[Nouveau] [PATCH v2] gr/gf100: provide a bit more info for various errors

[Ilia Mirkin]

Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
---

v1 -> v2: undo slight change to SKED stat handling, which would cause writes
where there previously might not have been any.

 drm/nouveau/nvkm/engine/gr/gf100.c | 78 +++++++++++++++++++++++++++++++++-----
 1 file changed, 69 insertions(+), 9 deletions(-)

diff --git a/drm/nouveau/nvkm/engine/gr/gf100.c b/drm/nouveau/nvkm/engine/gr/gf100.c
index 414521b..1f81069 100644
--- a/drm/nouveau/nvkm/engine/gr/gf100.c
+++ b/drm/nouveau/nvkm/engine/gr/gf100.c
@@ -856,7 +856,41 @@ gf100_gr_units(struct nvkm_gr *base)
 	return cfg;
 }
 
+static const struct nvkm_bitfield gf100_dispatch_error[] = {
+	{ 0x00000001, "INJECTED_BUNDLE_ERROR" },
+	{ 0x00000002, "CLASS_SUBCH_MISMATCH" },
+	{ 0x00000004, "SUBCHSW_DURING_NOTIFY" },
+	{}
+};
+
+static const struct nvkm_bitfield gf100_m2mf_error[] = {
+	{ 0x00000001, "PUSH_TOO_MUCH_DATA" },
+	{ 0x00000002, "PUSH_NOT_ENOUGH_DATA" },
+	{}
+};
+
+static const struct nvkm_bitfield gf100_unk6_error[] = {
+	{ 0x00000001, "TEMP_TOO_SMALL" },
+	{}
+};
+
+static const struct nvkm_bitfield gf100_ccache_error[] = {
+	{ 0x00000001, "INTR" },
+	{ 0x00000002, "LDCONST_OOB" },
+	{}
+};
+
+static const struct nvkm_bitfield gf100_macro_error[] = {
+	{ 0x00000001, "TOO_FEW_PARAMS" },
+	{ 0x00000002, "TOO_MANY_PARAMS" },
+	{ 0x00000004, "ILLEGAL_OPCODE" },
+	{ 0x00000008, "DOUBLE_BRANCH" },
+	{ 0x00000010, "WATCHDOG" },
+	{}
+};
+
 static const struct nvkm_bitfield gk104_sked_error[] = {
+	{ 0x00000040, "CTA_RESUME" },
 	{ 0x00000080, "CONSTANT_BUFFER_SIZE" },
 	{ 0x00000200, "LOCAL_MEMORY_SIZE_POS" },
 	{ 0x00000400, "LOCAL_MEMORY_SIZE_NEG" },
@@ -866,6 +900,8 @@ static const struct nvkm_bitfield gk104_sked_error[] = {
 	{ 0x00040000, "TOTAL_THREADS" },
 	{ 0x00100000, "PROGRAM_OFFSET" },
 	{ 0x00200000, "SHARED_MEMORY_SIZE" },
+	{ 0x00800000, "CTA_THREAD_DIMENSION_ZERO" },
+	{ 0x01000000, "MEMORY_WINDOW_OVERLAP" },
 	{ 0x02000000, "SHARED_CONFIG_TOO_SMALL" },
 	{ 0x04000000, "TOTAL_REGISTER_COUNT" },
 	{}
@@ -1035,12 +1071,16 @@ gf100_gr_trap_intr(struct gf100_gr *gr)
 {
 	struct nvkm_subdev *subdev = &gr->base.engine.subdev;
 	struct nvkm_device *device = subdev->device;
+	char error[128];
 	u32 trap = nvkm_rd32(device, 0x400108);
 	int rop, gpc;
 
 	if (trap & 0x00000001) {
 		u32 stat = nvkm_rd32(device, 0x404000);
-		nvkm_error(subdev, "DISPATCH %08x\n", stat);
+
+		nvkm_snprintbf(error, sizeof(error), gf100_dispatch_error,
+			       stat & 0x3fffffff);
+		nvkm_error(subdev, "DISPATCH %08x [%s]\n", stat, error);
 		nvkm_wr32(device, 0x404000, 0xc0000000);
 		nvkm_wr32(device, 0x400108, 0x00000001);
 		trap &= ~0x00000001;
@@ -1048,7 +1088,11 @@ gf100_gr_trap_intr(struct gf100_gr *gr)
 
