1 files changed, 45 insertions, 10 deletions

diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c
index 9a765c8..5f69f40 100644
--- a/kernel/gsp_sema_service.c
+++ b/kernel/gsp_sema_service.c
@@ -1898,6 +1898,35 @@ struct svc_push_state {
 /* ------------------------------------------------------------------ */
 
 /*
+ * Translate a CE operand VA. Host-pinned translations (injected truth) are
+ * consulted before the content-scanned map: the CE push buffer and the UVM
+ * semaphore pool only exist in the pin tables (their VAs have no entry in
+ * the merged map), so map-only translation dropped every push-data copy
+ * (observed: `memcopy dst=0x204020 src=0x12000007c rc=-2` for UVM's inline
+ * notifier writes, and the same for the HtoD payload copy). Returns the CE
+ * aperture for the translated address.
+ */
+static int svc_op_translate(struct nv_vgpu_device *vgpu, u64 va, u64 *addr,
+			    u32 *aperture)
+{
+	u64 pa = 0;
+	int ap = SVC_APERTURE_SYS;
+
+	if (!svc_pin_lookup_any(va, &pa, &ap)) {
+		*addr = pa;
+		*aperture = (ap == SVC_APERTURE_FB) ? CE_APERTURE_VID :
+						      CE_APERTURE_SYS;
+		return 0;
+	}
+	if (!svc_map_lookup(va, &pa)) {
+		*addr = pa;
+		*aperture = CE_APERTURE_SYS;
+		return 0;
+	}
+	return -ENOENT;
+}
+
+/*
  * Write `count` elements of `value` (element size `elem`) starting at
  * `addr`, in vidmem or guest RAM depending on the aperture DST_TYPE
  * selected. Returns 0 on success.
@@ -1919,13 +1948,15 @@ static int svc_fill(struct nv_vgpu_device *vgpu, u64 addr, u64 value, u32 elem,
 	memcpy(pat, &value, elem);
 
 	if (!phys) {
-		/* Virtual destination: translate through the VA->PA map. */
+		/* Virtual destination: translate through the pinned map first,
+		 * then the VA->PA map. */
 		u64 pa;
+		u32 aper;
 
-		if (svc_map_lookup(addr, &pa))
+		if (svc_op_translate(vgpu, addr, &pa, &aper))
 			return -ENOENT;
 		addr = pa;
-		aperture = CE_APERTURE_SYS;
+		aperture = aper;
 	}
 
 	if (aperture == CE_APERTURE_VID) {
@@ -1971,19 +2002,21 @@ static int svc_copy(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t total,
 
 	if (!dst_phys) {
 		u64 pa;
+		u32 aper;
 
-		if (svc_map_lookup(dst, &pa))
+		if (svc_op_translate(vgpu, dst, &pa, &aper))
 			return -ENOENT;
 		dst = pa;
-		dst_aperture = CE_APERTURE_SYS;
+		dst_aperture = aper;
 	}
 	if (!src_phys) {
 		u64 pa;
+		u32 aper;
 
-		if (svc_map_lookup(src, &pa))
+		if (svc_op_translate(vgpu, src, &pa, &aper))
 			return -ENOENT;
 		src = pa;
-		src_aperture = CE_APERTURE_SYS;
+		src_aperture = aper;
 	}
 
 	for (o = 0; o < total; ) {
@@ -2046,11 +2079,13 @@ static void svc_run_ops(struct nv_vgpu_device *vgpu, struct svc_memset_op *ops,
 			svc.memsets_nomap++;
 		else
 			svc.memsets_fail++;
-		if (svc.memsets_done <= 8)
-			pr_info("sema-svc: %s dst=0x%llx value=0x%llx elem=%u count=%u bytes=%llu phys=%u ap=%u src=0x%llx rc=%d\n",
+		if ((rc == 0 && svc.memsets_done <= 8) ||
+		    (rc != 0 && svc.memsets_fail + svc.memsets_nomap <= 8))
+			pr_info("sema-svc: %s dst=0x%llx value=0x%llx elem=%u count=%u bytes=%llu phys=%u ap=%u src=0x%llx srcphys=%u srcap=%u rc=%d\n",
 				m->type == 1 ? "memcopy" : "memset",
 				m->dst, m->value, m->elem, m->count,
-				m->bytes, m->phys, m->aperture, m->src, rc);
+				m->bytes, m->phys, m->aperture, m->src,
+				m->src_phys, m->src_aperture, rc);
 	}
 }