GSPRM CLK legacy controls: populate the domain queries the runtime clock helper gates on

The static CUDA runtime's device init failed with cudaErrorInitializationError
even though every kernel ioctl returned 0: cuInit succeeded and then
cuDeviceGetAttribute(ECC_ENABLED=36) returned CUDA_ERROR_NOT_FOUND (500).
Tracing libcuda's clock-domain helper (libcuda+0xe8c410) shows it issues the
CLK legacy controls 0x20809009/0x20809064/0x20809001 and bails with 500 when
the replies do not carry the expected domain bits/entries; the dependent
attribute getters (ECC_ENABLED, TOTAL_CONSTANT_MEMORY) then abort init.

Reply to the three queries with the minimal fields the helper checks: bits
32|34 present in 0x20809009, non-zero domain entries at offsets 12/20/28/36
of 0x20809064, and a matching selector in 0x20809001. No frequencies are
fabricated (all other fields stay zero).

Verified in the guest with the stock 570.144 static-cudart vectorAdd client:
cudaGetDeviceCount -> rc=0 (ndev=1), cudaSetDevice, cudaFree, three
cudaMallocs and both HtoD copies all return rc=0; the client then parks in
cudaDeviceSynchronize waiting for the (not yet forwarded) kernel launch.

1 files changed, 62 insertions, 0 deletions

diff --git a/kernel/gsp_rm_emulation.c b/kernel/gsp_rm_emulation.c
index ff65275..d6aef9e 100644
--- a/kernel/gsp_rm_emulation.c
+++ b/kernel/gsp_rm_emulation.c
@@ -2031,6 +2031,60 @@ static void gsp_rm_handle_fb_get_info_v2(struct gsp_emu_context *ctx,
 }
 
 /*
+ * CLK legacy domain queries (0x20809001 / 0x20809009 / 0x20809064)
+ *
+ * libcuda's clock-domain helper (libcuda+0xe8c410) walks these three queries
+ * before it will report clock/ECC/perf data. It requires:
+ *   - 0x20809009: bits 32 and 34 set in the 8-byte reply (domain presence);
+ *   - 0x20809064: non-zero domain entries at reply offsets 12, 20, 28, 36;
+ *   - 0x20809001: a selector mask overlapping those entries.
+ * With the generic zero reply the helper returns CUDA_ERROR_NOT_FOUND, which
+ * makes the dependent device-attribute queries (ECC_ENABLED = attr 36,
+ * TOTAL_CONSTANT_MEMORY = attr 13) abort device initialization with
+ * cudaErrorInitializationError. Populate only the fields the checks read;
+ * clock rates stay zero (no fabricated frequencies).
+ */
+static void gsp_rm_handle_clk_domain_info(struct gsp_emu_context *ctx, u32 cmd,
+					  u32 req_size,
+					  u32 *out_status,
+					  u8 *out_params,
+					  u32 *out_params_size)
+{
+	u32 n = req_size ? req_size : 8;
+	u32 one = 1;
+
+	if (n > 4096)
+		n = 4096;
+	memset(out_params, 0, n);
+
+	if (cmd == TU104_CTRL_CLK_DOMAIN_MASK) {
+		u64 mask = (1ULL << 32) | (1ULL << 34);
+
+		if (n >= sizeof(mask))
+			memcpy(out_params, &mask, sizeof(mask));
+	} else if (cmd == TU104_CTRL_CLK_DOMAIN_DATA) {
+		if (n >= 16)
+			memcpy(out_params + 12, &one, sizeof(one));
+		if (n >= 24)
+			memcpy(out_params + 20, &one, sizeof(one));
+		if (n >= 32)
+			memcpy(out_params + 28, &one, sizeof(one));
+		if (n >= 40)
+			memcpy(out_params + 36, &one, sizeof(one));
+	} else { /* TU104_CTRL_CLK_DOMAIN_SEL */
+		if (n >= sizeof(one))
+			memcpy(out_params, &one, sizeof(one));
+	}
+
+	*out_params_size = n;
+	*out_status = 0;
+
+	if (gsp_sema_dbg)
+		nv_vgpu_info(ctx->vgpu, "GSPRM: CLK domain info cmd=0x%x size=%u",
+			     cmd, n);
+}
+
+/*
  * BUS_GET_INFO_V2 (0x20801823)
  *
  * Same flat index/data list pattern as FB_GET_INFO_V2. UVM's
@@ -2641,6 +2695,14 @@ static void gsp_rm_handle_rm_control(struct gsp_emu_context *ctx,
 			reply + GSP_RM_CTRL_HDR_SIZE, &out_params_size);
 		break;
 
+	case TU104_CTRL_CLK_DOMAIN_MASK:
+	case TU104_CTRL_CLK_DOMAIN_DATA:
+	case TU104_CTRL_CLK_DOMAIN_SEL:
+		gsp_rm_handle_clk_domain_info(ctx, req.cmd, req.paramsSize,
+			&out_status,
+			reply + GSP_RM_CTRL_HDR_SIZE, &out_params_size);
+		break;
+
 	case TU104_CTRL_INTERNAL_DISPLAY_GET_STATIC_INFO:
 		gsp_rm_handle_display_get_static_info(ctx, sequence,
 			&out_status,