sema-svc: learn the tracking-sem VA from readable pushes for the release fallback

1 files changed, 42 insertions, 14 deletions

diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c
index 343981c..2127f38 100644
--- a/kernel/gsp_sema_service.c
+++ b/kernel/gsp_sema_service.c
@@ -139,6 +139,7 @@ struct svc_channel {
 	u32	last_get;	/* next GPFIFO entry we have still to run */
 	u64	submissions;	/* GPFIFO entries executed */
 	u32	expected_sem_payload;	/* next semaphore payload we expect for this channel's tracking sem */
+	u64	last_rel_va;	/* last tracking-sem VA seen in a readable push */
 	u32	logged_expected;	/* last pool value logged (rate limit) */
 	unsigned long rescue_last;	/* rate limit for the full-RAM rescue scan */
 	u32	rescue_fail;	/* consecutive failed rescues for the entry */
@@ -1728,8 +1729,7 @@ static void svc_rebuild(struct nv_vgpu_device *vgpu)
 /* push application                                                   */
 /* ------------------------------------------------------------------ */
 
-/* C5B5 (TURING_DMA_COPY_A) methods, from the class interface. */
-#define CE_M_SET_SEMAPHORE_A	0x240
+/* C5B5 (TURING_DMA_COPY_A) methods, from the class interface. */#define CE_M_SET_SEMAPHORE_A	0x240
 #define CE_M_SET_SEMAPHORE_B	0x244
 #define CE_M_SET_SEMAPHORE_PAY	0x248
 #define CE_M_SET_SRC_PHYS_MODE	0x260
@@ -1989,6 +1989,14 @@ static bool svc_method_stream_plausible(const u8 *buf, u32 len)
  * translation is executed at the push's release method (see below); pages
  * decoded from raw sweeps never execute work at all.
  */
+/*
+ * Last tracking-semaphore VA decoded from a readable push. The pump copies
+ * it into the channel so the unreadable-push fallback targets the semaphore
+ * the client actually waits on (its pool index is allocated by UVM and does
+ * not match the channel rank).
+ */
+static u64 svc_rel_va_seen;
+
 static void svc_decode_stream(struct nv_vgpu_device *vgpu, const u8 *buf,
 			      u32 len, u64 page_pa, bool allow_ops,
 			      u32 ch_h, u32 ch_eng, u64 push_va)
@@ -2119,6 +2127,7 @@ static void svc_decode_stream(struct nv_vgpu_device *vgpu, const u8 *buf,
 				st.rel_va = va;
 				st.rel_payload = payload;
 				st.have_rel = 1;
+				svc_rel_va_seen = va;
 				/* The push's work always runs; see the
 				 * REPORT_SEM comment below. */
 				if (allow_ops && n_ops) {
@@ -2188,6 +2197,7 @@ static void svc_decode_stream(struct nv_vgpu_device *vgpu, const u8 *buf,
 					st.rel_va = va;
 					st.rel_payload = st.rs_payload;
 					st.have_rel = 1;
+					svc_rel_va_seen = va;
 					/*
 					 * The push's collected work always runs on
 					 * real hardware; the semaphore release only
@@ -3800,12 +3810,19 @@ static void svc_pump_channels(struct nv_vgpu_device *vgpu)
 						 * need but their other methods may
 						 * alias CE encodings.
 						 */
-						svc_decode_stream(vgpu, pb,
-								  len, 0,
-								  ch->engine_type != 0,
-								  ch->h_object,
-								  ch->engine_type,
-								  va);
+						{
+							u64 rel_before = svc_rel_va_seen;
+
+							svc_decode_stream(vgpu, pb,
+									  len, 0,
+									  ch->engine_type != 0,
+									  ch->h_object,
+									  ch->engine_type,
+									  va);
+							if (svc_rel_va_seen != rel_before)
+								ch->last_rel_va =
+									svc_rel_va_seen;
+						}
 						ch->submissions++;
 						svc.gpfifo_entries_done++;
 						/* Advance expected payload for this channel's tracking semaphore. */
@@ -3841,12 +3858,20 @@ static void svc_pump_channels(struct nv_vgpu_device *vgpu)
 								 * fill vidmem with garbage). Only
 								 * the release is applied.
 								 */
-								svc_decode_stream(vgpu, pb,
-										  len, 0,
-										  false,
-										  ch->h_object,
-										  ch->engine_type,
-										  va);
+								{
+									u64 rel_before =
+										svc_rel_va_seen;
+
+									svc_decode_stream(vgpu, pb,
+											  len, 0,
+											  false,
+											  ch->h_object,
+											  ch->engine_type,
+											  va);
+									if (svc_rel_va_seen != rel_before)
+										ch->last_rel_va =
+											svc_rel_va_seen;
+								}
 								ch->submissions++;
 								svc.gpfifo_entries_done++;
 								if (ch->engine_type && ch->expected_sem_payload)
@@ -3901,6 +3926,8 @@ static void svc_pump_channels(struct nv_vgpu_device *vgpu)
 							    ch->engine_type &&
 							    ch->expected_sem_payload) {
 								u64 pool_va =
+									ch->last_rel_va ?
+									ch->last_rel_va :
 									0x121000000ULL +
 									4ULL * ch->ce_index;
 								u64 spa = 0;
@@ -3984,6 +4011,7 @@ static void svc_pump_channels(struct nv_vgpu_device *vgpu)
 				 */
 				svc.chan[k].expected_sem_payload =
 					ch->expected_sem_payload;
+				svc.chan[k].last_rel_va = ch->last_rel_va;
 				svc.chan[k].rescue_last = ch->rescue_last;
 				svc.chan[k].rescue_fail = ch->rescue_fail;
 				break;