Since the power driver probes from device tree, sandbox has no poweroff
or restart handler when started with a --dtb that lacks a
barebox,sandbox-power node.
The poweroff and reset commands will then end up in hang(), which spins
at 100% CPU until the process is killed.
Let's register the device ourselves when the device tree has no such
node.
Fixes: d2748c1ae720 ("sandbox: poweroff: migrate to driver probed from device
tree")
Signed-off-by: Ahmad Fatoum <[email protected]>
---
arch/sandbox/board/power.c | 20 ++++++++++++++++++++
1 file changed, 20 insertions(+)
diff --git a/arch/sandbox/board/power.c b/arch/sandbox/board/power.c
index 9caaea5be2f0..a95e25b57215 100644
--- a/arch/sandbox/board/power.c
+++ b/arch/sandbox/board/power.c
@@ -2,6 +2,8 @@
#include <common.h>
#include <driver.h>
+#include <init.h>
+#include <of.h>
#include <poweroff.h>
#include <restart.h>
#include <mach/linux.h>
@@ -68,6 +70,9 @@ static int sandbox_power_probe(struct device *dev)
if (IS_ENABLED(CONFIG_SANDBOX_REEXEC))
restart_handler_register(&power->rst_reexec);
+ if (!dev->of_node)
+ return 0;
+
power->reset_source_cell = of_nvmem_cell_get(dev->of_node,
"reset-source");
if (IS_ERR(power->reset_source_cell)) {
@@ -100,3 +105,18 @@ static struct driver sandbox_power_drv = {
.probe = sandbox_power_probe,
};
coredevice_platform_driver(sandbox_power_drv);
+
+/*
+ * A device tree passed with --dtb need not have a sandbox-power node.
+ * Without one, poweroff and reset end in hang(), which busy-loops forever,
+ * so register the device ourselves in this case.
+ */
+static int sandbox_power_fallback(void)
+{
+ if (!of_find_compatible_node(NULL, NULL, "barebox,sandbox-power"))
+ add_generic_device("sandbox-power", DEVICE_ID_SINGLE, NULL, 0,
+ 0, 0, NULL);
+
+ return 0;
+}
+coredevice_initcall(sandbox_power_fallback);
--
2.47.3
