r576 diag: launch-decision + event-post payload witnesses (gsp_phys_gr_dec_log, gsp_rm_ev_log)

The GR launch-decision dump logs, for the first N draws whose fill saw
dropped pages / a partial argument set or whose decision point carries a
dstpend/copy-pending verdict: the decision verdicts, the launch's resolved
data pages (VA -> guest PA), the live deferred-write records and the pending
CE opdef entries - the operand evidence needed to explain a red draw of the
readback lottery (the wrong-but-similar page hypothesis).

The event-post witness logs every EVENT_POST_EVENT record actually sent with
its full payload (hClient/hEvent/notifyIndex/data) and send result, plus a
sentinel when post_all_events found valid events but posted none while none
were dedup-skipped. This is the evidence needed when a client parks waiting
for a completion wake-up that never arrives (the 2nd-client device-state
spin): it shows whether the post was generated, for whom, and whether the
status-queue write accepted it or dropped it (burst cap / ring full).

Both are inert by default (budget 0).

1 files changed, 68 insertions, 0 deletions

diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c
index fbc0772..dfe7c3f 100644
--- a/kernel/gsp_sema_service.c
+++ b/kernel/gsp_sema_service.c
@@ -20986,6 +20986,65 @@ static bool svc_copy_pending(void)
 	return pending;
 }
 
+/*
+ * r576 DIAG: the GR launch-decision witness (gsp_phys_gr_dec_log=N budget
+ * per boot). Logged from svc_gr_request right after the fill, for launches
+ * whose fill saw dropped pages / a partial argument set or whose decision
+ * point carries a dstpend/copy-pending verdict - i.e. exactly the draws the
+ * r566/r567/r568 guards can defer or (on their blind spots) forward. The
+ * data-page list is the operand evidence: each VA -> guest PA the engine
+ * will read; a wrong-but-valid PA here (vs the page the client's H2D wrote
+ * for the same VA) is the failure mechanism the readback lottery draws.
+ */
+unsigned gsp_phys_gr_dec_log_max;
+static unsigned svc_gr_dec_log;
+
+static void svc_gr_dec_dump(const struct svc_channel *ch)
+{
+	bool dstpend, copypend;
+	u32 i, n = 0;
+
+	if (svc_gr_dec_log >= gsp_phys_gr_dec_log_max)
+		return;
+	svc_gr_dec_log++;
+
+	dstpend = svc_gr_dstpend_overlap();
+	copypend = svc_copy_pending();
+
+	pr_info("vgsp-gr: GR-DEC h=0x%x push=0x%llx n_data=%u dropped=%u partial=%d stalled=%d dstpend=%d copypend=%d defer_exp=%d\n",
+		ch->h_object, svc_gr_cand.push_va, svc_fwd_req.gr_n_data,
+		svc_fwd_req.gr_dropped,
+		svc_fwd_req.gr_arg_partial ? 1 : 0,
+		svc_gr_data_stalled ? 1 : 0,
+		dstpend ? 1 : 0, copypend ? 1 : 0,
+		svc_gr_defer_expired(svc_gr_cand.push_va) ? 1 : 0);
+
+	for (i = 0; i < svc_fwd_req.gr_n_data && i < 32; i++)
+		pr_info("vgsp-gr: GR-DEC data[%u] va=0x%llx pa=0x%llx ap=%u reloc=0x%llx\n",
+			i, svc_fwd_req.gr_data[i].va,
+			svc_fwd_req.gr_data[i].pa, svc_fwd_req.gr_data[i].ap,
+			svc_fwd_req.gr_data[i].reloc_va);
+
+	for (i = 0, n = 0; i < svc_dstpend_n && n < 16; i++) {
+		if (!svc_dstpend[i].va_page)
+			continue;
+		n++;
+		pr_info("vgsp-gr: GR-DEC dstpend va=0x%llx age_ms=%u\n",
+			svc_dstpend[i].va_page,
+			jiffies_to_msecs(jiffies - svc_dstpend[i].since));
+	}
+
+	for (i = 0, n = 0; i < SVC_OPDEF_SLOTS && n < 8; i++) {
+		if (!svc_opdef[i].active || !svc_opdef[i].n)
+			continue;
+		n++;
+		pr_info("vgsp-gr: GR-DEC opdef h=0x%x entry=%u n=%u age_ms=%u\n",
+			svc_opdef[i].h_object, svc_opdef[i].entry_idx,
+			svc_opdef[i].n,
+			jiffies_to_msecs(jiffies - svc_opdef[i].since));
+	}
+}
+
 static bool svc_gr_request(struct nv_vgpu_device *vgpu,
 			   const struct svc_channel *ch)
 {
@@ -21084,6 +21143,15 @@ static bool svc_gr_request(struct nv_vgpu_device *vgpu,
 		return false;
 	}
 	/*
+	 * r576 DIAG: witness the decision inputs of the draws the guards
+	 * care about (see svc_gr_dec_dump).
+	 */
+	if (gsp_phys_gr_dec_log_max &&
+	    (svc_fwd_req.gr_dropped > 0 || svc_fwd_req.gr_arg_partial ||
+	     svc_gr_data_stalled || svc_gr_dstpend_overlap() ||
+	     svc_copy_pending()))
+		svc_gr_dec_dump(ch);
+	/*
 	 * r541: launch-ordering guard. The fill's data pages are what the
 	 * engine will read; if any of them is the target of a device copy
 	 * that decoded but has NOT executed yet (svc_dstpend_note - the