vgsp: eng15 diagnostics - 3D window 48->256, WINPROBE 64->2048, NOWRITE-PBDMA dump

Diag-only commit (eng15 receipts: /tmp/opencode/eng15-receipts-c260.txt).
- gr3d_win_log 48 -> 256: keep the raw 3D window dump for the gl_frames-era
  replays (the 48/boot budget was consumed by earlier launches).
- winprobe_log 64 -> 2048: keep WINPROBE samples for every stream3d launch.
- NEW NOWRITE-PBDMA dump (stream3d-only, 256/boot): all 12 PBDMA instances'
  intr0/gp_put/pbfetch/off-in-block/hdr/pbcnt at the zero-write verdict.
  Produced the decisive c260 data: pbfetch == 0x101000+words*4 exactly
  (deposit fully fetched per launch), intr0=0.

No behaviour change; logging only.

1 files changed, 63 insertions, 2 deletions

diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c
index f1cd2cb..8e3f25f 100644
--- a/kernel/gsp_phys_booter.c
+++ b/kernel/gsp_phys_booter.c
@@ -30982,7 +30982,9 @@ gr_page_unmapped:
 			static u32 gr3d_win_log;
 			u32 wi;
 
-			if (gr3d_win_log < 48) {
+			/* eng15: 48/boot was consumed before the gl_frames
+			 * replays; keep the raw window on record for them. */
+			if (gr3d_win_log < 256) {
 				gr3d_win_log++;
 				pr_info("vgsp-gr: 3D window dump h=0x%x push=0x%llx words=%u\n",
 					req->h_object, req->push_va, win3d_n);
@@ -31744,7 +31746,15 @@ gr_page_unmapped:
 			static u32 winprobe_log;
 			u32 pi;
 
-			if (winprobe_log < 64) {
+			/*
+			 * eng15: the 64/boot budget was consumed by X's
+			 * launch train before the gl_frames replays, so the
+			 * one witness that localizes a stalled walk (pbfetch
+			 * vs the module block range) never fired for the
+			 * launches under investigation. Raised so every
+			 * stream3d launch keeps both samples.
+			 */
+			if (winprobe_log < 2048) {
 				winprobe_log++;
 				pr_info("vgsp-gr: WINPROBE +%ums push=0x%llx words=%u block_dma=0x%llx grfifo=%08x trapped=%08x trapped_lo=%08x\n",
 					sked_tick * GSP_PHYS_RMCALL_POLL_MS,
@@ -32931,6 +32941,57 @@ gr_page_unmapped:
 				READ_ONCE(gsp_phys_gr_nw_streak) + 1u,
 				gsp_phys_gr_wedge_quiet,
 				gsp_phys_gr_wedge_heals);
+			/*
+			 * eng15: PBDMA state at the zero-write verdict. The
+			 * WINPROBE's 150 ms/1.2 s pbfetch samples are budget
+			 * bounded; this one is stream3d-only, bounded
+			 * 256/boot, and names WHERE the PBDMA walk of the
+			 * just-accepted replay stopped when the stores never
+			 * happened: off inside [0,words) = still walking our
+			 * block; off >= words = whole stream decoded and the
+			 * stall is downstream; no fetch at all = the entry
+			 * was never consumed.
+			 */
+			if (stream3d && bar0) {
+				static u32 nwpd_log;
+				u32 pi2;
+
+				if (nwpd_log < 256) {
+					nwpd_log++;
+					for (pi2 = 0; pi2 < 12; pi2++) {
+						u32 b2 = 0x40000u +
+							  pi2 * 0x2000u;
+						u32 i0 = ioread32(bar0 + b2 +
+								  0x108u);
+						u32 gpf = ioread32(bar0 + b2 +
+								   0x050u);
+						u32 pbl = ioread32(bar0 + b2 +
+								   0x054u);
+						u32 pbh = ioread32(bar0 + b2 +
+								   0x058u);
+						u64 f64 = ((u64)(pbh & 0xffu)
+							   << 32) |
+							  (u64)(pbl & ~3u);
+						s64 wo = -1;
+
+						if (!i0 && !gpf && !pbl &&
+						    !pbh)
+							continue;
+						if (gsp_phys_gr_last_pb_dma &&
+						    f64 >= gsp_phys_gr_last_pb_dma &&
+						    f64 < gsp_phys_gr_last_pb_dma +
+							   (u64)gsp_phys_gr_last_words * 4u)
+							wo = (s64)(f64 -
+							     gsp_phys_gr_last_pb_dma) / 4;
+						pr_warn("vgsp-gr: NOWRITE-PBDMA P%u intr0=%08x gp_put=%u pbfetch=0x%llx off=%lld hdr=%08x pbcnt=%08x h=0x%x push=0x%llx words=%u\n",
+							pi2, i0, gpf, f64, wo,
+							ioread32(bar0 + b2 + 0x118u),
+							ioread32(bar0 + b2 + 0x088u),
+							req->h_object,
+							req->push_va, words);
+					}
+				}
+			}
 
 			/*
 			 * m180 (lottery discriminator, agent-order H1/H2).