[v2] i2c: stm32f7: Do not prepare/unprepare clock during runtime suspend/resume

Commit Message

Marek Vasut Sept. 30, 2024, 7:27 p.m. UTC

In case there is any sort of clock controller attached to this I2C bus
controller, for example Versaclock or even an AIC32x4 I2C codec, then
an I2C transfer triggered from the clock controller clk_ops .prepare
callback may trigger a deadlock on drivers/clk/clk.c prepare_lock mutex.

This is because the clock controller first grabs the prepare_lock mutex
and then performs the prepare operation, including its I2C access. The
I2C access resumes this I2C bus controller via .runtime_resume callback,
which calls clk_prepare_enable(), which attempts to grab the prepare_lock
mutex again and deadlocks.

Since the clock are already prepared since probe() and unprepared in
remove(), use simple clk_enable()/clk_disable() calls to enable and
disable the clock on runtime suspend and resume, to avoid hitting the
prepare_lock mutex.

Acked-by: Alain Volmat <[e-mail]>
Signed-off-by: Marek Vasut <[e-mail]>
---
Cc: Alain Volmat <[e-mail]>
Cc: Alexandre Torgue <[e-mail]>
Cc: Andi Shyti <[e-mail]>
Cc: Christoph Niedermaier <[e-mail]>
Cc: Maxime Coquelin <[e-mail]>
Cc: Pierre-Yves MORDRET <[e-mail]>
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
---
V2: - Update the error message
    - Add AB from Alain
---
 drivers/i2c/busses/i2c-stm32f7.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

Comments

Andi Shyti Oct. 1, 2024, 2:40 p.m. UTC | #1

On Mon, Sep 30, 2024 at 09:27:41PM GMT, Marek Vasut wrote:
> In case there is any sort of clock controller attached to this I2C bus
> controller, for example Versaclock or even an AIC32x4 I2C codec, then
> an I2C transfer triggered from the clock controller clk_ops .prepare
> callback may trigger a deadlock on drivers/clk/clk.c prepare_lock mutex.
> 
> This is because the clock controller first grabs the prepare_lock mutex
> and then performs the prepare operation, including its I2C access. The
> I2C access resumes this I2C bus controller via .runtime_resume callback,
> which calls clk_prepare_enable(), which attempts to grab the prepare_lock
> mutex again and deadlocks.
> 
> Since the clock are already prepared since probe() and unprepared in
> remove(), use simple clk_enable()/clk_disable() calls to enable and
> disable the clock on runtime suspend and resume, to avoid hitting the
> prepare_lock mutex.
> 
> Acked-by: Alain Volmat <[e-mail]>
> Signed-off-by: Marek Vasut <[e-mail]>

I think we also need:

Fixes: 4e7bca6fc07b ("i2c: i2c-stm32f7: add PM Runtime support")
Cc: <[e-mail]> # v5.0+

I'm adding them, please, let me know if you think they are not
needed.

For now I merged this patch in i2c/i2c-host-fixes.

Thanks,
Andi

Patch

diff --git a/drivers/i2c/busses/i2c-stm32f7.c b/drivers/i2c/busses/i2c-stm32f7.c
index cfee2d9c09de3..0174ead99de6c 100644
--- a/drivers/i2c/busses/i2c-stm32f7.c
+++ b/drivers/i2c/busses/i2c-stm32f7.c
@@ -2395,7 +2395,7 @@  static int __maybe_unused stm32f7_i2c_runtime_suspend(struct device *dev)
 	struct stm32f7_i2c_dev *i2c_dev = dev_get_drvdata(dev);
 
 	if (!stm32f7_i2c_is_slave_registered(i2c_dev))
-		clk_disable_unprepare(i2c_dev->clk);
+		clk_disable(i2c_dev->clk);
 
 	return 0;
 }
@@ -2406,9 +2406,9 @@  static int __maybe_unused stm32f7_i2c_runtime_resume(struct device *dev)
 	int ret;
 
 	if (!stm32f7_i2c_is_slave_registered(i2c_dev)) {
-		ret = clk_prepare_enable(i2c_dev->clk);
+		ret = clk_enable(i2c_dev->clk);
 		if (ret) {
-			dev_err(dev, "failed to prepare_enable clock\n");
+			dev_err(dev, "failed to enable clock\n");
 			return ret;
 		}
 	}