sema-svc: detect a decoded release by count, not by the (unchanging) VA

The pump decided "this push carried a release" by comparing the last
decoded release VA before/after svc_decode_stream. Consecutive pushes on a
channel release to the SAME tracking-semaphore VA (UVM's tracker pattern),
so every push after the first read as release-less: the entry was held for
the full CE grace window (observed: exactly one push consumed per 8 s,
gating the whole runtime init; the release itself had already been applied
by the decode, only the pump's bookkeeping stalled).

svc_rel_count now increments on every decoded release (host sem block,
REPORT_SEM block, CE launch-DMA release) and the pump compares counts at
all three sites (initial decode, rescue re-route, rescued-page retry).

1 files changed, 19 insertions, 7 deletions

diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c
index dc8e085..30c5561 100644
--- a/kernel/gsp_sema_service.c
+++ b/kernel/gsp_sema_service.c
@@ -2534,6 +2534,15 @@ static bool svc_method_stream_plausible(const u8 *buf, u32 len)
  * not match the channel rank).
  */
 static u64 svc_rel_va_seen;
+/*
+ * Monotonic count of releases decoded. svc_rel_va_seen alone cannot tell
+ * whether a decode produced a release: consecutive pushes on a channel
+ * release to the SAME tracking-semaphore VA (UVM's tracker pattern), so the
+ * VA is unchanged and the pump read "no release" for every push after the
+ * first - holding each one for the full CE grace window. The count changes
+ * on every decoded release regardless of its target.
+ */
+static u32 svc_rel_count;
 static u32 svc_dbg_streams;
 
 static void svc_counter_learn(u32 h, u64 sem_va);
@@ -2999,6 +3008,7 @@ static void svc_decode_stream(struct nv_vgpu_device *vgpu, const u8 *buf,
 				st.rel_payload = payload;
 				st.have_rel = 1;
 				svc_rel_va_seen = va;
+				svc_rel_count++;
 				/* The push's work always runs; see the
 				 * REPORT_SEM comment below. */
 				if (allow_ops && n_ops && !gsp_sema_skip_ops) {
@@ -3082,6 +3092,7 @@ static void svc_decode_stream(struct nv_vgpu_device *vgpu, const u8 *buf,
 					st.rel_payload = st.rs_payload;
 					st.have_rel = 1;
 					svc_rel_va_seen = va;
+					svc_rel_count++;
 					/*
 					 * The push's collected work always runs on
 					 * real hardware; the semaphore release only
@@ -3435,6 +3446,7 @@ static void svc_decode_stream(struct nv_vgpu_device *vgpu, const u8 *buf,
 					} else if (!svc_sem_write(vgpu, sem_va, cur)) {
 						svc.releases_applied++;
 						svc_rel_va_seen = sem_va;
+						svc_rel_count++;
 						svc.post_hint_until = jiffies + 10 * HZ;
 					}
 				}
@@ -5213,7 +5225,7 @@ static void svc_pump_channels(struct nv_vgpu_device *vgpu)
 						 * alias CE encodings.
 						 */
 						{
-							u64 rel_before = svc_rel_va_seen;
+							u32 rel_n_before = svc_rel_count;
 							bool rel_seen;
 
 							svc.op_tr_pending = 0;
@@ -5244,7 +5256,7 @@ static void svc_pump_channels(struct nv_vgpu_device *vgpu)
 								break;
 							}
 							ch->op_retry = 0;
-							rel_seen = (svc_rel_va_seen != rel_before);
+							rel_seen = (svc_rel_count != rel_n_before);
 							/*
 							 * A CE push that should carry this
 							 * channel's next tracking release but
@@ -5261,7 +5273,7 @@ static void svc_pump_channels(struct nv_vgpu_device *vgpu)
 							 */
 							if (!rel_seen && ch->engine_type &&
 							    ch->expected_sem_payload) {
-								u64 rel2 = svc_rel_va_seen;
+								u32 rel_n2 = svc_rel_count;
 
 								if (!svc_rescue_push_page(vgpu, ch, va, len) &&
 								    !svc_read_va_trusted(vgpu, va, pb, len))
@@ -5269,7 +5281,7 @@ static void svc_pump_channels(struct nv_vgpu_device *vgpu)
 											  len, 0, false,
 											  ch, idx, va, true);
 								rel_seen =
-									(svc_rel_va_seen != rel2);
+									(svc_rel_count != rel_n2);
 							}
 							if (rel_seen)
 								ch->last_rel_va =
@@ -5424,8 +5436,8 @@ if (gsp_sema_fake_rel &&
 								 * the release is applied.
 								 */
 								{
-									u64 rel_before =
-										svc_rel_va_seen;
+									u32 rel_n_before =
+										svc_rel_count;
 
 									svc_decode_stream(vgpu, pb,
 											  len, 0,
@@ -5434,7 +5446,7 @@ if (gsp_sema_fake_rel &&
 											  SVC_ENTRY_NONE,
 											  va,
 											  false);
-									if (svc_rel_va_seen != rel_before)
+									if (svc_rel_count != rel_n_before)
 										ch->last_rel_va =
 											svc_rel_va_seen;
 								}