vgsp: sema_gl_mme_nop_drop (slots in mask have CALL runs dropped entirely - r1144-r1177 proven-safe retire) + force-stub semantics (nop_drop slot stubbed in drop filter + SA end-noop stub kept - r1186 lesson: keep|57 removes SA[57] stub, mis-detected kept run executes faulting macro) - validation: (a) keep=5 retires PROVEN (7/14/15 retires, poll break ch=0xc0ffee02 after 94us); (b) keep=5|1<<57 PARKS confirmed; (c) nop-retarget VALIDATED NEGATIVE (DATA-port run without open macro context parks); (d) all-real unresolved (bad boots baseline ILLEGAL_OPCODE too); boot lottery 7 healthy of ~28; stores 0/900-901 all session; next: apply nop_drop=1<<57+sa_stub_all=N within ~10s of first retire on a healthy boot, bisect slots via nop_drop (34=146650112133693440) - SESSION 153 r1180-r1208

1 files changed, 46 insertions, 1 deletions

diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c
index da3ea83..5ca854d 100644
--- a/kernel/gsp_phys_booter.c
+++ b/kernel/gsp_phys_booter.c
@@ -17625,6 +17625,26 @@ MODULE_PARM_DESC(sema_gl_mme_nop_param_mask,
 	"Kept slots whose CALL_MME_MACRO trigger is rewritten to a plain CALL_MME_DATA port write (params conserved, macro not executed; u64 bitmask)");
 
 /*
+ * SESSION 153-CONTINUED-80: the r1181 live A/B validated the retarget above
+ * and it PARKS the PBDMA exactly like the raw kept call - with
+ * keep=5|1<<57 + nop=1<<57 the first post-change launch froze
+ * (MME-FAST begin grfifo=0x00000001, end events=0, no poison, warm sem
+ * WEDGED) while keep=5 (slot 57 call+params elided) retired in 94-106 us
+ * on the same boot (r1176/r1181 baseline). The bare DATA-port run without an
+ * open macro context is not a safe landing for the trigger. With this knob
+ * set, a nop-mask slot joins the drop path instead (the r1144-r1177
+ * proven-safe behavior: whole run elided, retires clean, params NOT
+ * conserved). Use for the all-real-macro A/B: sa_stub_all=0 leaves
+ * 34/39/40/48/51 real while this drops only 57. Default 0 preserves the
+ * retarget behavior so the r1179 transform evidence stays reproducible.
+ */
+static u64 gsp_phys_rmcall_gr_nop_drop;
+module_param_named(sema_gl_mme_nop_drop, gsp_phys_rmcall_gr_nop_drop, ullong,
+		   0644);
+MODULE_PARM_DESC(sema_gl_mme_nop_drop,
+	"Slots whose call runs are dropped entirely (proven-safe retire, SA[slot] stubbed) instead of retargeted to the CALL_MME_DATA port (u64 bitmask; default 0)");
+
+/*
  * The same stubbed-slot predicate the launch builder applies (see the
  * sa_stub emit decision): a slot is stubbed when stubbing is on and either
  * the all-slots knob is set, the capture holds no SA entry, or the IR
@@ -17635,6 +17655,16 @@ static bool gsp_phys_gr_slot_stubbed(u32 j, u64 sa_mask, const u32 *sa_ent)
 {
 	if (j >= 64u || !gsp_sema_gr_sa_stub_value())
 		return false;
+	/*
+	 * SESSION 153-CONTINUED-80: a nop-drop slot is force-stubbed even
+	 * when it is also kept or covered - the r1186 evidence shows a kept
+	 * slot 57 loses its SA end-noop stub (the stub builder skips kept
+	 * slots) and any call-run encoding the drop filter misses then
+	 * executes the real macro 57 (the park class). Dropping the runs and
+	 * stubbing SA[57] together is the safe combination.
+	 */
+	if (READ_ONCE(gsp_phys_rmcall_gr_nop_drop) & (1ull << j))
+		return true;
 	if (READ_ONCE(gsp_phys_rmcall_gr_keep_mask) & (1ull << j))
 		return false;
 	if (READ_ONCE(gsp_phys_rmcall_gr_sa_stub_all))
@@ -17685,6 +17715,19 @@ static u32 gsp_phys_gr_mme_drop_calls(u32 *win, u32 n, u64 sa_mask,
 				     (1ull << j)) &&
 				    (READ_ONCE(gsp_phys_rmcall_gr_nop_param_mask) &
 				     (1ull << j))) {
+					if (READ_ONCE(gsp_phys_rmcall_gr_nop_drop)) {
+						/*
+						 * SESSION 153-CONTINUED-80: the
+						 * retarget parks the PBDMA
+						 * live (see the
+						 * sema_gl_mme_nop_drop
+						 * comment) - drop the whole
+						 * run instead.
+						 */
+						iscall = true;
+						dmask |= 1ull << j;
+						break;
+					}
 					/*
 					 * SESSION 153-CONTINUED-79: kept +
 					 * nop-param slot - keep the run but
@@ -18006,7 +18049,9 @@ static u32 gsp_phys_rmcall_build_gr_launch(u32 *pb,
 					 */
 					if (sa_mask & (1ull << j)) {
 						if (!READ_ONCE(gsp_phys_rmcall_gr_sa_stub_all) &&
-						    gsp_sema_mme_ir_covered(sa_ent[j])) {
+						    gsp_sema_mme_ir_covered(sa_ent[j]) &&
+						    !(READ_ONCE(gsp_phys_rmcall_gr_nop_drop) &
+						      (1ull << j))) {
 							for (t = 0; t < sa_ncov; t++)
 								if (sa_cov[t] == j)
 									have = true;