sema-svc: operand resolver gains the targeted rm_mem fallback; per-side translation-miss tracing (the DtoH copy operand is the failing side)

1 files changed, 35 insertions, 2 deletions

diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c
index 8b5f4af..b025dde 100644
--- a/kernel/gsp_sema_service.c
+++ b/kernel/gsp_sema_service.c
@@ -354,6 +354,7 @@ struct sema_svc {
 	u64			ring_ops;		/* op runs from ring-decoded pushes */
 	u64			lda_log;		/* LAUNCH_DMA inventory budget */
 	u64			sem256_log;		/* 256-byte op trace budget */
+	u64			op_tr_log;		/* operand targeted-resolve budget */
 	u64			copy_watch_log;		/* sema_copy_watch log budget */
 	u64			copy_pat_log;		/* sema_copy_pat log budget */
 	u64			copy_dst_log;		/* sema_copy_dst log budget */
@@ -1996,6 +1997,26 @@ static int svc_op_translate(struct nv_vgpu_device *vgpu, u64 va, u64 *addr,
 						      CE_APERTURE_SYS;
 		return 0;
 	}
+	/*
+	 * Targeted resolve: the RAM scan for the rm_mem record that owns this
+	 * VA. svc_lookup_trusted has always had this fallback (the GRCE HtoD
+	 * resolves the client's device pointer through it), but this operand
+	 * path did not: a boot whose content-scanned map or published page
+	 * directories happen not to cover the operand failed the copy with
+	 * -ENOENT (observed: the DtoH copy of the cdt2 payload, rc=-2, while
+	 * the HtoD for the same source VA resolved). That is the boot-to-boot
+	 * 64/64 vs 0/64 variance.
+	 */
+	if (!svc_targeted_resolve(vgpu, va & ~0xfffULL, &pa)) {
+		*addr = pa + (va & 0xfff);
+		*aperture = CE_APERTURE_SYS;
+		if (svc.op_tr_log < 64) {
+			svc.op_tr_log++;
+			pr_info("sema-op-tr: va=0x%llx -> pa=0x%llx\n",
+				va, *addr);
+		}
+		return 0;
+	}
 	return -ENOENT;
 }
 
@@ -2077,8 +2098,14 @@ static int svc_copy(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t total,
 		u64 pa;
 		u32 aper;
 
-		if (svc_op_translate(vgpu, dst, &pa, &aper))
+		if (svc_op_translate(vgpu, dst, &pa, &aper)) {
+			if (svc.op_tr_log < 64) {
+				svc.op_tr_log++;
+				pr_info("sema-tr-miss: DST va=0x%llx h=0x%x\n",
+					dst, svc.cur_ch);
+			}
 			return -ENOENT;
+		}
 		dst = pa;
 		dst_aperture = aper;
 	}
@@ -2086,8 +2113,14 @@ static int svc_copy(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t total,
 		u64 pa;
 		u32 aper;
 
-		if (svc_op_translate(vgpu, src, &pa, &aper))
+		if (svc_op_translate(vgpu, src, &pa, &aper)) {
+			if (svc.op_tr_log < 64) {
+				svc.op_tr_log++;
+				pr_info("sema-tr-miss: SRC va=0x%llx h=0x%x\n",
+					src, svc.cur_ch);
+			}
 			return -ENOENT;
+		}
 		src = pa;
 		src_aperture = aper;
 	}