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.

3 files changed, 179 insertions, 39 deletions

diff --git a/device_info_procfs.c b/device_info_procfs.c
index 1fc0586..b1e58b1 100644
--- a/device_info_procfs.c
+++ b/device_info_procfs.c
@@ -31,7 +31,7 @@ struct file_operations nvdebug_read_reg32_file_ops = {
 
 // Called to start or resume a sequence. Prior to 4.19, *pos is unreliable.
 // Initializes iterator `idx` state and returns it. Ends sequence on NULL.
-static void* device_info_file_seq_start(struct seq_file *s, loff_t *pos) {
+static void* device_info_file_seq_start_previous(struct seq_file *s, loff_t *pos) {
         static int idx;
         // If start of sequence, reset `idx`
         if (*pos == 0)
@@ -42,9 +42,21 @@ static void* device_info_file_seq_start(struct seq_file *s, loff_t *pos) {
         return &idx;
 }
 
+static void* device_info_file_seq_start_ampere(struct seq_file *s, loff_t *pos) {
+        static int idx;
+        // If start of sequence, reset `idx`
+        if (*pos == 0)
+                idx = 0;
+        struct nvdebug_state *g = &g_nvdebug_state[seq2gpuidx(s)];
+        // Number of possible info entries is fixed, and list is sparse
+        if (idx >= (nvdebug_readl(g, 0x0224fc) >> 20))
+                return NULL;
+        return &idx;
+}
+
 // Steps to next record. Returns new value of `idx`.
 // Calls show() on non-NULL return
-static void* device_info_file_seq_next(struct seq_file *s, void *idx,
+static void* device_info_file_seq_next_previous(struct seq_file *s, void *idx,
                                        loff_t *pos) {
         (*pos)++; // Required by seq interface
         // Number of possible info entries is fixed, and list is sparse
@@ -53,12 +65,50 @@ static void* device_info_file_seq_next(struct seq_file *s, void *idx,
         return idx;
 }
 
+// Steps to next record. Returns new value of `idx`.
+// Calls show() on non-NULL return
+static void* device_info_file_seq_next_ampere(struct seq_file *s, void *idx,
+                                       loff_t *pos) {
+        (*pos)++; // Required by seq interface
+        // Number of possible info entries is fixed, and list is sparse
+        struct nvdebug_state *g = &g_nvdebug_state[seq2gpuidx(s)];
+        if ((*(int*)idx)++ >= (nvdebug_readl(g, 0x0224fc) >> 20))
+                    return NULL;
+        return idx;
+}
+/*
+// Steps to next record on Ampere GPUs. Returns new value of `idx`.
+static void* device_info_file_seq_next_ampere(struct seq_file *s, void *idx,
+                                       loff_t *pos) {
+        (*pos)++; // Required by seq interface
+        // Number of possible info entries is fixed, and list is sparse
+        while(1) {
+            struct nvdebug_state *g = &g_nvdebug_state[seq2gpuidx(s)];
+            if ((*(int*)idx)++ >= (nvdebug_readl(g, 0x0224fc) >> 20))
+                    return NULL;
+            ptop_device_info_t curr_info;
+            curr_info.raw = nvdebug_readl(g, NV_PTOP_DEVICE_INFO_AMPERE(*(int*)idx));
+            if(!curr_info.raw && !*info_type) continue;
+            (*info_type)++;
+            break;
+        }
+        while(1) {
+            struct nvdebug_state *g = &g_nvdebug_state[seq2gpuidx(s)];
+            if ((*(int*)idx)++ >= (nvdebug_readl(g, 0x0224fc) >> 20))
+                    return NULL;
+            ptop_device_info_t curr_info;
+            curr_info.raw = nvdebug_readl(g, NV_PTOP_DEVICE_INFO_AMPERE(*(int*)idx));
+            if(curr_info.raw & 0x80000000) continue;
+            break;
+        }
+        return idx;
+}
+*/
 // Print info at index *idx. Returns non-zero on error.
-static int device_info_file_seq_show(struct seq_file *s, void *idx) {
+static int device_info_file_seq_show_previous(struct seq_file *s, void *idx) {
         ptop_device_info_t curr_info;
         struct nvdebug_state *g = &g_nvdebug_state[seq2gpuidx(s)];
-        
-        curr_info.raw = nvdebug_readl(g, NV_PTOP_DEVICE_INFO(*(int*)idx));
+        curr_info.raw = nvdebug_readl(g, NV_PTOP_DEVICE_INFO_PREVIOUS(*(int*)idx));
         // Check for read errors
         if (curr_info.raw == -1)
                 return -EIO;
@@ -104,24 +154,88 @@ static int device_info_file_seq_show(struct seq_file *s, void *idx) {
         return 0;
 }
 
