0644);
+MODULE_PARM_DESC(gsp_phys_gr_retry_fast,
+		 "On the first GR launch timeout, run the extended re-arm + a fresh ring entry immediately and re-poll briefly (default 1; 0 = legacy ladder timing)");
+
+int gsp_phys_gr_retry_fast_value(void)
+{
+	return gsp_phys_gr_retry_fast_param;
+}
+EXPORT_SYMBOL_GPL(gsp_phys_gr_retry_fast_value);
+
+/*
+ * F2 first-poll window (gsp_phys_gr_first_poll_ms; default 2500, clamped to
+ * [1000, 30000]). With gsp_phys_gr_retry_fast=1 the FIRST poll window of a
+ * forwarded GR launch (before any re-arm) uses this instead of the fixed 5 s,
+ * so the fast re-arm path above starts earlier; with retry_fast=0 the launch
+ * keeps today's exact behavior and this knob is not read. The later ladder
+ * windows keep their own constants.
+ */
+static unsigned int gsp_phys_gr_first_poll_ms_param = 2500u;
+module_param_named(gsp_phys_gr_first_poll_ms,
+		   gsp_phys_gr_first_poll_ms_param, uint, 0644);
+MODULE_PARM_DESC(gsp_phys_gr_first_poll_ms,
+		 "First GR launch poll window in ms, clamped [1000,30000], read when gsp_phys_gr_retry_fast=1 (default 2500)");
+
+unsigned int gsp_phys_gr_first_poll_ms_value(void)
+{
+	unsigned int v = gsp_phys_gr_first_poll_ms_param;
+
+	if (v < 1000u)
+		v = 1000u;
+	if (v > 30000u)
+		v = 30000u;
+	return v;
+}
+EXPORT_SYMBOL_GPL(gsp_phys_gr_first_poll_ms_value);
+
+/*
  * m69 window-0 shadow pool (gsp_phys_rmcall_gr_shadow_win0; default 0). When
  * set, the forwarded GR launch copies the QMD's program/param pages into the
  * window-0 shadow pool (module pages at VAs 0x108000..0x10f000, PTEs planted