vgsp: fail4 r235 - A-TRACE dir tag + 64KB copies exempt from bigcopy budget

r235: only 4 of the probe's ~9 64KB copies reached the ring (the
sema_bigcopy_log budget had absorbed the rest - the A H2D is missing
entirely), and the events lack direction/buffer identity. Now:

- every total==65536 copy bypasses the bigcopy budget (its own 128-line
  A-TRACE budget still bounds volume) so the probe's H2D and all D2H
  readbacks always appear;
- A-TRACE carries dir=H2D/D2H (device=VID low-pa, host=SYS high-pa) plus
  pt=P1/P2 (upload read/landed) or P4/P4R (readback source/result).

r235 decode so far: the A upload lands ([0 1 2 3] in device pa 0x7800000
at t=.94) and the SAME device page is later rewritten to
[16382 16383 8193 8194] - h_A's own values at shifted offsets - with the
pattern repeated across consecutive device pages (0x7800000/0x7810000/
0x7820000), i.e. a repeated-page/offset-wrapped write inside the emulated
copy or engine path, not client-side. The trace fix makes the next capture
name the exact path (missing H2D included).

(cherry picked from commit 9b35145136c949405c7fb4d22b3405c174ecb055)

1 files changed, 34 insertions, 9 deletions

diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c
index 487bb4d..1a8e45a 100644
--- a/kernel/gsp_sema_service.c
+++ b/kernel/gsp_sema_service.c
@@ -5199,6 +5199,8 @@ static int svc_copy(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t total,
 	bool bigcopy_this = false;
 	bool audit = total >= SVC_COPY_AUDIT_MIN;
 	static u64 bigcopy_log;
+	bool trace64 = (total == 65536u);
+	const char *dir = "?";
 
 	if (done_out)
 		*done_out = 0;	/* 0 until the first byte lands */
@@ -5341,8 +5343,33 @@ static int svc_copy(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t total,
 	 */
 	if ((total >= SVC_COPY_AUDIT_MIN ||
 	     (dst_virt && dst_orig >= SVC_VAS_FLOOR)) &&
-	    bigcopy_log < gsp_sema_bigcopy_log_max) {
+	    (bigcopy_log < gsp_sema_bigcopy_log_max || total == 65536u)) {
 		u8 sb[16] = {0}, db[16] = {0};
+		/*
+		 * r235: 64 KB copies are exempt from the bigcopy budget - the
+		 * probe's A upload/readbacks must ALWAYS appear (the budget
+		 * belongs to the multi-MB traffic and had absorbed every 64 KB
+		 * probe copy in the r235 capture).
+		 */
+		bool trace64 = (total == 65536u);
+		const char *dir =
+			(dst_aperture == CE_APERTURE_VID &&
+			 src_aperture == CE_APERTURE_SYS) ? "H2D" :
+			(src_aperture == CE_APERTURE_VID &&
+			 dst_aperture == CE_APERTURE_SYS) ? "D2H" :
+			src_shadow ? "shadow->" : "phys";
+
+		/*
+		 * r235: 64 KB copies are exempt from the bigcopy budget - the
+		 * probe's A upload/readbacks must ALWAYS appear (the budget
+		 * belongs to the multi-MB traffic and had absorbed every 64 KB
+		 * probe copy in the r235 capture).
+		 */
+		dir = (dst_aperture == CE_APERTURE_VID &&
+		       src_aperture == CE_APERTURE_SYS) ? "H2D" :
+		      (src_aperture == CE_APERTURE_VID &&
+		       dst_aperture == CE_APERTURE_SYS) ? "D2H" :
+		      src_shadow ? "shadow->" : "phys";
 
 		bigcopy_log++;
 		bigcopy_this = true;
@@ -5379,7 +5406,7 @@ static int svc_copy(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t total,
 		 * (the engine-authored page as ints). The expected content is
 		 * ints 0,1,2,3,... for the probe A buffer.
 		 */
-		if (total == 65536u && svc_atrace_log < 128u) {
+		if (trace64 && svc_atrace_log < 128u) {
 			u32 i0, i1, i2, i3, j0 = 0, j1 = 0, j2 = 0, j3 = 0;
 
 			svc_atrace_log++;
@@ -5391,9 +5418,8 @@ static int svc_copy(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t total,
 			memcpy(&j1, db + 4, 4);
 			memcpy(&j2, db + 8, 4);
 			memcpy(&j3, db + 12, 4);
-			pr_info("vgsp: A-TRACE pt=%s dst_va=0x%llx dst_pa=0x%llx dst_prov=%u | src_va=0x%llx src_pa=0x%llx src_prov=%u | src_int4=%d %d %d %d dst_int4_before=%d %d %d %d\n",
-				dst_shadow || dst_aperture == CE_APERTURE_VID ?
-				"P1" : "P4",
+			pr_info("vgsp: A-TRACE pt=%s dir=%s dst_va=0x%llx dst_pa=0x%llx dst_prov=%u | src_va=0x%llx src_pa=0x%llx src_prov=%u | src_int4=%d %d %d %d dst_int4_before=%d %d %d %d\n",
+				!strcmp(dir, "H2D") ? "P1" : "P4", dir,
 				dst_orig, dst, dst_prov, src_orig, src,
 				src_prov, i0, i1, i2, i3, j0, j1, j2, j3);
 		}
@@ -5772,7 +5798,7 @@ static int svc_copy(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t total,
 		 * the copy - for H2D this is the landed upload (P2), for D2H
 		 * the host-side readback result (P4R).
 		 */
-		if (total == 65536u && svc_atrace_log < 128u) {
+		if (trace64 && svc_atrace_log < 128u) {
 			u32 i0, i1, i2, i3;
 
 			svc_atrace_log++;
@@ -5780,9 +5806,8 @@ static int svc_copy(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t total,
 			memcpy(&i1, da + 4, 4);
 			memcpy(&i2, da + 8, 4);
 			memcpy(&i3, da + 12, 4);
-			pr_info("vgsp: A-TRACE pt=%s dst_va=0x%llx dst_pa=0x%llx dst_int4_after=%d %d %d %d\n",
-				dst_shadow || dst_aperture == CE_APERTURE_VID ?
-				"P2" : "P4R",
+			pr_info("vgsp: A-TRACE pt=%s dir=%s dst_va=0x%llx dst_pa=0x%llx dst_int4_after=%d %d %d %d\n",
+				!strcmp(dir, "H2D") ? "P2" : "P4R", dir,
 				dst_orig, dst, i0, i1, i2, i3);
 		}
 	}