+// Print info at index *idx for Ampere GPUs. Returns non-zero on error.
+static int device_info_file_seq_show_ampere(struct seq_file *s, void *idx) {
+        ptop_device_info_t curr_info;
+        struct nvdebug_state *g = &g_nvdebug_state[seq2gpuidx(s)];
+        curr_info.raw = nvdebug_readl(g, NV_PTOP_DEVICE_INFO_AMPERE(*(int*)idx));
+        // Check for read errors
+        if (curr_info.raw == -1)
+                return -EIO;
+        int info_type = -1;
+        if(curr_info.raw) {
+                if(*(int*)idx < 1 || !nvdebug_readl(g, NV_PTOP_DEVICE_INFO_AMPERE(*(int*)idx) - 1)) info_type = 0;
+                if(*(int*)idx < 2 || !nvdebug_readl(g, NV_PTOP_DEVICE_INFO_AMPERE(*(int*)idx) - 2)) info_type = 1;
+                if(*(int*)idx < 3 || !nvdebug_readl(g, NV_PTOP_DEVICE_INFO_AMPERE(*(int*)idx) - 3)) info_type = 2;
+        }
+        // Parse and print the data
+        switch(info_type) {
+        case 0:
+                seq_printf(s, "|           instance %d\n", curr_info.inst_id_ampere);
+                seq_printf(s, "| Fault ID:        %3d\n", curr_info.fault_id_ampere);
+                seq_printf(s, "| Engine Type:      %2d (", curr_info.engine_type_ampere);
+                if (curr_info.engine_type_ampere < ENGINE_TYPES_LEN)
+                        seq_printf(s, "%s)\n", ENGINE_TYPES_NAMES[curr_info.engine_type_ampere]);
+                else
+                        seq_printf(s, "Unknown Engine, introduced post-Ampere)\n");
+                break;
+        case 1:
+                seq_printf(s, "| BAR0 Base %#.8x\n", curr_info.pri_base_ampere << 12);
+                seq_printf(s, "| Reset ID:         %2d\n", curr_info.reset_enum_ampere);
+                break;
+        case 2:
+                seq_printf(s, "| Host's Engine ID: %2d\n", curr_info.engine_enum_ampere);
+                seq_printf(s, "| Runlist ID:       %2d\n", curr_info.runlist_enum_ampere);
+                break;
+        default:
+                // Device info records are sparse, so skip unset or unknown ones
+                return 0;
+        }
+
+        // Draw a line between each device entry
+        if (!curr_info.has_next_entry_ampere) {
+                seq_printf(s, "+---------------------+\n");
+        }
+        return 0;
+}
+
+
 static void device_info_file_seq_stop(struct seq_file *s, void *idx) {
         // No cleanup needed
 }
 
-static const struct seq_operations device_info_file_seq_ops = {
-        .start = device_info_file_seq_start,
-        .next = device_info_file_seq_next,
+static const struct seq_operations device_info_file_seq_ops_previous = {
+        .start = device_info_file_seq_start_previous,
+        .next = device_info_file_seq_next_previous,
         .stop = device_info_file_seq_stop,
-        .show = device_info_file_seq_show,
+        .show = device_info_file_seq_show_previous,
+};
+
+static const struct seq_operations device_info_file_seq_ops_ampere = {
+        .start = device_info_file_seq_start_ampere,
+        .next = device_info_file_seq_next_ampere,
+        .stop = device_info_file_seq_stop,
+        .show = device_info_file_seq_show_ampere,
 };
 
-static int device_info_file_open(struct inode *inode, struct file *f) {
-        return seq_open(f, &device_info_file_seq_ops);
+static int device_info_file_open_previous(struct inode *inode, struct file *f) {
+        return seq_open(f, &device_info_file_seq_ops_previous);
 }
 
-struct file_operations device_info_file_ops = {
-        .open = device_info_file_open,
+static int device_info_file_open_ampere(struct inode *inode, struct file *f) {
+        return seq_open(f, &device_info_file_seq_ops_ampere);
+}
+
+struct file_operations device_info_file_ops_previous = {
+        .open = device_info_file_open_previous,
         .read = seq_read,
         .llseek = seq_lseek,
         .release = seq_release,
 };
+
+struct file_operations device_info_file_ops_ampere = {
+        .open = device_info_file_open_ampere,
+        .read = seq_read,
+        .llseek = seq_lseek,
+        .release = seq_release,
+};
diff --git a/nvdebug.h b/nvdebug.h
index f6e057e..3ccdcfe 100644
--- a/nvdebug.h
+++ b/nvdebug.h
@@ -554,34 +554,30 @@ static const char* const ENGINE_TYPES_NAMES[ENGINE_TYPES_LEN] = {
   See dev_top.ref.txt of NVIDIA's open-gpu-doc for more info.
 */
 
-#ifdef NV_BUILD_FOR_AMPERE
-#define NV_PTOP_DEVICE_INFO(i) (0x00022800+(i)*4)
+#define NV_PTOP_DEVICE_INFO_AMPERE(i) (0x00022800+(i)*4)
+#define NV_PTOP_DEVICE_INFO_PREVIOUS(i) (0x00022700+(i)*4)
 #define NV_PTOP_DEVICE_INFO__SIZE_1 64
 typedef union {
         struct {
-                uint32_t fault_id:7;
+                uint32_t fault_id_ampere:7;