vgsp: fail4 r233 - budgeted REC-DUMP + READ-DIVERGE-WIT (fill vs copy-path vs ledger)

r231/r232: the probe-class launches produced zero REC-DUMP lines while the
transformer's 8+2 appeared - the per-line pr_info_ratelimited was
suppressed wholesale during the probe's post-stall logging burst. REC-DUMP
now uses plain pr_info under a boot-wide 512-line budget, so a <=64-rec
launch always dumps its full rec table.

READ-DIVERGE-WIT (sema service, rec-build loop): for small launches the
three resolutions of each VA are compared side by side - the fill's
dg->pa (the page whose content the copy-in stages; exactly what the engine
reads), svc_op_translate (the H2D/D2H copy path's resolver) and the
ledger's write landing. A page-level split is logged with op_src/op_rc and
an op-miss token. Bounded: first 8 recs, boot budget 128.

Static permutation check (negative): reloc_va is assigned sequentially in
fill order (svc_gr_data_fill), copy-in/out iterate recs in order, the
engine reads reloc-base + linear offset, and the per-page copy loop folds
in-page offsets correctly on every resolver path; no sort/rotate/LIFO
order found in the page-table or copy paths. The remaining split suspect is
the m171/m180 ledger DIVERGE preference - exactly what READ-DIVERGE-WIT
compares.

Build green.

(cherry picked from commit 59e50207d60067b5d1e3a0546f20e9b22c92e112)

1 files changed, 36 insertions, 0 deletions

diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c
index 02c100f..d9f5895 100644
--- a/kernel/gsp_sema_service.c
+++ b/kernel/gsp_sema_service.c
@@ -12666,6 +12666,7 @@ static unsigned long svc_fwd_state;
 static u64 svc_fwd_requests;
 static u64 svc_fwd_dropped;
 static u32 svc_grvaudit_launches;	/* GRVAUDIT: per-launch audit counter */
+static u32 svc_readdiv_log;		/* r233 READ-DIVERGE-WIT budget */
 /*
  * Pause gate shared by the pump's request path and the backend release path:
  * quiesce takes this lock, sets paused and waits for the in-flight work item;
@@ -14388,6 +14389,41 @@ static void svc_fwd_gr_work(void)
 			}
 		}
 		/*
+		 * r233 READ-DIVERGE-WIT: the engine's read of this VA uses
+		 * exactly what the fill resolved (dg->pa: the page whose
+		 * content was copied into the bounce). The copy paths (H2D
+		 * writes, D2H reads) resolve independently through
+		 * svc_op_translate/ledger. For small launches (the probe
+		 * class) compare the three for the first recs and flag any
+		 * page-level split - the direct witness for a read/write
+		 * permutation. Bounded: first 8 recs, 4 lines per launch,
+		 * boot budget 128.
+		 */
+		if (svc_fwd_req.gr_n_data <= 64u && i < 8u &&
+		    svc_readdiv_log < 128u) {
+			u64 opa = 0, lpa2 = 0;
+			u32 oap = 0, lap2 = 0;
+			bool op_miss, have_led;
+
+			op_miss = svc_op_translate(svc_fwd_req.vgpu, dg->va,
+						   &opa, &oap) != 0;
+			have_led = !svc_ledger_lookup(dg->va, &lpa2, &lap2);
+			if (op_miss || (!op_miss &&
+					(opa & ~(u64)(SEMA_PAGE_SIZE - 1)) !=
+					(dg->pa & ~(u64)(SEMA_PAGE_SIZE - 1))) ||
+			    (have_led &&
+			     (lpa2 & ~(u64)(SEMA_PAGE_SIZE - 1)) !=
+			     (dg->pa & ~(u64)(SEMA_PAGE_SIZE - 1)))) {
+				svc_readdiv_log++;
+				pr_info("vgsp-gr: READ-DIVERGE-WIT va=0x%llx fill_pa=0x%llx op_pa=0x%llx op_src=%u op_rc=%d ledger_pa=0x%llx%s\n",
+					dg->va, dg->pa, opa,
+					op_miss ? 0xffffffffu : svc.op_tr_src,
+					op_miss ? -ENOENT : 0,
+					have_led ? lpa2 : 0,
+					op_miss ? " (op-miss)" : "");
+			}
+		}
+		/*
 		 * C3 (data class): after the run the backend copies this
 		 * page's results back to dg->pa (gsp_phys_booter.c copy-out).
 		 * Record that page as the truth for this VA so the guest's