gpu: nvgpu: Debugfs support for Railgating stats.

This patch calculates:
-Total time spent by GPU with rails gated.
-Total time spent by GPU with rails ungated.
-Total Railgating Cycles.
and dumps this information in debugfs file.

This feature requires CONFIG_DEBUG_FS set to true.

Bug 200195100

Change-Id: I1379f11237ce4900076947e18524caaa3304c7cb
Signed-off-by: Deepak Goyal <dgoyal@nvidia.com>
Reviewed-on: http://git-master/r/1178308
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>

1 files changed, 97 insertions, 0 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.c b/drivers/gpu/nvgpu/gk20a/gk20a.c
index ab417386..9255c847 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.c
@@ -94,6 +94,71 @@ u32 gk20a_dbg_ftrace;
 
 static int gk20a_pm_prepare_poweroff(struct device *dev);
 
+#ifdef CONFIG_DEBUG_FS
+static int railgate_residency_show(struct seq_file *s, void *data)
+{
+        struct device *dev = s->private;
+        struct gk20a_platform *platform = dev_get_drvdata(dev);
+        struct gk20a *g = get_gk20a(dev);
+        unsigned long time_since_last_state_transition_ms;
+        unsigned long total_rail_gate_time_ms;
+        unsigned long total_rail_ungate_time_ms;
+
+        if (platform->is_railgated(dev)) {
+                time_since_last_state_transition_ms =
+                                jiffies_to_msecs(jiffies -
+                                g->pstats.last_rail_gate_complete);
+                total_rail_ungate_time_ms = g->pstats.total_rail_ungate_time_ms;
+                total_rail_gate_time_ms =
+                                        g->pstats.total_rail_gate_time_ms +
+                                        time_since_last_state_transition_ms;
+        } else {
+                time_since_last_state_transition_ms =
+                                jiffies_to_msecs(jiffies -
+                                g->pstats.last_rail_ungate_complete);
+                total_rail_gate_time_ms = g->pstats.total_rail_gate_time_ms;
+                total_rail_ungate_time_ms =
+                                        g->pstats.total_rail_ungate_time_ms +
+                                        time_since_last_state_transition_ms;
+        }
+
+        seq_printf(s, "Time with Rails Gated: %lu ms\n"
+                        "Time with Rails UnGated: %lu ms\n"
+                        "Total railgating cycles: %lu\n",
+                        total_rail_gate_time_ms,
+                        total_rail_ungate_time_ms,
+                        g->pstats.railgating_cycle_count - 1);
+        return 0;
+
+}
+
+static int railgate_residency_open(struct inode *inode, struct file *file)
+{
+        return single_open(file, railgate_residency_show, inode->i_private);
+}
+
+static const struct file_operations railgate_residency_fops = {
+        .open           = railgate_residency_open,
+        .read           = seq_read,
+        .llseek         = seq_lseek,
+        .release        = single_release,
+};
+
+int gk20a_railgating_debugfs_init(struct device *dev)
+{
+        struct dentry *d;
+        struct gk20a_platform *platform = dev_get_drvdata(dev);
+
+        d = debugfs_create_file(
+                "railgate_residency", S_IRUGO|S_IWUSR, platform->debugfs, dev,
+                                                &railgate_residency_fops);
+        if (!d)
+                return -ENOMEM;
+
+        return 0;
+}
+#endif
+
 static inline void set_gk20a(struct platform_device *pdev, struct gk20a *gk20a)
 {
         gk20a_get_platform(&pdev->dev)->g = gk20a;
@@ -1133,10 +1198,25 @@ static int gk20a_pm_railgate(struct device *dev)
 {
         struct gk20a_platform *platform = dev_get_drvdata(dev);
         int ret = 0;
+#ifdef CONFIG_DEBUG_FS
+        struct gk20a *g = get_gk20a(dev);
+
+        g->pstats.last_rail_gate_start = jiffies;
+
+        if (g->pstats.railgating_cycle_count >= 1)
+                g->pstats.total_rail_ungate_time_ms =
+                        g->pstats.total_rail_ungate_time_ms +
+                        jiffies_to_msecs(g->pstats.last_rail_gate_start -
+                                        g->pstats.last_rail_ungate_complete);
+#endif
 
         if (platform->railgate)
                 ret = platform->railgate(dev);
 
+#ifdef CONFIG_DEBUG_FS
+        g->pstats.last_rail_gate_complete = jiffies;
+#endif
+
         return ret;
 }
 
@@ -1144,6 +1224,18 @@ static int gk20a_pm_unrailgate(struct device *dev)
 {
         struct gk20a_platform *platform = dev_get_drvdata(dev);
         int ret = 0;
+#ifdef CONFIG_DEBUG_FS
+        struct gk20a *g = get_gk20a(dev);
+
+        g->pstats.last_rail_ungate_start = jiffies;
+        if (g->pstats.railgating_cycle_count >= 1)
+                g->pstats.total_rail_gate_time_ms =
+                        g->pstats.total_rail_gate_time_ms +
+                        jiffies_to_msecs(g->pstats.last_rail_ungate_start -
+                                g->pstats.last_rail_gate_complete);
+
+        g->pstats.railgating_cycle_count++;
+#endif
 
         trace_gk20a_pm_unrailgate(dev_name(dev));
 
@@ -1153,6 +1245,10 @@ static int gk20a_pm_unrailgate(struct device *dev)
                 mutex_unlock(&platform->railgate_lock);
         }
 
+#ifdef CONFIG_DEBUG_FS
+        g->pstats.last_rail_ungate_complete = jiffies;
+#endif
+
         return ret;
 }
 
@@ -1531,6 +1627,7 @@ static int gk20a_probe(struct platform_device *dev)
 
         gr_gk20a_debugfs_init(gk20a);
         gk20a_pmu_debugfs_init(&dev->dev);
+        gk20a_railgating_debugfs_init(&dev->dev);
         gk20a_cde_debugfs_init(&dev->dev);
         gk20a_alloc_debugfs_init(dev);
         gk20a_mm_debugfs_init(&dev->dev);