 	if (trap & 0x00000002) {
 		u32 stat = nvkm_rd32(device, 0x404600);
-		nvkm_error(subdev, "M2MF %08x\n", stat);
+
+		nvkm_snprintbf(error, sizeof(error), gf100_m2mf_error,
+			       stat & 0x3fffffff);
+		nvkm_error(subdev, "M2MF %08x [%s]\n", stat, error);
+
 		nvkm_wr32(device, 0x404600, 0xc0000000);
 		nvkm_wr32(device, 0x400108, 0x00000002);
 		trap &= ~0x00000002;
@@ -1056,7 +1100,10 @@ gf100_gr_trap_intr(struct gf100_gr *gr)
 
 	if (trap & 0x00000008) {
 		u32 stat = nvkm_rd32(device, 0x408030);
-		nvkm_error(subdev, "CCACHE %08x\n", stat);
+
+		nvkm_snprintbf(error, sizeof(error), gf100_m2mf_error,
+			       stat & 0x3fffffff);
+		nvkm_error(subdev, "CCACHE %08x [%s]\n", stat, error);
 		nvkm_wr32(device, 0x408030, 0xc0000000);
 		nvkm_wr32(device, 0x400108, 0x00000008);
 		trap &= ~0x00000008;
@@ -1064,7 +1111,8 @@ gf100_gr_trap_intr(struct gf100_gr *gr)
 
 	if (trap & 0x00000010) {
 		u32 stat = nvkm_rd32(device, 0x405840);
-		nvkm_error(subdev, "SHADER %08x\n", stat);
+		nvkm_error(subdev, "SHADER %08x, sph: 0x%06x, stage: 0x%02x\n",
+			   stat, stat & 0xffffff, (stat >> 24) & 0x3f);
 		nvkm_wr32(device, 0x405840, 0xc0000000);
 		nvkm_wr32(device, 0x400108, 0x00000010);
 		trap &= ~0x00000010;
@@ -1072,7 +1120,11 @@ gf100_gr_trap_intr(struct gf100_gr *gr)
 
 	if (trap & 0x00000040) {
 		u32 stat = nvkm_rd32(device, 0x40601c);
-		nvkm_error(subdev, "UNK6 %08x\n", stat);
+
+		nvkm_snprintbf(error, sizeof(error), gf100_unk6_error,
+			       stat & 0x3fffffff);
+		nvkm_error(subdev, "UNK6 %08x [%s]\n", stat, error);
+
 		nvkm_wr32(device, 0x40601c, 0xc0000000);
 		nvkm_wr32(device, 0x400108, 0x00000040);
 		trap &= ~0x00000040;
@@ -1080,7 +1132,16 @@ gf100_gr_trap_intr(struct gf100_gr *gr)
 
 	if (trap & 0x00000080) {
 		u32 stat = nvkm_rd32(device, 0x404490);
-		nvkm_error(subdev, "MACRO %08x\n", stat);
+		u32 pc = nvkm_rd32(device, 0x404494);
+		u32 op = nvkm_rd32(device, 0x40449c);
+
+		nvkm_snprintbf(error, sizeof(error), gf100_macro_error,
+			       stat & 0x1fffffff);
+		nvkm_error(subdev, "MACRO %08x [%s], pc: 0x%03x%s, op: 0x%08x\n",
+			   stat, error, pc & 0x7ff,
+			   (pc & 0x10000000) ? "" : " (invalid)",
+			   op);
+
 		nvkm_wr32(device, 0x404490, 0xc0000000);
 		nvkm_wr32(device, 0x400108, 0x00000080);
 		trap &= ~0x00000080;
@@ -1088,10 +1149,9 @@ gf100_gr_trap_intr(struct gf100_gr *gr)
 
 	if (trap & 0x00000100) {
 		u32 stat = nvkm_rd32(device, 0x407020) & 0x3fffffff;
-		char sked[128];
 
-		nvkm_snprintbf(sked, sizeof(sked), gk104_sked_error, stat);
-		nvkm_error(subdev, "SKED: %08x [%s]\n", stat, sked);
+		nvkm_snprintbf(error, sizeof(error), gk104_sked_error, stat);
+		nvkm_error(subdev, "SKED: %08x [%s]\n", stat, error);
 
 		if (stat)
 			nvkm_wr32(device, 0x407020, 0x40000000);
-- 
2.4.10