vgsp: eng43 - keep the MME pack completion record and packed cursor across an owner channel re-register (sema_gl_mme_keep_done_across_recycle, default 0)

c579: the first eng42-module completion (2799/2799) was followed 1.09 s later
by the beef owner re-register; the eng40 chan_ctx_reset treated the recycle
as a context loss, rewound the pack cursor and wiped pack_done_gen/ch, so
every later prelude launch re-emitted (pre_words_out > 0 -> the r1222 clause
correctly withheld the release) and MME-PACK-REARM could never fire.

The knob keeps pack_done_h/gen/ch and the cursor (pack_off/pack_sa/load
accounting) across the re-register so the first prelude launch on the
re-incarnated channel is non-emitting and the REARM clause releases the held
calls against the engine-resident program. The per-incarnation
latch/live/release state is still cleared, and every genuine context wipe
(WEDGE-RECOVER/CWD-CLEAR, park recovery, iram-gate failure, non-keep_gen
generation rewind, FLR via reload) clears the record on its own path.
Default 0 = legacy invalidate (validation pending).

1 files changed, 52 insertions, 7 deletions

diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c
index 9f1efdd..7a7af61 100644
--- a/kernel/gsp_phys_booter.c
+++ b/kernel/gsp_phys_booter.c
@@ -19452,6 +19452,30 @@ MODULE_PARM_DESC(sema_gl_mme_pack_keep_gen,
 	"Keep the packed-part cursor across prelude-generation changes when the golden overlay is forced (deterministic image; lets a multi-part beef program complete across interleaved state captures; default 0)");
 
 /*
+ * eng43: keep the completion record + packed cursor across an owner channel
+ * RE-REGISTRATION. The client recycles its beef channel family per fire: the
+ * engine context (and the MME IRAM program loaded into it) is not wiped by a
+ * mere channel-handle recycle, but the eng40 chan_ctx_reset treated the
+ * re-register as a context loss. c579: the first eng42-module completion
+ * (2799/2799) was followed 1.09 s later by `beef chan register #12`, the
+ * reset rewound the pack cursor, every later prelude launch re-emitted
+ * (pre_words_out > 0 -> the r1222 clause correctly withheld the release) and
+ * the post-completion REARM could never fire. With this knob on, the
+ * re-register keeps pack_done_h/gen/ch and the cursor (pack_off/pack_sa/
+ * load accounting) so the first prelude launch on the re-incarnated channel
+ * is non-emitting and MME-PACK-REARM releases the held calls against the
+ * engine-resident program. Genuine context wipes (WEDGE-RECOVER/CWD-CLEAR,
+ * park recovery, iram-gate failure, the non-keep_gen generation rewind, FLR
+ * via module reload) still clear the record on their own paths. Default 0 =
+ * legacy (the re-register invalidates). A/B knob.
+ */
+static bool gsp_phys_rmcall_gr_mme_keep_done_across_recycle;
+module_param_named(sema_gl_mme_keep_done_across_recycle,
+		   gsp_phys_rmcall_gr_mme_keep_done_across_recycle, bool, 0644);
+MODULE_PARM_DESC(sema_gl_mme_keep_done_across_recycle,
+	"Keep the MME pack completion record (pack_done_h/gen/ch) and the packed cursor across an owner channel re-register (the client recycles beef channels per fire without wiping the GR context); genuine wipes still clear. Default 0 = legacy invalidate");
+
+/*
  * SESSION 153-CONTINUED-107: separate knob for the per-launch chronology
  * probe lines. They write the MME debug port DURING a live launch, and the
  * first three probe-bearing cycles (r1369..r1371) all parked the engine on
@@ -20874,6 +20898,12 @@ static void gsp_phys_gr_ctx_wipe_latches(void)
  * the post-completion call release cannot fire against a non-resident
  * program. Gated by the same A/B knobs as the release itself; with the
  * knobs off this is a no-op (legacy).
+ *
+ * eng43: the "fresh context" premise is what c579 falsified - the client
+ * recycles its channel family per fire while the GR context (and the MME
+ * IRAM program in it) survives. sema_gl_mme_keep_done_across_recycle=1
+ * keeps the completion record + packed cursor across the re-register and
+ * still clears the per-incarnation latch/release state (see below).
  */
 void gsp_phys_gr_mme_chan_ctx_reset(u32 h)
 {
@@ -20889,13 +20919,28 @@ void gsp_phys_gr_mme_chan_ctx_reset(u32 h)
 	if (h != READ_ONCE(gsp_phys_gr_mme_pack_load_h) &&
 	    h != READ_ONCE(gsp_phys_gr_mme_pack_done_h))
 		return;
-	WRITE_ONCE(gsp_phys_gr_mme_pack_off, 0);
-	WRITE_ONCE(gsp_phys_gr_mme_pack_sa, false);
-	WRITE_ONCE(gsp_phys_gr_mme_pack_load_h, 0);
-	WRITE_ONCE(gsp_phys_gr_mme_pack_load_words, 0);
-	WRITE_ONCE(gsp_phys_gr_mme_pack_done_h, 0);
-	WRITE_ONCE(gsp_phys_gr_mme_pack_done_gen, 0);
-	WRITE_ONCE(gsp_phys_gr_mme_pack_done_ch, 0);
+	/*
+	 * eng43: sema_gl_mme_keep_done_across_recycle=1 - the re-register
+	 * recycles the client handle but leaves the GR engine context (and
+	 * the MME IRAM program loaded into it) untouched; c579 showed the
+	 * legacy wipe turned every later launch into a re-emission, so the
+	 * post-completion REARM could never fire. Keep the completion
+	 * record and the packed cursor; a genuine context wipe clears them
+	 * on its own path (WEDGE-RECOVER/CWD-CLEAR, park recovery, the
+	 * iram-gate failure, the non-keep_gen generation rewind). The
+	 * per-incarnation latch/release state below is always cleared - the
+	 * new incarnation must re-earn residency through the REARM clause.
+	 * Default 0 = legacy (identical to the pre-knob behaviour).
+	 */
+	if (!READ_ONCE(gsp_phys_rmcall_gr_mme_keep_done_across_recycle)) {
+		WRITE_ONCE(gsp_phys_gr_mme_pack_off, 0);
+		WRITE_ONCE(gsp_phys_gr_mme_pack_sa, false);
+		WRITE_ONCE(gsp_phys_gr_mme_pack_load_h, 0);
+		WRITE_ONCE(gsp_phys_gr_mme_pack_load_words, 0);
+		WRITE_ONCE(gsp_phys_gr_mme_pack_done_h, 0);
+		WRITE_ONCE(gsp_phys_gr_mme_pack_done_gen, 0);
+		WRITE_ONCE(gsp_phys_gr_mme_pack_done_ch, 0);
+	}
 	WRITE_ONCE(gsp_phys_gr_mme_prelude_gen, 0);
 	WRITE_ONCE(gsp_phys_gr_mme_pre_emit_off, 0);
 	WRITE_ONCE(gsp_phys_gr_mme_pending_gen, 0);