vgsp-gl: raster-enable prelude (CT_WRITE/depth/stencil/scissor) + DS reset fix

1 files changed, 65 insertions, 3 deletions

diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c
index 213aa22..0fc2ee7 100644
--- a/kernel/gsp_phys_booter.c
+++ b/kernel/gsp_phys_booter.c
@@ -5634,7 +5634,14 @@ static bool gsp_phys_mme_fast_sample(void __iomem *bar0, const char *tag,
 		u32 esr = ioread32(bar0 + 0x00405840u);
 
 		if (esr) {
-			iowrite32(esr, bar0 + 0x00405840u);
+			/*
+			 * SESSION 155 CONT 14: RESET (bit30, RESET_TASK)
+			 * must be set with the W1C clears so the HWW
+			 * latch is actually reset (reference-RM idiom);
+			 * without it the per-draw attribution is unsound.
+			 */
+			iowrite32(esr | 0x40000000u,
+				  bar0 + 0x00405840u);
 			if (ds_post_log < 24u) {
 				ds_post_log++;
 				pr_info("vgsp-gr: DS-ESR post-clear=%08x h=0x%x tag=%s\n",
@@ -18680,6 +18687,32 @@ u32 gsp_phys_rmcall_gr_sph_gs_mode_value(void)
 {
 	return READ_ONCE(gsp_phys_rmcall_gr_sph_gs_mode);
 }
+
+/*
+ * SESSION 155 CONT 14: raster-enable prelude. The client's per-frame 3D
+ * window carries no SET_CT_WRITE / SET_DEPTH_TEST / SET_DEPTH_WRITE /
+ * SET_STENCIL_TEST / SET_SCISSOR_ENABLE - those are channel state on the
+ * client's own channel - so the replay channel keeps the hardware
+ * defaults. If CT_WRITE defaults to 0 every rasterized fragment is
+ * discarded and the surface stays black even though the draw executes.
+ * Emit the safe render-enable state ahead of each replayed 3D window:
+ *   SET_CT_WRITE(0..7)    0x1a00+i*4 <- 0x1111 (R|G|B|A write enable)
+ *   SET_DEPTH_TEST        0x12cc      <- 0
+ *   SET_DEPTH_WRITE       0x12e8      <- 0
+ *   SET_STENCIL_TEST      0x1380      <- 0
+ *   SET_SCISSOR_ENABLE(0) 0x0e00      <- 0
+ * Default 0 = legacy bytes (A/B).
+ */
+static bool gsp_phys_rmcall_gr_raster_pre;
+module_param_named(sema_gl_raster_prelude, gsp_phys_rmcall_gr_raster_pre,
+		   bool, 0644);
+MODULE_PARM_DESC(sema_gl_raster_prelude,
+	"Emit CT_WRITE(0..7)=0x1111 + depth/stencil/scissor disable ahead of replayed 3D windows (default 0; A/B)");
+
+bool gsp_phys_rmcall_gr_raster_pre_value(void)
+{
+	return READ_ONCE(gsp_phys_rmcall_gr_raster_pre);
+}
 static bool gsp_phys_gr_mme_pairs_emitted;	/* launch emitted a pair run */
 static u32 gsp_phys_gr_mme_pairs_gen;		/* gen whose pairs retired OK */
 static u32 gsp_phys_gr_mme_launch_gen;		/* last built launch's gen */
@@ -21279,7 +21312,8 @@ if (gsp_phys_gr_slot_kept(j, sa_mask,
 					u32 esr = ioread32(ds_bar0 + 0x00405840u);
 
 					if (esr)
-						iowrite32(esr, ds_bar0 + 0x00405840u);
+						iowrite32(esr | 0x40000000u,
+							  ds_bar0 + 0x00405840u);
 					if (ds_mask_log < 24u) {
 						ds_mask_log++;
 						pr_info("vgsp-gr: DS-ESR pre-clear=%08x (prev window's latch; this window starts clean) h=0x%x push=0x%llx\n",
@@ -29740,6 +29774,8 @@ gr_page_unmapped:
 				READ_ONCE(gsp_phys_rmcall_gr_sph_gs_img);
 			u32 gs_mode =
 				READ_ONCE(gsp_phys_rmcall_gr_sph_gs_mode);
+			bool emit_raster =
+				READ_ONCE(gsp_phys_rmcall_gr_raster_pre);
 
 			/*
 			 * SESSION 155 CONT 9: the prelude is prepended ahead of
@@ -29882,6 +29918,31 @@ gr_page_unmapped:
 				     (0x1128u >> 2);
 			in[nfin++] = 0xffffffffu;
 			in[nfin++] = 0xffffffffu;
+			/*
+			 * SESSION 155 CONT 14: the render-enable state the
+			 * client's channel carried from its init push
+			 * (see the sema_gl_raster_prelude comment).
+			 */
+			if (emit_raster) {
+				u32 ri5;
+
+				in[nfin++] = 0x20000000u | (8u << 16) |
+					     (0x1a00u >> 2);
+				for (ri5 = 0; ri5 < 8u; ri5++)
+					in[nfin++] = 0x1111u;
+				in[nfin++] = 0x20000000u | (1u << 16) |
+					     (0x12ccu >> 2);
+				in[nfin++] = 0u;
+				in[nfin++] = 0x20000000u | (1u << 16) |
+					     (0x12e8u >> 2);
+				in[nfin++] = 0u;
+				in[nfin++] = 0x20000000u | (1u << 16) |
+					     (0x1380u >> 2);
+				in[nfin++] = 0u;
+				in[nfin++] = 0x20000000u | (1u << 16) |
+					     (0x0e00u >> 2);
+				in[nfin++] = 0u;
+			}
 
 			/*
 			 * SESSION 153-CONTINUED-89: the client's init-push
@@ -29943,7 +30004,8 @@ gr_page_unmapped:
 			    (emit_j2 ? (bind_binding ? 8u : 7u) : 0u) +
 			    (emit_j4 ? (bind_binding ? 8u : 7u) : 0u) +
 			    (emit_j5 ? (bind_binding ? 8u : 7u) : 0u) +
-			    (emit_gsimg ? (9u + ((gs_mode & 1u) ? 11u : 0u)) : 0u)
+			    (emit_gsimg ? (9u + ((gs_mode & 1u) ? 11u : 0u)) : 0u) +
+			    (emit_raster ? 17u : 0u)
 			    <= 96u) {
 				/*
 				 * SESSION 155 CONT 9: pair_safe emits each