added offsets for lce mapping in nvdebug.h and code to read lce for each pce in nvdebug_entry.c

2 files changed, 19 insertions, 2 deletions

diff --git a/nvdebug.h b/nvdebug.h
index b79ede1..6aa0f11 100644
--- a/nvdebug.h
+++ b/nvdebug.h
@@ -687,6 +687,8 @@ typedef union {
   Also see dev_ce.ref.txt of NVIDIA's open-gpu-doc for info.
 */
 #define NV_PTOP_SCAL_NUM_CES 0x00022444
+//defined LCE->PCE mapping offset from nvgpu (same as ce_pce2lce_config_r(i) in nvgpu) 
+#define NV_LCE_FOR_PCE(i)(0x00104040+(i)*4)
 
 /* Physical Copy Engine (PCE) information
   On Pascal GPUs or newer, this register complements the above information by
diff --git a/nvdebug_entry.c b/nvdebug_entry.c
index 0abe658..84e8a6a 100644
--- a/nvdebug_entry.c
+++ b/nvdebug_entry.c
@@ -203,7 +203,7 @@ int create_tpc_mask_files(int device_id, struct proc_dir_entry *dir) {
 int __init nvdebug_init(void) {
         struct proc_dir_entry *dir, *preempt_entry, *disable_channel_entry,
                               *enable_channel_entry, *switch_to_tsg_entry, *device_info_entry,
-                              *num_gpcs_entry;
+                              *num_gpcs_entry, *lce_for_pce_entry;
         int rl_create_err, tpc_masks_create_err;
         // Check that an NVIDIA GPU is present and initialize g_nvdebug_state
         int res = probe_and_cache_device();
@@ -261,8 +261,23 @@ int __init nvdebug_init(void) {
                         (void*)NV_FUSE_GPC);
                 // In both nouveau and nvgpu, the PCE_MAP register is only available on Volta+
                 if (g_nvdebug_state[res].chip_id >= NV_CHIP_ID_VOLTA) {
+                        
+                        //create a pce mask for iteration
+                        u32 ce_pce_map = nvdebug_readl(&g_nvdebug_state[device_id], NV_CE_PCE_MAP);
+                        u32 num_pce =  U32(hweight32(ce_pce_map));
+                        u32 disabled_pce_mask = ~ce_pce_map;
+                        char file_name[20];
+                        for (int pce_id = 0; pce_id < num_pce; pce_id++){
+                                //if pce is disabled, do nothing
+                                if ((1 << pce_id) & disabled_pce_mask)
+                                        continue;
+                                snprintf(file_name, 20, "lce_for_pce%d",pce_id);
+                                lce_for_pce_entry = proc_create_data(
+                                file_name, 0444, dir, compat_ops(&nvdebug_read_reg32_file_ops),
+                                (void*)NV_LCE_FOR_PCE(pce_id));                                 
+                        }
+
                         // TODO: Redo to num_pces
-                        // Create file `/proc/gpu#/pce_map`, world readable
                         num_gpcs_entry = proc_create_data(
                                 "pce_map", 0444, dir, compat_ops(&nvdebug_read_reg32_file_ops),
                                 (void*)NV_CE_PCE_MAP);