diff options
| -rw-r--r-- | kernel/gsp_rm_emulation.c | 62 | ||||
| -rw-r--r-- | kernel/gsp_rm_tu104.h | 3 |
2 files changed, 65 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, diff --git a/kernel/gsp_rm_tu104.h b/kernel/gsp_rm_tu104.h index 7cd5a4b..b5ee4d7 100644 --- a/kernel/gsp_rm_tu104.h +++ b/kernel/gsp_rm_tu104.h @@ -728,6 +728,9 @@ static const struct tu104_internal_device_info tu104_internal_device_table[] = { #define TU104_CTRL_FIFO_GET_DEVICE_INFO_TABLE 0x20801112 #define TU104_CTRL_INTERNAL_INTR_GET_KERNEL_TABLE 0x20800a5c #define TU104_CTRL_GPU_GET_CONSTRUCTED_FALCON_INFO 0x208001b0 +#define TU104_CTRL_CLK_DOMAIN_MASK 0x20809009 +#define TU104_CTRL_CLK_DOMAIN_DATA 0x20809064 +#define TU104_CTRL_CLK_DOMAIN_SEL 0x20809001 #define TU104_CTRL_INTERNAL_FIFO_GET_NUM_CHANNELS 0x20800a61 #define TU104_CTRL_BUS_GET_PCI_BAR_INFO 0x20801803 #define TU104_CTRL_INTERNAL_DISPLAY_GET_STATIC_INFO 0x20800a01 |