[PATCH] i2c: cadence: fix clk disable/unprepare underflow warning on remove

[Cong Nguyen]

cdns_i2c_probe() enables id->clk via devm_clk_get_enabled(), which
auto-disables/unprepares it on unbind. Runtime PM autosuspend
separately enables/disables the same clock as the device idles and
wakes. remove() never forces the device back active first -- if it's
runtime-suspended at removal (routine after any idle period, not a
rare race), the devm cleanup then disables an already-disabled clock,
hitting the "already disabled"/"already unprepared" WARN().

i2c-imx.c has the identical shape and already gets this right:
pm_runtime_get_sync() first, balanced with pm_runtime_put_noidle()
before the final disable. Apply the same fix here.

Fixes: 3d36dd1161ca ("i2c: cadence: Simplify using devm_clk_get_enabled()")
Assisted-by: LLM
Signed-off-by: Cong Nguyen <congnt264@xxxxxxxxx>
---
 drivers/i2c/busses/i2c-cadence.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/i2c/busses/i2c-cadence.c b/drivers/i2c/busses/i2c-cadence.c
index da8770182a18..d0919593456b 100644
--- a/drivers/i2c/busses/i2c-cadence.c
+++ b/drivers/i2c/busses/i2c-cadence.c
@@ -1625,14 +1625,21 @@ static int cdns_i2c_probe(struct platform_device *pdev)
 static void cdns_i2c_remove(struct platform_device *pdev)
 {
 	struct cdns_i2c *id = platform_get_drvdata(pdev);
+	int ret;
+
+	ret = pm_runtime_get_sync(&pdev->dev);
+	if (ret < 0)
+		dev_err(&pdev->dev, "Failed to resume device (%pe)\n", ERR_PTR(ret));
 
-	pm_runtime_disable(&pdev->dev);
-	pm_runtime_set_suspended(&pdev->dev);
 	pm_runtime_dont_use_autosuspend(&pdev->dev);
 
 	i2c_del_adapter(&id->adap);
 	clk_notifier_unregister(id->clk, &id->clk_rate_change_nb);
 	reset_control_assert(id->reset);
+
+	pm_runtime_put_noidle(&pdev->dev);
+	pm_runtime_disable(&pdev->dev);
+	pm_runtime_set_suspended(&pdev->dev);
 }
 
 /**
-- 
2.25.1