diff --git a/drivers/gpu/nvgpu/os/linux/platform_gk20a_tegra.c b/drivers/gpu/nvgpu/os/linux/platform_gk20a_tegra.c
index 0977e63ae..c39e4f0e6 100644
--- a/drivers/gpu/nvgpu/os/linux/platform_gk20a_tegra.c
+++ b/drivers/gpu/nvgpu/os/linux/platform_gk20a_tegra.c
@@ -799,6 +799,7 @@ static int gk20a_tegra_probe(struct device *dev)
 
 	if (joint_xpu_rail) {
 		nvgpu_log_info(g, "XPU rails are joint\n");
+		platform->can_railgate_init = false;
 		__nvgpu_set_enabled(g, NVGPU_CAN_RAILGATE, false);
 	}
 
diff --git a/drivers/gpu/nvgpu/os/linux/platform_gp10b_tegra.c b/drivers/gpu/nvgpu/os/linux/platform_gp10b_tegra.c
index c5527f2d7..9bf8d637c 100644
--- a/drivers/gpu/nvgpu/os/linux/platform_gp10b_tegra.c
+++ b/drivers/gpu/nvgpu/os/linux/platform_gp10b_tegra.c
@@ -141,6 +141,8 @@ static void gp10b_tegra_scale_exit(struct device *dev)
 static int gp10b_tegra_probe(struct device *dev)
 {
 	struct gk20a_platform *platform = dev_get_drvdata(dev);
+	bool joint_xpu_rail = false;
+	struct gk20a *g = platform->g;
 #ifdef CONFIG_TEGRA_GK20A_NVHOST
 	int ret;
 
@@ -163,6 +165,17 @@ static int gp10b_tegra_probe(struct device *dev)
 	platform->g->gr.ctx_vars.force_preemption_gfxp = false;
 	platform->g->gr.ctx_vars.force_preemption_cilp = false;
 
+#ifdef CONFIG_OF
+	joint_xpu_rail = of_property_read_bool(of_chosen,
+				"nvidia,tegra-joint_xpu_rail");
+#endif
+
+	if (joint_xpu_rail) {
+		nvgpu_log_info(g, "XPU rails are joint\n");
+		platform->can_railgate_init = false;
+		__nvgpu_set_enabled(g, NVGPU_CAN_RAILGATE, false);
+	}
+
 	gp10b_tegra_get_clocks(dev);
 	nvgpu_linux_init_clk_support(platform->g);
 
diff --git a/drivers/gpu/nvgpu/os/linux/platform_gv11b_tegra.c b/drivers/gpu/nvgpu/os/linux/platform_gv11b_tegra.c
index ac1958a25..6c9d0f5cd 100644
--- a/drivers/gpu/nvgpu/os/linux/platform_gv11b_tegra.c
+++ b/drivers/gpu/nvgpu/os/linux/platform_gv11b_tegra.c
@@ -73,6 +73,8 @@ static int gv11b_tegra_probe(struct device *dev)
 {
 	struct gk20a_platform *platform = dev_get_drvdata(dev);
 	int err;
+	bool joint_xpu_rail = false;
+	struct gk20a *g = platform->g;
 
 	err = nvgpu_nvhost_syncpt_init(platform->g);
 	if (err) {
@@ -94,6 +96,17 @@ static int gv11b_tegra_probe(struct device *dev)
 	platform->g->gr.ctx_vars.force_preemption_gfxp = false;
 	platform->g->gr.ctx_vars.force_preemption_cilp = false;
 
+#ifdef CONFIG_OF
+	joint_xpu_rail = of_property_read_bool(of_chosen,
+				"nvidia,tegra-joint_xpu_rail");
+#endif
+
+	if (joint_xpu_rail) {
+		nvgpu_log_info(g, "XPU rails are joint\n");
+		platform->can_railgate_init = false;
+		__nvgpu_set_enabled(g, NVGPU_CAN_RAILGATE, false);
+	}
+
 	gp10b_tegra_get_clocks(dev);
 	nvgpu_linux_init_clk_support(platform->g);
 
diff --git a/drivers/gpu/nvgpu/os/linux/sysfs.c b/drivers/gpu/nvgpu/os/linux/sysfs.c
index 2ab296498..84a311de3 100644
--- a/drivers/gpu/nvgpu/os/linux/sysfs.c
+++ b/drivers/gpu/nvgpu/os/linux/sysfs.c
@@ -259,13 +259,19 @@ static ssize_t railgate_enable_store(struct device *dev,
 	unsigned long railgate_enable = 0;
 	/* dev is guaranteed to be valid here. Ok to de-reference */
 	struct gk20a *g = get_gk20a(dev);
+	struct gk20a_platform *platform = dev_get_drvdata(dev);
 	bool enabled = nvgpu_is_enabled(g, NVGPU_CAN_RAILGATE);
 	int err;
 
 	if (kstrtoul(buf, 10, &railgate_enable) < 0)
 		return -EINVAL;
 
-	if (railgate_enable && !enabled) {
+	if (!platform->can_railgate_init) {
+		nvgpu_err(g, "Railgating is not supported");
+		return -EINVAL;
+	}
+
+	if (railgate_enable) {
 		__nvgpu_set_enabled(g, NVGPU_CAN_RAILGATE, true);
 		pm_runtime_set_autosuspend_delay(dev, g->railgate_delay);
 	} else if (railgate_enable == 0 && enabled) {