From e4c3f8554d70dcf22567621528a4601d7444666b Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Wed, 30 Sep 2026 04:04:22 -0400 Subject: gsp-rm: fill GR info indices 7/9/0x17/0x22 (libcuda clock-table path; e94 TPC count) --- kernel/gsp_rm_emulation.c | 16 +++++++++++++++- 1 file changed, 15 insertions(+), 1 deletion(-) (limited to 'kernel/gsp_rm_emulation.c') diff --git a/kernel/gsp_rm_emulation.c b/kernel/gsp_rm_emulation.c index 777c63c..a9b63b1 100644 --- a/kernel/gsp_rm_emulation.c +++ b/kernel/gsp_rm_emulation.c @@ -1775,11 +1775,25 @@ static void gsp_rm_handle_kgr_get_info(struct gsp_emu_context *ctx, /* TU104 litter topology: 6 GPC x 4 TPC x 2 SM = 48 SMs */ info->infoList[TU104_GR_INFO_IDX_LITTER_NUM_GPCS].data = TU104_NUM_GPC; - info->infoList[TU104_GR_INFO_IDX_LITTER_NUM_TPC_PER_GPC].data = TU104_TPC_PER_GPC; + info->infoList[TU104_GR_INFO_IDX_LITTER_NUM_TPC_PER_GPC].data = + TU104_GR_INFO_LITTER_NUM_TPC_PER_GPC; info->infoList[TU104_GR_INFO_IDX_GPU_CORE_COUNT].data = TU104_NUM_SM; info->infoList[TU104_GR_INFO_IDX_LITTER_NUM_SM_PER_TPC].data = TU104_SM_PER_TPC; info->infoList[TU104_GR_INFO_IDX_FAMILY_MAX_TPC_PER_GPC].data = TU104_TPC_PER_GPC; + /* Indices libcuda fetches via NV2080_CTRL_CMD_GR_GET_INFO when it + * builds the ucode-image clock table (GR_GET_INFO copies + * pGrInfo->infoList[index].data). Observed on real TU104: index 9 + * -> 24 (devstate+0xe94 TPC count; 0 makes FUN_003e35f0 return + * NV_ERR_INVALID_ARGUMENT and cuCtxCreate fail 999), index 7 -> 6, + * index 0x17 -> 6 (devstate+0xb0/+0xb4 SM/TPC), index 0x22 -> 48. */ + info->infoList[TU104_GR_INFO_IDX_SHADER_PIPE_COUNT].data = + TU104_GR_INFO_SHADER_PIPE_COUNT; + info->infoList[TU104_GR_INFO_IDX_SHADER_PIPE_SUB_COUNT].data = + TU104_GR_INFO_SHADER_PIPE_SUB_COUNT; + info->infoList[TU104_GR_INFO_IDX_RT_CORE_COUNT].data = + TU104_GR_INFO_RT_CORE_COUNT; + /* subcontext (VEID) counts: 64 per hw */ info->infoList[TU104_GR_INFO_IDX_MAX_SUBCONTEXT_COUNT].data = TU104_MAX_SUBCONTEXT_COUNT; info->infoList[TU104_GR_INFO_IDX_MAX_LEGACY_SUBCONTEXT_COUNT].data = TU104_MAX_SUBCONTEXT_COUNT; -- cgit v1.2.3