vgsp-gl: author the PS/TCS stage images for the emitted mask slots

3 files changed, 26 insertions, 2 deletions

diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c
index 43e4e11..733937c 100644
--- a/kernel/gsp_phys_booter.c
+++ b/kernel/gsp_phys_booter.c
@@ -19810,6 +19810,18 @@ bool gsp_phys_rmcall_gr_sph_bind_value(void)
 }
 
 /*
+ * SESSION 155 CONT 25: the synth slot mask must be readable by the sema
+ * service so it authors exactly the stage images the booter binds. With
+ * sema_gl_sph_bind=0 + sema_sph_bind_mask=0x35 the booter emits j0/j2/j5,
+ * but the service's PS/TCS authoring loop was gated on the all-slots bool,
+ * so the PS image could be left stale/zero at draw time (r1575).
+ */
+u32 gsp_phys_rmcall_gr_sph_bind_mask_value(void)
+{
+	return READ_ONCE(gsp_phys_rmcall_gr_sph_bind_mask);
+}
+
+/*
  * SESSION 154c (viewport wall): the replayed 3D window carries no
  * viewport family state - the colour target, raster enable, shader
  * binds, attributes, index buffer and draws are present, but the
diff --git a/kernel/gsp_phys_gpu.h b/kernel/gsp_phys_gpu.h
index 7de4fd5..d84b080 100644
--- a/kernel/gsp_phys_gpu.h
+++ b/kernel/gsp_phys_gpu.h
@@ -347,6 +347,7 @@ int gsp_phys_rmcall_gr_shadow_win0_value(void);
  * author the matching program images at those family VAs.
  */
 bool gsp_phys_rmcall_gr_sph_bind_value(void);
+u32 gsp_phys_rmcall_gr_sph_bind_mask_value(void);
 
 /*
  * SESSION 155 CONT 11: bind a synthesized header-only GEOMETRY SPH to
diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c
index 6972172..e42e23d 100644
--- a/kernel/gsp_sema_service.c
+++ b/kernel/gsp_sema_service.c
@@ -10587,14 +10587,25 @@ static void svc_gl_flush(struct nv_vgpu_device *vgpu,
 				  ~(u64)(SEMA_PAGE_SIZE - 1);
 			u32 k;
 
-			for (k = 0; gsp_phys_rmcall_gr_sph_bind_value() &&
-			     k < 2u && n_recs < SVC_GL_PROG_MAX; k++) {
+			/*
+			 * CONT 25: author exactly the slots the booter emits.
+			 * The booter emits j5 (PS) on mask 0x20 and j2 (TCS)
+			 * on 0x04 when the all-slots bool is off; gate each
+			 * image on the same condition instead of the bool
+			 * alone (the r1575 zero-PS re-capture).
+			 */
+			for (k = 0; k < 2u && n_recs < SVC_GL_PROG_MAX; k++) {
+				u32 slot_bit = k ? 0x04u : 0x20u;
 				u64 dst = fam + (k ? 0x200u : 0x0u);
 				const u8 *b = k ? svc_gl_vs : svc_gl_ps;
 				u32 bl = k ? svc_gl_vs_len : svc_gl_ps_len;
 				bool dup = false;
 				u32 pi3;
 
+				if (!(gsp_phys_rmcall_gr_sph_bind_value() ||
+				      (gsp_phys_rmcall_gr_sph_bind_mask_value() &
+				       slot_bit)))
+					continue;
 				if (!b || bl < 4u)
 					continue;
 				for (pi3 = 0; pi3 < svc_gl.n_progs; pi3++)