This commit is to update the repo for display during a meeting.

- Added an Ampere version of the device info data.
- Added Ampere versions of auxillary functions.
- Modified display functions to accommodate Ampere data.
- Made other various small modifications.

1 files changed, 36 insertions, 5 deletions

diff --git a/nvdebug_entry.c b/nvdebug_entry.c
index 8e8266c..d82c648 100644
--- a/nvdebug_entry.c
+++ b/nvdebug_entry.c
@@ -23,12 +23,14 @@ extern struct file_operations preempt_tsg_file_ops;
 extern struct file_operations disable_channel_file_ops;
 extern struct file_operations enable_channel_file_ops;
 extern struct file_operations switch_to_tsg_file_ops;
-extern struct file_operations device_info_file_ops;
+extern struct file_operations device_info_file_ops_previous;
+extern struct file_operations device_info_file_ops_ampere;
 extern struct file_operations nvdebug_read_reg32_file_ops;
 
 // Bus types are global symbols in the kernel
 extern struct bus_type platform_bus_type;
 struct nvdebug_state g_nvdebug_state[NVDEBUG_MAX_DEVICES];
+int g_architectures[NVDEBUG_MAX_DEVICES];
 unsigned int g_nvdebug_devices = 0;
 
 // Starting in Kernel 5.6, proc_ops is required instead of file_operations
@@ -123,6 +125,7 @@ int probe_and_cache_device(void) {
                         return -EADDRNOTAVAIL;
                 }
                 g_nvdebug_state[i].chip_id = ids.chip_id;
+                g_architectures[i] = ids.architecture;
                 printk(KERN_INFO "[nvdebug] Chip ID %x (architecture %s) detected on PCI bus and initialized.",
                        ids.chip_id, ARCH2NAME(ids.architecture));
                 // TEMP
@@ -138,7 +141,7 @@ int probe_and_cache_device(void) {
 }
 
 // Create files `/proc/gpu#/runlist#`, world readable
-int create_runlist_files(int device_id, struct proc_dir_entry *dir) {
+int create_runlist_files_previous(int device_id, struct proc_dir_entry *dir) {
         ptop_device_info_t info;
         struct proc_dir_entry *rl_entry;
         int i, rl_id;
@@ -148,7 +151,7 @@ int create_runlist_files(int device_id, struct proc_dir_entry *dir) {
         // registers. Runlists are always numbered sequentially, so we just have
         // to find the highest-valued one and add 1 to get the number of runlists.
         for (i = 0; i < NV_PTOP_DEVICE_INFO__SIZE_1; i++) {
-                info.raw = nvdebug_readl(&g_nvdebug_state[device_id], NV_PTOP_DEVICE_INFO(i));
+                info.raw = nvdebug_readl(&g_nvdebug_state[device_id], NV_PTOP_DEVICE_INFO_PREVIOUS(i));
                 if (info.info_type != INFO_TYPE_ENUM || !info.runlist_is_valid)
                         continue;
                 if (info.runlist_enum > max_rl_id)
@@ -167,6 +170,34 @@ int create_runlist_files(int device_id, struct proc_dir_entry *dir) {
         return 0;
 }
 
+// Create files `/proc/gpu#/runlist#`, world readable
+int create_runlist_files_ampere(int device_id, struct proc_dir_entry *dir) {
+        ptop_device_info_t info;
+        struct proc_dir_entry *rl_entry;
+        int i, rl_id;
+        char runlist_name[12];
+        int max_rl_id = 0; // Always at least one runlist
+        // Figure out how many runlists there are by checking the device info
+        // registers. Runlists are always numbered sequentially, so we just have
+        // to find the highest-valued one and add 1 to get the number of runlists.
+        for (i = 0; i < (nvdebug_readl(&g_nvdebug_state[device_id], 0x0224fc) >> 20); i++) {
+                info.raw = nvdebug_readl(&g_nvdebug_state[device_id], NV_PTOP_DEVICE_INFO_AMPERE(i));
+                if (info.runlist_enum_ampere > max_rl_id)
+                        max_rl_id = info.runlist_enum;
+        }
+        // Create files to read each runlist. The read handling code looks at the
+        // PDE_DATA associated with the file to determine what the runlist ID is.
+        for (rl_id = 0; rl_id <= max_rl_id; rl_id++) {
+                snprintf(runlist_name, 12, "runlist%d", rl_id);
+                rl_entry = proc_create_data(
+                        runlist_name, 0444, dir, compat_ops(&runlist_file_ops),
+                        (void*)(uintptr_t)rl_id);
+                if (!rl_entry)
+                        return -ENOMEM;
+        }
+        return 0;
+}
+
 // Create files /proc/gpu#
 // TODO: Don't run this on unsupported GPUs
 int create_tpc_mask_files(int device_id, struct proc_dir_entry *dir) {
@@ -212,7 +243,7 @@ int __init nvdebug_init(void) {
                 if (!(dir = proc_mkdir_data(device_id_str, 0555, NULL, (void*)device_id)))
                         goto out_nomem;
                 // Create files `/proc/gpu#/runlist#`, world readable
-                rl_create_err = create_runlist_files(device_id, dir);
+                rl_create_err = g_architectures[device_id] == 0x17 ? create_runlist_files_ampere(device_id, dir) : create_runlist_files_previous(device_id, dir);
                 // Create files `/proc/gpu#/gpc#_tpc_mask`, world readable
                 tpc_masks_create_err = create_tpc_mask_files(device_id, dir);
                 // Create file `/proc/gpu#/preempt_tsg`, world writable
@@ -233,7 +264,7 @@ int __init nvdebug_init(void) {
                         (void*)device_id);
                 // Create file `/proc/gpu#/device_info`, world readable
                 device_info_entry = proc_create_data(
-                        "device_info", 0444, dir, compat_ops(&device_info_file_ops),
+                        "device_info", 0444, dir, compat_ops(g_architectures[device_id] == 0x17 ? &device_info_file_ops_previous : &device_info_file_ops_ampere),
                         (void*)device_id);
                 // Create file `/proc/gpu#/num_gpcs`, world readable
                 num_gpcs_entry = proc_create_data(